Processing of Particulate Solids

Powder Technology Series

EDITED BY
BRIAN SCARLETT and GENJI JIMBO
Delft University of Technology Chuba Powtech Plaza Lab
The Netherlands Japan

Many materials exist in the form of a disperse system, for example

powders, pastes, slurries, emulsions and aerosols. The study of such systems
necessarily arises in many technologies but may alternatively be regarded as
a separate subject which is concerned with the manufacture,
characterization and manipulation of such systems. Chapman & Hall were
one of the first publishers to recognize the basic importance of the subject,
going on to instigate this series of books. The series does not aspire to
define and confine the subject without duplication, but rather to provide a
good home for any book which has a contribution to make to the record of
both the theory and the application of the subject. We hope that all
engineers and scientists who concern themselves with disperse systems will
use these books and that those who become expert will contribute further to
the series.

Particle Size Measurement

Terence Allen
5th edn, hardback (0412753502), 2 volume set, 552 and 272 pages

Chemistry of Powder Production

Yasuo Arai
Hardback (041239540 1), 292 pages

Particle Size Analysis

Claus Bernhardt
Translated by H. Finken
Hardback (0412558807), 428 pages

Particle Classification
K. Heiskanen
Hardback (0412493004), 330 pages

Powder Surface Area and Porosity

S. Lowell and Joan E. Shields
3rd edn, hardback (0412 39690 4), 256 pages
Pneumatic Conveying of Solids
G.E. Klinzing, R.D. Marcus, F. Rizk and L.S. Leung
2nd edn, hardback (0412724405), 624 pages

Principles of Flow in Disperse Systems

O. Molerus
Hardback (0412406306), 314 pages

Processing of Particulate Solids

J.P.K. Seville, U. Tiiziin and R. Clift
Hardback (0751403768), 384 pages
JOIN US ON THE INTERNET VIA WWW, GOPHER, FTP OR EMAIL:
WWW: http://www.thomson.com
GOPHER: gopher.thomson.com Iri'\®
A service of I(!)P
FTP: ftp.thomson.com
EMAIL: [email protected]
 Processing of
Particulate Solids

Jonathan Seville
Professor of Chemical Engineering
School of Chemical Engineering
University of Birmingham, UK

Ugur Tuzun
Professor of Process Engineering
Department of Chemical and Process Engineering
University of Surrey
Guildford, UK

and

Roland Clift
Professor of Environmental Technology
Centre for Environmental Strategy
University of Surrey
Guildford, UK

BLACKIE ACADEMIC cSt PROFESSIONAL

An Imprint of Chapman & Hall

London· Weinheim . New York· Tokyo· Melbourne· Madras

Published by B1ackie Academic & Professional, an imprint of
Chapman & Hall, 2-6 Boundary Row, London SEt 8HN, UK

Chapman & Hall, 2-6 Boundary Row, London SEl 8HN, UK

Chapman & Hall GmbH, Pappelallee 3, 69469 Weinheim, Germany
Chapman & Hall USA, Fourth Floor, 115 Fifth Avenue, New York
NY 10003, USA
Chapman & Hall Japan, ITP-Japan, Kyowa Building, 3F, 2-2-1
Hirakawacho, Chiyoda-ku, Tokyo 102, Japan
DA Book (Aust.) Pty Ltd, 648 Whitehorse Road, Mitcham 3132, Victoria,
Australia
Chapman & Hall India, R. Seshadri, 32 Second Main Road, CIT East,
Madras 600 035, India

First edition 1997

© 1997 Chapman & Hall
Typeset in 10112pt Times by Cambrian Typesetters, Frimley, Surrey

ISBN-13:978-94-010-7152-9 elSBN:978-94-009-1459-9
DOI:I0.I0071978-94-009-1459-9

Apart from any fair dealing for the purposes of research or private study,
or criticism or review, as permitted under the UK Copyright Designs and
Patents Act, 1988, this publication may not be reproduced, stored, or
transmitted, in any form or by any means, without the prior permission
in writing of the publishers, or in the case of reprographic reproduction
only in accordance with the terms of the licences issued by the Copyright
Licensing Agency in the UK, or in accordance with the terms of licences
issued by the appropriate Reproduction Rights Organization outside the
UK. Enquiries concerning reproduction outside the terms stated here
should be sent to the publishers at the London address printed on this
page.
The publisher makes no representation, express or implied, with
regard to the accuracy of the information contained in this book and
cannot accept any legal responsibility or liability for any errors or
omissions that may be made.
A catalogue record for this book is available from the British Library
Library of Congress Catalog Card Number: 96-83857

GPrinted on permanent acid-free text paper, manufactured in

accordance with ANSIINISO Z39.48-1992 (Permanence of Paper).
Contents

Preface xi

1 Particle characterisation 1
1.1 Particle size distributions 1
1.1.1 Definitions 1
1.1.2 Methods of presenting size data 3
1.1.3 Averages 7
1.1.4 Weighted distributions 9
1.1.5 Model distribution functions 16
1.2 Particle size measurement 18
1.2.1 Particle size 18
1.2.2 Particle shape 21
1.2.3 Sampling 26
1.2.4 Methods of particle size measurement 30
1.3 Nomenclature 48
A1 Derivation and properties of the log-normal distribution 49
References 52

2 Particles in fluids 53
2.1 Single particles 53
2.1.1 Fluid-particle drag 53
2.1.2 Fluid motion 56
2.1.3 Free fall or rise 59
2.1.4 Non-continuum effects 76
2.2 Unsteady motion of single particles 79
2.2.1 Drag in unsteady motion 79
2.2.2 Acceleration 80
2.2.3 Curvilinear motion 81
2.3 Assemblages of particles 85
2.3.1 Settling and particulate fluidisation 85
2.3.2 Flow through packed beds 86
2.4 Flow of particlelfluid mixtures in pipes 87
2.4.1 Continuity 88
2.4.2 Momentum balance 91
viii Contents

2.4.3 Mechanical energy balance 95

2.5 Nomenclature 96
References 98

3 Particle mechanics 99
3.1 Interparticle forces 99
3.1.1 van der Waals forces 99
3.1.2 Liquid bridge forces 109
3.1.3 Electrostatic forces 120
3.1.4 Comparison of the magnitude of interparticle forces 123
3.2 Effects of interparticle forces at contacts 125
3.2.1 Contact mechanics 125
3.2.2 Assembly mechanics 127
3.3 Friction at a single contact 132
3.4 Impact and rebound of particles 135
3.4.1 The coefficient of restitution 137
3.4.2 The effect ofliquid layers 140
3.5 Nomenclature 142
References 143

4 Characterisation of bulk mechanical properties 145

4.1 Introduction 145
4.2 Empirical measurements of coefficients of friction 146
4.3 Angles of friction of cohesionless materials 148
4.4 Angles offriction of cohesive materials 151
4.5 Characterisation of 'flowability' of cohesive powders
(Jenike design method) 157
4.5.1 Material flow function (FF) 158
4.5.2 Jenike's hopper flow factor (ff) 159
4.5.3 Jenike's minimum critical outlet span for flow 159
4.6 Equations of stress equilibrium in bulk solids 160
4.7 Bulk failure criterion 163
4.8 Description of bulk solids stress states 165
4.9 Bulk solids flow rules 167
4.10 Empirical measurements of compressibility of bulk solids 168
4.11 Nomenclature 171
References 172

5 Assembly mechanics 174

5.1 Introduction 174
5.2 Assembly modelling of wall friction in a flowing particulate bed 174
5.2.1 Normal stress and dilatancy 176
5.3 The distinct element method 179
5.3.1 General features 179
 Contents ix

5.3.2 History and current applications 180

5.3.3 Methodology 182
5.3.4 Forces 187
5.3.5 Particle size and shape 190
5.3.6 Stress analysis 193
5.4 Nomenclature 196
A5 Frictional forces 197
A5.1 Frictional forces 197
A5.2 Single contacts 198
A5.3 Particle assemblies 198
References 199

6 Fluid-particle systems 201

6.1 Hydraulic and pneumatic conveying 201
6.1.1 Differences and similarities 201
6.1.2 Hydraulic transport 205
6.1.3 Pneumatic transport 221.
6.2 Fluidisation 226
6.2.1 Types offluidisation behaviour 227
6.2.2 General description of group behaviour 233
6.2.3 Criteria for group boundaries 236
6.2.4 Temperature and pressure effects 239
6.2.5 Defluidisation and cohesive effects 240
6.2.6 Bubbling fluidised beds 242
6.2.7 Agglomeration 250
6.3 Nomenclature 257
References 258

7 Gas/solid separation 261

7.1 Gas and particle properties 261
7.2 Inertial separators 263
7.2.1 Introduction 263
7.2.2 Analysis of cyclone performance 267
7.2.3 Effects of solid loading 272
7.2.4 Effects of temperature and pressure 273
7.2.5 Concluding remarks 274
7.3 Filtration 274
7.3.1 General features of filtration behaviour 274
7.3.2 Depth filtration 277
7.3.3 Barrierfiltration 285
7.3.4 Concluding remarks 294
7.4 Nomenclature 294
References 295
x Contents

8 Storage and discharge of particulate bulk solids 298

8.1 Introduction 298
8.1.1 General design considerations for 'process' silos 299
8.1.2 Design considerations for large storage silos 301
8.2 Flow regimes in bins and hoppers 302
8.2.1 Mass flow versus funnel flow 302
8.2.2 Other possible flow regimes 306
8.2.3 Choice of flow regime 306
8.3 Velocity distributions in bins and hoppers 308
8.3.1 Steady-state flow fields for 'free-flowing' materials 308
8.3.2 Transient flow fields of 'free-flowing' materials 326
8.4 Discharge rates from bins and hoppers 330
8.4.1 Introduction 330
8.4.2 Early empirical work and discharge rate correlations 331
8.4.3 Theoretical predictions of mass flow rate 337
8.4.4 Effect of interstitial pressure gradients 343
8.5 Stress distributions in bins and hoppers 348
8.5.1 Introduction 348
8.5.2 Approximate analysis: the method of differential slices 350
8.5.3 More realistic analysis of the stresses at the wall 356
8.5 .4 Janssen-Walker analysis in other hopper geometries 358
8.5.5 Walters' switch stress analysis 361
8.6 Nomenclature 363
References 364

Index 369
Preface

Well over half the products of the chemical and process industries are sold in
a particulate form. The range of such products is vast: from agrochemicals to
pigments, from detergents to foods, from plastics to pharmaceuticals.
However, surveys of the performance of processes designed to produce
particulate products have consistently shown inadequate design and poor
reliability. The science of fluid mechanics is usually dated back to Archimedes
and developed through the work of such as Galileo, Newton, D' Alembert,
Darcy and Euler. By the end of the last century, fluid handling had become a
scientifically based technological industry. In contrast, it was only in the 1950s
that the study of solids handling was seriously attempted and many solids
processing devices are still designed on a more-or-Iess empirical basis.
Particle technology is thus a very new subject and it faces new challenges. It
is apparent that chemical and process engineering is becoming less concerned
with the design of plants to produce generic simple chemicals (which are often
single phase fluids) and are more concerned with speciality 'effect' chemicals
which may often be multi-phase and multi-component. Chemical and process
engineers are also being recruited in increasing numbers into areas outside
their traditional fields, e.g. the food industry, pharmaceuticals and the
manufacture of consumer products of all kinds. We intend this book to meet
their needs.
We have aimed at a comprehensive coverage of the technology of
particulate solids in a form which is both sufficiently accessible and
sufficiently concise to be useful to engineering and science students in the
final year of an undergraduate degree and at Masters' level. Although it was
written with students of chemical engineering in mind, concern with granular
solids and powders is not limited to one engineering discipline. We therefore
hope that this book will be of use and interest to students of other disciplines.
The content and style follow a pattern which we have found useful in teaching
undergraduate and graduate students at the Universities of Surrey, Birmingham
and British Columbia.
The book is divided into two halves: fundamentals and applications. The
behaviour of single spheres in simple fluids is a standard part of any chemical
engineering degree; the challenge that we have attempted to meet here is to
widen that treatment to include non-spherical particles over a wide range of
xii Preface

conditions, generalising the approach in order to make it simpler to

understand. Particle-particle and particle-surface interactions are much less
well understood, at least by most engineers, and there is much less agreement
about what and how much it is proper to include in a book of this kind. We
are convinced that contact mechanics and the study of interfaces in particulate
systems is vital to the understanding of the subject and have included those
topics which we have ourselves found helpful. Whilst continuum treatments
of the behaviour of materials in bulk have proved useful - and we have
covered these fully here - progress in modelling their behaviour depends on
the development of assembly mechanics, particularly distinct element
methods. We have therefore summarised the distinct element approaches in
the confident expectation that they will help us to achieve major advances in
the next decade.
The applications of particle technology are extremely diverse. We have
necessarily been selective and chosen to cover examples which illustrate the
fundamental areas from the first part of the book. These examples include
several multi-phase flow processes, the separation of particles from gases and
the storage and discharge of bulk particulate materials. We have deliberately
chosen not to consider colloidal suspensions since this would involve
consideration of surface forces in a depth beyond that possible in a book of
this size. For similar reasons, we have had to exclude liquid/solid filtration,
comminution and aggregation/agglomeration.
Most of the examples that we have chosen to include here have been
developed from our personal contact with engineers in industry, struggling to
design and operate equipment for processing of particulate solids, often with
inadequate understanding of the fundamentals of the subject. It is our hope
that this book will be useful to them also.
One of the attractions of working in a relatively new subject such as particle
technology is that one is always breaking new ground, making new
connections with other subjects and finding new uses for principles and
analyses from other disciplines. We hope that our enthusiasm and excitement
in pursuing this subject will be apparent to readers of the book and that it will
bring new and fresh minds into an important area of human endeavour.
The authors are hugely indebted to the staff and students of the
Universities of Birmingham, Surrey and British Columbia for their suggestions
and encouragement during the time that this book was written, especially Dr
Paul Langston for his significant contribution to Chapter 5, to a succession of
long-suffering secretarial staff who have helped us to assemble the contents
and, perhaps chiefly, to our families for their understanding of our frequent
prolonged absences in spirit and, from time to time, in body.

J.P.K. Seville
U. Tiiziin
R. Clift
1
Particle characterisation

The most important characteristics of a particle are its size, its shape and its
density. At first sight these may seem simple unambiguous properties which
should be straightforward to determine. In practice, this is seldom the case,
particularly if the particle in question is smaller than, say, the diameter of a
human hair. This chapter is an introduction to the science of particle
characterisation; for a more detailed examination of this diverse subject the
reader is referred to Allen (1996). In practice, particles seldom occur singly,
and the first part of this chapter is concerned with the properties of particle
size distributions.

1.1 PARTICLE SIZE DISTRIBUTIONS

1.1.1 Definitions
A particle may be defined as a single entity comprising part of a solid or liquid
discontinuous phase. According to this definition, a particle may have any
size. It is common to refer to a suspension of particles in a gas as an aerosol
and to particles in suspension in a liquid as a sol (hydrosol if the liquid is
water). Clearly, when we are considering the stability of a suspension, particle
size becomes important. For example, a suspension of 1 ftm (one micrometre
= 1 ftm = 10-6 m) particles in air may remain stable for many minutes,
whereas 100 ftm particles will settle out in seconds. Similarly, flow rates of
particles from hoppers and bins, reaction rates of particles in reactors,
separation efficiencies in filters and separators and most other aspects of
particle behaviour depend on particle size.
Figure 1.1 shows the size ranges for a variety of commonly occurring
materials. It is important to note that the ranges of particle size are often very
wide, and that they sometimes exceed the ranges of application of physical
laws. For example, for a suspension in a gas of particles smaller than 1 ftm in
diameter, the gas can no longer be treated as a continuum because the size of
the particles becomes comparable to the mean free path of the gas molecules.
(The effect which this has on particle settling velocities is considered in
Chapter 2.) The limit of resolution of an optical microscope is related to the
wavelength of the illuminating light, 0.55 ftm for white light. Thus, particles

J. Seville et al., Processing of Particulate Solids

© Chapman & Hall 1997
 (1 mil) Particle diameter, microns (11) (1 mm) (1 cm)
0.0001 0.001 0.01 0.1 1 10 100 1000 10000
I 234 68,1 ~ ~1,?,~,1 ~ ~1'?'~lLJ ~iJRJ 2 34 68 23468,1 f~1,M,1 fr1,?,~,1 2
I I I 115000 I
1 10 100 1000 10000 2500
Equivalent I I I
sizes Angstrom Units (A) Theoretical mesh I I I I oI 50 130 116 I 8 I 41 318'1 314'
(used very infrequently) ::28 c::?8 I I _ II I I I I
'" '"~ ~ (0
U.S. Screen Mesh
1 I I I I III I I I I
I Visibl~ I
Electromagnetic 1------- X-rays )r Ultraviolet ~~ Near infrared --1-<1-0--- I
Far :tared I etc)
--------+.- Microwaves (ralar,
I
waves J
r- I ~ Solar radiation ~
Gas I Solid: Fume Dust~-----------l-----'
Technical dispersoids I Liquid: Mist Spray --------::i
definitions 'I' Anerberg or International Std.. Ch;~.ssj!icatjon System -----.-t- Clay )I! i( Silt ~ Fine sand -¥- Coarse sand ~ Gravel--
.--1.....-
S01. Iadopted by Intemat. Soc. Soli SCI. Since 1934 I
Common atmospheric f-- ·----r--Smog I .. k--C[OUd5andfOg~Mist)l~riZZI~I' Rain-----)Io
dispersoids _1
~ Fertilizer, groundlimestone~
ROSin smoke
~Ollsmoke ~ '(~ Coal dust
Fly ash

