REVIEW

Reviews [Link]

Review on Modeling and Simulation of Continuous

Casting
Brian G. Thomas*

fully three-dimensional domain and in-

Continuous casting is a mature, sophisticated technological process, used to volve wide ranges of both length and time
produce most of the world’s steel, so is worthy of fundamentally-based scales. For example, initial solidification
computational modeling. It involves many interacting phenomena including at the meniscus, which govern surface
defects, involves a size scale much less than
heat transfer, solidification, multiphase turbulent flow, clogging, electromagnetic
one millimeter, while nucleation occurring
effects, complex interfacial behavior, particle entrapment, thermal-mechanical there is well known to affect centerline
distortion, stress, cracks, segregation, and microstructure formation. Further- segregation, that occurs more than 10 m
more, these phenomena are transient, three-dimensional, and operate over away. The time scale of turbulent eddies
wide length and time scales. This paper reviews the current state of the art in that govern the capture of particles between
dendrites is in milliseconds, while insta-
modeling these phenomena, focusing on practical applications to the formation
bilities in mold surface level caused by
of defects. It emphasizes model verification and validation of model predictions. bulging instabilities between rolls in the
The models reviewed range from fast and simple for implementation into online secondary cooling zones are known to
model-based control systems to sophisticated multiphysics simulations that evolve over several hours. An accurate
incorporate many coupled phenomena. Both the accomplishments and computational model of phenomena, such
remaining challenges are discussed. as these must find ways to bridge these
great gaps in length and time scale, and is a
daunting task.
The staggering complexity of the contin-
uous-casting process makes it impossible
1. Introduction to model all of these phenomena together at once. Thus, it is
necessary to make reasonable assumptions and to uncouple or
The continuous casting process is used in making over 96%
neglect the less-important phenomena, according to a specific
of steel products in the world, as of 2015.[1] It is a mature
purpose chosen for the model. Choosing the model purpose is
and sophisticated technological process that has undergone
thus the most critical modeling step. One major purpose of
decades of improvements, based mainly on experience with
useful models is to gain new insights into understanding some
the commercial operation, aided by physical water modeling to
aspect of the process, such as explaining the mechanism(s) of
understand the fluid flow behavior. Further advances will require
how a particular defect likely formed, and how its incidence is
better understanding of the fundamental behaviors, which is
affected by controllable process variables.
difficult with experiments, owing to the harsh environment.
Having chosen the domain and governing equations, they can
This makes continuous casting an ideal process for advanced
be discretized and solved using numerical methods, such as
computational simulation, which has improved significantly in
finite-difference for fluid flow and finite-element for stress
recent years.[2]
analysis. It is important that adequate numerical validation be
The challenge for modeling is to accurately predict something
conducted. Numerical errors commonly arise from too coarse a
of practical use to designers and practitioners, such as the
computational domain or incomplete convergence when solving
formation of defects. Unfortunately, this is very difficult,
the nonlinear equations. Solving a known test problem and
because like most real commercial processes, the continuous
conducting mesh refinement studies to a achieve grid indepen-
casting of steel is governed by an extremely complex system
dent solution is an important first step to verify the model.
of inter-related phenomena, a few of which are illustrated in
Finally, a model must be calibrated and validated against
Figure 1.[3] Furthermore, most of these phenomena require a
experimental measurements on both the laboratory and plant
scales before it can be trusted to gain quantitative understanding
Prof. B. G. Thomas of the real process from parametric studies.
Professor of Mechanical Engineering This paper reviews some of the recent advances in modeling
Colorado School of Mines, Brown Hall W370-B, steel continuous casting, which have been accomplished by
1610 Illinois Street, Golden, CO 80401
E-mail: bgthomas@[Link]
many research groups around the world. Space permits just a
few examples, which are taken mainly from slab casting with
The ORCID identification number(s) for the author(s) of this article mold slag, and from the results of my own research group, owing
can be found under [Link]
to familiarity with this work. Space also permits referencing only
DOI: 10.1002/srin.201700312 a fraction of the high-quality research that has been done on this

steel research int. 2017, 1700312 1700312 (1 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

Dr. Brian G. Thomas is Professor of

Mechanical Engineering at the
Colorado School of Mines, Golden,
CO, USA, and Director of the
Continuous Casting Center. He is also
Gauthier Professor Emeritus at the
University of Illinois, Urbana, Illinois.
His research efforts focus on
computational modeling of continuous
casting of steel and related processes.

via several different numerical techniques, which are reviewed

elsewhere.[4,5]
The most difficult task in development of an accurate
solidification heat-transfer model of continuous casting is to
accurately model heat removal from the strand surface. This is
especially an issue in the mold, where heat transfer is governed
mainly by heat conduction across the interfacial gap between the
shell and the mold wall. Heat transfer below the mold also
presents challenges, owing to the complex phenomena such as
droplet impingement and film boiling that occur at the strand
surface during spray and air-mist cooling, and from roll contact.
The simplest approach is to calibrate the model to match
plant measurements, which is both popular and effective.[6–12]
To extend the model to other casting conditions, however, a more
fundamental approach is desired as well.
Calibration of solidification heat-transfer models in the mold can
exploit measurements such as cooling-water heatup for a global
Figure 1. Fluid-flow phenomena in the mold region of a steel slab caster. energy balance, thermocouple temperatures in the mold walls for
Reproduced with permission.[3] 2002, TMS. the heat-flux profile, and shell thickness measurements from
breakout shells or composition profiles after trace-metal injection.[6–
8,13–16]
very large and popular subject. An emphasis is made on model This can be accomplished using trial and error[9,13,17,18] or an
validation, because many recent models combine together inverse model,[8,9,19] with either a full 3D model[7,9,15–17,19–21] or an
several complex phenomena for the first time, and comparison accurate reduced-order model of the mold wall geometry.[13,14,22,23]
with actual plant measurements is the only way to check if these An example is shown in Figure 2, where interfacial parameters in the
complex models are reasonable. Every aspect of every model model were calibrated to match mold temperatures, and then
should be both verified with known solutions and validated with comparison with the breakout shell profile represents model
measurements in order to trust the predictions of computational validation.[13] Note that the prediction for steady-state conditions
models and suggested improvements arising from modeling differs somewhat from the profile predicted under the transient
studies. Such efforts are worthwhile, because plant experiments conditions, when the breakout occurred, which naturally matches
aiming to improve this mature process take a long time to better with the measurements.[13]
evaluate, and at significant cost, especially if they are wrong. Model validation with plant measurements below the mold is
more difficult. Surface temperature can be measured in the
spray zones, but pyrometer readings are lowered by steam and
surface oxide scale formation. Shell thickness typically involves
2. Heat Transfer and Solidification indirect observations, such as tracer injection,[6] strand thermo-
Continuous casting is fundamentally a heat-transfer process, couples,[22] relative roll force,[24] roll movement,[24] internal
removing the superheat, latent heat, and some sensible heat hot-tears induced by inserting shims between the strand and
from the molten metal to produce a solid strand with a desired the rolls,[14] the melted tips of nails shot into the strand,[25] or
cross-sectional shape and size. Models to predict temperature the serious defect of whale formation, which occurs when the
distribution and growth of the solidifying steel shell are needed metallurgical length exceeds the supported length of the
for basic design, trouble shooting, and control of the process. strand.[10,24] For example, the shell thickness predictions in
They are also a crucial building block for many advanced Figure 3 were validated indirectly by the model agreement
computational models, which couple heat transport together with both roll movement measurements,[24] and with many
with other phenomena. These models solve the transient plant instances with and without whale formation.[10] The
heat conduction equation, where latent heat may be introduced surface temperature predictions only agree with the pyrometer

steel research int. 2017, 1700312 1700312 (2 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

measurements lower in the caster and during reheating after

caster exit.[10] When accurate pyrometer measurements are
available, an inverse model can be applied to match the predicted
surface temperature profile with the peaks of those measure-
ments, to overcome dips caused by intermittent scale.[12] Finally,
the shell thickness predictions near final solidification of several
models have been successfully validated with nail shooting
measurements.[12,25–28]
Once validated, simple heat transfer solidification models
have a wide range of powerful practical applications, which have
had great impact on improving the process in commercial
practice. For example, post-mortem analysis of the location and
cause of internal hot-tear cracks can be inferred from model
predictions of the location of the critical temperature range near
the solidification front. Another important example is determi-
nation of the solid fraction profile near final solidification.
In addition to finding the maximum casting speed to avoid
whale formation,[24] such models can be used to help choose
the best position of soft reduction equipment[27,29] or final
electro-magnetic stirring equipment.[12,25,26,28] for the mitigation
of centerline defects.