Tobacco smokf dusts and fumes ~+-- Cement dust

MetarIlU~:Onlum chloride fume SulfurIC
2 2 6H
O CO C 6 Contact concentrator mist
CI,
H2~' _ ., ••• _---><
" '®
\ .. . . olecules
Gas - Carbon InseC!
k--- Paint~p,gment: dusts
cide
Typical particles
•• Plant
and Z'ncox'defume~ ~Gro"ndtalc
2
gas dispersoids 80 C, H
N" CH ']"1 0 .-----.
Spray dned m,Ik poresPoilens --')
~~~ '~~
O ~~
H 0 HCI
.~ .~,'_ ",~'"'_~" _
C , Atmosphenc dust --C=-~-~~-~I~~e;~rops ;'::umatlC nozzle
"",•. "~,,,~" I
m ___ --- Sea salt nucle, Lung damag,ng _I rops
---- --r - - ~ ~;t<-;:- ~
r-- ++.,
~- - - - 1'( dust dults) 7 5).t±O 3).t
•
Combus~lon I nucle, .. ed
Rblood diameter
cell , Human hair--Ji.>j
~ Viruses--+l I" I Bacteria

Figure 1.1 Particle size ranges (reprinted courtesy of SRI International).

 Particle size distributions 3

much below 0.55 [lm in diameter cannot be resolved using this method, which
has important implications for particle sizing, as will be discussed later. Most
particles show a distribution of particle size, i.e. they are polydisperse. Those
which do not are described as monodisperse, i.e. they are all of the same size.
In practice, it is only possible to approximate to the latter condition, although
in nature, plant pollen, and in industry, polymer latex spheres formed under
zero gravity are close approximations to monodisperse particles.
Fluid particles (i.e. bubbles or drops) are often spherical if they are small
enough (Clift et al., 1978) whereas solid particles are normally not. The term
'diameter' is therefore ambiguous and it is necessary to define an equivalent
diameter of a non-spherical particle as the diameter of a sphere having the
same value of a particular physical property as the particle of interest. Many
equivalent diameters are used. The choice, and the methods of measurement
are discussed later in this chapter.

1.1.2 Methods of presenting size data*

Table 1.1 is a set of particle size data for 1000 particles, the size being
measured by some unspecified method. It consists of particle counts within a
number of discrete size intervals and is therefore an example of a number
distribution. Some size measuring devices allow the limits of the sizing
intervals to be defined by the user, in which case it is preferable to keep the
resolution (the interval width divided by the mean interval size) approximately
constant, for example by using a geometric progression of interval limits: x,
x 2 , x 3 . . . xn. Frequently, the limits of the intervals are not user-defined, but
this does not prevent an accurate representation of the particle size
distribution from being obtained. The data of Table 1.1 are most simply
represented as a frequency histogram (Figure 1.2). The problem with this
representation is that the heights of the rectangles must depend to some
extent on the width of the interval, which makes comparison difficult. It is
preferable, therefore, to divide each interval count by the total number of
counts, to obtain the fractional count in each size class, and then to divide this
fraction by the interval width (usually in micrometres). If fractioni[lm is now
plotted as a histogram against particle size (Figure 1.3) the reSUlting
representation has the important property that the area under each rectangle
represents the fraction of particles in that size interval. The total area is thus
equal to one. Putting this in mathematical terms,

ni
I".
Jt = - = h.!l.d.
I l
(1.1a)
N

*This section draws on the extremely useful chapter on this subject in w.e. Hinds' book Aerosol
Technology, Wiley (1982), to which the reader is referred for a more comprehensive account of
the subject.
4 Particle characterisation

'2:.1i = '2:. (hib..d i)

i
=1 (1.2a)

where ni is the number of particles in each interval, N is the total number of

particles, hi is the height of the ith interval, of width b..di, and fi is known as
the frequency distribution function. Ii is defined as the fraction of the total
number of particles with diameters between d i and d i +1 (the ith interval).
The frequency distribution may be discrete, as in Figure 1.3, or a smooth

Table 1.1 Example of a particle size distribution

Diameter range (!-lm)

Count Fraction Percent Cumulative
Lower Upper frequency per !-lm percent

0 5 39 0.00780 3.9 3.9

5 10 175 0.03500 17.5 21.4
10 20 348 0.03480 34.8 56.2
20 30 187 0.01870 18.7 74.9
30 40 112 0.01120 11.2 86.1
40 60 89 0.00445 8.9 95.0
60 80 27 0.00135 2.7 97.7
80 100 13 0.00065 1.3 99.0
100 150 8 0.00016 0.8 99.8
150 200 2 0.00004 0.2 100.0
1000 100.0

-
300 -

- '-

100 -
fo-
I-

o Jr 111 1 1 1 1
~

1 1 I I
o 20 40 80 100 150 200
Particle diameter (microns)

Figure 1.2 Number frequency distribution.

 Particle size distributions 5

I=l 0.03~-
8u
·s
§ 0.02 - e-
'.p

£ e-
0.01 -

o-
n
I I I I I I I I I I I
o 20 40 80 100 150 200
Particle diameter (microns)

Figure 1.3 Fraction per !Lm versus particle size, discrete number distribution.

line may be drawn through the tops of the frequency rectangles at the mean
point of each interval to form a continuous distribution, as in Figure 1.4. (For
the mean of an interval it is usual to take the geometric mean, i.e. (d i d i +1)'h. If
the interval width is small, this differs little from the arithmetic mean, V2(di +
d i +1 ).
For a continuous distribution, the fraction of the total number of particles
having diameters between a and b is the integral under the distribution curve
between these limits
(LIb)

where f(d p) is the continuous frequency distribution function. The total area
is again unity:

(1.2b)

An alternative representation of particle size distributions is the cumulative

distribution, shown in Figure 1.5 for the same data as in Table 1.1. For a
continuous distribution, the cumulative distribution function, F(a), is defined
as the fraction of the total number of particles with diameters less than a,
F(a) = fa f(dp)dd p (1.3)
o
or
dF(dp )
fed ) = - - (1.4)
p ddp
 0.040 -ri- - - - - - - - - - - - - - - - - - - - - - ,

0.035

0.030

::i.
5 0.025
.=-=..... 0.020
~
~ 0.015

0.010

0.005

0.000 I v=---.; • i • • i

o 20 40 60 80 100 120 140 160 180 200

Particle diameter (J,Lm)
Figure 1.4 Fraction per ~m versus particle size, continuous number distribution.
 Particle size distributions 7

Thus the frequency function at any point can be obtained from the slope of
the cumulative distribution function.
Because the cumulative distribution is the integral of the frequency
function, it is less sensitive to 'scatter' in the data. 'Smoothing' of
measurements and interpolation between measured points on the distribution
are therefore simple and reliable. For these reasons, it is common practice to .
work with the cumulative distribution function rather than the frequency
function.

1.1.3 Averages
To define a particle size distribution completely requires a large amount of
information, such as Table 1.1, for example. It is obviously advantageous to
be able to approximate the distribution by some form of mathematical
function, and many such functions are available. Each requires at least two
calculated parameters: one to define the location of the distribution and one
to define its width. The first of these parameters is usually some form of
'average', such as the mean (strictly arithmetic mean), geometric mean,
median or mode.
The arithmetic mean, dp , is given by the sum of all the particle diameters,
divided by the total number:

_ '2:.d '2:.n;d;
dp = - = - - (1.5)
N '2:.n;
= 1000
dJ( dp)dd p (1.6)
The geometric mean, d g , is defined as
dg = [dnldn2dn3
2
I 3 ...
dni]'/N
I (1.7)

The importance of the geometric mean will become clear in later sections.
The median is the diameter for which one half of the total number of
particles are larger, and one half smaller. It divides the frequency distribution
into equal areas and is the diameter which corresponds to F = 0.5 on the
cumulative distribution curve. The advantage of the median is that it is less
affected by skewness (lack of symmetry) of the distribution than the mean.
The mode is the most frequent size, i.e. the highest point on the frequency
curve. For monodisperse distributions all four of these averages are equal.
For distributions which are skewed towards larger sizes as in Figure 1.4,
which is frequently the case for particles,
·mode < median < mean.
Very often the particle sizing device which is used does not provide a direct
measurement of particle diameter but measures some property which is
 100
-.
~
'-' 80
~

~
= 60
==~
~
~ 40
••
~
] 20
U=

o
o 20 40 60 80 100 120 140 160 180 200
Particle diameter (J.lm)
Figure 1.5 Cumulative number distribution.
 Particle size distributions 9

related non-linearly to the diameter, such as mass or surface area. For

example, the average mass, liz, is given by

Lm Jtd~
Iiz=--=--Q (1.8)
N 6 p

Therefore, for particles whose shape is independent of size (see equation

1.13)

dm -_[6 ---Lm
QpJtN
]
'/3 [Ld
--
N
3
JY3 -_ [Ln;d~
--
N
] '/3 (1.9)

where Qp is the particle density, m is a single particle mass and d m is the

diameter of the sphere of average mass, also known as the third moment
average. Similarly, the diameter of the sphere of average surface, or second
moment average, is given by

ds -
_ [Ln;df
--
JY2 (1.10)
N
The first moment average is simply the arithmetic mean as defined in equation
(1.5).

1.1.4 Weighted distributions

So far we have considered only number or count distributions. Frequently we
are more interested in some weighted distribution, such as the mass or surface
distribution. For example, the commonly used method for obtaining the size
distribution of coarse particles is sieving, in which the test sample is placed on
the top of a stack or nest of sieves, with mesh sizes decreasing with height in
the stack, and the apparatus is shaken. Each sieve is then weighed, so that the
result is a distribution of the mass of particles with diameters between each
sieve size. While the number distribution gives the fraction of the total
number of particles in any size range, the mass distribution gives the fraction
of the total mass contributed by particles in any size range. It is important to
realise that the graphical representations and the values of the averages for
these two commonly used distributions are different. A product which is 90%
by number within the required size range may contain only 10% by weight of
saleable material!
The number mean diameter was given earlier as

(1.11)
7a Particle characterisation

Similarly, the mass mean diameter, d mm , is given by

(1.12)

where m; is the mass of particles in the ith interval and M is the total mass.
The ratio (m;lM) is a form of weighting factor in the averaging process. If
particle shape is not a function of particle size for the distribution, i.e. we can
write
(1.13)
where k is a constant for all values of d;, then

"Lm·d· "Ln;d'f
d =_'_' __ _ (1.14)
mm M

(The mass mean diameter should not be confused with the diameter of average
mass.) In a similar way, the surface mean diameter (also known as the
'volume-surface mean' or the 'Sauter mean') is given by

"Ls;d; "Ln;dr
(1.15)
d sm = S = "Ln;dr

where s; is the surface area of particles in the ith interval and S is the total
surface area. The surface mean diameter is the appropriate diameter to use
when calculating the pressure drop through a packed bed of particles at low
Reynolds numbers (see Section 2.3.2).
It is also possible to obtain moment averages of weighted distributions,
analogous to the moment averages of the count distribution defined in
equations (1.9) and (1.10).

Example 1.1 Averages of distributions

Problem
For the data of Table 1.1, obtain:
(a) the count mean diameter
(b) the count median diameter (CMD)
(c) the count mode
(d) the diameter of average mass
(e) the mass mean diameter
(f) the surface mean diameter

Solution
(Refer to Table 1.2)
Table 1.2 Solution to Example 1.1

Diameter range (~m)

Mean size, Count nd nd2 nd3 nd4
Lower Upper d (~m) frequency (~m) (~m)2 (~m)3 (~m)4

0 5 2.5 39 97.5 243.8 609.4 1523.4

5 10 7.5 175 1312.5 9843.8 7.38E+04 5.54E+05
10 20 15 348 5220.0 7.83E+04 1.17E+06 1.76E+07
20 30 25 187 4675.0 1.17E+05 2.92E+06 7.30E+07
30 40 35 112 3920.0 1.37E+05 4.80E+06 1. 68E+08
40 60 50 89 4450.0 2.23E+05 1.11E+07 5.56E+08
60 80 70 27 1890.0 1.32E+05 9.26E+06 6.48E+08
80 100 90 13 1170.0 1.05E+05 9.48E+06 8.53E+08
100 150 125 8 1000.0 1.25E+05 1.56E+07 1.95E+09
150 200 175 2 350.0 6.13E+04 1.07E+07 1.88E+09
1000 24085.0 9.89E+05 6.52E+07 6.15E+09
12 Particle characterisation

_ 'Lnidi 24 085
(a) d p = N = 1000 = 24 f,lm

(b) CMD = 18 f,lm

It is easiest to obtain this directly from the cumulative distribution curve
(Figure 1.5).
(c) Count mode = 11 f,lm
It is easiest to obtain this from the frequency distribution curve (Figure
1.4).
'Lnidf
(d) d m = [ -
J'/3
=
[6.52 X 107 J '13
= 40 f,lm
N 1000
'Lnidi 6.15 X 109
(e)d =-= =94f,lm
mm 'Lnid~ 6.52 X 107
'Lnidf 6.52 X 107
(f) d sm = - = = 66 f,lm
'Lnidr 9.89 X 105
Figure 1.6 shows these averages marked on the frequency distribution curve
of Figure 1.4.

Example 1.2 Conversion between differently weighted distributions

Problem
Convert the count distribution in Table 1.1 into a cumulative mass
distribution and plot the result on the same basis as Figure 1.5.

Solution
The fraction of the total number of particles in the ith interval is fin- The
weight of these particles (assuming spheres) is thus
ndr
6-
Nfin X - Qp

where di is the mean diameter of the interval.

The cumulative mass distribution function is then given by:

7,Nfin Cd')
-i i
-3
Qp 'L findi
1
Fim= ---- (1.16)

7Nfin
I (
-i
nd 3
)
Qp
I
-3
'L findi
1

where I is the total number of intervals. In fact it is not essential for the
 0.050
Count Mode
0.045
0.040 Count Median Diameter

0.035
e Count Mean Diameter
%0.030
.9
.... 0.025
~
E 0.020 Diameter of Average Mass
~
0.015
Surface Mean Diameter
0.010 Mass Mean Diameter
0.005
0.000
0 20 40 60
I 80 100 120 140 160 180 200
Particle diameter (J.lm)
Figure 1.6 Averages of the distribution.
14 Particle characterisation

particles to be spherical, but only necessary for their shape to be independent

of size,· i.e. for equation (1.13) to hold. Similarly, to convert a mass
distribution, represented by Jim' to a cumulative number distribution, Fin:

i
~
(f~:) I
Fin = (1.17)

(f~~)
I
~

Taking the data of Table 1.1 we now retabulate it as shown in Table 1.3.
The resulting cumulative mass distribution is plotted in Figure 1.7.
In converting from a number distribution to a mass distribution, or vice
versa, it is important to consider the accuracy of the extremes of the
distributions. A 10 i-!m particle has 1000 times the mass of a 1 i-!m particle.
Thus a small error in the upper size classes of a number distribution will have
a significant effect on the resulting mass distribution. Since the number of
particles in any upper size class is usually small it is essential to repeat the
count several times to avoid the usual problems of small-number statistics.
Conversely, when converting from a mass distribution to a number
distribution, small errors in the lower size classes of the mass distribution are
magnified by the conversion, producing large errors in the number distribu-
tion.

Table 1.3 Solution to Example 1.2

Diameter range (fAm) Cumulative

Interval Mean size Count fin dr mass below
no. Lower Upper (fAm) frequency upper bound
di di +! di fin (%) Fim

0 5 2.5 39 6.09E+02 0.00

2 5 10 7.5 175 7.38E+04 0.11
3 10 20 15 348 1.17E+06 1.92
4 20 30 25 187 2.92E+06 6.40
5 30 40 35 112 4.80E+06 13.77
6 40 60 50 89 l.l1E+07 30.83
7 60 80 70 27 9.26E+06 45.04
8 80 100 90 13 9.48E+06 59.58
9 100 150 125 8 1.56E+07 83.55
10 150 200 175 2 1.07E+07 100.00
1000 6.52E+07
 1.0

.9== 0.8
...
~
eu
==~ 0.6
eu
e
aI
~ 0.4
eu
-e=
= 0.2
U

0.0 I• .........-r
o 20 40 60 80 100 120 140 160 180 200
Particle diameter (f..lm)
Figure 1.7 Cumulative mass distribution.
76 Particle characterisation

Example 1.3 An instrument to measure surface mean diameter

Problem
Boeck and Leschonski (1984) have developed an instrument in which a
flowing aerosol passes simultaneously through a laser beam and an X-ray
beam. The attenuation (reduction in transmitted energy) of each beam is
measured. The attenuation of the laser is related to the total surface area, S,
for the aerosol in the path of the beam (this is discussed in section 1.2.4.3) and
the attenuation of the X-ray is a measure of the total mass, M, which is
present. Show how the surface mean diameter can be derived.

Solution
Assuming that the particles are spheres,
S = Jt"Lnid7"
and
QpJt
M= - - "LnidT
6
Substituting in equation (1.15)

d sm = "Lnid~ = [~J M (1.18)

"Lnidi Qp S

If the particles are not spheres, the constant (6/Qp) will change. This example
illustrates the ease with which it is sometimes possible to obtain weighted
mean diameters from macroscopic properties.

1.1.5 Model distribution functions

It is obviously useful to represent particle size distributions by some form of
mathematical model distribution and many such models have been used. The
commonest is the log-normal distribution, the derivation and properties of
which are discussed in the Appendix to this chapter. It should be emphasised
that this and other models have no theoretical justification for application to
particle size distributions; their use is empirical only.
The log-normal distribution has a number of convenient properties, one of
which is that it gives a straight line on special co-ordinates, using a logarithmic
scale for diameter and a probability scale for cumulative fraction. Graph
paper with these co-ordinates is routinely available as log-probability paper.
Figure 1.8 shows the data of Table 1.1 plotted on such co-ordinates. It can be
seen that this representation compresses the central portion of the data and
expands the extremes. The graph shows a cumulative count distribution, so
that the count median diameter (CMD) can be read directly from it. In the
Appendix it is shown that the log-normal distribution is completely defined by
the CMD, which defines the location of the distribution, and the geometric
 1000 ~I-----------------------------------------'

100
-.
e::::s..
'-' CMD
c..
"0

101 ~ I
50%

1
0.01 0.1 1 10 30 50 70 90 99 99.9
Percent less than indicated size
Figure 1.8 Log-normal plot-number distribution.
18 Particle characterisation

standard deviation (GSD), which defines its spread. For a log-normal

distribution
In Og = In d s4 % - In d so % (1.19)
Therefore

Og =--=--=
d so % d 16 %
[ Jd s4 %
d 16 %
Y,
(1.20)

So Og may also be obtained very simply from the graph. Note that unlike the
standard deviation of a normal distribution, Og has no units and must always
be greater than 1. A further important property of the log-normal
representation is that weighted distributions are parallel on a log-probability
plot. This means that the geometric standard deviation is the same for the
number and weighted distributions. This is shown for the data of Table 1.1 in
Figure 1.9. Transformation between different weighted distributions and
between the various averages discussed earlier in this chapter is made possible
by means of the Hatch-Choate equations (see Appendix AI), the transforma-
tion depending only on the value of GSD. Note however that mixtures (i.e.
sums) of log-normal distributions are not log-normal.
It is often the case that particle size distributions are only approximately
log-normal; even in these circumstances it may still be helpful to plot them on
log-probability paper, but it must be remembered that the special properties
described above apply only to true log-normal distributions.

1.2 PARTICLE SIZE MEASUREMENT

1.2.1 Particle size
The terms particle size and diameter are unambiguous only for spheres. For
non-spherical particles some form of equivalent diameter is used. It is
important to realise that the commercial instruments available for particle
sizing measure different equivalent diameters, so that one would expect them
to give different answers when measuring the same sample. With some
knowledge of the particle shape, however, it is often possible to convert from
one equivalent diameter to another. The following definitions are in common
use.
(i) equivalent-volume sphere diameter, d y
This is the diameter of the sphere with the same volume, v, as the
particle

dy
_ (6V)Y,
- - (1.21)
Jt
The value of d y is not dependent on the orientation of the particle.
 1000 - - . - - - - - - - - - - - - - - - - - - - - - - - - - - - - ,

I
100
MMD •
,.-.,

~
'-'
Q.
I CMD
"0
lO
V _/ I
50%

1
0.01 0.1 1 10 30 50 70 90 99 99.9
Percent less than indicated size
Figure 1.9 Log-normal plot-number and mass distributions.
20 Particle characterisation

(ii) equivalent-surface sphere diameter, ds

This is the diameter of a sphere with same surface area, S, as the particle

ds _(S)'h
- -
Jt
(1.22)

Again, ds is independent of orientation.

(iii) equivalent-projected-area circle diameter, d A

This is the diameter of a circle with the same area, A, as the projected
area of the particle.

dA
_ (4A)'/2
- - (1.23)
Jt
The value of dA is dependent on orientation. However, it can be
shown (Vouk, 1948) that, averaged over convex particles in random
orientation,
S )Y2
dA = ds = ( --; (1.24)

where S is the total surface area of the particle.

(iv) Stokes' diameter, d st

This is the diameter of a sphere with the same settling velocity as the
particle (in the Stokes regime; see Chapter 2). This definition is rather
different from the earlier ones in that the particle size is defined in terms
of its aerodynamic behaviour rather than its geometrical properties. It is
shown in Chapter 2 (equation 2.27) that
d st = [18 !lVtf(Qp - Q)g(' (1.25)
where V t is the particle terminal velocity. For small particles (dp :::;: 1 !lm)
in gases it is necessary to account for non-continuum effects in the fluid
by introducing Cst, the Cunningham slip-correction factor (section
2.1.4)
(1.26)
It is also shown in Chapter 2 that the Stokes diameter and the equivalent-
volume sphere diameter are simply related,
_ [3Jtd~ ] '/,
d st - -- (1.27)
c
where c is the hydrodynamic resistance of the particle (3Jtd for a
sphere). c can be calculated exactly for various model shapes in preferred
or random orientations. For a non-spherical particle, dst is always greater
 Particle size measurement 21

than d y , but the difference is relatively small for compact particles, such
as cubes or spheroids.

(v) aerodynamic diameter, d a

This is the diameter of a sphere of density 1000 kg/m3 and the same
settling velocity as the particle (in the Stokes regime).
By analogy with equation (1.26),
d a = [18 !!V/Ca(Qo - Q)gt' (1.28)
where Qo = 1000 kg/m3 and Ca is the appropriate Cunningham slip-
correction factor for d a . From (1.26) and (1.28),

da = d st [ Cst (Qp - Q) ] '/2 (1.29)

Ca (Qo - Q)
For a particle in a liquid, Cst = Ca = 1, so that
da = d st [ (Qp - Q) ] Vz
(1.30)
(Qo - Q)
For a particle in a gas, Cst = Ca, Qp > > Q and Qo >> Q so that
da = d st V Qp/Q (1.31)
The choice of equivalent diameter depends on the use to which the data
are put. For example, if the efficiency of an inertial separating device
such as a cyclone (Chapter 7) is required, it is appropriate to use the
Stokes diameter, since this best describes the behaviour of particles
suspended in a fluid when inertial effects are dominant.

1.2.2 Particle shape

There have been numerous attempts to devise a comprehensive system for
definition of particle shape, none of which has been wholly successful. One of
the simplest is that due to Wadell (1932), who defined sphericity, 'ljJ, as
surface area of sphere having same volume as particle
surface area of particle
(1.32)
Clearly, 'ljJ ~ 1.
There is obviously a limit to how well a single parameter can characterise
particle shape. Heywood (1938) recognised that the shape of a particle has
two parts: the form, such as cuboid, spheroidal, etc. and the relative
proportions, which distinguish one shape from another of the same form.
Heywood went on to define surface and volume shape coefficients, as and
22 Particle characterisation

U v which he related to the equivalent projected-area diameter, d A , obtained

from microscope analysis
S = USA ce.. = nd§ (1.33)

v = uvAd~ =-i-
nd3
(1.34)

Here the extra subscript A emphasises that the shape coefficients are related
to the projected-area diameter; it is equally valid to derive expressions for the
shape coefficients which relate them to some other equivalent diameter.
Heywood now considered a particle resting on its plane of maximum
stability (Figure 1.10) and defined its dimensions in three orthogonal
directions: thickness, T (perpendicular to the plane of maximum stability),
length, L, and breadth, B. The shape is now described by
the elongation ratio, n = LIB
and the flakiness ratio, m = BIT.
The surface and volume shape factors can be combined with the shape ratios
as follows. Consider a rectangular parallelepiped of dimensions L X B X T
(Figure 1.10). The proj ected area of the particle,
ndi
A = - = rABL (1.35)
4
where rA is the 'area ratio', ndi/4BL. The volume of the particle is:
v = prABLT (1.36)
which is equal to uvAdl from the definition of UvA in equation (1.34). This
serves to define the 'prismoidal ratio' , p - another measure of particle shape -
as, after some algebraic manipulation, 4uvA d AlnT.
Combining equations (1.35) and (1.36) gives

nVn p
(1.37)
UvA =---
8 mvr;:n
If the particle is 'isometric', i.e. B =L= T and n =m = 1, then
nVn p

8 vr: (1.37a)

The term UvAI can therefore be used as an indication of particle form. When
the particle is not isometric, UvA is given by uvA1lmYn. In other words, the
form is modified by the relative proportions, represented by m and n.
Heywood tabulated values of rA and p for four general forms, as shown in
 Particle size measurement 23

I
I
T I
Minimum
Dimension

Plane of maximum stability

~I"'--B --~~I

T T
.. . ..

.
~ . .. ..... . ...
. . . . ..
. .. . ..
........ . .

......•.•........
..
....•... ::.::::.:.: . :> ... ::" . ":-:: ... :.
. ..

~~ .... ~
~~L
Figure 1.10 Heywood's shape factors.
24 Particle characterisation

Table 1.4. One practical application of Heywood's shape classification

approach is given in Example 2.3.
Numerous methods for the characterisation of two-dimensional particle
profiles have been proposed (see, for example, Hawkins, 1990). One of the
simplest is to plot the polar coordinates of the particle profile with the centre
of gravity of the silhouette as the origin (Figure 1.11). The resulting plot may
be approximated by a truncated harmonic series. There is a multiplicity of
approaches to the analysis of such series in order to obtain some set of
coefficients describing particle shape, most of which must be treated with
some caution in order to avoid overcomplicating an already complex
problem.
An interesting method for describing the roughness of a particle profile
relies on the use of fractal geometry, and is entertainingly explained by Kaye
(1989), who is largely responsible for popularising the technique. One might

Table 1.4 Values of 'A and p for particles of various

shapes (Allen, 1996)

Shape 'A P

Angular
tetrahedral 0.5--0.8 0.4--0.53
prismoidal 0.5--0.9 0.53--0.9
Subangular 0.65--0.85 0.55--0.8
Rounded 0.72--0.82 0.62--0.75

R ,- ,-,
'0·- . '- ,_.0. .....
-e..
.... .•..•. •..•.. .-
e
Figure 1.11 A particle outline in polar coordinates (Hawkins, 1990).
 Particle size measurement 25

imagine 'stepping' around a particle profile using a pair of dividers, as shown

in Figure 1.12, in order to measure the length, P, of its perimeter. If the 'step
length', A., is now decreased and the operation repeated, P will, in general,
be larger, as smaller details of the profile can now be detected. In fact, as A.
decreases, P(A.) increases without limit, which is sometimes called the
'coastline of Britain problem', reflecting the geographical origins of the
subject. It is sometimes the case that
P(A.) = n(1 - 6) (1.38)
where 0 is the so-called 'fractal dimension', which takes a value between 1
for a perfectly smooth line and 2 for an infinitely re-entrant (or 'space-filling')
line. A plot of log P versus log A. enables 0 to be obtained. In practice, the
utility of this approach is limited by the fact that for real particle profiles 0 is
not often constant, although distinct changes in the slope of the log Pilog A.
plot can sometimes be used to indicate changes in the nature of the particle
morphology with scale (Kaye, 1989). The real problem with fractal analysis of
particle shapes remains one of interpretation: despite numerous attempts to

(a) (b)
6.0
5.0 8= 1.17
4.0
P
3.0

2.0
~..
0.05 0.10 0.20 0.50
A
6.0
5.0 8=1.37

~ ~
4.0
P
3.0

2.0
0.05 0.10 0.20 0.50
A
6.0
5.0 0 8=1.42

~
4.0
P
3.0

2.0
0.05 0.10 0.20 0.50
A

Figure 1.12 Examples of fractal plots (Kaye, 1989).

26 Particle characterisation

establish them, the links between fractal dimensions and other measures of
particle properties, such as friction factors and flow behaviour, remain
somewhat tenuous.
The effect of particle shape on motion can be accounted for by means of the
dynamic shape jactor, X, which is defined as the ratio of the actual resistance
force acting on a particle, F o , to the resistance force acting on a sphere of the
same volume and velocity (in the Stokes law regime). Hence
x= F o /3JtflVd v (1.39)
Putting
Jtd~
Fo = - (Op - O)g,
6
we obtain the terminal settling velocity,

(Op - O)~g
Vt =------ (1.40)
18 flX

Hence, X = (d)d s )2 (1.41)

Experimental determinations of X are listed in Table 1.5; approaches to the
theoretical determination of X for various model shapes are given in Chapter
2.

Table 1.5 Values of dynamic shape factor

X (after Davies, 1979) (averaged over all
orientations, except where stated)

sphere 1.00
cube 1.08
cylinder (LID = 4)
axis horizontal 1.32
axis vertical 1.07
bituminous coal 1.05-1.11
quartz 1.36
sand 1.57
talc 2.04

1.2.3 Sampling
Only very rarely is it possible to perform a size analysis on the entirety of the
powder of interest. The aim of sampling is to obtain a fraction which is
representative of the whole.
 Particle size measurement 27

Allen (1996) states the 'golden rules' of sampling as

(a) a powder should always be sampled while in motion;
(b) it is better to take the whole of the powder for a short time than some of
the powder for a longer time.
The enemy of good sampling is segregation (see Williams, 1990) which is the
word used to describe the separation of particles due to differences in physical
properties such as shape and density. One of the most common forms of
segregation occurs when particles are tipped onto surfaces or into bins and the
coarser particles tend to roll down the sloping sides of the tip, leaving a
greater concentration of fine particles near the centre. Methods and devices
for reliable sampling in these and other circumstances are given by Allen
(1996). When obtaining a small sample for analysis, it is advisable to use a
device such as a 'riffler' to divide the bulk into many smaller samples, and to
further subdivide these samples by the same technique until a sufficiently
small sample is obtained.
In many of the examples in this book, it is necessary to sample particles
from a flowing suspension, where in practice it is impossible to collect all of
the suspended material. Two general methods of measurement are employed:
those in which the sample is extracted from a duct and analysed outside it,
and those in which some form of measuring instrument is inserted into the
duct and the measurement is performed in situ. In the former case, the sample
is physically defined: it is contained in the fraction of the total flow which is
extracted. In the latter case, the sample may not be physically separated; for
example, if optical methods are used, the sample may be that fraction of the
powder which is illuminated by the beam. The two size distributions
(delivered and in situ) may not be the same (see Chapter 6). Whichever
measurement method is chosen, it is advisable to use multiple measurements
at different points within the duct in order to guard against effects arising
from spatial variations in flow velocity and/or particle loading and size
distribution.
The advantages and disadvantages of extractive and in situ approaches are
compared in Table 1.6. In extractive sampling, a sharp-ended sampling probe
is inserted into the flow (Figure 1.13) and a sample of the suspension is
withdrawn at a known flow rate, usually through a filter or other collection
device. The concentration of particles in the fluid can then be obtained from
the mass of particles collected from a given volume of fluid. As in bulk
sampling, care must be taken to avoid size-selective removal. Sampling must
be isokinetic, i.e. the fluid velocity at the tip of the sampling probe must be
the same as the free-stream velocity just up-stream. If the sampling rate is too
high (Figure 1.14), the sample size distribution will be biased towards smaller
sizes and the measured concentration will be too low. If the sampling rate is
too low, the converse will apply.
The errors resulting from anisokinetic sampling are discussed by Durham
28 Particle characterisation

Table 1.6 Advantages and disadvantages of sampling methods

Advantages Disadvantages

Extractive sampling
Equipment can be relatively simple and Analysis results are not immediately
cheap. available.
A portion of the sample removed can be Extractive equipment may disturb gas
retained. flow, producing an unrepresentative
sample.
Several forms of analysis can be Particulates can agglomerate in
performed and repeated if necessary. sampling and conversely agglomerates
can be dispersed during measurement,
producing an unrepresentative sample
for size analysis.

In situ sampling
Generally, sample extraction equipment Calibration can be difficult.
is not required.
Particles can be monitored as they exist in Sampled material not retained for further
the duct. analysis.
Rapid data return. Sensors may be exposed to 'hostile'
conditions and may need to be cleaned
periodically .

..
::;;;;;ZZ/7ZZZZZZtzZZ7ZtZZZ7Z7ZZlr

... Gas StreamlInes ...

..
U

..
...
~
...
~/Z21//ZZZZZ;:7ZZZZZ/?ZZZZZ2/

----~......------------ ...

Uo=U

Figure 1.13 Isokinetic sampling (after Hinds, 1982).

and Lundgren (1980) and Hinds (1982), among others. Figure 1.15 shows the
fractional error in measured particle concentration (CICo) as a function of the
fractional departure from isokinetic conditions (Uo/U), where the subscript
zero refers to the free-stream conditions and the unsubscripted variable to
conditions at the probe tip. The error in measured concentration is seen to
 Particle size measurement 29

Vo ....
Gas streamlines
entering probe {
(a)

Vo
---$q?zzvvzzzzzzV/z/zz'jlzzvm
...
Gas streamlines { ..
~
----; ..
: -: I1 :
entering probe
~ ---.. ..
~ ~27ZZ7Z7VVVZ7ZZV)jVZZVZ
..
(b) ....
--------~-------~

--
--~~--~.----------- ..
----'u"-'Q'--------:-~?ZZZZZZZZ/zzzzzz;;zzzzzzzzz/
..--- .
{.--11..---------
---------
Gas streamlines
entering probe
. ..
---------1:-- .-_-----
U ..

..
..

(c)
--
_________:_-~ZZ7Z7ZZ7ZZZZZZ/~ZZZZZ7ZZZ7Z

Figure 1.14 Anisokinetic sample. (a) Misalignment, e=

:-
0; (b) U> Uo; (c) U < Uo
(after Hinds, 1982).

depend on the value of a dimensionless group known as the Stokes number,

St, which may be considered as the ratio between the forces due to particle
inertia and viscous drag,
Qpdff,UoCst
St=---- (1.42)
91l D s
where Qp = particle density (kg m-3 ), d p = particle diameter (m), Uo = free
stream velocity (m S-l), Cst = Cunningham slip-correction factor (-), Il =
fluid viscosity (Pa s) and Ds = probe diameter (m). (The Stokes number is
30 Particle characterisation

1.0
0.5

0.2
0.1
~o
r:5
.~ 1.0 Stk = O.D1
___=:::;;;~~~:::..____-- 0.01
c:
o
.z; 0.1
e 0.2
~
"oc: 0.5
U
1.0

0.1 L-_.L---L_--L_...L..-L-L..L.l-L.L-_.L..--L_---'_-'---L...J
0.1 1.0
Velocity ratio, l'olU

Figure 1.15 Effect of velocity ratio on concentration ratio for several values of the
Stokes number (8 = 0°) (after Hinds, 1982).

considered in more detail in section 2.2 and the Cunningham slip-correction

factor in section 2.1.4.) Errors in concentration are small for Stokes' numbers
below 0.1 and negligible for Stokes' numbers below 0.01. For a probe
diameter of 1 cm in ambient air at a free stream velocity of 10 m S-l, these
correspond to particle diameters of about 2.5 !lm and 0.8 !lm, respectively,
for a particle density of 2500 kg m-3 . In general, therefore, it is important to
maintain sampling conditions as close to isokinetic as possible. For further
guidance on sampling from gases, including choice of sampling points, the
reader is referred to the review by Van Santen and Gannon (1987).

1.2.4 Methods of particle size measurement

New methods and instruments for measurement of particle size and
concentration appear every year in response to the increasing demands of
scientists and engineers concerned with the many aspects of particle
technology. The ultimate requirement is for a portable on-line instrument
having a fast response and producing the size distribution of the particle
diameter which is most closely related to the phenomenon under investiga-
tion, for example, the Stokes diameter for the study of inertial collection
 Particle size measurement 31

processes. The intention in this section is to provide a brief account of some of

the more common methods for particle sizing of distributions which are
substantially above 1 !-tm in size (Table 1.7) and to explain which diameter or
equivalent diameter they measure in practice. Because the physical laws
which govern the behaviour of sub micron particles are often different from
those which operate for particles substantially above a micron in size, the
methods used for sizing submicron particles form a distinct group which will
not be considered here.

1.2.4.1 Sieving. Sieving has been used as a method of particle separation

since ancient times: by using a 'nest' or stack of sieves of different sizes it is
possible to sort particles by size and so to obtain a distribution by mass. The
. method has the advantage that it enables the sorted fractions to be retained
and subjected to further analysis. The apparent simplicity of the method is
deceptive: particles are sorted by their two smallest dimensions only, so that if
both size and shape vary the result of a sieve separation may be complex
(Figure 1.16).
In practice, size determinations are normally performed using an automated
sieve shaker, to reduce operator bias. Nevertheless, errors can occur due to
'blinding' or sieve blocking, particle breakage and mesh stretching caused by
overloading.

1.2.4.2 Microscopy. Microscopy is the only method in which particles are

directly observed and measured, and this is its chief attraction. Sampling and
preparation are critical, which makes it time-consuming. Particles may be
deposited onto the viewing surface by filtration (if the surface is a
microporous filter, for example), sedimentation, electrostatic precipitation or
thermal precipitation.
For particles larger than about 0.8 !-tm, optical microscopy is possible,
although there are severe depth-of-focus problems at high magnifications.
For this reason, scanning electron microscopy is often preferred, even for
particles well above this limit.
The diameter obtained is the diameter of the circle of equivalent projected
area, d A . If the sample is carefully deposited (Ghadiri et aI., 1987) it is
possible to obtain the maximum equivalent projected area diameter.
The use of microscopic methods has been greatly stimulated in recent years
by the emergence of cheap image analysis computers which will rapidly and
automatically count and size particles (and determine a great many other
user-defined features). Figure 1.17 shows an example of a possible procedure
for determining particle size from a microscope image using image analysis
software.
New methods such as confocal laser scanning microscopy and image
reconstruction techniques will undoubtedly make three-dimensional particle
imaging routinely possible in the near future.
Table 1.7 Methods for particle size measurement

Method Diameter measured Range Distribution type Comments

Sieving 'sieve diameter' 20 !tm upwards mass slow; cheap, errors due to
blinding

Microscopy can be interfaced to image

analysis computer;
optical dA 0.8-150 !tm number depth of focus problems;
identification possible
electronic dA 0.01-20 !tm number chemical analysis possible

Light interaction
extinction
single particle dA 2-1000 !tm number mainly used in liquids; limited
concentration range
multiparticle dA 2-1000 !tm average only total surface measurement
scattering 'equivalent scattering
single particle diameter' 0.2-20 !tm number fast; limited concentration
range; sensitive to refractive
index and shape
multi particle dA (for larger particles) 4--1000 !tm volume fast; has been used for on-line
measurement and control

Settling dst or da 1-1000 !tm mass slow; can be automated (in

liquid)

Impaction dst or da 0.5-50 !tm mass errors due to bounce,

breakage, re-entrainment;
cheap

Electrozone or Coulter dv 0.3-200 !tm number particles must be dispersed in

electrolyte; shape-
independent
 Particle size measurement 33

j Mass
.1\ distribution

Equivalent
volume diameter

Equivalent
mesh size

Figure 1.16 The effect of shape on various types of distribution (adapted from
Scarlett, 1984).

1.2.4.3 Light scattering. There are three basic types of instrument III
which light scattering is employed (Figure 1.18).

(a) Single particle counters. The suspension to be sampled is drawn

through a small view volume where it is illuminated by light from a laser or
incandescent light source. The forward-scattered light is collected by a lens
or, in some instruments, a mirror, and focused onto a photomultiplier where
it is converted into a voltage pulse. Successive pulses are classified by peak
height and counted to provide information on particle number in a certain
size range and so, over a preselected period, a size distribution by number is
built up. Different instruments differ in sampling rate, view volume and the
angles of the optical arrangement and these choices have a strong influence
on the instrument response, as discussed below.

(b) Fraunhofer scattering instruments. Also known as 'field scattering'

instruments, these convert the angular distribution of the forward-scattered
light intensity from a multiparticle 'field' measurement into a size distribution.
Like the single particle counters, this type can be used to measure particle
sizes in suspension in a liquid or a gas. The suspension is illuminated using a
parallel beam of light, usually from a low-power HeINe laser. The scattered
light is collected and, in the most common embodiment, focused on to a
34 Particle characterisation

(a)

Figure 1.17 Image analysis. (a) Initial image.

position-sensitive detector at the focal point of the lens. The distribution of

scattered light is then converted mathematically into a particle volume
distribution. The instrument gives a fast response and can be used in situ; it
has been used as a sensor for on-line process control.

(c) Extinction meters. This type, which is commonly used in situ to

follow particulate emissions to atmosphere from stacks, for example
measures the attenuation of a beam of light passing through the sample. As
will be seen below, the instrument can only give a measure of total loading or
an average particle diameter, rather than a complete size distribution.
However, the type is rugged, cheap and well-adapted to industrial use.

The theory behind the light scattering technique is as follows. Consider a

particle in a plane formed by an incident light beam and the direction of
 Particle size measurement 35

(b)

Figure 1.17 (b) Digitised image ready for manipulation.

observation (Figure 1.19). The incident light energy may be deflected or

absorbed. Deflection is considered to arise from reflection, refraction and
diffraction on the particle cross-sectional area; the deflection process is
referred to as 'scattering'. The scattered light-intensity, 1(8), may be defined
as the amount of electromagnetic energy which crosses unit area perpendicular
to the flow per unit time (W m-2 ) at angle 8 to the incident beam. The
incident illumination is considered parallel, laterally coherent and unpolarised
and account is taken of the scattered flux regardless of polarisation. The
general theory of light scattering by spheres is that derived by Mie and is a
complete formal solution to Maxwell's equations for the incident light wave,
the wave inside the particle, and the scattered wave, subject to a set of
boundary conditions at the particle surface. For a derivation and explanation
of the theory see Kerker (1969).
The Mie theory is complex and two simplifications are possible. For small
36 Particle characterisation

/'
/'
1
/'

/'

light
Source '-': trap
" "- "" Collecting
Particle
at focus "" aperture

(a) 1

Collection lens Detector

Aerosol

Pinhole Pinhole
apertu re aperture

;~t*lT: ~~
(c) ~L~
Figure 1.18 Instruments for particle sizing by light scattering. (a) forward-scatter
from a single particle; (b) diffraction sizing (reproduced by permission of Malvern
Instruments Ltd.); (c) extinction measurement (Hinds, 1982).

particles (nd/A < 0.3, where d is particle diameter and A is the wavelength
of incident light, i.e. d < 0.05 f!m in visible light) Rayleigh scattering theory
can be used, in which the relative angular distribution of scattered light, /(8),
is the same for all particle sizes, shapes and refractive indices. /(8) is then
proportional to d6 /'A 4. For large particles (nd/A > 3/(m - 1), d > = 2 f!m in
visible light) the forward scattered light intensity (8 < 90°) can be
approximated by Fraunhofer diffraction theory (see later).
 Particle size measurement 37

scattering
plane

incident scattered light 1(9)

light 10

Figure 1.19 Light scattering from a single particle.

At a distance R in the direction 8 from a spherical particle illuminated with

unpolarised light of intensity la, the scattered intensity is given (Hinds, 1982)
by

(1.43)

where i 1 and i2 are the Mie intensity parameters for scattered light polarised in
the perpendicular and parallel directions, i1 and i2 are complicated functions
of refractive index, m, size parameter red/A, and scattering angle 8. Two
general observations are of importance:
(a) As the scattering angle, 8, increases towards 90°, the scattered light
intensity becomes more sensitive to changes in refractive index, m,
(particularly whether or not the particle is absorbing) and to departures
from sphericity (Hodkinson, 1966).
(b) As the size parameter, a = red/A, increases, the angular scattering
pattern becomes more complicated. A plot of Mie intensity parameters
(i1 + i 2 ) against a will show variations in intensity of several orders of
magnitude over small changes in a for a > 10 (Figure 1.20) (Hinds,
1982).
For these two reasons most commercial single particle counters integrate the
scattered light over the forward direction (8 < 90°), ensuring the least
dependence of response on particle refractive index and shape. Integration
over a number of incident wavelengths (i.e. by using an incandescent white
light source) also serves to damp out oscillations in the Mie response curves.
Cooke and Kerker (1975) obtained the theoretical responses of a number of
counters by integrating the Mie intensities over the collecting aperture and
showed that, for those integrating the response in the forward direction, there
is comparatively little dependence on non-absorbing (real) particle refractive
index for d > 1 !lm. For absorbing (complex) refractive indices, however, the
38 Particle characterisation

1000

:"J<
+
'-e 100
OJ
E
~c:.
::-
'in
c
OJ

.~
OJ 10
~
.s'"
~r,;
OJ
>
.;::;
'"
0;
a: 1.0

0.1 L...I.._'--_'--..L-L.....l.._-'-_-.L...-'-....L-J
1.0 10 100
Size parameter, C(

0.17 1.7 17
Particle diameter for).. = 0.52 I'm (I'm)

Figure 1.20 Relative scattering (Mie intensity parameter) versus size parameter, for
water droplets (m = 1.33) at scattering angles of 30 and 90° (Hinds, 1982).

responses are very different, being in general much lower than for a real
refractive index of the same (real) magnitude. Instruments with less
convergent illumination (larger y; see Figure l.1Sa) and smaller collection
angles (smaller j3 - b; see Figure l.1Sa) show a multivalued response curve,
the principal uncertainty being in the range between indicated particle
diameters of 0.5 and 1 fLm.
For aerosols of industrial interest, which are in general both irregular and
absorbing (and also frequently of inhomogeneous refractive index), theoretical
light scattering predictions are too complicated to be used to predict the
 Particle size measurement 39

response of a particular single-particle counter, especially when the vagaries

of a particular instrument (such as non-linearity of the photomultiplier) are
taken into account. Even for spherical aerosols of known refractive index,
theoretical response prediction is tedious, so it is usual practice to calibrate
each instrument in the factory using mono-sized polystyrene latex spheres
(refractive index = 1.59) and to incorporate a calibration device, usually a
'light chopper', into the instrument. If the instrument is to be used on an
aerosol which is non-spherical and/or of a different refractive index, it is
essential for it to be recalibrated for the aerosol of interest. However, it is
useful to be able to predict qualitatively what the response of the counter will
be to non-spherical absorbing particles, in order to relate the 'light-scattering
diameter' to other particle diameters of interest (see section 1.2.1).
Predictions of scattering intensities for non-spherical particles are generally
inexact and incomplete. In the range 0.3 < a < 20, particles with no great
inequality between their different dimensions have a scattering pattern
similar to that of spheres of equal volume, although the upper size limit for
which this generalisation can be made depends on the nature of the particle.
For a randomly-oriented non-spherical particle which is large compared
with the wavelength of the incident light, the forward diffraction lobe of the
scattering pattern depends largely on projected area and is not significantly
affected by particle shape (Hodkinson, 1966). Outside the forward diffraction
lobe, the main deviation in the scattering behaviour of a large irregular
particle arises from internally transmitted and refracted light, whose
behaviour is very sensitive to shape. The effect is to increase the scattering
intensity at large scattering angles well above that for a sphere of equivalent
projected area (Pollack and Cuzzi, 1979; Zerull et al., 1979). The scattering
pattern for large opaque particles should deviate less from the spherical case
since in this case there is scattering by diffraction and external reflection only
(Hodkinson, 1966). Thus the response of a single particle counter is sensitive
to both refractive index and shape of the particles sampled. For example,
changing the refractive index from 1.54 (non-absorbing) to 1.54 - 0.5i
(strongly absorbing) causes one commercial instrument to undersize by a
factor of 3 to 4 (Cooke and Kerker, 1975). In addition to this effect, a
correction factor for particle shape must be applied, which will be near unity
for small particles but larger than one for larger particles. Larger absorbing
particles are therefore undersized less than equivalent spheres of the same
refractive index.
It should be noted that single particle counters as a group are also subject to
count losses due to the simultaneous presence of more than one particle in the
sensing volume ('coincidence'), the inability of the electronics to process
closely spaced pulses, and 'spreading' of the observed distribution whereby all
particles of the same size are not counted in the same channel. For an
examination of these problems see Pisani and Thompson (1971).
As noted above, several instruments have been developed which convert
40 Particle characterisation

the angular distribution of forward scattered light intensity from a multiparticle

field measurement into a size distribution. Their operation is reviewed by
Bayvel and Jones (1981).
In theory it is possible to obtain the particle volume distribution for
spherical particles directly from the measured light energy distribution. In
practice the system of equations which must be solved is unstable and leads to
oscillating size distribution functions for small errors in measured light
energy. Alternative methods are therefore employed: either a model for the
particle size distribution is assumed and fitted in some 'least squares' sense,
or a set of weight bands is set up and an iterative heuristic technique used to
obtain a satisfactory fit (the 'model independent' method). Use of inter-
changeable collection lenses enables a theoretical size range of 1.2 to 1800 !lm
to be covered. In order to reduce multiple scattering effects, it is essential not
to have too many particles in the illuminated zone; in practice this condition is
easily met, even for fuel sprays. There are two theoretical difficulties with the
technique: the range of applicability of the Fraunhofer diffraction assumption,
and the so-called 'anomalous diffraction'. Fraunhofer diffraction is a good
approximation to Mie theory for particles which are much larger than the
wavelength. Jones (1977) has investigated the error in particle sizing
associated with the Fraunhofer assumption and concludes that for an error
on particle diameter of less than 20%, a must be greater than 20 if ml (real)
> 1.3 (for non-absorbing particles) and a must be greater than 8 for ml (real)
> 1.5 (for strongly absorbing particles). For HeiNe laser light
(A. = 0.633 !lm) these limits correspond to 4.0 !lm and 1.6 !lm, respectively;
the limits are increased as the relative refractive index becomes closer to
unity, for example in a liquid. Anomalous diffraction occurs for comparatively
large transparent particles when the refractive index is close to 1. It is due to
interference between relatively undeflected refracted rays and the diffracted
component. The more recent software includes some compensation for this
effect. Theoretically, Fraunhofer diffraction instruments produce a volume
distribution derived from the distribution of spheres of equivalent diffraction
pattern. For many practical applications this will approximate to the volume
distribution of projected area diameter, in either random or preferred
orientation, depending on the hydrodynamic conditions.
In summary:
(a) Single particle light-scattering counters are very sensitive to shape and
refractive index effects and must be precalibrated using the aerosol of
interest by two independent techniques. Instruments based on Fraunhofer
diffraction are rather less sensitive to refractive index and shape effects,
except at the lower extent of their range where the assumptions upon
which the instrument design is based become invalid.
(b) The single particle light-scattering counters, since they count individual
particles, measure the number distribution directly and unambiguously
 Particle size measurement 41

(apart from the 'spreading' effect) providing that their upper concentration
limits are not exceeded. This means that, for accurate filtration
measurements, which require comparison of inlet and outlet concentra-
tion and size distribution, they are preferable to diffraction instruments
which only measure volume distribution. However, single particle
counters require special dilution systems for measuring particles at high
concentrations. Single particle counters are effectively subject to no
lower concentration limits, whereas diffraction instruments are subject to
signal-to-noise ratio problems at low concentrations.
(c) Diffraction instruments do not physically sample the suspension and
therefore cause no disturbance in the measured flow and their operation
is relatively insensitive to duct temperature and pressure. For these
reasons, they are preferable to the single particle counters as on-line
instruments. The requirement that the aerosol to be measured must pass
within the focal length of the collection lens is, however, a restriction.
In contrast to the devices which measure the intensity of scattered light, as
described above, extinction instruments measure the intensity of the light
which is not scattered out of the beam, i.e. the attenuation of the incident
light. According to the Lambert-Beer law (Hinds, 1982) the transmitted light
intensity is given by
(1.44)

where 10 is the intensity of incident radiation (W m-2 ), L is the path length

(m) and De is the 'extinction coefficient' (m- I ). The total extinction arises
from the sum of all the effects of individual particles so that

Ad2
De = L N;--' (Qe); (1.45)
, 4

where N; is the number of particles per unit volume in size class i (m-3 ), d; is
the mean diameter of particles in class i (m) (which is to be interpreted as the
projected area diameter, dA ) and Qe is the 'particle extinction efficiency' for
particles in size class i, which equals the radiant power scattered and absorbed
by a particle divided by the radiant power geometrically incident on the
particle.
Qe is thus a measure of the ability of a particle to remove light from the
beam by scattering and absorption. In general, it is a complicated function of
both refractive index, m, and size parameter, a (Figure 1.21), but for a
greater than about 12 (particles in air of diameter greater than about 2 !-lm)
Qe approaches a constant value of 2. If this is the case
Qe
De = -4 L; N(JuE)
I I
= S/2 (1.46)
42 Particle characterisation

6,----,-----,----,----,-----,----,-----,----,----,----,

5
~
>-
"~ 4
'u
:;:
;:;
c
.'2 3
U
c
.;::;
x
w
2

Size parameter, 0:

o 2 3
Particle diameter for A = 0.52 11m (11m)

Figure 1.21 Extinction efficiency versus particle size for spheres (Hodkinson, 1966).

where S is the total surface area of particles per m 3 of suspension. Hence,

combining this with equation (1.44)
2
S = --In(1o11) (1.47)
L
Thus, if the particles are sufficiently coarse, an average measure of the
suspension properties may be simply obtained. It is clear, however, that an
extinction measurement alone cannot provide more than some indication of
either particle concentration or particle size. If both can vary simultaneously,
this approach is not sufficient and must be used in conjunction with some
other measuring method (see Example 1.3).

7.2.4.4 Electrozone. The 'electrical sensing zone' or 'Electrozone' method

was invented by Wallace Coulter in 1950, and is in world-wide use,
particularly in medical applications where it is used for counting blood cells.
The sample is dispersed in an electrolyte and forced to flow through a small
(10-400 ~m diameter) orifice, having an immersed electrode on each side
(Figure 1.22). In a sufficiently dilute suspension, particles pass through the
orifice one at a time. As each particle passes through, it changes the electrical
resistance across the orifice. Since the current is kept constant, this change of
 OSCILLOSCOPE
r------+l-I ~ PULSE DISPLAY

VOLUMETRIC I "' ..
METERING ",:::,:",." ,.""",«
DEVICE i "'.

OUTPUT

STIRRED SUSPENSION
OF PARTICLES IN
ELECTROLYTE SOLUTION

COUNTER START I STOP

Figure 1.22 'Coulter' counter (reproduced by permission of Coulter Electronics Ltd.).

44 Particle characterisation

resistance generates a voltage pulse whose amplitude is related to the particle

volume (see below). The instrument therefore yields a measurement of the
equivalent volume sphere diameter, which is independent of particle
orientation. The voltage pulses resulting from the passage through the orifice
of a known volume of suspension are electronically scaled and counted, from
which a particle size distribution can be derived. Appropriate aqueous or
organic electrolytes can be found for most particle types. The instrument
must be calibrated with a dispersion of known particle size, but this does not
need to be of the same material as the particles of interest; standard
dispersions of polymer latex are usually used. The theory of the method has
been the subject of much work; a simple approach is given by Deblois and
Bean (1970). According to Maxwell, the effective electrical resistivity of
dilute suspension of insulating spheres, Qefh can be expressed as

Qeff =Q( 1 + ~ f + .. .) (1.48)

where f is the volume fraction occupied by the spheres and Q is the resistivity
of the carrier fluid. For a cylindrical sensing zone of length L and diameter D
filled with the fluid the resistance is
4QL
R1=-- (1.49)
nD2
When a sphere of diameter d is introduced, the value of f becomes

Vsphere 2d3
f= =- (1.50)
Vcylinder 3D2L
Hence the new resistance R2 is now

R 2 = 4QL
nD2
[1 + ~+
D2L
... J (1.51 )

Subtracting the original resistance, R 1 , gives

4Qd3
!1R = R2 - Rl = - - (1.52)
nD4
The change of resistance due to the presence of the particle, !1R, is thus
independent of the orifice length, L, and, more importantly, proportional to
the volume of the particle.
The facts that the method is independent of particle shape and that it relies
on single particle counting are considerable advantages. The disadvantage,
compared with some light-scattering techniques, is that careful sample
redispersion is usually required.
 Particle size measurement 45

7.2.4.5 Settling and aerodynamic methods. The settling of particles in a

fluid is an important subject in itself and is considered in detail in section
2.1.3. It is clear that particles of different sizes and densities* will settle at
different rates, and this is the basis for a large class of size-measuring devices,
usually for liquid suspensions, from the simple 'Andreassen pipette' to more
sophisticated devices in which the settling suspension is probed using laser
light or X-rays (Allen, 1996). The most significant problems associated with
these methods are the necessity for good particle dispersion, anomalous
hindered settling behaviour and the long time taken to obtain a distribution.
(In some devices centrifugal force is used to accelerate the settling process.)
For particles in suspension in a flowing gas, a number of aerodynamic
methods may be used for sizing. The most basic of these is the inertial
impactor (Figure 1.23). As a jet of gas emerges from the impactor nozzle the
streamlines bend sharply. Larger particles deviate from the streamlines and
may impact on the collector plate, while smaller particles follow the
streamlines and are carried out (or on to the next stage, if the plates are
arranged in a 'cascade'). To a large extent, the efficiency of an impactor can
be predicted theoretically; the important parameter is again the Stokes'
number as in equation (1.42):

Qp~UCst
St= - - - (1.53)

Impaction nozzle
or jet

Streamlines

Impaction plate

Figure 1.23 (a) Inertial impactor.

*The appropriate density is relatively easily measured for non-porous particles using a specific
gravity bottle or gas pyknometer. Measurement of the appropriate density for porous particles,
the so-called 'envelope' density, is discussed by Geldart (1990). The important area of powder
porosity and surface area determination, which is outside the scope of this book, is
comprehensively covered by Lowell and Shields (1991).
46 Particle characterisation

100,---------,----------r------~_,--------_,

80

20

o 2 3 4
Aerodynamic diameter (f.lm)

o 0.47 0.94
.J Stk
Figure 1.23 (b) Grade efficiency for inertial impactor.

where U is the average nozzle exit velocity and D j is the diameter of the
nozzle. Single impactors are easily designed and constructed; design
information is given by Marple and Willeke (1976). A good impactor
efficiency curve is shown in Figure 1.23b; this is a 'grade efficiency' curve,
showing the collection efficiency of particles as a function of their size. (A
theoretical approach to the prediction of impactor efficiency is given in
example 2.4 in section 2.2.3.) In practice, impactors rarely achieve as sharp a
'cut-off' as shown here because particles may bounce or break up on impact
or may deposit at various intermediate points between successive stages of a
cascade. The stages can be coated with a thin layer of sticky liquid to prevent
bouncing. A series of suitably sized miniature cyclones (see section 7.2) may
also be used to separate particles by their inertial behaviour.
 Particle size measurement 47

Aerosol In Outer Nozzle

5'1_

Inner Nozzle
1 L/mln
Sheath Air Valve
4 'I.....

Pressure
Transducer

Figure 1.24 Aerodynamic particle sizer (reproduced by permission of Biral Ltd. and
TSI Inc.).

In principle, any device which measures the response of a particle to

suitably rapid changes in gas velocity may be used to determine its
aerodynamic diameter (or Stokes' diameter if its density is known). Such a
device is the 'aerodynamic particle sizer' in which a suspension of particles is
drawn through a fine nozzle, where they are accelerated at a rate which
depends on their aerodynamic size (Figure 1.24). Fine particles accelerate at a
rate almost equal to that of the gas; coarse particles accelerate more slowly.
The particle exit velocity from the nozzle is measured using 'time of flight'
between two closely-spaced laser beams and individual measurements are
used to build up a size distribution. The theory behind this approach is
summarised by Wilson and Liu (1980). Because the instrument counts
individual particles, it is subject to the same coincidence effects as in all other
single particle counters, which means that the gas stream entering the
instrument must often be diluted by as much as 100: 1. In theory, this method
provides an absolute measurement of particle size; in practice, calibration is
necessary using particles of known size. In a further variation of this general
approach to particle sizing, Tate et al. (1986) have proposed a method of
particle sizing in situ by observation of particle velocities as they approach a
spherical obstacle along the stagnation streamline. Particles with a small
Stokes' number reduce their velocity as they approach the obstacle, at a rate
which is close to that of the gas; coarser particles maintain a velocity which is
closer to that in the free stream.
48 Particle characterisation

1.3 NOMENCLATURE
A Area (m 2 )
Cst Cunningham slip-correction factor
CMD Count median diameter (m)
c Hydrodynamic resistance (m) (equation (1.27))
D Dimension, diameter (m)
d Particle diameter (m) (equation (1.23))
dA Equivalent projected area diameter (m)
da Aerodynamic diameter (m)
dg Geometric mean diameter (m) (equation (1.7))
di Representative diameter of particles in interval i (m)
dIll Diameter of the particle of average mass (m) (equation (1.9))
dmm Mass mean diameter (m) (equation (1.12))
dp Arithmetic mean diameter (m) (equation (1.5))
ds Equivalent surface diameter (m) (equation (1.22))
dsm Sauter or surface-volume diameter (m) (equation (1.15))
d st Stokes diameter (m) (equation (1.25))
dv Equivalent volume diameter (m) (equation (1.21))
F Cumulative distribution function (equation (1.3))
FD Drag force (N)
f Frequency distribution function (equations (1.1a and b))
g Acceleration due to gravity (m S-2)
h Height of element of a frequency histogram (equation (1.1a))
I Number of intervals, radiation intensity (W m-2 )
Mie intensity parameter (equation (1.43))
M Total mass (kg)
MMD Mass median diameter (m) (equation (A1.7))
m Heywood flakiness ratio; refractive index (section 1.2.2)
Mass in interval i (kg)
Total number of particles
n Heywood elongation ratio (section 1.2.2)
Number in interval i
Perimeter (m)
Heywood prismoidal ratio (equation (1.36))
Particle extinction coefficient (equation (1.45))
Distance; electrical resistance (ohm)
Heywood area ratio (equation (1.35))
Total surface (m 2 )
Stokes number (equation (1.42))
Amount of surface in interval i
Gas velocity (m S-I)
Particle velocity (m S-I)
Particle terminal velocity (m S-I) (equation (1.25))
Particle volume (m 3 )
 Appendix A 1 49

Greek letters
= nd/A, size parameter in light scattering
Heywood surface and volume coefficients (equations (1.33)
and (1.34»
Dynamic shape factor
Fractal dimension (equation (1.38»
Step length (m) (equation (1.38»
Angle (0)
Viscosity (N s m-2)
Fluid density (kg m-3 )
Particle density (kg m-3 )
Standard deviation
Extinction coefficient (m-l)
Geometric standard deviation
Wadell's sphericity

APPENDIX A1 DERIVATION AND PROPERTIES OF

THE LOG-NORMAL DISTRIBUTION
A commonly used statistical distribution is the normal distribution, for which
the number frequency function, df, is given by

1
df= - - e x p
ov2n
[(d
-
p -

202
dp ?] dd p (ALI)

where df is the fraction of the total number of particles having diameters

between d p and d p + ddp , dp = arithmetic mean diameter and 0 = standard
deviation, where

In practice this is not a very good representation of particle size distributions

because, as we have already observed, most particle size distributions are
skewed to the right as in Figure 1.4, but also because representations of wide
size distributions would require a fraction of the particles to have a negative
size. However, it is an experimental observation that if particle size
distributions are plotted as (fractionl~log dp ) against log dp the resulting plot
frequently appears roughly normal in shape. (The data from Table 1.1 have
been plotted in this way in Figure ALl.) This means that the logarithms of the
diameters are normally distributed and implies that a logarithmic transforma-
tion of the normal distribution will give a better representation of the true
distribution. This also overcomes the earlier problem with negative diameters.
The normal distribution is transformed by replacing dp with the arithmetic
mean of In dp
 0.6

0.5 ""
/
-. "-
'0 0.4
-=
~
Q
..........= 0.3 \
(oJ \
c:
0.2
'=
-- 0.1
""- ..........
--,
a
1 10 100 1000
Particle diameter (I-lm)

Figure A1.1 Frequency distribution curve (logarithmic size scale).

 Appendix A 7 57

Ln;ln d;
lndp = -N- =lndg (A1.2)

where d g is the geometric mean diameter (equation (1.7)). But d g is the

arithmetic mean of the distribution of In dp, which is a symmetrical
distribution, so
mean of In d p = median of In d p = median of d p or
count median diameter (CMD)
o is replaced by the 'standard deviation of the logarithms', or geometric
standard deviation, Og or GSD,
Ln;(1n d; - In d g )2] liz
In 0 = [ (A1.3)
g N-1
Applying these transformations to equation (ALI) we have

1 [(In d p -In CMDf]

df = exp - d(ln d p) (AlA)
v21tln Og 2(ln ogf
df is now the fraction of particles having diameters whose logarithms lie
between In dp and In dp + d(1n d p).
But
ddp
d(In d) = -
p d
p
Therefore,
1 [ (In dp -In CMD)2]
df= exp - ddp (A1.5)
v'2idpln Og 2(ln Og)2
This is the log-normal distribution; particle distributions which are log-normal
give a straight line on log-probability paper.
The important properties of the log-normal distribution have been
summarised in section 1.5:
(i) the distribution is completely defined by the CMD and GSD;
(ii) weighted distributions are parallel on a log-probability plot.
For a normal distribution, 95% of particles lie between (dp - 20) and (dp +
20); for a log-normal distribution, 95% of particles lie between CMD/o~
and CMDo~ (i.e. not a symmetrical interval).
Transformation between the CMD and any other average diameter is made
possible by use of the Hatch-Choate equations (Hinds, 1982). In general, the
p-moment average of the q-weighted distribution, (dqrnL is given by:
p

(dqrn)p = CMD exp [( q +~) In2 Og ] (A1.6)

52 Particle characterisation

For example, to convert from the count median diameter (p = 0, q = 0) to the

mass median diameter (p = 0, q = 3):
In MMD = In CMD + 3ln2 Og (Al.7)

REFERENCES
Allen, T. (1996) Particle Size Measurement, 5th edn., Chapman and Hall, London.
Bayvel, L.P. and Jones, A.R. (1981) Electromagnetic Scattering and its Applications,
Applied Science Publishers, London.
Boeck, Th. and Leschonski, K. (1984) Proceedings of the 3rd European Symposium on
Particle Characterisation (PARTEC), Nilrnberg, 581.
Clift, R., Grace, J.R. and Weber, M.E. (1978) Bubbles, Drops and Particles,
Academic Press, New York.
Cooke, D.D. and Kerker, M. (1975) Appl. Optics 14, 734.
Davies, C.N. (1979) 1. Aerosol. Sci. 10, 477.
DeBlois, R.W. and Bean, C.P. (1970) Rev. Sci. Instr. 41 (7),909.
Durham, M.D. and Lundgren, D.A. (1980) 1. Aerosol. Sci. 11, 179.
Geldart, D. (1990) Powder Technol. 60, 1.
Ghadiri, M., Seville, J.P.K., Raper, J.A. and Clift, R. (1986) 1st World Congress of
Particle Technology, Nilrnberg 1, 203.
Hawkins, A.E. (1990) in Principles of Powder Technology (M.J. Rhodes, ed.), Wiley,
Chichester.
Heywood, H. (1938) Proc. Inst. Mech. Engrs 140, 257.
Hinds, W.C. (1982) Aerosol Technology, Wiley-Interscience, New York.
Hodkinson, J.R. (1966) in Aerosol Science (C.N. Davies, ed.), Academic Press,
London.
Jones, A.R. (1977) 1. Phys. D. 10, 163.
Kaye, B. (1989) A Random Walk Through Fractal Dimensions, VCH, Weinheim.
Kerker, M. (1969) The Scattering of Light and Other Electromagnetic Radiation,
Academic Press, New York.
Lowell, S. and Shields, J.E. (1991) Powder Surface Area and Porosity, 3rd edn.,
Chapman and Hall, London.
Marple, V.A. and Willeke, K. (1976) in Fine Particles (B.Y.H. Liu, ed.), Academic
Press, New York.
Pisani, J.F. and Thompson, G.H. (1971) 1. Phys. E. 4,359.
Pollack, J.B. and Cuzzi, J.N. (1979) in Light Scattering by Irregularly Shaped Particles
(D.W. Schuermann, ed.), Plenum Press, New York.
Scarlett, B. (1984) Proceedings of the 3rd European Symposium on Particle
Characterisation (PARTEC), Nilrnberg, 1.
Tate, A.H.J., Seville, J.P.K., Singh, A. and Clift, R. (1986) I. Chern. E. Symp. Ser.
99,89.
Van Santen, A. and Gannon, E.J. (1987) Filt. Sepn., Sept./Oct. 10, 328.
Vouk, V. (1948) Nature 162, 330.
Wadell, H. (1932) 1. Geol. 40, 443.
Williams, J.C. (1990) in Principles of Powder Technology (M.J. Rhodes, ed.), Wiley,
Chichester.
Wilson, J.C. and Liu, B.Y.H. (1980) 1. Aerosol. Sci. 11, 139.
Zerull, R.H., Giese, R.H., Schwill, S. and Weiss, K. (1979) in Light Scattering by
Irregularly Shaped Particles (D.W. Schuerman, ed.), Plenum Press, New York.
2
•
Particles In fluids

Much of particle technology is concerned with the interactions between

particles and fluids. This chapter is intended to introduce basic concepts
which underlie many of the applications of particle technology, and to
provide some useful results which can be applied immediately, for example to
the calculation of terminal and hindered settling velocities. The topic of
particle/fluid interaction is broad in itself; for a detailed treatment of single
rigid particles, liquid drops and gas bubbles, covering heat and mass transfer
as well as fluid mechanics and momentum transfer, the reader is referred to
Clift et al. (1978).

2.1 SINGLE PARTICLES

2.1.1 Fluid-particle drag
Consider first a single rigid particle held stationary in a fluid flowing steadily
at velocity U, as shown in Figure 2.1. Any element of the surface, such as dS
in Figure 2.1, will in general experience both a shear stress (acting
tangentially to the surface) and a normal stress. If there is no motion of the
fluid relative to the particle, i.e. if U = 0, then the shear stress is zero. It is
convenient to distinguish between two components making up the normal
stress: the stress present if the fluid is stationary and resulting from body
forces acting on the fluid, and the additional normal stress arising from the
fluid motion. Usually the only body force is gravity. The first component of
normal stress is then the hydrostatic pressure at dS. The integral of this stress
over the surface of the particle is the buoyancy force. It acts vertically
upwards, and is evaluated from Archimedes' principle as the weight of fluid
displaced: Qgv where Q is the fluid density, g the gravitational acceleration,
and v the volume of the particle. The second component of normal stress,
known as the 'modified' or 'reduced' pressure, represents the difference
between the actual pressure in the flowing fluid at dS and the hydrostatic
pressure. The integral of the reduced pressure over the surface of the particle
gives the hydrodynamic pressure force acting on the particle. Similarly, the
integral of the shear stress over the surface is the skin friction force. The
components of these forces parallel to the fluid approach velocity U are

J. Seville et al., Processing of Particulate Solids

© Chapman & Hall 1997
54 Particles in fluids

BUOYANCY
LIFT
DRAG

~
~ ___ Shear
~\ Stress
I I
- --
Normal
Stress
Figure 2.1 Forces on a stationary particle in a flowing fluid.

known respectively as form drag and skin friction drag and are the two
components of the hydrodynamic drag, F D , on the particle. Any component
of the total hydrodynamic force normal to U, F L , is known as lift.
The distinction between buoyancy and hydrodynamic force is useful for
several reasons beyond the basic point that only the buoyancy force exists
when the fluid is stationary. The few theoretical solutions available for fluid
motion and drag are obtained by eliminating the hydrostatic pressure so that
the equations describing fluid motion are written in terms of the reduced
pressure (see Batchelor, 1967). The drag force is frictional in the sense that it
corresponds to dissipation of mechanical energy within the fluid (although the
fluid can do no work directly on a rigid stationary particle). Buoyancy, on the
other hand, is not associated with mechanical energy dissipation.
There is no universal theoretical solution available for calculation of F D . It
is therefore necessary to proceed by applying dimensional analysis to
empirical results. In general, if the fluid is incompressible and Newtonian, FD
depends on U and Q, on the fluid viscosity I-l, and on some characteristic
dimension of the particle d, i.e.
FD = f[U, d, Q, I-l) (2.1)
for a particle of fixed geometric shape and orientation. Equation (2.1)
contains five parameters with three basic dimensions (mass, length and time)
so that it can be expressed in general in terms of two dimensionless groups.
These groups can be selected at will, provided they are independent.
 Single particles 55

O.ll=-----j----+---+---+---+---+---t--;.oj

0.1 10 100
Re
Figure 2.2 Drag coefficient as a function of Reynolds number: ~~-, sphere; -'-'-,
cylinder in crossflow; - - - - -, disc with flow parallel to axis.

Conventionally the two groups chosen are the particle drag coefficient,
(2.2)
and the particle Reynolds number
Re = QdU/!J, (2.3)
In equation (2.2), A is the cross-sectional area of the particle, and therefore
proportional to tP. For a spherical particle, A is JttP/4 so that
Co = 8Fo/JtQU2tP (2.2a)
The particle Reynolds number is useful because it indicates the nature of flow
around the particle: a low value of Re indicates that the fluid is in creeping
flow, i.e. its motion is dominated by viscous effects in the fluid. At high Re
the inertia* of the fluid is dominant. Numerical ranges corresponding to 'low'
and 'high' are discussed below.
In dimensionless form, equation (2.1) now becomes
CD = f[Re] (2.4)
This type of relationship is summarised here as Co[Re]. Empirical drag
measurements can therefore be expressed in this form. Figure 2.2 shows the

*Some authors use the term 'turbulent' to describe high Re flows around particles. The term
appears to derive from a mistaken analogy with high-Re internal flows, such as in pipes. Its
application to external flows, such as those around particles, is misleading and is therefore to be
avoided.
56 Particles in fluids

relationship between CD and Re for spheres, for long cylinders in crossflow,

and for flat discs normal to the direction of flow. For a long cylinder of
diameter d, the drag coefficient is most conveniently expressed as
CD = (FD/L)/(dgU2 /2) (2.2b)
where FD/ L is the drag per unit length.

2.1.2 Fluid motion

2.1.2.1 Spheres. The shape of most interest for particle processing is the
sphere. The flow patterns around a spherical particle will therefore be
discussed with reference to the drag curve for the sphere. Figure 2.2 shows
that, at low Re, CD is inversely proportional to Re. This is the only range for
which a full theoretical solution is available. The resulting relationship is
Stokes' law,
CD = 24/Re (2.5)
or
FD = 3rq.,tdU (2.6)
so that, in this range, the drag force is directly proportional to U. Of the total
drag, two-thirds results from the pressure distribution over the sphere and
one-third from shear stresses, i.e. even in this creeping flow range, form drag
on a sphere is twice the skin friction drag.
Equations (2.5) and (2.6) are reasonable approximations for Re < 0.1: the
error in CD is about 2% at Re = 0.1 and nearly 20% at Re = 1. Outside this
range, it is necessary to estimate CD by some empirical result, for example
that due to Turton and Levenspiel (1986) given in Table 2.1. The changes in
drag coefficient shown by Figure 2.2 correspond to changes in the fluid flow
around the sphere. At Re = 20 flow separates from the surface, to form a
closed recirculatory wake which becomes larger as Re increases.
At higher Re the downstream tip of the wake oscillates until, for Re > 100,
fluid is shed from the wake. Throughout this range, the point at which flow
separates from the surface moves forward with increasing Re. For Re > 6000,
the separation point reaches a steady position somewhat forward of the
equator. In this range, skin friction drag is negligible and form drag depends
most strongly on the position of flow separation. Therefore the drag
coefficient varies very little with Re: over the range 750 ~ Re ~ 3.4 X 105 , CD
only varies by ± 13% about a value of 0.445 (or 4/9): this is usually known as
the Newton's law range, because Newton's concepts of fluid dynamic drag
implied a constant drag coefficient.
As Re increases above 6000, changes occur in the flow field which do not
immediately reflect in the drag coefficient. The fluid layers which have
detached from the surface of the sphere become turbulent, and the point at
 Single particles 57

Table 2.1 Correlations for drag and terminal velocity

A. Spheres
(a) Drag

Range of Re Correlation

24 0.413
~ 3.38 X lOs Co = -(1 + 0.173 ReO. 6S7 ) + ---------,---:-:-
Re 1 + 1.63 X 104 Re-1.09
3.38 X lOs to 4 X 105 CD = 29.78 - 5.3 IOg1O Re
4 X lOs to 106 CD = 0.1 IOglO Re - 0.49
~ 106 CD = 0.19 - 8 X 104 Re- l

(b) Free fall or rise

Correlations for dimensionless terminal velocity, Vis, as a function of dimensionless
diameter, d*, after Grace (1986). These are defined in equations 2.12 and 2.14.

Range of Correlation
d*

(a) ~ 3.8 ~ 0.624 ~ 2.37 Vis = (d*)z/18

- 3.1234 X 10-4 (d*)S
+1.6415 X 1O-Q(d*)8
- 7.278 X 10-10 (d*)l1
(b) 3.8-7.58 0.624-1.63 2.37-12.4 X = -1.5446 + 2.9162w
- 1.0432w2
(c) 7.58-227 1.63-28 12.4-6370 x = -1.64758 + 2.94786w
- 1.09703w2 + 0.17129w 3
(d) 226-3500 28-93 6.373 X 103 - X = 5.1837 - 4.51034w
3.26 X lOs + 1.687w2 - 0.189135w3
Note: w = IOg1Od*; x = log1OVis

B. Cylinders E = Lid = aspect ratio

(a) Drag (E » 1): CD = 9.689 Re -D·78
0.1 < Re < 5 CD = CD (1 + 0.147Reo. 82 )
5 < Re < 40 CD = CD (1 + 0.227Reo. sS )
40 < Re < 400 CD = CD (1 + 0.0838Reo. 82 )
(b) Free fall or rise (E ~ 1, d* < 20, Vi < 25, Re < 100)
IOg1O Vi = A o + Alw + Azw2 + A3W3, where
Ao = -0.66140 - 0.461431E - 0.032461E2
Al = 1.38545 + 1.375421E - 0.455261E2
Az = 0.21234 - 1.273551E - 0.575921E z
A3 = -0.03436 + 0.375251E - 0.189691E2
58 Particles in fluids

which they become turbulent approaches the surface of the sphere as Re is

increased. When the layer becomes turbulent very close to the surface, it
reattaches and the position of final separation moves abruptly towards the
rear of the sphere. This 'critical transition' results in a sharp drop in CD, as
shown in Figure 2.2. Estimates given by Clift et ai. (1978) for CD in the range
Re > 3.38 X 105 are given in Table 2.1. This range is not normally
encountered in particle technology, but is relevant to such objects as weather
balloons, large hailstones, projectiles and the balls in some sports.

2.1.2.2 Other shapes. Steady crossflow past a long cylinder shows much
the same features as flow past a sphere, but the transition Reynolds numbers
are different. Flow separation occurs for Re > 5. The closed recirculatory
wake oscillates for Re > 30, and wake shedding occurs for Re > 40. Over the
range 70 < Re < 2500, the wake sheds alternately from the two sides of the
cylinder to form a regular series of vortices alternating in their sense of
rotation; this is often termed a 'Von Karman vortex street'. Above Re = 105 ,
the critical region is entered, with flow transitions similar to those occurring
for a sphere. Table 2.1 gives some expressions for CD(Re) for long cylinders
in crossflow, from Clift et ai. (1978).
The comparison between discs and spheres illustrates some of the effects of
departure from a spherical shape. Because a disc in a fluid flowing parallel to
its axis has no surface parallel to the direction of flow, it can experience only
form drag. At very low Re, drag on a disc is lower than on a sphere because of
the absence of skin friction. Wake formation occurs much sooner for a disc,
around Re = 1, because the sharp edge encourages the flow to separate. Disc
wakes start to oscillate at about Re = 100 but, because the position of flow
separation is fixed, the drag coefficient is constant, at CD = 1.17, for
Re > 133. At high Re, flow patterns continue to be qualitatively similar to
those around a sphere except that, because the separation position is fixed by
the particle shape, there is nothing equivalent to the critical transition. Table
2.1 gives results due to Potter et ai. (1973) and Clift et al. (1978) for estimating
CD for discs fixed normal to the direction of flow.
For other particle shapes, it is usually helpful to compare with particles of
well-characterised shape, such as spheres, cylinders, discs, or spheroids (see
section 2.1.3.2). In the creeping flow range, the drag on a particle is always
less than or equal to that on a body which encloses it, and greater than or
equal to that on any body contained within it (Hill and Power, 1956). Drag in
this range is most influenced by surfaces on which shear and normal stresses can
act: for example, creeping flow parallel to the axis of a long fibre, needle or
extremely prolate spheroid causes half the drag arising in crossflow. At
intermediate and high Re, drag is most likely to be influenced by edges which
define the position of flow separation. Thus drag at high Re is reduced by
'streamlining', i.e. profiling the particle shape to prevent boundary layer
separation. Prolate (i.e. elongated) particles demonstrate the difference
 Single particles 59

between these flow regimes: at low Re, flow parallel to the axis of a prolate
particle causes higher drag than on a sphere of the same volume, whereas at
high Re the drag on the prolate body is lower and can be very much lower if
flow separation is prevented. The effects of shape on free fall or rise of
particles are considered below.

2.1.3 Free fall or rise

2.7.3.7 Spheres. Figure 2.3 shows a spherical particle of density Qp

moving freely under the action of gravity through a fluid of density Q and
viscosity f.!. When the particle is moving at steady velocity, usually known as
the terminal velocity, V t , the fluid drag exactly balances the immersed weight
of the particle, i.e. its weight minus the Archimedean buoyancy. Therefore,
for a particle with Qp > Q so that it falls through the fluid, the terminal
velocity is defined by
rtd3
FD=-(Q - Q)g (2.7)
6 p

where rtd3 /6 is the volume of the sphere, v. The drag coefficient then follows
from equations (2.2a) and (2.7) as
(2.8)
where Ret = QdVt/f.! is the particle Reynolds number at the terminal velocity.
For the creeping flow range, Ret < 0.1, in which Stokes' law applies, the
terminal velocity follows from equations (2.6) and (2.7) as
(2.9)

Buoyancy, pgv,
+ drag, FD

Weight, ppgv

Figure 2.3 Spherical particle falling at its terminal velocity.

60 Particles in fluids

In the Newton's law range (750 :;::; Re :;::; 3.4 X 105 ) where Co is close to 0.445,
the terminal velocity follows as
(2.10)
The interest in particle technology is usually in the intermediate range of Ret,
between creeping flow and Newton's law, say 0.1 < Ret < 750. There is no
explicit result for V t in this range. It is possible to calculate V t from equation
(2.8) using a correlation for Co[Re] such as that in Table 2.1; however, this is
inconvenient, because the equation is implicit and therefore an iterative
solution is needed. It is therefore preferable to express the dimensionless drag
relationship in terms of more convenient groups, remembering that the use of
Co and Re results solely from historical convention and that any two
independent dimensionless groups can be used.
From equation (2.8), a particle at its terminal velocity satisfies the
relationship
(2.11)
The group CoRe;, and various multiples of it, is known as the 'Archimedes
number', or the 'Best number'; it is a convenient dimensionless group, and an
independent variable for estimation of terminal velocities because it involves
the relevant particle and fluid properties but not the terminal velocity. Here,
rather than using the group in equation (2.11), we shall use the dimensionless
particle diameter which follows from it,

Similarly, we note that equation (2.7) or (2.8) can be rearranged to

RetCo = 3Q2V~/4(Qp - Q)gft (2.13)
This group, and multiples of it, is sometimes called the 'Galileo number'.
Equation (2.13) suggests that, as the dependent variable, it is convenient to
use the dimensionless terminal velocity defined by

Retfd* (2.14)

The dimensionless drag correlation can then be expressed in the form Vi[ d*];
the resulting relationship is shown in Figure 2.4. Table 2.1 gives useful
expressions for Vi[ d*], developed by Grace (1986) from correlations for
Ret[Co Re;] given by Clift et at. (1978).
 Single particles 61

10-2 L--...L...LJLLliJIlol.--L...Ll...L1JJlL---1---L.LLI.lill_L....l....L.L.llllL-....L-L..LJ...J...LW
0.1 10 10 2 103 104

Figure 2.4 Dimensionless correlation for terminal velocities of spheres.

Figure 2.4 also shows the simplifications for the creeping flow and
Newton's law ranges. In dimensionless form, with the ranges over which they
estimate V t within about 10%, they are:
Stokes'law
(d* < 2.5, Vi < 0.32, Ret < 0.8):
Vi = (d*?/18 (2.9a)
Newton's law
(60 < d* < 3500, 13 < Vi < 100, 750 < Ret < 3.4 X 105 ):
Vi = 1.73 v'd* (2.l0a)
Equation (2.9) shows that, in the creeping flow range, the terminal velocity
depends strongly on both particle size and fluid viscosity. For both liquids and
gases, viscosity is very sensitive to temperature (but is essentially independent
62 Particles in fluids

of pressure). On the other hand, the terminal velocity in the Newton's law
range is not strongly dependent on particle diameter, and depends on fluid
density rather than viscosity. Therefore measurements in which particles are
classified by terminal velocity are carried out at low Reynolds number, with
close temperature control. For the 'critical' range, roughly 2000 < d* < 3500,
Figure 2.4 shows that three terminal velocities are possible. The intermediate
value, portion AB of the curve corresponding to the range of Re in which CD
decreases sharply (see Figure 2.2), is unstable. Particles in this range have
been observed to fluctuate between the two stable terminal velocities, giving
a rather erratic motion. The curve beyond B represents supercritical motion.
The dimensionless diameter d* proves to be a convenient independent
variable in describing a range of fluid-particle operations, beyond just
estimation of terminal velocities. It is therefore used throughout this book.
Similarly, velocities made dimensionless in the form of equation (2.14) will be
used in a number of other places.

Example 2. 1 Terminal velocities of spheres in water

Problem
Calculate the terminal velocities of spheres with the density of sand
(2650 kg m-3 ) in water at 293K (20°C). The properties of water at this
temperature are Q = 998.2 kg m-3 , fA, = 1.002 X 10-3 N s m-2. Cover the
diameter range from 10 fA,m to 0.1 m in decades.

Solution
This type of calculation is particularly facilitated by using a correlation of the
form Vi[d*] because the property groups in equations (2.12) and (2.14) are
constants.

d*
d
[ Q(Q'"~ Q)g T
= [998.2 X (2650 - 998.2) X 9.81 J:;'
(1.002 X 10-3 )2
= 2.526 X 104
V *t
[ 2 ]11'
fA,(Qp ~ Q)g .
998.22 . ] Y3
[
1.002 X 10-3 X (2650 - 998.2) X 9.81
= 39.44
 Single particles 63

It is also instructive to calculate the terminal Reynolds numbers as

Ret = d*Vi
and also the estimates obtained from Stokes' law (equation 2.9) and Newton's
law (equation 2.10).
Taking the 1 mm (10-3 m) particle as a specific example
d* = 2.525 X 104 X 10-3 = 25.26
so, from the appropriate correlation in Table 2.1,
Vi = 6.328
Vt = VU39.44 = 0.160 m S-1

Proceeding similarly for all the diameters:

d(m) d* Re

Stokes' Newton's
law law

10-5 0.2526 3.544 X 10-3 8.99 X 10-5 8.95 X 10-4 8.99 X 10-5
10-4 2.526 0.3251 8.24 X 10-3 0.821 88.99 X 10-3
10-3 25.26 6.328 0.160 160 0.899 0.220
10-2 252.6 28.99 0.735 7320 0.697
10-1 2526 82.69 2.10 2.09 X 105 2.20

Comments
This particle density is representative of many naturally and industrially
occurring materials. The inaccuracy in Stokes' law is already apparent for a
particle 10-4 m (i.e. 100 IJ.m) in diameter; Stokes' law is typically inapplicable
to particles bigger than about 100 IJ.m in liquids. Newton's law gives rough
estimates, but only rough estimates, for particles larger than a few millimetres
in diameter. Particles as large as 1 cm in diameter are in the critical range
where two terminal velocities are possible: the estimate above refers to the
lower, sub critical value.

Example 2.2 Effect of temperature and pressure on terminal velocities of

particles in air
Problem
Plot the terminal velocities in air of particles of diameters 10 IJ.m, 100 IJ.m and
1 mm with density 2650 kg m-3 , as functions of temperature from 300K
(27°C) to llOOK (827°C) for pressure of 1 bar and 10 bar. Relevant properties
of air are (Mayhew and Rogers, 1972):
64 Particles in fluids

T(K) f-l (N s m-2) e (kg m-3)

1 bar 10 bar

300 1.846 X 10-5 1.177 11.77

500 2.670 X 10-5 0.706 7.06
700 3.332 X 10-5 0.5043 5.043
900 3.897 X 10-5 0.3922 3.922
1100 4.396 X 10-5 0.3209 3.209

Solution
The general procedure follows that used in Example 2.1, but the property
groups must be calculated for each set of conditions.

T(K) d*/d V7IVt

1 bar 10 bar 1 bar 10 bar

300 4.477 X 104 9.633 X 104 1.424 6.619

500 2.953 X 104 6.356 X 104 0.8956 4.160
700 2.277 X 104 4.903 X 104 0.6647 3.087
900 1.886 X 104 4.062 X 104 0.5335 2.477
1100 1.628 X 104 3.507 X 104 0.4483 2.082

which leads to the following estimates for terminal velocity (in m S-l):

T(K) d = 10-5 m d=lO-4 m d = 10-3 m

1 bar 10 bar 1 bar 10 bar 1 bar 10 bar

300 7.82 X 10-3 7.75 X 10-3 0.583 0.341 7.02 2.51

500 5.41 X 10-3 5.39 X 10-3 0.473 0.320 8.08 3.06
700 4.33 X 10-3 4.32 X 10-3 0.406 0.303 8.68 3.46
900 3.71 X W-3 3.70 X 10-3 0.357 0.282 9.04 3.76
1100 3.28 X 10-3 3.28 X 10-3 0.321 0.266 9.26 3.99

The values for 100 f,lm and 1 mm particles are plotted in Figure 2.5, together
with the intermediate diameter, 300 f,lm, for comparison.

Comments
Note how the terminal velocities of fine particles in the Stokes law range
decrease with increasing temperature, being inversely proportional to gas
 Single particles 65

Vt
(m/5)
1 mm.

- - J-

--
o. 2 L....-~--'----L..----J'--..I....-.....L.-.......L..-----I

400 700 1000

Temp.(K)

Figure 2.S Terminal velocities of spherical particles of density 2650 kg m-3 in air, as
functions of temperature: - - - , 1 bar; - - - - -, 10 bar.

viscosity, but are essentially unaffected by pressure. The implication is that

the difficulty of removing fine particles from gases increases with increasing
temperature.
Particles which are large enough to be outside the creeping flow range show
a more complex effect on V t of process conditions. In the Newton's law range,
increasing temperature reduces the fluid density and so increases V t . For
particles smaller than about 10 ftm in gases, it is necessary to take account of
non-continuum effects; see section 2.1.4.

2.1.3.2 Non-spherical particles. Non-spherical particles present problems

not present for spheres, as introduced in section 1.2. In particular:
66 Particles in fluids

(a) Whereas a sphere has a single unambiguous diameter, the appropriate

length scale to describe an irregular particle is not obvious a priori.
(b) The drag on and hence the terminal velocity of a non-spherical particle
depends on its orientation.
(c) In some Re ranges a particle can maintain any orientation, whereas in
other ranges it may tend to adopt a preferred orientation, while in yet
other ranges a secondary motion can be superimposed on steady fall or
nse.
These problems will be illustrated by reference to some well-defined non-
spherical shapes. In addition to cylinders and discs, spheroids prove to be
useful 'reference shapes', especially in the creeping flow range. A spheroid is
an 'ellipsoid of revolution', obtained by rotating an ellipse about its major or
minor axis as shown in Figure 2.6. Rotation about the minor axis gives an
oblate spheroid, i.e. a 'compressed' sphere. If the axis of symmetry is the
major axis, then the spheroid is prolate, i.e. elongated from pole to pole. The
shape of the spheroid is characterised by its aspect ratio, E, defined here as
the ratio of the polar diameter (2b) to the equatorial diameter (2a). Thus
E < 1 for an oblate spheroid and E > 1 for a prolate spheroid. The volume of
a spheroid is (4na 3 E/3). Many industrially and naturally occurring particles
tend to be flat, platy or lenticular; oblate spheroids are convenient 'model
shapes' for such particles. The limit E ~ 0 represents a very thin disc. Many
other particles are fibrous or needle-shaped, and the prolate spheroid is then
a convenient 'model shape'.

2.1.3.2.1 Creeping flow. We concentrate here on particles which can be

treated as axisymmetric, and which have 'fore-and-aft' symmetry, i.e. they
have a plane of symmetry normal to the axis. In addition to genuinely
axisymmetric particles like cylinders, discs and spheroids, this group of

IAXIS
I
IAXIS
I, 0 I--
I-O--l

J
=- -r
- - - PLANE OF
b SYMMETRY
b
1 J_
OBLATE
PROLATE
Figure 2.6 Oblate and prolate spheroids.
 Single particles 67

particles includes shapes such as bars whose sections are regular polygons.
For more detailed accounts of the behaviour of particles in creeping flow,
including results for other shapes, see Clift et al. (1978) and Happel and
Brenner (1983).
In creeping flow, it is useful to define the principal resistances to translation
of the particle. For motion parallel to the axis of an axisymmetric particle at
velocity U1 , the drag force is parallel to the relative velocity and given by
(2.15)
where C1 is the resistance to axial motion. Similarly, for motion normal to the
axis at velocity U2 , drag and velocity are again parallel with
(2.16)
where C2 is the resistance to motion normal to the axis. For a sphere of
diameter d = 2a, C1 == C2 = 3nd = 6na. Expressions for the principal
resistances of spheroids, thin discs and cylinders are given in Table 2.2.

Table 2.2 Principal resistances of spheroids, discs and cylinders in creeping flow

Analytic
Spheroids: Oblate (E < 1) Prolate (E > 1)
8Jta(E2 - 1)

[{(I - 2E')cos-1E}I--I(1 - E2)] +E [{(2E2 - 1)lo(E + --I(E2 - 1))}--I(E' - 1)]- E

Disc (E -'> 0): 16a

Normal to 16Jta(1 - E2) 16Jta(E' - 1)
axis, C2:
[{(3 - 2E')cos-1E}I--I(1 - E2)] - E [{(2E2 - 3)lo(E + --I(E2 - 1))}1--I(E' - 1)] + E

Disc: 32AI3
Mean, in
random 6JtaV(1 - E2) 6JtaV(E2 - 1)
orientation,
E:
cos-IE In(E + V(E2 - 1»
Disc: 12a
Empirical (from Heiss and Coull, 1952)
Cylinders E = Lid; 0.1 < E < 10
Axial, 5.82dE- Y' V(2E + 1) exp[0.367e-o.448 V(2E + 1){(3EI2)'h - I}]
1)]
Cl

Normal to 0.939E'i4(l.0145g-1I6 -
axis, C2: 6.57dE 1I6 V(2E + 1) exp [
V(2E + 1)
Note that, for a sphere, d = 2a and C1 = C, = 3nd or 6na.
68 Particles in fluids

Because drag and relative velocity are parallel when the motion is parallel
to one of the principal directions, axisymmetric particles with fore-and-aft
symmetry settle on a vertical path if the axis is either vertical or horizontal.
The corresponding terminal velocities, for a particle of volume v, are
(2.17)
Oblate bodies generally have C2 < Cl, so that they fall faster with the axis
horizontal. The converse is true for prolate shapes. As noted in section
2.1.2.2, a very prolate (i.e. needle- or fibre-like) axisymmetric particle has C2 =
2CI, so that it falls twice as fast when oriented vertically than when its axis is
horizontal.
For any other direction of motion, drag is not parallel to the relative
velocity. Thus, while an axisymmetric particle with fore-and-aft symmetry
moves on a straight trajectory without rotating, the trajectory is not in general
vertical. Figure 2.7 shows such a particle settling with its axis at some angle <\>
to the vertical. The path of the particle lies in the vertical plane through the
axis, but is at some angle J3 to the vertical. The components of relative
velocity parallel and normal to the axis are respectively Ucos( <\> - J3) and
Usin( <\> - J3). The two components of the drag force are then

FDI = [lC1Ucos(<\> - J3); FD2 = [lC2 Usin (<\> - J3) (2.18)

acting in the directions shown in Figure 2.7. For steady motion, the total
horizontal and vertical forces on the particle must be zero, i.e.

FDl sin <\> - FD2 cos <\> =0 (2.19)

FDJ cos <\> + FD2 sin <\> = v(gp - g)g (2.20)

v(pp-p)g
u
Figure 2.7 Axisymmetric particle moving at terminal velocity in creeping flow.
 Single particles 69

Solving equations (2.18) to (2.20) (see Clift et al., 1978) gives

(2.21)

(2.22)

(2.23)

Equation (2.21) shows that, for a prolate particle with Cl < C2, J3 is positive,
i.e. the particle tends to 'sideslip' on a trajectory between the vertical and the
axis. Conversely, an oblate body with Cl < C2 hasJ3 negative, i.e. its trajectory
lies between the vertical and the equatorial plane. Figure 2.7 represents both
these cases. For a large number of identical particles in random orientation,
the mean resistance and terminal velocity are:
(2.24)

(2.25)

Particles such as regular polyhedra, for which the principal resistances are
equal in three orthogonal directions, have the same resistance in every
direction and settle on a vertical trajectory whatever their orientation. The
terminal velocity can be estimated reasonably reliably from the empirical
correlation (see Clift et al., 1978):
V t = V ts (1 + 0.3671n '\jJ) (2.26)

where V ts is the terminal velocity in creeping flow of a sphere with the same
volume and density, and '\jJ is the sphericity of the particle.
One of the applications of these results is in estimating relationships
between volume-equivalent and Stokes' diameters of non-spherical particles
(see section 1.2.1). The Stokes diameter d st is, by definition, the diameter of
the sphere with the same density and terminal velocity as the particle, i.e.
from equation (2.9),
'12
dst = [18""V/(Qp - Q) g] (2.27)

Using equation (2.17) to eliminate V t ,

d st = [
18v/c]
1/2
(2.28)
70 Particles in fluids

where c is the appropriate hydrodynamic resistance of the particle. In terms

of the volume-equivalent diameter of the particle, d v = (6vl:l1f3, equation
(2.28) becomes
(2.29)
Thus the ratio of the Stokes to volume-equivalent diameter can be evaluated
for a 'model' shape for which expressions for c are available. For example, for
a spheroid
(2.30)
so that the result for oblate spheroids in random orientations in Table 2.2
leads to
dstfd v = [E'/'cos-1Etv'(1 - E2) t2 (2.31)
Corresponding results can be derived for motion parallel and normal to the
axis, for prolate spheroids, and for discs and cylinders (Ghadiri et at., 1986).
Figure 2.8 shows the results for spheroids. The results for other shapes are
very similar, illustrating the value of the spheroid as a 'model particle'. This
derives from the general result, introduced in section 2.1.2.2, that the drag on
a particle is less than or equal to that on a circumscribing body and more than
or equal to that on an inscribed body; thus the proportions and overall
dimensions of the particle are of most importance, the detailed shape less so.
Figure 2.8 also shows that dstfdv remains close to unity except for very platy or
very elongated shapes. Thus the volume-equivalent diameter can be used as a
good first approximation to the Stokes diameter for a broad range of particle
shapes.

;' "-
;'
3 "-
;' '''-~
ds/de ;' "-
"-
:2' , "-
"-
"-
"-

0.5

0;1 to 10.0 100

Figure 2.8 Ratio of Stokes' to volume-equivalent diameter for spheroids. 1. motion

parallel to axis; 2. motion normal to axis; 3. random orientation.
 Single particles 71

Example 2.3 Settling of discs and oblate spheroids

Problem
Compare the terminal velocities in creeping flow of thin discs and oblate
spheroids with the same volume and diameter.

Solution
A disc of diameter d and thickness kEd has volume (rr,d 3EI4) X Ed, i.e. v =
rr,d3E/4.
Similarly, a spheroid with equatorial diameter d and aspect ratio E' has
volume
v = rr,3E'/6
so that, if the disc and spheroid have the same volume,
E' = 1.5E
From equation (2.17) the terminal velocity of the particle is
Vt = v(Qp - Q)g/f.!C
where C is the hydrodynamic resistance in the appropriate orientation.
Therefore the ratio of the terminal velocity of the disc to that of the sphere
IS

V/V; = c'/c
where c is the resistance of the disc of aspect ratio E and c' is that of the
spheroid of aspect ratio E' (= 1.5E). Consider three aspect ratios, E = 0.2
(E' = 0.3), E = 0.1 (E' = 0.15) and E = 0.01 (E' = 0.015). For the first two
shapes, the Heiss and Coull (1952) empirical results can be used; the third
shape lies outside the range of these correlations, but can be approximated as
a vanishingly thin spheroid (E ~ 0) because the drag on the curved sides can
be neglected. For the oblate spheroids, the results in Table 2.2 are used in
each case:

Motion Motion Random

parallel normal orientation
to axis to axis

(a) E = 0.2 (E' = 0.3)

Resistance of disc 8.765d 7.069d 7.556d
Resistance of spheroid 8.237d 6.643d 7.101d
V/V; 0.940 0.940 0.940
(b) E = 0.1 (E' = 0.15)
Resistance of disc 8.096d 6.206d 6.724d
Resistance of spheroid 8.972d 6.000d 6.561d
VtlV; 0.997 0.967 0.976
72 Particles in fluids

Table Continued

Motion Motion Random

parallel normal orientation
to axis to axis

(c) E = 0.01 (E' = 0.015)

Resistance of disc 8d 5.333d 6d
Resistance of spheroid
V/V; 1.0 1.01 1.Ol

Comments
These calculations further illustrate the general point that, in creeping flow,
terminal velocities are insensitive to particle shape. In fact, the differences
between the estimates obtained by treating lenticular particles as discs and as
oblate spheroids are within the sort of accuracy which can be reasonably
expected of an empirical correlation. The difference between the two shapes
increases, to pass through a maximum around E = 1.0, but is never very large
(see Clift et al., 1978).

2.1.3.2.2 Intermediate Reynolds numbers. At Reynolds numbers inter-

mediate between the creeping flow and Newton's law ranges, a non-spherical
particle in free motion generally tends to adopt a preferred orientation with
the maximum direction normal to the direction of motion relative to the fluid.
Thus a disc or oblate spheroid, for example, falls with its axis vertical,
whereas a long cylinder or prolate spheroid tends to fall with its axis
horizontal. The lower limit of the intermediate range itself depends on the
shape of the particle: Re is around 0.1 for a disc and 0.01 for a cylinder.
Similarly the upper limit of the intermediate range depends on particle shape
(see below); Re is typically of order 100. The results summarised in this section
refer to free fall in the preferred orientation. However, it must be
remembered that, especially when QpiQ » 1 (i.e. primarily for particles in
gases), a non-preferred orientation can be maintained over large distances;
the drag and relative velocity may then be quite different.
Because the motion is steady and rectilinear, measurements of drag on a
rigidly-mounted particle in the preferred orientation can be used to estimate
terminal velocity. Thus the results in Table 2.1 can be used to estimate the
terminal velocities of long cylinders or discs. For cylinders with E = Lid> 1;
it is convenient to base calculations on the dimensionless diameter and
terminal velocity defined by equations (2.12) and (2.4). Table 2.1 also gives
explicit forms for Vi[ d*] in the intermediate range, derived from the
correlations given in slightly different form by Clift et al. (1978).
For particles of arbitrary or irregular shape, the fact that the particle
 Single particles 73

presents its maximum cross-section to the direction of motion helps to define

the appropriate length dimension to be used to characterise it. Because form
drag is generally much larger than skin friction in this range, the most
important parameter is the cross-section. Therefore, as Heywood (1938) first
pointed out, the most appropriate dimension is the projected-area-equivalent
diameter, dA , i.e. the diameter of the sphere with the same cross-sectional
area as the particle. Because a particle in its equilibrium orientation on a flat
horizontal surface presents the maximum cross-section to view normal to the
surface, the appropriate area and hence dA can be measured by microscopic
observation or image analysis. If the particle size is determined by sieve
analysis, dA is somewhat less than the screen opening. The difference depends
on the shape of the maximum section; if it is roughly isometric, then it is
frequently a reasonable approximation to estimate dA as the screen opening.
These ideas led Heywood to suggest that particles can be characterised by a
'volume shape coefficient' (see section 1.2.2):
<J.vA = (volume of particle)/d~ (2.32)
so that <J.vA is 0.524 for a sphere. Representative values for other particle
shapes are given in Table 2.3. An 'isometric particle' is one for which all three
orthogonal dimensions are equal; for example, the sphere is the isometric
shape corresponding to a non-isometric spheroid. Table 2.3 gives the shape
factor, <J.vAI, for various isometric shapes based on the shape factors in Table
1.4, and calculated using equation (1.3.7a). Values for non-isometric shapes
can be obtained as
(2.33)
where Land B are the two major dimensions of the particle (normally at right
angles in the plane of maximum projected area) while T is the thickness in the
direction normal to this plane. The value for <J.VAI to be used in equation
(2.33) is that for the isometric body of homologous form, for example the
sphere for a spheroid, the cube for a cuboid, etc. In general, the more flaky
the particle, the lower is the value of K. Table 2.3 gives typical values for the
volumetric shape factor for selected mineral particles.
Heywood's procedure to estimate the terminal velocity of a particle in the
intermediate range is then as follows:
(a) Estimate V ts , the terminal velocity of the sphere of diameter dA with the
same density as the particle.
(b) Estimate the terminal velocity of the particle, V t , as
(2.34)
where the velocity correction factor, K A , is correlated empirically as a
function of d'A and <J.vA, as shown in Figure 2.9.
The factor KA attempts to account for two influences on V/Vts : the volume
74 Particles in fluids

Table 2.3 Typical values for volumetric shape

factor

Isometric particles (aVAI)

Regular:
sphere 0.524
cube 0.696
tetrahedron 0.328
Irregular:
rounded 0.54
subangular (partly rounded) 0.51
angular:
(i) tending to cuboidal 0.47
(ii) tending to tetrahedral 0.38
Mineral particles (typical aVA)
Sand 0.26
Sillimanite 0.23
Bituminous coal 0.23
Blast furnace slag 0.19
Limestone 0.16
Talc 0.16
Plumbago 0.16
Gypsum 0.13
Flake graphite 0.023
Mica 0.003

Figure 2.9 Velocity correction factor for non-spherical particles at intermediate

Reynolds numbers. Ratio of terminal velocity of non-spherical particle to value for
sphere, aVA, as a function of dimensionless diameter d*.
 Single particles 75

of the particle is actually less than that of the sphere of diameter d A , while its
drag coefficient may be somewhat different from that for the sphere.
Generally the former influence is the stronger, which explains why KA is
always less than unity and probably helps to explain why this approach is the
most reliable of those available for estimating V t for irregular particles (see
Clift et al., 1978). The estimates are usually within ±20% and frequently
within about 10%. Given the impossibility of describing particle shape by a
single parameter, this is about the best which can be expected of any simple
empirical correlation.

2.1.3.2.3 Newton's law range. At higher Re, such that the wake behind
the particle sheds intermittently, secondary motion sets in. In addition to
shape, size, and (Qp - Q), the mean terminal velocity then depends on Qp
because the particle inertia determines how strongly it is influenced by wake
shedding. Even spheres show the influence of QpiQ in the Newton's law
range, and can follow zig-zag or spiral trajectories, although the effect on V t is
small. The effects are much stronger for other shapes.
Discs are probably the best understood. In addition to the dimensionless
diameter, conveniently defined for a disc of thickness Ed as

d* =d [ E Q( Q:: Q) g J!h (2.35)

it is also necessary to define the dimensionless moment of inertia,

J* = nEQpi64Q (2.36)
Figure 2.10 shows a map of the possible regimes of motion, constructed by
Clift et al. (1978) from the results of Willmarth et al. (1964) and Stringham et
al. (1969). Three types of secondary motion have been observed, although
the transitions between them are not sharp. When unsteady motion first sets
in, a disc shows regular oscillations about a diameter; the amplitude and
horizontal 'sideslip' increase with d* and decrease with [*. The frequency of
oscillation, f, is indicated in Figure 2.10 in terms of the Strouhal number,
Sr= fdlVt (2.37)
At higher d*, in the glide-tumble range, a disc 'flies' in a succession of curved
arcs, shedding a wake vortex at the end of each arc. At high [*, the disc can
tumble about a diameter, following a trajectory which is roughly rectilinear
but not vertical. Once the motion becomes unsteady, the mean drag can be
quite different from that on a fixed disc in a flowing fluid: it is generally higher
in free fall, but may be lower around the transition from glide-tumble to
tumbling motion. Confusion between fixed and freely-moving discs probably
explains why some authors erroneously show a maximum in CD [Re] around
Re = 300 (see Figure 2.2).
76 Particles in fluids

i
0,01

I,
c
15
<1>
I
~ 10' STEAOf

III<1> '
c:
,Q
'"c
<1>
E
o

Figure 2.10 Regimes of motion for discs in free fall or rise: _._._, contours of
constant Re; - - - , contours of constant Sr. Bold curves indicate approximate
boundaries between regimes of motion.

These observations help to understand the behaviour of other shapes

whose motion is not so well characterised. Cylinders with E > 1 oscillate and
'sideslip' in a vertical plane, the amplitude decreasing with increasing E. A
cylinder with E = 1 'tumbles' on a trajectory inclined to the vertical. Particles
of other shapes with the ratio of maximum to minimum dimension < 1.7 can
be regarded as isometric, and the terminal velocity estimated as

Vt = 0.49 ( Qp )1136 [_d_vC_Qp_--Q-)g-J 'Ii (2.38)

Q Q(1.08 - 1jJ)

where d v is the volume-equivalent diameter and 1jJ the sphericity. For other
shapes in the Newton's law range, it is generally advisable to measure V t
rather than attempting to estimate it.

2.1.4 Non-continuum effects

For most processes involving particles in fluids, the fluid can be treated as an
incompressible continuum. However this simplification breaks down for very
fine, usually submicron (but see below), particles in gases. It follows that non-
continuum effects are only significant if the particle Reynolds number is very
small (see Clift et al., 1978). Non-continuum effects can be significant in some
gas/solid separation processes, as discussed in Chapter 7. Compressibility
 Single particles 77

effects, arising when the relative velocity between the particle and the fluid is
comparable with the speed of sound in the fluid, can usually be ignored in
processing equipment, except in processes like atomisation which use high-
velocity gas jets.
The significance of non-continuum effects is indicated by the Knudsen
number, Kn, defined as the ratio of the molecular mean free path in the fluid
(/...) to some characteristic dimension of the particle, usually the diameter (d):
Kn =/.../d (2.39)
Equations using Kn are at best estimates, so that there is little point in
questioning in detail which measure of d should be used in equation (2.39) for
non-spherical particles; if in doubt, the volume-equivalent diameter should be
used. Clift et al. (1978) discuss some results for the effect of slip on non-
spherical particles.
The mean free path depends on the gas temperature and pressure, and can
be estimated as (Beard, 1976)
/. . = 3.2 X 10-4f.tYJt/2QP (2.40)
For air, equation (2.40) can be approximated as
/. . = 2.15 X lO-4f.tr'h/P (2.41)
Note that /... is in metres and P in bars in both these equations. As a general
approximation from equation (2.4), /... is roughly proportional to T/P,
showing that non-continuum effects become more significant at elevated
temperature and reduced pressure. To indicate typical values, /... for air is
0.069 f.tm at 300 K and 1 bar, 0.32 f.tm at llO0 K and 1 bar and 0.032 f.tm at
1100 K and lO bar.
Non-continuum effects reduce the fluid/particle drag, and for this reason are
sometimes termed 'slip effects'. The drag (always at low Reynolds number;
see above) can be estimated for near-continuum slip flows as
FD = 3Jtf.tdUlC (2.42)
where C is the slip correction factor or 'Cunningham coefficent' defined as

Drag on particle in continuum flow (at same Re)

C = (2.43)
Drag on particle in presence of 'slip'

The value of C for a given particle and gas actually depends on the
'accommodation coefficient', i.e. the fraction of gas molecules undergoing
diffuse reflection from the particle surface. Clift et al. (1978) summarise
results which can be used to estimate C if the accommodation coefficent is
known. Usually it is not known and C must be estimated empirically from
Davies' (1945) modification of a form proposed by Knudsen and Weber
(1911):
78 Particles in fluids

c = 1 + Kn [2.514 + 0.8 exp (-0.55/Kn)] (2.44)

Some representatives values for C for air are shown in Figure 2.11, together
with the corresponding terminal velocities for particles of density 2500 kg m-3 .
It is worth noting that, whereas C only departs significantly from unity for
sub micron particles at ambient conditions and at elevated pressure, slip
effects are significant for particles several microns in diameter at elevated
temperature and ambient pressure. In the absence of slip effects, the terminal
velocity depends on gas temperature, but not on pressure, through the effects
on gas viscosity (see Example 2.2 and Chapter 7).
10-2 10-8,--------,---------,

\
10-3 10-9
\
\
\
\
\
'iU) ~
E 10"'" E 10-10 <..)
C!l
~ C

10-5 10

10-6 -------
c

0.1 1 10
dp (ll m)

Figure 2.11 Variation of terminal velocity, diffusion coefficient and Cunningham slip
correction factor with particle diameter (Clift et ai., 1981). For particles of density
2500 kg/m3 in air and 300 K, 1 bar (--); 1100 K, 1 bar (_._); and 1100 K, 10 bar
(----).
 Unsteady motion of single particles 79

Figure 2.11 also shows values for the Brownian diffusivity , DAB, associated
with the random motion of the particle in the gas due to collisions with gas
molecules. The values are calculated from the Stokes-Einstein equation (see
Clift et at., 1981)
DAB = Ck B Tl3nfld (2.45)
where kB is Boltzmann's constant, 1.380622 X 10-23 JK-1 . Brownian
diffusivity is a significant parameter in some gas/solid separation processes, as
discussed in Chapter 7.

2.2 UNSTEADY MOTION OF SINGLE PARTICLES

2.2.1 Drag in unsteady motion
The preceding sections of this chapter were concerned with steady motion of
a particle in a fluid which is either stagnant or moving at steady uniform
velocity. Two dimensionless groups, e.g. CD and Re or d* and Vi, then
described the problem. However, if the motion is unsteady, then the inertia
of the particle is important as well as its immersed weight, i.e. Qp is
significant as well as (Qp - Q). Furthermore, accelerations as well as steady
velocities must be considered. Thus the number of potentially significant
variables proliferates so that dimensional analysis ceases to be helpful. It is
therefore most constructive to consider some specific types of unsteady
motion, with a view to deriving guidelines of general value. A detailed
discussion is given by Clift et at. (1978) but some general points will be
summarised here.
In the creeping flow range, the drag on a sphere in unsteady motion
through a stagnant fluid can be written

nd3 ]
[ __ _
dU + _d2-vJtQiA
3 t [dUJ
_ __ ds
6 dt 2 roo dt t=s-vt=S
(2.46)

where U(t) is the instantaneous particle velocity. Thus the total instantaneous
drag is made up of three terms, of which the first is the steady (Stokes) drag at
the instantaneous velocity. The second term on the right of equation (2.46) is
known as the 'added mass' or 'virtual mass' term. It arises from the entrained
fluid which accelerates with the particle; for a sphere, this 'added mass' of
fluid corresponds to a volume equal to half that of the sphere. The final term,
known as the 'Bassett term' or 'history term', involves an integral over all past
accelerations of the sphere. The form of this term arises from generation of
vorticity at the surface of the particle, to diffuse into the surrounding fluid.
The history term can be thought of as describing the state of development of
80 Particles in fluids

the flow around the particle: if it is small, then the relaxation time of the flow
is sufficiently small that the instantaneous flow field differs little from steady
flow at the instantaneous value of U.
Equation (2.46) can be developed to give an equation of motion of the
sphere, for example to describe the initial motion of a particle released from
rest. The form of equation (2.46) only applies in creeping flow. Nevertheless,
and slightly surprisingly, a modified form with empirically-determined
coefficients of drag, added mass and history gives a good description of
accelerated motion at higher Reynolds numbers; see Clift et al. (1978).
Solution is obviously complicated by the history integral, but some useful
general conclusions emerge:
(a) The history term is usually larger than the added mass term. Therefore it
is futile to ignore the inconvenient history term but to include the added
mass term.
(b) For particles in gases, with Qp » Qg, the added mass and history terms
are often small, so that the drag can be approximated by the steady-state
drag at the instantaneous velocity. However, this is by no means
universally true, even for particles in gases. It is therefore advisable to
estimate U(t) by making the quasi-steady approximation, and then
evaluate the two other terms to verify the validity of the approximation.
(c) For particles in liquids, the quasi-steady assumption is almost never
justified. Therefore the kind of calculations made for particles in gases,
for example in inertial devices for separating particles from gases (see
section 7.1), cannot be applied to particles in liquids.

2.2.2 Acceleration
As shown earlier (equation 2.9), the terminal velocity of an isolated particle
in free-fall under gravity, in the Stokes law range and in the absence of slip
effects, can be expressed as
(2.47)

where

(2.48)

'tis known as the 'relaxation time', so called because it characterises the time
taken for a particle to adjust to any changes in flow velocity or direction. This
approach can be generalised to give the terminal velocity of a particle
subjected to any constant external force, F,
F
V tF = 't- (2.49)
m
 Unsteady motion of single particles 81

During acceleration in a gas (see section 2.2.1), the drag force may still be
approximated by the Stokes law, using the instantaneous velocity, so that for
acceleration from rest in a gravity field

dV(t)
m - = mg - 3Jt!lV(t)d (2.50)
dt
or
dV(t)
,; - - = ,;g - V(t) (2.51)
dt

which can be solved to give

V(t) = V t (1- e-tlt ) (2.52)
By similar reasoning, if Vo is the initial particle velocity at t
final (steady-state) velocity
= °and V f is the
V(t) = Vf - (Vf - Vo)e- tlt (2.53)
Integrating this equation, we obtain the distance travelled
x(t) = Vft - (Vf - Vo),;(1 - e-t/') (2.54)
In the case of a particle projected into a stagnant gas at velocity Yo, so that V f
= 0, this reduces to
(2.55)
so that the 'stopping distance' or total distance travelled (for t » ,;) is
simply
(2.56)
This is of importance in estimating how far a particle can be 'thrown' into a
stagnant gas. (For a 1 !lm particle of density 103 kg m-3 in air, Xs is about
3.6 !lm per (m S-l) initial velocity, i.e. a small multiple of its diameter.)

2.2.3 Curvilinear motion

In general, curvilinear motion occurs when the forces on a particle act in more
than one direction or when there is a change in the direction of the fluid
streamlines. The latter may arise when a fluid encounters an obstacle, in
passing through a bed of fibres or particles for example, and the motion of
particles which are carried by that fluid is of considerable importance in the
study of filtration phenomena (see section 7.2).
It is possible to analyse the motion of particles independently in two
orthogonal directions if that motion is within the Stokes law regime. For
82 Particles in fluids

o ------;--------- --->0-
X

y V

Figure 2.12 Particle trajectory for a settling particle with an initial horizontal
velocity.

example, consider a particle projected horizontally into a stagnant gas at a

velocity Vo in a gravitational field (Figure 2.12). From equations (2.55) and
(2.54)
x(t) = Vo't(l - e- tft ) (2.57)
yet) = Vtt - V t't(1 - e-I/'t) (2.58)
If the particle motion is outside the Stokes regime, so that drag and relative
velocity are not linearly related, the motion in one direction is coupled with
the motion in orthogonal directions and a more complicated analysis is
required.
For flow around an obstacle, it is first necessary to define the flow field,
using appropriate assumptions if necessary. The equation of motion of the
particle is then solved numerically.
Motion of particles in fluids with curved streamlines or of rapidly changing
velocity is characterised by the use of a Stokes number, St, which is the ratio
of the stopping distance of the particle to a characteristic dimension of the
flow, such as a dimension of the obstacle causing the flow disturbance, de.
xs 'tVo
St = - = - - (2.59)
de de
Here Vo is interpreted as the undisturbed flow (and therefore particle)
velocity at infinite distance upstream of the flow disturbance. Thus the Stokes
number can be interpreted as the ratio of the effects of particle inertia to fluid
drag, a measure of the tenacity with which particles 'hold' to their course
rather than following the streamlines. Small values of St indicate good
 Unsteady motion of single particles 83

'tracking' of the streamlines, while large values indicate that the particle will
'resist' changes in direction or velocity.

Example 2.4 Inertial impaction (after Hinds, 1982)

Problem
Inertial impaction is one of the main mechanisms of particle capture in filters
(section 7.2) and also a method of sizing and classifying particles (section
1.2.4). A flow of gas is directed to impinge on a flat plate as shown in Figure
2.13. Calculate the critical Stokes number for which 50% of the particles in
the flow through a rectangular slot nozzle are collected by the impaction
plate, given that
(Qp - Q)rf2u 'tU
St= =-- (2.60)
18f!(dP) (dP)
where U is the mean gas velocity in the nozzle and (dP) is the half-width of
the nozzle.

Nozzle

r
~~_r--_ _ _ _ _~U
-Impac';o" pi",

I
I
I

Figure 2.13 Simplified rectangular impactor model (after Hinds, 1982).

84 Particles in fluids

Solution
A full analysis of the problem would require numerical solution of the
complex flow field. Consider a rectangular impactor nozzle of half-width (d/
2) and infinite length (i.e. a two-dimensional problem) and assume:
(a) uniform flow across the entry;
(b) that the fluid streamlines are arcs of a circle centred on A;
(c) that departures from the streamlines are small, so that the radial velocity
can be taken as constant.
The terminal radial velocity can then be taken as

(2.61)

where ar is the radial acceleration, U2 /r, and r is the radius of curvature of the
streamline. The velocity, U, is assumed to remain constant throughout the
curved region, which is a gross assumption but adequate for the present
purpose. Hence,

(2.62)

and the total radial displacement,

~ = Vrt~t (2.63)
where ~t is the time taken to traverse the quarter circle. Therefore,

~ = cU2 (2nr) = ncU (2.64)

r 4U 2
and the 'impaction efficiency', EJ, for this particle diameter will be given by
~ ncU n
E1 = - = -.- = - St (2.65)
d/2 dj 2

Any particle in the flow which is closer to the centre line of the nozzle than ~
will be collected by the impaction plate. Now the critical Stokes number, for
which 50% of the particles in the flow are collected, will be given by
n 1
-St=- (2.66)
2 2
or
Stcrit = 0.32 (2.67)
 Assemblages of particles 85

Comments
The assumptions made about the fluid and particle motion in this example are
gross and therefore good agreement with experimental values would not be
expected. Nevertheless, the experimental value of Stcrit for the rectangular
nozzles (0.53) is of the same order as that obtained from this first-order
calculation. Note that the nozzle-to-plate separation distance is relatively
unimportant in impactor design because the gas jet expands only slightly
before coming within one jet diameter of the impaction plate. The key
parameter is therefore the slot width, or diameter if circular holes are used.

2.3 ASSEMBLAGES OF PARTICLES

2.3.1 Settling and particulate fluidisation
The preceding discussions were concerned entirely with single particles. For
an assembly of particles (e.g. a slurry of particles in a liquid) the settling
velocity is typically much lower than the single-particle terminal velocity. The
reduction in settling velocity arises from two complementary effects:
(a) the displacement of fluid by the settling particles causes an upflow
through the interparticle voids;
(b) the drag on each individual particle is increased by the effect of the
neighbouring particles on the velocity profile in the interstitial fluid.
Provided that direct particle-particle interactions are negligible (see
below), the combined effect is described by the well-known Richardson-Zaki
correlation (see Kay and Nedderman, 1985),
(2.68)
where V~ and V t are, respectively, the 'hindered settling velocity' and the
single-particle terminal velocity, and E is the void fraction, i.e. (1 - Cy ),
where Cy is the volumetric concentration of particles. The index n depends
on the single-particle Reynolds number or on d*, and values are given in
Table 2.4. Note that the effect of solids concentration is strongest in the
creeping flow range. Equation (2.61) has been shown to describe the

Table 2.4 Values for the Richardson-Zaki index

Rep at terminal velocity Value of n

Rep :'S 0.2 4.6

0.2 < Rep < 1 4.4 R ep-D·03
1 :'S Rep < 500 4.4 Re p-{)·1
500 :'S Rep 2.4
 86 Particles in fluids

expansion of particulately-fluidised beds (see section 6.2.2.3) as well as

sedimentation.
The Richardson-Zaki correlation only takes account of hydrodynamic
effects, in systems where the particles are randomly distributed as an
unstructured assemblage. Therefore it does not apply to very fine particles in
aqueous suspension, where the behaviour is dominated by interaction
between the charged regions adjacent to the particle surfaces; in these cases,
the hindered settling velocity can be higher or lower than V~ from equation
(2.61), depending on whether the particles agglomerate or form a structured
assembly. Similarly, it applies only to a very limited extent to particles in
gases, due to the fact that a uniform dispersion of particles in a fluid is
. unstable so that, if the density difference between the two phases is large, the
. dispersion rapidly readjusts to give regions of high and low particle
concentration. This is the origin of the 'bubbles' which are characteristic of
aggregative gas-fluidised beds, and of the 'clusters' of particles which form in
lean-phase fast-fluidised systems. These phenomena are discussed in section
6.2.6.

2.3.2 Flow through packed beds

The viscous (low Reynolds number) flow through a packed bed has been
considered in detail by a number of authors (see, for example, Dullien, 1992;
and Kay and Nedderman, 1985) and those arguments will not be reiterated
here. By analogy with the Hagen-Poiseuille law for viscous flow in pipes, it is
possible to derive a relationship for flow in packed beds which is known as the
Carman-Kozeny equation and takes the form

(2.69)

where D..p is the pressure drop, S is the particle surface per unit volume of
the bed, U is the superficial flow velocity (the volumetric fluid flow divided by
the cross-sectional area of the bed) and L is the thickness of the bed in the
direction of flow. The constant k has been found to be about 5 in many
practical cases.
If the bed consists of spherical particles of diameter d, then the number of
particles per unit volume is
6 (1 - E)/nd3

so that

6(1 - E) 6(1 - E)
S=ncPX =--- (2.70)
nd3 d
 Assemblages of particles 87

Hence from equation (2.69), taking k = 5,

(1 Ef f!UL
I1p = 180--- (2.71)
10 3 d2
or in terms of the pressure gradient

dp (1 - f!U Ef
- - = 180--- (2.72)
dl 103 d2

Thus, the pressure gradient is directly proportional to the flow rate, a result
first obtained by Darcy in 1846.
If the particles are not spherical or mono-sized, then (Kay and Nedderman,
1985)

6 (1 - E)
S=---- (2.73)

where tV is the particle sphericity defined in equation (1.32). The forms of

equation (2.70) and (2.73) show why the appropriate measure of particle size
in packed beds is the volume-surface mean (or Sauter mean), d sm , defined in
equation (1.15).
Equation (2.72) applies at low Reynolds numbers, where the flow is
dominated by viscosity. At higher Reynolds numbers, taking the analogy with
pipe flow once more, one would expect that the pressure gradient would
become proportion to gU2 , and this is indeed what happens. An empirical
equation by Ergun (1952) fits the experimental data over a wide range of
Reynolds numbers. For spherical particles this takes the form

(2.74)

For beds with a distribution of particle sizes, it is usual to use d sm in both

terms in equation (2.74), although it should be noted that this is only clearly
justified for the first, creeping flow term. Equation (2.74) provides the basis
for estimation of minimum fluidisation velocities, discussed in section 6.2.

2.4 Flow of particle/fluid mixtures in pipes

As a general background to topics covered elsewhere in this book, a brief
account will be given here of the continuity, momentum and energy equations
describing the flow of particle/fluid mixtures in pipes. Only steady flow will be
considered, i.e. all derivatives with respect to time at fixed location are taken
88 Particles in fluids

to be zero. However, we do not limit the discussion to fully-developed flow,

i.e. we allow conditions to vary along the pipe. The intention is to explain
some of the main features of particlelfluid flow at the macroscopic (and partly
qualitative) level, by exploring the facts that the velocities and concentrations
of the two phases are generally not equal and may vary across the pipe
section. For more complete and detailed discussions, see for example, Wallis
(1969), Govier and Aziz (1972) or Soo (1989).

2.4.1 Continuity
Figure 2.14 illustrates schematically the flow being considered. Fluid and
particles flow at volumetric rates Qf and Qs along a pipe of cross-sectional
area A, inclined at e to the horizontal. The superficial velocities of the two
phases are then defined to be
Uf = QflA (2.75)
Us=QJA (2.76)
The delivered volume concentration of the solid is defined to be
(2.77)
Now consider the mixture flowing through BB', a section normal to the axis
of the pipe. Consider an element dA of the cross-sectional area, as shown in
Figure 2.14. The local volume fraction of the solid in the element is C\; this
can be envisaged as the average fraction of time for which the element dA is
occupied by the solid phase. The mean in situ concentration of solids across
BB' is then defined as

Cyt = -1 f CydA
A
(2.78)
A A

Figure 2.14 Flow of a particle/fluid mixture in a pipe - continuity.

 Assemblages of particles 89

which can be interpreted as the average fraction of the section which is

occupied by solids. Similarly, the mean in situ fluid fraction at BB' is

-
If (1 - Cy)dA
A

=1- If
- , CydA
A

=1- C Yt (2.79)
A A A A

Consider now the flows of the phases along the pipe. Suppose that the
average velocities of the two phases parallel to the pipe axis at the element are
Vs and Vf, where each average is evaluated over the times when the element is
occupied by the phase in question. The flows of the two phases through dA
are then C\VsdA and (1 - Cy)VfdA. The total volumetric flows across BB'
are then

(2.80)

Qf = 1(1 - CV)VfdA (2.81)

A

and the mean phase velocities across BB' are

Vs = Q.lCvtA = [ f Cv VsdA ] / CvtA

A
(2.82)

(2.83)

The mean slip velocity between the two phases across BB' is defined as (Vf -
Vs) whereas the local slip velocity at dA is Vf - Vs. From equations (2.78) to
(2.83),

(2.84)

Equation (2.84) shows that there is no simple relationship between the

average and local slip velocities. As an extreme case, if Vs and V f are equal
everywhere across BB',

- - 1[ 1 f
Vf - Vs=-
A
- - ' (1- Cv)VdA- -I
1 - C Vt A
A A

C Vt
f CvVdA J
A
AA
(2.85)
90 Particles in fluids

where
Vf = Vs = V
Hence, from equation (2.85),

= 1 J (CYt -
..
Cv)VdA (2.86)
ACyt(l - Cyt) A

From equation (2.86), the mean slip velocity (Vf - Vs ) is still positive if V is
relatively large where (CYt - C\) is positive (i.e. Cy < Cyt ), and is relatively
small where (CYt - Cy ) is negative (i.e. Cy > Cyt ). To give a specific
example, applicable to both pneumatic and hydraulic conveying (see Chapter
6), there is mean slip, i.e. (Vf - Vs ) is positive, if the fluid flows at higher
velocity through a part of the pipe where the local solid concentration (Cy ) is
relatively small, even if there is no significant local slip between the phases.
At a different level, consider the case where Cy is essentially constant
across the section. In this case, it follows from equation (2.78) that Cy = Cyt .
This behaviour is approached by some vertical or high-velocity flows. Then
the above equations simplify to

VS = -1 J' VsdA; V-f = - J.

1 . VfdA (2.87)
A A A A
and hence, if the local slip velocity is constant across the pipe,

Vf - Vs = Vf - Vs (2.88)

For flow in a vertical pipe, the local slip velocity, Vf - V" may sometimes be
related to the hindered settling velocity of the particles in the fluid. Equation
(2.88) is used below to describe some particle/fluid flows.
Equation (2.80) illustrates a further general point. A given solids flow (Qs)
can occur at high Cy and relatively low Vs, or low Cy and relatively high Vs.
The former case corresponds to dense phase flow and the latter to lean phase
flow. It may be possible for a flow to switch between these two states. This is
the condition known as choking in pneumatic conveying (see Chapter 6), and
is usually to be avoided.
 Assemblages of particles 91

Returning to the general case, with arbitrary distributions of velocity and

concentration across the pipe section, the mean mixture velocity in the pipe is
Vm = (Qs + Qf)A (2.89)
From equations (2.70) and (2.75),
CYd = A VsCy/A Vm
= (V/Vm)CYt (2.90)
which serves to emphasise that, where there is mean slip between the phases
so that Vs =1= Vm, the in situ and delivered particle concentrations differ. The
velocity difference between the mixture and the solids, sometimes called the
'lag' or 'slip' velocity, is Vm - Vs and the 'lag ratio' is
(2.91)
In terms of A,
CYt = Cyd/(I- A) (2.92)

2.4.2 Momentum balance

For the flow shown schematically in Figure 2.14, the mass flows of the two
phases through the element dA are respectively QsCyVsdA and Qt<1 -
Cy) Vf dA. The total solids momentum flux across the section is therefore

(2.93)

while the fluid momentum flux is

JMf =f Qf(1 - Cy)VrdA (2.94)
A

and the total momentum flux across BB' is

(2.95)
Suppose that the total shear stress at the wall is 't, taken to be positive if it
acts on the flowing mixture in the direction opposite to the direction of flow
(see Figure 2.15). For a cylindrical pipe of diameter D, so that A = nD2/4,
the total force acting in this direction on the mixture in the volume element
from section 1 to section l' in Figure 2.15 is
nD2 nD2 dP
dF = nD'tdL + - - dL[ CytQs + (1 - Cyt)Qtlg sine + -- - dL
4 4 dL
(2.96)
92 Particles in fluids

~ D\
af~
as --------~---------------------------.
Figure 2.15 Flow of a particlelfluid mixture in a pipe - momentum.

where the first term in equation (2.96) arises from shear at the wall, the
second represents the axial component of the weight of the mixture in the
element in question and the last term arises from the pressure gradient along
the pipe. The linear momentum balance along the pipe is then

nD2 dP
g sin 8[ CytQs + (1 - Cyt)Qd - --
4 4 dL
(2.97)

The general significance of equation (2.97) can be approached by considering

the case of a homogeneous flow, in which Cy and Vs are constant across the
pipe section so that C y = C yt , Vs = Vs and Vt = Vt (see above). The
momentum flux terms can then be written

JM =JMs + JMf
= MsVs + MfVt (2.98)

where Ms and M f are the mass fluxes of the two phases,

Ms = f
A
QsCy VsdA = A QsCYt Vs
4
(2.99)

(2.100)
 Assemblages of particles 93

Unless gas or solids are added to or removed from the pipe at the section in
question, continuity requires Ms and M f to be constant along the pipe.
Therefore, for a homogeneous flow,

dIm d dVs dVf

- =-(JMs+JMf)=Ms - + M f - (2.101)
dL dL dL dL
From equations (2.97), (2.100) and (2.101)

dP 41: dVs dVfJ

-=- [ Ms- + Mf -
dL D dL dL
(2.102)

Equation (2.102) shows that the pressure gradient along a straight pipe can be
expressed as the sum of three terms. The first arises from the shear stress at
the pipe wall; in this macroscopic formulation, it does not matter to which
phase the stress is imparted. The second term arises from the weight of the
mixture in the pipe, and can therefore be regarded as a hydrostatic term. The
last term arises from the acceleration of the two flowing phases.
We can now consider some further simplifications. For a fully-developed
flow, whose parameters do not change along the pipe, the acceleration term is
zero. Then
dP 41:
- = - - - g sin 8[CytQs + (1 - Cyt)Qr] (2.103)
dL D
This equation is used below to describe fully-developed flows with particles
transported by a liquid (hydraulic conveying) or a gas (pneumatic conveying).
For a vertical flow, the hydrostatic term commonly dominates. It should be
noted that it depends upon the in situ solids concentration, C yt , rather than
the delivered concentration, C yd .
As a further simplification, consider a horizontal flow [8 = 0; sin 8 = 0].
Equation (2.102) then takes the simpler form

dP = _ 41: _ [MsdVS + MtdVtJ (2.104)

dL D dL dL

A particular case of interest is shown schematically in Figure 2.16: the solids

are introduced to the fluid flowing along the pipe. This feeding arrangement
is conventional in pneumatic conveying (see below), although in hydraulic
conveying the fluid and solids are more commonly introduced to the pipe
together. In this case, the solids enter the pipe with Vs = 0 and accelerate over
the 'entry length' to reach their equilibrium conveyed velocity. Over the entry
length,
94 Particles in fluids

(2.105)

For a lean phase flow, the pressure profile along the pipe therefore typically
takes the form shown in Figure 2.16. Upstream of the feed point, the pressure
gradient is small because 't arises solely from the shear on the gas and is
therefore relatively low. Downstream of the feed point, where the solids are
being accelerated, the term in Ms!1 Vs dominates and so the pressure drops
rapidly. As the solids reach their steady mean conveying velocity, the
acceleration term dies away and the pressure gradient is again dominated by
't, although the mean wall shear is larger than that for the gas alone and
transmitted primarily to the particles. By estimating the acceleration pressure
drop, !1PAccn in Figure 2.16, it may then be possible to estimate the mean
velocity of the conveyed solids.
In addition to pressure changes close to the feed point, this qualitative
aspect also explains why sharp bends have a strong effect on the pressure drop
in lean phase pneumatic conveying: the solids must accelerate, following the
bend, to a conveying velocity in a different axial direction.
In a vertically-upwards flow, the entrance effects are even larger. Low Vs in
the immediate vicinity of the feed point gives high solids concentration (CYt )
so that the hydrostatic term also contributes to the local high pressure
gradient.

L
i, -------------------_

,
,, L

~ __ 1Ms
---L-------------------------------~
Mf (Ms+ Mf )

Figure 2.16 Consequences of solids addition to a flowing fluid in a long pipe.

 Assemblages of particles 95

2.4.3 Mechanical energy balance

Referring again to Figure 2.16, the flux of kinetic energy through section BB is I

IKE = f [CvQsVsCV;/2) + (1 -
A
Cy)QfVf(Vr/2)]dA

= -Qs f'CyVsdA + -Qf f

'3 "3
(1 - Cy)VfdA (2.106)
2 A 2 A

For a steady flow, the gradient of potential energy in the direction of flow is
g sin 8(Ms + Mf). The flowing mixture does 'displacement work' against the
pressure gradient given by (Qs + QD (dPldL). Putting these terms together,
the rate of mechanical energy dissipation from the flowing mixture per unit
length of pipe is

dP
- (Qs + Qf)- - g sin 8(Ms + Mf )
dL

(2.107)

To interpret this energy dissipation, we associate it with an irreversible (or ir-

recoverable) pressure gradient,

dP1rr dP (Ms + MD
- dL = - dL - g sin 8 (Qs + Qf)

1 -d [Qs
- f'CyVsdA + -Qf f (1 -
'3 Cy)VfdA
"3 ] (2.108)
(Qs + Qf) dL 2 A 2 A

To illustrate the significance of equation (2.108), we consider again the 'flat

profile' case (i.e. Cv = C yt , Vs = V" and V f = Vf). The kinetic energy flux
term now becomes

IKE = A QsCYt Vs!2

-3
+ AQf(1 - -3
C yt) Vi
-3 -3
IKE = AQsCvt Vs/2 + AQf(1 - Cyt) V f
= Ms V;/2 + MfVr/2 (2.109)

Equation (2.108) then simplifies to

96 Particles in fluids

-dP1rr -dP (Ms + Mf )

~ = dL - g sin 8 (Qs + Qa

[ - dVs
-
1
MsVs- + MfV- f dVfJ
- (2.110)
(Qs + Qf) dL dL
For a fully-developed flow, with Vs and Vf constant along the pipe,
-dP1rr dP (Ms + Mf )
(2.111)
dL = - dL + g sin 8 (Qs + Qf)
Inserting the expression for the pressure gradient from equation (2.103),

(2.112)

The first term on the right of equation (2.112) corresponds to the frictional
(and therefore dissipative) wall shear stress, and is in the form familiar for
single-phase flow. The other term can best be investigated by noting that
[gsCYt + gf(1 - Cyt) ] is the mean in situ mixture density, gMt, whereas
the term (Ms + Mf)/(Qs + Qf) is the mean delivered mixture density g. In
these terms, equation (2.105) becomes
-dP~ ~ _
---= - + g sin 8[gMt - g] (2.113)
dL D
Note also that
gMt - g = (gs - gf)( C Vt - Cyd)
= (gs - gf)Cyd[A/(l - A)] (2.114)
from equation (2.92).
For an upflowing mixture, in the common case where gs > gh the
direction of slip is such that A is positive (see equation (2.91)): therefore the
density difference term in equation (2.112) is positive, representing a
contribution to mechanical energy dissipation. If the mixture is in downflow,
A is negative (see equation (2.91)) so that Vs > Vm. However sin 8 is now
negative, so that the term is again positive. Thus in both cases, slip of the
solids relative to the fluid leads to dissipation of mechanical energy - as
expected, given that particle/fluid drag is a dissipative frictional process.

2.5 NOMENCLATURE
A Area (m 2 )
C Cunningham slip correction factor (equation (2.43)
 Nomenclature 97

CD Drag coefficient (equation (2.2»

c resistance to motion (m) (equation (2.15))
DAB Diffusion coefficient (m2/s)
d Particle diameter (m)
d* Dimensionless particle diameter (equation (2.12))
dy Volume equivalent diameter (m)
E Aspect ratio
F Force (N)
FD Hydrodynamic drag (N)
FL Lift (N)
f Frequency of oscillation (Hz)
g Acceleration due to gravity (m S-2)
I* Dimensionless moment of inertia (equation (2.39))
J Momentum flux (N)
Kn Knudsen number (equation 2.39))
kB Boltzmann's constant (JK-1)
L, I Length (m)
m Mass (kg)
n Richardson-Zaki index in Table 2.4
P Total pressure (Pa)
p Pressure (Pa)
Q Volumetric flow rate (m 3 S-1)
Re Reynolds number (equation (2.3))
S Surface area (m2)
Sr Strouhal number (equation (2.37))
St Stokes number (equation (2.59))
T Temperature (K)
t Time (s)
U Gas velocity (m S-l)
V Particle velocity (m S-l)
Vt Terminal particle velocity (m S-l)
Vi Dimensionless terminal particle velocity (m S-l)
(equation (2.14))
v Volume (m 3 )

Greek letters
a Shape coefficient in section (2.1.3.2.2)
f3 Angle between particle trajectory and vertical (Figure 2.7)
E Void fraction
Angle between axis of symmetry and vertical (Figure 2.7)
Mean free path (m)
Viscosity (Pa s)
Fluid density (kg m-3 )
98 Particles in fluids

Qp Particle density (kg m-3 )

tV Sphericity
L Relaxation time (s) in equation (2.48); shear stress (Pa)

REFERENCES
Batchelor, G.K. (1967) An Introduction to Fluid Dynamics, Cambridge University
Press, Cambridge.
Beard, K.V. (1976) 1. Atmos. Sci., 33, 851-64.
Clift, R., Grace, 1.R. and Weber, M.E. (1978) Bubbles, Drops and Particles,
Academic Press, New York.
Clift, R., Ghadiri, M. and Thambimuthn, K.V. (1981) in Progress in Filtration and
Separation, Vol. 2 (ed. R.l. Wakeman), Elsevier, Amsterdam, p. 75.
Davies, C.N. (1945) Proc. Phys. Soc, 57, 259-70.
Dullien, F.A.L. (1992) Porous Media, 2nd edn., Academic Press, London.
Ergun, S. (1952) Chem. Eng. Prog. 48, 89.
Ghadiri, M., Seville, 1.P.K., Raper, 1.A. and Clift, R. (1986) Estimation of the
Hydrodynamic Diameter of Fine Particles from Equivalent Area and Volume
Diameters, Proc. 1st World Congress of Particle Technology, Nurnberg, 36--54.
Govier, G.W. and Aziz, K. (1972) The Flow of Complex Mixtures in Pipes, Van
Nostrand Reinhold, New York.
Grace, 1.R. (1986) Can. 1. Chem. Eng. 64,353.
Happel, 1. and Brenner, H. (1983) Low Reynolds Numbers Hydrodynamics, Martinus
Nijhoff, The Hague.
Heiss, 1.F. and Coull, J. (1952) Chem. Eng. Prog. 48, 133.
Heywood, H. (1938) Proc. Inst. Mech. Engrs. 140,257.
Hill, R. and Power, G. (1956) Quart. 1. Mech. Appl. Math. 9, 313.
Hinds, W.C. (1982) Aerosol Technology, Wiley-Interscience, New York.
Kay, 1.M. and Nedderman, R.M. (1985) Fluid Mechanics and Transfer Processes,
Cambridge University Press, Cambridge.
Knudsen, M. and Weber, S. (1911) Ann. Phys, 36, 981-94.
Mayhew, Y.R. and Rogers, G.F.C. (1972) Thermodynamic and Transport Properties
of Fluids, 2nd edn., Basil Blackwell, Oxford.
Potter, R.L., Pruppacher, H.R. and Hamielec, AE. (1973) 1. Atmos. Sci. 30, 125.
Soo, S.L. (1989) Particulates and Continuum, Hemisphere Publishing, New York.
Stringham, G.E., Simons, D.B. and Guy, H.P. (1969) U.S. Geological Survey,
Professional Paper 562-C.
Turton, R. and Levenspiel, O. (1986) Powder Technol. 47, 83.
Wallis, G.B. (1969) One-dimensional Two-phase Flow, McGraw-Hill, New York.
Willmarth, W.W., Hawk, N.E. and Harvey, R.L. (1964) Phys. Fluids 7,197.
 3
Particle mechanics

3.1 INTERPARTICLE FORCES

3.1 .1 van der Waals forces
It is not the purpose of this section to review the various intermolecular forces
which can occur, but to explain the way in which they can give rise to forces
between particles, which are in general many orders of magnitude larger than
the molecules of which they are made up. For a more extensive and very
readable review of the subject, the reader is referred to Intermolecular and
Surface Forces by Israelachvili (1991). A useful short review of adhesion of
solids is given by Tabor (1987).
All intermolecular forces are essentially electrostatic in origin, although
they manifest themselves in such different ways that subclassification has
become common. Intermolecular forces may arise from the following,
interactions:
• covalent • charge-non-polar
• charge-charge (Coulomb) • dipole-non-polar
• charge-dipole • non-polar-non-polar (dispersion
• dipole-dipole forces)
• hydrogen bonding
Dispersion forces arise from the local polarisations produced in molecules by
the random fluctuation of electrons. They are therefore due to attractions
between transient induced dipoles, and are the only interactions of those
listed above which are always present; the other interactions mayor may not
occur, according to the nature of the materials. The 'van der Waals force' is
taken to include dipole-dipole, dipole-non-polar and dispersion forces.
In its simplest form, the interaction between two molecules is represented
by a 'pair potential', which can be differentiated to give the interaction force
(Figure 3.1). For larger values of the molecular separation, r, the interaction
force is positive, due to the van der Waals attraction between the molecules.
At very small separations, however, the electron clouds associated with the
molecules overlap, resulting in a very strong repulsion. The total inter-
molecular pair potential is obtained by summing the attractive and repulsive

J. Seville et al., Processing of Particulate Solids

© Chapman & Hall 1997
100 Particle mechanics

>-
0\
'-
QJ
C
QJ

d
....c
.l'! \ Repulsion
o \
CL
\
\. Separation r
0.111111111111111111111 1I11I1~III1IIIII1III1I1III1II1UIIIJIIIIIIIIIIII 111111111111111111111

_ro_7
I .
I Attraction
I

Figure 3.1 Potential energy as a function of the separation between two atoms
attracted by van der Waals forces. Similar curves apply to all types of interactions
though they may differ in scale and shape. (ro indicates equilibrium separation.)
(Tabor, 1987.)

potentials, as shown in Figure 3.1. For our purposes, however, it is sufficient

to neglect the repulsive potential, since the molecules in two particles, for
example, are usually very far apart by comparison with the equilibrium
separation shown in Figure 3.1.
The non-retarded* van der Waals pair potential for two molecules in
vacuum can then be represented as
W(r) = -C/rn (3.1)
where C is a constant, r is the distance between the molecules and n is usually
taken as 6. The resulting force between the molecules is simply equal to
dW(r) Cn
F(r) = - -d = n+1
- (3.2)
r r
Thus, the van der Waals force between molecules decays so rapidly with
distance that it is essentially a contact force only. In contrast, we shall show
that for 'large' bodies (i.e. much larger than the molecular scale):
(i) the net interaction energy is proportional to the radius of the body;
(ii) the net interaction energy is large at contact but still appreciable at fairly
large separations (compared with the molecular dimension).

*See section 3.1.1.6 for an explanation of retardation effects.

 Interparticle forces 107

There are essentially two theories for predicting the van der Waals
interaction between large bodies: the microscopic (Hamaker) and the
macroscopic (Lifshitz) theories. The Hamaker theory is easier to grasp and its
main results are summarised here; the Lifshitz theory is more rigorous but the
form of its predictions is very similar to those of the cruder Hamaker theory.

3.1.1.1 Interaction between a molecule and a plane surface. Consider an

isolated molecule at a distance a from a plane wall (Figure 3.2). The Hamaker
theory assumes that the interaction energies between the isolated molecule
and all the molecules in the large body are additive and non-interacting. Thus
the net interaction energy can be found by integrating the molecular
interactions over the entire body. Consider a circular ring within the large
body, of cross-section dx dz and radius x. The interaction energy between the
isolated molecule and any molecule in the ring is given by
(3.3)
The number of molecules in the ring is 2n X dx dz Q where Q is the number
density of molecules in the solid. Thus the net interaction energy is

(3.4)

• z

Figure 3.2 Interaction between a molecule and a plane surface (after Israelachvili,
1991 ).
702 Particle mechanics

Note that
X=OO _ _ l 1
] (3.5)
n-2
x=o
Thus,
-2JtCg
W(a) = for n > 3 (3.6)
(n - 2)(n - 3)a n - 3

Therefore, for n = 6,
W(a) = -JtCg/6a3 (3.7)
and
-dW(a)
F(a) = - - = -JtCg/2a4 (3.8)
da

3.1.1.2 Interaction between a sphere and a plane. Now consider a large

sphere of radius R at a distance a from a plane surface (Figure 3.3).
By simple geometry
x 2 = (2R - z)z (3.9)
The volume of a thin circular element of the sphere, of area Jtx 2 and
thickness dz is
Jtx 2 dz = Jt(2R - z)z dz (3.10)

dz

a
z=R

2R-z
a+z

Figure 3.3 Interaction between a sphere and a plane (after Israelachvili, 1991).
 Interparticle forces 103

so that the number of molecules in this section is n(2R - z)z dz Q. All these
molecules are at a distance (a + z) from the planar surface. Therefore, from
equation (3.6),

=f
z= 2R - 2nCQ nQ(2R - z)z
W(a)
z = 0 (n - 2)(n - 3)
. (a + Z)"-3
dz

-2n 2Cg 2 2R(2R - z)z dz

(3.11)
= (n - 2)(n - 3)10 (a + Z)"-3
For a « R, we may write (notice the limits)

W(a) =-
2n 2CQ2 f 00 2Rz dz
-- (3.12)
(n - 2)(n - 3) 0 (a + Zy-3
Therefore,
n 2CQ2R 00 z
(3.13)
W(a) =- (n _ 2)(n - 3)1'0 (a + Z)"-3 dz
Note that, by substitution y = (a + zt-3
00 z 1 1
(3.14)
fo (a + Zy-3 dz = (n - 4)(n - 5) . a"-s

Therefore,
-4n 2 Cg 2 R
W(a) = - - - - - - - - - - - - cS
(n - 2)(n - 3)(n - 4)(n - 5)a n -
which for n = 6 becomes
W(a) = -n2CQ2R/6a
=-AR/6a (3.15)
where A is the Hamaker constant, n 2 Cg 2 •
For a» R, however, equation (3.11) becomes

2n 2CQ2 j2R(2R - z)z

W(a) =- dz (3.16)
(n - 2)(n - 3) 0 an - 3

2nCQ( 4nR 3 (/3)

(3.17)
(n - 2)(n - 3)a n- 3

which is essentially the same as equation (3.6), since 4nR 3 g/3 is simply the
104 Particle mechanics

number of molecules in the sphere, i.e. at large separations the sphere now
behaves like a single point body located at its centre.

3.1.1.3 Other geometries. It can be shown that for two spheres of equal
radii, R, at small separation, a (<< R), the interaction energy is half of that in
equation (3.15), while at large separations W(a) is proportional to l/an, as for
two molecules.
In practice, the results at small separations are of most interest. In general,
for two unequal spheres of different materials,
-rr?0J1Q2R
W(a)=---- (3.18)
6a
where R = RIR2/(Rl + R 2) and R1 and ~h are the radius and molecular
number density, respectively, for sphere 1. Other interaction laws for
common geometries are given in Figure 3.4.

3. 1. 1.4 Lifshitz theory. The Hamaker theory is deficient in a number of

respects, chief of which is the assumption of 'pairwise additivity'. If one
considers the interaction between two atoms or molecules in the presence of
other atoms, the instantaneous field due to atom 1 not only induces a dipole
in atom 2 but also in all other atoms, which in turn act on atom 2. The more
sophisticated Lifshitz theory avoids the problem of additivity by ignoring the
atomic structure of the bodies and treating them as continua. The interaction
is calculated in terms of bulk properties such as the dielectric constant and the
refractive index. However, the results of the theory are of the same form as
Hamaker's; effectively, the Lifshitz theory provides a rigorous method for
calculation of the Hamaker constant. Some values for the Hamaker constant
in various media are given in Table 3.1.
Note that the range of values for the Hamaker constant is comparatively
small. For most solids interacting across a vacuum (or air) A - 4-40 X
10-20 J. The value of A is much reduced for interaction across water (although
other forms of interaction can occur in liquids, such as electrical double-layer
effects, which may greatly exceed the van der Waals interaction).

3. 1. 1. 5
Combining relations. Approximate relationships are available
which enable unknown Hamaker constants to be obtained from known ones.
If we define A 132 as the Hamaker constant for media 1 and 2 interacting
across medium 3, then
A132 = ± V(A l3l A 232 ) (3.19)
or, if medium 3 is a vacuum (or air)
A12 = V(A l1 A 22 ) (3.20)
 Interparticle forces 705

Molecule-surface Sphere-surface

D p

w = -ltCp I 60' W =-AR/60

Two spheres Two cylinders

AL (R'R' )112
W = - 12.,/20'12 R, + R,

Two crossed cylinders Two flat surfaces

R,

:
W = -A..JR:R, 160 W = -A I 12ltO' per unit area

Figure 3.4 Non-retarded van der Waals interaction free energies between bodies of
different geometries calculated on the basis of pairwise additivity. The Hamaker
constant A is defined by A = n 2 C!.hQ2, where Ql and Q2 are the number of atoms
per unit volume in the two bodies and C is the coefficient in the atom-atom pair
potential. (Israelachvili, 1991.)
106 Particle mechanics

Table 3.1 Hamaker constants for two media (1 and 2) acting across another medium
(3) (after Israelachvili, 1991)

Interacting media Hamaker constant, A

(10-20 J)

1 3 2 Calculated Measured

Quartz Vacuum (air) Quartz 6.5 5-6

Mica Vacuum (air) Mica 10 13.5
Metals Vacuum (air) Metals 30-50
PTFE Vacuum (air) PTFE 3.8
Quartz Water Quartz 8.83
Mica Water Mica 2.0 2.2
PTFE Water PTFE 0.33
Air Water Air 3.7
Quartz Water Air -1.0

Also,
A131 = A313 = (VAu - vA33)2 (3.21)
Combining equations (3.21) and (3.19),
Al32 = (vAll - VA 33 ) (VA22 - VA 33 ) (3.22)
A132 is negative if All < A33 < A22 or All > A33 > A 22 . Thus it is possible to
have a repulsive van der Waals force, as in the case of the quartz/water/air
system. These relationships apply only when dispersion forces dominate; in
particular, they are not reliable where water is concerned. Nevertheless,
some general rules may be derived:
(i) the force between any two bodies in vacuum (or air) is always attractive;
(ii) the force between two identical bodies in a medium is always attractive,
while that between different bodies in a medium can be attractive or
repulsive;
(iii) the Hamaker constant for two identical media acting across a second
remains unchanged if the media are interchanged, i.e. two water droplets
in air will attract with the same force as two air bubbles in water (all
other things being equal).

3.7.7.6 Retardation effects. The van der Waals force between two atoms
arises because of the temporary dipole set up in atom 2 in response to a field
produced by atom 1. If the time taken for the electric field of the first atom to
 Interparticle forces 107

reach the second and return is comparable with the period of the fluctuating
dipole itself, the force arising from the interaction will be less than it would
otherwise be, i.e. it will be retarded. Hence the accurate value of the
Hamaker constant is reduced at larger separations, and in exceptional cases
can even change sign. This is accounted for by the full Lifshitz theory, but
obviously means that the Hamaker 'constant' is not really a constant at all,
but is itself a function of the surface separation.

3.1.1.7 Interaction energies and forces: the Oerjaguin approximation. It

is useful to be able to relate the force between two curved surfaces to the
interaction energy of two plane surfaces (since the former is more useful and
the latter usually easier to obtain theoretically). This can be done using
Derjaguin's approximation (Israelachvili, 1991), as follows: For two flat
surfaces, from Figure 3.4
W(a) = -A/12Jta2 per unit area (3.23)
while for a sphere of radius R near a plane surface (equation 3.15),
W(a) = -AR/6a (3.24)
so that the force between the sphere and the plane is given by
-dW(a) AR
F(a) = - - = - (3.25)
da 6a 2
Hence,
F(a)sphere-plane = 2JtRW(a)plane-plane (3.26)
More generally,

F(a)curved surfaces = 2Jt ( RIR2 ) W(a)plane-plane (3.27)

Rl + R2
where Rl and R2 are the radii of the two contact surfaces. Derjaguin's
approximation applies for any type of force law, whether attractive, repulsive
or oscillatory. Note that
(i) for R2 » Rb F(a) = 2JtR 1 W(a), i.e. a sphere near a plane surface;
(ii) for R = Rl = R 2, F(a) = JtRW(a), i.e. two equal spheres;
(iii) for spheres in contact, taking W(a) = 2y, where y is the surface energy
per unit area ('surface tension' for a liquid), the force of adhesion, Fad, is
given by
(3.28)
y may be regarded as half the van der Waals energy needed to separate
two surfaces from contact to infinity.
708 Particle mechanics

3. ,. 1.8 Practical calculations. The expressions listed above give the van
der Waals interaction forces up to the point of contact between bodies, but
what is meant by 'contact' is unclear since if a is allowed to go to zero, the
interparticle forces become infinite. Many older references take a to be 4 X
10-10 m, which is close to the average interatomic distance for many common
solids. However, Israelachvili recommends the use of a 'cut-off distance', ao,
substantially less than the interatomic distance. He shows that a 'separation'
of 1.65 X 10-10 m gives values for surface energies within ±20% of
experimental values. From Figure 3.4 and equation (3.28),
y=A/24Jta5 (3.29)
Substituting for ao,
A = 2.1 X 1O-18y (3.30)
Real particle surfaces are, of course, rough and the radii of the asperities may
determine the van der Waals force. Fine particles may act as spacers between
the larger particles and therefore reduce the forces between them, or may fill
in gaps and thus increase the interaction force, depending on their relative
SIze.
Real particles are not rigid and will deform elastically and/or plastically at
the contact point, even under zero external load. The effect of elastic
deformation is considered in section 3.2. For plastic deformation, the total
interaction force must include a term for the extended contact area. From
Figure 3.5, for two spheres, Visser (1989) recommends

T
2r I
I
+

_l~
I I

.-.~1 ~_a -

Figure 3.5 Two spheres, deformed by contact.

 Interparticle forces 109

AR A
F(a) =- + - . n,2 (3.31)
12a2 6na 3
where r is the radius of the extended contact area.
From the calculations leading to equation (3.15) it is apparent that the total
interaction energy for two bodies almost in contact is dominated by the
interactions between the surfaces of the bodies, the so-called 'screening'
effect. In fact, the interaction energy is determined almost entirely by the two
surface layers of depth equal to the surface separation. Thus thin layers of, for
example, oxide or adsorbed gases may have a strong effect on the observed
interparticle force.

3.1.2 Liquid bridge forces

If the surface of a particle has a film of mobile liquid then at points of contact
with other particles and surfaces 'liquid bridges' will form. It will also be the
case that if the partial pressure of some suitable vapour in the surrounding gas
is sufficiently high, condensation will occur at points of contact. Again, the
result is the formation of a liquid bridge. In this section, the interparticle
forces due to the presence of liquid bridges are derived, and some
consideration is given to how these depend on the contact geometry.
The forces acting between two particles due to a liquid bridge between
them may be both capillary and viscous in nature, i.e. static and dynamic,
respectively. In this section only the lowest state of saturation, the pendular
state (Figure 3.6) is considered. Both capillary and viscous forces are
calculated and their magnitudes are compared under various conditions of
contact geometry and separation rate.

PENDULAR FUNICULAR

CAPILLARY

Figure 3.6 Distribution of liquids in agglomerates (after Capes, 1980).

110 Particle mechanics

3.1.2.1 The capillary force between two spheres. In general, the capillary
force between two particles is the sum of three terms: the axial component of
the surface tension force at the solid/liquid/gas interface, the force due to the
reduced hydrostatic pressure in the bridge itself, and the buoyancy force due
to the partial immersion of each particle. Princen (1968) and Picknett (1969)
demonstrated for spheres that both the buoyancy term and the distortion of
the shape of the bridge due to gravity can be neglected if the particle size is
small, say < 1 mm.
Consider two identical spheres of diameter 2R, joined by a liquid bridge of
half-angle p and separated by a distance 2a (Figure 3.7). In the arguments
which follow, zero contact angle (perfect wetting) will be assumed, although
the derivations can be readily extended to non-zero contact angles (Coughlin
et aI., 1982). According to the Laplace equation (Shaw, 1992), the reduced
hydrostatic pressure within the bridge, !1P, is given by

(3.32)

where rl and r2 are the principal radii of curvatur~f the bridge, as shown in
Figure 3.7, and y is the liquid surface tension. This equation requires that the
liquid surface must have constant total curvature, (ri1- r2:I), so that the surface
of the bridge in the plane of the paper is not a circular arc.
However, Fisher (1926) used a toroidal approximation for the shape of the
liquid bridge (i.e. assuming that the curvature of the bridge in the plane of the
paper is constant) and this has been shown theoretically (Orr et al., 1975) and
experimentally (Cross and Picknett, 1963; Mason and Clark, 1965) to lead to
an accurate estimate of bridge strength provided that the particles are smooth
and truly spherical. Consider first the case of zero sphere separation (a = 0),

Figure 3.7 Liquid bridge between two equal spheres (perfect wetting). a = half
particle separation.
 Interparticle forces 111

SPHERICAL
PARTICLE

~--4-----------~ "
r,
Figure 3.8 Liquid bridge joining two spheres (only one shown; zero contact angle).

which is the situation examined by Fisher (1926). The axial surface tension
force acting at the dividing plane is given by
(3.33)
while the hydrostatic force evaluated from the reduced pressure in the liquid
bridge at the dividing plane is

(3.34)

The total capillary force is then given by

(3.35)

By trigonometry,
r1 = R(secj3 - 1) (3.36)
r2 = R(1 + tanj3 - secj3) (3.37)
Putting t = tan (j312) , and substituting for r1 and r2 in (3.35) the following
simple result is obtained

2nRy
F =----- (3.38)
c 1 + tan(j312)

This result differs slightly from that obtained by Adams and Perchard (1985)
who argued that it is correct to evaluate the surface tension and hydrostatic
forces at the surface of the sphere, so that equations (3.33) and (3.34) become
(3.39)
112 Particle mechanics

and

(3.40)

i.e. the reduced hydrostatic pressure acts over the projected wetted area. The
main effect of this modification is to alter the proportion of the total force
which is attributed to each origin; the sum F1 + F2 remains almost the same as
that given by equation (3.38), as it should if the spheres are at equilibrium.
Adding (3.39) and (3.40), and substituting for '1 and '2 as before, we obtain
the result

Fe = 2nRy [
2f! - t + 1J (3.41)
(1 + f!)2

where t = tan(!3/2). This gives values for Fe within a few percent of those given
by equation (3.38) for values of j3 within the range of interest; the discrepancy
is attributable to the error in the toroidal approximation.
The variation of interparticle force with increase in the bridge half-angle j3
is shown in Figure 3.9, for the following values of the variables:
2R = 922!-tm
y = 0.072 N m- 1 (water at 25°C)
a = 0
Values of the surface tension and capillary forces and the total force are given
according to both approaches. It is clear from Figure 3.9 that the total force is
virtually independent of j3 at small values of j3; as predicted by equation
(3.38), Fe tends to a limiting value of 2nRy. At larger values of j3, Fe reduces
slightly. The range of interest of j3 extends· only to about 40°, since the
coalescence limits for liquid bridges between spheres are 30° and 45° for close
packed and cubic arrangements, respectively (Coughlin et al., 1982). It is
apparent that for values of j3 below about 10° the contribution of the surface
tension force is negligible.
Equation (3.38) predicts an increase in the magnitude of Fe for a decrease
in the size of the liquid bridge (represented by j3) until a maximum value is
reached at zero liquid content. Although careful laboratory experiments,
such as those of Cross and Picknett (1963) and Mason and Clark (1965),
reproduced this trend down to very low liquid volumes, this behaviour is
nevertheless the opposite of what one would intuitively expect, i.e. wetter
powders commonly appear stronger than drier ones (until the capillary state is
reached, at which point the strength is reduced). Pietsch (1968) attempted to
resolve this apparently paradoxical conflict between theory and experimental
result by suggesting that all real contacts are rough and that an effective
 Interparticle forces 113

- - - Fisher coalescence
limits
Adams & Perchard
300 45°
~
z:
5x104·r---~r-~~~-r,,~----~---r~TT'
UJ
~
a
u..
UJ r-~~-==---=_==:--_:::::::-.!TO~TAL FORCE

---
-I
I..J
t= HYDROSTATIC
a::
Cf.
a:: 10-4 FORCE
UJ 2R = 922}1m
I-
z: If = 0·072 N/m
5x 10-5 Q= 0 /
/
/
/ SURFACE
/ TENSION
FORCE

10-5 L...-_--'-_--1--L-~L_L_..L-L-'---<--____'__ __L__'_~

1 5 10 so
BRIDGE HALF-ANGLE ~ (0)

Figure 3.9 Variation of hydrostatic, surface tension and total capillary forces with
liquid bridge half-angle, jJ.

sphere separation, a, should be included in the theory, as shown in Figure

3.10.
The effect of finite separation distances can be considered using Adams and
Perchard's (1985) approach, rewriting equations (3.36) and (3.37) as
rl = R[(1 + a')secf3 - 1] (3.42)
and
r2 = R[1 + (1 + a')tanj3 - (1 + a')secf3] (3.43)
where a' = aiR. Figure 3.11 shows the variation in Fc with f3 for half-
separations, a, of 0, 1 f!m and 10 f!m, illustrating the dramatic difference that
even small separations can make to the interparticle force if f3 is small. The
predicted total capillary force now shows a maximum; the value of f3 at which
this maximum occurs increases as the sphere separation increases.

3.1.2.2 The capillary force for cone and plate contact. Most particles of
practical interest cannot be considered spherical, especially when viewed on a
114 Particle mechanics

Figure 3.10 Schematic representation of a liquid bridge between two particles with
surface asperities (after Pietsch, 1968).

a./R= 0

F
2TTRa'

0·1

5 10 50
BRIDGE HALF-ANGLE 13 (")

Figure 3.11 The dependence of normalised adhesion force, F12Jtry, between two
equal spheres on the bridge half-angle, j3, as a function of dimensionless separation,
aiR.

microscopic scale. It is therefore of interest to consider the forces arising from

other contact geometries. Figure 3.12 shows the contact between a cone, of
half-angle a and wetted length f, making contact with a plane. (Adams and
Edmondson (1987) consider the more complicated case of contact between
two non-spherical but axisymmetric particles.) By a similar approach to that
employed for spherical particles, neglecting buoyancy, taking zero contact
 Interparticle forces 715

CONICAL

r LIQUID BRIDGE

Figure 3.12 Liquid bridge at the contact between a cone and a plane.

angle and assuming the surface of the liquid describes an arc of a circle in the
plane of the paper as shown, the force acting perpendicular to the plane is
given by
Fe = 2n:yy cosa + I1Pn: y 2 (3.44)
where

(3.45)

and y is the radius of the cone at the three-phase line.

Now
y = Isina (3.46)
and

cosa ]
r=! [ (3.47)
1 + sina

Substituting in equation (3.44) we obtain

Fe = n:Y/tana[l + cos2 a + sina - cosa] (3.48)
which is the result obtained by Coughlin et al. (1982). For this geometry,
therefore, the capillary force is simply proportional to the wetted length and a
geometric factor depending only on the cone half-angle. In contrast to the
sphere-sphere case, the interparticle force increases monotonically with
increase in the volume of the liquid bridge; for a wetted length lequal to the
116 Particle mechanics

radius of the sphere previously considered, values of Fe will be comparable in

magnitude.
The capillary forces so far discussed are not the breakage forces; the
stability of liquid bridges and the critical separation limits are discussed by
Lian et al. (1993) while the energy required to extend a liquid bridge to the
rupture point is considered by Simons et al. (1994). However equations (3.38)
and (3.48) do enable maximum interparticle forces for quasi-static separation
to be calculated, which might be expected to be important in the behaviour of
wet granular materials. If separation rates are high, however, viscous forces
can become appreciable, as shown below.

3.7.2.3 Dynamic forces. The viscous force, F y , between two spheres of

radius R, and separation 2a, being separated at a rate 2V (where V = da/dt)
can be determined as a special case of the Reynolds lubrication equation
(Cameron, 1981; Adams and Edmondson, 1987),
(3.49)
where fl is the viscosity of the (Newtonian) liquid.
The ratio of the viscous force, F y , to the static force previously considered,
Fe, has been calculated, as a function of the separation rate, for the following
variable values:
f3 = 5° and 30°
a = 1 flm and 10 flm
2R = 922 flm
y = 0.072 N m- I (water at 25°C)
fl = 10-3 kg m S-1 (water at 25°C)
The results are presented in Figure 3.13, showing that in this case viscous
forces can become significant by comparison with capillary forces for
separation rates above about 1 cm S-1 if the particle separation is small.
However, for larger separations and larger bridge angles, viscous forces only
become comparable with capillary forces for separation rates above about
1m S-1.

3.7.2.4 Condensation and evaporation of liquid bridges. The vapour

pressure (P) above a curved liquid/gas interface may be more or less than that
above a plane surface (Po) according to whether the surface is concave or
convex (Shaw, 1992). For example, the vapour pressure above a surface
which is concave, with respect to the gas, and of radius r is given by
(3.50)
where y is the liquid surface tension and VL is its molar volume. This is a
special case of the Kelvin equation, which when applied to the liquid bridge
shown in Figure 3.7, takes the form
 Interparticle forces 177

W
\..J
IX
o
1L

>-
IX
~
-J
0:::
«
\..J

10~~~~~~~~~~~~~~--~~~~
(}OO1 0·01 (}1
SEPARATION RATE (m/s)

Figure 3.13 Ratio of viscous to capillary force, FvfFc> as a function of separation rate.

-YVL
In(p/po)=-- (1 1)
--- =. ·k (3.51)
RT '1 '2
where the term in brackets is the total curvature of the bridge surface, k.
Equation (3.51) expresses the equilibrium between the liquid in the bridge
and that in the surrounding vapour; in other words, it provides a link between
the vapour pressure in the surrounding gas and the extent of capillary
condensation at the contact points between the particles in a powder.
Consider two spherical particles as shown in Figure 3.7, at contact (a ---c> 0).
By simple geometry,
?z + 2'1'2 - 2'1R = 0 (3.52)
Substituting for '2 in terms of the total curvature k, we have
rj(2k + 2k2 R) - '1(3 + 4kR) + 2R = 0 (3.53)
which is a quadratic in '1 with the following physically realistic solution:
118 Particle mechanics

3 + 4kR - (9 + SkR)'/2
'1 = 4k(1 + kR)
(3.54)

This equation enables us to determine the influence of vapour pressure on

bridge dimensions and interparticle force. For a given value of p/Po, k can be
determined from equation (3.51), from which'l and'2 can also be obtained,
via equation (3.54), and the interparticle force, Fe, from equation (3.35).
Results of such computations, for spheres of 0.1 and 10 f1m joined by water
bridges, are shown in Figures 3.14(a) and 3.14(b). Equation (3.51) predicts a
limit to the size of an equilibrium bridge. As P tends to Po, from equation
(3.52),
'1 ='2 = 2R/3 (3.55)

(a) Particle radius = 0.1 J.l

0
8s=0
5.0
Total force
2ynr - - 4.5
4.0

-----
3.5
4ynr
~ 3.0 3
00

-
<::>
~
0.>
2.5
....0 2.0
<S
0.>
U l.5
.~

e-
~ 1.0
Surface tension force
0.5
0
~--------- ---------------
0 10 20 30 40 50 60 70 80 90 100
Relative pressure, p/Po(%)

Figure 3.14 (a) Surface tension, pressure and total forces as a function of (P/Po);
contact angle = 0° (Coughlin et at., 1982).
 Interparticle forces 179

(b) Particle radius = 10 f.!

es=0°
5.0-
Total force
2yrrr -4.5

i --
Pressure force
4.0-
____ 3.5-
2S 4yrrr
~ 3.0- 3
;<
~
.....
2.5-
<8
~ 2.0-
'i
e- 1.5-
OJ

:E 1.0 -

0.5-
Surface tension force
)
0 I I I I I I I I I

0 10 20 30 40 50 60 70 80 90 100

Relative pressure, p/Po(%)

Figure 3.14 (b) Surface tension, pressure and total forces as a function of (P/Po);
contact angle = 0° (Coughlin et ai., 1982).

At this point, from equation (3.35)

Fe = 4rryR/3 (3.56)
and
f3 = cos-1 0.6 = 53.1° (3.57)
(Note that equations (3.52) to (3.57) all apply for zero contact angle only, but
may easily be adapted for non-zero angles.) In practice, f3 would probably not
be able to reach a value as large as that predicted in equation (3.57) because
of coalescence with neighbouring bridges, as noted earlier.
Note that for spheres, Figure 3.14 shows a decrease in interparticle force for
increasing humidity; a similar development for cone-on-plate geometry (see
Coughlin et al., 1982) shows quite different trends, with an increase in
interparticle force with increasing humidity (Figure 3.15). The contrasting
behaviour of 'smooth' (spherical) and 'rough' (conical) contacts is again
apparent.
120 Particle mechanics

a = Cone half-angle

130
12
llO
100
a =80°
90
*...,
....
oJ 80
~
<E
.,
"0 70
<)

'"
"0 60-
~
50
40-

30
20
10

0
0 10 20 30 40 50 60 70 80 90 100
Relative pressure, p/Po(%)

Figure 3.15 Cone-on-plate contact. Effect of cone half-angle (a) on the interparticle
adhesion force, F*, for e = 0° (Coughlin et al., 1982). (F* = Jt"?vdRT).

3.1.3 Electrostatic forces

Most aerosol particles carry some electric charge and very large charges may
accumulate either accidentally, in working of stone for example, or
deliberately, as in electrostatic precipitation. For coarser particles, electro-
static changes are frequently generated in transport and handling and if not
properly controlled may lead to electrical discharges and the very real danger
of powder explosions (see, for example, Palmer, 1990).

3.1.3.1 Types of electrostatic forces. Three types of electrostatic forces

are distinguished below, and illustrated in Figure 3.16.

(i) Coulombic forces. If two particles are charged, as in Figure 3.16(a),

the force between them is given by

-QA QB
FE = -----=-- (3.58)
4JtErEoa
2
 Interparticle forces 121

(a) A B

Attraction
(0 0
0)
a
• ~

Repulsion
(0
(b)
Image
Charge
/~

(0 I
\
'"
'"

_/ /
\

a a

(c)

a r
. .
+Q -Q +Q

Figure 3.16 Electrostatic forces. (a) Coulombic force; (b) image charge force - semi-