2.1. Interfacial Heat Transfer

Many detailed models have been developed to predict heat

transfer across the interfacial gap between the strand and the
mold wall in a fundamental manner.[13,15–17,19,22,30–37] Interfacial
heat transfer depends on the thermal conductivity of the
interfacial layers and the size of the gap, which requires both
detailed models of the gap and measurements. Lab measure-
ments of mold slag thermal conductivity must be interpreted
using models to extract the property data from the raw
measurements. Both thermal conductivity and slag viscosity
Figure 2. a) Temperature in the mold wall, comparing predictions with depend greatly on the extent of crystallization, which is a
thermocouple measurements.[13] b) Solidified shell thickness in the mold, complex process that can be characterized as C-shaped
comparing predictions (lines), and breakout shell measurements (symbols). curves in time–temperature–transformation diagrams.[13,32]
Reproduced with permission.[13] 2006, The Iron and Steel Institute of Japan. This means that the slag properties properly require more

Figure 3. Strand surface-temperature predictions compared with pyrometer measurements and shell thickness profile. Reproduced with permission.[10]
2011, TMS.

steel research int. 2017, 1700312 1700312 (3 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

than simple temperature-dependent functions, which complicates

gap models.
Contact resistance between the solidified, typically de-vitrified,
slag layer, and the mold wall depends on the roughness of the
solid slag surface, and is greatly affected by slag composition. One
approach to model this challenging problem is to consider that the
slag against the mold wall should continue to flow and flatten until
it either crystallizes or its viscosity becomes extremely high, as
characterized by its cooling below the glass transition tempera-
ture.[13] Surface roughness is then defined by the distance from the
mold wall to the isotherm in the slag layer corresponding to that
transition temperature.[13]
Over most of the strand surface, the gap thickness depends
on the balance between liquid slag infiltration at the meniscus
due to mold oscillation and the consumption of the slag as it
is dragged downward by the strand at the casting speed, while
being squeezed by the ferrostatic pressure from the internal
liquid pool transmitted through the shell. Thus, a model of
thermal-flow and mass conservation of the slag is useful for
tracking the thickness of the liquid and solid slag layers along the
gap,[13,15,22,32,37] and can be validated with measured slag
films.[13,37] The oscillation marks greatly affect the gap and
consume a significant amount of slag, so deeper oscillation
marks or surface depressions have higher slag consumption,
and should be taken into account as well.[22,38,39]
Toward the corners of the strand in the mold, the ferrostatic
pressure is unable to overcome shrinkage of the strand to
maintain good contact between the shell and the mold. Thus,
when the corner regions are important, a two-way coupled
thermo-mechanical model can be combined with the interfacial
gap model to determine the size of the gap as it varies in space Figure 4. Model predictions of dynamic surface temperature, metallur-
and time. Such models should also consider taper of the mold gical length, and thermal shrinkage, compared with measured roll loads,
walls and thermal distortion of both the mold and shell, as during a dip in casting speed. Reproduced with permission.[24] 2015,
discussed later. AIST.

2.2. Online Control temperature histories predicted by the model, which are shown
in the remaining three frames of this figure. While the strand
Fundamentally-based computational models of heat transfer and has a liquid core, the internal liquid ferrostatic pressure
solidification are now sufficiently accurate and efficient, that they generates high loads in the support rolls. During the casting
are being used as part of online control systems.[10,24,26,40–42] speed dip, frame 3 shows that the metallurgical length decreases,
For typical commercial casting speeds, the advection of heat by so that the strand beneath all four rolls is completely solid, and
the moving strand is much larger than heat conduction in generates low measured loads.
the casting direction.[22] This enables Lagrangian thermal- The predicted TLE, which is reported relative to zero, when
solidification models of a horizontal slice moving down through liquid is present in the core, has the same qualitative behavior.
the strand, which are efficient enough to run in real time.[10,24] More remarkable is the agreement between the predicted
While validation of steady-state thermal models has become dynamic transition times and the measurements. Both the
standard practice in development of most models of continuous measured loads and the predicted TLE signals experience a
casting of steel, validation under transient conditions is rare. gradual decline, with a short delay after the initial drop in speed,
Figure 4 shows an example comparing indirect plant measure- which begin in sequence from rolls. Then, both signals slowly
ments of metallurgical length under transient conditions increase while the speed stays low. Finally, a relatively fast rise in
with predictions of a dynamic thermal solidification model in both signals follows in the opposite order of rolls from 75 to 79,
a conventional thick slab caster.[24] The top frame shows the and with a much longer delay after the final speed rise. As the
casting speed history, where attention is focused on a dip in model was calibrated only for steady-state conditions,
casting speed. The second frame shows strain gauge measure- this quantitative match with the measured transition times
ments taken in four rolls located near the metallurgical length under transient conditions indicates model validation.[24]
during this time. The speed drop changes the total thermal Advanced thermal-solidification models are also used online
linear expansion (TLE), or average shrinkage strain through for other control purposes, such as determining the solid
the strand thickness, the metallurgical length, and the surface fraction profile near final solidification, for positioning of

steel research int. 2017, 1700312 1700312 (4 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

dynamic soft reduction.[29] The use of fundamentally-based injection on turbulent flow inside the nozzle,[62–64] and also in
models in control systems will continue to grow. the mold.[58,60,61,66,69–88] These models depend on the bubble
size distribution, which is very difficult to determine, even
experimentally, as water models are unreliable owing to their
3. Fluid Flow Modeling property differences compared with molten metals. Models
show that argon bubbles in steel are larger than air bubbles in
Many investigations of fluid flow in the steel continuous casting
water for similar conditions, for example.[85,89] Computational
nozzle and mold have been conducted using three-dimensional
approaches to handle this problem are discussed later.
computational fluid dynamics (CFD) models, and has been the
subject of previous reviews.[43,44] These models solve the
continuity equation for conserving mass, and the Navier–Stokes
3.2. Nozzle Clogging
equations for conserving the momentum of incompressible
Newtonian fluids, such as steel, using classical finite-volume, or
The attachment and buildup of inclusion particles to the
finite-difference methods, often with commercial CFD software,
refractory walls of the nozzle leads to nozzle clogging. In
such as the ANSYS-FLUENT code.[45] Results in the form of
addition to lowering productivity, clogging is detrimental to steel
quantitative versions of Figure 1 are numerous, and include
quality due to the releasing of clogs, which generates both large
many realistic animations of the turbulent flow.
inclusions and unstable flow in the mold, leading to excessive
The most popular numerical approach is to assume steady-
surface level fluctuations.[90] Nozzle clogging has been investi-
state, single-phase flow, using a Reynold’s Averaged Navier–
gated using models to simulate the inclusion attachment process
Stokes (RANS) method, together with a turbulence model such
by a few researchers.[64,90–92] In these models, particle transport
as k-e, or k-ω SST, which is designed to estimate the average effect
and attachment can be modeled using a fluid flow model for
of turbulence,[43] based on solving extra transport equations
the velocity and pressure fields, combined with a Lagrangian
which depend on calibrated empirical parameters.[46] RANS
particle tracking method, such as the Discrete Particle Method
models are very computationally-efficient and are reasonably
(DPM)[93–96] for the inclusions. Argon gas can disrupt this
accurate for estimating the steady-state flow pattern, owing to
clogging process, by changing the flow pattern.[90] Clogging also
their use of empirical wall-law boundary conditions which can
depends greatly on the compositions of the inclusion and nozzle
capture the sharp velocity gradients in the boundary layers with a
refractory surfaces, and whether the inclusions are liquid or
coarse grid.[44] Although transient behavior is less accurate with
solid, which requires thermodynamic modeling.[97]
RANS models,[47] unsteady RANS models have been applied
Clogging can also be caused by inclusions generated from
recently to estimate transient flow, with some success.[48–50]
reoxidation, via exposure of the steel stream during open-stream
Large Eddy Simulation (LES) can more accurately resolve the
pouring or by aspiration of air through cracks and joints in the
details of transient flow, owing to its use of finer grids and
nozzle in submerged casting. Reoxidation has been modeled
smaller time steps to resolve the turbulent flow, and sub-grid-
in free stream pouring by adding an oxygen transport model
scale turbulence models which lessen the importance of the
and an oxide formation model to a fluid flow model.[98] The
empirical turbulence-model parameters as the grid becomes
reoxidation rate depends on the exposed surface area, and the
increasingly refined. However, this required grid refinement
saturation extent of the molten steel.
makes LES much more computationally expensive. As computer
Modeling the pressure distribution inside the nozzle can
hardware, and software tools such as parallel solvers using
provide important insights into nozzle clogging, as aspiration-
graphics processing units,[51] have become more powerful, the
based clogging is most likely when the pressure drops below
application of LES models to continuous casting has become
atmospheric pressure. Several previous models have applied
more popular in recent years.[52–61]
CFD models to predict these pressure distributions.[62,63] These
models found that aspiration can be mitigated by decreasing
the tundish depth, decreasing the upper tundish nozzle,
3.1. Fluid Flow in the Nozzle and submerged entry nozzle inner-bore diameters, and/or by
increasing the argon gas flow rate, thereby causing the flow
The flow velocity and pressure distributions in the metal delivery control to open more which lessens the local pressure drop.
system to the caster, including the tundish, flow control system
(stopper rod or slide gate), and the submerged entry nozzle are
very important to steel quality. This is because they control: 1) air 3.3. Reduced-Order Flow Models
aspiration which is a source reoxidation and inclusions, if the
pressure becomes negative near a joint in the refractories, 2) It is often useful to obtain a quick estimate of certain
nozzle clogging due to inclusion particle agglomeration on the phenomena, which are significant but peripheral to a project,
nozzle walls, and 3) the flow rate in the entire process, including without developing a full, three-dimensional computational
the flow pattern and level fluctuations in the mold. model of that aspect of the process. This can be accomplished
Computational models of the nozzle have included the using analytical solutions or with “reduced order” models. For
slide-gate,[62–64] the stopper rod,[65,66] or the refractory-funnel- example, the pressure distribution and corresponding flow rate
nozzle semi-open-pouring flow control system,[67] which greatly through the liquid metal delivery system, is needed for online
affects the flow in both the nozzle and the mold cavity.[68] Many control models, parametric studies, and as input to mold flow
models have simulated the important effects of argon gas models. This pressure distribution can be estimated using a