infinite surface; (c) image charge force - confined surface.
J22 Particle mechanics

where QA and QB are the total charges on the two particles, separated by a
centre-to-centre distance a, and EO and Er are the permittivity of free space
and the relative permittivity, respectively. FE is thus positive in attraction.
(ii) Space charge forces. Each particle in a cloud of charged particles is
affected by its interactions with all other particles. If, as is often the case, all
particles have like charge, this will result in mutual repulsion. The effect is
generally only of importance at high charge levels and high mass loadings, as
in electrostatic precipitation. The maximum electrical field which can be
sustained in air due to space charging is about 3 X 106 V m- I , the breakdown
strength of air; this same limit effectively limits the maximum charge which
particles can carry.
(iii) Image-charge forces. When a charged particle approaches a surface,
it induces an 'image charge' in the surface, as shown in Figure 3.16(b). For a
charged particle and a semi-infinite surface, the attraction is the same as if
there were a second particle of opposite sign on the other side of the
interface. For a conducting surface,
Q2
F 1C = ------ (3.59)
4JtErEo(2a?

If the neutral surface is limited in extent, the strength of the image-charge

force depends on the extent of charge separation within it, so that

(3.60)

Thus if r « a, the image-charge force is negligible, because little charge

separation is then possible. In filtration applications (section 7.3), it is
therefore generally more effective to charge the aerosol, which can then
induce an image charge on the much larger collecting fibre or particle, than to
charge the collecting surface alone and attempt to induce an image charge in
the aerosol. (It would be better still to charge both to opposite sign, but this is
usually difficult to arrange.)
It is worth noting in this context that electrostatic forces are often
important in filtration up to the first contact between the aerosol particle and
the collecting surface, but are seldom responsible for holding the aerosol
particle in place thereafter, because the stored electrical charge can usually
leak away through the contact. In other words, electrostatic charges can be
responsible for deflecting a particle trajectory, but can seldom hold the
particle against a surface, unless that surface is an insulator.
3.1.3.2 Magnitude of electrostatic forces. In principle, the magnitude of
an electrostatic force can be calculated using equations (3.58) to (3.60) but in;
 Interparticle forces 123

practice the electrostatic charge is very difficult to measure (Coury et al.,

1991) and even more difficult to predict a priori. Three mechanisms can
contribute to charging a solid aerosol particle: static electrification (tribo-
charging), diffusion charging and field charging. Static electrification depends
very much on the specific materials involved and can be quite unpredictable.
Diffusion charging results from the Brownian motion of ions and particles and
therefore does not require an external electric field. Field charging results
from the movement of unipolar ions in a strong electric field. The charge
acquired is proportional to the square of particle diameter in field charging
and to particle diameter in diffusion charging so that field charging is the
dominant mechanism for particles larger than about 1 f.lm (Hinds, 1982).
By way of example, for fly ash particles from fluidised bed combustion
redispersed for filtration testing using a tribocharging feeder, the charge level
was found to be about 10 f.lC m-2 (equivalent to about 200 excess electrons for
a 1 f.lm particle; Coury et al., 1991). From equation (3.58), with Er = 1 (air)
and EO = 8.9 X 10-12 c2 N-1 m-2 , the force between two particles carrying
equal and opposite charges is
(3.61)
At its maximum, a = d, so that
(3.62)
This is compared with the magnitude of the other interparticle forces
considered in the next section.

3.1.4 Comparison of the magnitude of interparticle forces

Theoretical interparticle forces for single-point contact between equal
spheres (in air) are plotted as functions of particle diameter in Figure 3.17,
with single particle weight for comparison.
The first point to note is that the capillary and van der Waals forces
increase in proportion to particle diameter, or are independent of it if
asperity contact is assumed, while the electrostatic force increases in
proportion to the square of particle diameter and particle weight depends on
the cube. Thus all three interparticle forces can potentially exceed particle
weight for particles below about 1 mm in size. In practice, maximum values of
capillary and van der Waals forces imply perfect smooth contact, which
rarely, if ever, occurs. It is probably more appropriate to take the asperity
contact results for capillary and van der Waals forces, in which case, for the
conditions considered in Figure 3.17, interparticle force and particle weight
are comparable for particle sizes of order 100 f.lm. This is in accordance with
common experience: dust particles of about this size and below are commonly
found adhering to all kinds of surfaces, whereas particles of 1 mm or so will
only adhere to surfaces in the presence of some additional 'glue', perhaps
124 Particle mechanics

z:
w 10-6
w
0::
o
LL

W
-'
w
i=
0::
rt

7l
0::
~ 10-7 -
CAPILLARY
- - -
FORCE (MAX,)
--1---

a=1·65A-
VAN DER WAA

- - -0=4;"
10 100 10c0
PARTICLE DIAMETER (,urn)

Figure 3.17 Comparison of the magnitude of interparticle forces (dashed lines

indicate asperity-to-plane contact). Theoretical interparticle forces for single-point
contact between equal spheres (in air), with particle weight plotted for comparison.

van der Waals (i) A = 6.5 X 10-20 J (quartz)

(ii) values presented for interparticle separations of 1.65 A and
4.0 A
(iii) dashed lines assume asperity-to-plane contact with asperity
radius 0.1 [.lm
Capillary (i) y = 72.8 X 10-2 N m-1 (water)
(ii) values are maximum (j3 -,) 0)
(iii) dashed lines indicate asperity contact as above
Electrostatic (i) maximum force (opposite sign)
(ii) lOr = 1; £0 = 8.9 X 10-12 C2 N- 1 m-2
(iii) charge density = 10 [.lC m-2
Weight Qp = 3 X 103 kg m-3
 Effects of interparticle forces at contacts 125

arising from a chemical bond. Clearly, much larger interparticle forces than
those considered here can occur between particles if there is a chemical
reaction between the surfaces, or if sintering occurs, for example.

3.2 EFFECTS OF INTERPARTICLE FORCES

AT CONTACTS
3.2.1 Contact mechanics*
Consider two elastic spheres brought into contact under an external load Was
shown in Figure 3.18. In the absence of surface forces such as those
considered in the earlier sections of this chapter, classical elasticity theory can
be used (Timoshenko and Goodier, 1951) to determine the radius of the
contact spot, a, in terms of the applied load and the materials properties,

(3.63)

where
1- vT
k1=---
E1

aI-I-a l

Figure 3.18 Contact between two elastic spheres in the presence (contact radius al)
and absence (contact radius ao) of surface forces (after Johnson et ai., 1971).
W = normal load; [) = displacement.

*The essential general text is Contact Mechanics (Johnson, 1985).

126 Particle mechanics

and
1 - Y~
k2=---
E2
R1 and R2 are the radii of the spheres and Yj and Y2 are the Poisson's ratios
of the materials. For equal spheres of the same material and diameter d,
3 (1 - y2)
a3 = - Wd (3.64)
8 E
This analysis was first performed by Hertz (1882), after whom it is named.
Hertz also obtained the relationship between the displacement, 0, and the
normal load

(3.65)

These equations hold approximately for contact between large spheres,

where surface forces need not be considered, and may often be used when the
surfaces are rough or dirty and the surface forces are therefore not able to
contribute. However, at low loads and when the surfaces are relatively
smooth, van der Waals forces become comparable with particle weight and
can have a large influence on the contact mechanics, as argued by, for
example, Briscoe and Adams (1987). In effect, the surface attractive forces
pull the two surfaces together as shown in Figure 3.18, thus increasing the
radius of the contact spot. Furthermore, this also happens in the absence of
any applied load, W. By balancing the interfacial energy, due to the
intermolecular forces, and the elastic strain energy stored in the deforming
spheres, Johnson et at. (1971) showed that in the absence of any applied load
(W = 0), the radius of the contact circle is now given by
9 (1 - y2)
a3 = - nrd2 (3.66)
16 E
where r is the adhesion energy, defined as the work required to separate two
surfaces of unit area from contact to infinity (i.e. twice the surface free
energy, y).
It is of interest that this contact radius is the same as that produced in the
absence of surface forces by an externally imposed contact load, from
equations (3.64) and (3.65), of
3nrd
W'=-- (3.67)
2
The theory of Johnson et al. (1971), commonly known as the JKR theory,
also gives the deflection of the two spheres due to elastic deformation of the
contact spot under a small load (i.e. W« W'),
 Effects of interparticle forces at contacts 127

dW
(3.68)
do
and the force necessary to separate the two spheres, the 'pull-off' force is
3nfd
W=- (3.69)
p 8
Johnson et al. (1971) were able to confirm these results by experiment using
optically-smooth rubber spheres.

3.2.2 Assembly mechanics

In principle, if the microscopic mechanical properties of the single contacts
making up a particle assembly are known, it should be possible to calculate
the macroscopic properties of the assembly. In practice, however, there are
considerable difficulties with this approach. One of the most serious is that
the stresses transmitted through a real particle assembly are not carried
homogeneously (Drescher and de Josselin de Jong, 1972; Thornton and
Barnes, 1986) (Figure 3.19) but are concentrated in preferred paths, so that
some particles experience very high loads, well in excess of the average for
the assembly cross-section, while others 'see' no load at all and may even be

Figure 3.19 Force transmission lines in a compressed bed of particles; the width of
the lines indicates the relative magnitude of the local contact force; principal stress
vertical (Thornton and Barnes, 1986).
128 Particle mechanics

removed from the assembly without disturbing its mechanical properties

(Troadec et al., 1991). This problem can only be addressed in detail by use of
computer simulation (see Chapter 5). Here we consider simple approximate
models for two important macroscopic properties: elasticity and fracture
strength, following the approach set out by Kendall (1987).

3.2.2. 1 Elasticity. The link between the elastic nature of the interparticle
contact and the macroscopic elasticity of an assembly has been analysed by
Kendall et at. (1987). For uniaxial normal stress on the assembly * , 0*, the
mean load per contact is
W= o*ln (3.70)
where n is the number of particle-particle contacts per unit area across a
plane through the assembly. By considering a number of regular packings
with known void fractions, £, Kendall et at. (1987) derived results which are
equivalent to
n = 13.3 (1- £)4a2 (3.71)
where d is the diameter of spherical particles making up the assembly. (This
may be compared with an earlier result due to Rumpf (1962) which gives
similar values for n for values of £ of about 0.4, as shown by Abdel-Ghani et
at., 1991. The extensive subject of packing geometry is considered in detail by
Cumberland and Crawford, 1987.) An external stress causes deformation of
the contacting particles. If the average relative displacement of the particles
under load is 0, then the mean uniaxial strain is, by simple geometric
arguments,
s* = old (3.72)
The assembly elastic modulus is then
do* dW
E* = - = nd- (3.73)
ds* do
assuming that the strain causes no change in the contact density. Since W is a
non-linear function of 0, this equation is applicable only to small deformations
about the unloaded state. Substituting for (dWldo) from equation (3.68),

E* =n [
9Jtf E 2 d 5 J1jj
(3.74)
16(1 - y2)2
or, using equation (3.71) for n,
9JtfE2 ] 1"3
E* = 13.3(1 - £)4 [ (3.75)
16d(1 - y2?

* Asterisks are used to denote the properties of the assembly, so that, for example, E* refers to
the elastic modulus of the assembly and E is the modulus of an individual particle.
 Effects of interparticle forces at contacts 129

Kendall et at. (1987) showed good agreement between this equation and
measured elastic moduli of compacts of submicron titanium dioxide powder,
where particle-particle contact might be expected to be good. However,
Abdel-Ghani et at. (1991), working with much coarser glass spheres,
measured elastic moduli which were much smaller than those predicted from
equation (3.75) and they proposed possible reasons for this disagreement. It
is likely that both the inhomogeneous distribution of stresses (see earlier) and
the imperfect nature of the interparticle contact in the case of larger particles
playa role.

3.2.2.2 Fracture. The classical analysis of the tensile strength of an

agglomerate, due to Rumpf (1962), assumes that the compact 'fails' (i.e.
breaks in tension) when all particle-particle contacts across the failure plane
rupture simultaneously. Failure occurs when the tensile stress exceeds a
critical value given by

(3.76)

where n is the particle contact density and Wp is the force required to

separate two particles. According to the JKR analysis, Wp is given by equation
(3.69) for the special case of equal-sized spheres. -
However, it is well known that macroscopic solids fail in tension under
loads much lower than the stresses needed to cause simultaneous failure of
the interatomic or intermolecular forces across the failure surface. By
analogy, therefore, it should be expected that the failure mechanism analysed
by Rumpf should overestimate the strength of a compact (and Rumpf himself
recognised this). Starting with the classic work of Griffith (1920), the science
of fracture mechanics has developed to explain the failure of solids, and the
role of flaws or imperfections in their structure (e.g. Parker, 1981; Williams,
1984t More recently, various workers (e.g. Kendall et at., 1986; Mullier et
at., 1987; Adams et at., 1989) have shown how the ideas of fracture mechanics
can be carried over into the behaviour of powder compacts. For an elastic
solid, fracture occurs by propagation of cracks, initiated at points of high
stress concentration. Consider, as a simple example, an elastic body of
thickness B containing a sharp crack of length A (Figure 3.20). When a tensile
stress is applied, elastic strain energy is stored throughout the body except in
the immediate vicinity of the crack. Failure of the body in simple tension -
usually known as 'Mode I failure' - occurs when the crack propagates across
the body, as elastic strain energy is released to provide the energy required to
create the new surfaces. Thus for the crack to propagate,

t Anextremely entertaining layman's account of the subject is contained in J.E. Gordon's book,
The New Science of Strong Materials (Gordon. 1976).
130 Particle mechanics

Strain-
free

Figure 3.20 Elastic body containing a crack under tensile stress, showing strain
release around the crack. (B is the thickness, perpendicular to the plane of the paper).

1 aU (3.77)
- =2y*
B oA
=Gc
where a U/ aA is the 'rate of release of strain energy' with respect to crack
extension, y* is the free energy per unit area of fresh surface (i.e. the
'fracture surface energy'), and G c is known as the 'critical strain energy
release rate'. (Note that G c is not a 'rate' in the more conventional sense of
the term.)
Following Kendall et al. (1986), the fracture surface energy of a compact
might be expected to be related to the interactions between the constituent
particles as
2y* = nu (3.78)
where u is the energy required to separate two particles. For smooth spheres,
u follows from the JKR analysis as
SrSd4 ] v:,
u = [ 0.296 Jt E2 (1 - y2f (3.79)

so that, with n given by equation (3.71)

rS(1 - Y2)2J '/3
2y* = 59.7(1 - £)4 [ (3.80)
E 2 d2
 Effects of interparticle forces at contacts 131

In practice, equation (3.80) does not give a good prediction of the fracture
energy except in those rare cases where the particle interactions can be
considered entirely elastic. It is well known that there is a region of stress
concentration at any notch or flaw. When the local stress exceeds the elastic
limit, the compact deforms inelastically. Relative movement between the
particles may also dissipate mechanical energy by interparticle friction
(Adams et at., 1987). In macroscopic terms, a plastic 'process zone' develops
ahead of the notch in which energy is dissipated by plastic deformation rather
than being stored elastically (Mullier et at., 1987; Adams et at., 1989). In this
case, equation (3.77) becomes
(3.81)
where y; is the additional energy dissipated. It is common in many materials,
including polymers and metals, for this term to be considerably greater than
the interfacial energy. The formation of a process zone increases the effective
crack length due to perturbation of the stress field.
Adams et at. (1989) discuss methods of measuring G c and related
parameters. All tests involve loading a specimen in some way until failure
occurs; the commonest example is the 'three-point bend'test (Figure 3.21).
For 'stable fracture', in which the crack extends only as external work is done
on the specimen, G c can be obtained from a simple energy balance
U
G=--- (3.82)
c B(D -A)

t Applied force

Figure 3.21 Three-point bend test.

132 Particle mechanics

where U is the work done on the specimen in the fracture test (the area under
the force-deflection curve) and B(D - A) is the fracture area. If fracture is
'unstable', sufficient energy is stored in the system at the critical point to
propagate the crack completely across the specimen and
U
(3.83)
Gc = BD¢>

where U is the work done on the specimen up to the fracture point and ¢> is a
calibration factor which accounts for the change in compliance (the rate of
change of displacement with change in load) of the specimen with notch
length; ¢> is tabulated for bars containing notches of various lengths by Plati
and Williams (1975).
Abdel-Ghani et al. (1991) used a modification of the three-point bend test
to measure the fracture energy of assemblies of glass spheres, and confirmed
the earlier conclusions of Adams et al. (1989) that the fracture of assemblies
or compacts of fine particles can be characterised using the principles of
fracture mechanics. However, the measured values of G c were much larger
than those calculated from equation (3.80), although this had been found
adequate to explain the fracture behaviour of submicron powders, albeit with
a fitted value for r (Kendall et aI., 1986). As discussed above, the explanation
is that most of the work done on the specimen is dissipated, rather than being
stored as elastic energy which can be released to create new surface.

3.3 FRICTION AT A SINGLE CONTACT

The traditional approach to describe the friction between two solid bodies in
contact is through the use of the widely accepted Amonton's friction law,
which states that the friction force acting between the two rigid bodies is
independent of the 'apparent' contact area and is also proportional to the
applied normal load,

F= !AwW (3.84)

A corollary to equation (3.84) is the Coulomb friction model, which describes

the bulk mechanical behaviour of 'cohesionless' granular solids where
attractive interparticle forces are negligible and the coefficient of friction is
independent of the level of consolidation (see Chapter 4). Not all materials
follow Amonton's friction law, however. Smooth particles like polyethylene
pellets and glass spheres, for example, exhibit high values of friction under
very low compressive loads (Tiiziin et aI., 1988). As the load increases, the
friction coefficient is found to reduce to a much smaller value, as shown in
Figure 3.22. Particles with rough surfaces, on the other hand, such as
agricultural seeds, show no such load dependence.
 Friction at a single contact 133

0.30 ,--.,---,------,---,---,
(a)

0.20

0.10 .
• ••
-----~-.----.--------
•••••••
• ••

o ~~~==::==~~
0.60 I I 0.75 r : . : - - - - - , - - , - - - - , - - - - - , - - - ,
~ (b) (c)
0.64
+
I
I
0.54
0.40 ~ • -
\.

0.20
. '......
I-- ,
\
,
,
..
-
:l
0.43

0.32 \
..
'
•
I- • •- ... ----..--.\-- -
•• • 0.21 ~! ---
-

O~ __L_I__L_I__L_I__ L_I~

o 3 6 9 12 3 6 9 12 15
(Jw (kPa) (Jw (kPa)

Figure 3.22 Experimental wall friction coefficients: (a) mustard seeds; (b) glass
beads; (c) polyethylene beads (Tiiziin et al., 1988).

Highly polished metallic surfaces also exhibit very large effective coefficients
of friction at low normal loads. Tabor and his coworkers (Tabor, 1955;
Bowden and Tabor, 1955) explained such phenomena via an 'adhesion
model' requiring significant tangential force to break bonds formed between
the contacting surfaces. Briscoe et al. (1984, 1985) extended the same concept
to describe the interfacial shearing mechanism of thin polymeric films
wherein new 'junctions' continually form as old ones are ruptured.
The inclusion of an interfacial adhesive component in the friction model
results in a modification of Amonton's law as
F='t,A + aW (3.85)
where F is the shear force, W the applied normal load, A the interfacial
contact area, 'to the interfacial shear strength and a the pressure coefficient,
the last two being material properties. The effective coefficient of friction can
be obtained by dividing equation (3.85) by the normal load to give

(3.86)
734 Particle mechanics

where p = WIA is the contact normal pressure. Such a model produces a

coefficient of friction that is very large at small values of the contact pressure.
The wall friction coefficient tends towards the asymptotic value given by the
pressure coefficient, a, at large values of the contact pressure. Such frictional
behaviour is followed very closely by the wall friction data measured for
granular materials composed of smooth particles such as glass ballotini and
polyethylene presented in Figure 3.22.
Adams et al. (1987) and Tiiziin et al. (1988) have adapted the basic form
given by equation (3.86) above to explain the wall friction data measured with
smooth particles (see the first column of Figure 3.23). Assuming elastic
Hertzian contacts between spheres and that the mean normal pressure, p, acts
uniformly over the Hertzian contact area (of radius a), they obtained

!lw = (3.87)
W

where a = (E*WR)Y3 (see equation 3.63), R is the effective radius of

curvature at contact given by 11R = (1/R1) + (1IR 2 ) and E* is calculated as

E*=-
3
4
(1-V---
E1
T 1-V~)
+- --
E2

Here v and E are the Poisson ratio and Young's modulus of elasticity,
respectively, and subscripts 1 and 2 refer to the particle and the vessel wall.
Equation (3.87) has been used successfully to predict the load dependence of
friction at very low loads (Tiiziin et ai., 1988) and has the advantage of
relating the frictional force to measurable material and geometric quantities.
Adams et ai. (1987) show that it can be regarded as equivalent to the
empirical equation

F=kW' (3.88)

where the parameters k and n are load-dependent. It should be noted that the
load index, n, can vary between 213 and 1 even for a Hertzian (elastic) contact
when surface forces are taken into account.
The foregoing analysis applies to perfectly smooth particles; in practice it is
commonly the case that real particles show surface asperities. This case has
been considered in detail by Archard (1957) and his approach is summarised
by Adams et al. (1987). Archard's model is shown schematically in the second
column of Figure 3.23. The important feature is that the number of contacts
between the rough particle and the plane surface increases as the load
increases, so that the mean pressure at the asperity contacts is insensitive to
load. Hence, for a rough particle
 Impact and rebound of particles 735

~ (Hertz) (Archard)
SURFACE SMOOTH ROUGH
EXAMPLES GLASS MUSTARD SEED
POLYETHYLENE
FRICTION LOAD-DEPENDENT NOT LOAD-
COEFFICIENT DEPENDENT

Figure 3.23 Modelling of single particle friction.

(3.89)

where P is a mean asperity contact stress, which is close to the flow stress of
the softer of the two contacting bodies, and a is a constant. Hence the
frictional force for rough particles is directly proportional to the load.

3.4 IMPACT AND REBOUND OF PARTICLES

Impact and rebound are important wherever particles come into contact.
Their importance in agglomeration and filtration will be illustrated in
Chapters 6 and 7. The important questions are usually: Will the particle 'stick'
or 'bounce'? How much energy will be lost in the collision?
Consider a particle of mass m approaching a plane surface in the direction
normal to the surface and at velocity Vi (Figure 3.24)*. As it nears the surface
it will fall into a 'potential well' of depth E i , which arises from the attractive
force between the surface and the particle. If it rebounds from the surface
with insufficient energy to escape the 'rebound potential' it will be captured.
Note that the initial (precontact) potential E i , may be different from the
rebound potential, En because the surface of the particle may deform during
the contact, thus making it more difficult for the particle to escape. This
'deepens' the potential well out of which the particle must 'climb'.
The kinetic energy of the approaching particle is given by 112m Vr, so that
the total energy just before impact is (1jzmVr + Ei). We may now define a
'coefficient of restitution', e,

*This analysis draws on that of Dahneke (1971).

136 Particle mechanics

Vi
+--

0 Approach

0 Rebound

- V,

.......... ...... -

Initial surface
potential
-_ .. _----_ .. Rebound surface
potential

Figure 3.24 Surface potentials for impact and rebound.

normal particle velocity at instant of rebound

e =------------------------------------
normal particle velocity at instant of contact

The total energy at the instant of rebound is then

(3.90)
Upon rebound, the particle must exchange kinetic energy for potential energy
as it 'climbs out' of the potential well. The final kinetic energy of the particle,
beyond the influence of the surface potential, is

112m V2r = (V2m V2] + E)e

I
2 - Er (3.91)

The particle will be captured when all the kinetic energy after rebound is used
up in trying to escape from the rebound surface potential; the limiting value
of Vi for which capture occurs is therefore given by
W2mVt 2 + Ei)e 2 - Er = 0 (3.92)
or

(3.93)
 Impact and rebound of particles 137

There are two special cases:

(1) Er =E =Ej (i.e. no surface deformation, perfectly elastic behaviour)

.'. vt= [2E (3.94)

m
(2) E j « Er (i.e. much deformation, plastic behaviour)
.'. vt = [2E/me 2 ] V2 (3.95)
We now consider an example, that of the impact of quartz particles on quartz
surfaces. (We shall assume the collision to be perfectly elastic - but see later.)
Taking special case (1) above, and assuming only van der Waals forces act,
Ej = Er = E = ARI6a = Adl12a (3.96)
where A is the Hamaker constant (see section 3.1.1), R is the particle radius
(= d/2) and a is the closest surface separation.
Therefore

vt = [AD 6am
1- e2 ] v,
e2
=~
d
[_A_
JtaQp
1- e2 ] v,
e2
(3.97)

Putting A = 6.5 X 10-20 J, a = 1.65 A, Qp = 3000 kg m-3 and e = 0.99 (see

later), then
vt = 2.9 X 1O-81d
Particles with a velocity greater than vt will escape capture. For a 1 !lm
particle this critical velocity is then 2.9 cm s-l, well in excess of its terminal
settling velocity in air. For a 100 !lm particle, however, vt is 0.029 cm S-l,
which is much less than its terminal settling velocity. We can therefore expect
a 1 !lm particle but not a 100 !lm particle to adhere where it falls.
Note that in the calculation above, all fluid effects have been neglected. In
practice, viscous drag will reduce the particle velocity prior to impact and will
also cause energy to be dissipated after rebound. This is considered in section
3.4.2.
Figure 3.25 illustrates the effect of adhesion on impact. Immediately after
contact is made, at (a), adhesive forces draw the surfaces together, from (b)
to (c). Rebound then follows the same path (assuming no deformation has
occurred) until (b), but the surfaces must continue to separate as far as (d)
before separation can occur.

3.4.1 The coefficient of restitution

Although often taken as a constant value, the coefficient of restitution
(e) depends on d, V j and the materials properties of both surfaces. Impact
138 Particle mechanics

+a b c d

--->o..G..-<-cx=_o-->G<---/o-->...8.-<-_8<--+:
cx",----

Separation Distance of
point d approach cx

Rebound

Figure 3.25 Schematic representation of the impact of a particle, including the effects
of adhesion (Reed, 1987).

will only result in plastic deformation if the maximum pressure exerted

during the impact exceeds the elastic yield limit, y, for the softer surface.
This occurs (Reed, 1987) when the impact velocity exceeds a certain value,
V y , or

Via> Vy = 2Jt]Z[
[ 3K
- _2 ]Y.> y'h (3.98)
5~?1

where 01 is the density of the impacting particle,

(3.99)

and

K=-
4 [1---.VI+1-- -
V~ ] -1

3 E1 Ez

where v = Poisson's ratio, E = Young's modulus, subscript 1 = particle and

subscript 2 = surface.
Typical values of Vy are as follows:
 Impact and rebound of particles 139

glass on glass 23 m S-l

glass on stainless steel 2.3 m S-1
copper on stainless steel 0.003 m S-l
In general, plastic deformation will greatly increase the energy loss on
impact and therefore make capture more probable. Figure 3.26 shows
the critical impact adhesion velocities for glass impacting on stainless and
mild steel. For the smaller critical impact velocities associated with larger
particles, both target materials show elastic behaviour, but at larger critical
impact velocities (smaller particle sizes) the behaviour of the two materials is
quite different. The difference between curves A and B shows how plastic
deformation increases the range of particle sizes for which adhesion occurs.
The velocity required for a 10 !Am particle to escape a mild steel surface is
much greater than that required for it to escape a stainless one, because of the
plastic deformation which occurs in the former case. These considerations are
important in determining whether aerosol particles will stick to, and therefore
foul, pipes and heat exchange surfaces.
If the impact is elastic, the only energy dissipation mechanism is
propagation of an elastic wave, unless the surface struck is very thin and
flexible, in which case flexural work can also be done. For elastic impact,
Dahneke (1971) estimates that the coefficient of restitution should be in the
range 0.98 to 0.99. According to Reed (1987), the fractional loss of energy on
impact is given by

(3.100)

....
c
o
·iii
<D ~
..
€';" 10- 1
_ E
co
CI)

o~

~~
EO
._ 0
(ij Qi 10-2
.. .
o >
."

<:5
.
10·3L---------~.----~--------~~~
1 10
Particle radius (Jlm)

Figure 3.26 Critical impact adhesion velocities for glass particles impacting onto steel
surfaces: A, stainless steel (elastic); B, mild steel (elastic-plastic). The broken line
shows the terminal velocity of the glass particles in air (Reed, 1987).
740 Particle mechanics

where

and
Co = (E2/Q2)'/2
Using the properties of glass, Aw = 0.025 so that for V io = 1 m S-l, e = 0.99.

3.4.2 The effect of liquid layers

In some cases of impact of particles on surfaces or with each other, the
particles are covered with a thin layer of liquid. This is the case in
agglomeration, for example, where liquid binders are deliberately added in
order to promote agglomerate growth. As an example of the effect that this
layer can have upon the fate of impacting particles, we here consider a
simplified analysis proposed by Barnocky and Davis (1988).
Consider a spherical particle, partially immersed in a liquid layer of
thickness 0 (Figure 3.27) and approaching a flat surface at velocity V. By
definition,
da
-Vet) (3.101)
dt
while
dV
m - = -F(t) (3.102)
dt
Now assume that the pressure in the liquid layer is not so large that the
surfaces are able to deform, and that the Reynolds number based on the gap
dimension, a, is small:
Re = QVal1J. « 1 and that a « R.
If these conditions are met then the viscous force, F = 6:JtIJ.R2Vla. This is
Reynolds' 'lubrication equation' (see section 3.1.2.3).
In general, the viscous force dominates over all others, such as gravity
(unless the particle is very large) and van der Waals forces, while the particle
is within the viscous layer.
From equations (3.49), (3.101) and (3.102)

dV
m =-=--- (3.103)
da V a
 Impact and rebound of particles 141

Figure 3.27 A spherical particle approaching a plane surface covered by a thin liquid
layer.

which can be integrated with the initial condition V = Vo at a = ao to give

VIVo = 1 - In(aola)/St (3.104)
where St = m VoI6:n:!lR2. St is a Stokes number, a measure of the inertia of
the particle relative to the viscous forces (see section 2.2 for other definitions
of the Stokes number). Note that for equation (3.103) to hold, the nose of the
particle must be wetted over an area with radius (Ra)'h, which is true when ao
= 20/3. We shall neglect the slowing of the particle up to that point. Also
note that the static component of the capillary force has been neglected
(which will not always be justified - see section 3.1.2.3).
For perfectly smooth spheres, significant deformation occurs when the
particle has sufficient inertia to penetrate the liquid layer to a separation
equivalent to an 'elasticity length scale' for the surface, at> i.e. capture occurs
if
St < St* = In(aolal) (3.105)
where al = [4K!l Vo R 3hf'S, with K as in equation (3.99).
A more complete analysis, taking into account elastic deformation of the
sphere on approach and possible rebound thereafter, gives the approximate
result
St* = 1.36In(aolal) - 1.25 (3.106)
For a particle to escape, it is only necessary (if the surfaces are perfectly
smooth) for initial rebound to occur. Barnocky and Davis (1988) show that
the pressure deficit is then such that the resistance to motion in travelling out
through the liquid layer is negligible since cavitation of the liquid must occur.
In practice, it is much more likely that at least one of the contacting surfaces
will be 'rough', with asperities of ~1 !lm. If this is the case, and the asperity
height a2 exceeds at> from equation (3.105), the critical value of St below
which particles will be captured before making contact with the asperities is
St* = In(aola2) (3.107)
142 Particle mechanics

but in this case we must also take into account the resistance of the liquid
layer after rebound from the surface, because the pressures generated in the
liquid are not low enough to cause cavitation. At the point of first contact
with the asperities, from equation (3.104)
(3.108)
Therefore, just after rebound
Vro = -eVio = -eVo [1 -In(aola2)ISt] (3.109)
Now integrating equation (3.103) with V = Vro at a = a2 and - V> 0 at a = ao
gives

St* = (1 + -e1 )In(aola2) (3.110)

= 2In(aola2)
if e = 1.
Barnocky and Davis (1988) obtained results for impact of particles on
smooth and rough surfaces, for a range of liquid layer thicknesses and
viscosities, showing good agreement with equations (3.106) and (3.110).
This approach to the prediction of the outcome of collision events opens
up the possibility of achieving a better understanding of 'capture' of particles
in some forms of filtration (see section 7.3.2.2) and in the various processes
which are used to manufacture agglomerated particles (see section 6.2.7).

3.5 NOMENCLATURE
A Hamaker constant (J), contact area (m 2)
a Separation between bodies (m); contact radius (m)
a' Dimensionless separation, aiR
d Particle diameter (m)
E Elastic modulus (Pa)
Ej,Er Surface potential energies (J)
E* Elastic modulus of assembly (Pa)
e Coefficient of restitution
F Force (N)
Gc Critical strain energy release rate (J m-2)
k Total curvature (m-I )
m Mass (kg)
n Index in pair potential; particle-particle contact density (m-2)
p Pressure (Pa)
Q Electric charge (C)
R Radius (m)
r Separation between molecules (m); radius (m)
St Stokes number
 References 143

u Energy (J)
u Energy to separate two particles to infinity (J)
V Velocity (m S-1)
Liquid molar volume (m3 kmol-1)
Pair potential (J); normal load (N)
Pull-off force (N)

Greek letters
a Pressure coefficient
f3 Half-angle e)
y Surface energy (J m-2 )
r Adhesion energy (J m-2 )
o Displacement (m)
E Void fraction; permittivity (F m-1)
!l Viscosity (Pa s)
lAw Wall friction coefficient
Q Number density of molecules (m-3 )
o Stress (Pa)
v Poisson ratio
A Parameter in equation (3.100)
<j> Compliance correction in equation (3.83)
'to Shear stress

REFERENCES
Abdel-Ghani, M., Petrie, J.G., Seville, J.P.K., Clift, R. and Adams, M.J. (1991)
Powder Techno/. 65, 113.
Adams, M.J. and Perchard, V. (1985) 1. Chern. E. Symp. Ser. 91, 147.
Adams, M.J. and Edmondson, B. (1987) in Tribology in Particulate Technology (B.J.
Briscoe and M.J. Adams, eds), Adam Hilger, Bristol and Philadelphia.
Adams, M.J., Briscoe, B.J. and Pope, L. (1987) in Tribology in Particulate
Technology (B.J. Briscoe and M.J. Adams, eds) , Adam Hilger, Bristol and
Philadelphia.
Adams, M.J., Williams, D. and Williams, J.G. (1989) J. Mater. Sci. 24, 1772.
Archard, J.F. (1957) Proc. Roy. Soc. (London) A243, 190.
Barnocky, G. and Davis, R.H. (1988) Phys. Fluids 31, 1324.
Bowden, F.B. and Tabor, D. (1955) Friction and Lubrication of Solids, Cambridge
University Press, Cambridge.
Briscoe, B.J. and Adams, M.J. (eds) (1987) Tribology in Particulate Technology,
Adam Hilger, Bristol and Philadelphia.
Briscoe, B.J., Pope, L. and Adams, M.J. (1984) Powder Technol. 37, 169.
Briscoe, B.J., Fernando, M.S.D. and Smith, A.C. (1985) J. Phys. D: Appl. Phys. 18,
1069.
Cameron, A. (1981) Basic Lubrication Theory, Ellis Horwood, Chichester.
Capes, C.B. (1980) Particle Size Enlargement, Elsevier, Amsterdam.
Coughlin, R.W., Elbirli, B. and Vergara-Edwards, L. (1982) J. Colloid Interface Sci.
87,18.
144 Particle mechanics

Coury, J.R., Raper, J.A., Guang, D. and Clift, R. (1991) Trans. l. Chern. E. 69(B),
97.
Cross, N.L. and Picknett, R.G. (1963) International Conference on the Mechanism of
Corrosion by Fuel Impurities, Marchwood Engineering Laboratories, Butterworths,
London, p. 383.
Cumberland, D.J. and Crawford, R.J. (1987) The Packing of Particles, Elsevier,
Amsterdam.
Dahneke, B. (1971) J. Colloid Interface Sci. 37,342.
Drescher, A. and de Josselin de Jong, G. (1972) J. Mech. Phys. Solids 20, 337.
Fisher, R.A. (1926) J. Agric. Sci. 16, 492.
Gordon, J.E. (1976) The New Science of Strong Materials, 2nd edn., Penguin,
London.
Griffith, A.A. (1920) Phil. Trans. Roy. Soc. (London) A221, 163.
Hertz, H. (1882) in Miscellaneous Papers by H. Hertz (1896) (Jones and Schott, eds),
Macmillan, London.
Hinds, W.C. (1982) Aerosol Technology, Wiley-Interscience, New York.
Israelachvili, J.N. (1991) Intermolecular and Surface Forces, 2nd edn., Academic
Press, London.
Johnson, K.L. (1985) Contact Mechanics, Cambridge University Press, Cambridge.
Johnson, K.L., Kendall, K. and Roberts, A.D. (1971) Proc. Roy. Soc. (London)
A324,301.
Kendall, K. (1987) in Tribology in Particulate Technology (B.J. Briscoe and M.J.
Adams, eds), Adam Hilger, Bristol and Philadelphia, 110.
Kendall, K., Alford, N.McN. and Birchall, J.D. (1986) Inst. Ceram. Proc. Special
Ceramics No.8, Institute of Ceramics, Stoke on Trent, p. 255.
Kendall, K., Alford, N.McN. and Birchall, J.D. (1987) Proc. Roy. Soc. (London)
A412,269.
Lian, G., Thornton, C. and Adams, M.J. (1993) J. Colloid Interface Sci. 161,138.
Mason, G. and Clark, W.C. (1965) Chern. Engng. Sci. 20, 859.
Mullier, M.A., Seville, J.P.K. and Adams, M.J. (1987) Chern. Engng. Sci. 42,667.
Orr, F.M., Scriven, L.E. and Rivas, A.P. (1975) J. Fluid Mech. 67,723.
Palmer, K.N. (1990) in Powder Technology (M.J. Rhodes, ed.), Wiley, Chichester.
Parker, A.P. (1981) The Mechanics of Fracture and Fatigue, Spon, London.
Picknett, R.G. (1969) J. Colloid Interface Sci. 29, 173.
Pietsch, W.B. (1968) Nature 217,736.
Plati, E. and Williams, J.G. (1975) Polym. Eng. Sci. 15, 470.
Princen, H.M. (1968) 1. Colloid Interface Sci. 26, 249.
Reed, J. (1987) in Tribology in Particulate Technology (B.J. Briscoe and M.J. Adams,
eds), Adam Hilger, Bristol and Philadelphia, 123.
Rumpf, H. (1962) in Agglomeration (W.A. Knepper, ed.), Interscience, New York,
p.379.
Shaw, D.J. (1992) Colloid and Surface Chemistry, 4th edn., Butterworths, London.
Simons, S.J.R., Seville, J.P.K. and Adams, M.J. (1994) Chern. Eng. Sci. 49,2331.
Tabor, D. (1955) Proc. Roy. Soc. (London) A229, 198.
Tabor, D. (1987) in Tribology in Particulate Technology (B.J. Briscoe and M.J.
Adams, eds), Adam Hilger, Bristol and Philadelphia, 206.
Thornton, C. and Barnes, D.J. (1986) Acta Mechanica 64, 45.
Timoshenko, S. and Goodier, J.N. (1951) Theory of Elasticity, McGraw Hill, New
York.
Troadec, J.D., Bideau, D. and Dodds, J.A. (1991) Powder Technol. 65, 147.
Tiiziin, U., Adams, M.J. and Briscoe, B.J. (1988) Chern. Engng. Sci. 43, 1083.
Visser, J. (1989) Powder Technol. 58, 1.
Williams, J.G. (1984) Fracture Mechanics of Polymers, Ellis Horwood, Chichester.
4
Characterisation of bulk
mechanical properties

In storage, flow and handling of bulk solids, accurate determination of the

angles of internal and wall friction is essential both for the efficient design of
equipment and for reliable predictions of flow behaviour. Current methods of
measurement of the frictional behaviour of bulk solids in conventional shear
cell equipment suffer serious limitations when investigating the dynamic
stress and velocity fields observed in practice. Accurate modelling of
frictional behaviour requires measurements of friction angles as functions of
both the individual particle properties and also of changes in the state of flow,
such as those observed in filling and discharging storage vessels.

4.1 INTRODUCTION
The frictional behaviour of powder materials in all engineering applications is
described by two independent parameters: the coefficient of internal friction
and the coefficient of wall friction. The former determines the stress
distribution within a bed of powder undergoing strain and the latter describes
the magnitude of the stresses between the material bed and the walls of its
container.
Accurate knowledge of these two friction coefficients is essential to the
design of all bulk solids storage, handling and transportation equipment.
Both the structural strength of such equipment and the stability of the
material flow within them rely heavily on the accuracy of the values of the
friction coefficients used in design calculations (British Materials Handling
Board, 1985).
Values of friction coefficients are in general determined empirically, and
these empirical values are then used in all subsequent calculations to predict
stress and flow fields. The accuracy of such predictions will be affected
significantly by the bias and errors associated with a particular measuring
technique.
There have been few attempts to predict analytically the coefficients of
internal and/or wall friction. This scarcity is proof of the difficulty of such a

J. Seville et al., Processing of Particulate Solids

© Chapman & Hall 1997
146 Characterisation of bulk mechanical properties

task, which almost certainly requires explicit definitions of the functional

relationships between the friction coefficients and both the single particle
properties such as particle size, shape, surface roughness and modulus of
elasticity, and some assembly characteristics such as packing fraction,
compressibility and cohesive strength.
This chapter provides firstly a critical review of the current techniques
employed in measuring the coefficients of friction. It then proceeds to
question the validity of the empirical data resulting from such tests in the light
of the reported laboratory and industrial scale observations made during
filling and discharge of granular materials.

4.2 EMPIRICAL MEASUREMENTS OF

COEFFICIENTS OF FRICTION
The internal and wall coefficients of friction of powder materials may be
measured indirectly in a number of ways. The most common method is the
use of a 'shear cell' or 'shear tester'. The principles of operation of these
devices are described and compared in great detail in the literature (see, for
example, Jenike, 1961; Carr and Walker, 1968; Schwedes, 1979; Roberts,
1984), and will therefore not be repeated here. For the present purposes, it is
sufficient to say that the measurement involves a simple procedure of filling a
known quantity of powder into a rectangular or circular box comprising a
shoe and a detachable lid section, as shown in Figure 4.1. In the design
attributed to Jenike (1967) and Jenike et al. (1973) the shear is applied
transversely to the powder sample which is held under the action of a known
normal load, W, as in Figure 4.1(a). In the so-called 'annular shear cell'
attributed to Walker et al. (see Walker, 1967; Carr and Walker, 1968) the

(a) (b) (c)

~
Lever
arm
v~ Top ring
Shear Base
Retaining force 5
_~====d:6d..pln ~ F
F I n put
orced
t rans Vibration
ucer

I
Sample

Figure 4.1 Shear testing equipment for granular solids: (a) Jenike shear box (from
Jenike, 1961); (b) annual shear cell (from Carr and Walker, 1968); (c) vibrating shear
cell (from Roberts et at., 1984).
 Empirical measurements of coefficients of friction 147

friction, F, is caused by rotating the lid relative to the shoe, as in Figure

4.1(b).
By measuring the frictional force, F, required to move a flat block of
granular material against another under various normal loads, W, the
material yield locus (MYL) is obtained,
T = f(a) (4.1)
where T is the shear stress parallel to the planar failure surface and a is the
compressive stress normal to that surface. The slope of the material yield
locus determines the coefficient of internal friction, f-l,
f-l = tan<l> (4.2)
where <I> is known as the angle of internal friction. A similar procedure is
adopted to determine the wall coefficient of friction, f-lw, but in this case the
base of the lid is lined with the wall material to be tested. The resulting yield
locus is referred to as the wall yield locus (WYL). The slope of the wall yield
locus determines the coefficient of wall friction, f-lw,
f-lw = tan<l>w (4.3)
where <l>w is known as the angle of wall friction.
In both cases, the accuracy of the measurements depends largely on the
creation and maintenance of a well-defined failure plane within the cell
throughout the tests. A great deal of research has already gone into the
possible modifications to the conventional shear testing equipment mentioned
above so as to ensure well defined and reproducible failure surfaces; see, for
example, Williams and Birks (1965). To this effect 'true' biaxial and triaxial
testing equipment has been designed and developed in recent years, as
described by, for example, Haaker and Rademacher (1982) and Schwedes
and Harder (1985). It is possible with such equipment to consolidate the
material samples either isotropically or non-uniformly prior to shear and to
measure directly the extent of volumetric deformation during material
failure.
The internal angle of friction can sometimes be approximated by the so-
called angle of repose of the material. As will be demonstrated below, this is a
, reasonable approximation only when the cohesive strength of the material is
believed to be negligible. Two alternative ways have been used in the past to
determine the angle of repose: materials are either discharged from a flat-
bottomed container (see Rose and Tanaka, 1959; Tiiziin and Nedderman,
1982) or are allowed to form a heap on a horizontal surface (see Smid, 1983).
In the former case, the angle of repose is given by the angle between the
surface of the material left behind in the container and the bottom wall at the
end of discharge, as shown in Figure 4.2(a). In the latter, the half-angle ofthe .
free-standing heap measured at the top is taken as the angle of repose, as
shown in Figure 4.2(b). Both of these methods of measurement refer to the
148 Characterisation of bulk mechanical properties

materials at rest, and hence they can only be taken as a likely indication of the
'static' value of the internal angle of friction.
The angle of wall friction, <Pw, however, can be determined 'quasi-
statically' by accommodating a layer of the bulk material to be tested on an
inclined surface and measuring the angle of inclination of the surface at the
point of initiation of particle sliding (Augenstein and Hogg, 1974; Nedderman
and Laohakul, 1980), as shown in Figure 4.2(c).
A more direct determination of the wall friction is made possible by the
simultaneous measurement of the normal and shear loads at the walls of a
container using twin-axis cells (Tiiziin and Nedderman, 1983, 1985). In this
case, the wall friction corresponding to the static material fill can be measured
separately and then compared with the dynamic values resulting from a
flowing bed (Tiiziin et al., 1986).
However, such in situ measurements of the static and dynamic values of the
internal angle of friction are not possible. The shear cell tests described above
can at best be called quasi-static as the so-called yield loci obtained refer to
the stress distribution at incipient failure, which is analogous to the initiation
of motion in a static material bed. Unlike the dynamic wall friction
measurements described above, there is as yet no direct method of
measurement of the effect of the state of flow on the magnitude of the
internal friction. Such information would be invaluable to the successful
coupling of the stress and strain rate fields in particulate materials, as will be
demonstrated below.

Dead

-~
bead
~zone
))

(a) (b) (c)

Figure 4.2 Indirect measurements of angles of friction of granular solids: (a) angle of
repose in a flat-bottomed bin; (b) angle of repose of a static heap; (c) angle of friction
of an inclined wall.

4.3 ANGLES OF FRICTION OF COHESION LESS

MATERIALS
The term 'cohesionless' is used to describe those granular materials whose
intrinsic shear strength under zero consolidation load (a = 0) is negligible. In
this case the yield locus for plastic failure is given by
 Angles of friction of cohesionless materials 149

't = otan<j> (4.4)

where 't is the shear stress and <j> is the angle of internal friction. Equation
(4.4) is often referred to as the Coulomb failure criterion, in recognition of
this early pioneer (Coulomb, 1776). Lack of cohesion in granular materials
suggests that the effects of interparticle forces (see Chapter 3) are negligible.
Dry granular materials of particle size greater than about 100 ~m and which
do not carry significant levels of electrostatic charges will usually behave as
'Coulomb type' materials. Figure 4.3(a) shows the internal yield loci of
mustard seeds, table salt and fine dry sand obtained in the Jenike-type shear
box described above. All three materials behave like the Coulomb type, i.e.
the internal yield locus is given by a straight line through the origin, whose
slope determines the angle of internal friction (equation 4.4). More
significantly, the yield loci corresponding to the different levels of pre-
consolidation of the material samples prior to shear are almost indistinguish-
able for the materials seen in Figure 4.3(a). Hence, for all practical purposes,
a Coulomb-type material is believed to have a coefficient of internal friction
which is independent of its bulk density and hence of the magnitude of the
interstitial voidage of the material bed. The implication therefore of the
results of shear cell tests with Coulomb-type materials is that in all
engineering applications involving filling of materials into storage units,
discharge of material from storage units onto mechanical conveying equipment
or into pipe assemblies, the frictional behaviour of the material will remain
essentially independent of its packing conditions.
The wall friction behaviour of Coulomb-type materials is characterised by
the lack of adhesive forces between the particles and the wall surface. In
practical applications, this is known as 'non-stick' behaviour. With non-
sticking materials, there will be no permanent deposition of the bed contents
on the walls of its container. In the absence of adhesion, the wall yield locus
can be reasonably described by an equation of the form
(4.5)
where Ow and 'tw are the normal and shear stress at the wall, and <j>w is the
angle of wall friction. The angle of wall friction of Coulomb-type materials is
often assumed to be independent of the magnitude of the normal stress acting
on the wall and also of the bed void age adjacent to the wall. This assumption
is inherent in the method of measurement using a layer of particles on an
inclined wall surface as described above. Figure 4.3(b) shows wall yield loci
obtained by different researchers (Cooker, 1980; Roberts et al., 1984; Tiiziin
et al., 1986) in equipment similar to the shear testers illustrated in Figure 4.1.
The wall yield loci in Figure 4.3(b) all appear to go through the origin,
which confirms that the adhesive component of the wall friction is negligible.
More significantly, in some cases the wall yield locus is slightly curved
indicating a dependence of the wall friction angle on the magnitude of the
normal load on the wall.
150 Characterisation of bulk mechanical properties

10r----------------------.

8 4

~6
E
z
~4
\-'

2 4 6 8 10 12 14 2 10
(a) cr (kN m- 2) (b)

0.5~

I
0.4~
\
.'
0.3~~,
..+.'-.
.J. ",. -?.........
'.!'I.
..•.• - ' -
+'_ a=0.14
0.2 f-... + + ··'"-f.·····-+~-···T~-;;:0.20
... ... ...
...... ... ...... A ... A a = 0.15
0.1 I-

OL--~I~I---L-I-L-I-~I~I---L-I-­
o 2 4 6 8 10 12 14
(e) crw (kN m- 2 )

Figure 4.3 Angles of friction of cohesionless materials. (a) Effective internal yield
loci of three cohesionless materials. Data from Tiiziin et at. (1986) with values of mean
particle size d (!im): 0, mustard seed (2290); *, table salt (360); 6, fine sand (160).
(b) Yield loci of cohesionless solids on various wall surfaces: 0, mustard seed on
aluminium alloy (Tiiziin et at., 1986); ., black coal on stainless steel (Roberts et at.,
1984); 6, polymer powder on glass (Cooker, 1980); ., polymer powder on steel
(Cooker, 1980). (c) Variation of wall angle of friction with normal stress on a smooth
wall (from Tiiziin et at., 1988). The curves are derived from theory using equation
(4.6). The data and curves are: ,;., full curve, mustard seeds; +, double broken curve,
glass ballotini; . , broken curve, polyethylene pellets.
 Angles of friction of cohesive materials 151

The variation of the wall friction angle during filling and discharge of bulk
solid containers is best measured by the use of the double-axis load cells
described above. Tiiziin et al. (1988) have shown that with Coulomb-type
materials, the magnitude of the wall friction can in certain cases change
considerably once flow is initiated, and during flow, significant variation of
the wall friction angle is observed along the height of the container. Figure
4.3(c) shows the experimental values of the wall friction with mustard seeds,
glass ballotini and polyethylene beads as functions of the normal load acting
on the smooth aluminium side wall of a plane-strain silo. Clearly, the wall
friction behaviour is markedly different in each case; even though all three
materials behave as Coulomb-type, differences in the particle surface
roughness are shown to give rise to different functional dependence of !!w on
the wall stress Ow (see Tiiziin et al., 1988).
Conventionally, the continuum mechanics approach is used to explain the
variation of the internal angle of friction with consolidating load at the point
of initiation of failure within the material. The continuum mechanics
approach is therefore also the underlying theory of the shear-testing devices
described in the previous section. In practice, however, this approach cannot
fully explain the frictional behaviour of Coulomb materials during sustained
flow. Its apparent deficiencies result from neglect of effects of the key particle
properties such as those described above. Where the wall friction is
concerned, these effects can be investigated to a certain extent by carrying out
single-particle friction experiments. A description of one type of apparatus
for performing these experiments is given by Briscoe et al. (1973). The
apparatus is designed to slide a single particle against a well-defined surface
under the action of a given normal load. The theory of single contact friction
is summarised in section 3.3, while section 5.2 describes some recent attempts
at assembly models of friction.

4.4 ANGLES OF FRICTION OF COHESIVE

MATERIALS
Particulate materials are referred to as 'cohesive' if their intrinsic shear
strength under zero consolidation load is large. Using the Coulomb criterion
given by equation (4.4), the yield locus at incipient failure can be written as
1: = (tan<j»o + c (4.6)
where c is termed the cohesion and is simply the intercept of the internal yield
locus on the shear stress axis, as shown in Figure 4.4. The form of the yield
locus given by equation (4.6) is a straight line. However, experimental
measurements with shear cells sometimes indicate significant curvature of the
yield locus. Figure 4.4(a) shows three yield loci, corresponding to three
different levels of preconsolidation of a polymer powder (Cooker, 1980). All
152 Characterisation of bulk mechanical properties

8
.. '

a
(a) (b)

't

o a
(c)

Figure 4.4 Angles of friction of cohesive materials. (a) Incipient yield loci of a
polymer powder « 1400 !lm). (1) Ql = 380 kg m-3 , aLl = 0.928 kN m-2 ; (2) Q2 =
444 kg m-3 , aL2 = 3.32 kN m-2 ; (3) Q3 = 484 kg m-3 , aL3 = 8.101 kN m-2 • Here Q
is the initial bulk density and aL is the termination point of the incipient yield locus.
(b) Mohr-Coulomb failure of cohesive materials: IYL, incipient yield locus; EYL,
effective yield locus; CVL, critical voids line. (c) Termination locus of incipient yield
and consolidation surfaces: YL, yield line; CL, consolidation line; CVL, critical voids
line (= termination locus).

three loci have a non-zero intercept on the shear axis and all are distinctly
curved. An equation of the form

(4.7)

°
can be used to fit the data in Figure 4.4(a) for < 0L, i.e. for stress values
lower than the termination point of the yield locus. Here T is termed the
tension or tensile strength of the material and is given simply by the intercept
ofthe yield locus on the normal load axis. Equation (4.7) is well known as the
Warren-Spring equation in recognition of the pioneering work on granular
 Angles of friction of cohesive materials 153

materials carried out at the Warren-Spring Laboratories in England in the

late 1960s.
The cohesive materials whose frictional behaviour is described by equation
(4.7) are also referred to as Warren-Spring-type materials. Experiments with a
range of granular materials have shown that the load index n varies between 1
and 2. Nedderman (1978) has also shown theoretically that the value of the
index n cannot be larger than 2 if Mohr's circle is to be tangential to the
material yield locus everywhere (see Figure 4.4(b)).
Figure 4.4(a) demonstrates quite clearly that with cohesive materials the
individual yield loci are strong functions of the interstitial voidage, E, of the
bulk. This is in contrast to the internal yield locus of a cohesionless material
which is independent of the voidage (Figure 4.3(a)). With a cohesive
material, the end points, 0L and 'tL, of the individual yield loci are all found
to lie on a straight line which goes through the origin. This is known as the
termination locus. An effective yield locus can be drawn by taking a tangent to
the Mohr's circles containing these end points, 0L and 'tL. As seen in Figure
4.4(b) , the effective yield locus is a straight line through the origin whose
slope gives the effective angle of internal friction
(4.8)
The theoretical arguments that give rise to the effective yield locus are based
on the concept of the critical voids ratio. It is argued that at the termination
point of the yield locus, the bulk achieves a critical or limiting value of the
interstitial voidage beyond which it proves impossible to shear the material
under the given normal load, irrespective of the magnitude of the shear load
acting on the shear plane.
Further shearing is only possible if the normal load is reduced and therefore
the material is allowed to dilate prior to shear. Similar arguments can also be
advanced for the consolidation of cohesive materials. Roscoe et al. (1958)
suggest, based on the earlier work of Hvorslev (1937), that these materials
can be consolidated to some limiting value of the interstitial voidage which is
given by the termination point of the incipient yield locus, as shown in Figure
4.4(c). An increase in the shear force beyond the limiting value would simply
result in material failure along the incipient failure line before further
consolidation could take place. Hence it is possible to plot a family of
consolidation lines corresponding to different values of initial bed void age by
carrying out a series of consolidation tests in a 'consolidometer'. The exact
nature of the relationships between the consolidation load, p, and the shear
force, q, applied to the material during the test and the stresses acting on
some surface would depend very much on the design of the apparatus in
question. Figure 4.5 shows a family of consolidation lines superimposed on
the so-called internal yield loci obtained in a shear tester. It is most important
to note here that the termination points of both the yield and the
consolidation lines lie on the same straight line which is defined as the
154 Characterisation of bulk mechanical properties

(JA (J

Dilation prior to
Consolidation prior ~Iastic failure
to plastic failure 1:2
I:
--.
1:4

'tA 'tL4 'tL5 'tA

Incipient failure
at £5
y y

(a) (b)

Figure 4.5 Flow initiation in a static bed of cohesive solids: (a) plastic failure of an
underconsolidated fill; (b) plastic failure of an overconsolidated fill.

termination locus above. It now becomes relatively straightforward to

advance the argument that when flow is initiated in a static bed of material,
the underconsolidated material (where aA < ad will consolidate uniformly
under the action of an increased normal load until it reaches the so-called
critical state (Figure 4.5(a». The overconsolidated material however will
firstly have to fail along a failure line which will result in local dilation of the
material bed under a shear strain 'to Further shear will simply take the
void age to its critical value which marks the termination point of the failure
locus (aL' 'td as in Figure 4.5(b). Hence, regardless of the distribution of
voidage in the static material fill, upon initiation of flow, the voidage
everywhere will assume its critical value appropriate to the limiting stress
distribution given by the termination points of the incipient failure loci in
Figure 4.4(a). In steady flow with varying voidage, all Mohr's circles (see
section 4.6 below) will be tangential to the end points of their respective yield
loci but will also have a common tangent that passes through the origin, i.e.
the effective yield locus. Thus a cohesive material in steady motion will
 Angles of friction of cohesive materials 155

appear to be cohesionless even though individual slip planes will occur within
the flow field given by the point of tangency with the individual incipient yield
loci; these points are illustrated in Figures 4.4 and 4.5.
The result of the critical state analysis, as put forward by Roscoe et al.
(1958), Roscoe (1970) and Schofield and Wroth (1967), is that it allows us to
define the frictional behaviour of a cohesive powder in steady flow as being
independent of the value of the interstitial voidage. However, the material
properties cohesion, c, and tensile strength, T, are both strong functions of
the voidage and are proportional to the termination point, 0L, of the
individual yield loci, as seen in Figure 4.4(a):
c = feE,
ad (4.9a)
T= feE, ad (4.9b)
Roberts et al. (see under Roberts, 1984) have developed a vibrating shear cell
in which the material yield loci can be obtained as a function of the amplitude
and frequency of the vibration applied to the cell; the device is illustrated in
Figure 4.1(c). This work is significant in that it demonstrates a 'dynamic' way
of altering and controlling the interstitial void age distribution in a bed of
material, thereby significantly altering the material yield strength under
shear. Their experimental results show a strong correlation between the
cohesive strength, c, of the material samples and the frequency of the applied
vibration.
The void age distribution that results from the filling of the material into a
container will decide the magnitude of the cohesive forces at different
sections in the static fill. For failure under compression, flow will only occur if
the magnitude of the compressive stresses acting on a failure plane are larger
than the cohesive strength of the material. Once flow is initiated, the critical
state analysis described above suggests that it can be sustained indefinitely
since the bulk will act like a cohesionless material at its critical state (see
Figure 4.4(b».
In dealing which cohesive materials, the continuum mechanical approach
based on the assumption of a critical voidage state suffers from two major
drawbacks:

(i) variation of bulk cohesive strength during continuous flow from

containers is ignored;
(ii) no attempt is made to quantify cohesion in terms of the particle
properties even though it is recognised that it results from the interactive
forces between the particles making up the bulk.

As a result, even though the continuum mechanics approach can be used to

design equipment so that flow initiation in a static bed can be guaranteed
(see, for example, the work of 1enike (1967) and lenike et al. (1973)), it
756 Characterisation of bulk mechanical properties

nevertheless fails to explain the observed differences in the flow behaviour of

different materials. Nor it seems can it be used to predict the discharge rates
of cohesive materials from orifices, see Chapter 8 for details.
The field is therefore wide open for the development of an analysis which
can interpret the dynamic frictional behaviour of a cohesive material. Such an
analysis will require an understanding of the contact frictional behaviour of
cohesive particles as well as recognising the observed protraction of the flow
fields into plastic and elastic failure zone.
The effect of the adhesive forces between the particles and the walls of the
container on the wall friction behaviour of the material can be accounted for
by modifying the wall yield locus for a Coulomb-type material (equation 4.5)
to give
(4.10)
where Cw is referred to as adhesion. Equation (4.10) is a straight line with an
intercept on the shear axis of a conventional plot of 1: against o. The wall
yield locus is therefore expected to be very similar to the internal yield locus
given by equation (4.6) above for a cohesive Coulomb material. However,
there are few experimental measurements of wall friction with cohesive
materials to support the validity of equation (4.10). The scarcity of
experimental data is believed to be partly due to the difficulty of measuring

10

--
0..
..:>C

14

Figure 4.6 Wall yield loci of cohesive solids from Roberts et al. (1984). The full
curves are for brown coal with 65% moisture content and the broken curves are for
black coal with 18.7% moisture content. The curves are for the following materials: A,
D, stainless steel; B, C, polyethylene.
 Characterisation of 'f/owability' of cohesive powders 157

the wall yield locus of a cohesive material in the conventional shear cell
equipment described above, as the magnitude of the so-called adhesive (<Pw)
forces will depend strongly on the local voidage adjacent to the wall surface,
which is very difficult to measure. Figure 4.6 shows the experimental wall
yield loci of moist pulverised coals measured by Roberts et al. (1984), in
comparison with those of a Coulomb-type material given by equation (4.10)
above. Some of the experimental wall yield loci in Figure 4.6 are significantly
curved, indicating that the wall friction angle, <Pw, is a function of the normal
stress acting on the wall. This behaviour is unlike that described by equation
(4.10) which gives rise to a constant value of the wall friction angle over the
entire normal load range. In general, the Coulomb failure criterion given by
equation (4.10) fails to explain the wall frictional behaviour of a cohesive
powder material. Moreover, as the cohesive strength increases, as in very wet
materials for example, the adhesion term Cw in equation (4.10) can become
much more significant that the term <Jwtan<pw. Wet materials handled in
industry have a strong tendency to stick to the surfaces of storage vessels and
the mechanical conveying equipment, thereby causing numerous material
handling problems, as described by Bransby and Heywood (1984). Yet very
little is currently understood about the wall frictional behaviour of wet
powder beds. The effects of the variation of the values of the stress normal to
the wall and the bed voidage adjacent to the wall have been observed in
practical applications but are not as yet accounted for by a theoretical
analysis.

4.5 CHARACTERISATION OF 'FLOWABILlTY' OF

COHESIVE POWDERS (JENIKE DESIGN
METHOD)
The measurements of the internal yield loci as described in section 4.4 above
cannot by themselves be used to predict how easily (if at all) a powder of
known cohesion c (refer to Figure 4.4) will flow in a given vessel geometry.
Jenike (1961, 1964) has proposed a semi-empirical approach to quantify the
'flow ability' of a cohesive powder which to this day represents the basis for all
known design codes for mass-flow hoppers, as discussed in Chapter 8.
Jenike's method is based on the calculation of a 'flow function' (FF) for the
powder and a 'flow factor' (ff) for a given hopper geometry. These two
functions are then used to calculate the minimum critical outlet size of the
discharge vessel which will allow the powder to flow without forming a 'stable
arch' above the outlet plane. It is a well-known fact that cohesive powders
often discharge intermittently or do not discharge at all due to their tendency
to form a 'stable arch' within the hopper; see, for example, Enstad (1975) for
a detailed treatment of the subject.
J58 Characterisation of bulk mechanical properties

4.5.1 Material flow function (FF)

1enike defines the flow function,
ac unconfined yield strength
FF =- = (4.11)
01 major consolidating stress
Here the unconfined yield strength, ac , is obtained by drawing the Mohr's
stress circle which is tangential to the material yield locus and goes through
the origin, i.e. the minor principal compressive stress, au, is zero; see Figure
4.7(b). In. this Mohr's circle, the magnitude of the major principal
compressive stress, 01, defines the unconfined yield stress a c where ac = 01· ac

Active
state of Vertical
stress part

Discontinuity
Transition
Passive
state of Hopper
stress
wall

Virtual "-'"--Outlet
apex

(a)

Shear Stress 1:

Mohr's Stress Circles

,/

-
C ['------!:----::------!c=_____
o (Ix. Normal Stress 0
(b)

Figure 4.7 Jenike's analysis of stable arch formation in hoppers.

 Characterisation of 'f!owability' of cohesive powders 159

therefore gives the yield strength of the material confined between two
retaining walls under a stress-free top surface, as shown in Figure 4.7(b).
The major principal compressive stress, ar, corresponding to the flowing
state of the material is obtained by drawing a Mohr's circle through the
termination point of the internal yield locus; see Figure 4.4. The procedure is
repeated for the family of yield loci obtained at different levels of
preconsolidation during the Jenike shear cell tests (refer to section 4.2
above).

4.5.2 Jenike's hopper flow factor (ff)

Jenike defines the hopper flow factor,
ar major consolidating stress in flow
ff = - = ------------ (4.12)
a mean compressive stress on an arch
The mean compressive stress on a stable arch in a conical or wedge-shaped
hopper is calculated from

a = Qbg Yr
X-1
[a _
rr
Qbg YR ]
X-1 .
[~] x
R
(4.13)

where the variables (X, Y) are functions of the hopper half-angle, a, and the
internal and wall angles of friction, and arr is the radial component of
compressive stress, as shown in Figure 4.7, which also definesj3, the direction
angle of the principal compressive stress

2m sin<l>e [sin(2j3 + a) ]
X= +1 (4.14)
1 - sin<l>e sina

[2(1 - cos(j3 + a))]m (j3 + a)l-m sina + sinj3 sinm+l(j3 + a)

Y=-----------------------------------------------
(4.15)
where m = 0 for a wedge-shaped (plane strain) hopper and m = 1 for a
conical (axially-symmetric) hopper.

4.5.3 Jenike's minimum critical outlet span for flow

This is calculated by superimposing the hopper flow factor (ff) line over the
material flow function (FF) as shown in Figure 4.8. For flow to occur, aI/ae >
ff; that is in Figure 4.8, flow will take place when the material flow function
(FF) lies below the hopper flow factor (ft). The critical condition occurs when
a = a e , i.e. at the intersection of the two lines. The critical hopper opening
size B is given by equation
160 Characterisation of bulk mechanical properties

critical

No
Flow
Flow
o
Figure 4.8 Jenike's flow/no-flow criterion.

aH(a)
B=--- (4.16)
Qbg
where the empirical function H( a) is provided in graphical form in Figure
4.9. The major limitation of this approach is that it often predicts a minimum
outlet width of the order of a metre or more depending upon the cohesive
strength of the powder. This then means that the gravity discharge of the
powder from such a wide opening is controlled by the design parameters of
the mechanical conveyor system placed underneath. While being very
effective in the design of large-scale storage installations, the lenike method
is of limited use when designing pneumatic dosing and flow metering units
often used in the chemical, pharmaceutical and food industries where quite
small discharge rates are required coupled with strict requirements for the
solids fraction discharged.

4.6 EQUATIONS OF STRESS EQUILIBRIUM IN

BULK SOLIDS
The continuum mechanics approach to modelling bulk solids flow requires
the definition of the bulk stress (normal and shear) distributions prior to the
onset of bulk deformation (i.e. flow). The derivation of the differential
 Equations of stress equilibrium in bulk solids 161

3.0

2.5 V

--
1 -I-"'"

.,./
...- - ~ ...;.--

...;.--
E..
n

H(a) 2.0

1.5
-- -I-"'"

':J -
10' 20· 30' 40' 50' 60'
(l'

CURVE i_Circular
2_ Square
3_ Rectangular
(L ~3B)

Figure 4.9 lenike's empirical function H(a) to determine minimum critical outlet
size for flow for different orifice shapes: B = orifice width, L = orifice length.

equations of equilibrium for a general three-dimensional element in space

with dimensions x, y, z in cartesian coordinates is considered in Figure 4.10.
Figure 4.1O(a) shows the normal and shear components of the stress tensor
acting on the six faces of the rectangular block. It is possible to sum the
stresses acting in the three directions. For example, resolving the stresses in
the x-direction yields,

[oxx - (oxx + O;;X] oyox ~ ['tXy - ('tXy + °a't;y) ] oxoz

o'txz ] Ox oy + (ox oy oz) Xo = 0
+ [ 'txz - 'txz - Tz (4.17)

The other two differential equations of equilibrium in the y and z-directions

are obtained in the same way. If we now divide these equations by Ox, oy,
and oz, respectively, we reduce our rectangular element to a point in space
with coordinates x, y, z for which the following set of three equations must be
satisfied:
 y

z
~x
(jyy
~'tyx
'tyz .c t . (jzz
(jyy
(jyy

't:--±
'txy
/
/ O(jxx 0 +Xo
(jxx ~ I (jxx + ax x (j--H
~(jxx
xx
Oy )-
"0+
~
'tXyV~
l 'tyx

(jyy
o
1,,0 >-
X
+
"0'" r4'
6"';;'
'C/o'\, >,
X \:)~I rt>
rt> (b)
6"'' ' +
;f
(a)

Figure 4.10 Stress equilibrium: (a) in 3D; (b) in 2D.

 Bulk failure criterion 163

oOxx O"tXY o "txz

+ + - - +Xo=O (4. 18a)
Ox Oy oz
OOyy o"txy o"tyz
+ + - - +Yo=O (4. 18b)
oy ox oz
oOzz o"txz o"tyz
+ + - - +Zo=O (4. 18c)
oz ox oy
where X o, Yo and Zo are the components of the point body force in the three
directions. Equations (4.18) must be satisfied for all points throughout the
entire volume of the rectangular element, and at the surface they must be in
equilibrium with the external forces acting on the element.
The corresponding differential equations of equilibrium in planar two-
dimensional problems can be obtained by considering a balance over a small
rectangular block of unit thickness (Figure 4.10(b». Resolving the forces in
the x-direction, we obtain

+ = 0 (4.19)

Note that in this case -"tyx = "txy ; see Figure 4.1O(b).

Denoting the weight density of the rectangular block as y = Qg, we can
write a similar balance in the y-direction with the result

(4.20)

Equations (4.19) and (4.20) together define the state of stress equilibrium in
planar geometry.

4.7 BULK FAILURE CRITERION

The stresses acting on all planes of a solid body can be represented graphically
using a Mohr circle of stresses as shown in Figure 4.11. The Mohr circle is the
locus of all stresses OJ, OJ and "tij acting on each plane of the solid body.
Hence, a point on the Mohr circle defines the necessary normal and shear
stress components acting on a single plane. In order to initiate motion within
the solid body at least one point on the Mohr circle should correspond to a
failure plane. The location of the failure plane on the Mohr circle is obtained
by the tangency of the material yield locus to the Mohr circle as shown in
Figure 4.11. If the material yield locus is given by a straight line as described
in section 4.3 above, then the material is referred to as a Coulomb-type
J64 Characterisation of bulk mechanical properties

,
I
I
,
I
I
I
:0:
0: :,7l
7~ Element of bulk solid
t,Ojl OJ

(a)

Shear stress 't

Mohr's stress circles

o (JI Normal stress (J

(b)

Figure 4.11 Mohr-Coulomb failure criterion. 0 = effective internal friction angle.

See Nomenclature.

material and the equation of material failure is given by equation (4.4). The
construction in Figure 4.11 is often called the Mohr-Coulomb failure
criterion. This is the most popular theoretical model used to describe the
onset of plastic failure (i.e. flow) in granular solids. The model is essentially
two-dimensional and it does not take into account the influence of the
intermediate normal stress (Jzz) component in a three-dimensional element
(see Figure 4.10). In dealing with axially-symmetric geometries, the Mohr-
Coulomb failure criterion can be applied by assuming that

or

where (Jxx and Oyy are the principal normal stresses. This assumption is often
referred to as Harr and Von Karman's hypothesis; see for example
Nedderman (1992). More sophisticated failure criteria (e.g. Von Mises)
which take into account the variations in the intermediate principal stress and
the 'conical' nature of the yield surfaces in the three-dimensional flow have
been applied to predictions of granular flow fields in mass flow. Cleaver and
 Description of bulk solids stress states 165

Nedderman (1993) report little difference between the predictions based on

the Mohr-Coulomb criterion and more sophisticated yield criteria. However,
they do question the validity of Von Karman's hypothesis in axially-
symmetric flow. For a more comprehensive account of the three-dimensional
bulk failure criteria, the reader is referred to Drescher (1991) and Nedderman
(1992).

4.8 DESCRIPTION OF BULK SOLIDS STRESS

STATES
Depending on the direction of the major principal compressive stress, OJ, the
material inside a bin-hopper system can either be in the 'active' or 'passive'
state of failure. The concept of stress states was first introduced by Rankine
(1857) for a semi-infinite body of soil retained behind a single wall.
According to Rankine (1857), for a given weight of static material,
Oyy = Qgh (4.21)
there exists a stable range of values of 0xx which will not cause failure. The
lower limit of 0xx is given by the 'active state' of failure, which determines the
least force required on the retaining wall to stop the material flowing
outwards (Figure 4.13). In 'active' failure, therefore, a divergent flow pattern
will occur, i.e. the material will expand horizontally and as a result contract
vertically.
The upper limit of 0xx is given by the 'passive state' of failure, which is
associated with the force on the wall required to push the material retained
behind it inwards. Hence, in 'passive' failure, a convergent flow pattern will
be established, i.e. the material will contract horizontally and hence be forced
to expand vertically (Figure 4.13).

Figure 4.12 Rankine (1857) stress states.

166 Characterisation of bulk mechanical properties

Figure 4.13 shows the Mohr's stress circles corresponding to 'active' and
'passive' states of failure. Note that in the 'active' state, the direction of the
major principal stress, Or, is near to the vertical (i.e. Oyy> oxx), while in the
'passive' state, Or is close to the horizontal (i.e. Oyy < oxx).
In practice, the filling of a bunker results in the bunker walls bulging
slightly outwards and consequently, the initial material fill is generally
associated with an active state of stress. However, the material in the
converging section of a bin-hopper system is forced to contract horizontally
and thus expand vertically during discharge. This implies that in such a
geometry, the direction of the major principal compressive stress is near to
the horizontal, suggesting a passive state of failure.
On the basis of the above arguments, a discontinuous change from one
stress state to the other is to be expected in the bunker during flow initiation.
The experiments of Handley and Perry (1965,1967), Perry and Jangda (1970)
and Rao and Venkateswarlu (1974, 1975) confirm the above hypothesis.
These workers were able to measure point values of vertical and horizontal
stress components at various points along and across a bunker by tracing the
signals of a radio-pill. Their results show that after the initial filling, the
material in the bunker is in the active state, i.e. the direction of the larger
stress component is nearly vertical. However, during steady flow, they found
that the material in the parallel-sided bin section remains in the active state
while in the converging hopper section, the major principal stress rotates to

(a) (c)

Passive
Active

Figure 4.13 Major principal compressive stress direction 01 corresponding to

different stress states in a silo. (a) Static fill; (b) flow initiation; (c) mass flow.
 Bulk solids flow rules 167

the horizontal direction giving a passive stress field. Near the bottom corners
of the bin section, known as the 'transition corners' (see Figure 4.13), their
results also show 'peak stress' values due to transition from the 'active' to the
'passive' stress state, as predicted by lenike (1961, 1964), and later by Walters
(1973a, b).

4.9 BULK SOLIDS FLOW RULES

Continuum mechanics analyses of bulk solids flow attempt to predict the
velocity distributions from an a priori stress distribution given by the assumed
failure criterion and the bulk material stress state as described in the
preceding sections. In order to generate strain rate, and hence velocity fields
from the given stress distributions, it is necessary to define flow rules which
are the constitutive equations relating the bulk stresses to the bulk strain rates
during flow. This is often expressed mathematically in terms of an angle
(sometimes referred to as the angle of dilation) which relates the principal
strain-rate directions to the orientation of the principal normal stresses in the
flowing bulk. The values of the angle of dilation range between nl2 and zero,
where the former limit refers to the perfectly plastic flow (i.e. infinite
dilation) flow limit and the latter to the incompressible (i.e. zero dilation)
flow limit. For a comprehensive discussion of the use of flow rules in coupling
stress and strain-rate fields in granular flow, the reader is referred to Drescher
(1991) and Nedderman (1992).
With real granular materials, the angles of dilation appear to have finite
values which can be shown readily to fall within the range nl4 ± <1>/2 where
<I> is the angle of internal friction. Figure 4.14 marks the direction of the

CJ

b-slip line

Figure 4.14 The direction of the slip planes in cohesionless granular material at
incipient flow.
768 Characterisation of bulk mechanical properties

material 'slip planes' during flow in relation to the direction of the major and
minor principal stress planes. Unlike the experimental procedures described
above for measuring internal and wall friction angles of bulk materials, no
established experimental technique currently exists which can be used to
quantify the so-called 'angle of dilation'. It is this angle and its prediction
which holds the key to the definition of a characteristic flow rule for a given
granular material.

4.10 EMPIRICAL MEASUREMENTS OF

COMPRESSIBILITY OF BULK SOLIDS
The compressibility of a granular material bed has a significant effect on the
measured values of the friction angles and bulk cohesion obtained in the shear
cell measurements described in section 4.4 above. Both the magnitude of the
internal friction and of the bulk cohesion are shown to increase considerably
when the bulk material is consolidated, as illustrated in Figures 4.5(a) and
(b).
Traditionally, the compressibility of the bulk material is measured by
confining a sample of known weight in a cylinder (where the diameter
should exceed typically 30 particle diameters) and applying known weights to
the lid as shown in Figure 4.15. With fine powders which are susceptible to air
entrapment during filling, it is important to use a perforated lid to allow the
trapped air to escape during consolidation.
The initial bulk density of the material in the 'poured-in' state is given by

Om = W/A
t 0,

~
°2=°3
T """ ..,r
o
-- "-

(a)
t
(b)

Figure 4.15 Schematic diagram of uniaxial compression apparatus.

 Empirical measurements of compressibility of bulk solids 169

Qbo = (4.22)
AHbo
where Wb and Hbo are the weight and the initial height of the bulk sample and
A is the cross-sectional area. The mean value of the bed void age is then given
by
Qbo
1--

= 1 - (4.23)
QsAHbo
The values of bed void age corresponding to different uniaxial consolidation
loads are measured by monitoring the reduction in the sample height Hb in
the cylinder, i.e.

(4.24)

which allows a plot of the mean bed voidage as a function of the uniaxial
consolidation load. Figure 4.13 shows some typical results obtained with
different granular materials. In general, over a relatively small range of
consolidation loads, a linear relationship of the form
(4.25)
can be used as a best fit to experimental data, where the empirical coefficient,
a, is a material property and Oc is the applied load divided by the cross-
sectional area. However, over a larger range of consolidation loads, a
logarithmic relationship of the form
(4.26)
is more appropriate. The distribution of the applied load across the bulk
material sample will undoubtedly be affected by the angle of the material top
surface as well as by the wall friction. To minimise these effects, it is
important to use cylinders with a height-to-diameter aspect ratio >2 and a
ratio of cylinder diameter to average particle size >30.
Equation (4.20) approaches a rather 'ill-defined' asymptotic limit at the
large values of the applied load, so that it can be rewritten as
1
--- = Vo - AlnP (4.27)
1- E
where Vo is an empirical constant and the coefficient A is often referred to as
the compressibility coefficient in the soil mechanics literature (Roscoe et at.,
1958). An analogous expression in terms of the form is also possible
770 Characterisation of bulk mechanical properties

1 Vo
- =- - (A/Qs)lnP (4.28)
Qb Qs

where Qb and Qs are the bulk and solid densities, respectively, and P is the
applied pressure. With cohesionless bulk materials such as glass ballotini,
coarse sand and table salt (see Figure 4.16), typical values of the compress-
ibility coefficient are reported to be in the range 0.1-0.01 over a range of
consolidation loads up to 10 bar; see for example Crewdson et al. (1977) and
Thorpe (1992). With fine powders «100 !-lm) of irregular shape, and bulk
materials which exhibit multimodal size distributions (sauce mix and icing
sugar in Figure 4.16), the compressibility of the bed of bulk material is much
higher; as much as 30-40% change in bulk density is obtained at consolidation
pressures as small as 0.02-0.05 bar. With cohesionless materials, the change
in bulk density under such small pressures is often <1 or 2%. Hence with
cohesionless materials, the coefficients of friction measured in the shear cell
experiments described earlier will be independent of the applied consolidating
pressure. When the bulk material is significantly compressible, the values of
bulk cohesion and of the angles of bulk friction are strongly affected by the
degree of consolidation.
The compressibility of a bulk material bed also affects its permeability to

C
0
U
jg
en
:!2
(5
en
c
nl
Q)
:::i:

Mean consolidating stress (om)

Figure 4.16 Typical compressibility curves for various granular materials. 1. Table
salt, granulated sugar, fine sand (100 !-tm < d < 500 !-tm). 2. Sauce mix (d < 50 !-tm). 3.
Flour, corn flour, silica (d < 20 !-tm). 4. Icing sugar (d < 50 !-tm).
 Nomenclature 171

fluids, thereby influencing its fluidisation behaviour (see Chapter 6) as well as

its discharge rate from storage vessels (see Chapter 8).

4.11 NOMENCLATURE
A Cross-sectional area
B Orifice width, critical outlet width
C Cohesion
Cw Adhesion
D Diameter
d Particle diameter
Hb Initial height of the bulk
H (a) lenike's empirical function
L Length
MYL Material yield locus
P Pressure = W/A
r Radial distance
T Tension/tensile strength of the material
W Normal load
Wb Weight of the bulk
X, Y Empirical constants
X o, Yo, Zo Components of gravity in 3D space

Greek letters
a Hopper half-angle
j3 Angle defining principal compressive stress direction, pressure
coefficient
y Weight density = Qg
E Interstitial voidage
Voidage corresponding to zero consolidating stress
Compressibility coefficient
Coefficient of internal friction
Coefficient of wall friction
Empirical constant
Q Bulk density
Qb Initial bulk density
Qs Solid density
a Normal stress
OC Unconfined yield stress
0L, 'tL Termination point of the incipient yield locus
am Mean compressive stress
Ow Normal stress at wall
172 Characterisation of bulk mechanical properties

Major and minor principal compressive stress

Shear stress
Shear stress at wall
Angle of internal friction
Effective angle of internal friction
Angle of wall friction

REFERENCES
Augenstein, D.A. and Hogg, R. (1974) Powder Technol. 10,43.
Bransby, P.L. and Heywood, N.J. (1984) Needs for research on the handling of wet
particulate solids. Warren-Spring Laboratories, UK Department of Trade and
Industry (CHISA '94 Prague, Czechoslovakia).
Briscoe, B.J., Scruton, B. and Willis, F. R. (1973) Proc. Roy. Soc. (London) A333,
99.
Briscoe, B.J., Pope, L. and Adams, M.J. (1984) Powder Technol. 37, 169.
British Materials Handling Board (1985) Draft Code of Practice for the Design of
Silos, Bins, Bunkers and Hoppers.
Carr, J.F. and Walker, D.M. (1968) Powder Technol. 1,370.
Cleaver, J.A.S. and Nedderman, R.M. (1993) Chem. Engng. Sci. 48, 3693.
Cooker, B. (1980) PhD Thesis. Cambridge University.
Coulomb, C.A. (1776) Memoires de Mathematique de I'Academie Royal des Sciences,
vol. 7, Paris, p. 343.
Crewdson, B.J., Ormond, A.L. and Nedderman, R.M. (1977) Powder Technol.
16, 197.
Drescher, A. (1991) Analytical Methods in Bin-Load Analysis, Elsevier Science,
Amsterdam.
Enstad, G. (1975) Chem. Engng. Sci. 30, 1273.
Haaker, G. and Rademacher, F.J.C. (1982) in European Symposium on Storage and
Flow of Particulate Solids, EFCE, Braunschweig, Germany.
Handley, M.F. and Perry, M.G. (1965) Rheologica Acta 4, 225.
Handley, M.F. and Perry, M.G. (1967) Powder Technol.l, 245.
Hvorslev, M.J. (1937) Ingenior Vidensk Skr. 45.
Jenike, A.W. (1961) Gravity Flow of Bulk Solids, Utah Engineering Experimental
Station, University of Utah, Bulletin 108.
Jenike, A.W. (1964) Storage and Flow of Solids, Utah Engineering Experimental
Station, University of Utah, Bulletin 123.
Jenike, A.W. (1967) Powder Technol. 1,237.
Jenike, A.W., Johanson, J.R. and Carson, J.W. (1973) Trans ASME. B95, 6.
Nedderman, R.M. (1978) Powder Techno!. 19, 287
Nedderman, R.M. (1992) Statics and Kinematics of Granular Materials, Cambridge
University Press, Cambridge.
Nedderman, R.M. and Laohakul, C. (1980) Powder Technol. 25, 91.
Perry, M.G. and Janda, H.A.S. (1970) Powder Techno!. 4,89.
Rankine, C. (1857) Phil. Trans. Royal Soc. (London), 147,9.
Rao, V.L. and Venkateswarlu, D. (1974) Powder Technol. 10, 143.
Rao, V.L. and Venkateswarlu, D. (1975) Powder Technol. 11, 133.
Roberts, A.W. (1984) in Handbook of Powder Science and Technology (M.E. Fayed
and L. Otten, eds), Chapter 6, Van Nostrand Reinhold, New York.
Roberts, A.W. Ooms, M. and Scott, O.J. (1984) Trans. Aust. Institute Mech. Engrs
C(355) , 123.
 References 773

Roscoe, K.H. (1970) Geotechnique 20, 122.

Roscoe, K.H., Schofield, A.N. and Wroth, C.P. (1958) Geotechnique 8,22.
Rose, H.E. and Tanaka, T. (1959) The Engineer (London) 208, October, 29.
Schofield, A.N. and Wroth, C.P. (1967) Critical State Soil Mechanics, McGraw-Hill,
London.
Schwedes, J. (1979) Vergleichende Betrachtungen sum Einsatz van Schergeraten sur
Messung von Schuttguteigenschaften, Institut fUr Mechanisehe Verfahrenstechnik.
Technische Universitat Braunschweig, Germany.
Schwedes, J. and Harder, J. (1985) in Proceedings of the International Symposium on
Particulate and Multiphase Processes, Hemisphere Publications, New York, USA,
Vol. 1, 703.
Smid, J. (1983) Grundl. Landtechnik Bd. 33, 72.
Thorpe, R. (1992) Chern. Engng. Sci. 47,4295.
Tiiziin, U. (1982) Post-Doctoral Dissertation, Department of Chemical Engineering,
Cambridge University.
Tiiziin, U. and Nedderman, R.M. (1982) Powder Techno!. 31, 27.
Tiiziin, U. and Nedderman, R.M. (1983) Bulk Solids Handling, Trans. Tech. Publ.
3,507.
Tiiziin, U. and Nedderman, R.M. (1985) Chern. Engng Sci. 40,337.
Tiiziin, U., Adams, M. and Briscoe, B.J. (1986) Proceedings of the Second
International Conference on Bulk Materials Storage, Handling and Transportation,
Institute of Engineering, Australia.
Tiiziin, U., Adams, M.J. and Briscoe, B.J. (1988) Chern. Engng Sci. 43, 1083.
Walker, D.M. (1967) Powder Techno!. 1, 228.
Walters, J.K. (1973a) Chern. Engng. Sci. 28, 13.
Walters, J.K. (1973b) Chern. Engng. Sci. 28,779.
Williams, J.C. and Birks, A.H. (1965) Rheologica Acta 4, 170.
5
Assembly mechanics t

5.1 INTRODUCTION
Study of the mechanical behaviour of an assembly of solid particles in contact
with each other or with a wall surface is essential to a significant number of
process engineering applications involving bulk solids. The effects of single
particle properties on the bulk mechanical behaviour of granular materials is
an area of significant interest to process and materials engineers which has,
however, received little attention in traditional civil engineering and soil
mechanics literature (see, for example, Jenike, 1964; Schofield and Wroth,
1967). Historically, this is a direct consequence of the overwhelming
popularity of a continuum mechanics approach in mathematical modelling of
the granular flow and stress fields (Spencer, 1964; Pariseau, 1969a,b). In civil
and mechanical engineering applications, the internal and wall friction
coefficients and cohesion are traditionally measured as 'bulk' properties in
planar shear test equipment such as the apparatus described in Chapter 4.
The past two decades have also seen the development of more sophisticated
bi-axial and tri-axial bulk testers in which it is possible to exercise more
accurate control of the stress distribution and the interstitial void age during
both consolidation and shear of the material samples. However, the data
from such equipment are more difficult to generate and invariably require
more complex theoretical analysis than the results of the simple planar shear
cell tests. The major shortcoming of the bulk mechanical failure tests is that
they cannot be used to explain the effects of the single particle frictional
properties on bulk flow.

5.2 ASSEMBLY MODELLING OF WALL FRICTION

IN A FLOWING PARTICULATE BED
It is well known that a static bed of granular material experiences dilation as
motion is initiated within the bed. Figure 5.1 shows (a) an assembly of closed-
packed spheres in contact with a flat wall, in comparison with (b) a dilated
assembly structure. Here S is taken as the distance between two point

tThis chapter was written in collaboration with Dr Paul Langston, Department of Chemical
Engineering, Nottingham University, UK.

J. Seville et al., Processing of Particulate Solids

© Chapman & Hall 1997
 Assembly modelling of wall friction in a flowing particulate bed 175

WALL WALL
• ••
•••••
•
•
•••••
•••••
• ••
.....
.:.:. ~
••
•••••
S>2R

Load Cell
Active Surface

Figure 5.1 Packing at the silo wall. (a) Close packed (static); (b) dilated (flowing).

contacts on the wall surface. It is possible to define the interstitial void space
at a distance R from the flat wall plane in terms of S
mrcR2
E=l---- (5.1)
SZ
where m = 1 for square packing and m = 1.155 for hexagonal close packing of
touching spheres, i.e. S = 2R. For the dilated assembly, S > 2R and the
voidage will scale as
1- El S~
----- (5.2)
l- E2 •• Si
where the subscripts 1 and 2 refer to the different states of interstitial voidage,
as shown in Figure 5.1(b). The maximum value of the flowing voidage
attainable by a bed of coarse spheres is given by the voidage at minimum
fluidisation, Emf (see section 6.3). Measurements of E made in two-
dimensional beds by numerical integration of the particle velocity profiles
next to a flat transparent wall seldom result in values in excess of 0.45 (see,
for example, Laohakul, 1978, and Tiiziin and Nedderman, 1979). Substitu-
tion of this maximum value for E in equation (5.1) yields a maximum value of
S = 2.4R for square packing and S = 2.6R for hexagonal packing. In general,
factors like the particle shape and the width of the size distribution are likely
to influence the limiting values of the voidage obtained. Furthermore, when
flow is initiated in a static bed, there is no reason to believe that the packing
geometry will remain unaltered. However, it is difficult to envisage
circumstances which will give S > 3R.
176 Assembly mechanics

The number of points of contact with the surface varies inversely with the
square of the spacing, S. Hence, for an area of surface A * (the area of a load
cell for stress measurement, for example)
N=A*/S2 (5.3)
Therefore, for a closely-packed bed N = A */4R2.
We can now modify the single particle friction equation resulting from the
adhesion model of friction for elastic point contacts (see section 3.3) to
include the effect of the number density of particles in contact with the silo
wall.
For each contact point, say
of= :rt'to(E*R) 'hOWl) + aoW (5.4)
For N point contacts, F = NoF and W = NoW and substituting for N from
equation (5.3) gives

(5.5)

Thus
E*R )2/1 1
( -S- - '11 +a (5.6)
~

It is noteworthy that this equation predicts that f.tw is independent of particle

radius. The above analysis assumes that both the single-particle and powder
friction are governed by a 'mean contact pressure', P where P = W/A.
Actually there is a parabolic Hertzian pressure distribution across the single-
particle contact (see Chapter 3) and powders exhibit multimodal pressure
profile. The interfacial shear stress 't is a linear function of the pressure and it
is possible to show that the use of a 'mean pressure' to compute the frictional
force is equivalent to introducing the true pressure distribution (see Appendix
AS).

5.2.1 Normal stress and dilatancy

Equation (5.6) should, in principle, allow for the wall friction coefficient to be
calculated at any height along the silo wall for both static and flowing
conditions. Experimental wall stress profiles presented in Figure 5.2 for three
free-flowing materials show quite clearly that the wall stress levels experienced
during flow are several times higher than those obtained in a static bed at the
same height. These results are therefore in agreement with the well-known
theory of the 'stress switch' from the active to the passive state when flow is
initiated in a static bed (see Chapter 8).
 Assembly modelling of wall friction in a flowing particulate bed 777

- - Mustard
seeds
- - - - - - Polythene
beads

.
.
2

-- 3
0
I
...
··
4 ;.
·
Static Static Flowing

50 6

0
Glass Ballotini

--
0
I
H 3

4
Static Flowing
5

60
-t 12

Figure 5.2 Variation of stress with height in a silo.

All of the stress profiles presented in Figure 5.2 approach an asymptotic

limit with depth below the material top surface. These stress asymptotes can
be predicted using the well-established theory of Janssen-Walker (see
Chapter 8) for specific values of the internal and wall coefficients of friction.
However, these analyses do not allow for the variation of the wall friction
coefficient with depth inside the silo; nor do they recognise a difference
between the static and flowing bed values. The predictive capacity offered by
equation (5.6) above need not, however, be restricted by any of the
assumptions essential to the Janssen-Walker analysis.
778 Assembly mechanics

To achieve a general predictive capacity, it is imperative to couple equation

(5.6) with a function for the spacing, S, of the form
(5.7)
If such a function could be defined explicitly then the wall friction coefficients
become predictable both as a function of height and the state of flow within
the silo.
The exact nature of the relationship between the local bed porosity E and
the compressive stress a still remains unresolved especially at the low stress
levels measured in experimental silos. However, there is some empirical
evidence to suggest a logarithmic dependence of voidage on the compressive
stress under very high loading conditions (see, for example, Schofield and
Wroth, 1967, and Roscoe et al., 1958). A relationship of the form

E = Eo - C In [:0 J (5.8)

will predict dilation as the vertical loads during flow, 0, are significantly less
than those given by the static bed values 0 0 • Equation (5.8) is very similar in
form to equation (4.26) except here the reference state (Eo, 0 0 ) is well
defined. Since 0/0 0 < 1 then E > Eo where Eo is the value appropriate for
a static bed of closely-packed spheres. Here C is an empirical coefficient
whose values lie in the range 0.1-0.01. A similar expression can be used to
explain the variation of voidage with depth inside the silo. However, in this
case, the experimental stress profiles presented in Figure 5.2 indicate an
almost negligible variation of bed voidage with height except (i) immediately
below the top surface and (ii) close to the silo outlet where the stress levels
are found to change significantly with depth. This is typical of 'cohesionless'
materials whose internal failure loci remain independent of the interstitial
voidage. In line with Janssen's analysis described in Chapter 8, the
relationship between the vertical stress (Oyy)w at the wall and its normal
component is given by
(5.9)
where Kw is known as Janssen's pressure coefficient at the wall plane and can
be calculated for the active, K wA , and passive, K wp , stress states, as described
in detail in section 8.5. We can therefore rewrite equation (5.8) as

(5.10)

where 0wo is the static bed wall stress and Ow is the normal stress measured
during flow. For most cohesionless materials, the internal angle of friction is
about 30° and this would give values of the order of 0.1 for the ratio KWA/ Kwp.
It is conventional in soil consolidation tests to relate the mean value of the
 The distinct element method 779

bulk void age to the mean isotropic stress so that effects due to anisotropic
loading could be averaged over the sample volume (see, for example,
Schofield and Wroth, 1967 and Roscoe et al., 1958). In equations (5.8) and
(5.10) above, the vertical component of the wall stress (Oyy)w is used rather
than the mean stress Om, where Om = 0.5(1 + Kw)ow' If the mean isotropic
stress is used then the term KWA1Kwp in equation (5.9) is replaced by (1 +
KwA/1 + Kwp) which has now a value of about 0.33 for the same material.
Equation (5.9) can be rewritten as

C = Co + C' - C In ( Ow ) (5.11)
Owo

where the new empirical constant C' is decided by the assumption about the
controlling stress component. If vertical stress is assumed to be controlling
then C' is about 0.23 with C lying close to the limiting value of 0.1 (refer to
equation (5.8». If however, the isotropic mean stress is controlling then C'
= 0.11. The experimental wall stress profiles presented in Figure 5.2 indicate
a three- to fourfold increase in the wall normal load due to flow initiation in a
static bed. Under these circumstances, equation (5.10) will predict c = co,
i.e. negligible or no dilation at the bed wall interface if the voidage is assumed
to be controlled by the mean isotropic stress at the wall. If, on the other hand,
the vertical stress component at the wall is assumed to be controlling then c
= 1.5-2.0co with a three- to fourfold increase in the wall normal load (see
Figure 5.2). It is well known that a dilated shear zone of some two to five
particles across is created when a bed of coarse spheres is allowed to flow next
to a vertical flat wall (see Chapter 8). The above results appear to suggest that
this observed dilation is predicted correctly if the vertical stress at the wall is
assumed to be controlling the interstitial void age at the bed wall interface.
However it must also be born in mind that equation (5.11) is an empirical
form with two adjustable parameters, C and C', which makes it difficult to
state unequivocally the precise nature of the relationship between the bulk
stresses and the interstitial voidage at the bed wall interface. Clearly, more.
systematic measurements of the interface voidage are required as a function
of the state of flow, and hence the value of interparticle separation distance S.
The above analysis is essentially two dimensional; a more complex analysis
is required to model assembly wall friction against a curvilinear surface in an
axially symmetric container. This would also require a more specific model of
the particle packing structure in three dimensions.

5.3 THE DISTINCT ELEMENT METHOD

5.3.1 General features
The distinct element (DE) method, sometimes referred to as the discrete
element method, is becoming increasingly popular for simulation of particle
780 Assembly mechanics

assemblies. Using this technique, the position of each element is evaluated

and the interactions between the elements subsequently calculated. The
interactions may then subsequently affect the element positions. Elements
usually represent individual particles, but they might represent clusters. The
simulation may apply to a static assembly structure, but more often it is used
in a dynamic way. The method is closely related to the established finite
element (and finite difference) technique, but with the important difference
that in DE models each particle is integrally separate in its behaviour from its
neighbours. Figure 5.3 illustrates the basic features.
There are many variations on the DE method and many areas of
application. This book does not include a consideration of molecular
dynamics models (Haile, 1992) which spawned the DE technique. The
cellular automata method is briefly described. Other methods include
Newtonian dynamics and stochastic techniques to determine the element
dynamics.
This chapter is restricted to applications of granular systems, i.e. it does not
cover atomic particles or, on the other scale, traffic flow problems. The
method can be used in applications such as granular flow in silos, and in the
analysis of assembly properties such as simulation of flow in a shear cell, for
example. The basic advantage of the method over continuum techniques is
that it simulates effects at the particle level. Individual particle properties can
be specified directly - such as size and shape variation - and the assembly
response is a direct output from the simulation. There is less need for global
assumptions, such as uniform stress at a certain depth in the assembly.
Certain phenomena such as particle segregation can be simulated directly,
whereas in the continuum method this is far more difficult. However, the
disadvantage is that considerable computational resources may be required to
use the DE method. Even with current computing power, simulations can
usually only reasonably be undertaken with a relatively small number of
particles, by comparison with the total number of particles in the real system
(10000 particles is a small salt cellar not a large hopper!). However, the
principal use of current DE models is to gain an insight into the physics of the
system, in translating single particle properties into bulk phenomena. The DE
model should be viewed as complementary to the continuum models and
experimentation described in Chapters 4 and 8.

5.3.2 History and current applications

Pioneering work in the application of the method to granular systems was
carried out by Cundall and Strack (1979). They developed the program BALL to
simulate assemblies of discs. Out of this model the program TRUBAL was
developed which is now in wide application (see, for example Thornton,
1993). Early work concentrated on the use of 'springs and dashpots' to
represent particle interactions. Some researchers have more recently applied
 The distinct element method 181

Wall Particles
(a)

(b)

(c)

(d)

Figure 5.3 Principle of the distinct element Newtonian method - see equations (5.11)
and (5.12). (a) Position at time t; (b) interaction forces; (c) net acceleration (including
gravity); (d) position at time t + nl'o.t (several time steps later).
182 Assembly mechanics

detailed contact mechanics to these interactions. Langston et al. (1994, 1995)

have compared such approaches and also applied generic interactions, well
established in molecular dynamics simulation, where the particle interaction
is on a particle sized scale which leads to a greater connectivity in assemblies.
The DE method (TRUBAL and other models) is now being used in areas
such as the simulation of packing geometries (Satake, 1993), agglom-
erate collisions (Kafui and Thornton, 1993), quasi-static assembly strength
(Ouwerkerk, 1991), vibrating beds (Taguchi, 1993), pneumatic conveying
(Tanaka and Tsuji, 1993), avalanches, hopper flow (Langston et at., 1995),
fluidised beds (Tsuji et at., 1993), mixing, trickle bed reactors (at Shell,
Amsterdam), stockpiles, powder forming and metallurgy (Aizawa, 1993).
These applications range from static to slow flow, to fast-flow situations, to
low-high stress/strain systems, some in a multiphase assembly, some with
chemical reactions and heat transfer. In short the technique is extremely
versatile. Figure 5.4 shows an example of filling and discharging a silo and
Figure 5.5 illustrates mass flow and funnel flow in two-dimensional silos.

5.3.3 Methodology
The basic methodology of DE models is shown in Figure 5.3 for a dynamic
Newtonian analysis. First, the particle positions are recorded. From these the

Figure 5.4 Example DE application - silo fill and discharge.

 The distinct element method 183

Figure 5.5 Illustration of funnel flow and mass flow in silos.

particle interactions are determined and then the subsequent dynamics are
evaluated. The particle positions are determined at certain time intervals.
These are often regular (t, t + I:1t, t + 21:1t, etc.) but some analyses are event-
based, e.g. conditions are assumed constant until the next collision. Constant
or linearly varying conditions are assumed over the time interval, and hence
the time interval must be small enough to maintain sufficient accuracy. The
effect of the magnitude of the time step is discussed later.
In Newtonian dynamics, all the forces acting on each particle are
determined (e.g. friction at contacts, gravity) and then the net acceleration of
the particle is determined, both linear and angular. The positions and
orientations at the end of the next time step are then determined using an
explicit numerical time integration such as

(5.12)

x t+M = Xt + (x t + xt+M )O.5!1t (5.13)

This is the Euler method of integration. Other techniques, such as Runge-

Kutta, are also used. These are called explicit methods because the positions
at t + At are obtained directly from the acceleration at t. In FE models
(Smith and Griffiths, 1982) an implicit method is sometimes adopted. In this,
an iteration is undertaken between t and t + I:1t to satisfy the assembly matrix
differential equations at t + I:1t. This is a more accurate method and can
tolerate a larger time step, but the analysis is more complex and requires
more computer memory. It may not be feasible with some DE models since
the geometric relationships between the elements can alter more than in FE
 184 Assembly mechanics

models. Haile (1992) gives more details of these methods as applied to

molecular dynamics.
In stochastic models, which are appropriate where chaotic effects are
significant, random numbers are used to displace particles. At appropriate
times, particles are given small random displacements or velocities and the
consequences for the assembly are then followed through. Figure 5.6
illustrates such an analysis on the effects of vibrations on a granular pile. A
vibration is represented by small vertical displacements scaled to the
magnitude of the disturbance, combined with a lateral displacement sampled
from a normal distribution. The particles are then allowed to settle without
being permitted to overlap; see Barker and Mehta (1992).
The method of Cellular Automata has been applied to silo discharge
analysis by Martinez et at. (1995), as illustrated in Figure 5.7. Here particles
move along a lattice grid according to certain rules. Each cell can be occupied
by one particle or it is empty. At each time step a particle is set to move to a
neighbouring cell as a result of a number sampled from a probability
distribution. Each empty cell is similarly set to receive a possible particle. The
particle is then actually moved if it has found a compatible cell. This
technique is efficient at simulating two-dimensional granular flow, but it does
not appear to have been widely used in three dimensions. It also seems that it
is more difficult to analyse stresses (such as those at a silo wall) using such a
model.
A critical aspect of DE models is that of contact identification, i.e. which
particle is touching which? Obviously to test every possible pair of particles in
a large system would be inefficient. Since the number of particles which can
. be simulated in a reasonable time is the principal limitation of DE methods,

(c)

Figure 5.6 Stochastic simulation of a vibrating assembly. Thc original pile (a) and
after a large shake (b) and after a small shake ( c). Insets show contact network of
highlighted cluster. Reproduced with permission from T.A.J. Duke, G.c. Barker and
A. Mehta. A Monte Carlo Study of Granular Relaxation, published by Institute of
Physics, 1990.
 The distinct element method 185

N
~
NW~2*1 O-rNE

SW
~3
4
5,.. SE
'f
S
Request process Exchange process
(a) (b)

Gravity
action

W Kinetic

~
Interactions energy

W ! W
~ ~

Probability
distribution

Reaction test with neighbouring

cell in direction k

Exchange
(d)
(c)

Figure 5.7 Use of Cellular Automata. (a) Orientation and numbering around a site.
(b) Request-exchange process governing grain displacements. (c) Transition rule
diagram. (d) Silo flow. Reproduced with permission from J. Martinez, S. Masson and
D. Deserable. Flow Patterns and Velocity Fields during Silo Discharge Simulation with
a Lattice Grain Model. PARTEe '95, published by Niirnbergmesse GmbH, D-90471
Nuremberg, 1995.
186 Assembly mechanics

an efficient algorithm for contact identification is important. There are two

common methods to speed up the analysis: near-neighbour schemes and
zoning. In the near-neighbour scheme the program identifies all particles
within a certain small distance of each particle. The detailed analysis of
possible interaction in subsequent time steps is then made with these particles
only. The 'neighbour lists' are recompiled whenever any particle has moved
more than a certain distance, such that the old lists might miss a particle-
particle contact. With geometrical zoning, the unit is divided into zones. Each
particle is tested against other particles in the same zone and the neighbouring
zones only. (The Cellular Automata method is a special case of zoning.) It is
also important to note that the history of the contacts may be required; for
example, in the friction force calculation, the tangential displacement vector
developed over several time steps will be needed. Hence it is important to
store these correctly in the neighbour lists. With 'infinite systems' the
application of periodic boundaries enables the problem to be simulated with a
relatively small number of particles as shown in Figure 5.8. A particle moving
out at the top boundary reappears at the lower boundary. This technique is
widely used in molecular dynamics.
Another important feature is that models may in general be two or three
dimensional. (Here two-dimensional implies planar elements moving in a
plane, i.e. x- and y-coordinates varying, with a constant z-coordinate). Two-
dimensional simulations are obviously simpler and faster, and since it is easier
to 'see what is happening' in the simulation, they can be very instructive.
However, very few industrial granular process units even approximate to two-
dimensional behaviour. A few process systems may be genuinely modelled in
this way: cigarette packaging and bottling lines, for example (although these
are beyond the scope of this book). Three-dimensional simulations are more
complex because the contact analysis is considerably more involved, the
number of contacts is greater and the number of particles for a similarly sized
unit is greater. Presentation of graphical results is also more difficult.
The time step is a critical parameter. If it is too large the errors become

L/. • • ~ .~
•

{j
•
• • ••
• •
• • • • •••
• • •
• • • • •
•• •

Figure 5.8 Periodic boundaries. A two-dimensional periodic cell is shown. As a

particle leaves the right-hand boundary, a mirror image enters from the left boundary.
Similarly for the upper and lower boundary.
 The distinct element method 187

significant and the results are inaccurate or the model becomes unstable
(because particles overlap too much and they undergo increasingly large and
unrealistic accelerations). If the time step is too small the computer run time
will be unnecessarily long. Rounding errors can propagate over the time steps
causing errors for time steps which are too small. Another effect of a small
time step is spurious oscillations. Disturbances can be propagated faster than
the wave speed through the medium. In terms of the Rayleigh wave speed the
corresponding time step is given by (Johnson, 1985)
M = (nR/a)(Q/G)'/2 (5.14)
where R is the particle radius, a is a constant (0.9-0.95), Qis the particle
density and G is the shear modulus. Some researchers relate the time step to
the natural period of oscillation of the system. Thompson and Grest (1991)
state that the time step should be < 0.12Y(mlk) for particle mass m and
stiffness k. Zhang and Campbell (1992) propose that it should be
< 0.075Y(mlk). These estimates should give a good idea of the required time
step, but selection should be followed by trial runs with different I'3.t values to
check that the results are stable and reasonably invariant over this range of
I'3.t.

5.3.4 Forces
Chapter 3 summarises models for forces between particles. This section looks
at the application of these forces in DE models. Langston et al. (1995) have
developed DE granular models for non-cohesive particles using three
different functional forms for the normal interaction. These are illustrated in
Figure 5.9. The continuous interaction curve is widely used in molecular and
colloidal simulations and is fairly stable. The Hertz curve is based on the
contact mechanics of spheres (see section 3.2). The simple linear spring is the
simplest mathematical form and is fairly widely used. In deciding how to
model particle normal interaction, the analyst needs to consider the eventual
application of the simulation. For example, since the characteristics of flow in
silos are dominated by geometrical effects, the velocity and stress profiles
may be realistically modelled using softened interactions which allow some
particle 'overlap'. The precise details of the contact mechanics may not be
necessary, and since the time step would have to be smaller for a more
realistic particle stiffness, this enables more particles to be simulated. Bearing
in mind that most real granular systems are irregular and rough, it seems
unnecessary to model granular flow with ideal Hertzian contacts, although
the interaction must not be so soft as to allow excessive overlap, or the
analysis will be inaccurate. If the analyst is interested in modelling a few
particles in a shear cell to obtain 'continuum' constitutive properties, then it
may be necessary to model the contact physics more closely. Frictional forces
are usually modelled using the Mindlin equation (Mindlin and Deresiewiez,
188 Assembly mechanics

o Overlap, r

Figure 5.9 Modelling normal forces. Non-cohesive normal force options shown.
(a) Continuous interacti?n (molecular dynamics), Fn oc r-n; (b) Hertz (contact
mechanics), Fn oc (1 - r) 12; (c) Hooke (spring contact), Fn oc (1 - r).

1953), as shown in Figure 5.10. Some other analysts use a linear spring for
friction, usually with a cut-off at the maximum force given by the friction
coefficient. At the other end of the scale, Thornton and Yin (1991) model the
normal and tangential contact mechanics in considerable detail in simulations
of elastic sphere impact.
Contact damping is an important consideration in granular dynamics. This
is most generally modelled by treating the contacts as second-order systems as
shown in Figure 5.11. The damping force is most generally characterised by
relating it to the ratio of critical damping. This ratio is typically in the range
0.1 to 0.5. Some analysts work in terms of coefficient of restitution. Zhang
and Campbell (1992) relate the coefficient of restitution to critical damping
for a second-order system.
The simulations can be extended to model cohesive forces such as those
due to electrostatic charge, using the interaction force equations in Chapter 3.
Gravitational forces can be defined in DE models using constant vectors.
Fluid forces such as those encountered in pneumatic conveying, fluidised
beds or air-assisted hopper discharge present other problems which can be
handled in a number of ways. Tsuji et al. (1993) has modelled fluidised beds
using a DE model for the particle dynamics and a fixed (Euler) grid for the air
flow. Mass and momentum balances are carried out on the gas for each zone,
using a finite element technique. The drag force on each particle is calculated
using the Ergun equation (see section 2.3.2), the region around each particle
being modelled as a localised packed bed with a voidage dependent on the
positions of the neighbouring particles. The drag force effect on the gas
 The distinct element method 189

dp
o

Figure 5.10 Modelling frictional forces. Mindlin friction equation.

Ft = flFn [1 - (1 - b/bmax)'h].

Figure 5.11 Modelling contact damping forces. Includes normal and tangential.
Retarding force, ex. relative velocity; mx +ex
+ kx = F; damping ratio ex. critical
damping; e ex. v' (mlk).

momentum balance is also accounted for. An example of the results of Tsuji's

model is shown in Figure 5.12. As an example of how fluid drag may be
incorporated into a DE model, Figure 5.13 illustrates a simulation method for
air-assisted discharge of a hopper without explicit simulation of the air flow,
i.e. without solving the momentum balance. Using the Ergun equation and
the overpressure specified above the hopper contents, the superficial air
velocity at the hopper orifice is evaluated at each time step. The model
assumes radial uniform flow of the air throughout the hopper, such that at any
190 Assembly mechanics

Figure 5.12 Simulation of a f1uidised bed. Shows particle velocity vectors for
different jet velocities. Reproduced with permission from Y. Tsuji, T. Kawaguchi and
T. Tanaka. Discrete Particle Simulation of Two-Dimensional Fluidized Bed, published
by Elsevier, 1993.

particle within the hopper the superficial air velocity is known. Then from the
local voidage at each particle, the interstitial air velocity is calculated and
hence the relative air-particle velocity is known. This gives a local drag force.
This is added to the other forces acting on the particle and its dynamic
response is modelled in the same way as in Figure 5.13. If the assumption of
radial flow is valid the FE method is not required. This approach is probably
reasonable for hopper flow and some pneumatic conveying applications
where spatial and temporal gradients of voidage are relatively small, but it
will not be sufficient for more complex applications such as fluidised beds.
An advantage of DE methods is that they can handle any container
geometry, for example a screw feeder, drum mixer or conveyor. These do not
present any geometrical difficulty (although their size may be a problem in
terms of the number of particles in the simulation). Particle---container
interactions can be handled in a manner similar to particle-particle contacts.
Frequently, the analysis is simplified by the presence of a fixed container
boundary.

5.3.5 Particle size and shape

A major potential advantage of DE models is that they can handle the effects
of particle shape and size directly. However, modelling non-spherical
particles can present significant problems. This is in marked contrast to the
comments above about the container geometry. Most granular systems have
 The distinct element method 797

Figure 5.13 Modelling air-assisted hopper discharge. (u o = f(llP) from Ergun

equation; uo(z) = f(uo, z) continuity equation; Uj = Uo(Z)lEl radial flow assumed; FD
= f( Ej, Uj - vp ).) This uses the Ergun equation and continuity equation to predict the
air superficial velocity. The local velocity at a particle is determined from the local
void age and assuming radial flow of gas. There is no restriction on the movement of
the particles. The drag force on the particles is then determined from the local voidage
and relative velocity using the Ergun equation.

some polydispersity in particle size. Campbell and Potapov (1993) have

shown that even a small polydispersity can significantly affect the packing
structure. Langston et al. (1995) have also shown this for packing of two-
dimensional discs in a silo (Figure 5.14) where a uniform linear 10% variation
in disc diameter has also been imposed. Other distributions could easily be
used, such as the log-normal distribution which is probably the most generally
applicable. Some industrial processes may include the handling of multimodal
mixtures, a set of large particles and a set of fines, for example. This can lead
to segregation during handling and storage leading to undesirable effects on
product quality, or handling problems such as erratic flow. The DE method is
ideal for simulation of these effects, unlike the continuum models.
Most researchers use discs in two dimensions and spheres in three
dimensions, but some have made efforts to investigate other shapes. The
contact identification is more complex here, as is the force analysis. The
three-dimensional analysis is even more complex for non-spheres both in
terms of contact determination and in the equations of momentum which
determine the variation of orientation with time. (Three-dimensional angular
motion is complex - see Fincham, 1992). Hogue (1993) uses a general method
of specifying radii with angle for two-dimensional particles. The necessity for
such a general representation is debatable. Mustoe and DePoorter (1993) use
792 Assembly mechanics

(a)

(b)

Figure 5.14 Effect of polydispersity on the packing structure. In a monosized system

a crystalline structure is set up. With only 10% polydispersity, it is more amorphous.
(a) Monosized, (b) polydisperse.

an elliptical equation to model two-dimensional shapes which has computa-

tional efficiency advantages. Other shapes which have been considered
include 'end-rounded cylinders' and squares. The choice of shape depends on
the application and the requirements of the model. For most compact
particles a softened sphere analysis, such as the continuous interaction model,
 The, distinct element method 193

is probably reasonable to simulate flow, especially since with spheres more

particles can be modelled. For needle-shaped particles the elongated shape
needs to be modelled explicitly, as a cylinder, for example. In some studies it
may be desirable to investigate shape in detail on a small scale. The
consequences of the shape for the small scale simulation might then be scaled
up. For instance, if small cubes are being handled, the effects of tessellation
may be significant. Figure 5.15 shows an example of two-dimensional non-
disc simulation by Walton (1983).

Figure S.lS Two-dimensional non-disc simulation of shale flow. Shows position and
velocity vectors. Reproduced with permission from O. Walton. Particle-Dynamics of
Shear Flow, published by Elsevier, 1983.

Another feature to be considered in this section is that of attrition and

fracture. If information is available on attrition, this can be incorporated in
order to reduce the particle size. In such a model it may be reasonable to
maintain the same particle shape and ignore the fines produced from the
attrition. With fracture, however, the consequences for residual particles are
not so straightforward. Data may be available on particle breakage under
load, but the shape and size distribution of the products is difficult to predict
and to incorporate into the model. Ouwerkerk (1991) has used TRUBAL to
model the bulk strength of a two-dimensional disc assembly as a function of
single particle properties as shown in Figure 5.16. Ning (1995) has used TRUBAL
to model fragmentation of single agglomerates due to impact as shown in
Figure 5.17.

5.3.6 Stress analysis

Both internal and wall stresses may be calculated using the DE method. The
problem here is to calculate stresses resulting from 'point' contact forces. For
the walls this is reasonably straightforward: the particle-wall contact forces
are summed over small elements of the wall and divided by the area
considered, both for normal and tangential forces. Figure 5.18 shows an
example of such calculations for discharging a two-dimensional hopper.
194 Assemblv mechanics

Figure 5.16 Modelling of assembly bulk strength. Two consecutive two-dimensional

TRUBAL simulations. Thickness of lines proportional to contact forces and broken
particles numbered in order of breakage. Reproduced with permission from C.E.D.
Ouwerkerk. A Micro-mechanical Connection between the Single-Particle Strength and
the Bulk Strength of Random Packings of Spherical Particles, published by Elsevier,
1991.

Figure 5.17 Fragmentation of polydisperse agglomerates due to impact (supplied by

Z. Ning, 1995).

The internal distributions of the normal and tangential bulk stresses have
been calculated by adopting a local averaging technique that works on the
values of the normal and tangential contact force vectors at various 'probe'
points within the particle assembly. The information thus generated is
analogous to the experimental measurements obtained by placing force
 The distinct element method 795

Figure 5.18 Wall stress in two-dimensional discharging hopper. Line lengths

proportional to simulated wall stress, normal and tangential, Langston et al. (1995).

transducer pills at different positions within the bulk (see, for example,
Handley and Perry, 1967). Normal and shear stresses for each probe point in
two dimensions are calculated from the following equations:

(5.15)
V 2DO y = L FnYnuYn + LFtYtuYn (5.16)
V 2D Oyx = L FtXtuYn + LFnxnuYn (5.17)
V 2D Oxy = L FtYtuxn + LFnYnuxn (5.18)

°
where V 2D is the sample volume, is the stress, Fis the force vector, Xn is the
contact x component vector (from particle-to-particle centre) and subscript u
indicates unit vector, n normal vector and t tangential vector. The three-
dimensional equations for the horizontal stress, On and the vertical stress an
are similar. This is essentially a numerical integration of the force vectors
within a small radius (e.g. two particle diameters) of the probe point to obtain
an average stress. Figure 5.19 shows an example of the internal normal stress
196 Assembly mechanics

CLLLL

lLL.LL.

LLLlL
LLLu
LLLL

LL'--Ll

Figure 5.19 Internal normal stress in a two-dimensional discharging hopper. Shows

vertical and horizontal stress at 'probe' points in simulation. Line lengths proportional
to stress (reorientated slightly for clarity).

distribution in a two-dimensional discharging hopper which illustrates the

switch from 'active' to 'passive' stress towards the hopper orifice; see
Nedderman (1992) and refer to Chapter 8 for further details.

5.4 NOMENCLATURE
A Area
c Damping constant
C Compressibility coefficient
D Elasticity constant; = E* in equation (3.88a)
E* Effective contact modulus (see equation (3.88a))
F Contact force
FD Air drag on particle
g Gravitational acceleration
G Shear modulus
h Height of packing
k Particle stiffness
K Janssen's coefficient
m Particle mass
N Number of particles
n Index in normal interaction curve
 Appendix A5 197

jj.p Overpressure
R Particle radius
r Separation of particle centres
S Separation distance
Sn Normal displacement at contact
t Time
jj.t Time interval
Vo Superficial air velocity
Vj Interstitial air velocity
vp Particle velocity
V 2D Volume of 'sensor' search area in two-dimensional analysis
x Particle position

Greek letters
Pressure coefficient
Tangential displacement at contact, maximum before sliding
Local void age
Friction coefficient
Poisson's ratio
Q Density
o Normal stress

Subscripts
j For each particle
n Normal
r Radial
Tangential
u Unit vector
w wall
x,Y,Z Axes

APPENDIX A5 FRICTIONAL FORCES

A.5.1 Frictional forces
The frictional force, F, can be written in terms of the surface integral
F= j;-r(r)dA (AI)
where from equation (0.00) L(r) is given by
L(r) = Lo + aP(r) (A2)
198 Assemblv mechanics

A.5.2 Single contacts

For a Hertzian contact, the pressure distribution is given by
,2 )Y7
Per) = Po ( I - b
2 (A3)

where b is the contact radius and the maximum pressure Po at r = 0 is given by

Po = 3W/2nb 2 (A4)
Evaluating equation (AI) after substituting equations (A2) and (A3) gives
(A5)
Equation (3.85) is recovered by substituting equation (A3), P = W/nb 2 and
A = nb2 into equation (A5); see also section 3.3.

A.5.3 Particle assemblies

The total frictional force F = NbF can be written in the following exact form
N N
F =I 't;A; = I ('to + aPJA; (A6)
i= 1 i= 1

Introducing the mean contact area per particle

1 N
bA = -IA;
Ni=l
and the relationship
W; = P;A;
gives

N
F = 'toNbA + a I Wi (A7)
;= 1

but

N
I Wi = NbW
;=1

where bW is the mean load per particle, hence

F = 'toNbA + aNb W (A5.8)
Equation (5.6) can then be obtained from equation (A5.8).
 References 199

REFERENCES
Aizawa, T. (1993) Granular modeler of two and three dimensional powder dynamics
in powder forming and metallurgy, in The First Nisshin Engineering Particle
Technology International Seminar, January 18-20, Nisshin Engineering Co. Ltd.,
p.27.
Barker, G.C. and Mehta, A. (1992) Physical Review A 45(6), 3435.
Campbell, C.S. and Potapov, A.V. (1993) Recent applications of computer simulation
to granular systems, in The First Nisshin Engineering Particle Technology
International Seminar, January 18-20, Nisshin Engineering Co. Ltd., p. 27.
Cundall, P.A. and Strack, O.D.L. (1979) Geotechnique 29(1),47.
Duke, T.A.J., Barker, G.C. and Mehta, A. (1990) Europhys. Letters 13(1),19.
Fincham, D. (1992) Molecular Simulations 8, 165.
Haile, J.M. (1992) Molecular Dynamic Simulation, Wiley, New York.
Handley, M.F. and Perry, M.G. (1967) Powder Technol. 1,245.
Hogue, C. (1993) PhD Thesis, University of Cambridge.
Jenike, A.W. (1964) Storage and flow of solids, Bulletin 123, Utah Enging. Exper.
Station, University of Utah.
Johnson, K.L. (1985) Contact Mechanics, Cambridge University Press, Cambridge.
Kafui, K.D. and Thornton, C. (1993) in Powders and Grains '93 (C. Thornton, ed.),
Balkema, Rotterdam, p. 401.
Langston, P.A., Heyes, D.M. and Tiiziin, U. (1994) Chem. Engng. Sci. 49, 1259.
Langston, P.A., Heyes, D.M. and Tiiziin, U. (1995) Chem. Engng. Sci. 50,967.
Laohakul, C. (1978) Velocity distributions of flowing granular materials, PhD Thesis,
University of Cambridge.
Martinez, J., Masson, S. and Dcserablc, D. (1995) in Proceedings of the Third
European Symposium on Storage and Flow of Particulate Solids, Niirnberg, March
1995. Niirnbergmesse GmbH, p. 367.
Mindlin, R.D. and Deresiewicz, H. (1953) J. Appl. Mech. 20,327.
Mustoe, G.G.W. and DePoorter, G. (1993) in Powders and Grains '93 (C. Thornton,
ed.), Balkema, Rotterdam, p. 357.
Nedderman, R.M. (1992) Statics and Kinematics of Granular Materials, Cambridge
University Press, Cambridge.
Ning, z. (1995) PhD Thesis, University of Aston.
Ouwerkerk, C.E.D. (1991) Powder Technol. 65, 131.
Pariseau, W.G. (1969a) Gravity flows of ideally plastic materials through slots, Trans.
Am. Soc. Mech. Engings., J. Eng. Ind. May.
Pariseau, W.G. (1969b) Discontinuous velocity fields in gravity flows of granular
materials through slots. Powder Technol. 3, 218.
Roscoe, K.H., Schofield, A.N. and Wroth, c.P. (1958) On the yielding of soils.
Geotechnique 8, 22.
Satak, M. (1993) in Powders and Grains '93 (C. Thornton, ed.), Balkema,
Rotterdam, p. 3.
Schofield, A.N. and Wroth, c.P. (1967) Critical State Soil Mechanics, McGraw Hill,
London.
Smith, I.M. and Griffiths, D.V. (1982) Programming the Finite Element Method,
Wiley, Chichester.
Spencer, A.J.M. (1964) A theory of the kinematics of ideal soils under plain-strain
conditions. J. Mech. Phys. Solids 12, 337.
Taguchi, Y. (1993) Convection, turbulence and anomalous diffusion in vibrated bed,
in The First Nisshin Engineering Particle Technology International Seminar, January
18-20, Nisshin Engineering Co. Ltd., p. 84.
200 Assembly mechanics

Tanaka, T. and Tsuji, Y. (1993) Cluster formation in gas-solid flows, in The First
Nisshin Engineering Particle Technology International Seminar, January 18-20,
Nisshin Engineering Co. Ltd., p. 84.
Thompson, P.A. and Grest, G.S. (1991) Phys. Rev. Lett. 67(13), 1751.
Thornton, C. (ed.) (1993) Proceedings of the Second International Conference on
Micromechanics of Granular Media. Published as Powders and Grains '93,
Balkema, Rotterdam.
Thornton, C. and Yin, K.K. (1991) Impact of elastic spheres with and without
adhesion. Powder Technol. 65, 153.
Tsuji, Y., Kawaguchi, T. and Tanaka, T. (1993) Powder Technol. 77,79.
Tiiziin, U. and Nedderman, R.M. (1979) Experimental evidence supporting kinematic
model of the flow of granular media in the absence of air-drag. Powder Technol. 24,
257.
Walton, O. (1983) Particle-dynamics calculations of shear flow, in Micromechanics of
Granular Materials, Elsevier, Amsterdam.
Zhang, Y. and Campbell, C.S. (1992) J. Fluid Mech. 237,541.
6
Fluid-particle systems

The behaviour of mixtures of more than one phase is important in many areas
of process engineering, particularly where particulate material is transported
by a fluid, in reaction engineering or where energy or mass transfer occurs
across phase boundaries:
Compared with single phase systems, multi-phase systems present inherent
experimental and theoretical difficulties, for example in the need to
characterise particle shapes, describe size distributions and their evolution,
treat surface layer effects including contamination, consider various
regimes of motion, and deal with a host of secondary factors such as
Brownian motion and electrostatic charges. (Grace, 1986)
Moreover, structures can develop in the flow with their own characteristic
length scales, different from those of either the particles or the equipment,
and this often presents further complications (Clift, 1996). The range of fluid-
particle contacting devices in use in industry is very extensive. The subject
matter of this book is restricted largely to gas/solid systems, but even here the
variety of devices is extremely wide. One selection, again due to Grace
(1986), is presented in Table 6.1. No uniform view of the fluid and particle
processes in these devices yet exists. (Molerus, 1993, gives one of the best
accounts of those currently available.) In this chapter, therefore, a small
selection of fluid-particle contacting devices is presented, illustrating the
applications of the theoretical developments presented in Chapters 1 to 5.

6.1 HYDRAULIC AND PNEUMATIC CONVEYING

6.1.1 Differences and similarities
Hydraulic and pneumatic conveying are both common techniques for
transporting particulate solids in a conveying fluid. Here we deal solely with
conveying in pipes. Hydraulic transport can also be carried out in open
flumes, so that there is significant common ground with sediment transport in
rivers. Pneumatic transport obviously needs closed ducts, and is less
obviously related to transport of particles in the natural environment.
However, the basic ideas developed by Bagnold (1941) in his classic studies
on blown desert sand have been taken up in analysing hydraulic conveying
(see e.g. Wilson et al., 1996) and to some extent in pneumatic conveying.

J. Seville et al., Processing of Particulate Solids

© Chapman & Hall 1997
Table 6.1 Comparison of principal continuous gas/solid contactors: usual ranges (Grace, 1986)

Property Moving Fluidised Spouted Circulating Entrained Rotary

bed bed bed bed bed drum

Mean particle diameter 0.6-300 0.03-3 0.8--6 0.05--0.5 0.02--0.08 1--63

(mm)
Superficial gas velocity <0.3 0.5-3 0.6-2 3-12 15-30 2--6
(m s- 1)
Diameter or width (m) 1-5 0.3-25 0.15-3 0.5-9 0.1--4 2--6
Gas motion up up up up up up
Gas mixing near plug flow complex: two two regions dispersed plug flow near plug flow near plug flow
phases
Solids motion down up and down up and down mostly up, some up tumbling
down horizontal
Solids mixing near plug flow usually near near perfect near perfect near plug flow dispersed plug
perfect mixing mixing mixing flow
Overall voidage 0.4--0.5 0.5--0.8 0.45--0.55 0.85--0.99 0.98--0.998 0.7--0.96
Temperature gradients large very small may be small may be large
significant significant
Typical bed- or
suspension-to-surface
heat-transfer coefficient 50-150 200-550 150-250 150-250 50-100 1-2 (dominated
(W m-2 K- 1) by radiation)
Attrition little some considerable some considerable considerable
Agglomeration may be serious some little problem no problem no problem no problem
 Hydraulic and pneumatic conveying 203

At a mechanistic level, the distinction between hydraulic and pneumatic

conveying lies in the ratio of the density of the conveyed solids to that of the
conveying fluid. In a hydraulic system, the solids density is normally greater
than that of the fluid but of broadly the same order of magnitude. However,
in a pneumatic system the solids are typically at least three orders of
magnitude more dense than the fluid. The corresponding inertia of the
particles ensures that in a pneumatic system the particles impinge at high
velocity on bounding walls and on the surfaces of blowers and fittings.
However, in hydraulic systems, lift forces on the particles act to reduce wall
collisions (see Wilson et al., 1996). One immediate result of this difference is
that particle attrition in pneumatic conveying results primarily from impacts
of particles with the pipe wall, whereas in hydraulic conveying attrition occurs
mainly when the solids pass through a pump (see below). These basic
differences lead to practical differences between hydraulic and pneumatic
systems, of which the most significant are as follows.

System design. Both centrifugal and positive-displacement pumps are used

as 'prime movers' in hydraulic systems (see below). In both cases, it is
possible for the solids to pass through the pump with the liquid (or, more
generally, the 'conveying vehicle' - see below), subject to the obvious
restriction that the solids must be small enough to pass through the passages
in the pump and any other essential fittings, such as the non-return valves of a
positive-displacement pump. Therefore the basic layout of a hydraulic system
takes the form shown in Figure 6.1(a). The slurry is prepared in a mixing
vessel or sump, and then fed to the pump. Of course, for in-process
applications, the slurry already exists so that the preparation step is absent.
Because the slurry itself is pumped, further booster pumps can be placed in
series downstream (although note that using pumps in parallel is to be
avoided, for reasons of operating stability; see Wilson et al., 1996). Therefore
it is possible to convey solids hydraulically over very large distances.
In a pneumatic system, it is normally necessary to ensure that the solids do
not pass through the blower, for the obvious reasons that catastrophic
damage will result to the blower or the solids or both. Therefore it is
necessary to introduce the solids downstream of the blower, as shown
schematically in Figure 6.1(b). The solids may be fed from a hopper, although
some positive feeding device (such as a rotary feeder) is usually needed. It is
impossible to use booster blowers (unless the solids are first removed from
the gas, and reintroduced after the blower). Therefore the distance over
which the solids can be conveyed is limited by the pressure attainable at the
blower discharge. This means, in turn, that pneumatic conveying can only be
used over relatively short distances, typically for in-plant materials handling.

Pressure drop. In a pneumatic system, the solids must be accelerated from

rest at the injection point up to the steady conveying velocity. This requires a
204 Fluid-particle systems

Solids

Slurry
preparation

Booster
Feed pump
(a)
pump

Solids

Conveying gas
(air)

(b)

Figure 6.1 Elements of conveying systems: (a) hydraulic; (b) pneumatic.

substantial drop in the pressure of the conveying gas, but the energy
transferred to the particles is not recoverable. Therefore the 'entry loss' is
usually significant in a pneumatic system (see sections 2.4 and 6.1.3.1), and
may even be the dominant contribution to the overall system pressure drop.
Similarly, there is usually a significant pressure drop following a bend in a
pneumatic system, because the particles must be accelerated up to the
transport velocity in the new direction.
On the other hand, because the solids pass through the pumps in a
hydraulic system, while the relative inertia of the particles is in any case small,
any 'entry loss' can usually be neglected. Similarly, the pressure drop
associated with flow through bends and fittings can be estimated to acceptable
accuracy from standard results for single-phase fluids. Therefore, estimation
of pressure loss in hydraulic systems concentrates on the effect of particles on
the shear stresses between the pipe wall and the flowing mixture.

Flow regimes. In both hydraulic and pneumatic systems, two general types
of flow are possible (see section 2.4.1):
Lean phase: low particle concentration and relatively high velocity.
 Hvdraulic and pneumatic conveying 205

Dense phase: high particle concentration (perhaps 50% by volume) and

relatively low velocity.
In dense-phase flow, the pipe is essentially filled by a moving bed of solids. In
lean-phase flow, the particles are partly or wholly dispersed in the conveying
fluid, although there may still be a limited bed at the bottom of the pipe in
horizontal flow.
Dense liquid/solid mixtures are difficult to pump, typically being subject to
blockages in the pump entry. Therefore dense-phase hydraulic conveying is
only used in very specific applications, primarily in the mining industry for
systems in which flow is maintained by gravity or by some indirect application
of the necessary pressure difference (for example, using lock-hoppers; see
Brown and Heywood, 1991). Therefore, the discussion of hydraulic convey-
ing here covers only lean-phase flow.
By contrast, dense-phase pneumatic conveying is very commonly used,
perhaps to the point of being the normal approach. In part, this is associated
with the fact that pneumatic conveying is primarily a means of material
handling rather than transport, using short pipe runs so that frictional
pressure drop is not a major concern. The lower air flows required are an
advantage of dense-phase conveying. A further important incentive for using
dense-phase flow arises from particle damage: it eliminates the high-velocity
wall impacts which are usually the main cause of attrition in lean-phase
pneumi:ltic conveying. Therefore, both lean- and dense-phase pneumatic
conveying will be discussed here.
A further kind of flow must be considered in hydraulic systems, however. If
the settling velocity is low enough and the interparticle forces are strong
enough - typically for fine particles at high concentration - then the settling
tendency of the particles can be ignored. This type of liquid/solid mixture is
termed 'homogeneous' if the solids remain fully dispersed when flow is
stopped, or 'pseudohomogeneous' if the particles settle but so slowly that the
particle concentration is uniform across the pipe even at relatively low
conveying velocities. In either case, the mixture can be treated as non-settling,
i.e. as if it were a single-phase fluid, albeit with non-Newtonian rheological
properties. Larger particles show settling behaviour, so that the solid and
liquid must be treated as separate phases. In pneumatic conveying, a non-
settling mixture would be an aerosol. Therefore there is no practical
equivalent of a non-settling mixture: the two phases must be considered
separately, as in hydraulic conveying of a settling mixture.
Because of these fundamental differences, hydraulic and pneumatic
conveying have developed separately. The distinction will be retained here.

6.1.2 Hydraulic transport

After many years in which no current book on hydraulic conveying was
available, four were published within a year. Shook and Roco (1991)
206 Fluid-particle systems

provided a very comprehensive review, primarily of the research literature.

Brown and Heywood (1991) edited a multi-author book which covers most
aspects of system design and component selection. The book by Wilson et al.
(1992), updated in 1996, is specifically concerned with systems using
centrifugal pumps, and is therefore directed mainly at the dredging, mining
and mineral processing industries with emphasis on the application of recent
fundamental developments. Jacobs (1991) takes a less fundamental approach
than the other three, and is generally more concerned with equipment
components.
In hydraulic conveying, the solid material is transported as a slurry in a
liquid 'vehicle'. Most commonly, the vehicle is water, but in processing
operations it may be a different liquid and in some applications the vehicle
may itself be a slurry of fine particles. An example of the latter is the
transport of drilling debris by a 'drilling mud'. There are two, generally
distinct, classes of application: short-distance conveying and long-distance
transport.

Short-distance conveying. This takes place over typically hundreds of

metres in processing operations or up to several kilometres in mining
operations, in dredging, or in tailings disposal. Usually the conveying system
must handle a material whose size distribution and other characteristics are
set by requirements imposed by other parts of the process or system.
Furthermore, these properties and also the throughput may vary widely. In
this kind of application, centrifugal pumps are most commonly used, because
they can provide the relatively modest pressures required and are also
relatively tolerant to changes in slurry properties.

Long-distance transport. This takes place over tens or even hundreds of

kilometres. Hydraulic transport may be economic where there is no
developed transportation system, typically over rough terrain where a
dedicated pipeline system can be installed, and buried to protect it from
damage. In this kind of application, where the energy cost of transport is a
major consideration, it may be desirable to 'design' the slurry by milling the
particles to a controlled size distribution with good flow properties. The
pressures required for long-distance transport are high, and therefore positive
displacement pumps are usually preferred. Positive displacement pumps
generally have less operating flexibility than centrifugal pumps, primarily
because of the limited clearances between moving parts in non-return valves
and around pistons or plungers. Thus positive displacement pumps require
control of the particle size distribution, a further reason why they are
encountered primarily in long-distance applications.

The distinction between settling and non-settling slurries has already been
introduced. In a non-settling slurry, the particles are sufficiently fine and
 Hydraulic and pneumatic conveying 207

concentrated that their tendency to settle out from the conveying liquid can
be ignored in designing the conveying system; an extreme example is a paste.
The slurry can then be treated as if it were a single-phase fluid. The pressure
drop for a non-settling slurry in a horizontal pipe then varies with mean
velocity as shown qualitatively by curve NS in Figure 6.2, with the gradient
discontinuity around point T corresponding to the transition from laminar
flow at lower velocities to turbulent flow at higher velocities.
Settling slurries, by contrast, contain particles sufficiently large or dense
that their tendency to settle out from the vehicle cannot be ignored. As a rule
of thumb, this case arises when the hindered settling velocity is greater than
about 1.5 mm S-l. For particles with the density of sand (2650 kg m-3 ) in
water, this means that the slurry will normally show settling behaviour if the
particles are larger than about 75 !-tm. The system characteristic for a settling
slurry flowing through a horizontal pipe at constant delivered concentration
(see below) typically takes the form of curve S in Figure 6.2. Point D
corresponds to the deposit velocity, V" below which the pipe contains a
stationary deposit of solids which have 'settled out' from the slurry. At
velocities above V., the solids are all in motion, but the pressure gradient
typically decreases with increasing mixture velocity, to pass through a
minimum and then to approach the pressure gradient for the vehicle alone at
high velocities.
Systematic design of a hydraulic conveying installation depends on
calculating system curves like those shown schematically in Figure 6.2. This

PRESSURE GRADIENT

NS

CARRIER LlaUID

Vs MEAN VELOCITY, Vm

Figure 6.2 Velocity/pressure drop characteristics for flow of slurries in pipes at fixed
delivered concentration (schematic): NS = non-settling slurry; S = settling slurry; T =
laminar/turbulent transition; Vs = deposit velocity_
208 F/uid~particfe systems

approach is developed in detail, with case studies, by Wilson et al. (1992 and
1996). An outline of their approach is given here.

6.7.2. 7 Parameters describing slurry flow. Although many industries

traditionally express slurry concentration in terms of the weight concentration
of solids, Cw , a better indication is the volume concentration Cy • Relation-
ships between Cy , Cw , and the slurry specific density
(6.1)
where Qrn is the mean slurry density and Qf the liquid density, are
summarised in Table 6.2. Values for volume concentration in the range
10-20% are fairly typical for lean-phase conveying. The operating concentra-
tion is usually limited by the handling characteristics of the pumps, and a
value of C v of 30% is high for lean conveying. In the rarer use of dense-phase
conveying, values of Cy as high as 50% may be used, but special pumping
arrangements are then needed as noted above.
For a settling slurry, the mean particle velocity in the pipe, V p , is lower than
the mean liquid velocity or the mean slurry velocity, V m (unless the pipe is
directed downwards ~ see below). A simple continuity balance (see section
2.4) shows that the mean in situ solids concentration in the slurry issuing from
the pipe, C yt , is then higher than the delivered concentration in the slurry
issuing from the pipe, Cyd . The two are related by
Vp
(6.2)

Table 6.2 Measures of solids concentration in a slurry

Expressions for: In terms of:

Sp(Sm - 1)

Sm(Sp - 1)

where Cv = volume concentration of solids in slurry, Cw = weight concentration of solids in

slurry, Sm = relative density of slurry (Qm/Qf), Sp = relative density of particle (Q/Qf)'
 Hydraulic and pneumatic conveying 209

One of the implications of this simple result is that in situ measurements of

concentration, for example by absorption of some form of radiation, do not
normally give the delivered concentration. If this approach is used, as it is
increasingly in the mining and dredging industries, then the instrument must
be located in a section of pipe where the 'slip velocity', (Vrn - V p ), is as small
as possible. This usually means placing it in a vertical pipe (see below).
For stability of operation, pipes inclined at more than about 10° from the
horizontal or vertical are normally to be avoided in hydraulic conveying
systems handling settling slurries. Inclined pipes are subject to intermittent
slumping and re-entrainment of the 'contact load' (see below), leading to
erratic and sometimes damaging surging and vibration. In an extreme case,
slumping of the solids on shut-down may cause the pipe to become plugged.
Because hydraulic conveying systems normally only contain horizontal and
vertical runs, this summary concentrates on these two orientations. Effects of
inclination on flow are reviewed by Wilson et al. (1996).
In steady fully-developed horizontal flow, the pressure gradient results
from shear stresses between the pipe wall and the vehicle and particles.
Largely because much of the past work on hydraulic conveying has been
carried out by civil and mining engineers, who cling more tenaciously than
others to older concepts in fluid mechanics, the pressure drop is conventionally
expressed as the friction gradient, i.e. the loss of 'head' per unit length of pipe
run. For a single-phase liquid or vehicle, the friction gradient is (in metres
head per metre of pipe)

( IlP) 1 (6.3)
if = - ilL f gQf

where (- IlP/ IlL)f is the pressure gradient along the pipe. For a slurry, two
measures of friction gradient are then possible depending on the density used
to convert pressure to head. In terms of the vehicle density

(6.4)

or, in terms of the slurry density,

(6.5)

so that
(6.6)
Both im and jrn will be used here, to illustrate different points.
Even for relatively short-distance conveying, the dominant operating cost
can be the power required to drive the pumps. A simple measure of the
efficiency of a conveying system is then the specific energy consumption, i.e.
210 Fluid-particle systems

the energy consumed per unit mass of solids transported per unit distance
travelled. In terms of the friction gradient, i m , the specific energy consumption
IS

SEC = imlSpCvd (6.7)

where Sp = (Qp/Qf) is the relative density of the particles.

6.1.2.2 Slurry flow in a horizontal pipe. In terms of slurry head, jm, the
friction gradient for a non-settling slurry varies with concentration and
velocity as shown schematically in Figure 6.3. To a good approximation, the
friction gradient in turbulent flow is independent of slurry concentration (and
close to the value for the vehicle alone). In the laminar flow regime, to the left
of the turbulent flow curve, the slurry friction loss depends on solids
concentration. Non-settling slurries almost always show non-Newtonian
properties, most commonly with a significant yield stress and with shear-
thinning properties (i.e. mean viscosity decreasing with increasing shear
rate). However, some slurries can show dilatant behaviour (mean viscosity
increasing with shear rate), and this behaviour is often desirable for drilling
muds. The rheological properties of a slurry depend on the solids concentration
and size distribution, and on the interaction forces between the particles
which depend inter alia on the pH and ionic strength of the carrier liquid and
on trace concentrations of surface active agents. Because of the complexity,

jrn

TURBULENT FLOW

VELOCITY, Vm

Figure 6.3 Friction gradient for a non-settling slurry: C1 < C2 < C3 < C4 .
 Hydraulic and pneumatic conveying 211

and the number of variables involved, the rheological properties cannot be

predicted with confidence. Therefore, to calculate jrn (Vrn) in the laminar flow
range, it is necessary either to measure the rheological properties of the slurry
or to scale up from measurements in a pipe of conveniently small size. These
matters are covered elsewhere (Wilson et at., 1996). Typically the yield stress
increases rapidly with solids concentration when Cv exceeds some critical
value (compare concentrations C2 , C3 and C4 in Figure 6.3). Therefore, if the
slurry concentration is a design parameter, it is necessary to select the most
economic concentration, for example to minimise the specific energy
consumption. The appropriate value is frequently just below the concentra-
tion at which the yield stress starts to increase sharply. To ensure stability
against changes in slurry concentration, it is desirable to operate slightly
above the laminar/turbulent transition (see Wilson et at., 1996).
The remainder of this review concentrates on settling slurries. For a settling
slurry in turbulent flow, the values of irn and jrn typically vary with V rn as
shown in Figure 6.4. It is commonly the case that the jrn curves cross the curve
for vehicle alone at a velocity which is only weakly dependent on solids
concentration (although it depends on solids size and pipe diameter). At this
point, EF, the friction gradient for the slurry is equal to that for a hypothetical
'equivalent fluid' with mean density equal to that of the delivered slurry
(similar to turbulent flow of a non-settling slurry in Figure 6.3). At very high
mixture velocities, irn typically approaches asymptotically to the curve for the
vehicle alone, if; i.e. in the high velocity limit, the solids do not contribute to
the friction gradient. At low velocities, the deposit velocity, V s , defines the
velocity above which all the solids within the pipe are in motion, i.e. below
V" the pipe is partly blocked by a stationary deposit. Except in very rare
cases where reducing pipe wear outweighs energy efficiency (see Wilson et al.,
1996), a system must therefore be designed to have V rn > Vs' The deposit
velocity depends on solids concentration, and normal practice is therefore to
design to keep V rn above the maximum value of Vs over all possible
concentrations, V srn . Figure 6.5 shows a nomograph developed by Wilson
and others for quick estimation of V srn . Values obtained from this nomograph
are conservative, and Wilson et at. (1996) give less conservative relationships
incorporating the effect of solids concentration. Figure 6.5 shows the
existence of a 'worst possible' particle size, dependent on pipe diameter but
around 400-500 !-lm, which usually limits V srn . This arises because smaller
particles can more easily be lifted into suspension by turbulent eddies in the
conveying fluid, whereas larger particles are more easily moved along the
pipe by rolling, sliding or saltation.
For a settling slurry with V rn > V s , it is useful to distinguish between two
distinct models of solids transport: suspended load and contact or bed load.

(a) Suspended load. These are particles which are carried by turbulent
suspension in the conveying liquid. The immersed weight (i.e. weight minus
212 Fluid-particle svstems

im

VELOCITY. Vm

jm

VELOCITY. Vm

Figure 6.4 Friction gradient expressed as head of vehicle Om) and head of slurry Urn)
for a settling slurry: EF = equivalent-fluid point; C I < C2 < C3 .

Archimedean buoyancy; see section 2.1) of these particles is supported by the

vertical pressure gradient in the liquid at any section in the pipe. Particles
suspended in this way can only affect the pressure gradient if they transfer
momentum to the pipe wall directly by collision, or if they affect the shear
stress between the wall and the conveying vehicle (see section 2.4). The
former effect is negligible, except in pipes of very small diameter, because the
fluid flow near the wall imparts a 'lift' force which drives particles away from
the wall (see Wilson et al., 1996) so that wall collisions are very infrequent. As
a result, modification of fluid/wall shear stress is only significant if the slurry
contains a significant proportion of particles small enough to be contained
within the laminar sublayer at the wall of the pipe. For the pipe sizes and
 Hydraulic and pneumatic conveying 213

Pipe diameter 0 (mm)

o 0 0 009 99 0 0 0
i\) :......
~ ~ ~
~ ~:...:......:......
o
.......
00 0) ~Wl\) ..... o

o
o 009~
:...... :...... N I\)
01 00001
Velocity at limit of stationary deposition
:-" :--'"
Vsm (m/s) S 2.65
:--"':-" -t.....L -t. f\)
= I\) UJ UJ ~ c.n -..JO
.....
..... I\) w~o,Cn 000 0, 00,00 00

Figure 6.5 Nomograph for estimation of maximum velocity at limit of stationary

deposition, V sm , from Wilson (1979).

velocities used in hydraulic conveying, the laminar sublayer thickness is

typically of order 50 'Am, i.e. only particles which are effectively non-settling
(see above) can affect wall shear and pressure gradient in this way. There is
indeed ample experimental evidence (see Wilson et al., 1996) that settling
solids in turbulent suspension do not contribute to pipe friction.

(b) Contact load or bed load. These are particles whose immersed weight is
supported by stresses transmitteci directly between particles and wall. This
214 Fluid-particle systems

includes particles which move through the pipe as a sliding bed, but also
extends to other types of particle motion, notably 'saltation', in which
particles are supported intermittently by collisions either with the wall of the
pipe or with other particles which in turn collide with other particles or with
the wall. The essential feature of the contact load is that, because it is
supported by solid/wall contacts which can be represented by a local mean
solid/wall normal stress, it gives rise to direct momentum transfer between
the particles and the wall which can be embodied in a mean solid/wall
tangential stress. The ratio of the mean tangential stress to the mean normal
stress can be regarded as a coefficient of dynamic friction between the solids
and the wall. The development of this model by Wilson and his coworkers has
simplified the contact load as a sliding bed but the general approach is not
limited to this type of motion: it is an approximation to any hydraulic
transport in which the immersed solids weight is supported by direct contact
between particles and wall, whether continuous or intermittent. Representa-
tion on a more detailed or sophisticated level - for example by single-element
modelling or by treating collisions as a kinematic process - is a very long way
off because, by definition, hydraulic transport is a case for which the
interstitial fluid cannot be neglected and the unsteady fluid drag terms (see
section 2.2.1) are usually significant.
The distinction between the modes of transport is illustrated in Figure 6.6,
which shows schematic sections through a horizontal pipe carrying a settling
slurry at different velocities corresponding to different ranges in Figure 6.4.
At or below the deposit point (Figure 6.6a), the pipe contains a stationary bed
with particles transported primarily by saltation along the top of the bed. If
the slurry contains fine particles, then these may be conveyed in turbulent
suspension even in the presence of a stationary bed. At some velocity V m >
Vs (Figure 6. 6b), the slurry is typically in heterogeneous flow, with part of the
solids transported as contact load and part as suspended load. The pressure
gradient is greater than that for the vehicle alone, due primarily to mechanical
friction with the contact load. As the velocity is increased, the proportion of
solids transported as contact load decreases so that, for very high velocities
(Figure 6.6c), virtually all the solids are in turbulent suspension. Because the
suspended solids do not contribute to the pressure gradient (see above),
im ~ iw at high V m, and hence, from equation (6.6), jm ~ iw/Sm < iw.
This representation of the flow of a settling slurry illustrates the importance
of the stratification ratio, i.e. the proportion of the conveyed solids which are
transported as contact load. For a slurry in which the suspended solids do not
contribute to wall shear, the friction gradient can be written

im = if + (Sm - 1) fn ( V m) (6.8)
Vso
where the (Sm - 1) term is proportional to delivered solids concentration
(see Table 6.2) and therefore accounts for variations in solids throughput.
 Hydraulic and pneumatic conveying 215

~-----~---------

Cv
------
(a)

------r--c-------
------~
(b)

------1---\-------
------~
(c)

Figure 6.6 Particle concentration profiles within a pipe for settling slurry (schematic):
(a) below deposit velocity (Vrn < Vs); (b) in heterogeneous flow (Vrn > V s); (c) high-
velocity pseudo-homogeneous flow.

The function of (Vm/VSO) is proportional to the stratification ratio, where Vso

is the mixture velocity at which the stratification ratio is 0.5 (i.e. 50% of the
solids are conveyed as stratified load). Ideally, Vso is to be determined by
direct measurement under controlled conditions, and used as a basis for
scaling the measurements to other concentrations and pipe sizes. However, if
this kind of information is not available, d so can be estimated for particles of
diameter d in a pipe of diameter D as (Wilson et at., 1996)

Vso = w ~
':yI[;
cosh[60d/D] (6.9)

where ff is the (Moody) friction factor for the flow of the vehicle alone in the
pipe, and w depends on the terminal velocity of the particle, V t ,
w = 0.9Vt + 2.7[(Qp - Qf)g~/Q~V3 (6.10)

and ~ is the shear viscosity of the vehicle. Equation (6.9) embodies the idea
that, for a particle to pass into turbulent suspension, it must be enveloped by
an eddy whose characteristic velocity exceeds the terminal velocity; the
wV(8/ff ) term provides a rough estimate for velocity in an eddy (see Wilson et
at., 1996). It also contains the idea that the eddy must be large enough to
envelop the particle, and unless diD is very small only a proportion of the
eddies will be sufficiently large; the term cosh [60d/D] is a semi-empirical
expression to account for this effect.
216 FIUld-particle systems

The function of (Vm/VSO) in equation (6.8) must decrease with increasing

V m because the contact load reduces as the mixture velocity is increased. A
form which has proved reliable for heterogeneous flow is

fn(Vm/VSO) = 0.22 (Vm/VSotM (6.11)

so that equation (6.8) becomes

im - if) = 0.22 ( ~)-M

( Sm (6.12)
-1 Vso

The index M depends on the width of the particle size distribution. For a
narrow size distribution with a geometric standard deviation less than about
1.22, M can be taken as 1.7. For broader size distributions, M is smaller, and
can be estimated from results given in Wilson et al. (1996). In this case, din
equation (6.9) and V t in equation (6.10) should be taken as the values for the
particle size such that 84% of the particles are smaller and 16% are larger (i.e.
the particle diameter which is one geometric standard deviation times the
geometric mean size; see Chapter 1) because stratification is dominated by
the coarsest particles present.

6.1.2.3 Slurry flow in a vertical pipe. In vertical slurry flow of both settling
and non-settling slurries, there is nothing corresponding to the limiting
velocity, Vs , in horizontal flow of a settling slurry - in principle, any
superficial fluid velocity above minimum fluidisation (see section 6.2) can be
used in vertical upwards flow, while conditions in downward flow are limited
by bridging or blocking. The solids distribute themselves fairly uniformly
across the pipe section, so that equations (2.87) to (2.92) apply. The slip
velocity (Vm - Vs ), can then be approximated by the hindered settling
velocity, V;, at the in situ concentration. From equations (2.92) and (6.2), the
in situ concentration is then related to the delivered concentration for a slurry
in up flow by

Cvd = ( + V;)
1 Vm Cvd (6.13)

provided that V; < < V m' The in situ concentration in vertical downflow is

Cvt = -Vm)
( Vm Cvd = (1 - --V:) Cvd (6.14)
+ V; Vm
--

For a settling slurry, V: is normally very much less than the slip velocity in
horizontal flow, so that in either case the in situ and delivered concentrations
differ less in vertical than in horizontal flow.
Because the solids are uniformly distributed across the pipe, they are
 Hydraulic and pneumatic conveying 277

analogous to the suspended load in horizontal flow. In particular, contact

between particles and the wall is negligible and there is no direct momentum
transfer between the wall and the conveyed solids. The pressure gradient in a
riser can therefore be written in the form of equation (2.103):

(6.15)

where Smt is the in situ slurry specific gravity and 1'0 is the shear stress
between the wall and the conveying fluid (or the vehicle in the case of a slurry
with a significant concentration of particles in the non-settling size range).
The first term on the right of equation (6.15) is the 'hydrostatic' term, and the
equation can be written in terms of the in situ concentration (see Table 6.2) as

(6.16)

(6.17)

from equation (6.14). The corresponding result for a slurry in downflow is

---;;:z:
( -IJ.P ] D = -g
[
Qf + (Qp - Qf)Cvd
( V ; )]
1 - V rn + D
41'0

(6.18)
because the 'hydrostatic term' now acts in the direction of the flow.
Equations (6.17) and (6.18) can be used to estimate pressure drop in
vertical conveying. They also provide a relatively simple means of measuring
the density (and hence concentration) of a slurry directly on-line. Subtracting
equation (6.18) from equation (6.17) gives

-IJ.P)
( - - - ( - IJ.P ) = 2g[Qf + (Qp - Qf)Cvd] = 2gQrnd
IJ.L u IJ.L D

(6.19)
Thus averaging the pressure gradient up a riser and down a downcomer gives
the mean delivered slurry density and hence the delivered solids concentration.
The result is used in practice by passing the slurry through a vertical U-Ioop,
with the pressure gradient measured in the two legs (see Wilson et al., 1996).
A more rigorous analysis, avoiding the approximation in equations (6.17) and
(6.18) and allowing for the dependence of hindered settling velocity on in situ
concentration, has been given by Clift and Clift (1981); it delineates the
conditions under which this measurement technique becomes inaccurate.
218 Fluid-particfe systems

6.1.2.4 Design and operation of hydraulic conveying systems. Figure 6.7

shows the minimum essential components of a slurry transport system. Sump
design can be critical, and the requirements are reviewed by Wilson et ai.
(1996). The feed pump should be located as close as possible to the sump
from which the slurry is pumped, to avoid cavitation. If more than one pump
is used, the 'booster' pumps may be located near the feed pump or may be
spaced along the line, depending on specific requirements and on whether the
total pressure generated exceeds the bursting strength of the pipe. A simple
arrangement is to allow the solids rate into the sump to vary - in mining
operations, for example, as batches of material are delivered - and to control
the make-up of the vehicle (or 'hydraulic water') to maintain the level in the
sump. In some industries, feed control is crude, with the level control effected
by occasional operator intervention or even by allowing the sump to
overflow. Particularly in long-distance transport operations, where the
particle size and concentration must be controlled, control schemes much
more sophisticated than Figure 6.7 are used.
The kind of cheap, rudimentary system illustrated by Figure 6.7 carries
with it disadvantages in increased pumping costs. The problems which arise
are illustrated in Figure 6.8. To reduce specific energy consumption, from
equation (6.7) it is desirable to operate at low velocity (to reduce im) and high
concentration, Cvd ' However, it is now necessary to consider the stability of
the system by comparing the system characteristics (i.e. head loss as a function
of velocity) with the pump characteristics (i.e. head generated as a function of
flow rate).
For a well-designed centrifugal slurry pump, the solids have little effect on
the head generated when measured in terms of the delivered slurry density
(although they have much more effect in reducing pump efficiency; see
Brown and Heywood, 1991; Wilson et ai., 1996). Thus the pressure
generated, or the head expressed in terms of the carrying liquid density,
varies almost in proportion to slurry density. Given that slurry pumps must be

SLURRY

Pump

Figure 6.7 Simple slurry preparation system.

 Hydraulic and pneumatic conveying 219

Head of Vehicle

---- - - - - -----~
'"
., ---
'"
....
-........ " At

- -- --
Discharge

Head of Slurry

• f
I

Discharge

Figure 6.8 System and pump characteristics for settling slurry conveyed with
centrifugal pumps: f = vehicle alone; 1 = concentration Cj ; 2 = concentration C2 (>
C j ) ; A = pump A; B = pump B (larger or faster than A).

relatively low-speed devices (see below), the pump characteristics for fixed
speed operation have the form shown schematically in Figure 6.8.
Consider first operation with Pump A, selected to ensure that the operating
point (defined by the intersection of the pump and system characteristics) is at
modest discharge rate or slurry velocity. If the slurry concentration is
controlled at levell, then the system will be operable. However, if the
concentration goes up to level 2, then the pressure loss through the system
goes up to characteristic 2. The head generated by the pump (characteristic
220 Fluid-particle systems

A 2 ) is now insufficient to maintain flow through the system, and the discharge
will reduce back into the deposit range. Thus, if the concentration is variable,
it will be necessary to use a pump like B, selected to give operation around
the 'equivalent fluid' point (jrn = if; see above). Variations in concentration
and throughput can now be accommodated with little variation in the steady-
state velocity or total discharge rate. A more detailed analysis, along these
lines but covering transients and changes in particle size, is given by Wilson et
al. (1996).
Figure 6.8 illustrates why the operating condition for a settling slurry with
centrifugal pumps is usually limited by operating stability rather than by the
deposit velocity. Positive displacement pumps, by contrast, deliver a fixed
discharge rate, so that their characteristics can be represented by vertical lines
on the coordinates of Figure 6.8. It is then possible to select the most
economic velocity - as is required for long-distance transport - and the
deposit velocity may be limiting.

6.1.2.5 Other aspects. As in any system handling particulate materials,

wear can present problems. In hydraulic conveying, the pumps are particularly
subject to wear so that special design measures are adopted. Elastomer-lined
pumps are rarely used except for chemical processing applications where the
vehicle itself is particularly aggressive. Otherwise, it is more common to use
hard-metal pumps. However, centrifugal pumps are normally restricted to a
tip speed of about 25 m S-1 to limit abrasive wear. Wear on the pipe itself is
limiting in long-distance transport (see above). Pipe wear can be significant
when a settling slurry is conveyed in heterogeneous flow at high velocities -
around 5 m S-l, for example - but this is necessary for stable operation in
some mining applications, especially with coarse or uncontrolled solids. It
may then be necessary to rotate the pipe at planned intervals, to redistribute
the wear, and to replace the pipe regularly when it becomes too worn.
Because wall collisions are infrequent in hydraulic conveying, attrition of the
particles is concentrated in the pumps.
Because attrition and wear are more localised than in pneumatic convey-
ing, there is less incentive to use dense phase conveying. Dense phase is used
rarely, for example in vertical hoisting; see Brown and Heywood (1991). In
horizontal flow, dense conveying is essentially fully-stratified flow with the
contact load almost filling the pipe. It can be treated for analysis and design
by an extension of the approach to stratified flow outlined above (see Wilson
et al., 1996).
Measurements of slurry density or solids concentration by piezometric or
radiometric methods have already been mentioned. Flow-rate measurement
is much more complex than for single-phase fluids, because conventional
devices such as orifice plates, venturi meters and turbine meters either do not
work or cannot survive the slurry flow. A crude but effective means of flow
measurement is the bend meter, in which flow rate is related to the pressure
 Hydraulic and pneumatic conveying 221

difference between the inside and outside of a 900 bend (see Wilson et al.,
1996). Magnetic and Doppler meters are widely used for more precise
measurement of flow rate.

6.1.3 Pneumatic transport

Perhaps because scientific understanding of pneumatic transport is less
coherent than of hydraulic transport, fewer recent texts are available,
although the interested reader will find Marcus et al. (1990) and Woodcock
and Mason (1988) useful. Some relevant material will be found in books on
fluidisation (Knowlton, 1986, for example) bearing in mind that 'fasf
fluidisation' (see section 6.2) is essentially vertical pneumatic conveying with
a fluidised bed as a feeder and with fines return. As for hydraulic conveying,