steel research int. 2017, 1700312 1700312 (5 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

simple energy-balance approach based on a Bernoulli-type

analysis of turbulent pipe flow.[99] Such simple models can
match reasonably well with both a multiphase Eulerian/Eulerian
numerical model of the flow, (Figure 5) and with plant
measurements (Figure 6), considering that the top surface
pressures in both tundish and mold are predicted to be near
1 atm for the given plant conditions.[100]
Both models predict negative pressure below the slide gate.
Considering that the grooves in the slide gate plates are supplied
with argon gas to avoid air aspiration, these joints will allow
passive flow of argon gas into the low-pressure region below the
slide gate. If leaks are excessive or if argon flow is insufficient,
severe reoxidation can occur. The argon flow rate needed
to sustain positive pressure inside the nozzle (above 1 atm) can
be predicted with these models.[100]

3.4. Argon Gas Injection

Argon gas is commonly injected into the Upper Tundish Nozzle

to help prevent clogging. After being delivered through a gas line
above the slide-gate, it typically diffuses through porous portions
of the nozzle refractory to exit the inner-bore surface into the
flowing steel, as shown in Figure 7a.
Figure 6. Pressure distribution down nozzle, comparing numerical, and
analytical models.[100]

In order to determine the bubble size and flow-rate

distribution that enters the steel, the gas flow distribution
through the hot refractory has recently been modeled,
including the effects of the refractory geometry, thermal
expansion of the gas, temperature-dependent gas viscosity, and
possible leakage into any unsealed joints. A one-way-flow
pressure boundary condition was formulated and implemented
to account for the pressure threshold needed to overcome
surface tension to create a bubble at the refractory-steel
interface, and to prevent unrealistic flow of steel into the
refractory.[101]
The complete model was validated with both analytical
solutions of 1-D test problems and observations of a water
bubbling experiment, which is shown in Figure 7b and c. [101]
For this particular nozzle geometry, note that no gas exits the
lower left of the nozzle, owing to this pressure threshold. In
the steel plant, both the pressure and flow rate of the gas are
measured. Because this porous-flow model requires only one
of these measurements as a boundary condition (the inlet
pressure), the predicted flow rate can be compared with the
measured flow rate, and the difference represents leak-
age.[101] Practical results suggest that the leakage can be
significant, exceeding 50%.[49] This model could serve as an
important initial step of a comprehensive multiphase model
system.

3.5. Multiphase Fluid Flow in the Mold

Particle entrapment associated with flow problems in the mold

Figure 5. Gas volume fraction in a slide-gate nozzle.[100] is responsible for most of the serious defects in final-rolled

steel research int. 2017, 1700312 1700312 (6 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

Figure 7. Comparison between predicted and measured gas flow rate at the upper tundish nozzle inner surface.[101] a) Schematic of gas injection
process, b) Underwater experiment showing bubbling locations, c) Predicted gas normal surface velocity distribution. Reproduced with permission.[101]
2015, TMS.

continuous-cast products.[102] These include pinholes and have been coupled together with the single set of fluid
blisters from entrapped bubbles, and the inclusions which continuity and transport equations to track the trajectories
often accompany them, especially in products with exposed of individual argon bubbles, and to include their local effects
(visible) surfaces, such as body panels for automotive and on the flow.[58,60,75,77,81,84] Using such an Eulerian–Lagrang-
appliance applications. They also include slivers from inclusions ian two-way coupled model of Ar-steel flow, with a k e
and the entrainment of mold slag, which accompany flow turbulence model, larger (0.25–2.5 mm diameter) bubbles
problems in the mold, and are aggravated by unstable single-roll were predicted to float directly upwards near the SEN to the
flow patterns, which is affected by the buoyancy of argon gas top surface upon exiting the nozzle ports, while smaller
bubbles,[103] excessive surface velocity which can entrain the (<0.25 mm) bubbles travel with the jet across the mold
mold slag,[104] interactions with argon gas bubbles, leading to cavity.[81]
their eventual capture into the solidifying shell,[105] the capture of With higher gas fractions, it is more accurate to solve an
small inclusions by the argon bubbles, leading to surface defects extra continuity equation and complete extra set of transport
such as slivers if they are captured, and surface level fluctuations equations for the gas phase, coupled together with the
that disrupt initial solidification in the meniscus region.[3,68] fluid continuity and transport equations, in order to enforce
Each of these problems is associated with a different type of the total volume fraction to equal 100% everywhere. This
multiphase flow, which can be modeled via many different Eulerian–Eulerian approach has been applied successfully in
approaches. many previous studies of argon-steel flow in the continuous
Firstly, argon gas injection modifies the flow pattern, due to casting mold region.[56,61–64,66,71,72,76,78,80,82] Asymmetric,
the two-way coupling between the bubbles and the turbulent oscillating flow is observed if gas fractions are excessive.[62]
flow field. The gas buoyancy lifts the jets exiting the nozzle ports, Using a RANS k e model for turbulence, parametric studies
while the bubbles are dragged by the momentum of the with this approach quantified how the detrimental transition
turbulent flow. This has been the subject of many previous from double-roll to single-roll could be avoided by keeping
modeling studies,[58,60–64,66,69–88] using several different compu- the argon gas volume fraction below a critical fraction,
tational approaches. If the gas fraction is small, the most efficient which is lower and easier to achieve with narrow mold
method is a simple mixture model, such as Algebraic-Slip[66] or a width.[71]
modification.[69,70] These quasi-multiphase approaches include The Eulerian–Eulerian approach has been extended to enable
the effect of the bubbles via an extra source term in a single set of spatial variations in bubble size, and size evolution according to a
momentum equations for the fluid phase, and track the gas population balance framework via the homogeneous MUlti-SIze
fraction by solving one extra transport equation. Results Group (MUSIG) model.[106] Models of flow in continuous
reveal that increasing Ar gas increases upward flow near the casting molds using this approach[73,74,87,88] match well with
Submerged Entry Nozzle, (SEN) and tends to transform the measurements of both flow pattern and bubble size in water
classic double-roll flow pattern to a more detrimental single-roll models, except near to the SEN. Like DPM, the MUSIG method
flow pattern with surface flows directed away from the SEN can correctly predict rising flow near the SEN that contains
toward the narrow faces.[70] large bubbles and the steel jet that traverses directly across the
To improve on this by considering the bubble size distribu- mold cavity that transports smaller bubbles.[73,87,88] However, as
tion, Lagrangian methods, such as the Discrete Particle Method with the Eulerian–Eulerian models, all bubble sizes in the

steel research int. 2017, 1700312 1700312 (7 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

homogeneous MUSIG model share a single gas velocity field, into detached eddy simulations of the transient turbulent flow,
and momentum interactions with the fluid phase are governed dithering of the flow control device (e.g., stopper rod) at the
by a local average bubble size. wrong frequency was shown to cause severe surface sloshing
To improve on these limitations, non-homogeneous MUSIG instabilities, which matched predictions from an analytical
models have been developed recently, which enable each of the model, and observations at the plant.[49,50] Using an Euler–
bubble size groups to have a different velocity field by solving Euler–Euler model for the molten steel, liquid slag, and air above
more than one set of continuity and momentum equations for the slag layer, and a Lagrangian DPM model for the argon
the gas phase.[85,86] Similar to other simpler multiphase models, gas particle transport, the behavior of four different phases was
results with this sophisticated method for handling bubbles recently modeled.[116] This model showed how argon bubbles
show generally good agreement with measurements in water raise the slag-steel interface near the SEN, and cause more level
models, using the standard k e model for turbulence.[85,86] All of fluctuations there, which matched well with physical model
these methods require the initial bubble size distribution to be experiments of a slab caster where air bubbles were injected into
input at the model inlets, so future work should be combined water.[116]
with output from the models of initial bubble size discussed in
the previous section.
3.7. Effect of Applied Electromagnetic Forces on Flow