pipes inclined at more than about 10° to the horizontal or vertical are to be
avoided. Therefore we concentrate here on horizontal and vertical flow.

6.1.3.1 Flow regimes and flow transitions. It has already been noted that
both lean- and dense-phase flow are used in pneumatic conveying, a(
distinction from hydraulic conveying where genuinely dense conveying is
rare. A further distinction lies in the difference between the two: the
transition from lean to dense flow with decreasing conveying velocity is sharp
in a pneumatic system, whereas there is rarely a sharp transition in a slurry. In
general terms, this difference mirrors the much-studied case of the stability of
fluidised systems (see for example Jackson, 1985) and arises from the
different solid/fluid density ratios in the two cases. A gas-fluidised bed shows
sharp transitions; for example, an expanded bed is usually unstable (but see
section 6.2.3.1), and it collapses into a bubbling bed with particle-free voids
(i.e. 'bubbles') distinct from the dense fluidised 'phase'. By contrast, a liquid-
fluidised bed can expand, but the two phases separate only gradually so that
'bubbles' only develop towards the top of deep beds.
In horizontal flow, the transition from lean- to dense-phase flow is usually
termed the saltation velocity (although the term is somewhat misleading,
because part of the transport in the lean phase may be by the process of
saltation, as for settling slurries). The transition is shown schematically in
Figure 6.9. At constant solids flow rate, M s , and starting with a high
superficial gas velocity, Uf, the mixture is in lean-phase flow. If the gas flow is
reduced, the pressure gradient at first decreases, much as for lean-phase flow
of a settling slurry. When the saltation velocity (Usa1t ) is reached, the pressure
gradient increases sharply, as an external indication of transition to dense-
phase flow. Further reduction in gas flow causes the pressure gradient to
increase. The reverse transition occurs on increasing flow, without appreciable
hysteresis if the flow is allowed to reach steady state. The saltation velocity
increases when the solids flow rate is increased. Therefore, the transition
222 Fluid-particle systems

...
.J::

'"c:
..9:!
H
.'1::
c: F
::I
:;;Q. ~ G MS2
Q. AL
0 E
~
MSl
~
::I
II>
II>
Q)

D: Ms=O
o

A USAlT for curve CDEF

Superficial gas velocity, U

Figure 6.9 Flow transitions in horizontal pneumatic conveying (schematic).

from lean to dense flow may be triggered by an increase in solids flow rate, if
the line is operating close to the saltation point.
Figure 6.10 shows schematically the corresponding flow transition in flow
vertically upwards. In this case the transition is somewhat less sharp. As the
gas velocity is reduced with the solids flow rate kept constant, the in situ
particle concentration goes up (see equations (2.77) and (6.13» so that the
'hydrostatic' contribution to the pressure gradient goes up (see equations
(2.103) and (6.15». This contribution eventually outweighs the decrease in the
frictional pressure gradient, so that the total gradient passes through a
minimum and then increases. Eventually it rises more sharply, to a condition
in which the riser pipe contains a slugging (or, more rarely, bubbling)
fluidised bed. This condition is called choking. A 'choked' vertical conveying
line is typically characterised by pressure fluctuations associated with the rise
and eruption of slugs. As for horizontal conveying, the transition velocity
depends on the solids flow rate. At higher solids flow and therefore higher
choking velocity, the transition can be to a fast-fluidised system, characterised
by 'clusters' or 'strings' of particles (i.e. with the dense regions dispersed in a
lean phase, rather than a slugging or bubbling bed in which lean regions are
dispersed through a dense fluidised phase). The pressure fluctuations in the
'choked' system are then less violent.

6.7.3.2 Horizontal conveying. Knowlton (1986) has reviewed the many

empirical and semi-empirical correlations for saltation velocity. A correlation
due to Zenz (1964) has been widely used. However, it is complicated to use, is
not particularly reliable (Leung and Jones, 1978), does not lend itself readily
to computer calculations and is devoid of any evident scientific basis;
 Hydraulic and pneumatic conveying 223

.EC>
"
..9!
'2
":;; Ap
a.
a. AL
~
."
2!
""'
"'2!
0..

Superficial gas velocity, U

Figure 6.10 Flow transitions in vertical pneumatic conveying (schematic).

therefore it will not be considered further here. Rizk (1973) proposed a semi-
empirical correlation which has become widely used. In SI units, the Rizk
correlation can be written
Ms Ms 10 (-1.96 - 1440d)Fr (2.5 + l1ood)
------- salt (6.20)
Mf QfUsaltA
where Frsal t is the pipe Froude number at the saltation point
(6.21)
The pressure loss in a horizontal pneumatic conveying line is obtained from
the macroscopic linear momentum equation. The pressure drop along the
same length L of pipe downstream of the solids feed point is made up of the
following terms, introduced in general terms in section 2.4.2 and illustrated
schematically in Figure 6.11:

(a) Acceleration or 'entry loss'. The mean gas velocity increases from Uf
to the mean value allowing for the fraction of the pipe occupied by the solids,
Uf /(l - Cvi ). This contributes a total pressure drop corresponding to the fluid
momentum term in equation (2.105):

(-1'l.PfA) = QfU; [ (1 - Cvi

1 f -lJ (6.22)
[= 2CviQfU; for lean phase conveying]
224 Fluid-particle systems

Solids

--+--~~~---------------
Gas • ~
:... l ------~.:
Pressure, p

• ••••
- !1 p •

..t· .
- !1 p,
..f .. ......... I!.. ...... _ ...... __ .................. __ ......... ..

Distance

Figure 6.11 Components of pressure loss in horizontal pneumatic conveying

(schematic)_ApA = acceleration or 'entry loss'; ApF = friction.

More significant is the pressure drop required to accelerate the solids to their
mean conveying velocity, V p , i.e. the solids momentum term in equation
(2.105)
(6.23)
These two terms together represent the entry loss (or acceleration term)
(6.24)

(b) Friction loss. The pressure gradient due to friction arises from shear
stresses between the wall and both the gas and the conveyed solids, as in
equations (2.104) and (2.105). Because the entry length is usually relatively
short, while the accelerational and frictional components cannot readily be
distinguished in this zone, the friction gradient is usually taken to be the value
for fully-developed flow even in the entry region. As for hydraulic conveying,
the contributions to friction from the solids and the conveying fluid are
evaluated separately and taken to be additive as in equation (6.8). At least for
lean-phase conveying, the contribution from the gas is evaluated as for single-
phase flow at the mean mixture velocity. The solids contribution can be
estimated from a correlation originally proposed by Hinkle (1953)
(-/},PsF)IL = 2JsMsVpIDA (6.25)
where (in SI units)
 Hydraulic and pneumatic conveying 225

Vp = Uf (l- 0.0638do. 3 QpO.5) (6.26)

and

In these equations Vp is an estimate for the conveyed solids velocity, while Co

is the gas/particle drag coefficient. For coarse particles, CD can be taken as
the 'Newton's law' value, 0.44.
For dense-phase conveying, the 'entry loss' term is normally negligible
compared to the friction term. Results are given in a useful review of dense
transport by Konrad (1986).

6.1.3.3 Vertical conveying. As for saltation in horizontal flow, various

correlations have been proposed for the choking velocity in vertical pipes (see
Knowlton, 1986). That due to Punwani et al. (1976) is empirical, but has the
merit of being based on a comprehensive set of data. The gas velocity (Uch )
and void fraction (lOch) at the choking point are obtained from (in SI units)
U ch
- - - Vt
Ms
=- - - - - (6.28)
lOch Qp(l - lOCh)
2250D(lO-j/ -1)
QfO. 77 = ------- (6.29)
[(Uch/lOch) - vtF
where V t is the single-particle terminal velocity and equation (6.28) assumes
that this can be taken as the slip velocity of the particles relative to the gas.
The pressure gradient in vertical conveying includes the acceleration and
friction terms which must be considered in horizontal flow, but also includes
the hydrostatic term introduced for hydraulic conveying in equations (6.15)
and (6.16)
(-/)"PH)/L = g[CyiQp + (1- Cyi)Qd (6.30)
The friction gradient can be estimated from a correlation due to Konno and
Saito (1969)

(-/)"Psp)/L = 0.057 - MsA

A
-
D
(6.31)

Clearly equation (6.31) takes no account of the characteristics of the solid

particles, and therefore cannot be satisfactory. The fact that so little attention
has been paid to friction in lean-phase vertical conveying results from the
dominance of the hydrostatic and acceleration terms. The gas contribution to
friction is evaluated, as usual, for a single-phase fluid at the mean mixture
velocity.
226 Fluid-particle systems

6.2 FLUIDISATION
When a gas is passed upwards through a settled bed of particles, several types
of behaviour are possible. Figure 6.12 (Grace, 1986) shows the six flow
regimes which are generally distinguished, although not all these regimes are
observed for all particles/gas systems, and some types of behaviour (such as
'spouting' and 'slugging' - see later) are dependent on the dimensions of the
apparatus. In general, as the gas flow increases, the bed passes through three
types of 'aggregative fluidisation', in which the particulate 'phase' is more or
less continuous, before entering a so-called 'fast fluidisation' type, in which
the fluid phase is continuous and the particles are suspended as clusters. In
this type of behaviour, 'elutriation' (the removal of particles from the
apparatus by the gas flow) is very great, so that the solids must be collected
(usually in a cyclone - see section 7.2) and returned to the bed as shown. This
gives rise to the so-called 'circulating fluidised bed' which is in wide use as a
reactor and in combustion processes. Despite extensive research work since
the early 1980s, a comprehensive understanding of the fluid mechanics of
circulating fluidised beds is not yet available. At higher gas velocities still, the
pneumatic conveying regime is reached; this type of behaviour has already
been discussed in section 6.1.3. In this section, attention is confined to the
bubbling and slug-flow regimes of fluidisation.
For a settled bed, there comes a point at which, as the fluid velocity is
increased, the drag force on the particles becomes equal to the buoyant
weight of the bed. The bed is then supported by the fluid flow and itself
possesses fluid-like properties: flowing easily, maintaining a horizontal level
and exhibiting an apparent viscosity, for example, so that this regime of
multiphase behaviour has become known as 'fluidisation'. By virtue of their
highly mobile character, fluidised beds are very good mixers and possess good

[5.9.'7;
;~;o
~.,{b6P
fiXED BED BUBBLING SLUG flOW TURBULENT FAST PNEUMATIC
OR ~IR_EG_I_M_E____~Vr-____R_EG_IM_E~1 FLUIDIZATION CONVEYING
DELAYED
BUBBLING AGGREGATIVE FLUIDIZATION

INCREASING U, t

Figure 6.12 Principal flow regimes of upward flow of gas through solid particulate
materials (Grace, 1986).
 Fluidisation 227

heat-transfer properties; they have become popular for a wide range of

process engineering applications, from catalytic reactors to particle dryers.
Although liquid-fluidised beds are becoming more widely used as bioreactors
and in biological separations, gas-fluidised beds are still of much greater
industrial importance and most of the available literature reflects this bias.
This section is concerned almost exclusively with gas fluidisation.
The literature on fluidisation is among the most voluminous on any single
engineering subject and there are several excellent books on the theory and
applications of the technique (see, for example, Geldart (1986), Davidson et
al. (1985) and the series of Engineering Foundation Conferences of which
the most recent are 0stergaard and SlZlrensen (1986), Grace et al. (1989),
Potter and Nicklin (1992) and Large and Laguerie (1996». Since the fluid
mechanics of fluidisation have been well described by so many authors, only
the briefest account of these matters will be given here. Rather more
attention will be given here to the effects of particle cohesion on fluidisation
behaviour, some aspects of which have been reviewed by Seville (1987).

6.2.1 Types of fluidisation behaviour

For a settled bed to be fully supported by an upward fluid flow, the total force
exerted by the fluid on the particles must equal the bed weight. Since, by
definition, the drag force on the particles due to the fluid flow is equal to the
manometric pressure drop across the bed (i.e. that part of the pressure drop
arising solely from the fluid motion) we can say that at this, the point of
incipient fluidisation, the manometric pressure drop, !1P, is equal to the
buoyant weight of the bed, or
(6.32)
where L is the bed height, EMF is the voidage at minimum fluidisation and
Qp and Qg are the densities of the particles and the fluid, respectively.
Umf, the superficial fluid velocity at minimum fluidisation, can be found by
combining equation (6.32) with a suitable expression for the pressure drop
through a settled bed of solids, such as the correlation due to Ergun (1952),
which was introduced in section 2.3,

(6.33)

Equating (6.32) and (6.33) gives

= 150

+ (6.34)
228 F/uid~partjc/e systems

The group on the left-hand side is dimensionless and known as the

Archimedes number, Ar. The main problem with using this equation directly
to predict Umf is that Emf is not known a priori. Wen and Yu (1966)
correlated all the available data and simplified equation (6.34) to give
Ar = 1650 Remf + 24.5 Re;;"f (6.35)
where
Remf = QgUmfd//l
from which a value of U mf may be easily obtained. For small particle sizes
(below about 100 /lm, depending on density) the second term on the right-
hand side of equation (6.35) can be neglected, so that

(6.36)

Figure 6.13 shows schematically how the pressure drop across the bed
increases with increasing superficial gas velocity (linearly if the viscous term
in equation (6.33) is the dominant one) up to the minimum fluidisation
velocity. A simple force balance shows that it is not possible for this pressure
drop to exceed the buoyant weight of the particles. At higher fluid velocities,
therefore, either the bed voidage must increase so as to maintain the pressure
drop at or below this level, or not all the fluid can flow interstitially. The
following types of behaviour are now possible:
• uniform expansion
• 'jetting', 'bubbling' or 'slugging'

liP

Possible "overshoot"

Bed weight
per unit area

Packed bad Fluidised bed

U
Umf

Figure 6.13 Dependence of the pressure drop liP on thc fluid velocity U through a
packed and f1uidised bed.
 Fluidisation 229

• 'spouting', in which the gas enters only through a single central orifice
(Mathur and Epstein, 1974)
• nucleation and growth of stationary cavities
• channelling ('rat-holing')
All of these types of behaviour, with the exceptions of spouting and
channelling, can be described as fluidisation, because both the bed and the
individual particles within it are wholly supported by the pressure drop.
Spouting and channelling cannot, because, in general, the pressure drop
during these types of behaviour is less than that required to support the bed.
There have been several attempts to devise theoretical and empirical
classifications of these behavioural types, most of which relate only to gas-
solid fluidisation. Of these, the most widely used is the empirical classification
of Geldart (1973), who divides fluidisation behaviour according to mean
particle size and density difference between the solids and the fluidising gas
(Figure 6.14). Geldart recognises four behavioural groups, designated A, B,
C and D. Typical fluidisation behaviour of groups A, Band C is illustrated in
Figure 6.15:

Group B. These particles fluidise easily, with bubbles forming at or only

slightly above the minimum fluidisation velocity.

"! 104
CI
.lO:

/
re' /
I
,/ B SPOUTABLE
cf"
,/
,/

UJ
,/
/
SAND- LIKE 0
u
z /
UJ /
0::
UJ
u.
u. 103
/
/ A
/
15 AERATABLE
>- "
/
!::
VI
"
/
Z ""
UJ
Cl ( "
COHESIVE
""""

1~~~~~~~--~~~~~~~~
10 50 100 500 100J
MEAN PARTICLE DIAMETER (,urn)

Figure 6.14 Geldart's (1973) classification of fluidisation behaviour (for air at

ambient conditions).
230 Fluid-particle systems

Group C. These particles are cohesive and tend to lift as a plug or channel
badly; conventional fluidisation is usually difficult or impossible to achieve.
Group A. These particles are intermediate in particle size and in behaviour
between groups Band C, and are distinguished from group B by the fact that
appreciable (apparently homogeneous) bed expansion occurs above the
minimum fluidisation velocity but before bubbling is observed. There is now
much experimental evidence (Seville, 1987) that group A particles are also
intermediate in cohesiveness between groups Band C, their interparticle
cohesive forces being of the same order as the single particle weight.
Group D. These particles are those which are 'large' and/or abnormally
dense. Such particles show a tendency to 'spout', rather than fluidise.
Other properties of the groups are summarised in Table 6.3 and are discussed
below.
It should be emphasised that the 'Geldart diagram' (Figure 6.15) is
applicable only to particles fluidised by air under ambient conditions, and in
the absence of artificially-enhanced interparticle forces, due to the presence
of liquid layers on the particles, for example. However, the grouping of
fluidised systems in terms of experimentally observed behaviour is of wider
application, even in cases where interparticle forces are artificially enhanced
(Seville and Clift, 1984). A later and more comprehensive classification due

GROUP B GROUP A GROUP (

Uf uf 6.P uf

H H H BEHAVIOUR ERRATIC
AND IRREPRODU5.!BLE
I'"
'" '" BUBBLING
HMF I------!"" - - - - - - 1--_ _--"1 : ~U~L.!..NG_
1 1
EXPANDING
1

u u
Figure 6.15 Typical fluidisation behaviour of particles in Geldart's (1973) groups A,
Band C (note that scales are different for each group) (Seville, 1987). H = bed height;
W/A = bed weight per unit area.
Table 6.3 Characteristic features of Geldart's (1973) classification of fluidisation behaviour (after Geldart, 1986)

Group

C A B D

Most obvious characteristic Cohesive, difficult to Bubble-free range of Starts bubbling at Umf Coarse solids
fluidise fluidisation
Typical examples Flour, cement Cracking catalyst Building sand, table Crushed limestone,
salt coffee beans
Property
1. Bed expansion Low when bed High Moderate Low
channels; can be high
when fluidised
2. Deaeration rate Can be very slow Slow Fast Fast
3. Bubble properties Channels Splitting and coales- Stable size large, and Size limited by vessel
cence predominate. may not be reached Small wake fraction
Limiting bubble size.
Large wake fraction.
4. Solids mixing Very low High Moderate Low
5. Gas back-mixing Very low High Moderate Low
6. Slug properties Solid slugs Axisymmetric; break- Asymmetric Horizontal voids.
down to turbulent Solid slugs.
fluidisation Wall slugs
7. Spouting No, except in very Shallow beds only Shallow beds only Yes, even in deep beds
shallow beds
232 Fluid-particle systems

to Grace (1986) is shown in Figure 6.16. This uses the dimensionless velocity,
U*, and particle diameter, d*, which were introduced in section 2.1.3, to
define the behavioural groups.
Figure 6.16 also shows the various processing options which might be
considered for particles of various sizes and gases of different properties.
Grace's classification successfully accounts for the effects of variation in gas
properties due to operation at elevated temperature and pressure (see later),

DILUTE CONVEYING

•
:J
T /' I

/ i GROUP 0
1 ;' I i PARTICLES
CONVENTIONAL FLUIDIZED BEDS
I I /GROUP B I
1' ,PARTICLES k
1 GROUP A / ' / '1 Proposed B- 0
1 PARTICLES Boundary
, I 1
, / II
I~
I~
: I.i~
<1;/ i
ly<>1
I I

Is ~I I

'" : ;1;'
,~ .~ ;; to.Approximote A-B 3
1 CI, I .t! 1 3
:~ Boundory lor 6(>· 10 - 2x 10

:0 L
'I fI
, .~ ~I
PACKED BEDS
'ff
Q. "
1>-
, .... I~
10-3~~~'~~__~~~~__~__~~~~~~~
10 10 2

Figure 6.16 Dimensionless superficial gas velocity versus dimensionless particle

diameter for upflow through solid particles, showing regions in which industrial
reactors operate, approximate boundaries between groups C, A and B in Geldart's
classification, and a proposed boundary between groups Band D. T indicates the
onset of turbulent fluidization (Grace, 1986).
 Fluidisation 233

but there is, as yet, no satisfactory classification which also takes into account
interparticle forces, which in many practical situations may be of considerable
importance.

6.2.2 General description of group behaviour

6.2.2.1 Group B. Many commonly-encountered experimental particles,

such as most samples of sand and glass ballotini, lie in group B, which, for a
particle density of about 3000 kg m-3 , encompasses the particle size range
from about 75 ~m to 600 ~m. In group B, as mentioned above, bubbles form
at about the minimum fluidisation velocity. Bed expansion is small, and the
bed collapses rapidly when the gas supply is cut off (Figure 6.17). Bubble rise
velocity (see section 6.2.6.1.2) depends on bubble size, but most bubbles
travel faster than the interstitial gas velocity, Vmf/Emf, so that gas tends to
circulate within the bubble, except during coalescence and splitting. There
may be a maximum bubble size but it is usually so large that bubbles will
continue to grow by coalescence until their size is limited by the size of the
apparatus.

1 Time

Figure 6.17 Typical de aeration curves (bed height, H, versus time) for powders in
groups A, Band C (Geldart, 1986). Fluidizing gas flow cut off sharply at time zero.
Uc = bed collapse rate.
234 Fluid-particle systems

6.2.2.2 Group C. Interparticle forces usually scale approximately with the

particle diameter (section 3.1) or, more precisely, the effective radius of the
particle contact, which may not be the same if there are surface asperities.
Particle weight scales with particle diameter to the third power, so it is clear
that as particle size decreases, interparticle forces become more significant.
The extent of group Con Geldart's diagram corresponds to van der Waals
forces alone; other forces, such as those arising from liquid bridges, can be
much stronger (see Figure 3.17), and group C-type behaviour can extend to
larger particle sizes if the interparticle forces are enhanced. In practice,
interparticle forces cannot be reliably estimated theoretically, and experiment
is essential. The hardness of the solids at the operating conditions is
important, soft solids being more cohesive, and the fluidising gas may have an
influence, particularly if it has a high humidity (see Chapter 3).
GrollP C powders will readily form stable channels from the distributor to
the surface, and may also lift as a cohesive plug, particularly if the apparatus
is small. The pressure drop across the bed usually remains below the bed
weight per unit area, and mixing and heat transfer are poor. Fluidisation can
sometimes be made possible by increasing the gas velocity to break up the
cohesive structure, or by mechanical stirring or vibration. For example,
Chaouki et al. (1985) were able to fluidise fine aerogels (d p < 20 !lm; Qp =
70 kg m-3 ) by causing spontaneous agglomeration into clusters of about 1 mm
in size. Fluidisation can sometimes be promoted by adding a small proportion
of fumed silica or some other sub micron powder; these reduce the
interparticle forces by modifying the contact geometry. Electrostatic forces
can be reduced by making the apparatus conducting and by (moderate)
humidification of the gas stream. Once group C powders do become aerated
their de-aeration times are comparatively long (Figure 6.17).

6.2.2.3 Group A. As mentioned earlier, group A particles are those which

exhibit a region of non-bubbling expansion for gas velocities above the
minimum fluidisation velocity. (In earlier literature, non-bubbling expansion
is known as 'particulate' fluidisation, by contrast with 'aggregative' bubbling
fluidisation.) Geldart (1973) defines a minimum bubbling velocity, Umb , and
designates group A particles as those for which UmblUmf > 1. The non-
bubbling expansion of a group A bed can be characterised in terms of the
Richardson and Zaki (1954) equation (section 2.3)
U
__ = En (6.37)
Vt
where V t is the particle terminal velocity in an infinite medium and n is a
function of the particle Reynolds number at the terminal velocity, normally
taking values between 4.65 and 2.4. This equation was first used to correlate
the homogeneous expansion of liquid fluidised beds (where the particle
 Fluidisation 235

Reynolds number is low, so that n is usually about 4.65; see section 2.3) and
the fact that it can also be used to characterise the bubble-free expansion of
gas fluidised beds has been used to suggest that the expansion mechanism is
the same in both cases. However, as Martin (1983) points out, the comparison
is misleading. In liquid fluidisation, expansion is uniform and particle-particle
contacts are transient. In group A fluidisation, bed expansion is thought to
occur by nucleation and growth of cavities whose sizes range from a few to
about ten particle diameters (Massimilla et al., 1972; Donsi and Massimilla,
1973). The surrounding particles maintain the surface contacts which are
essential for the stability of the structure. Further evidence for the non-
uniform mode of expansion of group A beds is given by Geldart and Wong
(1984) who showed that the Richardson-Zaki index, n, increases with
decreasing particle size below about 100 !-lm, as the interparticle forces are
able to stabilise the expanded structure more effectively. Other aspects of the
stability of expanded non-bubbling beds are reviewed by Seville (1987).
As the superficial gas velocity exceeds Umb , the passage of bubbles breaks
up the expanded structure, causing a decrease in bed height (Figure 6.15) as
the dense-phase voidage is reduced to somewhere between Emf and Emb.
When the gas supply is suddenly cut off, the bed initially collapses rapidly as
the bubbles leave (Figure 6.17) and then continues much more slowly. This
property of slow de aeration is responsible for the ease with which fluidised
group A solids are maintained in circulation around catalytic cracking plants,
for example, but is also responsible for their tendency to 'flood' on discharge
from hoppers (Geldart and Williams, 1985).
In bubbling group A beds, all bubbles travel faster than the interstitial gas,
but a tendency towards bubble splitting (see section 6.2.6.1.3) limits the size
to which they can grow by coalescence. Circulation and mixing are rapid, bed-
to-surface heat transfer is favourable, and gas exchange between the bubbles
and the dense phase is high. All of these factors, together with a larger solid
surface area per bed volume than for groups Band D, favour the use of group
A particles in fluidised bed reactors.

6.2.2.4 Group D. The distinction between groups Band D concerns the

rise velocity of the bubbles, which is, in general, less than the interstitial gas
velocity in group D beds, so that gas flows into the base of the bubble and out
of the top. Because of the size and density of the particles, the permeability of
the bed is high, so that the minimum fluidisation velocity is also high. Gas and
solids mixing is low, but cohesive solids can be fluidised because the greater
momentum of the particles on impact and fewer particle-particle contacts per
unit bed volume reduce the tendency towards agglomeration.
If gas is introduced over a small part of the distributor, group D particles
can be made to spout (Mathur and Epstein, 1974). In this flow regime, a high-
velocity lean-phase spout forms through the bed. The particles in the
annular region around the spout are not supported by the gas flow, but move
236 Fluid-particle systems

downwards to be entrained into the spout, so that there is a continuous

circulation of particles up in the spout and down in the annulus. In practice, it is
often advantageous to exploit this technique and to use a spouted bed rather
than a f1uidised bed when processing or handling group D particles.

6.2.3 Criteria for group boundaries

6.2.3. 7 AlB boundary. Purely hydrodynamic models for stability of homo-

geneously f1uidised beds (see, for example, Foscolo and Gibilaro, 1984) have
been successful in predicting the range of stability of liquid f1uidised beds, but
because they fail to take proper account of the role of interparticle forces such
models are inappropriate for prediction of the AlB boundary. Rietema and
Mutsers (1973) carried out a stability analysis on the expanded bed, assuming
expansion to be uniform and incorporating a dense-phase 'elasticity' term to
take into account interparticle cohesion. However, attempts either to predict
or to measure the elasticity have so far failed so that it is necessary to fall back
on rather simpler semi-empirical predictions.
As mentioned earlier, the criterion for group A type behaviour is that
UrnblUrnf > 1 (6.38)
Abrahamsen and Geldart (1980) correlated values for U rnb using
d p","g
0°. 06
U rnb = 2.07 exp (0.726F)--
flO. 347
(6.39)

where F is the mass fraction of the powder <45 flm in size (another
illustration of the effect of fines on interparticle forces). The constant is
dimensional, so SI units must be used. For air at ambient conditions and
F = 0.1, this reduces to
(6.40)
(Again, SI units must be used.) Geldart (1986) used his own correlation for
U rnf and substituting this and equation (6.39) into equation (6.38) obtained
the boundary expression for group A type behaviour

2300Q~·126flo.523 exp(0.716F)
--~~~------------ > 1 (6.41)
cfi,.8 gO.934(Qp _ Qg)O.934

For air under ambient conditions this reduces to

0°. 934 do. 8 < 1 (6.42)

"'"P P
 Group C properties
reported
~

m '"0
E GROUP 0
en Group A properties
.><: 3
m x reported
0
~ V):, ~ e
x '(//\ GROUP B
2'
CI
"
c£" 'J 'J ~ ~1jI IjI
() ()OO . xx I!I
10..
0- x x xl Il
~.~~ ~
ell
\/
u
C
ell
0
. 1!II!II!II!II!It!I •
... e
~
()
~
~ 0,5 Group B properties
-0 GROUpe GROUP A/' reported
>.
iii
-
c
••
~ 0,2 Group 0 properhes
reported
x gas/solid
.. liquid/solid
20 50 100 200 500 1000 2000

Mean particle size dp(\Jml

Figure 6.18 Boundaries between Geldarfs groups (Molerus, 1982),

238 Fluid-particle systems

In a rather different approach, Molerus (1982) suggested that the

differences between groups A, Band C arise from the relative importance of
the interparticle forces and the drag of the fluidising gas, and developed semi-
empirical conditions for the boundaries. He defined group B type behaviour
as that for which interparticle forces are negligible by comparison with fluid
drag. By definition, in a fluidised bed the drag on a particle must
counterbalance its immersed weight. Hence, on the NB boundary

-------------= K (6.43)

where FH is the adhesion force transmitted in a single contact between two

adjacent particles. Taking the case of 100 ~m glass ballotini, which lie
approximately on the boundary, and estimating FH due to van der Waals
forces alone, Molerus obtained K = 0.16. The condition resulting from
equation (6.43) is plotted in Figure 6.18 as a hatched region, the limits
representing the transitions to be expected for 'hard' and 'soft' materials. For
'soft' materials, group A extends further to the right on the diagram. Molerus'
values for FH were obtained from purely theoretical equations for van der
Waals forces, which are probably overestimates (see section 3.1.1). Other
estimates, and direct measurements (Seville, 1987), suggest that K is about an
order of magnitude too low; however, this does not alter the transition lines
on Figure 6.18 since equation (6.43) is forced to fit the case of 100 ~m
ballotini.
Grace (1986) has correlated data for particles near the AlB boundary in
gases other than air and at elevated pressures and temperatures. In terms of
the dimensionless particle diameter introduced in section 2.1.3, his expression
for the boundary is:

(6.44)

6.2.3.2 Ale boundary. The boundary between groups A and C is clearly

influenced primarily by the strength of the interparticle forces. By similar
reasoning to that which led to equation (6.43), Molerus (1982) proposed the
transition expression

= 10-3 (6.45)

This equation is also plotted in Figure 6.18. On the basis of experimental

work on a range of materials, Wong (1983) proposed
 Fluidisation 239

Experimental measurements in this region are usually hard to reproduce, and

possibly dependent on equipment scale, so that it is, at present, difficult to
generalise about the position of the boundary. As a strictly empirical way of
determining in which group a powder lies, Geldart et ai. (1984) have proposed
the use of a simple test procedure to determine the 'Hausner ratio', the ratio
of the highest and lowest bulk densities (i.e. the 'tapped' and 'poured' bulk
densities, respectively). The Hausner ratio is another crude indication of
interparticle force; looser pac kings can be stabilised by higher interparticle
forces, while the densest packing is unaffected by interparticle forces so that
higher values of the Hausner ratio imply larger interparticle forces. Geldart et
ai. (1984) estimated that the transition from group A to group C behaviour
occurs at Hausner ratios between 1.25 and 1.4.

6.2.3.3 BID boundary. As mentioned earlier, there are two criteria for a
particle system to be in group D:
(i) that the bubble velocity, Ub, is less than the interstitial gas velocity, i.e.

(6.47)

(ii) that the particles can be spouted.

Neither is a very good criterion, since (a) there is a range of bubble sizes in
any fluidised bed, so that not all bubbles will satisfy equation (6.47), even in a
group D bed, and (b) whether or not a powder can be spouted depends not
only on the particles themselves but also on the apparatus, specifically the
ratio between the particle diameter and the gas entry orifice diameter
(Chadnani and Epstein, 1986). However, Geldart (1986) showed that both
criteria (i) and (ii) above give roughly similar expressions: the bubble rise
condition demands that for group D behaviour U mf > 0.3-0.5 m S-l
(approximately), while the spouting condition requires that
(Qp - Qg)d~·24 > 0.23 (6.48)
The reader is referred to the original references for the (numerous)
assumptions which have been made in these derivations.

6.2.4 Temperature and pressure effects

For particles in all groups, the qualitative effects of temperature and pressure
on the minimum fluidisation velocity can be obtained from consideration of
240 Fluid-particle systems

the gas density and viscosity terms in the Wen and Yu (1966) correlation
(equation 6.35). For fine particles, Umf decreases with increasing temperature
and is hardly affected by pressure, reflecting the change in gas viscosity, while
for coarse particles Urnf increases with increasing temperature and decreases
with increasing pressure in response to changes in gas density. However, it
must be noted that equation (6.35) must be used with caution if there are
appreciable interparticle forces, because these may affect Emf, which is
subsumed within the constants of that equation (see later).
The effect of temperature and pressure on Umb are predicted by equation
(6.39), which is in reasonable agreement with the results of King and
Harrison (1980). However, Piepers et ai. (1984) showed that Umb and other
measured characteristics of a fine catalyst under pressure depend strongly on
the type of gas which is used; independent measurements suggested that this
was due to different levels of gas adsorption on the solid surface, which is
known to influence the van der Waals forces. It is therefore to be expected
that the position of the AlB boundary will depend not only on temperature
and pressure but also on gas adsorption onto the particles.
The effect of temperature on the AIC boundary will depend on whether the
sintering temperature has been exceeded. This is considered briefly in the
next section.
There is some evidence (King and Harrison, 1980) that the minimum
spouting velocity decreases with increasing pressure; hence, the BID
boundary should move to the right as the pressure is increased, bringing
larger particles into group B. Grace (1986) also considers the BID boundary,
and proposes another correlation to locate it.

6.2.5 Defluidisation and cohesive effects

Researchers have frequently worked with dry glass spheres or sand; real
particles may often be sticky due to the presence of free liquids (see section
3.1.2), or because they are above their sintering temperature, or for some
other reason. Increase in interparticle forces for whatever reason generally
gives rise to an increase in the minimum fluidisation velocity, through an
increase in Emf, i.e. the formation of a looser bed structure (Seville, 1987).
The increase in interparticle forces enables this looser bed structure to remain
stable. If interparticle forces become very large, defluidisation or 'quenching'
can occur; this may be thought of as a transition to Geldart's group C (Seville
and Clift, 1984). The defluidisation point is usually taken as the point at which
the bed quenches (almost all movement ceases) as the gas velocity is reduced.
This is often approximately the same as the point at which the bed starts to
fluidise, measured with increasing velocity, although there are usually
hysteresis effects, and sometimes time-dependent ones. Gluckman et ai.
(1976) plotted operating diagrams such as Figure 6.19, defining regimes of
fluidisation and defluidisation in the gas velocityltemperature plane, for
 Fluidisation 241

OPERATING
DIAGRAM
>-
~
u
o
...J
UJ
FLUIDISED
>
VI
«
l!J
...J
:;!;
~
LL
0:::
UJ
c... DEFLUIDISED
::>
VI

TEMPERATURE

DlLATOMETER
TRACE
:z:
t-
l!J
Z
UJ
...J

~
UJ
l!J
Z
«
:z:
u

Minimum
sintering
temperature

Figure 6.19 High-temperature defluidisations - operating region diagram and

corresponding dilatometer trace (Gluckman et al .• 1976).

group B copper, polymer and glass particles. In all cases, it was observed that
the temperature at which the measured defluidisation velocity departed from
the predicted behaviour in the absence of interparticle forces coincided with
the 'initial sintering temperature', as measured using a dilatometer. This
approach was developed further and applied by Tardos et al. (1984, 1985). It
was also observed that normal fluidisation could be recovered by increasing
the gas velocity, provided that the bed had not been allowed to remain
defluidised for too long.
It may be noted that the cohesive effects of, for example, liquid layers and
sintering are qualitatively different. The presence of a liquid layer may
influence whether the outcome of a particle collision is sticking or rebound
(see sections 3.4.2 and 6.2.7); this is an almost instantaneous effect. Sintering,
however, is time-dependent and in order to predict whether particles brought
into contact will fuse by sintering, it is necessary to consider the time they are
242 Fluid-particle systems

able to spend in contact, which will depend on the local bubble frequency and
therefore on the fluidising gas velocity.
There is experimental evidence (see Seville, 1987) that a small increase in
interparticle forces in a group B material can cause it to exhibit group A type
behaviour, i.e. Umb/Umf > 1. Such behaviour would be consistent with
Molerus' (1982) boundary expressions, and suggests that it is possible to move
from group B to A to C simply by an increase in interparticle forces alone.

6.2.6 Bubbling fluidised beds

Most gas-fluidised beds operate in the bubbling reg:me (also known,
misleadingly, as the 'aggregative' or 'heterogeneous' regime), by contrast
with liquid fluidised beds which usually show non-bubbling behaviour (also
known as 'particulate' fluidisation). To a first approximation, in a bubbling
fluidised bed only sufficient fluid to support the particles flows interstitially,
while the rest of the gas passes through the bed as distinct 'voids'. The two-
phase picture of a bubbling fluidised bed, therefore distinguishes between:
(a) the dense phase (alias 'particulate phase' or 'emulsion phase') consisting
of the bed particles fluidised by interstitial gas;
(b) the lean phase consisting of rising voids virtually free of bed particles.
Because these voids are superficially like gas bubbles in liquids (actually like
bubbles in engine oil rather than water - see below), the voids are usually
called 'bubbles' and the lean phase is also known as the 'bubble phase'. The
gas in the lean phase is not in direct contact with the particles. Hence, in a
fluidised bed in which a reaction is carried out involving both gas and solids, it
is important to understand the division of gas flow between the two phases,
and the processes by which material can be transferred between them.
A bubbling bed can be regarded as a bed in which the bubble phase is
dispersed and the particulate phase is continuous - as in a bubbling liquid. At
higher velocity, the proportion of the bed volume occupied by the bubbles,
Eb, increases. It may become sufficiently high that the bed can no longer be
described as 'lean phase dispersed/particulate phase continuous'; a state in
which the two 'phases' are so interspersed that neither can be described as
continuous is one way of interpreting the poorly-defined regime known as
'turbulent fluidisation' (see Figure 6.12). At higher velocities still, Eb
becomes so high that the 'lean phase' is continuous, with the 'particulate
phase' dispersed in it; this probably corresponds to 'fast fluidisation' although
there is, as yet, no universally-accepted definition of what constitutes a 'fast-
fluidised bed'.
It is important to realise that any resemblance between bubbles in fluidised
beds and in liquids is only superficial. Furthermore, the analogy is with liquids
of moderately high viscosity (in the range of automobile engine oils) and not
with low-viscosity liquids like water. The most important differences arise
 Fluidisation 243

from the nature of the bubble boundary. For a bubble in a liquid, the
boundary is a true interface dividing two phases. It has a measurable and
recognisable surface tension and, while material can diffuse across the
interface, there can be no bulk convection across the bubble boundary. By
contrast, a bubble boundary in a fluidised bed is not a sharp interface and has
no property really analogous to surface tension. It is just a boundary between
a region in which the particle concentration is very low, and a region in which
the particle concentration is of the same order as that in the whole bed at
incipient fluidisation. This boundary is permeable, so that the gas can, and
does, flow through it. Thus interphase transfer in a fluidised bed can result
both from diffusion and from bulk convection.

6.2.6. 1 Bubbles

6.2.6.1.7 Bubble shape and particle mlxmg. Figure 6.20 shows an

idealised section through a single bubble in a fluidised bed. The bubble
volume is Vb' The upper surface is approximately spherical, with radius of
curvature r. The base is typically slightly indented. The volume filling in the
sphere is called the wake and, to a good approximation, this volume Vw
consists of dense phase rising with the bubble. Bubble shapes vary according
to the properties of the bed particles. For large or irregular group B materials,
Vw is typically one-quarter of the sphere volume (i.e. one-third of the bubble
volume). In group A particles, the bubbles tend to be flatter so that the wake
is proportionately larger. In group D beds, the bubbles are much more round
and are roughly spherical. Further details, including experimental evidence,
are given by Clift (1986).
The wake volume has an important effect on particle motion. The rising
sphere, corresponding to the bubble plus its wake, displaces the surrounding

Bubble
volume Vb

Wake
volume Vw

Figure 6.20 Spherical cap bubble.

244 Fluid-particle systems

dense phase: the effect is roughly equivalent to dragging up a volume equal to

V2(Vb + vw), by the process known as drift transport. If Vw = Vb/3, then the
drift volume is roughly equal to 2Vb/3. Therefore, the total transport of dense
phase, by drift and in the bubble wake, is
(6.49)
i.e. to a first approximation, a bubble rising through a fluidised bed transports
its own volume of dense phase. It is this rapid turnover of the bed particles
which gives fluidised beds many of their important properties, such as good
temperature uniformity.
The foregoing refers to isolated bubbles. Bubbles whose dimensions
approach those of the bed behave rather differently and are known as slugs
(see below). When two bubbles coalesce, the lower one usually accelerates to
catch up with the leading bubble. In this process the leading bubble becomes
flattened sideways and the lower bubble elongates in the vertical direction
(Clift and Grace, 1986).

6.2.6.7.2 Rise of a single isolated bubble. Theoretically (Davies and

Taylor, 1950), a large bubble III a fluid of relative low viscosity rises at
velocity
(6.50)
Bubble velocities in fluidised beds vary erratically, but equation (6.50) seems
to give a good estimate for mean rise velocity. In a freely-bubbling bed, the
rise velocity is greater than the value given by equation (6.50), and this is
discussed further below.
Usually the radius of curvature, r, is not known. The characteristic
dimension frequently used is the volume-equivalent sphere diameter, d v , i.e.
the diameter of a sphere whose volume is equal to the bubble volume Vb.
The rise velocity of an isolated bubble is then given by
Vb = Kv'g(l;, (6.51)
where K depends on the relationship between rand d v . Commonly it is
assumed that K = 0.71 as for bubbles in water, giving
Vb=O.71~ (6.52)

(a) (b)

Figure 6.21 Bubble spitting (Clift, 1986).

 Fluidisation 245

Equation (6.52) appears to be a good approximation for group D particles,

but for groups A and B the values of K lie in the range 0.54).6 (Clift, 1986).

6.2.6.1.3 Bubble splitting. Bubbles in fluidised beds break up by the

process shown schematically in Figure 6.21. An indentation forms on the
upper surface of the bubble and grows as it is swept around the periphery by
the particle motion. If the curtain grows sufficiently to reach the base of the
bubble before being swept away, the bubble divides. Splitting dominates in
beds of relatively fine particles (typically group A) and also tends to become
more frequent at elevated pressures.
Splitting is generally taken to occur if the bubble exceeds a maximum stable
size, (dv)max. The analysis of Harrison and de Kock (see Davidson and
Harrison, 1963) is based on an erroneous model (see Clift and Grace, 1986)
but still seems to give reasonable estimates for (dv)max as
(dv)max = 2V;/g (6.53)
where V t is the terminal velocity of the bed particles in the fluidising gas.
Geldart (as quoted in Clift (1986)) has shown that equation (6.53) gives the
best estimates if V t is calculated for a particle diameter 2.7 times the mean
diameter present in the bed although any fundamental reason for this is not
clear. Figure 6.22 shows values of the maximum stable bubble diameter for

f
(dv)max
V lOOOK
300K
10m

$
'<l

f/
"-
1m 3OOK

,/
1~
0'1m

1em I

h
1mm L -_ _--'-_ _ _-L._ _-.J

10,um 100 p.m 1000 p.m

dp

Figure 6.22 Predicted maximum stable bubble size as a function of the mean size of
the powder and gas pressure or temperature (Clift, 1986).
246 Fluid-particle systems

particles of density 2500 kg m-3 , fluidised by air. Consistent with observed

bed behaviour, the bubbles remain small in group A materials, so that the
mean size of bubbles present is typically controlled by splitting. However, for
group D and large group B particles, the maximum stable bubble size is so
large that in practice it would not limit the size of bubbles present. Equation
(6.53) also predicts that increasing temperature and pressure tend to reduce
the maximum stable bubble size, making fluidisation 'smoother'. This is
qualitatively correct, although the strong effects sometimes observed at
elevated pressures probably arise at least partially from interparticle forces
(see earlier) rather than purely hydrodynamic effects.

6.2.6.2 Slugs
Provided that d v is less than about D18, where D is the bed diameter, the
shape and rise velocity of a bubble are unaffected by the walls. For larger
bubbles, wall effects reduce the rise velocity and change the bubble shape to
reduce the wake friction. For d)D greater than about 0.6, the bubble takes up
the characteristic shape shown in Figure 6.23(a) and is known as a slug.
Increasing slug volume simply adds further to the cylindrical body of the slug,
without changing the shape of the nose. Slugs sometimes adhere to the walls
of the bed, as shown in Figure 6.23(b).
These slug shapes are found typically in beds of groups A and B. For
coarse, angular or cohesive particles, or for groups A and B in very narrow
tubes, plugs rather than slugs tend to be formed. Plugs are slow-moving voids
which fill the tube so that the bed particles 'rain through' them rather than
flowing around. 'Plug flow' is usually undesirable and can be suppressed by
roughening the bed walls (Geld art et al., 1978).
Wall effects retard bubbles. For 0.125 < d)D < 0.6, the rise velocity can be
estimated roughly (Wallis, 1969) as

(a) (b) (c)

Figure 6.23 (a) Axisymmetric slug. (b) Asymmetric slug. (c) Square-nosed slug or
'plug' (Clift, 1986).
 Fluidisation 247

U;'= 1.13Ub exp(-d)D) (6.54)

where Ub is the rise velocity of the same bubble free from wall effects. For
d)D > 0.6, i.e. for true slugs, the rise velocity is independent of slug volume
and completely controlled by the walls (Clift et al., 1978)
Us = 0.35vgD (6.55)
For 'wall slugs', as in Figure 6.23(b), the rise velocity is about 1.4 times that
from equation (6.55).

6.2.6.3 Continuous bubbling and slugging

6.2.6.3.1 Overall flow patterns. When a bed of particles is fluidised at a

gas velocity above the minimum bubbling point, bubbles form continuously
and rise through the bed, which is said to be 'freely bubbling'. Bubbles
coalesce as they rise, so that the average bubble size increases with distance
above the distributor until the bubbles approach the maximum stable size.
Thereafter, splitting and recoalescence cause the average bubble size to
equilibrate at a value close to the maximum stable value. For the reasons
noted in section 6.2.6.1.3 above, this is a particular feature of group A
particles where the bubble size is typically constant throughout much of the
bed.
Bubble coalescence can also have an influence on circulation of the dense
phase. The effect is shown schematically in Figure 6.24(a). Bubbles usually
coalesce by overtaking a bubble in front (Figure 6.24(b )(i» and may move
sideways into the track of a bubble (Figure 6.24(b )(ii». Thus coalescence can
cause lateral motion of bubbles. Bubbles near a bed wall can only move
inwards, because bubbles surrounding them are only on the side away from
the wall, while bubbles well away from the walls are equally likely to move in
any horizontal direction. As a result of this preferential migration of bubbles
away from the wall, an 'active' zone of enhanced bubble flow forms a small
distance in from the wall. In this zone, coalescence is more frequent so that
the bubbles become larger than at other positions on the same horizontal
plane. Because the region between the 'active' zone and the wall is depleted
of bubbles, coalescence continues to cause preferential migration towards the
bed axis. Eventually the 'active' zone comes together to form a 'bubble track'
along which the lean phase rises as a stream of large bubbles. Large beds may
divide into several 'cells', with several preferential bubble tracks in the bed.
Because of the transport of dense phase by the bubbles (see section 6.2.6.1.1
above), the solids tend to move up in regions of high bubble activity and down
elsewhere. In the upper levels, the motion is up near the bubble tracks and
down near the walls. At lower levels, the particle motion is down near the axis
and outwards across the distributor; this motion can in turn enhance bubble
activity near the walls close to the distributor.
248 Fluid-particle systems

(a)

iii i i
0
a
(i)
C) (ii)

(b) 6
Figure 6.24 (a) Bubble and solids flow pattern and (b) coalescence models:
(i) bubble overtake; (ii) bubble sideways motion (Clift, 1986).

6.2.6.3.2 Bubble and slug flow rates. The bubble flow rate (also known
as the visible flow rate) in a fluidised bed, Qb, is defined as the rate at which
bubble volume crosses any level in the bed. A first estimate for Qb is given by
the 'two-phase theory of fluidisation' (Davidson and Harrison, 1963), which
conceives the bed as consisting of two phases: (a) a dense phase in which the
gas flow rate is equal to the flow rate for incipient fluidisation', i.e. the
voidage is constant at the minimum fluidisation value; (b) a bubble phase
which carries the additional flow of fluidising fluid.
The visible flow rate is then estimated as
(6.56)
where V is the superficial fluidising velocity and Vmf the value at minimum
fluidisation. In words, the simple two-phase theory can be stated as:
 Ffuidisation 249

The excess gas flow above that necessary for minimum fluidisation passes
through the bed in the form of bubbles.
Extensive experimental work since this theory was first proposed has
revealed certain shortcomings. As a first approximation, however, equation
(6.56) can be used as stated, noting that Qb is actually rather lower in
practice. The discrepancy is typically of order 10% for group A, 20-30% for
group Band 40-50% for group D (Clift, 1986; Clift and Grace, 1986). There
is some evidence that the discrepancy is less in slugging beds.

6.2.6.3.3. Bubble size and rise velocity. Correlations to estimate the

development of bubble size in a fluidised bed abound in the literature, and
are reviewed by Clift (1986) and Clift and Grace (1986). Most of them are
based on the same empirical data, so it is perhaps not surprising that most of
them give broadly similar predictions. However, most are also based on data
for quite small beds, so that their application to full-scale beds must be
treated with due scepticism. The result due to Darton et at. (1977) is probably
as reliable as any, and is relatively convenient to use. The mean bubble size
formed at the distributor dy,o is first estimated by an expression whose form
has a fundamental theoretical basis in the analysis of Davidson and Schuler
(see Davidson and Harrison, 1963),
dy,o = 1.63 [Ao(U - Umf )]0.4g-D.2 (6.57)
where Ao is the distributor area associated with one gas inlet orifice or nozzle.
The effect of coalescence on bubble growth above the distributor is then
given by
dy=0.54(U- Umf)0.4(z+4~)O.8g-0.2 (6.58)
where Ao is the distributor area associated with one inlet orifice or nozzle.
use equation (6.58) (or an equivalent) up to the height at which d y reaches the
maximum stable size, (dy)max, and to assume that dy = (dy)max at higher
levels. Typically the maximum bubble size is reached close to the distributor
in beds of group A particles. In group B beds, the maximum stable size is
usually only reached if the bed is very deep, is operated at high gas velocity,
and is sufficiently large that slug flow is not reached first. Equation (6.58) is
not applicable to group D materials, for which Cranfield and Geldart (1974)
have proposed an alternative correlation of equivalent form.
The rise velocity of bubbles in a freely-bubbling bed is problematic.
Davidson and Harrison (1963) suggested that the mean bubble rise velocity,
U A, is given by
(6.59)
where Ub is the rise velocity of a single isolated bubble (see above). Equation
(6.59) is pleasingly simple, and it does predict correctly that bubbles in freely-
250 Fluid-particle systems

bubbling beds rise faster than isolated bubbles of the same size. Unfortu-
nately, the theoretical basis for equation (6.59) is at best questionable (see
Clift and Grace, 1986). In fact, VA> Vb primarily as a result of the
increase in bubble velocity caused by coalescence. There is some evidence
that equation (6.59) accounts approximately for the average effect of
coalescence. In this case, it can be applied as a first estimate to a bed overall,
although instantaneous bubble velocities vary widely about this result and the
mean velocity for a given bubble diameter varies with position in the bed,
being highest in the 'active' bubbling zones where coalescence is frequent.

6.2.6.3.4 Bed expansion. Consider a section as AA' in Figure 6.24,

across which the visible bubble flow rate is Qb and the average bubble velocity
is VA. The average fraction of the bed area at this level occupied by bubbles is
then
(6.60)
The volume of dense phase in an elementary height dz at this level is then
A(l - tb)dz, so that the total volume of dense phase in the whole bed is
A J:H(l- tb)dz =A [,H(l- QbIAVA)dz (6.61)
o 0
According to the two-phase model, this volume is equal to the total dense
phase volume at the minimum fluidisation point, AHmf . Therefore, after
some algebraic manipulation

(6.62)

It is worth re-emphasising that equation (6.62) assumes that the bed

expansion, H - Hmf> results solely from the presence of bubbles in the bed. If
U A is estimated from equation (6.59) and a single average value is used, based
on the average bubble size in the whole bed, equation (6.62) simplifies to

H - Hmf V - Vmf
(6.63)
Hmf Vb
This result, first proposed by Davidson and Harrison (1963), has been very
widely used to predict bed expansion. For more precise estimates, it is
probably best to use equation (6.62) with V A based on estimates of bubble
size given, for example, by the correlation of Darton et al. (1977) (see above).

6.2.7 Agglomeration
Although this area is largely neglected by researchers into fluidisation,
agglomeration is both a major problem and a major application of
 Fluidisation 251

fluidisation. If particles are sticky, often due to the presence of free surface
liquid, they will tend to stick together. This can lead to defluidisation (see
section 6.2.5) but if controlled can be used as a method of manufacture of
agglomerated products.
Agglomeration of a powder can be used to change the shape, porosity and
bulk density of a material as well as the particle size. This can bring
improvements in the appearance, flow properties and solubility of the
material and reduce the amount of dust present. Agglomeration can be used
to ensure a well-dispersed non-segregating blend of, say, a high-cost material
with a low-cost filler or an active material with an inert filter. Comprehensive
reviews of particle-size enlargement processes including agglomeration have
been produced by Capes (1980), Sherrington and Oliver (1981) and Pietsch
(1991).
All methods of agglomeration have in common the mobilisation of
adhesive or 'binding' forces and disruptive or 'break-up' forces, and it is the
relationship between these which gives rise to the particular properties of the
products which are associated with the different agglomeration processes. In
most cases, the binding forces arise from the deliberate addition of a binding
agent (Krycer et al., 1983), usually in the form of a liquid spray. The binding
agent may remain in liquid form, or solidify/evaporate to leave solid layers or
bridges between particles. The binder can be sprayed into the bed through an
atomising nozzle which can be placed above, in or below the bed (see Figure
6.25). Circulation patterns in the bed cause particles to enter the spray region
and collide with atomised binder droplets. They then experience particle-
particle impacts as they circulate through the bed and return after some
circulation time to the spray region.
An attractive feature of fluidised bed agglomeration is that it can combine
the agglomeration and drying steps in one piece of equipment. Drying in
fluidised beds is extremely rapid so that the bed usually contains a large
proportion of dried agglomerates which periodically recycle through the
binder feed zone in order to pick up more binder.
During the agglomeration process, single particles or wet or dry agglomer-
ates are brought into contact with each other at a certain relative velocity.
Their kinetic energy may be converted to elastic or plastic deformation
(viscous dissipation) or, in the case of break-up, the creation of new surface.
As a result, the particles may coalesce, stick to each other, rebound and/or
break-up, depending on the kinetic energy of each partner and their
cohesional and adhesional strengths. Under different circumstances, and in
different types of agglomerator, agglomeration behaviour may be controlled
by the wet or dry agglomerate strength, and breakage may occur due to gross
fracture, particle attrition or some combination of these.
The agglomeration rate and the final size of agglomerates are both affected
by parameters which can change the balance of forces in the agglomerator
(Nienow and Rowe, 1985). When the binding forces (which depend on the
 252 Fluid-particle systems

(b)

Figure 6.25 Fluidised bed agglomerator (Glatt Protech Ltd.). (a) top spray;
(b) bottom spray ('Wurster' column).
 Fluidisation 253

binder properties as well as the particle size) are initially large compared with
the break-up forces, the embryonic agglomerate grows. This will be the case
with coarse binder atomisation, high concentration binder and low operating
gas velocities. If the binding strength is low compared with the break-up
forces, the embryonic agglomerate will break down before it can re-enter the
feed zone, often leaving the constituent particles with a coating or a patch of
dried binder. In time, this may lead to successive layers of coating, known as
'onion-ring' growth. One disadvantage of fluidised bed agglomeration by
comparison with other methods is this sensitivity to the formulation/operation
combination, although their relatively gentle action does mean that fluidised
beds can be used to produce agglomerates of relative low bulk density with
the open structure suitable for 'instant' products. Figure 6.26 illustrates the
effect of operating variables on growth rate for a strongly-agglomerating
system. Here, the use of a higher operating gas velocity or a larger starter
particle size leads to slower growth and a maximum stable agglomerate size.

6.2.7. 1 Modelling of agglomeration. Despite its widespread use,

agglomeration is poorly understood. For example, it is not generally possible
to proceed from a knowledge of the process parameters and the nature of the
particles and the liquid to a prediction of the mode of particle growth (e.g.
pair-wise agglomeration or 'layering') or the rate of growth. Attempts at
population balance modelling of agglomeration processes are clearly fruitless
until there is agreement on basic mechanisms.
The microscopic origins of agglomeration due to liquid addition are well
known and have been summarised in Chapter 3. For a full description of the
agglomeration process, it is important to be able to predict the initial
agglomeration behaviour, i.e. whether collisions between wet particles will
give rise to rebound or 'capture'. An analysis of the impact between a particle
and a wet surface (section 3.4.2) suggests that the outcome depends on the
value of a Stokes number which includes the relative velocity, the particle size
and the viscosity of the wetting fluid.
The Stokes number may be considered as proportional to the ratio of the
kinetic energy of particles to the energy absorbed in collision. Thus, there is a
critical value of the Stokes number St* (of order 1), above which rebound will
occur and below which particles will adhere. This approach has been
proposed by Ennis et al. (1991) to explain qualitatively the features of
agglomeration processes in general. In any agglomerator there will be a range
of particle velocities and possibly a maldistribution of binding liquid, giving
rise to a distribution of operating Stokes number, St. Considering now Figure
6.27(a), if the distribution of operating Stokes numbers falls below St*,
agglomeration will be 'non-inertial' and all collisions will be 'successful'. The
rate of coalescence is then independent of particle or granule kinetic energy
and the initial distribution of binder is all-important. This is typically the case
when the particle size is very small.
254 FlUid-particle systems

600 (a)

------;0::---°
500 o

400

300

0 60 120 '80 240 300

t Imin)

3·00 (b)

2·60

2·20
0- d po = 245 fJm
o

~"
-00.. '.80

'·40

,·00

0 60 120 lBO 240 300

t Imin)

Figure 6.26 Granulation in a strongly agglomerating system: (a) the effect of excess
gas velocity (glass powder; 5% Carbowax; constant U - Umf = 0.65 n S-l); (b) the
effect of starter particle size, d po (glass powder; 5% Carbowax; constant U - Umf =
0.525 m S-1) (Nienow and Rowe, 1985).

As the average operating Stokes number increases, by granule growth, for

example, the value of St associated with some collision events may exceed St*
(Figure 6.27(b)) and rebound will occur. Net growth will still be achieved but
agglomeration may be more controlled and there will be, in effect, a 'growth
limit'. This is the 'inertial regime' where increases in binder viscosity or
 Fluidisation 255

St 8t* 8t 8t
8t*

X x
8t*
----
X
(a) (b) (e)

Figure 6.27 Stokes numbers in agglomeration (x indicates a space coordinate) (Ennis

et al., 1991).

decreases in granule kinetic energy will increase the overall rate of growth as
intuitively expected. Further increases in St (Figure 6.27c) so that all values
exceed St* and all collisions are 'unsuccessful' will lead to a change of
behaviour to coating rather than coalescence. All of these qualitative aspects
of agglomeration behaviour are well-documented experimentally.
It should be emphasised that in general not only the agglomerate size but
also the binder properties change with time, due to evaporation. Other things
being equal, evaporation of the binder will make coalescence more likely,
through increase in the binder viscosity. A further factor is that the liquid
loading and liquid properties determine the rate at which an agglomerate is
able to deform and consolidate (Ennis et al., 1991) - if the binder becomes
viscous very rapidly the agglomerates may have insufficient time to
consolidate and may therefore remain as loose open structures (Figure
6.28(a)), which may have good 'instant' properties but poor resistance to
attrition (Mullier et al., 1991). This, of course, opens up the possibility of
controlling granule morphology by a combination of binder selection and
careful control of the evaporation rate.
There are several problems with the approach outlined above. One is that
the Barnocky and Davis (1988) analysis (section 3.4.2), on which the analysis
of Ennis et al. (1991) is based, considers only the energy loss on impact due to
viscous dissipation in the liquid layer between two approaching spheres. In
reality, there are other dissipation mechanisms if the 'particles' are them-
selves agglomerates. Impact may lead to extensive plastic deformation
throughout the agglomerate and/or to fracture and break-up. The liquid
content and properties within the agglomerates are crucial here, because they
determine the 'deformability' of the two bodies; small liquid bridges lead to
brittle behaviour while large ones allow larger extension of particle-particle
bonds before rupture occurs, leading to a more ductile behaviour. Kristensen
et al. (1985) have attempted to relate the strength and deform ability of
agglomerates to their composition, and have established experimentally a link
between these properties and their granulation behaviour which is consistent
with the reasoning above.
A further problem is that one does not know, in general, the distribution of
256 Fluid-particle systems

(a)

(b)

Figure 6.28 Products of fluidised bed agglomeration: (a) a rapidly solidifying binder;
(b) a slowly solidifying binder (Ennis et at., 1989).
 Nomenclature 257

particle impact velocities in a fluidised bed or any other particular piece of

process equipment. This aspect is not, at present, amenable to precise
theoretical prediction, although experimental techniques for tracing particles
and determining their velocities are developing rapidly (see, for example,
Stein et at., 1997).

6.3 NOMENCLATURE
A Pipe cross-sectional area (m 2 )
Ao Distributor area associated with one gas inlet point (m 2 )
Cv Volumetric concentration of solids
Cvd Delivered volume concentration
Cvt In situ volume concentration
Cw Solids concentration by weight
D Internal diameter of pipe (m), bed diameter (m)
d Particle diameter (m)
dp Mean bed particle diameter (m)
dy Volume-equivalent bubble diameter (m); Vb = Jtd~/6
(dv)max Maximum value of d v limited by stability (m)
FH Interparticle force (N)
If Moody friction factor
g Gravitational acceleration (m S-2)
H Bed depth
Hmax Maximum bed depth (m)
Hmf Bed depth at minimum fluidisation point (m)
if Hydraulic gradient for vehicle alone (m m-I )
im Hydraulic gradient for slurry in terms of vehicle head (m m-1)
jrn Hydraulic gradient for slurry in terms of head of delivered
slurry (m m- 1)
Length along pipe (m)
Mass flow rate of conveying fluid (kg S-1)
Mass flow rate of solids (kg S-1)
Pressure (Pa)
r Radius of curvature of bubble surface (m)
Sm Relative density of slurry
Ss Relative density of solids
SEC Specific energy consumption (equation (6.7»
V Superficial gas velocity (m S-I)
Vf Superficial fluid velocity (m S-1)
Vb Rise velocity of single isolated bubble (m S-1)
Vch Value of V f at choking point (m S-1)
VA Bubble velocity in freely bubbling bed (m S-1)
V rnb Gas velocity at minimum bubbling (m S-I)
258 Fluid-particle systems

Umf Gas superficial velocity at minimum fluidisation (m S-1)

Usalt Value of Uf at saltation point (m S-1)
US Rise velocity of single slug (m S-1)
Vt Terminal velocity of single particle (m S-I)
Vm Mean mixture velocity in pipe (m S-I)
Vp Mean solids velocity (m S-1)
VS Mixture velocity at limit of deposition (m S-l)
Vsm Maximum value of Vs over all solids concentrations (m S-l)
Vb Volume of bubble (m 3)
Vw Volume of bubble wake (m 3 )
w Velocity characterising particle behaviour, equation (6.9) (m S-I)
z Distance above distributor (m)

Greek letters
E Void fraction
Shear viscosity of fluid (kg m-1 S-l)
Particle density (kg m-3 )
Density of fluid (kg m-3 )
Gas density (kg m-3 )
Mean slurry density (kg m-3 )
Shear stress at pipe wall (Pa)

REFERENCES
Abrahamsen, A.R. and Geldart, D. (1980) Powder Technol., 26, 35.
Bagnold, R.A. (1941) The Physics of Blown Sand and Desert Dunes, Methuen,
London.
Barnocky, G. and Davis, R.H. (1988) Phys. Fluids 31 (6), 1324.
Brown, N.P. and Heywood, N.!. (eds.) (1991) Slurry Handling: Design of Solid-
Liquid Systems, Elsevier Applied Science, London.
Capes, C.E. (1980) Particle Size Enlargement, Elsevier, Amsterdam.
Chandnani, P.P. and Epstein, N. (1986) Fluidization. V, 0stergaard, K. and
S0rensen, A. (eds.), Engineering Foundation, New York, p. 233.
Chaouki, J., Chavarie, C., Klvana, D. and Pajonk, G. (1985) Powder Technol. 43,
117.
Clift, R. (1986) Hydrodynamics of Bubbling Fluidized Beds, in Gas Fluidization
Technology, (ed. D. Geldart), Wiley, Chichester.
Clift, R. (1996) Powder Technol., 88, 335.
Clift, R. and Clift, D.H.M. (1981) Int. f. Multiphase Flow 7, 555.
Clift, R. and Grace, J.R. (1986) in Fluidization, Davidson, J.F., Clift, R. and
Harrison, D. (eds.), Ch. 3, Academic Press, New York.
Clift, R., Grace, J.R. and Weber, M.E. (1978) Bubbles, Drops and Particles,
Academic Press, New York.
Cranfield, R.R. and Geldart, D. (1974) Chern. Eng. Sci. 27,2309.
Darton, R.C., LaNauze, R.D., Davidson, J.R. and Harrison, D. (1977) Trans. I.
Chern. E. 55,274.
 References 259

Davidson, J.R. and Harrison, D. (1963) Fluidised Particles, Cambridge University

Press, Cambridge.
Davidson, J.F., Clift, R and Harrison, D. (eds.) (1985) Fluidization, 2nd edn,
Academic Press, London.
Davies, R.M. and Taylor, G.I. (1950) Proc. Roy. Soc. A 200, 375.
Donsi, G. and Massimilla, L. (1973) A. 1. Chem. E. J. 19 (6), 1104.
Ennis, B.J., Seville, J.P.K., Mullier, M.A. and Adams, M.J. (1989) AIChE Annual
Meeting, unpublished abstract.
Ennis, B.J., Tardos, G. and Pfeffer, R (1991) Powder Technol. 65,257.
Ergun, S. (1952) Chem. Eng. Prog. 48,89.
Foscolo, P.U. and Gibilaro, L.G. (1984) Chem. Engng. Sci. 39,1667.
Geldart, D. (1973) Powder Techno!. 7, 285.
Geldart, D. (1986) in Gas Fluidization Technology, Geldart, D. (ed.), Wiley,
Chichester.
Geldart, D., Hurt, J.M. and Wadia, P.W. (1978) AIChE Symp. Ser. 74(176),60.
Geldart, D. and Williams, J.C. (1985) Powder Technol. 43, 18I.
Gluckman, M.J., Yerushalmi, J. and Squires, A.M. (1976) in Fluidization Tech-
nology, Keairns, D.L. (ed.), Hemisphere, Washington.
Grace, J.R. (1986) Can. J. Chem. Eng. 64,353.
Grace, J.R., Shemilt, L.W. and Bergougnou, M.A. (eds.) (1989) Fluidization. VI,
Engineering Foundation, New York.
Hinkle, B.L. (1953) PhD Thesis, Georgia Institute of Technology.
Jackson, R (1985) Hydrodynamic Stability of Fluid-Particle Systems, in Fluidization
Davidson, J.F., Clift, R. and Harrison, D. (eds.), Ch. 2, Academic Press, London.
Jacobs, B. (1991) Design of Slurry Transport Systems, Elsevier Applied Science,
London.
King, D. and Harrison, D. (1980) Powder Technol. 26, 103.
Knowlton, E.M. (1986) 'Solids transfer in fluidized systems', in Gas Fluidization
Technology, Geldart, D. (ed.), Ch. 12, Wiley, Chichester.
Konno, H. and Saito, S.J. (1969) Chem. Eng. Japan 2, 21I.
Konrad, K. (1986) Powder Technol. 49, 1.
Kristensen, H.G., Holm, P. and Schaefer, T. (1985) Powder Technol. 44,227.
Krycer, I., Pope, D.G. and Hersey, J.A. (1983) Powder Techno!. 34,39.
Large, J. -F. and Laguerie, C. (eds.) (1996) Fluidization VIII, Engineering Foundation,
New York.
Leung, L.S. and Jones, P.J. (1978) Paper Dl, Proceedings of Pneumotransport 4,
BHRA Fluid Engineering, Cranfield, UK.
Marcus, RD., Leung, L.S., Klinzing, G.E. and Rizk, F. (1990) Pneumatic Conveying
of Solids, Chapman and Hall, London.
Martin, P.D. (1983) Chem. Eng. Res. Des. 61, 318.
Massimilla, L., Donsi, G. and Zuccini, C. (1972) Chem. Engng Sci. 27,2005.
Mathur, K.B. and Epstein, N. (1974) Spouted Beds, Academic Press, New York.
Molerus, O. (1982) Powder Technol. 33, 8I.
Molerus, o. (1993) Principles of Flow in Disperse Systems, Chapman and Hall,
London.
Mullier, M.A., Seville, J.P.K. and M.J. Adams (1991) Powder Technol. 65, 32I.
Nienow, A.W. and Rowe, P.N. (1985) in Fluidization, 2nd edn., Davidson, J.F., Clift,
R. and Harrison, D. (eds.), Academic Press, London.
0stergaard, K. and S0renson, A. (eds.) (1986) Fluidization V, Engineering
Foundation, New York.
Piepers, H.W., Cottaar, E.J.E., Verkooijen, A.H.M. and Rietema, K. (1984) Powder
Techno!. 37, 55.
Pietsch, W. (1991) Size Enlargement by Agglomeration, Wiley, Chichester.
260 Fluid-particle systems

Potter, O.E. and Nicklin, D.J. (eds.) (1992) Fluidization VII, Engineering Foundation,
New York.
Punwani, D.V., Modi, M.V. and Tarman, P.B. (1976) International Powder and Bulk
Solids Handling and Processing Conference, Chicago.
Richardson, J.F. and Zaki, W.N. (1954) Trans. I. Chern. E. 32, 35.
Rietema, K. and Mutsers, S.M.P. (1973) Proceedings of the International Symposium
on Fluidisation, Toulouse, 28.
Rizk, F. (1973) Doktor-Ing Diss., Universitat Karlsruhe, Germany.
Seville, J.P.K. (1987) in Tribology in Particle Technology, Briscoe, B.J. and Adams,
M.J. (eds.), Adam Hilger, Bristol.
Seville, J.P.K. and Clift, R. (1984) Powder Technol. 37, 117.
Sherrington, P.J. and Oliver, R. (1981) Granulation, Heyden, London.
Shook, e.A. and Roco, M.e. (1991) Slurry Flow: Principles and Practice,
Butterworth-Heinemann, New York.
Stein, M., Martin, T.W., Seville, J.P.K., McNeil, P.A. and Parker, D.J. (1997) in
Non-invasive Monitoring of Multiphase Flows, Chaouki, J., Larachi, F. and
Dudukovic, M.P. (eds.), Elsevier, Amsterdam, p. 309.
Tardos, G., Mazzone, D. and Pfeffer, R. (1984) Can. 1. Chern. Eng. 62,884.
Tardos, G., Mazzone, D. and Pfeffer, R. (1985) Can. J. Chern. Eng. 63,377,384.
Wallis, G.B. (1969) One-Dimensional Two-Phase Flow, McGraw-Hill, New York.
Wen, e.Y. and Yu, Y.H. (1966) A. I. Chern. E. J., 12, 160.
Wilson, K.e. (1979) Proc. Hydrotransport 6, BHRA Fluid Engineering, 1.
Wilson, K.C., Addie, G.R. and Clift, R. (1992) Slurry Transport using Centrifugal
Pumps, Elsevier Applied Science, London.
Wilson, K.C., Addie, G.R., Sellgren, A. and Clift, A . .(1996) Hydraulic Conveying of
Solids with Centrifugal Pumps, 2nd ed., Chapman and Hall, London.
Wong, A.C.Y. (1983), PhD Dissertation, University of Bradford.
Woodcock, e.R. and Mason, J.S. (1988) Bulk Solids Handling, Chapman and Hall,
London.
Zenz, F.A. (1964) Ind. Eng. Chern. Fundamentals 3, 65.
7
Gas/solid separation

The separation of particles from gases is one of the fundamental technical

problems of engineering, finding increasing application in many industries,
both for environmental protection and prevention of fouling and damage to
equipment. The devices which have been proposed to achieve this separation
are astonishingly varied in design, in the physical principles upon which they
rely, and in their effectiveness. Classification of separation devices is
therefore difficult, but this is normally done on the basis of the physical
principles which are applied. For example, Svarovsky (1981) uses the
classification:
• aero-mechanical dry separators;
• aero-mechanical wet separators ('scrubbers');
• electrostatic precipitators;
• filters.
In this chapter, attention will be confined to the first and last of these
categories. Useful general views of gas/solid systems can be found in
Svarovsky (1981, 1986), Strauss (1975) and Brown (1993).

7.1 GAS AND PARTICLE PROPERTIES

In general, the gas and particle properties in a particular application must be
regarded as 'given' and these have a profound effect on the choice of
separation device and its effectiveness. For example, temperature and
pressure affect the gas properties, which has consequences for particle
motion. Temperature, pressure and chemical composition can also have
strong effects on particle-particle cohesion and on particle-surface adhesion
(see Chapter 3), which can also affect the performance of gas cleaning
equipment. These effects are to some extent system-specific, but some useful
general points can nevertheless be made.
Over the range of pressures currently conceived as relevant in industrial gas
cleaning, departures from ideal gas behaviour are negligible. Therefore the
density of a gas at pressure P and absolute temperature T can be
approximated by

J. Seville et al., Processing of Particulate Solids

© Chapman & Hall 1997
262 Gas/solid separation

=-=-- (7.1)
v RT
where M is the mean relative molecular mass of the gas, v the molar volume
and R the universal gas constant. With P in bars, T in Kelvin, and g in
kg m-3 , R is 8.3143 kJ kmol- 1 K-1 .
Elementary kinetic theory gives a first approximation for the effect of
temperature and pressure on gas viscosity. Viscosity is predicted to be
independent of pressure and proportional to the square root of absolute
temperature. In practice, the temperature dependence is usually stronger but,
at least over the pressure range of interest here, it can be assumed that
pressure has no significant effect on gas viscosity (Reid et al., 1977).
These general conclusions are useful for quick estimates of the pressure
drop associated with gas cleaning equipment. For filters in which the
Reynolds number of the gas flow is small, e.g. for most barrier filters (see
later sections), a form of the Carman-Kozeny equation applies (see section
2.3.2):
t!.P ex: [.tU (7.2)

where U is the actual face velocity through the filter. Therefore, for constant
volumetric throughput of gas,

(7.3)

where n is 0.5 or slightly larger. For constant mass throughput of gas,

t!.P ex: [.t/g ex: T(1+n) [Tl (7.4)
For high Reynolds number devices, i.e. inertial separators and some types of
filters with high face velocity,
t!.P ex: gU2 (7.5)
Therefore, for constant volumetric throughput,
t!.P ex: P T- 1 (7.6)
while, for constant mass throughput,
t!.P ex: g-l ex: T p-1 (7.7)
In addition to these macroscopic properties, the mean free path of gas
molecules, J..., is inversely proportional to density (see section 2.1.4), so that
A ex: T/P (7.8)
The resultant effects on fluid/particle drag and Brownian diffusivity are
reviewed in section 2.1.4. In summary, slip effects are significant for sub-
micron particles at ambient conditions and at elevated pressures, but are also
 Inertial separators 263

significant for particles several microns in diameter at elevated temperature

and ambient pressure. The Brownian diffusivity of a particle in a gas is
proportional to temperature but inversely proportional to gas viscosity and
particle diameter (equation 2.45»:
D AS ex: CTllA-d (7.9)
where C is the Cunningham 'slip correction' factor (see section 2.1.4). Taking
IA- ex: T' as above,
D AS ex:CT(1-n)a-1 (7.10)
Thus diffusivity increases with temperature and with decreasing particle size.
The effect of C is to make the dependence on T and d even stronger,
especially for sub-micron particles at high temperature.
Many particle-removal processes require the particles to migrate relative to
the gas, so that the drag of the gas on the particles is of prime interest. The
particle Reynolds number is defined as
(7.11)
where d is the particle diameter and V the velocity of the particle relative to
the gas. In most gas cleaning devices, Rep remains small. The drag force, FD
can then be estimated from Stokes law with a correction to allow for 'slip
effects' which arise when the particle diameter is comparable to the mean free
path of gas molecules (see section 2.1.4).
Some particulate collection devices, particularly barrier filters, rely on the
collected particles to form large agglomerates which have high terminal
velocity and are therefore removed easily from the gas when detached by
mechanical action. Other devices, notably granular bed filters, rely for their
operation on adhesion of dust to the filter medium and cohesion between dust
particles. Interparticle forces, which have been reviewed in Chapter 3, are
generally system-specific because they are strongly dependent on the
properties of the particle surface. Their effects on separation performance are
considered in the sections which follow.

7.2 INERTIAL SEPARATORS

7.2.1 Introduction
Inertial separators concentrate or collect particles by changing the direction
of motion of the flowing gas, in such a way that the particle trajectories cross
over the gas steamlines so that the particles are either concentrated into a
small part of the gas flow or are separated by impingement onto a surface.
Much the most widely used type of inertial separator is the cyclone, in which
the gas undergoes some kind of vortex motion so that the gas acceleration is
264 Gas/solid separation

centripetal, and the particles therefore move centrifugally towards the outside
of the cyclone. Almost all practical cyclones induce the vortex motion
'passively', by appropriate design of the gas flow channel so that the device
has no moving parts. Some rotating devices - perhaps more properly called
'gas centrifuges' - have been proposed but have seen very limited application.
The principal types of cyclone design are illustrated in Figure 7.1. Figures
7.1(a) and 7.1(b) show axial-flow cyclones, where the vortex motion is
induced by stationary vanes. Axial-flow cyclones are used for fairly
specialised duties, typically with many such cyclones mounted in parallel (for
reasons developed below) to give a relatively compact device to handle a
large gas flow; a widespread use is in the air intakes of diesel engines to
operate in dusty environments such as deserts. However, the commonest
cyclone in industrial use in the reverse-flow cyclone, shown schematically in
Figure 7.1(c), where the vortex motion is induced by introducing the gas
tangentially to a cylindrical section called the cyclone barrel. The gas exits
through an axial pipe sometimes called the vortex finder. The end of the
vortex finder extends beyond the gas outlet, so that the gas executes a helical
outer vortex in the barrel and tapered section or cone, before moving into the
much narrower inner vortex and leaving through the vortex finder. The
particles are flung out to the wall while, in a properly designed cyclone, the
helical motion in the outer vortex pushes them down towards the apex of the
cone. Most commonly, cyclones are mounted vertically for easier discharge of
particles from the cone, but they can be (and sometimes are) used in different
orientations. A disengagement hopper is sometimes provided to control
particle discharge, as shown in Figure 7.1(c). However, in some applications
where the particles are retained within the process, a dip-leg is used instead,
in the form of a pipe down which the disengaged particles move in dense-
phase flow. This is common practice in collecting particles from the gases
leaving fluidised bed reactors or crackers or combustors, and the dip-leg then
usually extends down below the surface of the bed.
The gas entry to the cyclone is critical. It is preferable to ensure a sufficient
length of straight pipe - at least several pipe diameters - before entering the
cyclone. The different gas entry configurations are exemplified by the
archetypal designs of Stairmand (see Stairmand, 1951; Stairmand and
Kelsey, 1955) shown in Figure 7.2, which are almost 'industry standards'.
A simple tangential entry, shown in Figure 7.2(a), is cheaper to construct but
may give higher pressure drop across the cyclone (see below); Stairmand used
the tangential entry for his 'high efficiency, medium throughput' design. For
lower pressure drop (but increased construction cost), a volute (or 'scroll' or
'wrap-around') entry can be used, as in Stairmand's 'high throughput,
medium efficiency' design shown in Figure 7 .2(b).
Other types of inertial separator have not seriously impinged on the
popularity of the cyclone, which remains the 'workhorse' of gas-particle
separation. The Cardiff Cyclone (Syred et al., 1986), which exists in a range of
 Inertial separators 265

(a)

f'l-ol-------Dust header -----~

Fixed vanes
Ta secandary
circuit

-
-
Cleaned
gas stream

Ta secondary
circuit

Clean gas out

Oust out

Figure 7.1 Cyclone designs. (a) Fixed impeller through-flow (Strauss, 1975);
(b) axial entry reverse flow (Strauss, 1975); (c) tangential entry reverse flow
(Institution of Chemical Engineers, 1985).
266 Gas/solid separation

0'50
h1

Cl
In
N

,
(a) (b)

Figure 7.2 Standard cyclone designs (Strauss, 1975, after Stairmand, 1951). (a) High-
efficiency, medium-throughput pattern. Nominal flow rate = 1.5D 2 m3 S-1.
(b) Medium-efficiency, high-throughput pattern. Nominal flow rate = 4.5D2 m 3 S-1.
Entrance velocity at these flows is approximately 15.2 m S-1 in both types.

variants, is essentially a Stairmand high-efficiency cyclone with distributed

gas entry, and with a separate 'vortex collector' in the gas entry to pre-collect
the larger particles. The air-shielded cyclone uses a secondary gas flow as an
annular vortex 'shield' between the dirty gas inlet and the vortex finder, to
reduce entrainment into the exit gas flow (Giles, 1982). The electrodyne
cyclone was an attempt to use electrophoretic migration (as in an electrostatic
precipitator) to enhance the centrifugal separation (Boericke et al., 1981).
None of these devices has really achieved acceptance, presumably because
the enhanced performance does not compensate for the increased complica-
tion and cost. In the specific case of the electrodyne cyclone, it was successful
in the sense that the efficiency was increased when the electric field was
activated, but unsuccessful in that the efficiency even then was lower than for
a cyclone with no electrode disturbing the flow pattern. Other, simpler
devices have been proposed - for example the wedge separator of Linhardt
(1981) - and are sometimes used as preliminary collectors for flows with very
high dust loadings.
Given that the cyclone remains the standard first-stage particle collector,
the following discussion is based on the reverse-flow cyclone. However, a
 Inertial separators 267

pragmatic approach is recommended, based on scaling using dimensionless

groups, so that the principles can readily be adapted to any other inertial
separator.

7.2.2 Analysis of cyclone performance

Perhaps because cyclones have been 'taken for granted' as reliable process
units, fundamental description of particle separation has not advanced very
far. Mechanistic models have been proposed by Leith and Licht (1972), Dietz
(1981) and Mothes and Laffler (1984). Clift et al. (1991) have shown that
there are basic errors in the models of Leith and Licht and of Dietz, and have
reformulated them on a correct conceptual basis; however, even the corrected
models have limited practical value. The model of Mothes and Laffler
requires a value for the particle diffusivity caused by fluid turbulence in the
cyclone body, and this is not usually available. It therefore might be used for
scale-up or assessing the effect of design modifications, but has limited use for
design. Boyson et al. (1982) initiated an approach in which the gas motion in
the cyclone is computed, enabling particle trajectories to be simulated to
generate collection efficiency predictions. This approach is promising, albeit
for low particle loadings only at this stage (see below), but the computing
resources are substantial. It has therefore led to a general computational fluid
dynamics (CFD) approach for swirling flows (and to some commercial
software packages), rather than an approach to cyclone design and analysis.
In the absence of detailed and accepted theory, the process engineer must
resort to the common practice of dimensional analysis. For most purposes,
this is quite adequate, because reputable cyclone suppliers have good data on
performance which can readily be scaled or extrapolated using appropriate
dimensionless groups.

7.2.2. 1 Pressure drop. Starting with the pressure drop across the cyclone,
I1P depends on the cyclone dimensions, the gas volumetric flow rate, Q, and
the gas density and viscosity, Q and !.l
I1P = f(D, Q, Q, !.l) (7.12)
where D is a characteristic cyclone dimension, typically the barrel diameter.
Equation (7.12) applies for the gas flow alone; particle loading effects are
discussed below. Given the five parameters in equation (7.12) and that the
system has the usual three dimensions (mass, length and time), the problem
can be expressed in terms of two (i.e. 5 minus 3) dimensionless groups. These
are, most conveniently, the pressure coefficient
1t = I1PI(QW2)
1t2 D 4 11P
(7.13)
268 Gas/solid separation

and a cyclone Reynolds number

QUeD
Ree=---
Il
4QQ
(7.14)
JtIlD
so that, for a given cyclone geometry,
Jt = f(Ree) (7.15)
In these equations, Ue represents a 'superficial velocity' through the cyclone

(7.16)

and is used here as a measure of gas velocity. The inlet velocity is typically of
order lOUe, the precise value depending on the geometric design of the
cyclone.
In practice, cyclones operate at very high Reynolds number so that, as for
many devices which operate in the range where flow is fully turbulent and
dominated by inertia, the effect of changes in Reynolds number can be