3.6. Level Fluctuations and Slag Entrainment In addition to nozzle geometry, strand cross section, casting
speed, and argon injection, flow in the mold can be greatly
Another important aspect of flow in the mold is the profile of the affected by the application of electromagnetic forces. Computa-
top surface interface between the molten steel and the slag tional models are ideal tools to investigate these effects and
layers, and the fluctuations of this level. Excessive variations of several important commercial configurations have been studied
the surface profile both in space (often called “standing wave”), in slab casting, including with local,[117,118] single-ruler,[119–121]
and in time, (level fluctuations), lead to surface defects by and double-ruler[54,58–60,72,107,122–124] ElectroMagnetic Braking
upsetting the uniform infiltration of liquid slag into the gap (EMBr).
between the shell and the mold, causing abnormal freezing of In the widely-used double-ruler configuration, or “FC-Mold”,
the meniscus to form hooks, the entrapment of mold slag, non- two rectangular-shaped magnetic fields are applied across the
uniform heat transfer, and other problems. Excessive flow mold width, with one above and the other below the nozzle ports.
velocity across this slag-steel interface may lead to emulsifi- Increasing the upper ruler strength tends to slow the surface
cation, vortexing, and the entrainment of slag droplets into the velocities, decrease level profile variations, decrease level
liquid pool. The simplest way to model surface profile variations fluctuations, and dissipate high-frequency turbulent fluctuations
from the results of a flow simulation with a fixed flat top surface inside the nozzle.[54,120,122,125] Increasing the strength of the
is via a simple energy balance, converting the pressure variations lower ruler field tends to deflect the jet upwards, causing surface
into the potential energy of the level variations.[52,66,83,107] This velocities and turbulence to increase.[54,122] Finding an optimal
method has been shown to be surprisingly accurate for many surface velocity and turbulence level is essential to avoid casting
casting conditions.[52,54,58,66,83,107] because the relative height of defects. Moving magnetic fields can create a stirring motion in
the level variations is usually small. In this approach, it is easy to the mold,[75,125–127] or can accelerate, or slow down the jets in
account for the molten steel displacing the slag layer, which can the mold,[103,125] and are only recently receiving attention by
amplify the level changes.[52] However, this effect is usually small modelers, owing to their complexity and the need for transient
relative to simple rising and falling of the slag with level models.
fluctuations, so the slag layer can be ignored when estimating It is important to include the effects of the conducting steel
level fluctuations.[58,83] shell in the model, especially for transient simulations. This is
More advanced methods have been applied to simulate because the current returning through the solid shell tends to
surface level and level fluctuations during continuous casting, stabilize the flow pattern and suppress unsteady low-frequency
including moving grid methods,[49,50,108–110] the free-surface oscillatory behavior of the jets that occurs when the current
Lattice–Boltzman method,[111] and the Volume of Fluid (VOF) returns through the fluid boundary layer, or Hartman layer.
method.[84,112–114] In addition to handling larger fluctuations This effect on stability has been confirmed with both LES
accurately, these methods can capture the effects of surface models[54,120,121] and with physical models using low-melting-
tension and complex interactions between gravity and momen- temperature metal alloys.[128] Positioning a field directly across
tum forces, which can cause surface waves, sloshing, and slag or in front of the ports should be avoided because this
entrainment. Simulations with VOF showed how surface destabilizes the flow by amplifying minor variations in jet
instabilities move from the narrow face region to near the angle: slightly low jets are deflected downward and slightly
SEN with decreasing casting speed, owing to the decrease in upward jets are deflected upward.[129] The RANS approach
dissipation of the jet momentum leaving the ports.[113,114] To cannot predict these transient phenomena.
understand the entrainment of slag droplets, starting from The electromagnetic field is often combined with argon gas to
protrusions beneath the slag layer, entrainment has been alter flow in the mold. Computational models are better-suited
modeled with VOF using LES on a fine grid and a special method than plant trials or lab experiments to study the complex
to identify droplets, and found to match droplet sizes observed in interactions between these effects in commercial casters, and
a water model.[115] Using a moving grid method implemented this important multiphysics problem is receiving more attention

steel research int. 2017, 1700312 1700312 (8 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

Figure 8. Comparison of surface velocity variations between LES model predictions and measurements a) without and b) with EMBr. Reproduced with
permission.[58] 2016, TMS.

in recent years.[58,60,72,75,130] Compensating for changing process surface velocities and level profiles measured in a typical slab
conditions such as argon gas is difficult, however, so real-time caster with argon gas, both with and without double-ruler
measurement of the mold level profile and casting conditions EMBr.[58] The LES model predictions agree well with nailbaord
combined with databases generated from computational-model measurements in the operating caster. With EMBr, the surface
results is being developed as a methodology for online control velocity is slightly lower on average, and the level profile is flatter.
of the electromagnetic field to generate optimal mold flow Of greater importance, EMBr causes a significant decrease in
conditions.[103] the time variations of both surface velocity and level, which
Transient models can reveal deeper insights into the are indicated in the figure by the range of the predicted
fundamental nature of the chaotic turbulent flow and electro- instantaneous profiles and the error bars on the measurements.
magnetic effects. For example, Figure 8 and 9 compare the A major cause of these fluctuations is revealed by LES results to

Figure 9. Comparison of surface level variations between LES model predictions and measurements a) without and b) with EMBr. Reproduced with
permission.[58] 2016, TMS.

steel research int. 2017, 1700312 1700312 (9 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

be flipping between clockwise and counterclockwise flow exiting

the nozzle ports. When flow from the slide gate opening, located
toward the outside radius, goes straight down the outer-radius
side of the nozzle, the jets exiting the ports swirl clockwise and
are very strong, leading a short time later, (4.3 s) to a larger
surface velocity with more level fluctuations. When flow from
the slide gate bounces across to the inside-radius side of the
nozzle, and thus exits the ports with a counter-clockwise swirl,
some of its momentum is lost, the jets deflect more toward the
outer radius of the mold, so the surface velocity a short time later
is a little slower and the level fluctuations are smaller. The EMBr
field of the upper ruler across the nozzle lessens the amount of
flipping between swirl patterns exiting the ports, leading to more
stable jet flow with less fluctuations in both surface velocity and
level.[58]
Electromagnetic effects also have an important effect on
particle capture, due to their influence on the general flow
pattern, local turbulence, and boundary-layer flow. A few studies
have investigated this, with both RANS models with random
walk particle trajectories[118] and LES models.[60,123] The overall
effect of EMBr on particle removal is small,[118] as most particles
are small and easy to capture. By deflecting the jet upwards,
EMBr can decrease the number of particles penetrating deep
into the caster, thereby improving internal cleanliness.[118]
However, this may increase particle capture high in the strand,
which increases surface defects.[118] Depending on the EMBr
orientation relative to the nozzle ports, sometimes the opposite
occurs, due to EMBr deflecting the jet downward.[122] By
generally increasing tangential velocity in the boundary layer,
EMBr may lessen the capture of large particles by increasing the
chances of bubbles and inclusions being washed off of the
solidification front and back into the bulk flow. This is significant
because the large bubbles and inclusions are of greatest concern
to steel quality. More work is needed to validate and quantify
these effects, coupling together the relevant phenomena in
parametric studies of realistic caster geometries. Figure 10. Instantaneous temperature calculated in centerplane with LES
model. Reproduced with permission.[132] 2005, TMS.

3.8. Superheat Transport

The transport of superheat by the turbulent fluid flow in the mold is

very important to both initial solidification at the meniscus, where
insufficient superheat can cause hook formation and surface
defects, and final solidification, where excessive, nonuniform
superheat distribution can change the solidification structure from
equiaxed toward columnar, and aggravate centerline segregation.
The superheat-temperature distribution is readily simulated by
adding a simple energy transport equation to the fluid flow
problem.[131,132]
These models show that most of the superheat is removed in
the mold region, or just below.[131,132] As shown in Figure 10, the
temperature decreases continuously with distance along the
path travelled by the flowing jet, with the lowest temperatures
found at the meniscus region, near the narrow faces and SEN,
for a typical double-roll flow pattern.[132] Comparisons with plant
measurements confirmed the model predictions that liquid in
the upper recirculation zone above the jet in the mold cavity Figure 11. Comparison of temperature calculated down vertical line in mold
retains less than one third of its superheat temperature, as with measurement from inserted probe. Reproduced with permission.[132]
shown in Figure 11.[132] 2005, TMS.

steel research int. 2017, 1700312 1700312 (10 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

The direction of the steel jets flowing from the nozzle 3.10. Particle Entrapment
controls the distribution of the superheat. If the nozzle port
angle is directed too deep, then the critical meniscus region Steel quality is greatly determined by the entrapment of
receives too little superheat, so the meniscus can solidify inclusions during continuous casting in the mold region, which
to form hooks, as discussed in the next section. In addition originate from upstream, the entrainment of slag from the top
to its effect on hook formation and surface defects, surface of the mold, or from the entrapment of argon gas
superheat dissipation in the mold is also important to the bubbles coated with inclusions. Many computational models
formation and survival of crystal nuclei, which greatly have been applied in recent years to simulate particle transport
influence microstructure development and segregation lower and entrapment into the solidifying steel shell, starting with a
in the caster. RANS or LES model of turbulent flow, and then solving a simple
one-way coupled Lagrangian DPM method to track the particle
motion.[53,57,60,77,79,81,85,118,123,130,139–141]
3.9. Thermal-Flow Models of Shell Solidification To predict the capture of particles, including both inclusions
and small gas bubbles, a suitable capture criterion is
Further coupling the fluid flow and energy transport model needed. The simple criterion that particles are entrapped if
with solidification enables direct study of how the superheat they touch the solidification front is often used. This is a
transported by the flowing molten steel affects shell growth. reasonable approximation for particles smaller than the primary
The numerical methodologies to couple these phenomena dendrite arm spacing.[140] However, small particles might flow
together efficiently have been pioneered by Beckermann, into the mushy zone, navigate between the dendrites, and escape
Voller, and coworkers.[133–135] Rather than solving for tempera- back into the bulk flow. This can be modeled by including
ture directly, many thermal-flow models with solidification solidification heat transfer into the fluid-flow model, but this
solve for enthalpy, with subsequent lookup of temperature, in computationally-intensive approach also needs a very fine
order to achieve better numerical stability. The tremendous mesh to resolve the mushy region[79] and empirical fitting
changes in thermal and flow properties from liquid to solid are parameter(s), in order to match with plant measurements.[142]
often handled with temperature- and/or phase- dependent Larger particles are more difficult to capture, and require a
properties. The large effective viscosity of the solid is more advanced capture criterion. Because it cannot fit between
often achieved by simply fixing the solid velocity to the casting the dendrites, a large particle touching the solidification front
speed. can be stopped, surrounded by dendrites that grow around it,
Together with an accurate model of heat transfer across the and captured, only if the tangential drag forces from the flowing
interfacial gap between the solidifying steel and the mold or fluid that try to rotate it away are insufficient to overcome a force
secondary-cooling chamber, a computational thermal-fluid flow balance with the other forces acting on the particle. In addition to
model can predict the shape of the solidifying steel shell, drag, lift, virtual mass, pressure-gradient, and buoyancy/gravity
including the local shell-thinning effect near the point of jet forces, which also affect particle transport in the bulk flow,
impingement onto the solidification front, for example at the additional forces acting in the boundary layer include the
narrow face in a slab caster,[79] or inner shoulder in a beam-blank lubrication, Van der Waals, and surface tension gradient
caster.[34,136] To accurately capture this behavior requires a very forces.[79,140] Surface tension gradients are caused by solute
refined mesh, however, in order to avoid exaggeration of the rejection at the solidification front, which produces concentra-
effect of the fluid flow on the shell growth.[79] Heat transfer tion gradients in highly-segregating, surface-active elements.
calculations are more sensitive to achieving adequate grid This lowers the surface tension near the solidification front,
refinement in the boundary layers than are velocity calcula- which generates a force toward the solidification front that
tions.[21,132] In addition, careful attention should the paid to increases particle capture in steel grades rich in elements such
choosing the terms in the transport equations so that the latent as S[140] or Ti.[103] More research is needed to study these
heat released during columnar solidification into the solid phenomena with validated, quantitative models.
should not be advected away with the flow.[79] This differs from With a steady RANS flow model, the chaotic motion of the
the solidification of equiaxed grains, which many models are turbulent particle trajectories can be approximated using the
based upon. Random Walk method, which adds a random velocity compo-
Other recent thermal-flow models with solidification showed nent to the particle velocity, that is, proportional to the local
that the flow pattern in the mold, as affected by electromagnetic turbulence level.[57,77,79,81,85,94,118,130,140,142] Unfortunately, this
braking, had a significant effect on the solidification profile, method assumes isotropic turbulence, which near solid walls or
which extended all the way to the shape of the final the solidification front, over-estimates the local velocity compo-
solidification front, or “crater end” that defines the metallurgi- nent toward those boundaries. This likely explains why this
cal length.[137] This crater-end shape has an important effect method has been observed to overestimate particle capture,
on centerline defects, and was found to exhibit a W-shape relative to plant measurements.[142]
profile across the strand width, with deeper fluid penetration A more accurate method is to simulate the chaotic motion of
near the microstructure triple-points, caused by downward the particles directly, using a transient Large Eddy Simulation
flow along the narrow faces, together with stronger for the turbulent fluid flow.[53,60,123,140,141] Figure 12 shows
surface heat removal from the water sprays along the strand that captured small particles are widely distributed.[60] These
centerline.[137,138] results for bubble capture match reasonably well with plant

steel research int. 2017, 1700312 1700312 (11 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

increases the removal rate (average number of inclusions removed

per unit time) of small particles to the top surface slag layer.[140]
Using an advanced capture criterion, only a very small fraction
(<0.02%) of larger bubbles (>1 mm) are captured (Figure 14).[130]
The simple capture criterion greatly overpredicts the capture of
large particles. Also, to achieve reproducible results, (removal rate
variations within 3%), more than 2500 particles of each size
should be tracked.[140]
The attachment of inclusions to bubbles, while they are being
transported through the molten steel pool is another important
phenomenon, which fundamental models of individual bubbles
can help to understand. One such study found that bubbles
circulating in the mold can entrap many small inclusions, even
though the chances of any given inclusion becoming attached to
the bubble is small.[143] Smaller bubbles are more efficient at
inclusion attachment and removal for a given gas volume, so
long as they are not captured into solidifying shell.[143] Another
model recently confirmed this finding and further quantified the
important contribution of the wake behind the bubble.[144]
Future modeling studies of turbulent flow, transport, and
capture of particles should combine more of these phenomena
together.

Figure 12. Location of small bubbles captured on wideface with advanced

criterion. Reproduced with permission.[130] 2016, TMS.
4. Initial Solidification at the Meniscus
measurements of capture fraction and distribution in the The initial solidification of the tip of the steel shell at the
solidified slab, and also with particle size distribution, as shown meniscus around the perimeter of the continuous-casting mold
in Figure 13. [60] Note that this model underpredicts particle is of paramount importance because it creates the surface of the
capture near the meniscus, which end up near the strand final product. The behavior of this region is governed by very
surface, perhaps due to its neglect of entrapment by solidified complex interrelated phenomena during each oscillation cycle.
hooks and/or its assumption of complete removal into the slag Recent advanced computational models focus either on thermal-
layer of every particle that touches the top surface. fluid flow behavior in the region or on thermal stress in the
Using flow fields obtained from both RANS and LES solidifying steel shell. To capture all of the known behaviors
simulations, most of the small inclusions (92% of those realistically would require combining both model types together
<40 μm[53] and 85% of those <80 μm[130]) are found to be into a true multiphysics model, which is beyond current
entrapped in the final product, as shown in Figure 14. Gas injection capabilities.