neglected and the pressure coefficient taken as constant.
The effective value of Jt depends on the cyclone design; it is affected by
both inlet and outlet designs, and the relative importance of inlet and outlet is
still not clear. The equation of Shepherd and Lapple (1940) is as good as any
(Hoffmann et aI., 1992)
Jt 2 D 4
Jt = k - - (7.17)
16abD~

where a and b are the dimensions of the rectangular gas entry and Dc is the
diameter of the vortex finder (see Figure 7.3). The parameter k depends on
the cyclone geometry, but is approximately 16 for tangential and volute
entries. Equation (7.17) should only be used for estimates, in the absence of
test data on any particular design being considered.

7.2.2.2 Col/ection efficiency. The efficiency of particle collection, 'Yj, by

an inertial separator, operated at given gas properties and throughput,
depends on particle diameter, d, and density, Qp. The relationship between
'Yj and d, for given Qp' is called the grade efficiency curve. Grade efficiency
curves for the two Stairmand designs of Figure 7.2 are shown in Figure 7.4.
These curves refer to low inlet particle loading; the effect of particle loading is
discussed below. As for pressure drop, a simple scaling approach can be used
to apply measured grade efficiency curves to different conditions.
For low inlet particle loading, and assuming that particles are sufficiently
 Inertial separators 269

T
s
h

Figure 7.3 Reverse flow cyclone with dimensions.

large that 'slip' effects can be ignored, the collection efficiency for a particle
of diameter d depends on the following variables,
'l1 = f[d, Qp' Q, [l, D, Q] (7.18)
where Qp is the particle density. Proceeding as above, the problem can be
generalised by writing it in terms of (six minus three), i.e. three dimensionless
groups. (Collection efficiency is already dimensionless.) Hence we can write
'l1 = f[St, Re, dID] (7.19)
where St is the Stokes number (see section 2.2) derived from the ratio of the
stopping distance to cyclone diameter
Qp d 2 Ue
St=--
18[lD
2 Qp~Q
(7.20)
9n [lD 3
However, equation (7.19) can be simplified. As noted above, variations in Re
have negligible effect. Similarly the size ratio, d/ D, is usually so small that its
effect can be ignored. This leaves
270 Gas/solid separation

100
90 -' l -
80 I
~ 70
g
>-
Q)
60
~ 50
~ 40
~
<!J
30
20
(a)

-
10

100
90
V
/
80
0 II
0 I
0 l
0
0'
20
(b)
0
0' I I
10 20 30 40 50 60 70
Particle size (11m)

Figure 7.4 Fractional efficiency curves for 200 mm diameter cyclones, inlet velocity
15.2 m S-l, solid density 2000 kg m-3 in air at 20o e. Strauss (1975) after Stairmand
(1951). (a) High-efficiency cyclone. (b) High-throughput cyclone.

1'] = f[St] (7.21)

Equation (7.21) shows that the grade efficiency can be expressed in a general
non-dimensional form by expressing 1'] as a function of St rather than d. In
particular, the particle size which can be collected by a cyclone is commonly
expressed in terms of the cut size, d so , i.e. the particle size which is collected
(at low particle loading) with 50% efficiency, 1'] = 0.5. From equations (7.21)
and (7.20), for a given geometric cyclone design,
QptPsoQ
= constant (7.22)
 Inertial separators 271

or
(7.23)
Equation (7.23) shows that the cut size decreases (i.e. the cyclone efficiency
improves) if the throughput increases, and also that the performance of small
cyclones is better than that of large cyclones, all other things being equal.

7.2.2.3 Cyclone systems. The preceding general results can be used to

obtain some useful conclusions on cyclone design, selection and operation.
For further results in this vein, see Svarovsky (1981, 1986).
For given gas and particle properties (i.e. Q, !l and Qp fixed) from
equation (7.13)

(7.24)

while, from equation (7.23),

d so ex (D 3fQ)'h
ex D'h/j"p-Y4 (7.25)

so that lower d so (i.e. improved collection performance) can also be achieved

at the expense of higher pressure drop.
Equations (7.23) to (7.25) also underline a common practice: the use of a
number of identical cyclones in parallel. For N cyclones handling a total gas
flow rate Q, the flow per cyclone should be
q=QfN (7.26)
so that, from equation (7.24)
/j"p ex q 2fD 4
ex Q2fN2D4 (7.27)
To maintain collection performance, from equation (7.23) D 3 fq must be kept
constant; i.e. ND 3fQ must be constant, implying that
(7.28)
for given total flow rate Q. From equations (7.27) and (7.28)
/j"p ex Q2Ai'/3 (7.29)
i.e. for given collection performance, the pressure drop can be reduced by
using a number of cyclones in parallel. In practice, the number of parallel
cyclones is limited by the difficulty of ensuring uniform gas distribution
between them, although this general result explains the interest in devices
using batteries of small cyclones in parallel, usually of some axial flow design
(see section 7.2.1).
272 Gas/solid separation

7.2.3 Effects of solid loading

The preceding development referred to operations in which the particle
loading in the gas entering the cyclone is low, less than about 10 g m-3 at
ambient conditions. Cyclones are rare amongst process plant in showing
improved performance with increased loading, i.e. AP goes down and 'I'J
goes up with increasing solids concentration. Therefore a design based on low
loading is conservative for high loading. The reasons for the effects of solids
loading are still not fully understood (see Hoffmann et al., 1992) but some
broad conclusions are worth noting.

7.2.3. 1 Pressure drop. According to the analysis of Muschelknautz (1980)

increasing particle loading increases wall friction, which in turn lowers the
tangential gas velocity and so reduces overall pressure drop. However, the
effect observed is much less than predicted by the Muschelknautz analysis
(Hoffman et al., 1992). It is therefore necessary to use empirical results to
estimate pressure drop, for example that due to Smolik (see Svarovsky, 1981)
which does appear to be reliable (see Hoffman et al., 1992)

(7.30)

where Jtc is the pressure coefficient at inlet loading C (kg m-3 ) and Jt o is the
pressure coefficient for gas flow alone.

7.2.3.2 Particle capture. Muschelknautz (1980) has also given an analysis

of the effect of solids loading on particle capture, by relating it to phenomena
in lean-phase pneumatic conveying. The idea is that there is a 'critical load' of
particles which can be transported by the gas; any material in excess of this
load is 'dumped' to the wall and captured by the cyclone. The part of the
solids above the 'critical load' is then completely collected, independent of
particle size, while the balance is collected with the 'inherent' grade efficiency
of the cyclone at low loading, 'l'Jo. The total collection efficiency is then given
by

'l'Jc = (1 - C/C) + 'l'JoC/C (7.31)

where Ct is the 'critical load' . This result is consistent with the observed effect
of loading on grade efficiency (Hoffman et al., 1992); in particular, the
collection efficiency is improved for the smaller particles, which are taken out
along with the larger particles in the 'supercriticalload'. However, the actual
values for the critical load do not match Muschelknautz' theory. At the
present time, it remains impossible to predict the effect of loading on capture
efficiency with any precision.
 Inertial separators 273

7.2.4 Effects of temperature and pressure

The effects of operating conditions on collection efficiency are most
conveniently seen through the effect on the cut size. From equation (7.1)
Q oc PIT (7.32)
and
(7.33)
where n is 0.5 or slightly larger (see section 7.1). From equation (7.23),
d so oc V(!-lD 3IQpQ)

(7.34)
so that the cut size increases (i.e. collection efficiency deteriorates) with
increasing temperature. The available experimental data generally support
this simple result (e.g. Abrahamson and Allen, 1986). However, the
temperature effect revealed by equation (7.34) is relatively slight; generally,
use of cyclones at elevated temperatures has been straightforward, except
sometimes for mechanical problems or problems in discharging particles from
the cyclone (see below).
The effects on pressure drop can be estimated in a similar way. From
equation (7.13),
!!.P oc QQ2fD4 (7.35)
while, from equation (7.34), keeping cut size constant implies
Q oc T'D 3 (7.36)
Therefore, if a cyclone of the same size is used at the extreme temperature
and pressure,
!!.P oc PT(2n-l)

(7.37)
On the other hand, if the same velocity is used (i.e. QfD2 constant) so that
smaller cyclones are needed, then from equation (7.34)
!!.pocPT- 1 (7.38)
On either basis, increasing pressure increases the pressure drop needed across
the cyclone. If cyclone size is kept constant, then increasing temperature
increases !!.P modestly, from equation (7.37). Equation (7.38) shows that if
smaller cyclones can be used for elevated temperature, then !!.P can actually
be reduced. However, this again raises the problem of distribution between
cyclones in parallel, noted above.
274 Gas/solid separation

The results for the effect of particle loading, discussed above, are untested
at conditions remote from ambient. The qualitative effects are almost
certainly the same at elevated temperature (and pressure), because the nature
of the physical processes should be the same. However, it is not likely that
Smolik's purely empirical equation (7.30) for the effect of loading on !1P will
be applicable at high temperature and pressure. At least there is some
comfort in the expectation that a cyclone designed for low particle loading
should perform better at high loading.

7.2.5 Concluding remarks

In practice, cyclones cannot be used to collect particles smaller than about
5 [tm. Although the results summarised above suggest that smaller particles
should be collected if D is small enough or Ue large enough, extrapolation to
extreme conditions invalidates the assumptions which form the basis of the
simple treatment summarised here.
Furthermore, nothing has been said here about the serious practical
problem of maintaining discharge (see Chapter 8) from the cyclone of the
particles collected. This can be critical, and is the practical limit on the solid
loading which a cyclone can handle. For example, blockage of the solids
discharge in a cyclone used for turbine protection can lead to particle
concentrations sufficiently high to damage the turbine before the problem is
detected. For this reason, combined cycle power plants tend to use one or two
stages of cyclone followed by some form of barrier filter.

7 .3 FILTRATION
7.3.1 General features of filtration behaviour
Filtration is the commonest method of particle removal from a fluid, and
refers to the process by which the particle laden gas passes through a
permeable filter 'medium', which consists commonly of a membrane or an
array of fibres or collector particles. The particles to be separated from the
gas - often known as the 'aerosol' - make contact with the collecting surfaces,
or other aerosol particles which have already been deposited thereon, and are
removed. Examples of filter media are shown in Figure 7.5(a). Most
commonly, industrial filters consist of arrays of vertically-hung cylindrical
elements ('bags' in the case of flexible fabric filters; 'candles' in the case of
rigid media), as shown in Figure 7.5(b). The flow is from the outside inwards,
and exits via a plenum chamber at the top. Many other geometries are
known.
A distinction can be drawn between two main types of filtration behaviour:
'depth' filtration and 'surface' or 'barrier' filtration (Figure 7.6). In depth
 Filtration 275

(a)

Clean gas
(b) /\~:~....
Nozzle for
reverse pulse
cleaning

Dirty
gas

Figure 7.5 (a) Example of a fibrous filtration medium: 'Cerafil' alumino-silicate

fibres. (Reproduced by kind permission of Cerel Ltd.) (b) Typical 'bag-house' type
filter (Institution of Chemical Engineers, 1985).
276 Gas/solid separation

Figure 7.6 Cross-section of a dust cake built up on a fibrous filter (surface filtration).
The scanning electron microscope image is taken from a resin block, in which the holes
represent where the particles originally were. The larger shapes at the bottom of the
picture are the filter fibres: (Reproduced by permission of Dr.-Ing. E. Schmidt,
University of Karlsruhe.)

filtration, collection of particles from the gas occurs throughout the filter
medium. In surface filtration, the medium acts as a barrier to the solids so
that a dust 'cake' is built up on the upstream surface, with no penetration into
the medium itself. In practice, filtration behaviour depends on the properties
of both the dust and the medium: not only the relative size of the dust
particles and the pores in the medium but also the surface properties (such as
adhesion) of both. In general, gas cleaning using membranes and the finer
grades of fibrous and granular filtration media approximates to surface
filtration. Perfect surface filtration is rare, and with a virgin 'surface filter' it is
usual for there to be a short period of penetration into the surface layers of
the medium before cake formation begins. During this short period, the
filtration efficiency may be slightly lower than in steady operation.
A further important practical difference between depth filtration and
barrier filtration is that in general it is not possible to remove particles
effectively from a depth filter after they have been captured, certainly not
'on-line' (i.e. without shutting down the filter and removing the filtration
medium). Barrier filters, however, are usually cleaned in situ, sometimes by
shaking or other mechanical action but more commonly by administering a
 Filtration 277

short pulse of pressurised gas in the opposite direction to that of the filtration
flow. This detaches the cake in relatively large pieces, which drop into a
collection vessel for disposal or recycle.
Since the considerations which govern filter design for depth and barrier
filtration are rather different, they will be dealt with separately in the sections
which follow.

7.3.2 Depth filtration

It is helpful to distinguish between three primary processes occurring in depth
filtration:
(i) collection of aerosol particles by the collector particles *;
(ii) retention of captured aerosol particles by the collector particles;
(iii) the effect of collected aerosol on the filter structure and subsequent
filtration efficiency.
In stationary filtration, collected particles either rebound instantaneously or
are permanently retained, and collection occurs on the surface of the collector
particles and not on the new structure formed by the collected aerosol.
Filtration efficiency therefore remains constant with time. The initial stages of
depth filtration can be considered stationary. If the filter performance
changes with time, it is said to be non-stationary. The extreme case of non-
stationary behaviour is barrier filtration, in which a dust cake forms on the
upstream face of the filter, so that the filter penetration (see equation (7.42)
below) decreases and the pressure drop increases approximately exponentially
with time.

7. 3. 2. J Collection. Consider a simple spherical collector particle of diameter

dc, in a gas stream of superficial gas velocity U (Figure 7.7(b)). The 'single
(collector) particle collection efficiency', E, is defined as:
number of dust particles collected
E=--------------------------------- (7.39)
number of dust particles in approach volume
E is a function of dust and collector size and electrostatic properties, dust
density, gas properties and bed voidage, £.
Now the volume of gas swept out by the collector in unit time is

*'Collector particles' here includes fibres. The derivations of the equations which follow are
slightly different for fibres, as are the correlations for the single-particle collection mechanisms
given in Table 7.1, but since the general approach is very similar, fibrous collectors are not
considered separately. For a comprehensive review of this subject, the reader is referred to
Brown (1993).
278 Gas/solid separation

(a)

(b) TRAJECTORY OF
PARTICLE WITH

APPROACH VOLUME GAS STREAMLI NES

Figure 7.7 Particle collection (a) on a single fibre (scanning electron microscope);
(b) on a single collector particle (schematic).

(some authors use U/E in place of U, which leads to numerically different

values for E). If the number concentration of particles in the gas approaching
the collector is C, then the number of aerosol particles collected per second is
EUC(Jtd~/4).
Now consider an assembly of such collector particles, making up a filter
 Filtration 279

bed. Unit volume of bed contains 6(1 - £)/Jtd~ collector particles. The
collection rate per unit bed volume, R y , is then given by
(7.40)

3EU(1 - £)
Ky = ----- (7.41)
2de
is the collection rate constant per unit bed volume. Equation (7.40) shows
that stationary filtration is a first-order process, i.e. the rate of filtration is
proportional to the dust concentration. For non-spherical collector particles,
particle shape will modify this result, but the effect will be slight provided that
the collector particles are roughly equi-axed, i.e. no one collector dimension
is very much greater than the others. If the collector particles cover a range of
sizes, the correct value for de is the surface-volume mean (also known as the
Sauter mean; see section 1.1.4).
We now consider the filter as a whole, and define a filter penetration, f
aerosol concentration in exit gas
(7.42)
f = aerosol concentration in entering gas
The overall filter efficiency, 'Y], is then given by
'Y]=l-f (7.43)
In order to obtain penetration in terms of E, we consider the case of deep bed
filtration in the stationary regime, for which E is constant throughout the
filter. We also consider a monodisperse aerosol or one size 'cut' in a
polydisperse dust.
Consider a fixed bed of height H and cross-sectional area A (Figure 7.8).
Let the challenging aerosol concentration be Co and the penetrating aerosol
concentration be C 1 . The rate of collection in a thin slice, of thickness dh, is
given by
3EU(1 - £)
RyAdh= CAdh (7.44)
2de
Now, aerosol entering per unit time = CUA and aerosol leaving per unit time
= (C + dC)UA so that
dC -3E(1 - £)C
------=--- (7.45)
dh U
Integrating over height H with C = Co at 'Y] = 0, and C = C 1 at 'Y] = H
C1
f= C = exp[-Ky H/U] (7.46)
o
280 Gas/solid separation

D cross-
section,
A

H
h

D
Figure 7.8 Filtration in a fixed bed (schematic).

It is often the case that additional collection occurs at the entrance to and exit
from the filter, for example on retaining screens, so that equation (7.46) is
modified as
/=f exp[-KvHIU] (7.47)
where f accounts for entrance and exit effects.
It remains to calculate E. The four purely mechanical processes which can
cause collection are shown in Figure 7.9 and reliable correlations for these are
listed in Table 7.1. When several processes occur together, it is common
practice simply to sum the single-mechanism values for E. This is valid if all
the individual efficiencies are much less than unity or if one collection
mechanism dominates.
Figure 7.10 shows individual and overall collection efficiencies for typical
values of bed and aerosol properties at ambient temperature plotted as
functions of aerosol size. For any filter operating conditions, there is a 'most
penetrating particle size' at which the collection efficiency is a minimum. For
 Filtration 287

(a)
, I
I (b)

,,, ,,
I

,,
, , PARTICLE
TRAJECTORY ,
\
\ \
\ \
\ \
\ \
\ \
\
,
\
,
, ,
, ;-
I
t ;
I

(d)
,
I

, I

\
\
\ , ,
\ , \
I
I ,
I

t
I

:. t ;
Figure 7.9 Mechanical mechanisms of collection: (a) diffusion; (b) inertia; (c) direct
interception; (d) gravity.

aerosol particles below this size, diffusional collection dominates, while, for
large aerosols, inertial impaction usually dominates. The effects of temperature
and pressure follow from the effects on gas properties and gas-particle
interaction discussed in sections 2.1.4 and 7.1. Increase in gas pressure has
relatively little effect on the combined efficiency, but increase in temperature
enhances collection of sub micron aerosols (because it increases Brownian
diffusivity) and reduces the collection of larger aerosols (because of the
increase in gas viscosity). The combined effect is to increase the most
penetrating particle size at higher temperatures as shown in Figure 7.11.
In addition to these purely mechanical processes, electrophoretic collection
282 Gas/solid separation

Table 7.1 Single-particle collection efficiencies

Mechanism Collection efficiency, E Source

(a) Diffusional collection 4.36 I DAB] % Wilson and

-1:- LUd e Geankoplis
(1966)
where U = superficial gas velocity
through filter, £ = void fraction
in filter, DAB = Brownian diffusivity
(m 2 S-l), de = collector particle
diameter
(b) Inertial deposition St3 .55 Coury et at. (1987)
1.1 X +
10-4 C 3 .55 St3 .55
where St = Qp d~ UC/9I,ld e ;
d a = aerosol particle diameter

(c) Direct interception IdaJ2 Paretsky (1972)

6.36c2 .4 Lde
(d) Gravitational settling
Upflow 0.0375 (VJU)°·5 Paretsky (1972)
Downflow 0.0375 (V/U)°5 + 0.21(V/U)°·78
where V t is the terminal velocity of
the aerosol particle

Note: numerous other correlations exist; those used here have been found reliable for granular
bed filters at ambient and high temperatures in air.

may be significant, as discussed in section 3.1.3. Coulombic and induced-

dipole mechanisms have both been used to enhance filter efficiency.
Coulombic attraction is costly and difficult to achieve because it requires a
potential to be maintained on the collector. By contrast, the charge-induced
dipole mechanism requires the dust to be charged but the collector to be
'earthed' - which would normally be the case anyway. Corona precharging
may be used, in which case the filter starts to have features in common with
an electrostatic precipitator with 'grounded' collector plates. In some
applications, the charge-induced dipole mechanism may occur unintention-
ally, if the aerosol to be collected is 'naturally' charged. Whether this is so in
many industrially operational filters remains unknown, because of the
difficulty in measuring charge distributions of aerosols (Coury et al., 1991).
In addition to these mechanisms, thermophoresis (i.e. the migration of
particles from high to low temperature regions) can affect deposition if the
collector medium is at a lower temperature than the gas. If the filter
temperature (in Kelvin) is less than about 90% of the gas temperature, then
 Filtration 283

de = 500J.,lm
Pa= 2650 kg/m3
air
>-
w 15°(
:z 10-1
l1.J
w
1 bar a.
u:::
u..
U = 0·05 m/s
l1.J

:z
0
;::: TOTAL E
W
l1.J
--.J
--.J
0
W

l1.J
--.J
W
;:::
ex:
ct.
l1.J
-'
\!J
:z 10-3
Vl

104~~~--wu~--~--~---,
0·1 1 10
AfROSOL DIA. (J.,Im)
Figure 7.10 Typical single-particle collection efficiencies.

thermophore sis increases the collection rate by two or more orders of

magnitude for particles in the diffusional collection range (Gokoglu and
Rosner, 1984). The resulting collection rate differs from the other mechanisms
in being almost independent of particle size.

7.3.2.2 Retention. In order to remove an aerosol particle from a gas

stream it is insufficient for it merely to be collected (by the mechanisms listed
above); it must also be retained, i.e. it must not rebound or be subsequently
dislodged (see section 3.4). There is ample experimental evidence (see Figure
7.12 for example) that poor retention limits the filtration efficiency of
284 Gas/solid separation

>-
u
z U= 0·08 rnls
LU
dc= 461 ,.urn
u
u.... E = 0·41
u....
LU
Pa= 1800 kg/rn 3
z:
0
~
nitrogen
u
LU
-I
1 bar a.
-I
0
10-1
u
LU
TEMPERATURE:
-I
U
~
0::
ct I
LU
-I
l!J
z 10-2
VI
-I

~
0
~

10
AEROSOL DIA. (,.urn)

Figure 7.11 Effect of temperature on overall single-particle collection efficiency

(Ghadiri,1980).

granular filters for larger particles and at higher gas velocities, i.e. in the
region in which inertial collection is most effective. For a particle which
adheres on impact, the adhesion force is normally sufficiently strong for fluid
drag alone not to cause re-entrainment (Stenhouse and Freshwater, 1976), so
that the problem is generally one of instantaneous rebound. However, in
some types of granular filters, subsequent attrition of collected particles may
also occur.
A theory of rebound has been described in section 3.4. Even though the
analysis remains to be developed further, some useful qualitative conclusions
can be drawn. Most obviously, any effects which dissipate energy, such as
 Filtration 285

GRADE
EFFICIENCY
1·0 ,-
/
I
/
I
I
I •
0·5 ••
t
I

I
0 0
0

0
0 Z 3 4 5 6 7 8 9
EQUIVALENT - VOLUME
DIAMETER (,.urn)

Figure 7.12 Grade efficiency for 922 (.tm silica sand filtering gasifier fines at 800°C
and superficial gas velocity 0.4 m S-1 (Ghadiri et at., 1993), d s ~ 0.89 d v . • , Sand
coated with a sticky 'retention aid'; D, clean sand.

plastic deformation or the presence of surface liquid, favour adhesion on

impact. This effect may be deliberately induced by covering the collector with
a sticky 'retention aid' (Nienow and Killick, 1987; Ghadiri et aI., 1993; see
Figure 7.12). Furthermore although collection efficiency due to inertial
impaction increases with increase in face velocity (for aerosol particles above
the most penetrating size), the probability of rebound also increases, and this
may eventually dominate.

7.3.3 Barrier filtration

7.3.3. 7 Cake formation and filter 'conditioning'. It is obvious that if the

pores available for gas flow through the filter are smaller than the aerosol
particle size, then the aerosol will be collected on the surface of the filter; this
is effectively 'sieving'. However, much coarser materials, such as woven
fabrics, are effective barrier filters because particles are able to bridge the
surface pores during the initial filtration period, and subsequent particle
capture occurs on these already deposited structures. A dust 'cake' then
builds up.
286 Gas/solid separation

Cohesive
and/or
low energy
particles

Uncohesive
and/or
low energy
particles

Figure 7.13 Filter cake structure. The larger circles indicate the filter medium
(schematic) (after Houi and Lenormand, 1986).

Once a cake forms, there is, as yet, no satisfactory method for predicting its
voidage and hence the pressure drop, which will increase rapidly as further
dust is collected. In general, retention of an aerosol particle at the point of
first contact gives a cake of higher void fraction (Figure 7.13) and therefore
lower resistance, because particles rest where they first contact the particles
deposited previously. Thus a high-voidage dendritic cake generally forms
when the dust is fine or cohesive or deposited at low face velocity, while dense
high-resistance cakes are typical of dense or uncohesive particles and high
approach velocities. Quantitative predictions are not yet possible so that
experiment is invariably required.
 Filtration 287

"p

/
/ Baseline
/
/
/
qj/
/

/
/
/

'---------- - - - - - -----..:..:.:.:-=-.~
No. of Cycles

Figure 7.14 Filter conditioning behaviour (schematic). See text (Seville, 1993).

In general, barrier filters are operated cyclically. During filtration, dust

builds up on the filter. After a prescribed time, or when the resistance to flow
reaches a prescribed level, the medium is cleaned. The usual cleaning action
is a reverse pulse of gas, applied to the clean side of the filter while it is on
line. This detaches the cake of deposited particles, which then falls into a
collecting hopper at the base of the unit, and the cycle is restarted.
Figure 7.14 shows schematically the behaviour of two hypothetical media
over many cycles of filtration and cleaning, assuming that the gas flow rate
remains constant during filtration. Medium 2 shows satisfactory performance:
it 'conditions' over relatively few cycles to give a stable 'baseline' resistance,
i.e. the resistance immediately following a cleaning pulse. Medium 1 shows
unsatisfactory behaviour: the baseline resistance continues to rise over many
cycles of filtration and cleaning and may fail to reach a steady equilibrium
value. Because filters must operate over many cycles, it is the 'conditioned'
resistance rather than that of the virgin element which is of most concern.
There are several possible reasons for the unsatisfactory behaviour of
medium 1. The filtration velocity may be high enough to cause penetration of
particles into the medium, which cannot be subsequently removed by the
cleaning pulse; alternatively, the cleaning system may not be properly
designed or operated to achieve the desired cleaning effect, or the adhesion of
the cake may be high enough to prevent its removal. These factors are
considered further below.
The conditioning process itself is poorly understood. However, it involves
288 Gas/solid separation

the establishment of a thin 'residual layer' of dust on the surface of the

medium and there is evidence that this may have a smaller mean particle size
than the challenging dust. This layer has an anomalously high flow resistance
compared with the remainder of the cake and is not removed at the end of the
cleaning cycle. Once it has formed, it dominates the subsequent filtration and
cleaning behaviour.

7.3.3.2 Resistance to flow. In general, the pressure drop through a planar

porous medium can be represented (see section 2.3.2) as

(7.48)

where (-dP/dz) is the pressure gradient in the direction of flow and U is the
superficial fluid velocity, i.e. the actual volumetric flow rate divided by the
area available for flow. In the case of the media considered here, the
Reynolds number (Ugdp/[!) is much less than unity, so that the second term
in equation (7.48) can be neglected and k j can be replaced by the Carman-
Kozeny expression (equation 2.69) provided that the void fraction E is not
too high (Kyan et aI., 1970),

(7.49)

where E is the void fraction, So the specific surface area of the medium and
kKis the Kozeny parameter, which depends on the geometrical structure. For
example, Seville et al. (1989b) investigated the resistance to flow of samples
of ceramic filter media prepared from fibrous and granular constituents mixed in
various proportions. Values of the Darcy's law resistance, k j , varied from 4 X
1010 to 5 X 10 11 m~2 but the value of kK remained approximately constant at
6.1 with a standard deviation of about 0.7, suggesting that equation (7.49) can
be used to 'design' a medium with the desired resistance characteristics. In
cases where E > 0.95, such as in many fibrous media, the prediction of
pressure drop is much more complex; reviews are presented by Strauss (1975)
and Brown (1993).
If Darcy's law applies, the flow resistance of an arbitrarily shaped filter
element can be predicted by solution of Laplace's equation (1). 2p = 0) with
appropriate boundary conditions, provided that the pressure drop across the
filter is small compared with the system pressure so that the gas density can be
regarded as constant.
As noted earlier, it is the conditioned rather than the virgin flow resistance
which is of most importance in practical applications. This depends on the
residual dust layer and is a great deal more difficult to predict. In principle,
the flow resistance attributable to the filter cake could be predicted by
application of equation (7.49) but in practice the void fraction is extremely
 Filtration 289

system-specific so that cake resistances must again be obtained by experiment.

Frequently, the 'cake resistance' (/:t,.P/flU) will be simply proportional to the
'areal' dust loading (the dust mass loading per unit area of filter surface) but
this cannot be assumed since filter cakes are frequently compressible
(Schmidt, 1993; H6flinger et al., 1993).

7.3.3.3 Mechanisms of filter cleaning. Effective cleaning is essential if the

conditioning behaviour of the filter is to be acceptable; it is clearly important
to be able to assess the magnitude of the required cleaning action. In some
circumstances, over-cleaning may be as damaging as under-cleaning since it
may lead to resuspension of the dust, which will not then separate into the
collection hopper.
Several analyses of the problem of cake detachment from flexible fabric
filters have been presented (see, for example, Leith and Allen, 1986; Sievert,
1988; Koch et al., 1993) but the common assumption in all of them is simply
that the dust cake detaches from the filter medium when it experiences a
tensile stress sufficient to overcome either the strength of the adhesive bond
between the cake and the medium (or a residual dust layer) or the internal
cohesion of the cake. Ideally, as soon as the strength of this adhesive or
cohesive bond is exceeded (by whatever cleaning mechanism) the cake
detaches everywhere simultaneously. In practice, however, neither the
adhesive/cohesive cake strength nor the applied stress is entirely uniform
across the filter surface so that 'patchy' cleaning results, as shown in Figure
7.15, i.e. cake is completely detached from some areas of the filter and
completely retained in others. This sort of behaviour is to be distinguished
from progressive removal of dust layers, which is not observed.
In a conventional fabric bag filter (Leith and Allen, 1986) it is usually
assumed that the required tensile cleaning stress is set up primarily by the
movement caused by the cleaning pulse or, in the case of mechanically-
cleaned filters, the shaking action. Pulse cleaning displaces the fabric
outwards. When it becomes taut, it decelerates sharply, normally at many
times gravitational acceleration. The cake then experiences a tensile stress
which depends on its areal density and on the deceleration. If the stress is
sufficient, fracture occurs so that the cake is thrown clear of the medium.
Rigid media such as ceramics (Seville, 1993) show no displacement on
cleaning. The tensile stress is therefore entirely the result of the pressure drop
imposed across the cake due to reverse flow of cleaning gas, as shown below.
Consider first the case of a filter medium on which a uniform cake has been
laid down. A cleaning flow is now set up in the opposite direction to the
filtration direction, as shown in Figure 7.16. During reverse-flow cleaning, a
pressure difference will be set up across the filter, consisting of contributions
from the cake and the medium,

(7.50)
290 Gas/solid separation

Figure 7.15 Patchy cake detachment from rigid ceramic filters (Seville et aI., 1991).
 Filtration 291

gas flow
-
medium cake

-1-

position

Figure 7.16 Pressure distribution in medium and cake during reverse flow (Seville et
at., 1989a).

The gas viscosity is effectively constant and the cake and medium thicknesses
can be incorporated into modified resistances, ke and km, so that
(7.51)
and
APe = keU (7.52)
Combining equations (7.50), (7.51) and (7.52)

APe=APT[~J
ke + k m
(7.53)

This is the pressure drop across the cake itself and also, as shown in Figure
7.16, the tensile stress acting at the cake/medium interface. It is therefore this
quantity which is of prime interest when investigating the cake removal
characteristics of a given dust/medium combination.
Equation (7.53) was developed for a uniform cake. The analysis above is
equally qpplicable to a partially cleaned filter (Seville et al., 1989a). Because
of its inhomogeneous resistance to flow, a patchily cleaned filter will show
regions of preferential gas flow (Figure 7.17). However, in the uncleaned
292 Gas/solid separation

gas flow
..
cake

medium

T T

Figure 7.17 Patchily cleaned filter during reverse flow (Seville et at., 1989a).

areas the total pressure drop across the filter must still be distributed across
the medium plus cake as shown in equation (7.53), provided that the flow is
rectilinear, i.e. in the uncleaned areas the gas velocity is the same in the
medium and in the cake. This approximation is valid provided that the
undetached cake patches are large compared with the cake thickness, which is
often but not always the case.
From the analysis above it is clear that when comparing the cleaning
behaviour of different dust/medium combinations it is I!!.Po the pressure drop
across the cake alone, which should be considered and not I!!.PT , the total
pressure drop. Indeed, comparison of values for I!!.PT necessary to detach the
cake may be misleading, because they depend on the cake loading, whereas
I!!.Pc does not. It is sometimes asserted that thick cakes are easier to clean
from filter media than thin ones. The foregoing analysis shows why this
appears to be so. For the total pressure drop across the cake to remain at its
critical value for detachment, a thicker cake requires a smaller pressure
gradient for removal and therefore less cleaning gas flow and less pressure
drop across the medium itself.
The approach outlined above considers only steady reverse-flow cleaning,
but it also applies to pulse cleaning, since the maximum stress to which the
cake is subjected corresponds to the steady flow cleaning value from equation
 Filtration 293

100 -l
I
J
% cake
J
remaining
I 'ideal' cake
_ . - - - detachment

-I
50 I.
J

-
I -
I
I
I
I -
Stress applied to cake (Pa)

Figure 7.18 Measurement of cake detachment stress - coupon test.

(7.53). In many industrial applications, the pressure rise associated with the
cleaning pulse is, in any case, so slow that the process can be considered in the
same way as steady reverse-flow cleaning.
Because such factors as the cohesion of the collected dust and the adhesion
of the dust to the medium cannot be predicted a priori, it is essential to carry
out experimental work to investigate the filter cleaning behaviour. Because of
spatial variations in both the filter medium and the cake, the critical stress at
which the cake detaches is not uniform across the surface of a filter so that
detachment occurs progressively as the cleaning pressure is increased.
It is possible to determine the range of tensile stresses over which the cake
detaches from the medium by progressively increasing llPT , and measuring
the fractional cake removal at each value of llPT by collecting and weighing
the dust removed from the filter. This is most conveniently carried out on a
small flat 'coupon' of filter medium, as described by Koch et al. (1993) or as
specified in the German VDI standard 3926 (1994). Not only are the
quantities of dust required for conditioning more manageable but since the
coupon is small it is comparatively easy to ensure that the imposed cleaning
stress is uniform across it; this is not usually the case for a candle or bag
(Berbner and Loffler, 1993). Results from the coupon test can be plotted in
the form of 'percentage cake remaining' versus 'applied stress' (Figure 7.18),
where the applied stress is the appropriate value of llPc for each point,
calculated from equation (7.53). (Values of kc and k m are obtained from
294 Gas/solid separation

pressure drop measurements before the test and after complete cake
removal.) This 'cake detachment curve' provides the information needed for
rational selection of cleaning pressure. For fabrics, the cake detachment
stress has been determined in a similar way but by using 'jerk tests', in which
progressively larger accelerations are applied and the proportion of cake
detached is measured at each acceleration level (Leith and Allen, 1986;
Sievert, 1988).

7.3.4 Concluding remarks

Prediction of the performance of filters draws equally on both fluid mechanics
and particle mechanics. In depth filtration, the problem posed is usually the
prediction of the grade efficiency as a function of the operating and design
variables. Experimental and theoretical studies of particle collection on fibres
and collector\particles have resulted in a reliable set of semi-empirical
expressions for single-particle collection efficiency. The prediction of reten-
tion efficiencies in practice is less tractable and makes overall efficiency
prediction more problematic.
In barrier filtration, the particle collection and retention efficiencies are of
less importance, since ideally all particles remain on the surface of the
medium. (Both collection and retention are clearly of importance in
determining whether a given filter with a given dust behaves as a depth or a
barrier filter, however.) For barrier filters, the important issue is the way in
which the dust cake forms, and its resulting cohesive and adhesive strength.
The cake structure, which can vary enormously depending on the properties
of the dust, even when the particle size remains unchanged, determines both
the subsequent resistance to flow and the ease with which the cake can be
removed at the end of each filtration cycle. At the present stage of theoretical
development, these can only be estimated through careful experiment.

7.4 NOMENCLATURE
A Filter cross-sectional area (m 2 )
a, b Dimensions of gas inlet to cyclone (m)
C Cunningham slip-correction factor; Inlet particle
loading (kg m- 3 )
D Cyclone barrel diameter (m)
DAB Brownian diffusivity of particle in a gas (m 2 S-l)
da Aerosol particle diameter (m)
de Collector particle diameter (m)
De Diameter of vortex finder (m)
d Particle diameter (m)
d 50 'Cut size' (m)
 References 295

E Single (collector) particle collection efficiency (equation (7.39))

f Penetration (equation (7.42))
H Height of filter bed (m)
k Dimensionless coefficient in equation of Shepherd and Lapple
Kv Collection rate constant per unit bed volume (S-l) (equation
(7.14))
kK Kozeny parameter
N Number of cyclones in parallel
p Gas pressure (Pa)
Q Volumetric gas flow rate (m 3 S-l)
q Gas flow rate through one cyclone in a battery in parallel
(m 3 S-l)
Cyclone Reynolds number (equation (7.14))
Collection rate per unit bed volume (kg m-3 S-l)
Cyclone Stokes number (equation (7.20))
Temperature (K)
'Superficial velocity' in cyclone, Q/(rrD2/4) (m S-l)
Superficial gas velocity, 'face velocity' (m S-l)
Particle velocity (m S-l)
Particle terminal settling velocity (m S-l)

Suffices
c Particle-laden flow
o Gas flow alone

Greek letters
Collection efficiency
Gas viscosity (N s m-2 )
Pressure coefficient (equation (7.13))
Gas density (kg m-3 )
Particle density (kg m-3 )
Void fraction

REFERENCES
Abrahamson, J. and Allen, R.W.K. (1986) First International Symposium on Gas
Cleaning at High Temperatures, Inst. Chem. Engrs. Symp. Ser. no. 99, 31.
Berbner, S. and Laffler, F. (1993) in Gas Cleaning at High Temperatures (R. Clift and
J.P.K. Seville, eds) Blackie, Glasgow, p. 225.
Boericke, R.R., Giles, W.G., Dietz, P.W., Kallio, G. and Kuo, J.J. (1981) US DoE
Meeting on 'High Temperature, High Pressure Particulate and Alkali Control in
Coal Combustion Process Streams', Morgantown, February 3-5, CONF-810249.
296 Gas/solid separation

Boyson, F., Ayers, W.H. and Swithenbank, J. (1982) Trans. Inst. Chem. Engrs 60,
222.
Brown, R. (1993) Air Filtration, Pergamon, Oxford.
Clift, R., Ghadiri, M. and Hoffmann, A.C. (1991) A.I. Chem. Eng. J. 37,285.
Coury, J.R., Thambimuthu, K.V. and Clift, R. (1987) Powder Techno!. 50,257.
Coury, J.R., Raper, J.A., Guang, D. and Clift, R. (1991) Chem. Eng. Res. Des.
69(B), 97.
Dietz, P.W. (1981) A. I. Chem. Eng. 1. 27,888.
Ghadiri, M. (1980) PhD Dissertation, University of Cambridge.
Ghadiri, M., Seville, J.P.K. and Clift, R. (1993) Chem. Eng. Res. Des. 71(A), 371.
Giles, W.B. (1982) 4th Symposium on the Transfer and Utilization of Particulate
Control Technology, Houston, October 11-15.
Gokoglu, S.A. and Rosner, D.E. (1984) Int. 1. Heat Mass Trans. 27,639.
Hoffmann, A.C., van Santen, A., Allen, R.W.K. and Clift, R. (1992) Powder
Techno!. 70, 83.
Hoflinger, W., Stocklmayer, Ch. and Hackl, A. (1993) Proceedings 'Filtech' 1993,
Karlsruhe, p. 563.
Houi, D. and Lenormand, R. (1986) Filtrat. Separ. (Proc. Filt. Soc.) (July/Aug. 1986),
238.
Institution of Chemical Engineers (1985) User Guide to Dust and Fine Control, 2nd
ed.,L Chern. E., Rugby.
Koch, D., Schulz, K., Seville, J.P.K. and Clift, R. (1993) in Gas Cleaning at High
Temperatures (R. Clift and J.P.K. Seville, eds.), Blackie, Glasgow, p. 244.
Kyan, C.P., Wasan, D.T. and Kinter, R.c. (1970) Ind. Eng. Chem. Fundam. 9,596.
Leith, D. and Allen, R.W.K. (1986) in Progress in Filtration and Separation (R.J.
Wakeman, ed.) Elsevier, Amsterdam, p. 1.
Leith, D. and Licht, W. (1972) A. I. Chem. Eng. Symp. Ser. 68, 196.
Linhardt, H.D. (1981) US DoE Meeting on 'High Temperature, High Pressure
Particulate and Alkali Control in Coal Combustion Process Streams', Morgantown,
February 3-5, CONF-810249.
Mothes, H. and Loffler, F. (1984) Chem. Eng. Process 18, 323.
Muschelknautz, E. (1980) VDI Berichte no. 363,49.
Nienow, A.W. and Killick, R.c. (1987) Powder Technol. 50,267.
Paretsky, L.C. (1972) PhD Thesis, The City University of New York.
Reid, R.C., Prausnitz, J.M. and Sherwood, T.K. (1977) The Properties of Gases and
Liquids, McGraw-Hill, New York.
Schmidt, E. (1993) Proceedings 'Filtech' 1993, Karlsruhe, p. 553.
Seville, J.P.K., Cheung, W. and Clift, R. (1989a) Filtrat. Separ. (May/June 1989),187.
Seville, J.P.K., Clift, R., Withers, C.J. and Keidel, W. (1989b) Filtrat. Separ. (July/
Aug. 1989), 265.
Seville, J.P.K., Legros, R., Brereton, C.M.H., Lim, C.J. and Grace, J.R. (1991)
Proceedings of the 11th International Conference on Fluidised Bed Combustion,
ASME, New York.
Seville, J.P.K. (1993) KaNA Powder and Particle (11), 41.
Shepherd, C.B. and Lapple, C.E. (1940) Ind. Eng. Chem. 32, 1246.
Sievert, J. (1988) Dissertation Universitat Karlsruhe Fortschr. Ber. VDI Reihe 3, Nr.
161.
Stairmand, C.J. (1951) Trans. 1nst. Chem. Engrs 29,356.
Stairmand, c.J. and Kelsey, R.N. (1955) Chem. Industry 1324.
Stenhouse, J.LT. and Freshwater, D.C. (1976) Trans. Inst. Chem. Engrs 54,95.
Strauss, W. (1975) Industrial Gas Cleaning, 2nd edn., Pergamon, Oxford.
Svarovsky, L. (1981) Solid Gas Separation, Elsevier, Amsterdam.
 References 297

Svarovsky, L. (1986) in Gas Fluidization Technology, Ch 8, Wiley, Chichester.

Syred, N., Biffin, N., Dolbear, S., Wright, M. and Sage, P. (1986) First International
Symposium on Gas Cleaning at High Temperatures, Inst. Chem. Engrs Symp. Ser.
no. 99,17.
VDI (Verein Deutscher Ingenieure) Standard 3926 (1994) Testing of Filter Media for
Cleanable Filters.
Wilson, E.J. and Geankoplis, C.J. (1966) Ind. Eng. Chem. Fund. 5,9.
8
Storage and discharge of
particulate bulk solids

8.1 INTRODUCTION
Particulate materials are stored in vessels which are referred to by different
names depending on their geometry; Figure 8.1 shows illustrations of some of
the most commonly used. Those containers with a flat bottom are commonly
known as 'bins' or 'bunkers'; if the vessel side walls are inclined then the term
'hopper' is used. Bins and hoppers can have either axially-symmetric or
rectangular cross-section as illustrated in Figure 8.1. With rectangular
containers, if the vessel thickness, I, is much less than the vessel half-width, a
(usually 1 < aI2), then the container may be referred to as 'planar' or 'plane'
type. The implication is that when the vessel thickness is quite small, then the
geometry is essentially two-dimensional and hence can be approximated by a
single plane.
The term 'silo' is also used especially in European literature to describe
bulk solids storage vessels irrespective of their geometry. Here, we will use
the word 'silo' to describe a storage vessel which comprises a bin section
placed on top of a hopper, a configuration often used for industrial containers
and illustrated in Figure 8.1 (d). Bulk solids storage vessels can be used for a
variety of purposes in different industrial sectors; a classification is presented
in Figure 8.2.
The vessels used in the 'primary' bulk solids handling sector, which includes
agricultural, mining, cement and refractory products, are essentially long-
term storage units for batches of material between 50 and 1000 tons. On the
other hand, when secondary processing or manufacture are involved, such as
in pharmaceuticals, chemicals, foodstuffs, plastics, dyes and pigment industries,
then bins and hoppers are more often used for intermediate storage of
materials in between other process units. Figure 8.3 shows two process flow
sheets where the storage container serves an additional 'functional' purpose
as well as holding a batch of particulate material for a given length of time. In
Figure 8.3(a), several hoppers are used to regulate the feeding of the
components of a bulk mixture which is continuously blended to form a paste.
In this case, the hoppers are used to feed different components of the mixture

J. Seville et al., Processing of Particulate Solids

© Chapman & Hall 1997
 Introduction 299

(a)
(b)

_2a-·- I -

(c)
(d)

Figure 8.1 Typical bulk solids storage vessel geometries used in industry. (a) Conical
or axisymmetric hopper; (b) plane-flow wedge hopper; (c) plane-flow chisel hopper;
(d) pyramid hopper.

at different rates; satisfactory discharge from these hoppers therefore has a

direct effect on the microstructural properties and the chemical composition
of the product. The second illustration, Figure 8.3(b), shows a storage unit
forming part of a recycle loop between a catalyst regenerator and a catalytic
reaction unit. In this case, both the regeneration rate of the catalyst and the
reaction rate in the reactor are controlled by the rate of material discharge
from the hopper.

8.1.1 General design considerations for 'process' silos

The silos used at the 'fine' end of the process industries generally tend to be
much smaller than those used by the primary bulk solids handling sector.
Process silos in many cases are less than 5 tons in capacity and therefore only
a few metres high and up to about 1 metre wide. Such vessels are usually built
from mild steel or some metallic alloy. Here, the structural stability of the
vessel is not affected by the stresses induced during storage and flow of the
particulate materials as these tend to be much smaller than the stresses
required to cause deformation of the vessel walls.
300 Storage and discharge of particulate bulk solids

No. of Annual Annual Annual Annual Annual

Industry establishments sales purchase industrial net output non-industrial
(£ million) (£million) (£ million) (£million) (£ million)

Major bulk
solids
producing
11310 30286.3 15658.6 521.2 14785.2 2514.7
Industries
Major
process
Industries 9007
using 29913.9 18978.9 669.9 9810.5 1990.0
granular
solids
Energy
industries
using bulk
454 21950.3 15935.0 494.1 6214.5 635.0
solids

Bulk solids
machinery, 4896.8 2256.5 209.3 2444.3 341.9
plant 2113
equipment

All UK
production 109595 193288.0 108955.4 5436.9 80680.0 12492.4
industries

Source: Business monitor, UK (1988)

% % %
100 100 100

75 75 75
"if.
"!
50 ;!!.
0
50 :;;:
0>
0
25 C\I 25

0 0
No. of establishments Net annual output Annual industrial costs

o All UK production industries [Z2] Energy industries

~ Bulk solids producing industries _ Service industries
!lEll Process industries using solids ~ Market share

Figure 8.2 Bulk solids handling sectors of UK industry (1988).

The process silos illustrated in Figure 8.3 are expected to operate at steady-
state flow conditions appropriate to the process in question. For reliable
operation, it is important to predict the steady-state flow fields set up by
different materials and their relation to the silo geometry as well as to the
physical properties of the particles. Furthermore, transient analyses may
become necessary in order to predict the dynamic changes in hopper flow
conditions as a function of certain step changes in various process variables
such as temperature, pressure, humidity and bulk composition.
 Introduction 307

(a) (b)

Figure 8.3 Process silos. (a) Feed forward mode: 1, feeder; 2, reactor/contactor; 3,
non-solid phase; 4, conveyor; 5, collector. (b) Recycle mode.

8.1.2 Design considerations for large storage silos

Generally, when storage capacities greater than 50 tons are required, then the
dimensions of the silo are simply too large for the construction of a metal
vessel to be cost-effective. In such cases, concrete structures are often used.
In the design of such concrete units, the wall stresses caused during filling and
discharge of the materials have to be taken into account during design as
these are often large enough to cause 'fatigue' and eventual failure of the
structure if they are underestimated. Figure 8.4 shows the photograph of a
silo collapse due to structural failure upon initiation of discharge. In this case,
the walls were designed to resist only the stresses induced by a static heap of
material. The silo collapse is due to the severalfold increase in the stress
acting on the walls following initiation of discharge. This sudden increase in
wall stress levels with the commencement of discharge is known as the switch
stress phenomenon and has been cited many times (see, for example,
Thiemer, 1969; lenike et ai., 1973; Walters, 1973b) as the cause of many
concrete and thin shell steel silo failures. Switch stress phenomena will be
investigated in section 8.5 as part of a more general discussion of stress
distributions in silos.
Large-scale concrete silos are often used for batch storage of materials over
a number of days and sometimes weeks. Practical experience (see, for
example, Arnold et al., 1982; Roberts, 1994) indicates that when particulate
materials are allowed to remain in storage containers over prolonged time
periods, then their yield strength Gc increases through a process known as
'time consolidation'. This effect becomes more significant as the particle size
302 Storage and discharge of particulate bulk solids

Figure 8.4 Silo collapse due to structural failure upon initiation of discharge.

is reduced and the cohesive forces between the particles are increased. The
reader is referred to Chapter 4 on the failure properties of particulate
materials for an illustration of consolidation effects with cohesive materials. If
the yield strength, 0c> of the material above the silo outlet reaches a critical
value during storage which is larger than the mean compressive stress, am,
acting in the bulk at the same point, then the material will not flow out when
the outlet is opened. Instead a 'stable arch' will form at the point where Oc >
Om and discharge will be prevented, as shown in Figure 8.5(a). The prediction
of the conditions which will give rise to stable arch formation is the basis for
the major design manuals used by industry and the equipment manufacturers;
see, for example, the Draft Code of Practice of the British Materials Handling
Board (1985). The design procedures outlined in such manuals are all variants
of a semi-empirical analysis due to Jenike (1961,1964,1967), Johanson (1965,
1968) and Jenike et ai. (1973) which is discussed in detail in Chapter 4. Table
8.1 compares the design aspects which are important to large-scale batch
storage silos and small-scale continuous process silos.

8.2 FLOW REGIMES IN BINS AND HOPPERS

8.2.1 Mass flow versus funnel flow
Two distinct kinds of flow behaviour are conventionally recognised in vessels
discharging particulate material. These are known as (i) mass-flow and
 Flow regimes in bins and hoppers 303

(a) (b)
Mass Flow Funnel flow

Figure 8.5 Mass flow VS. funnel flow.

Table 8.1 Factors relating to the design of bulk solids storage vessels

Parameter Storage type Process type

Capacity >50 tonnes <5 tonnes

Material of construction Concrete Metal
Installation cost High Low
Running cost Low High
Cost/unit solids discharge Low High
Wall stresses due to solids High Low
Mode of operation Quasi-static Dynamic
Flow problem Flow initiation Flow regulation

(ii) funnel or core-flow conditions. Figure 8.5(a) and 8.6(b) illustrate these
two types. In the so-called 'mass-flow' condition, particulate material is in
motion everywhere within the container during discharge. The flow region
boundary in this case coincides with the vessel walls and the entire cross-
sectional area of the vessel is made available for flow. As illustrated in Figure
8.6(a), such conditions are met in general when discharging free-flowing
materials from conical or wedge-shaped hoppers. To a first approximation,
the (included) hopper half-angle, u, measured clockwise from the vertical
has to be less than U e ,
304 Storage and discharge of particulate bulk solids

Jt
Uc =-- <Pw (8.1)
2
where <Pw is the angle of wall friction (see Chapter 4) so that particles next to
the wall are kept in motion. The limiting value of U c in equation (8.1) is given
by the so-called fully-rough wall condition, where <Pw = <p, i.e. the internal
and wall friction angles are identical. The case <Pw = <p becomes analogous
to the so-called 'core' or funnel flow illustrated in Figure 8.5(b) when the
material forms its own flow region boundary such as during discharge from a
flat-bottomed bin or a shallow hopper. In this case, we would expect a funnel
to form if
Jt
U
c
> -2 - <Pd (8.2)

where <Pd is the angle between the flow region boundary and the horizontal at
the bin outlet. If we can treat the flow region boundary as a fully-rough wall
then we would expect <Pd = <PR where <PR is the 'angle of repose' (see
Chapter 4) of the material. It is important to recall here that the angle of
repose will be equal to the internal friction angle <P when the material is
cohesionless, as discussed in Chapter 4. Hence with materials of negligible
cohesion, we would expect to see funnel flow inside a vessel if

(8.3)

Extensive experimental investigations of the flow region boundary in funnel-

flow bins by Tiiziin and Nedderman (1982), Takahashi and Yanai (1974) and
Nguyen et al. (1979), seem to show that if the vessel is of the planar type and
the flow is essentially two-dimensional then <Pd > <PR' With 'cohesionless'
materials, Tiiziin (1979) gives <Pd = 1.5 <PR. However, when axially-
symmetric containers are used, experimental evidence suggests <Pd = <PR;
see, for example, Takahashi and Yanai (1974). Equations (8.2) and (8.3) are
purely empirical and therefore cannot be relied upon to provide a very
accurate prediction of the flow regime that will occur in a hopper of a given
half-angle u. In general, the flow field inside the hopper will be a function of
the particle properties of the particulate material as well as of the wall surface
and the relative dimensions of the hopper and the discharge orifice. With so
many variables, it is not possible to establish a simple analytical 'mass-flow
criterion' for the discharge of bulk solids from silos. Instead, the national silo
design manuals in different countries (see, for example, German (DIN 1055),
Australian (1983) and British Draft Code of Practice (British Materials
Handling Board, 1985) use semi-empirical and often graphical procedures
to calculate the critical outlet size and the hopper half-angle as a function of
the internal and wall friction angles and the bin geometry. These procedures
are discussed in detail in Chapter 4.
 ~ ~
(b) (c)
(a)
(d) (e)

Figure 8.6 Mass flow, funnel flow and other typical flow regimes. (a) Mass-flow, (b) Semi-mass flow, (c) Funnel flow,
(d), (e) Expanded flow.
306 Storage and discharge of particulate bulk solids

8.2.2 Other possible flow regimes

In funnel or core flow, the material between the flow region boundary and the
vessel wall is referred to as the stagnant or dead zone, as shown in Figure
8.6(c). In funnel flow the stagnant regions on either side of the central flow
core will extend as far as the top surface of the material bed. This is typical of
the batch discharge of a free-flowing material from a shallow vessel where the
height to width ratio is less than 2. If however, an additional cylindrical or a
rectangular bin section is present above the discharge vessel, then the
stagnant zone boundary will extend only part of the way up the vessel. The
material in the top section will always be in mass flow regardless of the
geometry of the outlet. If stagnant regions exit only immediately above the
hopper outlet, then the material is said to be in 'semi-mass' flow condition as
shown in Figure 8.6(b).
Mass-flow hoppers are invariably of a steep half-angle (usually a < 30°)
which results in very tall vessels. To overcome this, manufacturers can
sometimes build a wide-angled hopper but reduce the half-angle only over a
small region immediately above the discharge orifice as seen in Figure 8.6(e).
In this case, the material in the steep wall section will be in 'mass' flow
whereas the stagnant zones can develop in the wide-angled top section. Such
a flow regime is known as 'expanded flow'. 'Expanded flow' requires a
transition of the material bed from funnel flow to the mass flow condition
before discharge from the silo. In contrast, 'semi-mass flow' requires a
transition from mass flow to funnel flow in the section above the outlet; the
two approaches are compared in Figures 8.6(c) and 8.6(d),(e).
A condition known as 'pipe flow' or 'channel flow' is sometimes observed
when discharging strongly cohesive materials from silos. In this case, the flow
is from a narrow section in the middle of the vessel. The flow region boundary
is almost vertical and extends as far as the top surface, as shown in Figure
8.6(c). This is obviously an extremely undesirable flow regime as it only uses
a very small portion of the available cross-sectional area of the vessel.
Furthermore, it is often impossible to remove the material in the 'dead zones'
without resorting to extreme measures such as inserting a mechanical agitator
or dismantling the vessel.

8.2.3 Choice of flow regime

Traditionally, bulk solids storage and discharge vessels are designed to ensure
mass-flow conditions. Current design procedures discussed in sections 8.5
and 8.6 are all concerned with designs to achieve mass flow. However,
experience suggests that other flow regimes occur much more often in
industrial practice and indeed in certain circumstances it may even be
advantageous not to have mass flow. Table 8.2 provides a summary of the
major advantages and disadvantages of the mass-flow and the funnel-flow
conditions.
 Flow regimes in bins and hoppers 307

It is worth noting that in industrial practice, the same vessel is often used to
discharge a number of different material batches with varying particle and
frictional properties. The changes in the physical properties of the materials
handled will often result in different flow regimes within the same container
which may have been in the first instance designed to provide mass flow
assuming a set of nominal values of material properties.
In Chapter 4, the frictional properties of powder beds and the experimental
techniques used for their determination were considered. The accuracy of
such measurements relies largely on the matching of particle-packing
conditions and the compressive load levels to those experienced in the silo.
Hence it is not difficult to over- or underestimate the 'effective' coefficients of
internal and wall friction used in the mass-flow design criterion.
In general, if the particulate material is not significantly cohesive then the
size of the stagnant zone in funnel flow will be quite small and hence the
volume of material trapped within this region will not be significant. This
disadvantage is easily compensated for by factors like the reduced wall
stresses, smaller vessel height to hold the same volume and reduced wall wear
and particle attrition. With cohesive materials, however, the funnel flow

Table 8.2 Comparison of mass flow and funnel flow of particulate materials

Mass flow Funnel flow

Characteristics
No stagnant zones Stagnant zone formation
Uses full cross-section of vessel Flow occurs within a portion of vessel
cross-section
First-in, first-out flow First-in, last-out flow

Advantages
Minimises segregation, agglomeration Small stresses on vessel walls during flow
of materials during discharge due to the 'buffer effect' of stagnant zones
Very low particle velocities close to vessel
walls; reduced particle attrition and wall
wear

Disadvantages
Large stresses on vessel walls during Promotes segregation and agglomeration
flow during flow
Attrition of particles and erosion and Discharge rate less predictable as flow
wear of vessel wall surface due to high region boundary can alter with time
particle velocities
Small storage volume to vessel height
ratio
308 Storage and discharge of particulate bulk solids

condition will promote overconsolidation of materials into rigid blocks within

the stagnant zones. This will in turn result in the narrowing of the central flow
zone and with time the flow will deteriorate to the condition known as 'pipe
flow' discussed in section 8.2.2 above. This is of course highly undesirable as
it will ultimately lead to the termination of the discharge due to the blockage
of the outlet by the overconsolidated material.

8.3 VELOCITY DISTRIBUTIONS IN BINS

AND HOPPERS
The prediction of the velocity profiles in different sections of a storage vessel
during steady discharge of a particulate material is essential to the
development of an analytical mass-flow criterion. Furthermore, a theoretical
prediction of the solids discharge rate is only possible if the velocity field
prevailing immediately above the outlet is known. In this section, we will
develop the theory of velocity distributions for the case of 'free-flowing'
materials. Discharge of fine powders and discharge of cohesive materials
must be treated as special cases. The reason is that the flow fields set up
during continuous discharge of material from a storage vessel can be analysed
using the steady-state assumption only when the material is free flowing
within the vessel. Fine powders and cohesive materials are 'difficult' materials
to discharge freely, and their flow fields can alter significantly with time.
Therefore, transient analyses become necessary in such cases.

8.3.1 Steady-state flow fields for 'free-flowing' materials

A particulate material is said to be 'free flowing' if a constant discharge rate
can be maintained indefinitely under a high material head (usually H > D;
see Figure 8.1). The flow field of such a particulate material can be divided
into three different regions within the storage vessel:
(i) plug-flow region;
(ii) converging flow region;
(iii) transition region.
Figure 8.7 provides a schematic illustration of the three flow regions in a flat-
bottomed container.

8.3.7.7 Plug flow. The material in the parallel-sided or cylindrical section

of a container moves in plug-flow fashion during discharge. Figure 8.8 shows a
number of velocity profiles measured in the plug-flow section of a container.
In plug flow, the horizontal component of the particle velocity is almost zero
and there is negligible variation of vertical velocity across the bin except for
the narrow shear zones adjacent to the walls. This 'shear zone' is the result of
 Velocity distribution in bins and hoppers 309

PLUG FLOW

Figure 8.7 Flow regions in a flat bottom silo.

12

• - Height No: 2
t::
<lJ o - Height No: 3
> x
VI
VI
•
<lJ
...J
c
.2
VI
c
<lJ
E
o

o 0.5 1.0
Dimensionless distance across the bin

Figure 8.8 Velocity profiles in plug flow, Laohakul (1978).

310 Storage and discharge of particulate bulk solids

the flow retardation of the particles in contact with the vessel wall surface. It
is analogous to the boundary layer set up next to a wall surface in laminar flow
of fluids.
Experimental investigations by Takahashi and Yanai (1973) in axially
symmetric bins, and by Nedderman and Laohakul (1981) in planar bins, have
however shown significant differences between the boundary layers in
laminar flow and the shear zones observed in plug flow of particulate solids.
In laminar flow of Newtonian fluids, the boundary layer thickness is a
function both of the distance from the leading edge, x, and of the fluid
velocity, Uf

(8.4)

where v is the fluid kinematic viscosity and 0 is the boundary layer thickness
(Kay and Nedderman, 1974). In particulate flow, the experimental evidence
suggests that the shear zone thickness is independent of the magnitude of the
plug-flow velocity and does not vary with height below the material top
surface. Instead, it is found to be a strong function of the particle size.
Davidson (1981) suggests a correlation of the form
v = vP - (v P - v W )f3 L -x/d (8.5)
where vp is the plug flow velocity and Vw is the particle velocity at the wall. f3L
is an empirical constant whose value is found to be between 1.5 and 2.0 (see
Laohakul, 1978). Equation (8.5) will predict the shear zone thickness to be of
the order of five to six particle diameters adjacent to the wall; see Figure 8.8
and Example 8.2.
Nedderman and Laohakul (1981) predict the particle velocities in the shear
regions using a hyperbolic tangent profile (see Figure 8.8). This is expressed
in the form
v = Vw + (vp - vw)tanh(x/o) (8.6)
where vp , and Vw are the plug-flow and wall velocities, respectively, and 0 is
the characteristic length closely related to the displacement thickness of the
conventional boundary layer theory of Newtonian fluids, as used in equation
(8.4). The particle velocity Vw at the wall is found to be a strong function of
the coefficient of wall friction. When the walls are smooth (i.e. [tw < 0.2)
then there is little retardation of the particles next to the wall surface and Vw
approaches vp ' the plug-flow velocity in the bulk. With fully-rough walls (i.e.
<Pw = <p), however, the particles adjacent to the wall are severely retarded
and the particle velocity at the wall, Vw, approaches zero, as shown in Figure
8.8.
The shear strain rate y across the shear region can also be related to the
velocity at the wall
 Velocity distribution in bins and hoppers 311

y = (vp - vw)/() (8.7)

where () is the shear zone thickness. The linearity of the shear strain rate is
the result of a constant shear zone thickness, (), and an almost constant ratio
vw1vp for particles in the plug-flow section of the bin. Equation (8.7) predicts
a maximum shear rate,
(8.8)
for the case of fully rough walls where the velocity at the wall Vw approaches
zero. Similarly, the shear strain rate y will approach zero with a very smooth
wall as Vw will be almost equal to vp'
As some dilation must inevitably accompany the shearing of a particulate
bed (refer to Chapter 4), these results therefore predict significant interstitial
voidage change across the shear zone boundary only when y is high, e.g. in
the case of the fully rough wall. With smooth-vessel walls, however, y
calculated using equation (8.6) is small and hence we should not expect a
significant voidage change in the particulate bed adjacent to the wall.
If a constant value of the interstitial voidage, £0, is assumed, then the plug-
flow velocity vp can be very simply related to the solids discharge rate, W from
the silo,
(8.9)
where A is the cross-sectional area of the plug-flow section appropriate to the
vessel geometry and Qs is the particle density. The term Qs(l - E) is in fact
the so-called 'bulk density' of the particulate bed during flow. As mentioned
earlier, some dilation of the material bed is inevitable during flow and this will
reduce the bulk density from its static value. Practical experience suggests
that free-flowing materials are essentially 'incompressible' and that there is
little change of the interstitial voidage during the flow of a loose material fill.
If, however, the material bed is compacted significantly during filling then the
dilation during flow will be more significant.
Equation (8.9) is also useful in estimating the residence time, t R , of
materials inside a rectangular or cylindrical silo as a function of the solids
discharge rate, W,
AHQs(l - E)
tR = Hlv = - - - - - - (8.10)
p W
where H is the height of the silo unit.

Example 8. 1 Calculation of storage time in batch discharge.

A stainless-steel silo is used to store and discharge agricultural seeds of solid
density Qs = 950 kg m-3 . The silo comprises a 5 m tall cylindrical top section
of 1 m internal diameter, above a conical section of a = 30°. In order to
ensure"a constant discharge rate of W = 5000 kg h-1 , it is essential that the
312 Storage and discharge of particulate bulk solids

head of material inside the silo should not fall below the top of the conical
section. If it is decided to fill the silo completely and then discharge its
contents in a batch manner, what would be the residence time of solids in the
cylindrical section of the silo?
Solution. Using equation (8.9)
tR = 5/vp
where vp can be calculated using equation (8.8). However, we need to guess
the interstitial voidage E. A good guess is E = 0.4 ± 0.05. Thus we get
5000 X 4
v = =3.10 mm S-1
p 3600 X Jt X 950 X 0.6
Hence tR = 5/3.10 X 10-3 = 1613 s = 0.45 h.
Comments. The magnitude of the residence time for the process in this
case is rather sensitive to the value of the voidage E used in the calculations.
tR = 0.52 h if E = 0.3 and tR = 0.37 h if E = 0.5.

Example 8.2 Calculation of the shear zone thickness.

The average diameter of the agricultural seeds used in Example 8.1 is about
1.5 mm. If the particle velocity at the wall during discharge is equal to
0.65 v p , calculate the shear zone thickness and the velocity profile across the
bin.
Solution. Davidson's correlation (equation 8.5) would give
v
- = 1 - (0.35f3L:xI 1. 5 )
vp

where x is the horizontal distance from the wall. The velocity profile vex) is
tabulated below for different values of the empirical constant f3L.

f3L = 1.5

x (mm) 0.5 1.0 2.0 4.0 8.0 10.0 12.0

v/vp 0.694 0.733 0.796 0.881 0.960 0.977 0.986

f3L = 2.0

x (mm) 0.5 1.0 2.0 4.0 8.0 10.0 12.0

v/vp 0.722 0.780 0.861 0.945 0.991 0.977 0.999

 Velocity distribution in bins and hoppers 313

These results indicate that the plug-flow velocity vp is reached within a

distance of about 10 mm from the wall. Hence we would predict a shear zone
thickness about equal to this value.

Comments. The shear zone thickness calculated here is of the order of six
to seven average particle diameters depending on the chosen value of the
empirical constant fh. Hence, in proportion to the bin cross-section, the size
of the shear region next to the wall is found to be insignificantly small.
Furthermore, according to the above table, it is only the first two particles
next to the wall whose velocities are severely retarded. This leads to the
conclusion that the shear zone phenomenon in plug flow of particulate solids
is purely a geometric effect resulting from the contact friction between the
wall surface and the particle layer next to the wall. The wall friction is almost
completely dissipated in this thin region. If a contact friction argument is
used, then it must simply follow that the ratio vw/v p will be independent of the
bed height so long as the wall friction angle <Pw does not vary significantly
with position along the wall. Furthermore, if vwlvp is a function of <Pw only
then it will be independent of the plug-flow velocity, vp.

8.3. 1.2 Converging flow. In general, discharge occurs through an outlet

whose size is much less than the bin cross-section. This requires the material
to 'converge' towards the outlet in the flow region immediately above the
discharge port. Convergence requires the particles to move horizontally
towards the centre as well as moving downwards towards the silo outlet. In
the converging flow zone, we therefore expect to see appreciable horizontal
component, U, of the particle velocities as well as a significant variation
olv/o x of vertical velocities across the flow field.
The size of the converging flow zone above a vessel outlet depends very
much on the hopper half-angle a. With cohesionless materials, mass flow will
occur when the (included) half-angle a of the hopper is sufficiently small (i.e.
usually a < 30°). In this case, the converging flow zone is bound by the
hopper walls and extends as far as the top of the hopper referred to Figure
8.6(a). As the hopper half-angle a is widened, funnel flow will develop. In
funnel flow, the converging flow regime is set up within the central funnel
region bordered on either side by the stagnant zones. The velocity profile
within the funnel is now closely linked to the shape and position of the
stagnant zone boundary, both of which are characteristic of the discharged
material as well as of the hopper dimensions.
In this section two simple analytical solutions for the converging velocity
profiles in mass-flow and funnel-flow hoppers are presented. The first of
these is due to Jenike (1964) and Johanson (1968), and is known as the 'radial
velocity field' solution, appropriate to the mass-flow condition. The second is
the 'kinematic flow model' developed by Nedderman and Tiiziin (1979),
which predicts velocity profiles and stagnant zone boundaries in funnel flow.
314 Storage and discharge of particulate bulk solids

Both analyses are based on a number of simplifying assumptions and hence

must at best be considered as first-order approximations (Tiiziin and
Nedderman, 19S2).

8.3. 7.3 Radial velocity field. Consider an element of a particulate material

in a mass-flow hopper as illustrated in Figure 8.9(a). The component stresses
am 0ee and \e acting on the element are given by

Orr = am (1 + sin<t> cos21jJ)

0ee = am (1 - sin<t> cos21jJ) (8.11)
1: e = - 1: = 8 m (sin<t> sin21jJ)
r f8

using the Mohr's circle construction in Figure S.9(b). (Refer to Chapter 4 for
the definition of the stress and strain states in granular materials undergoing
continuous deformation.) If the material is assumed to be essentially
incompressible during flow, then the equation of continuity in polar
coordinates (r, 8) is given by

-OV + ( v - r -OV) tan21jJ* =0 (S.12)

08 Or
where the angle 1jJ* denotes the major principal strain rate direction along
the radial coordinate r. The assumption of 'co-axiality' (see Chapter 4) of the
major principal stress and strain rate directions allows lenike to set 1jJ* = 1jJ,
i.e. the major principal stress direction in equations (8.11). Along a radial
streamline, the velocity will vary according to
Vr = f(8)r- 1- m (S.13)
where m = 0 in 2-D (plane-strain) and m = 1 in 3-D (axial symmetry).

df(8)
- + (2 + m)f(8)tan21jJ = 0 (8.14)
d8

It is now straightforward to generate the radial velocity profile by integrating

equation (8.14) and substituting the result of f(8) in equation (S.13).
However, the integration of equation (S.14) requires an assumption about the
prevailing stress field, i.e.
1jJ = 1jJ(8) (8.15)
which together with the condition of incompressibility characterises the
'radial stress field'. Substitution of equation (S.15) into (8.14) and integration
between limits 8 = 0 and 8 = 8 yields
Vr = Ar-1-mexp [- (2 + m) fo8tan21jJ(8)d8] (8.16)
 Velocity distribution in bins and hoppers 375

(J

1jI= 90° at e = 0
90 + ro+<I>w at e = a
o 1jI =
2

(a) (b)

Figure 8.9 Radial velocity and stress fields.

where A is the integration constant. This is the radial velocity field solution
first introduced by lenike (1964). The limiting value of the hopper half-angle
a for which the radial velocity field is possible is given by the conditions 'ljJ1
= rt/4 and 'ljJ2 = 3rt/4. In either case, equation (8.16) will produce Vr = O. It
can also be shown quite easily that in these limiting cases, aV/ a8 will either
be zero or infinity (see Jenike, 1961) and the resultant radial velocity profiles
are of the type seen in Figure 8.10. As noted earlier, the condition Vr = 0 at
the wall is met only when the hopper walls are fully rough (i.e. <l>w = <1».
With smooth hopper walls there will always be some slip at the wall, therefore
Vr > O. Hence the rays at which 'ljJ = 'ljJ1 and 'ljJ = 'ljJ2 are not likely in
hoppers whose walls are sufficiently smooth. This also implies that the radial
velocity field is likely in narrow-angled hoppers with half-angle a « 45°.
Furthermore, in his more recent work, lenike (1991), using a different form
of the mechanical failure function (known as Von Mises criterion) instead of
the Mohr-Coulomb criterion given by equation (8.11), has calculated much
smaller values of the hopper half-angle over which the radial velocity field will
apply; Nedderman (1992) gives a more detailed description of these
calculations.

8.3.7.4 Kinematic flow model. The radial velocity field solution due to
Jenike (1964) relies on the assumption of a radial stress field immediately
above the hopper outlet. In addition it also assumes that the particle
streamlines are radial and the flow region extends as far as the hopper walls.
316 Storage and discharge of particulate bulk solids

Hopper half angle a = 13.5" a= 3D'

(a) (b)

Figure 8.10 Radial velocity profiles. Comparison of Dosekun's (1981) experimental

velocity profiles in a wedge-shaped hopper with various theoretical predictions. _._._.,
Brown and Richards (1965); .... , lenike (1961), Nedderman and Tuzun (1979) (B
= lOd), - - , H = 40 cm; ------ (a) H = 20 cm, (b) H = 10 cm.

In funnel-flow hoppers (Figure 8.6(b », none of the above conditions is met.

The shape of the flow region boundary observed in practice is parabolic with
the parabola extending further up the silo in 'planar' or 'plane-strain' type
geometry. Figure 8.11(a) shows some funnel-flow region boundaries observed
in planar and axially-symmetric containers with coarse 'cohesionless' particles.
The particle streamlines seen in Figure 8.1l(a) are also considerably curved
and not radial. Figure 8 .11 (b) compares the vertical velocity profiles
measured inside the funnel region in the planar and axially-symmetric vessels
(Tiiziin, 1979). The profiles of vertical velocity in planar geometry are smooth
in shape while those in a cylindrical silo exhibit a plateau and a central peak.
Clearly in both cases, the velocity profiles are not radial.
Although significantly different in shape, the velocity profiles obtained
under plane-strain and axially-symmetric conditions have a number of
common features. Firstly, when discharging from a central outlet, the vertical
velocity is at a maximum on the centreline and is reduced to zero at the wall.
Furthermore, there is significant acceleration towards the outlet. Hence
unlike the plug-flow region discussed in section 8.3.1.1, the spatial derivatives
a a a a
of vertical velocity vi y and vi x are both significant. The derivative
a a
vi y is zero in the plug-flow region and increases towards infinity at the
a a
outlet. The derivative vi x is zero at the wall and must be zero at the
a
centreline because of symmetry. In fact, vl"a x goes through a maximum
between the centreline and the hopper wall. The horizontal velocity profile
 (i) (ii} (iii)

Cv) (vi tv(i

Figure 8.11 (a) Flow region boundaries of coarse 'cohesionless' materials in funnel-flow bins (Tiiziin, 1979). i. Flat-
bottom plane-strain bin (bin thickness = 2 cm) ii. Plane-strain bin; hopper half-angle = 60°. iii. Plane-strain bin; hopper
half-angle = 45°. iv. Mid-plane of a cylindrical bin showing tracers marking the stagnant zones. v, vi. Deformation
patterns of coloured particle layers in a cylindrical bin.
 318 Storage and discharge of particulate bulk solids

(b)

'0
'T
r TTT"''l7 1\ ~''''TTrrl'l 0..9

80

70
'
,1
.'/ ,,"
r-
0 0 "\

v.:"'.Jrom§.,;:) n.;:'\ ~
('I.
'

\',

('I.
' Height: 1200>
d p - 1.5mm
1
....... B ::: O.25cm
_.------B = O.35cm
0..8

0.7

6.0
ciJ ._ B ::: O.45cm 0.6
,r ~

~
'0 5 0.5
E
<>
>
4D 0..4
>
,~:
3.0 0..3

2D 7: "\ 0..2
'\
0..1
1.0

, "-
"\
~"".LUJ..U..L.UJJ.C"w.'J..'1w.,J..'~"w.1~
•• LLL-,-LLL.LU~L·LL.·
00
- , - 6. - 4, - '. 0. 1. 2. 3 " 5 6 7
Xem rem
(i) (ii)

Figure 8.11 (b) Vertical velocity profiles in planar and axially-symmetric bins
(Tiiziin, 1979), (i) Plane-strain bin with flat bottom, Bin thickness: 2 cm. Orifice
width: 2 cm. (ii) Cylindrical bin. Orifice diameter: 2 cm.

(e)

10 2O~~mmmm~'~I'I~'I~,m"~I'''I~''I'~'~I'~1
d p =1.5mm
Height: Scm
no
"
dp = 1.0mm
Height: 6.0cm

1.0 -
... ....
~ 00 ~ 0.5 -
~

.'.
E E
u -o..t <>
:J :J -'
.. •
-0.5 -

-no
-1.0 -

-10. -&. - 6, -2.

X em
0, 2. 4. .. .,.
r efT.
(i) (ii)

Figure 8.11 (c) Horizontal velocity profiles in planar and axially-symmetric bins
(Tiiziin, 1979). (i) Plane-strain bin with flat bottom. Bin thickness: 2 cm, Orifice
width: 2 cm, (ii) Cylindrical bin, Orifice diameter: 2 cm.
 Velocity distribution in bins and hoppers 319

for the particles is found to behave in an identical manner, as seen in Figure

8.11(c). Clearly, there is a strong basis for relating the horizontal component
of particle velocity, U, to the horizontal derivative of vertical velocity 0 vi 0 x.
If the observed flow fields in Figure 8.11 can be assumed to be independent
of the prevailing stress field, then it is possible to develop a velocity field
solution irrespective of bulk stresses. The stresses in the bulk of a flowing bed
of cohesionless particles are quite small and in converging flow they tend
towards zero at the plane of the outlet. With small bulk stresses and negligible
cohesive forces between particles, the velocity field is propagated by gravity
alone. Hence a purely kinematic solution ignoring bulk stresses is plausible
for the prediction of velocity fields.

Mathematical definition. Consider an assembly of particles moving under

gravity as seen in Figure 8.12. The equation of continuity for an essentially
incompressible material is

OV
- +
ov
-=0 (8.17a)
ox oy
in plane strain where (x,y) are Cartesian coordinates and
OV OU U
+ (8.17b)
oz or
+-=0
r

X-++ve

Figure 8.12 Kinematic flow model.

320 Storage and discharge of particulate bulk solids

in axial symmetry where (r,2,8) are the cylindrical polar coordinates. Motion
can be considered to be propagated by particles moving into the voids vacated
by others in the layer below. As seen in Figure S.12, if the two particles, B
and C, in the lower layer have different velocities, then there will be a
tendency for the particle A to move sideways in the direction of the faster
moving particle. Hence we expect the horizontal velocity, u, to be a function
of the gradient of the vertical velocity. Nedderman and Tiiziin (1979) chose
the simplest non-trivial form by allowing the horizontal velocity to vary
linearly with the vertical velocity gradient. This results in

u = -B-
ov in plane strain; u = -B -
ov in axial symmetry (S.lS)
ax Or
Substitution of equations (S.lS) into (S.17) results in
ov oZv
-=B- (S.19a)
oy oxz
in plane strain and

(S.l9b )

in axial symmetry. The constant B in equations (S.19) is the 'kinematic

constant', which has units of length. The physical arguments advanced above
would suggest that the kinematic constant should be of the order of a particle
diameter.
The form of equation (S.19b) is similar to that put forward in stochastic
theories of granular material flow, which are based on the notion of discrete
random motion of individual particles (see Litwiniszyn, 1971) or the random
propagation of discrete void spaces (see Mullins, 1974, 1979). Such models,
however, assume a priori probability functions similar to equations (S.19a)
and (S.19b) to describe purely random processes where the dependent
variable is probability and not velocity. Furthermore, it is not at all
straightforward to translate the resulting probability distributions into particle
velocity profiles.

Solutions. Two types of solution to equations (S.19) are possible

depending on the appropriate boundary conditions (refer to Nedderman and
Tiiziin, 1979).
1. Similarity solution: The flow towards the hypothetical point sink some
small distance below the single, central orifice at the bottom of a semi-infinite
body can be represented by
 Velocity distribution in bins and hoppers 321

v = v o e-x2/4By (S.20a)
in plane strain and
v = vo e-r2/4Bz (S.20b)
in axial symmetry. Here Vo is the centreline velocity which is given by
(S.21a)
in plane strain and
Vo = QI4rtBz (S.21b)
in axial symmetry. Hence we find that in axial symmetry, the centreline
velocity varies with liz as opposed to (ltVy) in the two-dimensional case for
the same volumetric flow Q.
2. Product solution: The product solution is obtained by considering a single
central orifice of finite size at the bottom of a vessel of finite cross-section.
These conditions result in a solution of the form

;) (S.22a)

in plane strain and

V = vp + LnAn1o(Ar)e -),.2Bz
n (S.22b)

in axial symmetry. The constants An are obtained from a specified velocity

distribution across the orifice. The vertical velocity, v, is assumed to be zero
at the side walls and at the base of the bin from the edge of orifice.

Particle streamlines and stagnant zone boundary. The predictions based

on the similarity and product-type solutions differ very little in cases where
the orifice size is of an order of magnitude less than the hopper cross-section
(Tiiziin and Nedderman, 19S2). For a given orifice size, it can easily be shown
that the similarity solution predicts parabolic streamlines whose widths
depend on the value of the kinematic constant, B, and hence on the particle
size d. Choosing the simplest case of a two-dimensional flow field, the stream
function I.; is given by
I.; = f v dx (S.23)
The substitution of equation (S.20a) into equation (S.23) above and
subsequent integration yields

I.; = Q erf[ _x ] (S.24)

2 2vBy
322 Storage and discharge of particulate bulk solids

Along a streamline 1; = constant; therefore equation (8.24) predicts

streamlines with shapes satisfying
x=AYBy (8.25)
where A is now an arbitrary constant whose value depends on the chosen
value of 1;. Hence equation (8.25) is found to predict the width of the
convergent flow region to vary as V d. A similar analysis is possible for the
case of axially-symmetric flow, Tiiziin (1979). It is also possible to use the
product solution in equations (8.21a) and (8.21b) to generate streamlines.
However this would require numerical computation and iterative techniques
(Tiiziin, 1979).
To calculate the flow region boundary, it is necessary to choose the value of
1; so that the arbitrary constant A in equation (8.25) can be evaluated. The
value of S equals Q12 at the wall, so that it is not possible to generate a
streamline in this case. The flow region boundaries seen in Figure 8.13 are
generated by allowing S = 0.99(Q/2). The stagnant zone boundaries
observed in experiments should lie quite close to the theoretical streamline S
= 0.99(Q/2), if the prevailing velocity field is predicted accurately by the
kinematic solution.
It is also possible to compare the experimental flow boundaries with the
contours of vertical velocity obtained from the product solution: see equation
(8.20b) above. Along a contour of vertical velocity, the value of v is kept
constant at a chosen value. Hence to generate velocity contours, it is first of
all necessary to generate velocity profiles at different heights above the orifice
and then generate the chosen contour by searching the velocity profiles for
the appropriate value of the vertical velocity v. The experimental evidence
presented by Nguyen et ai. (1979) and Tiiziin and Nedderman (1982) suggests
that along the flow region boundary, particles move at velocities an order of
magnitude less than the corresponding centreline velocity at a given height. If
Vboundary ::::; 0.1 vo , then this allows the approximate theoretical velocity
contours to be generated for comparison with the experimental flow region
boundaries (Figure 8.13).

The kinematic constant, B. The scaling of the flow fields predicted by the
model depends entirely on the chosen value of B. It has already been noted
that B might be expected to be a small multiple of particle size. Indeed, the
experimental profiles measured in planar containers (see Figure 8.11(b» do
seem to indicate B = 2d when funnel flow occurs (Tiiziin, 1979; Dosekun,
1981). In two-dimensional flow fields, both the shapes of the velocity profiles,