Figure 13. Number of bubbles captured on each layer of center sample (left) and corresponding average diameter (right),[130] where OR is outer radius
and IR is inner radius. Reproduced with permission.[130] 2016, TMS.

steel research int. 2017, 1700312 1700312 (12 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

Typical results for the changes in temperature, velocity, and

phase fraction contours during an oscillation cycle are shown in
Figure 15a.[149] A very fine mesh, 50 micron cell spacing in
this example,[149] is needed to capture the large gradients in
properties and field variables that occur over the short distances
involved. The shell tip, which forms the bottom of the eventual
oscillation mark, moves steadily downward with time. This
stretches the meniscus downwards until the ferrostatic pressure
of the molten steel can no longer be supported by the surface
tension, and the meniscus overflows the shell tip, during
Figure 15a frames d–f. This creates a small “hook”, where the
partly solidified meniscus extends into the molten steel. These
hooks are often observed near the strand surface in micrographs
such as Figure 15b, and are detrimental because they can capture
bubbles and inclusions, leading to surface defects.[150] This also
completes formation of the surface oscillation mark, which
moves down at the casting speed, and essentially retains its
shape in the final slab. This particular simulation was run for >8
oscillation marks. The pitch between oscillation marks, their
depth and shape, and the variations in these parameters for this
Figure 14. Fraction of bubbles captured for varying bubble diameter. and several similar simulations all match reasonably well with
Reproduced with permission.[130] 2016, TMS. plant measurements and their variations.[39,149] Some of these
variations are rooted in the classic chaotic behavior of complex
systems.[148] Predicted slag consumption rates and trends with
4.1. Thermal-Fluid-Based Meniscus Models operating parameters also agree well with measurements.[38,39]
Temperature evolution at different locations in the mold wall
Advanced thermal-fluid models of the meniscus region,[2,38,39,145–149] can be compared with thermocouple measurements for model
solve the transient heat conduction equation in the mold, validation.[38,39] Comparison with average heat flux based on
interfacial gap, solid and molten steel, and slag. Transient fluid heat-up of the cooling water is also useful for validation of heat
flow in the molten steel and slag is modeled with a set of flux profiles for runs which extend to mold exit. These models
momentum equations, mass conservation, phase fractions of can also predict tangential shear stress evolution along the mold
the slag and steel, and the VOF method to resolve the changing wall,[39,149] which potentially could be compared with mold
shape of both the meniscus and interfacial gap that separates friction measurements in future work. Improved insight into the
the slag and steel. Transient flow must properly include relationship between defect formation and real-time measured
turbulence in the molten steel bulk that diminishes to laminar parameters such as heat transfer and friction could enable better
flow toward the boundary layer near the solidification front, online diagnostic tools to benefit commercial operations.
which can be handled by k–ω Shear-Stress Transport (SST)
models[39,149] or by LES.
Mold powder added to the top surface in the mold heats, 4.2. Thermal-Mechanical-Based Meniscus Models
sinters, melts, and flows down to form a molten slag layer that
floats on the molten steel. The increase in thermal conductivity, Another approach to model initial solidification at the meniscus
initial increase in viscosity during sintering, and subsequent is to use a thermal-stress model, which is needed, for example, to
drop in viscosity during melting, are modeled via temperature- understand the well-known important effects of steel grade on
dependent property functions.[38,39,149] Slag in this liquid layer oscillation mark and crack formation. A few thermal-mechanical
moves slowly and intermittently toward the meniscus region, models of the initial solidifying shell have been developed[150–155]
where some solidifies against the oscillating mold wall to form a and applied to study distortion behavior during level fluctua-
rigid rim that pushes down on the meniscus during each tions,[151] the effect of steel grade,[11,152,154] and initial cracks.[155]
downstroke. Some slag is pushed down to be consumed into the These models, together with measurements, have revealed new
interfacial gap where it cools, may crystalize, or may undergo a insight into the mechanism of hook formation,[150] which build
glass transition, and some is dragged down by the moving solid on the overflow mechanism illustrated by the thermal-fluid
steel shell that is imposed to move downward at the casting model results in Figure 15a. In addition to overflow, thermal
speed. This requires a different set of temperature-dependent distortion can contribute to the shape of the lower (downstream)
properties for the cooling slag near the mold walls and in the surface of the oscillation mark.[150] Although mechanical
interfacial gap.[38,39,149] distortion of the liquid meniscus during oscillation is of critical
Temperature-dependent properties are also required in the importance to oscillation mark formation and slag consumption,
steel phase, including the great increase in viscosity (by >10 mechanical bending of the solidified meniscus hook is unlikely,
orders of magnitude)[39] that accompanies solidification from because it is so strong and brittle.[150] Ultralow carbon steels and
liquid to solid. This creates computationally difficulty, which can peritectic steel grades experience the greatest thermal distortion,
be handled by assuming the solid steel is rigid.[38,39,149] and consequently the deepest oscillation marks, both in the

steel research int. 2017, 1700312 1700312 (13 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

Figure 15. a) Thermal-fluid model of meniscus region behavior during an oscillation cycle, showing shape of meniscus and solidified steel shell outlined
in black.[149] b) Photograph of an as-cast microstructure cross section, showing a typical oscillation mark with a hook. Reproduced with permission.[150]
2006, TMS.

steel research int. 2017, 1700312 1700312 (14 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

model simulations and experiments.[152,154] This is because

these steels have more phase-transformation shrinkage from
delta to austenite in the solid state, while the solid steel shell is
both strong enough to resist ferrostatic pressure and thin
enough to experience significant thermal distortion.[152,154] The
next sections discuss thermal-mechanical models in more detail.

5. Thermal-Mechanical Behavior of the Mold

Thermal distortion of the mold from startup to steady operation
is important to residual stress, residual distortion, fatigue cracks,
and life of the mold, which in turn can greatly affect steel
quality.[9,156] Mold distortion is also important to heat transfer
and mechanical behavior of the solidifying shell, due to its great
effect on the interfacial gap between the mold and shell, and is a
critical consideration in the design of mold taper.[33,157] Mold
thermal distortion has been explored with computational
thermal-stress models for steady casting conditions in square
billet molds,[157,158] round billet molds,[17,30,33] conventional
thick-slab molds,[19,20,159] beam-blank molds,[160] and in thin-
Figure 16. Temperature field simulated in a thin-slab casting mold with
slab molds with [9,156,161,162] and without[9,161] a funnel. Each
thermal distortion (magnified 50). Reproduced with permission.[162]
mold shape has distinctive thermomechanical behavior. These 2012, TMS.
studies have revealed the importance of not oversimplifying the
geometry, applying appropriate contact constraints between
connecting parts,[159,162] and using appropriate loading con- through just the backing plate pull the copper plates back toward
ditions and material constitutive equations,[161] in order to the central region of the wideface. Thus, the maximum inward
simulate reasonable thermal-mechanical behavior of the system. distortion on the wideface is predicted near the location of
To simulate thermal distortion of the mold and its related highest temperature just below the meniscus and another
phenomena first requires solving the heat conduction equation, distortion peak is found near the mold bottom.
typically using measurements to help determine the interfacial The thermal expansion causes the narrow face to distort into a
heat flux needed to obtain an accurate temperature field. In trapezoidal shape that contacts the wide face along just a thin
addition, a thermal-mechanical model must solve the equilib- vertical line at the front corner of its hot face. This thin edge
rium equations which relate force and stress, the constitutive must transmit all of the clamping forces, so is susceptible to
equations which relate stress and strain, and the compatibility accelerated wear, especially during automatic width changes. If
equations which relate strain and displacement. Thermal strain clamping is too high, the hot edge can be crushed, causing a gap
is calculated from the temperature solution. Three-dimensional after cooling, which can lead to finning defects or even a sticker
finite-element models, including plasticity and creep if desired, breakout, if steel enters that gap during startup. In addition, the
can be developed using commercial finite-element packages widefaces may be scratched, leading to longitudinal cracks, and
such as ABAQUS,[163] which are well-suited to this nonlinear other defects on the strand surface.
thermal stress problem. Details are given elsewhere.[161,162] Another problem is permanent distortion and crack forma-
In order to accurately simulate the thermal-mechanical tion in the mold surface, especially in the meniscus region where
behavior and match the measured distortion, models should temperature is highest. Constraining the thermal expansion
include all important geometric features of the mold, which for a induces compressive stress and creep in the hot face of the
typical slab mold includes four copper plates with their water copper plates. This compression relaxes during operation, which
slots,[159] a reinforced steel water box assembly,[159] and tightened leads to residual tensile stress after cooling. Repeated cycles can
bolts.[162] Its four-piece construction causes a slab mold to behave build up significant distortion of the mold plates and fatigue
very differently from single-piece billet and bloom molds. cracks.[9] Their removal requires increased remachining and
Figure 16 shows example temperature contours and the reduces mold life.[9]
displaced shape calculated in one quarter of a commercial thin- Transient thermal-mechanical behavior of the mold has
slab funnel-mold caster under steady operating conditions.[162] received much less attention. One recent modeling study
The hot exterior of each copper plate attempts to expand, but is investigated the behavior of this same thin-slab mold during a
constrained by its colder interior and the constraint of the cold, startup.[96] Figure 17 and 18 show how the liquid level, position
steel water box which is attached to it. This makes each plate of the mold bottom being withdrawn downward by the dummy
bend inwards toward the solidifying steel, typically on the order bar, and thermal distortion all evolve with time.[96] The predictions
of one millimeter as observed on the narrow face in this mold. in Figure 18 are compared with two sets of inclinometer
The wideface distortion of this particular mold is very different, measurements obtained during startup and during operation,
however, owing to its waterbox being hollowed out to house an which confirm both the inward-curved shape of the mold narrow-
electromagnetic breaking system, so the shorter bolts going faces and the magnitude of the distortion.[96] These distortion

steel research int. 2017, 1700312 1700312 (15 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