as well as the magnitudes of the velocities are predicted with reasonable
accuracy when values of B of the order of 1-2d are chosen. However, in
axially-symmetric flow, the shape of the vertical velocity profile is no longer
bell-shaped (Figure 8.11). The velocities in the fast-moving central core are
predictable with values of B ::::; 2d; however in the surrounding plateau region,
 Velocity distribution in bins and hoppers 323

40~====~==~=?2;a:=~30n.~5~em~=+===+====~
-12a =7.6 em\--
35 2a= 12.7em
t<--t- 2a = 20.3 em -l-~

30

25
E
~
1: 20
0>
'w
I

15

10

o~~-~~~~~
-15.0 -10.0 -5.0 0 5.0 10.0 15.0
X (em)

Figure 8.13 Superimposition of the experimental flow boundaries in different bins

and comparison with the predictions of the kinematic model. -'-'-' Experimental flow
boundary (experimental time = 1 s). - - - - - - Theoretical contour of vertical
velocity (v = 0.2 cm S-1) ....... Theoretical streamline (1jJ = 0.99Q/2).

the profile flattens considerably which would require much larger values of B
(usually of the order of 5-lOd), in order to satisfy equations (S.20a) and
(S.20b). Moreover, an accurate prediction of the stagnant zone boundary
does also require B to be quite large (Figure S.U(b». This result cannot be
explained on the basis of the physical arguments advanced above. Hence we
must conclude that the proposed linear relationship between the horizontal
a a
velocity u and the gradient vi x (see equation S.lS) is far too simple and
restrictive to account for the velocity fields observed in axial symmetry. One
possible physical explanation for the observed variations in the value of the
kinematic constant could be the variation of local void age within the flow
field which is not included in the model. In an axially-symmetric flow field,
there is evidence to suggest the existence of a dilated core region separated by
a more compacted slow-moving annulus (McCabe, 1974).
324 Storage and discharge of particulate bulk solids

8.3.1.5 Comparison of radial velocity field and kinematic model solutions.

As noted earlier, if material is discharged under a high material head in the
silo, then a transition region is observed where the material originally in plug
flow starts accelerating towards the orifice. The size and the position of the
'transition zone' in the particle flow field depends noticeably on the vessel
geometry and to a lesser extent on particle properties such as size and shape.
The effect of the vessel geometry is illustrated in Figure S.13. In silos which
join a parallel-sided top section to a steep hopper section with inclined walls,
the 'transition corner' in Figure S.S(b) marks the change in vessel geometry.
The practical evidence suggests that when the material in the hopper is in
mass flow, the transition from plug flow to converging flow starts some way
below the bin-hopper transition. If the materials are discharged from a flat-
bottomed or a wide-angled container, then the size of the stagnant regions
formed in funnel flow will determine how high above the outlet plane the
transition region will lie (Figure S.7). As seen in Figure S.7, the flow
transition usually starts some way above the top of the stagnant zone
boundary and can also continue into the funnel region below, depending
upon the size and shape of the stagnant zones.
The transition to plug flow further up the silo is observed to be much faster
and sharper in axially-symmetric or 'plane-strain' type containers (McCabe,
1974; Tiiziin, 1979). The result is predictable using the kinematic model.
Figure S.14 compares the plots of (vwlvp) versus H (the height above the
orifice) which result from the plane-strain (equation S.22a) and the axially-
symmetric (equation S.22b) product solutions for two different values of the
kinematic constant B. The comparison of the plane-strain and axially-
symmetric curves corresponding to a certain value of B shows that in general,
the axially-symmetric solution gives rise to a more rapid transformation to
plug flow with height above the outlet than its plane-strain counterpart.
However, as the value of B gets larger (i.e. B = 6-7d) then the two
theoretical curves seem to move closer to each other. It is, however, possible
to produce this result much more simply using the similarity solution form
(equations S.20a and S.20b), from which it can be easily shown that

v 1
- ex in axial symmetry (S.26a)
vp Bz
and
v 1
- ex - - in plane strain (S.26b)
vp VBy
where v is the particle velocity at a given height within the converging flow
zone and vp is the plug-flow velocity.
Equations (S.26a) and (S.26b) above also show quite clearly that the size of
 Velocitv distribution in bins and hoppers 325

1.0

./
". .--- ---- ~~-

0.9 ~
/ ~
~

., ,
~
/
/ .,
0.8 ,,
/
, ,,
! 2,
,,
0.7
/
e-
I
I

'S / ,
I

"il 0.6 I
> / I
I
~ I I

~
I
I I
0.5
e; I I
I
I

.~ I I
I

0
"il 0.4 I I
I

> I I

, I

~ 0.3
I
I
I

I I
I
I I
I

I I
0.2 , I
I
I ,,
0.1
,/
I ,'

0.0
.(.r'I" I I I I I
0 10 20 30 40 50 60 70 80

Height (em)

Figure 8.14 Comparison of axially-symmetric (1) and plane-strain (2) solutions of

kinematic model. - - - , B = 0.75 cm; - - - - -, B = 0.25 cm.

the transition zone depends on the particle diameter. As the particle diameter
is increased in the same vessel, v tends to vp much faster with height above the
orifice. This is also consistent with the measurements discussed earlier of the
size of the stagnant regions in funnel flow of coarse granular materials in
plane-strain and axially-symmetric containers (section 8.3.1.4).
Nedderman (1992) points out that there are no radial flow solutions to
equations (8.19a) and (8.19b). This can be seen by expressing equation
(8.19a) in terms of the stream function in equation (8.23)

-v = -
as and hence u =
as (8.27)
ox oy
which results in

(8.28)
326 Storage and discharge of particulate bulk solids

In radial flow, the stream function 'i; is a function of the ratio x/y only.
Denoting a = x/y and replacing derivatives with respect to x and y in terms of
a gives

a,'i; d'i; aa x d'i;

= = y2 da
(8.29a)
ay da ay

a 2'i; d 2 'i; ( aa) 2 - -1 d 2 'i;

----- (8.29b)
ax 2 da 2 ax y2 da 2
Hence from equation (8.19a)
d'i; B d2 'i;
(8.30)

which shows that the stream function 'i; is not a function of a only and radial
flow does not apply. However, it is worthwhile to note here that the
magnitude of the kinematic scalar B/x in equation (8.30) is a function of the
cross-sectional distance within the flow field which is entirely consistent with
the observations of flowing voidage change across the bin cited in section
8.3.1.4.

8.3.2 Transient flow fields of 'free-flowing' materials

The flow fields discussed in section 8.3.1 are characteristic of continuous
material discharge from a bunker under a high material head. However, in
many industrial operations, bulk storage units will operate in 'batch' or 'semi-
batch' modes, i.e. the silo will either be refilled only after it has been
completely emptied or more commonly refilling will start only when the
material head in the bunker gets low enough to reduce the discharge rate
from the orifice (see section 8.4). In silos operating in the 'batch' or the 'semi-
batch' mode three different transient stages of flow will occur:

(i) initial transient stage;

Oi) pseudo steady-state;
(iii) final transient stage.

Figure 8.15 depicts the movements of the material top surface in a silo during
batch discharge and illustrates the resulting flow fields during each successive
transient stage of the batch discharge; see also Tuzun et at. (1982).

8.3.2. 1 Initial transient stage. Following the start of discharge from the
outlet, there exists a small time interval during which the 'opening up' of the
 Velocity distribution in bins and hoppers 327

flow zone is observed (Figures 8.15(a) and 8.15(b)). A dilation wave

propagates up the silo as air is drawn from the orifice into the material bed.
On the passage of this wave, the material starts to move and dilates from its
static fill voidage to one characteristic of the flowing material. During the
initial transient stage, the material bed is divided into a fast-moving 'flame-
shaped' dilated core in the middle surrounded by stagnant material outside
the core boundary. The so-called 'boundary' continues to move up the silo as
well as towards the vessel walls until it finally touches the walls, whereupon
no further expansion of the flow region is possible. Mullins (1979) has
predicted the motion of this wave from the kinematic model and Buchele and
Wynn (1980) have confirmed his predictions using the values of the kinematic
constant found by Tuzun (1979). In axial symmetry, the resulting equation of
the stagnant zone boundary at time T is given by

2Bz 2 rt!'l.Q
T = - - - - exp (
QQf
~
4Bz
) (8.31)

where Qf is the flowing bulk density and !'l.Q is the difference between the
flowing Qf and static bulk densities, Qo (Figure 8.16).
According to these calculations the duration of the initial transient with
free flowing coarse materials (i.e. d ;?: 1 mm), is predicted to be a strong
function of the particle size, d, as well as of the vessel geometry.
Throughout the initial transient stage of flow, the material top surface in
the bin will remain stationary. The material head will start to drop only after
the flow region boundary meets the vessel walls (Figure 8.15(c) and 8.15(d)).

8.3.2.2 Pseudo steady-state. When the dilation wave has passed through
the material, the initial transient phase of batch discharge is complete and
thereafter the velocity fields will be the same as those observed during steady-
state discharge. However, if the materials are discharged as a batch, then the
material top surface will continue to decline for some time within the so-
called 'plug-flow zone' without any change of shape (Figure 8.15(c)). Pseudo
steady-state will continue until the top surface descends to a level comparable
with the top of the stagnant zone (Figure 8.15(f)).

8.3.2.3 Final transient stage. As the material top surface approaches the
converging flow zone, the centreline velocity becomes greater than the
velocities to either side and a depression is formed in the top surface. This
depression grows as the surface descends until finally the slope exceeds the
angle of repose, whereupon material cascades down the top surface towards
the centre (Figures 8.15(g) and 8.15(h)).
These observations point to a time-dependent velocity field within the
converging flow zone as the velocity profiles below the material top surface
will now be a strong function of the changing shape of the top surface. If the
328 Storage and discharge of particulate bulk solids

(a) (b) (c) (d)

f--- Initial Transient - - l / IE- Pseudo"

(e) (f) (g) (h)

Steady State -1 IE- K- - - Final Transient -----if

(a) (b) (c) (d)

+-- Initial Transient + Pseudo"

(e) (f) (g) (h)

I
I
I
II
I
l ~:;;.--
I I
I
I III
I I

Steady State ~ ~ Final Transient

Figure 8.15 Transient stages of batch discharge from bins and hoppers.
 Velocity distribution in bins and hoppers 329

.............
",...... • •••••~.~ s
135,25
,,
: " ,
'\\ ,
,
,
: d p =2.07 mm I '"
.
" <Pw =1 8 ° ' *
122.00
.·
··
106.75
··

91.50

E
.s. 76.25
1:
OJ
·iii
I

61.00
23s

,, ,
.,.
45.75 ,,
..~~
30.50
52s
85s
150 s
15.25

,
L-____~~~~I~~_____ L_ _~
-10.16 -5.08 o 5.08 10.16
X (em)
Figure 8.16 Calculation of the transient flow region boundary in axially-symmetric
discharge. v = Particle velocity; v1 = propagation velocity of flow region boundary;
00 = static fill bulk density; Of = flowing bulk density. Do = 0.65 cm; D =20.3 em.
330 Storage and discharge of particulate bulk solids

material is in funnel flow, then the stagnant region extends up to the top
surface and therefore the continuous descent of the surface must also result in
the partial erosion of the stagnant region (Figures 8.15(f) and 8.15(g)). In
funnel flow, the completion of discharge from a vessel will not always ensure
complete emptying of the vessel contents. In wide-angled hopper sections and
flat-bottom containers, pockets of stagnant or 'dead' material may remain
inside the vessel on either side the discharge orifice (Figure 8.15(h)). The
amount of material remaining inside the vessel will be a function of the
internal and wall angles of friction and of the hopper half-angle (see Chapter
4). The observations cited above relating to the transient stages of batch
discharge from a tall vessel show quite clearly that it is the aspect ratio of the
container which determines to a large extent the relative durations of initial
and final transient stages. If the vessel aspect ratio is small (i.e. H < 1-2D)
then the so-called final transient will immediately follow the initial transient
as the material top surface will start to deform as soon as the flow region
boundary meets the vessel walls. In shallow bins and hoppers, therefore,
there will not be a pseudo steady-state.
In contrast, when the aspect ratio is high (i.e H > 3-4D) then we should
expect to see a significant pseudo steady-state as the material top surface
descends more or less horizontally in the plug-flow region of the container
(Figures 8.15(c) and 8.15(d)). In vessels of high aspect ratio, the duration of
the initial transient during which a dilation wave moves up the bunker will
also be substantially prolonged, although this period will also be affected
significantly by the ratio of the orifice width to the vessel width. In general, if
the aspect ratio is high and the vessel width is an order of magnitude or more
larger than the orifice width, then the initial transient stage becomes
significant. The model tests on tall concrete silos indicate initial flow
transients lasting as long as 0.5-1.0 h of discharge time (Nielsen, 1983a,b). If
the vessel is short and narrow, then the initial transient is reduced to a few
minutes (Tiiziin, 1979).

8.4 DISCHARGE RATES FROM BINS AND

HOPPERS
8.4.1 Introduction
The prediction of the rates of discharge of granular materials from storage
bins and hoppers is of primary importance in industrial processes which
involve bulk solids handling and transportation. At the design stage,
calculations of the material residence times in storage units as well as the
sizing of appropriate mechanical conveying equipment or transport pipelines
require reliable predictions of solids discharge rates. In plant operation,
reproducible and efficient handling demands an understanding of the effects
 Discharge rates from bins and hoppers 331

on discharge rates of altering the process variables such as particle-size

distribution, particle moisture content, air pressure and relative humidity and
vessel temperature.
In this section, an extensive review is presented of the existing correlations
for the prediction of discharge rates of free-flowing materials. Some simple
analytical theories are also discussed and their predictions are compared with
experimental observations.

8.4.2 Early empirical work and discharge rate correlations

The discharge of granular materials through orifices is best described in terms
of the typical cylindrical and conical hoppers shown in Figure 8.17, in which
the notation is also defined. The effect of the quantity of material in the
hopper, as expressed by the height H, on the flow rate has long been realised
to be slight. The stress analyses of Janssen (1895) and Shaxby and Evans
(1923) (see section 8.5.2) have shown that the weight of the material is borne
substantially by the walls of the hopper and the stress state near the orifice is
therefore independent of H. Most workers have reported no dependence of
the mass flow rate Won H, though Newton et al. (1945) reported that W is
proportional to J-fl.04. More typical results indicate that W is independent of
H provided H exceeds some critical value. Hinchley (1926) and Demming and
Mehring (1929) gave this critical value as 2.5D and Brown and Richards
(1959) reported that during batch discharge W remains constant until the
material head in the hopper is less than the hopper diameter, i.e. H < D. At

r+-- D
-- t
I
I

I,
\
j
--

Figure 8.17 Schematic diagram defining the notation in cylindrical and conical
hoppers. See also Nomenclature.
332 Storage and discharge of particulate bulk solids

these low values of H, the top surface of the material almost always shows a
central depression. Thus there seems to be substantial agreement that W is
independent of H until the hopper is almost empty though the precise value
of the critical height is not clear.
Similarly, there seems to be general agreement that the mass flow rate is
independent of the vessel diameter D provided this is not too small. Both
Ketchum (1929) and Brown and Richards (1960) reported that W is constant
provided D > 2.5Do and the latter gave correction factors for the smaller
values of D. Franklin and Johanson (1955), on the other hand, gave the
criterion as (D - Do) > 30d where d is the particle diameter. For smaller
values of D, larger flow rates are found and in the limit as D ~ Do, the whole
mass of material accelerates indefinitely under gravity.
Ignoring the possible effect of the particle diameter, d, dimensional
•
analysIs suggests that W ex: Qb g Y2 Do5/2. However, many earIy work ers
reported that W ex: D~ where n ;:,; 2.5. Under these circumstances, the
correlation must contain dimensional empirical constants or d to the power
(2.5 - n). Ketchum (1929) gives n = 3.0, Franklin and Johanson (1955) 2.93,
Rausch (1948) 2.8 and Brown and Richards (1959) 3.1. Here the dimensions
of the correlation are balanced by the insertion of a term in d(2.5 - n); this
rarely predicts the effect of particle size correctly.
Beverloo et ai. (1961) plotted W 2/, against Do and obtained a linear
relationship as shown in Figure 8.18. The interesting feature of this
presentation is the intercept Z. They correlated this with the particle
diameter, finding Z = kd. For spherical particles k is about 1.5 but somewhat
larger values were found for angular particles. In the latter case the value of k
depends crucially on the choice of typical dimension of the particle. The
constant k was found to be a function solely of the particle shape and no
dependence on other properties could be found. Beverloo therefore
correlated his results in the form
(8.32)
and this correlation has been applied successfully by many subsequent
workers. Though normally attributed to Berverloo, similar ideas had been
current prior to Beverloo's paper and indeed a correlation of this type had
been proposed by Hagen as early as 1852. Weighard (1952) also gave
W ex: (Do - Z)'h, and Brown and Richards (1960) devoted considerable
attention to the concept of the 'empty annulus'. No particle centre can
approach within a distance d/2 of the orifice edge and therefore all particle
centres must pass through a circle of diameter (D - d). This does not in itself
explain why k > 1 but Brown and Richards (1960) also found that there was a
decrease in the number of particles flowing per unit time in the zone adjacent
to the orifice edge. The measurements of Laohakul (1978) show that adjacent
to a solid boundary, there is a region of retarded flow a few particles in
thickness. This region clearly has much in common with the shear zone
 Discharge rates from bins and hoppers 333

Figure 8.18 Graphical representation of Beverloo et al. (1961) discharge rate

correlation.

described by Roscoe (1970) and the term kd may simply be the displacement
thickness of this zone.
Alternative formulations for the effects of particle and orifice diameters
include the correlations of Fowler and Glastonbury (1959),
\12
D )0.185
Wex: ADh ( - -
h
(8.33)
d
Rose and Tanaka (1959),

W ex: D'/~[:o -3 r· 3
(8.34)

and Harmens (1963)

Wex: D'/~ [f(<I»

0.45
0.35(dlD o)1.5
+ (dlD o )1.5
]
(8.35)

Though in many circumstances these give equally good correlations, the

physical arguments behind them are less satisfying than those leading to the
Beverloo correlation.
334 Storage and discharge of particulate bulk solids

All the above correlations break down for fine materials, i.e. for d <
500 !lm. According to Crewdson et at. (1977), this is due to the effects of
interstitial pressure gradients and this aspect is discussed in greater detail in
section 8.4.4.
For orifices that are less than about six particle diameters across, the flow is
intermittent and irreproducible. Equation (8.32) should therefore not be used
if Do < 6d, nor may it be assumed that putting the group (Do - kd) equal to
zero correctly predicts the orifice-to-particle diameter ratio at which flow
ceases. This is given by Brown and Richards (1959) as 2.5 for slots and 4.0 for
circular orifices. These results of Brown and Richards also support the earlier
experimental work by Langmaid and Rose (1966). However, for fine
powders, a different mechanism known as arching is found. A stable arch can
be formed over orifices of considerable size and this prevents flow taking
place. The prediction of such arches is considered in Chapter 4 when
discussing lenike hopper design procedure.
Clearly, the Beverloo correlation requires a density for dimensional
consistency but there is dispute as to the appropriate value to use since the
density within a hopper varies with respect to both time and position. Many
authors are not at all clear in their definition of density, but Beverloo's
original paper in 1961 specifies the use of the initial density Qi resulting from
the filling process. However, as described in section 8.3.2 above, on initiation
of flow the material dilates to some voidage characteristic of the flowing
material. Hence for more compacted beds, there is a longer period between
the initiation of flow and the time at which the top surface starts to descend.
There is therefore a need to define a 'flowing density', Qr, as the ratio of the
mass flow rate to the volumetric flow rate calculated from the observed rate
of descent of the top surface. This definition of Qf is found to be substantially
independent of the initial voidage or the mass flow rate. Using this density in
the Beverloo correlation, Nedderman (1992) reports a much smaller range of
values of C. In the original formulation,
W = CQig'h (Do - kd)'/2 (8.36)
and C was found to be in the range 0.55 < C < 0.65. In the modified
formulation
W= C'Qfg 'h ( Do - kd) 'I 2 (8.37)
and C' lies in the range 0.575 < C' < 0.595.
Note that C is dimensionless and thus does not depend upon choice of units
so long as they are consistent. The above values of C therefore differ from
those given in the paper of Beverloo et at. which uses mixed units.
Both X-ray investigations (Bosley et at., 1969; Bransby and Blair-Fish,
1979; Lee et at., 1974) and observations by transmitted light through narrow
bins (Tiiziin, 1979) clearly show considerable changes in voidage both on
initiation of flow and from point to point within the hopper. Tiiziin (1979)
 Discharge rates from bins and hoppers 335

gives an initial voidage of 0.32 and flowing voidage of 0.37 for glass ballotini.
Thus a potential error of up to 20% exists if the wrong density is used.
Clearly, some density characteristic of the flowing material is more
appropriate than the initial density. In more recent work by Hosseini-Ashrafi
and Tiiziin (1992), voidage profiles at the orifice were measured during
discharge from conical hoppers using a y-ray tomographic scanner. They
report a 10-20% reduction in flowing density in the vicinity of the orifice
plane due to the dilation of the flowing material.
The effects of the angle of friction <I> and the particle shape on the flow rate
have been investigated by Demming and Mehring (1929), Franklin and
Johanson (1955), Rose and Tanaka (1959), Harmens (1963) and Kotchanova
(1970). The effects however seem to be small. Rose and Tanaka recommend
the inclusion of a multiplicative factor of
exp (-7.7 X 10-6 c! d 3 Qg'l2)
to account for the effects of cohesion c. However C is a strong function of the
voidage and it is not specified at which voidage or stress level c should be
measured. It should also be noted that the group above is not dimensionally
consistent.
Orifice shape is clearly an important parameter and has been studied by
Fowler and Glastonbury (1959). They report that
W = A(Dh)V2 (D hld)o.185 (8.38)
where A is the area of the orifice and Dh is the hydraulic mean diameter. This
correlation preceded that of Beverloo and includes no allowance for the
empty annulus. Following the Beverloo approach it would seem reasonable to
work in terms of the area A * and hydraulic mean diameter Dt of the space
remaining after a zone of width kdl2 has been removed from the perimeter of
the orifice. The Beverloo correlation can be expressed in these terms as

4C 4C
W= Q0 *(gDt)'12 where = 0.75 (8.39)
Jt Jt

Thus with rectangular orifices of size b X 1, Dh = 2b. This substitution results

in the expression
W = 1.03Qbg V2(l- kd)(b - kd)'/' = 0.73QA *(gDt)'/2 (8.40)
This result is consistent with the modified Beverloo expression (equation
(8.39».
The major contribution to the understanding of the effect of wall
inclination, U, on the flow rate is the work of Rose and Tanaka (1959). For
larger values of u, a stagnant zone will occur and it seems reasonable to
assume that since the walls will be buried within this stagment material, their
336 Storage and discharge of particulate bulk solids

inclination will have little effect. Following these ideas, Rose and Tanaka
(1959) proposed the use of a multiplicative factor F( a, X) where
F(a, X) = (tana tanx)-D·35 a < 90° - X (8.41)

F(a, X)= 1 (8.42)

for use in conical hoppers. Here X is the inclination to the horizontal of the
stagnant zone boundary when core flow occurs. This is equivalent to the angle
of <Pd defined in equation (8.2). It is noteworthy that the flow rate increases
with decreasing a and more rapidly so as a approaches zero. Though X is
clearly defined in Rose and Tanaka's paper as the inclination of the stagnant
zone boundary, they unfortunately used the symbol <p, a symbol that is
normally reserved for the angle of internal friction. This has given rise to
much confusion, especially as in a cohesionless material the angle of internal
friction is closely equal to the angle of repose. It must be emphasised that the
quantity in the Rose and Tanaka correction factor is the inclination of the
stagnant boundary and not the angle of repose of a free surface. The two angles
are quite different. Tiiziin and Nedderman (1982), for example, quote values
of X of 45° in glass ballotini for which <P is about 25°. This result appears to
conform somewhat with the argument that the stagnant zone boundary is a
'slip plane' within the material and hence the angle of the slip plane is given
by E = Jt/4 ± <p/2 (refer to the Mohr-Coulomb failure criterion in Figure
4.14). Tiiziin and Nedderman's result is also confirmed by Nguyen et al.
(1979) who investigated flow boundaries in shallow wedge-shaped hoppers.
Zenz (1975) and Laird and Roberts (1979) also concluded that W does not
depend on a for a > 90° - X and they give values of X considerably greater
than <p. Both these papers present correction factors of the type (tanatn for
small a. Zenz gives n = 0.5 but Laird and Roberts could find no single value
of n that would correlate all their data and quote values ranging from 0.434
for 2 mm ballotini to 0.29 for sand.
While there is general agreement with the form of the dependence of mass
flow rate on wall inclination, further work is needed. The effect of wall
roughness has been studied by Nguyen et al. (1979). Intuitively, one would
expect the flow to decrease with increasing wall roughness and this was found
to be the case for narrow-angled hoppers (a < 90° - X), though the change in
flow rate was rarely more than 10%. However, for a wide-angled hopper, a
small increase in flow was obtained by roughening the wall. This is attributed
to an increase in the value of X so that a narrower flowing zone was formed
with resultant increase in flow rate.
Most investigations have been concerned with centrally placed orifices but
Kotchanova (1970) has also considered eccentrically placed orifices. In view
of the increase in mass flow rate caused by the close proximity of a vertical
wall (see the discussion of the effect of the ration DIDo), it is hardly surprising
that increased flows are obtained. A 10-20% increase in flow rate was found
 Discharge rates from bins and hoppers 337

for an orifice adjacent to a wall and up to a 35% increase for an orifice in the
corner of a rectangular hopper.
All the results discussed in this section have been otained in laboratory-
scale hoppers. There seem to be no published results for large orifices. One
reason for this may be that most hoppers discharge directly onto a conveyor
of some sort and the flow rate is controlled by the conveyor characteristics.
The reader is therefore referred to specialist literature in national design
codes for bulk solids transport equipment.

8.4.3 Theoretical predictions of mass flow rate

Many of the early theoretical analyses rely heavily on the concept of the 'free-
fall arch'. This is the surface in the neighbourhood of the orifice that forms
the lower boundary of the packed bed. Above the free-fall arch, the particles
are in contact with one another and analyses appropriate to packed beds must
be used. Below the free-fall arch, the particles are not in contact with one
another and accelerate freely under gravity unless impeded by some
hydrodynamic drag due to the interstitial medium. The concept of the free-
fall arch is the cornerstone of some earlier theoretical analyses and is implied
in later ones by the form of the boundary conditions since the normal stress
on the free-fall arch must be zero.
The observation that the flow rate from a hopper is independent of the
depth, H, of the material suggests that factors in the vicinity of the orifice
control the flow. This is supported by the observation that

(8.43)

compared with the general form for liquids

(8.44)

Many earlier theories started from the assumption that the flow was
independent of H and was determined solely by the nature of the free-fall
arch. The work of Harmens (1963) is typical of this approach. He assumes a
shape for the free-fall arch which is independent of scale so that the height
above the orifice plane is necessarily of order Do. Particles are assumed to
detach from the arch with considerable velocity and to accelerate freely under
gravity. Their velocity on passing through the plane of the orifice is therefore
of order (gDo)'Iz and the flow rate is thus proportional to g'lzDos/2, whatever
shape of arch is assumed. The constant of proportionality is determined by
the assumed shape of the arch and the unknown rate of detachment of
particles from the arch.
The 'minimum energy theorem' of Brown and Richards (1965) is likewise
crucially dependent on the details of the free-fall arch. They consider an
equation of the Bernoulli type for the total energy content, T, of unit mass of
338 Storage and discharge of particulate bulk solids

material as it flows rapidly towards a point sink somewhat below the orifice,
as shown in Figure 8.19.

v2 a
T = gr cosO + + (8.45)
2 g

they argue with conviction that the energy content must decrease with time
and hence that (dT/dr) must be positive. The stress term (alg) is dismissed
from the analysis using the argument of Janssen (1895) that, for a cylindrical
bunker, the stresses become invariant with depth and hence (daldr) is zero
(see section 8.5). This part of the argument is not sound, as the Janssen
analysis cannot hold right down to the free-fall arch where a must fall to zero.
The dismissal of the a term ensures that H does not enter the analysis and in
the absence of any other quantity with the dimensions of length, a result of
the form W ex gY2 D 5/2 is inevitable.
For radial flow, the velocity v is given by (Alr2) in axisymmetry (or (Air) in
plane strain) where A is a function of angular position only. Thus

dT
g cosO - (8.46)
dr ,s
While it is entirely reasonable to assume that (dT/dr) is positive throughout
the region above the free-fall arch, Brown and Richards make the additional
assumption that (dT/dr) falls to zero on the arch. Besides this unconfirmed
assumption, Brown and Richards also assume that the arch is spherical and
centred on the virtual apex as shown in Figure 8.19. Hence
A2 = ~ cosO (8.47)
and the radial acceleration at the arch will be

v -dv I g cosO (8.48)

dr ro

Carleton (1972) came to the same conclusion in saying that the particles are in
free-fall below the arch and hence their radial acceleration is g cosO. He
assumes that acceleration is a continuous function of position and hence the
acceleration immediately above the arch must also be g cosO. This argument
is open to question since there is no physical requirement that acceleration is
a continuous function of position. Sokolovskii (1965) has shown that in
granular materials discontinuous forces and accelerations are to be expected.
The radial velocity Vr is given by
 Discharge rates from bins and hoppers 339

Figure 8.19 Schematic diagram of 'free-fall arch' and 'virtual apex' at the outlet of a
cylindrical bin.

A
v=-
r ?
V2 5/
g r o"COS
V,
e
=------------ (8.49)

and the volumetric flow rate is given by

Q = (f3
)0
2nr sine Vrr de
4n
= (8.50)
3

where 13 is the included half-angle of the flowing zone, i.e. (900 - X).
Since the orifice diameter, Do = 2ro sin13 (Figure 8.19), hence

V2Jt
Q= g V, Do5/2 (1 - cos3/213)/sin5/213 (8.51)
6
340 Storage and discharge of particulate bulk solids

This theory provides a not unreasonable prediction of the mass flow rate,
commonly overpredicting by a factor of about two. However it must be asked
how much of this is due to coincidence. The suppression of the term in a
makes the correct dependence on Do inevitable and the remaining assump-
tions merely affect the dependence on /3, a quantity that is not often known
with certainty. However the prediction that the velocity distribution is given
by

g V2r 5/0 2 cos 1/28

Vr = ---?-,---

is not in agreement with experiment. Tiiziin (1979), for example, found a very
much stronger dependence on 8 (see Figure 8.10). Like all the other theories
considered in this section, the Brown and Richards analysis is based on
continuum mechanics and can therefore give no information on the effect of
particle diameter.
Three very similar analyses have been presented by Savage (1965), Sullivan
(1972) and Davidson and Nedderman (1973), to explain the lack of
dependence of the mass flow rate on the height H. Unlike the analyses of
Harmens (1963) and Brown and Richards (1965), which assume that W is
independent of H, these analyses include H and show that it has negligible
effect on the flow rate.
The analysis of Davidson and Nedderman (1973) considers a narrow-
angled, smooth-walled conical hopper as shown in Figure 8.20. Since the
shear stress is zero on the smooth wall and on the axis of symmetry, it is
assumed to be zero throughout, and the radial, tangential and circumferential
stresses are therefore principal stresses. As the flow is converging towards the
apex of the cone, a passive state of stress is assumed, and invoking the Haar-
Von Karman hypothesis and the Coulomb failure criterion (see Chapter 4) it
is shown that

= a

(1 +
'1''1'

= sin<j))
Orr (8.52)
1 - sin<j)

where <j) is the angle of internal friction.

Substituting these values into Euler's equation and assuming that the
velocities are radial and independent of angular position yields a first-order
differential equation in Orr that can easily be integrated. The two boundary
conditions, Orr = 0 on r = ro (the free-fall arch) and Orr = 0 on r = r1 (the top
surface) permit the evaluation of the constant of integration and the overall
flow rate. The mass flow rate is predicted to be
 Discharge rates from bins and hoppers 341

Figure 8.20 Schematic of radial polar coordinate system in a conical hopper.

1 + K ] Y2 [ 1 _ 3-2K] Y2
W = 2:rtgg Y2 (1 - cosa) [ - - - . /~ y (8.53)
2K - 3 1 - y-2-2K

where y = (rllro) and is normally large.

For commonly encountered materials <p is in the range 25-50° and hence K
is in the range 2.5-7.5 and y is therefore always raised to a large negative
power. The term including y is therefore always close to unity and can be
neglected. Replacing ro by the orifice diameter (Do = 2ro sina) yields

:rt gg'hv'~2 [ 1+ K ]Y2

(8.54)
W =4 sin'h a 2(2K - 3)

It must, however, be borne in mind that this analysis is only valid for small
values of a, since it has been assumed that the material is subjected to a
radial body force of gg.
342 Storage and discharge of particulate bulk solids

A similar analysis for an orifice of width b in a chisel-shaped hopper of

thickness I gives

w = _Q_g'-,-V21-;-b h
3
_ [ 1 +K ] v,
(8.55)
sin v2 a 2(K - 2)

The achievement of these analyses is the demonstration that the flow rate is
independent of y and hence H. The dependence on the orifice dimensions
follows automatically. The effect of angle of friction, via the term in K, is
shown to be small, but this does not seem to have been tested experimentally
in any satisfactory way. On the other hand, the analysis can give no
information on the effect of particle size and the dependence on the hopper
half-angle a is not in accord with experiment. In general this analysis
overpredicts the flow by a factor of 1.5-2. This method cannot predict a
velocity profile since it is necessary to assume that the velocities are
independent of angular position.
Savage (1967) attempted an alternative approach to the solution of the
equations of motion based upon the method of integral relations. By fitting
the stress and velocity distributions using polynomials and by integrating the
equations of motion across the width of the hopper, a set of ordinary
differential equations was obtained. Applying the appropriate boundary
conditions, these equations were integrated to yield closed form solutions for
the stress and velocity distributions as well as the flow rate. The predicted
flow rates in wedge-shaped hoppers using realistic values for the wall and
internal friction angles were lower than those for zero wall friction but still
considerably higher than the measured flow rates.
Williams (1977) proposed upper and lower limits to the flow rate which he
obtained by solving the equations of motion approximately along the
centreline and along the hopper walls. His analysis suggests that wall friction
has only a small effect on the velocity distribution and mass flow rate.
Savage and Sayed (1979) used a development of the method of integral
relations approach of Savage (1967) to investigate flows in wedge-shaped
hoppers. They found that small density variations near the aperture could
produce flow rates lower than those obtained when the bulk was assumed to
be incompressible. The analysis showed that roughening the walls could
increase the flow rate at large hopper half-angles and this prediction was
confirmed by experiments performed to investigate the effect. Savage and
Sayed (1981) more recently extended the method of integral relations
approach to solve for the stress fields and flow rates in rough-walled conical
hoppers.
Thorpe (1984) presented a somewhat simpler analysis to account for the
effect of wall friction, by relaxing the assumption of zero shear stress within
the flowing material. He incorporated momentum terms which take into
 Discharge rates from bins and hoppers 343

account the radial variation of particle velocities into Walker's (1966) stress
analysis for the conical hopper, as discussed in section 8.5 below. An equation
similar to that found by Davidson and Nedderman (1973) results, but the
constant K is replaced by a function of internal and wall angles of friction and
also the hopper half-angle u. The resulting predictions of the mass flow rate
are only slightly smaller than those of Davidson and Nedderman (1973).

8.4.4 Effect of interstitial pressure gradients

The internal stresses in a bunker containing a flowing material increase from
zero on a surcharge-free top surface to some maximum value at greater
depths and subsequently fall to zero again at the so-called 'free-fall arch'.
Since the material is believed to be shearing throughout, its bulk density will
be free to follow the stresses and the interstitial voidage will pass through a
minimum part of the way down the bunker. Thus, near the top, the material is
being compressed and air will be expelled, whereas somewhat above the
orifice, air will be drawn into the dilating material. The percolation of the air
will give rise to interstitial pressure gradients and it can be seen from the
above arguments that positive pressures will develop in the upper part of the
hopper whereas the pressure in the interstitial fluid will be below atmospheric
immediately above the orifice. Thus the material approaching the orifice will
be subjected to an adverse pressure gradient.
Besides the pressure gradients set up by the compressibility of the granular
material (such as in fine powder flow), pressure gradients can be deliberately
imposed by supplying air to the material within the hopper; in this case, the
discharge is described as air-augmented.
The pressure gradient is related to the superficial relative velocity Vr by
Darcy's law
(8.56)
where kp is known as the permeability of the bulk material. At low Reynolds
numbers, the Carman-Kozeny equation holds (see section 2.3.2) and for
spherical particles

(8.57)

For higher Reynolds numbers, the Ergun equation should be used.

Alternatively, the pressure gradient can be related to the relative velocity by
means of a drag coefficient CD defined by

2F
CD = --- (8.58)
ApQsgv;
344 Storage and discharge of particulate bulk solids

where Co is obtained from a correlation of the form CD = f(Re). A force

balance then gives

3 (1 - £)
grad p= 4 --d- QsgCov; (8.59)

The Carman-Kozeny equation indicates that

Vr
grad pex - (8.60)
cf

and the same result can be obtained from equation (8.59) at low Reynolds
numbers where CD ex 1/Re.
It is clear from equation (8.60) that interstitial pressure effects will be more
significant with fine powders. Moreover, in a conical hopper, both the gas and
solid velocities will vary as 1/,2 and therefore the pressure gradient will be
greatest near the orifice. This, coupled with the result of the previous section,
where it was shown that it is the features within the immediate vicinity of the
orifice that control the flow, suggests that any analysis of the effect of
interstitial pressures must concentrate on the behaviour near the orifice.
The adverse pressure gradients in the vicinity of the orifice are believed to
have a direct effect on the discharge rate and are claimed by Crewdson et al.
(1977) to be the cause of the well-known reduction of flow rate with
decreasing particle size.
By making sweeping assumptions about the voidage distribution within the
material and by assuming that the Carman-Kozeny equation was applicable,
Crewdson et ai. (1977) were able to obtain analytical expressions for the
interstitial pressure profile and the discharge rate from a conical hopper.
These showed substantial agreement with experiment. Spink (1976), working
with a wedge-shaped hopper, considered the problem in more detail and also
measured the voidage profile by a capacitance method.
Crewdson et al. (1977) simply postulate that there is an extra body force
equal to the pressure gradient at the orifice (dpldr)lr and therefore modified
the Beverloo correlation to the form 0

(8.61)

Harrison and Mushin (1979) incorporated a pressure gradient term of the

order of 1/,2 into Davidson and Nedderman's 'hour-glass theory' and
suggested that there should be an extra constant in the equation,
 Discharge rates from bins and hoppers 345

W = CQb [ g + -C' ~
P ] y,
(Do - kd)'/2 (8.62)
Qb dr ro

where they predicted that

2K- 3
C' = ---
2K-1

For most granular materials, K is large and C' does not differ greatly from
unity. In contrast, Carleton (1972) does not consider the relative motion
caused by the changing void age but adopts the alternative assumption of a
stationary fluid medium and uses the well-known relationship between
Reynolds number and drag coefficient, as outlined above, to deduce the
retardation of fine powders. Holland et al. (1969) used a similar approach to
modify the Minimum Energy Theorem of Brown and Richards (1965) by
incorporating such a fluid drag term in the velocity profile across the orifice.
They showed that

U2 = 2.4 [g cose + 1 -dP ] (Do - kd)sinj3 (8.63)

p Qs(1 - E) dr

where Up is the particle velocity and the pressure gradient term is given by
equation (8.56) above. This approach also seems to predict trends which are
in qualitative agreement with experimental discharge rate measurements.
While the argument of Crewdson et al. (1977) that the flow rate depends on
the pressure gradient near the orifice seems well founded, this result is rarely
directly useful in practice since it is most unlikely that the pressure gradient
can be measured. A much more common approach has been to relate the flow
rate to some pressure difference. This is measured between tappings just
below the orifice and at some arbitrarily selected point on the hopper wall.
Clearly any resulting theory or correlation will depend on the position of the
tappings but, fortunately, in many cases the bulk of the pressure change takes
place near the orifice and the precise position of the upper tapping is
relatively unimportant. This approach has been adopted by Bulsara et al.
(1964), McDougall and Evans (1966), McDougall and Knowles (1969),
Resnick et al. (1966) and many others. Commonly this method is coupled with
the use of Bernoulli's equation, either explicitly or implicitly (by the use of an
energy balance) and inevitably gives rise to a relationship of the form
W cc (/).p Y2 ).
Bulsara et al. give
W= C D (Qb/).P)Y2 (Do - kd? (8.64)
a result that yields a good correlation of experimental results for large values
of /)'P but incorrectly suggests that W tends to zero as /)'P ~ O.
346 Storage and discharge of particulate bulk solids

McDougall and Knowles (1969) corrected this by the introduction of a

constant I1P0 which represents the adverse pressure difference to stop the
flow. They give the result
W = K 1(I1P 0 + I1P)Vz D 1z
5
0 (8.65)
Resnick (1972) rightly pointed out that when I1P = 0, one simply has
gravitational flow and therefore KlI1P~ is nothing more than the group
CQbg'/ of the Beverloo correlation. His form of the relationship can be
2

written as
(8.66)
where K2 and K3 are empirical functions of the material properties.
Leung et ai. (1978), on the other hand, do not add the driving forces but
add the flow rates resulting from each separately. They therefore obtained a
relationship of the form
W= [C1Qb(gDo ) '/2 + C2 11PVz] Do2 (8.67)
Though different in principle from Resnick's form, this equation gives very
similar results for the limiting cases when I1P is either small or large. The
experimental results of Yuasa and Kuno (1972), however, suggest that the
Resnick/McDougall approach is the better of the two.
Resnick's result, equation (8.66), is very similar in form to that obtained by
Crewdson et at. (1977), equation (8.61), and all that is required to reconcile
the two approaches is a satisfactory relationship between the pressure
gradient at the orifice and the overall pressure difference. For a conical
hopper, the analysis is trivial provided one can assume that the Carman-
Kozeny equation holds. The velocities of both gas and particles are
proportional to l/r and on integration it can be seen that

dPI
-
dr Yo
[1 1]
- I1P -
r0
- -
r1
(8.68)

where ro is the radius from the virtual apex to the orifice and r1 is the radial
coordinate of the upper tapping, as shown in Figure 8.20. Normally r1 » ro
so that (dp/dr)lr = (I1P/ro) and the equation of Crewdson et at. becomes

(8.69)

However, if the relative velocity is large, inertial effects cannot be neglected

and Nedderman et ai. (1983), starting with the Ergun equation (see section
2.3.2), showed that
 Discharge rates from bins and hoppers 347

i1P
dPI f(Re) (8.70)
dr r"

where f(Re) ~ 1 at low gas Reynolds numbers and f(Re) ~ 3 as Re ~ 00

According to equation (8.69), there is a linear relationship between W2 and
i1P in a conical hopper which allows for the air-augmented flow rate of
powder from a pressurised vessel to be expressed in terms of a multiplicative
factor

(8.71)

where Wo is the gravity-induced mass discharge rate given by

Wo = CQbg'h(Do - kd)5h (8.72)
With fine powders discharging from a closed-top conical hopper, the
reduction in mass discharge rate due to airdrag resulting from the substitution
of the Carman-Kozeny equation (equation (8.57» into the equation of
Crewdson et ai. (1977) (equation (8.61» results in

Material: Glass Ballotini

Hopper orifice diameter: 86mm
Hopper half-angle: 30'
Minimum fluidisation voidage: 0.45
32

28

~
0 1
:'A 24 '--
Oil .-----:::--::". 1. Beverloo e/ ars correlation (1961)
C 2/0 - - 0"'" -3. Crewdson e/ ars theory (1977)
t;;--
--
'1 20
0 B :----

I
o Open
><
Ei:
" 16
___ - - ia
I
"tij
~

"~oS 12
* ,/ ~Carrnan-KOZeny's packed bed model
..c:
~
"6 8
I B/ . .o~
3/ (P~",0 o D p~latm
p~O.latm

4
/ p~ Total pressure
/
0
0' 0.3
Average Particle Size a(mm)
Figure 8_21 Variation of discharge rate with average particle size in a conical hopper.
348 Storage and discharge of particulate bulk solids

W2= W 2
o
[ 1 -C"W]
--
Qbgro
(8.73)

where
180!lg(1 - E)2 [2K - 3]
C" = -2-j[-~-Q-scf2-(-1---c-o-sa-)-E3- 2K - 1
The correction factor C" is inversely proportional to the square of particle
size, which for discharge in air results in significant reduction in discharge rate
when the particle size is significantly smaller than 500 !lm (see Figure 8.21).
With a cylindrical bin, the relationship between the orifice pressure
gradient dp/drl ro and the pressure drop I:!..P within the bin requires the
numerical solution of Laplace's equation; hence no simple analytical forms
are available for the prediction of interstitial air effects on bulk solids
discharge rates in flat bottom containers.

8.5 STRESS DISTRIBUTIONS IN BINS AND

HOPPERS

8.5.1 Introduction
Design and construction of bulk solids storage and handling equipment
require a knowledge of the stress distributions prevailing in static and moving
beds of granular material. Structural stability of the storage vessels can only
be ensured if the limiting values of the stresses on vessel walls are predicted
correctly. Furthermore, the changes that occur in bulk stress distributions at
the onset of discharge from a storage vessel can in many cases impose further
constraints with regard to the choice of critical vessel dimensions such as
vessel height-to-diameter ratio, shape and size of vessel orifice, and hopper
half-angle. Frequently, flow-promoting inserts are placed inside storage
vessels to facilitate easy discharge and prevent blockages. Such inserts have to
be designed so that they can withstand the stresses exerted by the material
without any structural failure. Shapes and relative sizes of these inserts and
their positioning inside the vessels are quite critical in achieving the desired
modifications in the bulk stress distributions and the material flow behaviour.
Bulk stress distributions also affect the magnitude of wall friction between the
material bed and vessel walls as well as the friction along the external surfaces
of the flow-promoting inserts. Wall friction is in many cases responsible for
the abrasion and wear of the mechanical surfaces and in some cases the
attrition and size degradation of the particulate material. The coupling of
storage vessels with mechanical conveying equipment such as belt, screw or
bucket type feeders will also require reliable predictions of static and moving
 Stress distnbutions in bins and hoppers 349

bed loads so that feeder dimensions and power requirement can be

optimised.
The bulk stresses in a bed of granular material are in general decided by
(i) the physical properties of the container, (ii) the physical properties of the
particles such as solid density, size and shape, surface roughness and surface
moisture and also (iii) the state of flow (i.e. the strain history) of the material
bed. However, there is currently no general theoretical understanding of the
interactions between these three sets of parameters. Most of the theoretical
work to date has concentrated on the prediction of the limiting load levels on
bulk solids containers so that structural stability can be guaranteed during
unit operations. This is the basis of the civil and structural engineering
approach, which relies on incremental force balances over an element of the
container of a given geometry. By making certain assumptions about the
variation of the bulk stresses across the vessel cross-section, it is then possible
to calculate the stresses on vessel walls as a function of the coefficients of
friction of the granular materials and the wall surface. Such calculations are
known as 'approximate solutions' of the bulk stress distributions and are
generally sufficient for the purpose of the structural design of the bulk solids
containers. It is also possible to arrive at exact solutions for stress
distributions by considering the yield (failure) behaviour of the material as in
a shear tester of the type described in Chapter 4, and in turn substituting
directly the resulting failure criterion into the appropriate equations of stress
equilibrium in two or three dimensions. The resulting equations are then
solved numerically subject to the appropriate boundary conditions using one
of the finite difference or finite-element methods available. In the soil
mechanics literature, a very popular method is known as the method of
characteristics or Sokolovskii's method, named after the pioneering work of
Sokolovskii (1965) on the extensive application of the method in solving a
wide range of boundary value problems. Calculations relating to more
complicated physical boundary conditions including those which involve
three-dimensional analyses are beyond the scope of this book. The reader
should therefore refer to the publications cited in sections 8.5.2 and 8.5.3 for
a more detailed treatment of the subject.
The analyses described in sections 8.5.2 and 8.5.3 are quasi-static in nature
in that they only consider the limiting values of the bulk stresses prior to
prolonged shear, i.e. prior to sustained flow. Prediction of the stress fields
corresponding to the steady-state flow fields observed in chemical and process
engineering applications would require theoretical coupling of the particle
velocities with the prevailing stress field. This is done by evoking certain
constitutive relationships known as flow rules, which relate the bulk stresses
to the bulk strain rates and hence to particle velocities. Such theoretical
analyses are, however, far too involved to be included in this book. The
reader is referred to Drescher (1991) for a detailed account of solid stress and
velocity coupling in granular flow.
350 Storage and discharge of particulate bulk solids

8.5.2 Approximate analysis: the method of differential slices

The method of differential slices was first proposed by Janssen in 1895 and is
still commonly referred to as Janssen's method, though it has been extended
beyond recognition since his time. Like Coulomb's method of wedges (see
Drescher, 1991) this method attempts to provide a prediction of the stress
distribution by means of an approximate analysis. In the method of wedges,
the shape of the failure surface is assumed; the method of differential slices,
on the other hand, makes assumptions about the distribution of stress on
horizontal planes. It is shown by Nedderman (1992), for example, that one of
the characteristic features of the full set of stress equations is that information
is transmitted only in certain directions, and as a result, stress distributions of
considerable complexity can arise. Such distributions are unlikely to be
guessed correctly ab initio and consequently predictions based upon assumed
stress distributions are always suspect. There do, however, exist some simple
situations in which realistic assumptions of the stress distribution on
horizontal planes can be made and for such situations the method of
differential slices is the simplest and clearest method of predicting the
variation of stress with vertical position. The success of the method in these
situations must not be taken as evidence of its general applicability and great
caution must always be taken when applying it to a new situation.
Like all other stress analyses mentioned in this book, the method of
differential slices can only be used to investigate certain limiting situations,
namely when slip is about to occur. Such situations are commonly referred to
as the active and passive states and have been considered in section 4.8; they
will be considered further below. These situations will only occur if
considerable strain has taken place but may normally be regarded as limiting
states between which the actual stress distribution will lie.

8.5.2.1 Basic Janssen analysis for a cylindrical bunker. Consider a vertical

cylindrical bunker of diameter D filled with a cohesionless material with an
angle of internal friction, $, and weight density, y. Take axes z, vertically
downwards and r horizontally from the axis of symmetry, as shown in Figure
8.22.
The basic Janssen analysis is confined to cohesionless Coulomb materials
and two asumptions are made about the stress distribution:
(i) The stresses on horizontal, and hence on vertical surfaces are assumed to
be principal stresses.
(ii) The stress is assumed to be uniform across any horizontal plane. Thus all
stresses are assumed to be independent of r and hence a function of z
only.
These two assumptions are relaxed in later versions of the method as
explained in subsequent sections of this chapter. They are, however, only
 Stress distributions in bins and hoppers 357

~-------- D --------~

Figure 8.22 Schematic description of Janssen's method of differential slices.

replaced by more realistic assumptions; at no stage does it become possible to

avoid making assumptions of this type.
As a result of these assumptions, a force balance can be made on the
differential element of thickness bz as shown in Figure 8.23.
rr
The force on the top surface =--D2 azz
4
rrD2
The force on the bottom surface = - - -
4

The force on the periphery = rrD-cwbz

rrD2
The weight of the contents = - - ybz
4
Hence by addition

rrD2 dazz rrD2

-- -- + rrD-cw = --y (8.74)
4 dz 4
352 Storage and discharge of particulate bulk solids

cr zz + d crzz 8 z
dz

I+---------D --------~

Figure 8.23 Differential force balance on a cylindrical element.

or
dozz 4'tw
--+--=y (8.75)
dz D
Defining the Janssen constant K as the ratio of the principal stresses (refer to
Mohr's stress circle in Figure 8.24),
Orr OIl Or
K = -- = -- or = -- (8.76)

Finally, combining these relationships, we find that

dozz 4!lwK
--- + - - Ozz =y (8.77)
dz D
This step has involved the use of assumption (ii), that the stress is constant
over any horizontal plane, and the force is therefore simply the product of the
stress and the area.
Since the analysis considers only cohesionless materials, the wall shear
stress, 'tw, is related to the normal stress on the wall (orr)W by
 Stress distributions in bins and hoppers 353

Figure 8.24 Mohr's circle of stress depicting the major and the minor principal
compressive stresses. YL = yield locus.

(8.78)
This is inconsistent with assumption (i) which specifies Orr to be a principal
stress and is considered in more detail elsewhere (Drescher, 1991; Nedder-
man, 1992).
Equation (8.77) is a first-order differential equation of standard form, with
solution

Ozz = --
yD
+A ( 4!-lWKZ)
exp - - - - (8.79)
4!-lwK D
where A is an arbitrary constant.
If the top surface of the fill is stress-free, we can use the boundary condition
Ozz = 0 on Z = 0
giving

Ozz = ~(1
4!-lwK
_ ex p (-4!-lW KZ )]
D
(8.80)

and substitution in equations (8.76) and (8.77) yields

Orr = -yD- [ 1 - exp (-4!-lWKZ)] (8.81)

4!-lw D

Tw = yD [
-4- 1 - exp
(-4!-lWKZ)]
D (8.82)
354 Storage and discharge of particulate bulk solids

At great depth below the material top surface, i.e. z ~ 00,

(8.83)

yD
(8.84)
00
au

00 yD
Tw=-- (8.85)
4
This last result, equation (8.85), can be derived solely by a force balance and
the assumption that a steady-state stress situation is approached. It has
general validity, not only for granular materials but also for fluids in motion.
The apparent anomaly of hydrostatics results solely from the rate of approach
to this asymptote being infinitely slow. Equation (8.85), which can be derived
from equation (8.84) and the Coulomb failure criterion, can be taken as
correct for cohesionless granular materials, but equation (8.83), involving the
Janssen constant K, has less general validity.
Referring to the Mohr's circle construction in Figure 8.24, there are two
limiting values of the stress ratio,

1 - sin<l>
KA =----
1 + sin <I>
and
1 + sin<l>
Kp=----
1 - sin <I>
Thus in the passive state the value of K is the reciprocal of its value in the
active state. It is therefore convenient to differentiate between these values of
the Janssen constant with the symbols KA in the active state and Kp in the
passive state. However, since most equations involving the Janssen constant
are equally applicable for active and passive states, the undifferentiated
symbol K is used above, and this takes the values Ka and Kp as appropriate.
For typical materials, i.e. 30° < <I> < 50°, the constant Ka lies in the range
0.33-0.13 and the corresponding values of Kp are therefore in the range
3-7.5. It will be readily seen that the analysis leading to equations (8.80) to
(8.85) is equally applicable for both the active and the passive states provided
the appropriate value of K is used. Thus the same asymptotic values of Orr
and Tw are obtained but the asymptotic value of Ozz is changed by a factor of
K2. The rate of approach to these asymptotes is seen to be a function of K and
is therefore more rapid in the passive state. Figure 8.25 shows the variation of
 Stress distributions in bins and hoppers 355

'tw

,---------------------------------------------r------

0.50 50 z

Figure 8.25 Variation of wall shear stress with depth in a cylindrical bunker.

the wall shear stress with depth for a material with <I> = 30° and Ilw = 0 for
both states. It is seen that the stress reaches 90% of its asymptotic value at a
depth of 4.8 bin diameters in the active state and of only 0.53 diameters in the
passive state. It is seen that even in the active state the approach to the
asymptote is comparatively rapid and that the asymptotic stresses will occur
for the larger part of all but the shallowest bunkers.

8.5.2.2 Effect of a surcharge. If the top surface is subjected to a uniform

externally applied stress Qm the boundary condition for Janssen's differential
equation becomes
Ozz = Qo at z =0
and hence equation (8.80) takes the form

= -yD
- - [ 1 - exp (4!lWKZ)]
- --- + Qo exp [4!lWKZ]
- --- (8.86)
OZZ

4!lwKz D D
This differs from equation (8.80) solely by the addition of the extra term in Qo
which, since it contains the exponential factor, will die away with increasing
depth. Thus the surcharge only affects the stress distribution in a compar-
atively narrow region near the top of a bunker. This is an important result for
the understanding of the flow of powders from an orifice in the base of a
hopper. Unless the hopper is very shallow, the stress state near the orifice is
356 Storage and discharge of particulate bulk solids

independent of any surcharge and hence the flow rate is unaffected by

externally applied forces.
Figure 8.26 shows the type of stress distribution predicted for various
values of Qo and it can be seen that for the particular case of
yD
Qo= - - -
4flwK

the stress is constant throughout the hopper.

Since the Janssen analysis assumes that all stresses are independent of
radial position, no predictions can be made about the effect of a non-uniform
surcharge, though in these cases Qo is often taken to be the average applied
stress.

Z
Figure 8.26 Effect of surcharge on stress distribution below the material top surface.

8.5.3 More realistic analysis of the stresses at the wall

The analysis of the section above is based on the inconsistent assumptions
that (a) the stresses on vertical and horizontal planes are principal stresses
and (b) that the shear stress on the vertical wall is non-zero.

8.5.3.1 Janssen-Walker analysis. The first of the two assumptions of

Janssen's theory was relaxed by Walker (1966), who calculated the magnitude
of the wall stresses by taKing into account a finite value of the wall friction
angle <Pw, which results in redefinition of Janssen's stress ratio K as:
(orr)w
Kw= - - - (8.87)
(ozz)w
 Stress distributions in bins and hoppers 357

where (ozz)w and (orr)W are the vertical and horizontal components of the
compressive stress on the wall. Referring to the Mohr's stress circle
construction in Figure 8.27 we get

1 - sin<p cos( CD - <Pw)

K WA = (8.88)
1 + sin<p cos( CD - <Pw)
in the active stress state (corresponding to the static fill condition) and
1 + sin<p cos( CD + <Pw)
Kwp = (8.89)
1 - sin<p cos( CD + <Pw)

in the passive stress state, which corresponds to the flowing (discharge)

condition. Here the angle
e= (CD ± <Pw)/2 (8.90)
defines the angle between the wall stress and the major principal compressive
stress. The case of e = 0 corresponds to Janssen's original theory. Here
CD = sin~l(sin<pw/sin<p) (refer to Figure 8.27).
With Walker's modification of Janssen's theory, equation (8.80) above
becomes

(ozz)w = -yD - - [ 1 - exp (-4f,lWKwZ)] (8.91a)

4f,lwKw D

(J

Figure 8.27 Mohr's stress circle depicting 'active' and 'passive' failure states at the
vessel wall. MYL = material yield locus; WYL = wall yield locus.
358 Storage and discharge of particulate bulk solids

and
(8.91b)

(8.91c)

8.5.3.2 Walker's distribution factor. An attempt was made by Walker

(1966) to resolve the difficulty of knowing which value of K was required by
introducing the concept of a distribution factor. Though an improvement on
earlier versions of the Janssen analysis, Walker's method still involves
assumptions that can lead at times to serious errors (see Nedderman, 1992).
Walker notes that equation (8.91) is absolutely correct for cylindrical
bunkers provided Ozz is taken as its mean value over the whole cross-section.
The Janssen constant, evaluated at the wall, correctly relates 'tw to the value
Ozz at the wall, and Walker therefore introduces his distribution factor!iJ
which is the ratio of (ozz)w to ozz, the mean value of Ozz.
Thus by definition,

!iJ = (ozz)w
(8.92)

Equation (8.77) therefore takes the form,

da zz
--+ (8.93)
dz
and is correct.
The solution,

(8.94)

however, presupposes that !iJ is not a function of z. This is true only when the
stresses are no longer varying with depth, i.e. at great depth. Detailed stress
analysis shows that !iJ is far from constant, and this is the main deficiency in
Walker's analysis. Nevertheless, Walker's method predicts the wall stresses
adequately in many cases and is to be recommended if a rapid estimate is
required. Figure 8.28 gives a plot of the great depth values of the distribution
factor!iJoo as a function of the internal and wall angles of friction.

8.5.4 Janssen-Walker analysis in other hopper geometries

In a two-dimensional (plane-strain) bin with vertical parallel walls equation
(8.86) is replaced by
 Stress distributions in bins and hoppers 359

3.0,-------------------------------~~=F__.

2.8

2.6

2.4

,0
~ 2.2
"3
""""iJ
..2- 2.0
II
tA
o 1.8
N
§ 1.6
:;:0
::J
.0
·c
.~ 1.4
""0
(/J
"Qj
~ 1.2
:s:
1.0

0.8

0.6

0.4L---~----~--~----~--~----~--~-----L----J
o 10 20 30 40 50 60 70 80 90
<p (0)

Figure 8.28 Great depth values of Walkers' distribution factor (1966), as a function
of the internal friction angle <p for different values of the wall friction angle <Pw in the
active and passive stress states.

(Ozz)w = ~[1
IlwKw .
- exp ( - IlwKwZ)]
a

(8.95)

where 2a is the separation distance between the parallel walls (see Tiiziin and
Nedderman, 1985 (a) and (b)). In conical (three-dimensional) and wedge-
360 Storage and discharge of particulate bulk solids

(a)

(b)

Figure 8.29 Walters' analysis of stress distribution in a conical hopper.

shaped (two-dimensional) hoppers, Walter's analysis results in an equation

of the form

(h)m yh [ (h )m-l] (8.96)

°hh = QO\h o + m _ 1 1 - \ho

where h is the vertical distance from the virtual apex, as shown in Figure
 Stress distributions in bins and hoppers 361

20r-~r------------------------------.

o m= 4.0

• m=2.0
o m= 1.5
'l'
E
z • m=0.8
C ¢ m=O.O
lfl10
~ • m = -0.5
Cii

O~~~------------~--------------~.
o 1 2
Height (m)

Figure 8.30 Wall stress profiles in conical hoppers (Nedderman, 1992).

8.29(b), and ho is the distance to the orifice plane. The value of the parameter
m is given by
+ x<J>w - 2xa)
nsin<J> sin( w
m= (8.97)
tana[1 - xsin<J> cos(w + x<J>w + 2xa)]
where x = -1 in the active case and x = + 1 in the passive case (see Figure
8.29(b)). Here n = 2 in conical hoppers and n = 1 in wedge-shaped hoppers.
Figure 8.30 shows the wall stress distribution in conical hoppers, for different
values of m, see Nedderman (1992) for a detailed discussion of the
relationship between 0hh and the stress at the wall which calculates the stress
at A Band C in Figure 8.29(a).
8.5.5 Walters' switch stress analysis
Walters (1973 (a) and (b)) presented an analysis for the transient switch stress
that occurs during flow initiation within a bunker. His argument is that the
material filled into the bunker is in an active state prior to discharge. On
initiation of flow, a dilation wave will pass upwards, separating the static
material from the dilated flowing material. Since the flowing material is being
pushed inwards towards the central orifice, a passive stress state will occur
within the flowing region.
Walters (1973a) assumes that at any moment the boundary between the
static and dynamic zones is a horizontal plane at depth Z, as seen in Figure
8.31. Using the Janssen-Walker analysis outlined above, we obtain

(8.98)
362 Storage and discharge of particulate bulk solids

T
cr

z z

1 "
,,
T''''''''' -.-.-..-.,-...-......-..........-..-
.,,~ ....... '"
,
O'rr

..
,#### I

,
, f

"
,f

: yO
;, : 4 Jlw ~iJ KwA

:~
,,, , 4 Jlw (j)
z ' yO

Figure 8.31 Schematic diagram of Walters' switch stress analysis.

as the value of the mean vertical stress at the active zone boundary where
z = Z. Below the zone boundary seen in Figure 8.31, the passive form of
equation (8.98) applies, with the surcharge Q being given by (ozz)z. Thus, in
this zone, ozz decreases from (ozz)z to its great depth asymptote (yDI
4~wKwAg). This variation is shown by the full line in Figure 8.3l.
At all points within the static zone above the boundary Orr = GzzIKwA' and
within the dynamic zone below the boundary, Orr = 0zzIKwp. Hence, we can
also plot the variation of the normal stress 0xx as shown by the dotted line in
Figure 8.31.
Walters' (1973b) analysis would, therefore, predict a so-called wall stress
jump at the zone boundary of magnitude equivalent to KwplKwA . For the
model granular material with the internal angle of friction <I> = 30°, the
magnitude of the stress jump could be as large as 9-1O-fold (refer to section
8.5.3 above). Details of Walters' analysis are far from perfect and no reliance
should be placed on the exact numerical values because horizontal switch
planes do not occur in real granular materials. However, localised large
stresses are frequently measured especially after initiation of discharge
(Tiiziin and Nedderman, 1985b). Such high values of wall stresses resulting
from the switch phenomena are also often recorded as being responsible for
industrial silo collapses, as discussed in section 8.1.
 Nomenclature 363

8.6 NOMENCLATURE
a Hopper half-width (m)
Ap Cross-sectional area of particle; constant of integration (m 2 )
b Breadth of slot outlet; orifice half-width (m)
B Kinematic constant
Co Drag coefficient
c Cohesion
C, C', C" Empirical constants
d, dp Particle diameter (m)
D Diameter (m)
Dh Hydraulic mean diameter (m)
Do Outlet diameter (m)
!if Distribution factor
gradp Pressure gradient (bar m-1)
h,H Height (m)
10 Bessel f]Jnction
kp Permeability
KA Janssen constant in the active state
Kp Janssen constant in the passive state
I Thickness (m)
p Pore pressure (Pa)
Q Volumetric flow (m 3 S-1)
Qo Surcharge (uniform applied compressive stress at the top
surface) (Nm-2)
Radial co-ordinate (m)
Reynolds number
Residence time (s)
Time/total energy content (s/Joules)
Particle velocity (m S-1)
Fluid velocity (m S-1)
Horizontal velocity (m S-1)
Vertical velocity (m S-1)
Fluid velocity (m S-1)
Centreline velocity (m S-1)
Plug-flow velocity (m S-1)
Radial velocity/relative velocity (m S-1)
Particle velocity at wall (m S-1)
Gravity-induced mass discharge rate (kg S-1)
Mass flow rate (kg S-1)
x Horizontal coordinate (m)
y Vertical coordinate (m)
z Axial coordinate (m)
Z Depth (m)
11 Po The adverse pressure difference at the hopper orifice (Nm-2 )
364 Storage and discharge of particulate bulk solids

Greek letters
0. Hopper half-angle (0)
Critical hopper half-angle for mass-flow (0)
The included half-angle of the flowing zone (0)
Empirical constant (1.5-2.0)
The inclination to the horizontal of the stagnant zone boundary
when core flow occurs (0)
Boundary layer thickness/shear zone thickness (m)
E Interstial void age/angle of slip plane CO)
Angle of internal friction CO)
Angle between the flow region boundary and the horizontal at
bin outlet CO)
<PR Angle of repose CO)
<PW Angle of wall friction CO)
y. Weight density (kg m-2 S-2)
y Shear strain rate (S-1)
A Eigenvalue
iJ, Coefficient of internal frictionlfluid viscosity (Nm-2 S-l)
iJ,g Fluid viscosity (Nm-2 S-1)
f-lw Coefficient of wall friction C)
e Polar co-ordinate (0)
Qb Bulk density (kg m-3 )
Qf Flowing bulk density (kg m-3 )
Qo Initial density (kg m-3 )
Qs Solid density (kg m-3 )
~Q Qf - Qo (kg m-3 )
a Normal stress (Nm-2)
Yield strength (Nm-2)
Mean compressive stress (Nm-2)
Normal stress at wall (Nm-2)
Angle between the major principal stress direction and the r-
direction C)
Angle between the major principal strain rate direction and the
r-direction (0)
Shear stress (Nm-2)
Shear stress at wall (Nm-2)
Stream function

REFERENCES
Arnold, P.e., McLean, A. G. and Roberts, A.W. (1982) Bulk Solids: Storage, Flow
and Handling, Tunra Research Associates, NSW, Australia.
 References 365

Australian Institute of Engineers (1983) National Code of Practice for Bulk Solids
Handling.
Beverloo, W.A., Leniger, H.A. and Van de Velde, J. (1961) Chern. Engng ScLI5, 260.
Bosley, J., Schofield, C. and Shook, e.A. (1969) Trans. Inst. Chern. Engrs. 47,147.
Bransby, P.L. and Blair-Fish, P.M. (1979) Powder Technol. 8, 197.
British Materials Handling Board (1985) Draft Code of Practice for the Design of Silos,
Bins, Bunkers and Hoppers.
Brown, R.L. and Richards, J.C. (1959) Trans. Inst. Chern. Engrs 37, 108.
Brown, R.L. and Richards, J.e. (1960) Trans. Inst. Chern. Engrs. 38, 243.
Brown, R.L. and Richards, J.C. (1965) Rheo!. Acta 4, 153.
Buchele, M.V. and Wynn, P. (1980) Chemical Engineering Tripos. Part 2. Research
Project Report, University of Cambridge.
Bulsara, P.U., Zenz, F.A. and Eckert, R.S. (1964) Ind. Engng Chern. Proc. Des.
Dev. 3,348.
Carleton, A.J. (1972) Powder Techno!. 6, 91.
Crewdson, B.J., Ormond, A.L. and Nedderman, R.M. (1977) Powder Techno!. 16,
197.
Davidson, J.F. (1981) Private communication; see also Nedderman and Laohakul
(1981).
Davidson, J.F. and Nedderman, R.M. (1973) Trans. Inst. Chern. Engrs. 51, 29.
Demming, W.E. and Mehring, A.L. (1929) Ind. Engng Chern. 29,661.
Dosekun, R. (1981) PhD Thesis. Department of Chemical Engineering, University of
Cambridge.
Drescher, A. (1991) Analytical Methods in Bin-Load Analysis. Elsevier Science,
Amsterdam.
Fowler, R.T. and Glastonbury, J.G. (1959) Chern. Engng Sci. 10,150.
Franklin, F.C. and Johanson, L.N. (1955) Chern. Engng Sci. 4, 119.
Hagen, E. (1852) Ber. Preuss. Akad. d. Wiss. 35.
Harmens, A. (1963) Chern. Engng Sci. 18,297.
Harrison, A. and Mushin, S.A. (1979) Chemical Engineering Tripos. Part 2. Research
Project Report, University of Cambridge.
Hinchley, J.W. (1926) Chernical Engineering. Encyclopaedia Britannica.
Holland, J., Miller, J.E.P. Schofield, e. and Shook, C.A. (1969) Trans. Inst. Chern.
Engrs. 47, 154.
Hosseini-Ashrafi, M.E. and Tiiziin, U. (1992) Chern. Engng Sci. 48, 158.
Janssen, H.A. (1895) Z. Ver. dt. Ing. 39, 1045.
Jenike, A.W. (1961) Gravity Flow of Bulk Solids. Utah Engineering Experimental
Station, University of Utah. Bulletin 108.
Jenike, A.W. (1964) Storage and Flow of Solids. Utah Engineering Experimental
Station, University of Utah. Bulletin 123.
Jenike, A.W. (1967) Quantitative design of mass flow bins. Powder Techno!. 1,237.
Jenike, A.W. (1991) Powder Technol. 50,229.
Jenike, A.W., Johanson, J.R. and Carson, J.W. (1973) Bin loads - part 3. Mass flow
bins. J. Engng Ind. Trans. ASME B 95,6.
Johanson, J.R. (1965) Trans. Soc. Min. Engrs AIME 232, 69.
Johanson, J.R. (1968) Powder Technol. 1,328.
Kay, J.M. and Nedderman, R.M. (1974) An Introduction to Fluid Mechanics and Heat
Transfer, Cambridge University Press, Cambridge.
Ketchum, M.S. (1929) The Design of Walls, Bins and Grain Elevators, McGraw-Hill,
New York.
Kotchanova, 1.1. (1970) Powder Techno!. 4,32.
Laird, B.W. and Roberts, P.M. (1979) Chemical Engineering Tripos. Part 2. Research
Project Report, University of Cambridge.
366 Storage and discharge of particulate bulk solids

Langmaid, R.N. and Rose, H.E. (1966) J. Inst. Fuel 30, 157.
Laohakul, C. (1978) PhD Thesis. Department of Chemical Engineering, University of
Cambridge.
Lee, S., Cowin, S.C. and Templeton, I.S. (1974) Trans. Soc. Rheology 18, 247.
Leung, L.S., Jones, P.J. and Knowlton, T.M. (1978) Powder Technol. 19,7.
Litwiniszyn, J. (1971) Symposium Franco-Polonais, Problemes des Rheologie,
Warsaw.
McCabe, RP. (1974) Geotechnique 1, 45.
McDougall, I.R and Evans, A.C. (1966) Trans. Inst. Chern. Engrs 44, 15.
McDougall, I.R. and Knowles, G.H. (1969) Trans. Inst. Chern. Engrs 47, 73.
Mullins, W.W. (1974) Powder Technol. 9, 29.
Mullins, W.W. (1979) Powder Techno!. 23,115.
Nedderman, RM. (1992) Statics and Kinematics of Granular Materials, Cambridge
University Press, Cambridge.
Nedderman, R.M. and Laohakul, C. (1981) Powder Technol. 25, 91.
Nedderman, R.M. and Tiiziin, U. (1979) Powder Techno!. 22,243.
Nedderman, R.M., Tiiziin, U. and Thorpe, RB. (1983) Powder Technol. 35,69.
Newton, R.H., Dunham, G.S. and Simpson, T.P. (1945) Trans. Am. Inst. Chern.
Engrs 41,215.
Nguyen, T.V., Brennen, C. and Sabersky, R. (1979) Mechanics Applied to the
Transport of Bulk Materials. ASME, AMD-31.
Nielsen, J. (1983a) International Conference on Bulk Materials Storage, Handling and
Transport, Newcastle, Australia.
Nielsen, J. (1983b) Proceedings of the 15th Annual Symposium of the Fine Particle
Society, Hawaii, USA.
Rausch, J.M. (1948) PhD Thesis, Princeton University.
Resnick, W. (1972) Trans. Inst. Chern. Engrs 50, 289.
Resnick, W., Heled, Y., Klein, A. and Palm, E. (1966) Ind. Engng Chern. Funds 5,
392.
Roberts, A.W. (1994) First International Particle Technology Forum, Denver, USA,
Part 3,64.
Roscoe, K.H. (1970) The influence of strains in soil mechanics. Tenth Rankine
Lecture. Geotechnique 20, 122.
Rose, H.F. and Tanaka, T. (1959) The Engineer (London) 208, October 23.
Savage, S.B. (1965) Br. J. Appl. Phys. 16, 1885.
Savage, S.B. (1967) I. J. Mech. Sci. 9, 651.
Savage, S.B. and Sayed, M. (1979) Mechanics Applied to the Transport of Bulk
Materials, ASME, AMD-31.
Savage, S.B. and Sayed, M. (1981) Z. Angew. Math. Phys. 32, 125.
Shaxby, J.H. and Evans, J.C. (1923) Trans. Faraday Soc. 19,60.
Sokolovskii, V.V. (1965) Statistics of Granular Media, Pergamon Press, Oxford.
Spink, C.D. (1976) PhD Thesis, Department of Chemical Engineering, University of
Cambridge.
Sullivan, W.N. (1972) PhD Thesis, California Institute of Technology.
Takahashi, H. and Yanai, H. (1973) Powder Technol7, 205.
Takahashi, H. and Yanai, H. (1974) Kagaku Kogaku 38, 74.
Thiemer, O.F. (1969) Trans. ASME J. Engng 1ndust. Series B, 91, 460.
Thorpe, R.B. (1984) PhD Thesis, Department of Chemical Engineering, University of
Cambridge.
Tiiziin, U. (1979) PhD Thesis, University of Cambridge.
Tiiziin, U. and Nedderman, R.M. (1982) Powder Techno!. 31,27.
Tiiziin, U. and Nedderman, RM. (1985a) Chern. Engng Sci. 40, 325.
Tiiziin, U. and Nedderman, R.M. (1985b) Chern. Engng Sci. 40,337.
 References 367

Tiiziin, U., Houslby, G.T., Nedderman, R.M. and Savage, S.B. (1982) Chern. Engng
Sci. 37, 1691.
Walker, S.M. (1966) Chern. Engng Sci. 21, 975.
Walters, J.K. (1973a) Chern. Engng Sci. 28, 13.
Walters, J.K. (1973b) Chern. Engng. Sci. 28, 779.
Weighhard, K. (1952) Ingen.-Arch. 20, 109.
Williams, J.C. (1977) Chern. Engng Sci. 32,247.
Yuasa, Y. and Kuno, H. (1972) Powder Techno/. 6,97.
Zenz, F.A. (1975) Fluidisation Technol. 2,239.
Index

Acceleration eccentrically placed orifices 336

of particle-fluid mixtures 93-4, 223-5 effect of orifice shape 335
of single particles 80-5 effect of wall inclination 335-6
'Added mass', in particle motion 79-80 free-fall arch 337, 343
Aerosol 1 'hour glass' theory 340-1, 344-5
Agglomeration 250-7 interstitial pressure gradients 334, 343-8
Angle of repose, see Friction in bulk solids 'minimum energy' theorem 337-8, 345
Archimedes number 60, 228 theoretical predictions 337-43
Archimedes' principle, see Buoyancy see also Flow in bins and hoppers
Assembly mechanics 127-32 Distrinct element (DE) method
cellular automata 184
Bassett term, in particle drag, see History term contact forces 187-90
Best number, see Archimedes number in general 179-196
Bin, see Bulk storage vessels, design Newtonian dynamics 183-4
Brownian diffusivity 79, 262-3 particle size and shape 190-3
Bulk storage vessels, design stochastic models 184
for large storage silos 301-2 stress analysis 193-6
for 'process' silos 299-30 Drag on a particle
types 298-9 for a cylinder 56, 58
Buoyancy 53-4 for a disc 58
Bunker, see Bulk storage vessels, design form drag 53-4
in general 53-6
Carman-Kozenyequation 86,262,288,343-7 for a sphere 56--8
Compressibility of bulk solids 168-71 for spheroids 58-9, 65-6
Coulomb failure, see Friction in bulk solids Newton's law 56
Creeping flow 55, 66--72 non-continuum effects 76--9
Cunningham slip correction factor 20, 77-9 particle drag coefficient 55-62
Curvilinear motion, of a particle 81-5 principal resistance to translation 67-72
Cyclones skin friction drag 53-4
collection efficiency 268-71 Stokes' law 56
pressure drop 267-8 unsteady motion 79-85
solids loading effects 272 see also Terminal velocity
Stairmand designs 264-7, 270
system design 271 Electrostatic forces
temperature and pressure effects 273 coulombic 120-2
types 263-7 diffusion charging 123
vortex finder 264 image charge 121-2
space charge 121-2
Darcy's law 87, 288, 343 Ergun's equation 87,227,343,346--7
Diffusion, see Brownian diffusivity Extinction of light, see Light scattering;
Diffusional collection, in filtration 280-3 Particle size measurement
Direct interception, collection mechanism in
filtration 280-3 Filtration
Discharge rates from bins and hoppers cake formation and filter conditioning
discharge correlations 331 285-8
370 Index

Filtration (contd): angle of friction, cohesive materials 151-7

collection efficiency in depth filtration angle of repose 147-8
277-85 angle of wall friction 147
collection mechanisms 280-3 consolidation 154
depth and surface filtration 274-7 Coulomb failure 149, 156-7
fabric or bag filter 275 critical state 154-5
filter cleaning mechanisms 289-94 critical voids ratio 153
pressure drop 288-9 measurement 146-8
retention 283-5 modelling of wall friction 174-9
Flow in bins and hoppers Warren-Spring material 153
arching 334 Friction at a single contact 132-5
choice of flow regime 306-8
comparison of radial velocity field and Galileo number 60
kinematic model solutions 324-6 Geldart's groups, see Fluidization
converging flow 313-26 Granulation, see Agglomeration
dilation wave 327
funnel flow 302-5 Hatch-Choate equations 18, 51-52
kinematic flow model 313-14, 315-26 see also Particle size distribution
mass flow 302-5 Hertz analysis, of particle contact 125-7
pipe or channel flow 306 History term, in particle drag 79-80
plug flow 308-13 Hopper, see Bulk storage vessels, design;
radial velocity field solution 313-15 Stresses in bins and hoppers
shear zone 308-13 Hydraulic conveying
stagnant or dead zones 306 comparison with pneumatic conveying
transient flow 326-30 201-5
velocity distributions 308-30 deposit velocity 207,211-12
see also Discharge rates from bins and design and operation 218-21
hoppers effect of distance 206
Flow through packed beds 86-7 flow in horizontal pipes 209-16
see also Carman-Kozeny equation; Darcy's flow in vertical pipes 216-17
law; Ergun's equation friction gradient 209
Flowability, 1enike design method 157--60 settling and non-settling slurries 206-7
Flow rules for bulk solids 167-8 slurry concentration 208, 215
Fluidization stratification ratio 214
bubble flow patterns 247-8 system characteristic 207, 218-20
bubble rise velocity 244-50 Hydrodynamic resistance, of a particle 20
bubble splitting 245--6 Hydrosol 1
bubbling beds 242-50
criteria for behavioural boundaries 236-9 Image analysis, see Particle size measurement
de aeration 233 Impact and rebound
defluidization and cohesive effects 240-2 in agglomeration 253-7
expansion of bubbling beds 250 in general 135-40
expansion without bubbling 234-5 on liquid-covered surfaces 140-2
fast-fluidisation 222, 226 Inertial collection, in filtration 280-3
Geldart's classification 229-36 Inertial impaction 83-5
minimum fluidization velocity 227-8 Inertial impactor, see Particle size
mixing due to bubbles 243-4 measurement
slugging 222, 226, 246-7 Inertial separators 263
temperature and pressure effects 239-40 see also Cycloncs
Fracture mechanics of particle compacts Interparticle forces, see Van der Waals forces;
129-32 Liquid bridges; Electrostatic forces
Fraunhofer diffraction, see Light scattering;
Particle size measurement JKR theory, for contact mechanics 126-7
Frequency distribution, see Particle size
distribution Kelvin equation 116
Friction in bulk solids Kozeny-Carman equation, see
angle of friction, cohesionless materials Carman-Kozeny equation
148-51 Knudsen number 77
 Index 377

Lambert-Beer law, see Light scattering count median diameter (CMD) 16-17,
Laplace equation 110 49-52
Light scattering diameter of the sphere of average mass 9
extinction 41-2 diameter of the sphere of average surface 9
Franhofer theory 36-41 geometric mean 7
Lambert-Beer law 41 geometric standard deviation (GSD) 16-18,
Mie theory 35-8 49-52
particle shape effects 39 mass mean diameter 10-12
refractive index 37-42 median 7
Rayleigh theory 36 mode 7
see also Particle size measurement moment averages 9-10
Liquid bridges Sauter mean 10, 279
capillary forces between spheres 110-13 surface mean diameter 10-12, 16, 279
capillary state 109 Particle size distribution
condensation and evaporation 116-20 conversion between weighted distributions
cone and plate contact 113-16 12-14, 51-2
dynamic forces 116-17 cumulative 5-9
effect of surface roughness 112-14, 119 discrete and continuous 4-5
funicular state 109 frequency distribution function 4-7
pendular state 109 frequency histogram 3-5
toroidal approximation 110 in general 3
Log-normal distribution, see Particle size log-normal distribution 16-8, 49-52
distribution log-probability axes 16-19
Log-probability axes, see Particle size mass distribution 9-15
distribution number distribution 3
surface distribution 9
Mean free path 1,77 weighted distributions 9-14
Microscopy, see Particle size measurement Particle size measurement
Mohl-Coulomb failure criterion 164 'aerodynamic particle sizer' 47
Mohr's circle of stresses 163-4 coincidence errors 39
Momentum balance, for particle-fluid 'Coulter counter' 42-4
mixtures in pipe flow 91-4 'electrozone' method 42-4
extinction meters 34, 41-2
Newton's law (for drag), see Drag on a particle Fraunhofer scattering instruments 33
Normal distribution 49 inertial impactor 45-6
methods compared 32
Particle, definition of 1 microscopy 31
Particle density 45 settling methods 45
Particle diameter sieving 31
aerodynamic diameter 21 single particle optical counters 33
equivalent diameters in general 3, 18 see also Light scattering
equivalent-projected-area circle diameter Pipe flow of particlelfluid mixtures 87-96
20 see also Pneumatic conveying; Hydraulic
equivalent-surface sphere diameter 20 conveying
equivalent-volume sphere diameter 18 Pneumatic conveying
relationship between Stokes' and choking 222
equivalent-volume diameters 69-70 comparison with hydraulic conveying
Stokes' diameter 20-1 201-5
Particle shape entry loss due to acceleration 223-5
dynamic shape factor 26 flow in horizontal pipes 223-5
fractal dimension 25-6 flow in vertical pipes 225
Heywood's approach 21-4 flow regimes and transitions 221-2
Wadell sphericity 21 saltation 221-2
Particle size, see Particle size, averages; Pyknometer, see Particle density
Particle size distribution; Particle size
measurement Resolution
Particle size, averages of optical microscope
arithmetic mean 7 of size measuring devices 3
372 Index

Restitution, coefficient, of, see Impact and Stresses in bins and hoppers
rebound active and passive states 165-7,350,354--61
Reynolds' lubrication equation 116 effect of a surcharge 355--6
Reynolds' number Janssen-Walker analysis 356--61
for a cyclone 268 method of characteristics 349
at minimum fluidization 228 method of differential slices (Janssen's
for a particle 55, 263 analysis) 350--6
Richardson-Zaki correlation 85--6, 234 Silo collapse 301-2, 362
stress and velocity coupling 349
Sampling switch stress (Walkers' analysis) 361-2
extractive 27-8 Walkers' distribution factor 358
in general 26--30 Strouhal number 75--6
in situ 27-8
isokinetic 27-30 Terminal velocity
Sauter mean, see Particle size, averages effect of temperature and pressure 63-5
Segregation 27 glide/tumble motion 75
Settling for non-spherical particles in
collection mechanism in filtration 280-3 creeping flow 65-72
of particle assemblies 85--6 for non-spherical particles at intermediate
Shape, see Particle shape Reynolds' numbers 72-5
Shear cell, see Friction, in bulk solids for non-spherical particles in the Newton's
Sieving, see Particle size measurement law range 75-6
Silo, see Bulk storage vessels, design; Stresses sideslip 76
in silos and hoppers for spheres 59-65
Size, see Particle size, averages; Particle size see also Drag on a particle
distribution; Particle size measurement
Skewness, of a distribution 7 Unsteady motion of single particles, see Drag
Slip correction factor, see Cunningham slip on a particle
correction factor
SolI Van der Waals forces
Spouting, of particles 226, 229-30, 235--6, 239 combining relations 104--7
Stability, of a suspension 1 Derjaguin's approximation 107
Stokes diameter, see Particle diameter geometrical effects 105
Stokes-Einstein equation 79 Hamaker constant 103
Stokes' law, see Drag on a particle Lifshitz theory 104
Stokes number pair potential 99-101
in acceleration and curvilinear motion 82-5 practical calculations 108-9
in agglomeration 253-7 Von Karman vortex street 58
in filtration 281 Virtual mass, see' Added mass'
in impaction 45--6
in isokinetic sampling 29 Wadell sphericity, see Particle shape
Stopping distance, for a particle 81
see also Stokes number Yield locus, see Friction in bulk solids