Because the liquid level at the top free surface in the mold is
free to rise and fall, the mass contained in the computational
domain for the mechanical analysis can change with time during
the process, and it is usually too expensive to include this
free surface in the model domain. This issue can be addressed
in several other ways, such as by introducing another strain
component for the fluid flow,[164] leaving space for shrinkage
in the central portion of the domain,[34,164,165] or by adopting an
Arbitrary Lagrangian–Eulerian (ALE) formulation, where the
internal fluid flow region is solved with an Eulerian methodology
and the exterior mechanical behavior of the moving solid shell is
solved with a Lagrangian method.[166]
Modeling the constitutive behavior of the solid steel is
difficult, owing in part to the challenge in finding accurate
experimental measurements of thermal-mechanical properties
at the high temperatures (including solidification), low strains,
(less than 2%), and low strain rates (10 6 to 10 2 s 1) relevant to
this casting process. The problem is further complicated by the
growing number of different steel grades of commercial
importance, such as advanced high-strength steels, which have
Figure 17. Mold level and dummy bar positions during startup.[96] 2014, very different thermal-mechanical properties. A promising
ASMET. approach to simplify the problem of conducting mechanical
tests on so many different grades is to consider that the
mechanical properties are governed mainly by the phases
predictions are important for designing mold taper to avoid present.[11,167–170] Delta-ferrite is much softer and prone to creep
detrimental air gap formation, especially for new mold designs. than austenite.[168,169] The effect of steel grade on stress
generation during initial solidification has recently been studied
using such a phase-dependent constitutive-model approach.[11]
Accurate computational modeling requires verification with
6. Thermal-Mechanical Modeling known analytical solutions followed by comparison with plant
Thermal-mechanical models are needed to investigate many measurements. Unlike models of heat transfer and fluid flow,
important phenomena in steel continuous casting. Examples where such validation has become routine, it is much less
include gap formation and crack formation in the mold, bulging common to see proper verification and validation of thermal-
between rolls, bending and unbending, the formation of surface mechanical models prior to their application to practical
depressions, and internal cracks, and segregation. In solving the continuous casting problems. An excellent analytical solution
same equations needed for mold distortion, the total strain is of solidification of an elastic-perfectly-plastic material exists,[171]
divided into elastic, thermal, and inelastic components, the latter which has been used for verification of a few thermal-mechanical
needed to handle plasticity and creep in the solidifying steel. models of steel continuous casting.[11,34,35,164,165] These studies

Figure 18. Simulated narrow face mold shape (curved lines) and inclinometer measurements (straight lines) just after startup and after a width
change.[96] 2014, ASMET.

steel research int. 2017, 1700312 1700312 (16 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

reveal that achieving mesh refinement is still an obstacle to shell leads to corresponding liquid level fluctuations at the
accurate modeling, as a much finer mesh is needed for meniscus region in the mold, causing surface defects if the level
mechanical analysis than for thermal modeling alone. Compari- control system is unable to overcome the problem. This dynamic
son of thermal-mechanical model predictions with stress bulging problem is receiving attention recently with thermal-
measurements is rare. A few models have successfully matched mechanical models.[181] Similar models are being applied to
lab measurements of temperature and force evolution in study the effects of taper of the machine below the mold: in
solidifying steel specimens including the submerged split-chill particular, the process of soft reduction to mitigate centerline
contraction test[172] and the gleeble test.[173] Direct plant problems.[27,182,183]
measurements of stress during continuous casting are more
difficult.
6.2. Crack Formation

6.1. Surface Shape Problems Crack formation has plagued the continuous casting process
since its inception, and limits the steel grades that a particular
Surface shape problems that have been studied with thermal- caster can produce. Severe cracks are responsible for cata-
mechanical computational models include deep oscillation strophic breakouts. Because strains are always small during the
marks and transverse depressions,[151,154] longitudinal depres- casting process, cracks only arise when local tensile stresses are
sions,[165,174] off-corner gutters,[174] rhomboidity in square accompanied by severe metallurgical embrittlement. Thus, crack
billets,[175] ovalization of round billets,[35,176] and bulging in modeling typically involves evaluating the results of a macroscale
large blooms and slabs.[3,166,177,178] Unless they are extreme, the thermomechanical model using a failure criterion.
shape problems themselves are less important than the cracks Most cracks in continuous-casting are caused by hot
and macrosegregation problems that often accompany them. tearing, which has been characterized by empirical cracking
Predicting surface shape is an easier first step for models to criteria,[170,184–187] that involve a critical level of strain accumu-
tackle, however, and comparison with plant measurements is lation over a critical temperature range near final solidification,
easier as well. and depends on steel composition and strain rate. A few recent
In the mold, the shape of the shell governs the gap size and studies have evaluated the accuracy of different hot tearing
heat transfer in the corner region, and depends on how well criteria by modeling lab experiments, which deform a solidifying
the taper profile of the thermally-distorted mold balances the steel ingot.[170,186,188] Specific types of hot-tear crack defects
shrinkage, considering that ferrostatic pressure and creep have been the focus of several successful modeling studies in
both act to minimize gap formation. Insufficient taper leads continuous casting of steel,[35,176,189] The microstructure greatly
to excessive gap formation, locally-decreased heat transfer, affects hot tearing, as columnar structures that predominate
higher shell temperatures, recrystallization, strain concentra- just beneath the strand surface are more susceptible than
tion, and crack formation. Excessive taper is equally problematic, equiaxed structures, but models rarely include this.[35,170] On the
as it can cause the shell either to jam in the mold leading microscale, models are just beginning to tackle the detailed hot
to transverse cracks, or to buckle, leading to longitudinal tearing phenomenon that involves steel dendrites pulling apart
depressions inside the mold, which can grow below the mold, when interdendritic fluid flow is too constrained. As discussed
and form subsurface cracks. Thermal-mechanical models are an later, this remains a modeling challenge.
excellent tool to investigate such problems, and to optimize Some cracks, such as transverse cracks, which typically open
mold taper.[37,174] Coupling them with a detailed model of the up during bending or unbending of the strand, are due mainly to
interfacial gap model enables better accuracy in the corner intermediate-temperature embrittlement, due to strain concen-
regions, by accounting for phenomena such as a larger local air tration at the austenite grain boundaries, made worse when
gap decreasing heat transfer across the gap, resulting in lower many small precipitates are present. These cracks occur at much
mold temperature, hotter shell surface temperatures, and a higher local strains and depend even more strongly on the
thinner shell.[15,31,35–37,179] microstructure, so have received less attention by modelers.
Even better accuracy can be achieved if the effects of fluid flow These cracks especially would benefit from microstructure
are taken into account by coupling with a thermal-flow model. A modeling, focusing on solid-state transformations and grain-
few models have demonstrated such multiphysics approaches, boundary embrittlement.
which are especially important in the mold, where flow has a
large influence on shell growth.[36,37,179] Thermal distortion of
the mold should be taken into account as well, which further
7. Segregation
complicates the multiphysics model.[34]
Below the mold, excessive bulging between the rolls, such Macrosegregation is one of the most serious defects in
as caused by either ferrostatic pressure with a large roll continuous-cast steel. Unlike internal cracks, voids, and porosity,
pitch or misaligned rolls, can cause subsurface transverse cracks, which are greatly improved by the subsequent rolling
called radial streaks, and contribute to centerline macro- process, the related problem of macrosegregation cannot be
segregation. Many models have simulated this steady bulging removed and always ends up in the final product. Modeling of
problem.[3,166,177,178] Even more problematic is transient bulg- macrosegregation is extremely difficult because it requires an
ing, which causes the volume of the entire internal liquid pool to accurate thermal-solidification model of the caster, modeling of
change with time.[180] This periodic or chaotic squeezing of the microsegregation between dendrites which generates the solute,

steel research int. 2017, 1700312 1700312 (17 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

macroscale fluid flow which redistributes the solute, volumetric scales of the entire caster, and involves three dimensions.
thermal shrinkage, and mechanical deformation which induce the Moreover, it depends heavily on the evolving temperature field,
flow, and finally failure of the microstructure, which accumulates fluid flow, solute flow, complex phase transformations in multi-
the solute locally into a defect. Many models of microsegregation at component alloys, mechanical stress and strain, and segregation.
the local scale between secondary dendrite arms have been In recent years, a few models have started to simulate dendrite
developed for steel, and match reasonably well with measure- formation in steel, using cellular automata[196,197] or phase-field
ments, although more work is needed for complicated multicom- methods.[198–200] Applications include the prediction of the
ponent alloys, especially peritectic steels.[190,191] columnar to equiaxed transition,[196] secondary dendrite arm
Although significant progress has been made since a review of spacing,[197] the delta to austenite massive-like phase transfor-
the modeling challenges over a decade ago,[192] macrosegrega- mation,[200] fracture strength during solidification,[198] and how
tion modeling is still in its infancy. Most macrosegregation alloys affect hot tearing.[199] Difficulties include finding material
models have adopted a thermal-fluids modeling approach, properties, such as interface energies, which depend on the
solving mass and momentum transport equations for the fluid, models used to extract them, and the huge computational
combined with conservation and transport equations for the resources needed for fine-grid 3D domains of sufficient size.
solute, and including methodologies to handle the transport Thus, empirically-based models are very useful alternatives for
and removal of latent heat, the phase transformation from microstructure parameters such as dendrite arm spacing and
liquid to solid, movement of the solid, and a micro- grain size.[21] In conclusion, microstructure modeling deserves
segregation model.[193–195] Both liquid and solid are assumed more attention in the context of continuous casting of steel, to
incompressible. contribute to other models and/or to commercial practice.
A few macrosegregation models have been applied to
simulate fluid flow, solute transport, and segregation in the
mold region, to predict macrosegregation at the strand surface. 9. Conclusions
One recent model simulated the effect of electromagnetic
Significant progress has been made in the ability of computa-
stirring on turbulent fluid flow, superheat transport, free-surface
tional models to accurately predict fundamental phenomena in
motion, temperature, solidification, and solute distribution in a
the continuous casting process (nozzle, mold, and strand),
continuous bloom-casting mold with a bifurcated nozzle.[127]
including temperature, solidified shell growth, turbulent fluid
The stirring increased top surface level fluctuations, but also
flow together with multiphase phenomena, electromagnetic
caused faster removal of mold superheat, leading to a thinner
effects and particle transport, microstructure and grain
shell in the mold, and a slight segregation profile, with negative
structure, thermal-mechanical behavior, distortion, and stress.
segregation (less alloy) at the shell surface, which was roughly
Indeed, basic heat transfer models are being used as part of
the same trend as measurements of carbon content made near
model-based online control systems, for the control of spray
the bloom surface.
water flow rates and breakout-detection warning systems in the
Other thermal-fluid-solute models have focused on centerline
commercial process. Significant inroads are being made toward
segregation that develops lower in the caster, and include a
predictions under transient conditions, which are often when
model to prescribe bulging between the support rolls and
defects form. However, the accurate prediction of real defects
mechanical soft reduction.[193–195] The results show strong
that affect the cast product is still in its infancy. Model
positive segregation at the strand centerline, with nearby regions
predictions of clogging, inclusion entrapment, segregation, and
of negative segregation.
cracks need much further validation and testing. Finally, the
One multiphysics model used an ALE approach to simulate
coupling together of the different phenomena into multiphysics
the fluid flow, solute transport, and mechanical deformation of
models, including the vastly different length and time scales of
the solidifying shell and predicted both the bulging between rolls
interest will remain a challenge.
during secondary cooling, and the resulting macrosegregation
On the bright side, the increasing power of computer
near the centerline.[166] Other models are needed to include
hardware and modeling software has enabled significant
electromagnetics. More work is needed on quantitative valida-
advances, leading to models that incorporate many of the
tion of macrosegregation models in continuous casting.
important fundamental phenomena that are relevant to solving
Furthermore, the important effects on solute transport of the
practical problems. Validation of heat transfer and fluid-flow
columnar and equiaxed microstructure, including its local
models with plant measurements is routine, and progress is
deformation and failure, remain a challenge for macrosegrega-
being made on all aspects of the challenges mentioned above.
tion models to include in the future.
Future advances to the real commercial processes will require
intelligent combination of all tools available: plant experiments,
physical modeling, laboratory experiments, and increasingly:
8. Microstructure computational modeling.
Modeling microstructure during steel solidification is important
to augment the macroscale modeling of continuous casting,
especially in the prediction of defects such as cracks and Acknowledgements
segregation. The task of detailed simulation of microstructure The author thanks the member companies of the Continuous Casting
formation is daunting, however, as the evolving columnar- Center at the Colorado School of Mines and the National Science
equiaxed dendritic structure depends on all time and length Foundation Grant CMMI 15-63553 for funding to make this work possible.

steel research int. 2017, 1700312 1700312 (18 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

Thanks are also extended to Hyunjin Yang for help with the multiphase [27] C. Ji, C. H. Wu, M. Y. Zhu, JOM 2016, 68, 3107.
flow references. [28] H. H. An, Y. P. Bao, M. Wang, L. H. Zhao, Metall. Res. Technol. 2017,
114, 405.
[29] Z. W. Han, D. F. Chen, K. Feng, M. J. Long, ISIJ Int. 2010, 50, 1637.
Conflict of Interest [30] J. E. Kelly, K. P. Michalek, T. G. Oconnor, B. G. Thomas, J. A. Dantzig,
Metall. Trans. A 1988, 19, 2589.
The author declares no conflict of interest. [31] B. G. Thomas, A. Moitra, R. Mcdavid, Iron Steelmaker (ISS Trans)
1996, 23, 57.
[32] Y. Meng, B. G. Thomas, Metall. Mater. Trans. B 2003, 34, 707.
Keywords [33] Y. Hebi, Y. Man, Z. Huiying, F. Dacheng, ISIJ Int. 2006, 46,
546.
computational models, continuous casting, fluid flow, heat transfer, [34] S. Koric, L. C. Hibbeler, R. Liu, B. G. Thomas, Numer. Heat Transfer
review, solidification, steel, stress analysis, validation B-Fundam. 2010, 58, 371.
[35] M. R. Ridolfi, S. Fraschetti, A. De Vito, L. A. Ferro, Metall. Mater.
Received: July 23, 2017 Trans. B 2010, 41, 1293.
Revised: October 22, 2017 [36] M. Y. Zhu, Z. Z. Cai, H. Q. Yu, J. Iron Steel Res. Int. 2013,
Published online: 20, 6.
[37] Z. Z. Cai, M. Y. Zhu, ISIJ Int. 2013, 53, 1818.
[38] A. S. M. Jonayat, B. G. Thomas, Metall. Mater. Trans. B 2014,
45, 1842.
[1] World Steel in Figures 2016; World Steel Association, [Link]:
[39] C. Blaes, MS Thesis, Colorado School of Mines, 2017.
2016, p. 1.
[40] M. Jauhola, E. Livela, J. Konttinen, E. Laitinen, S. Louhenkilpi, Steel
[2] K. C. Mills, P. Ramirez-Lopez, P. D. Lee, B. Santillana, B. G. Thomas,
Transl. 1995, 25, 19.
R. Morales, Ironmaking Steelmaking 2014, 41, 242.
[41] R. A. Hardin, K. Liu, A. Kapoor, C. Beckermann, Metall. Mater. Trans.
[3] B. G. Thomas, Metall. Mater. Trans. B 2002, 33, 795.
B 2003, 34, 297.
[4] V. R. Voller, C. R. Swaminathan, B. G. Thomas, Int. J. Numer.
[42] A.-G. Hou, Q.-L. Zhang, G.-D. Xu, M.-F. Jiang, J. Iron Steel Res. Int.
Methods Eng. 1990, 30, 875.
2015, 22, 98.
[5] V. R. Voller, Adv. Numer. Heat Transfer 1997, 1, 341.
[43] B. G. Thomas, L. F. Zhang, ISIJ Int. 2001, 41, 1181.
[6] J. E. Lait, J. K. Brimacombe, F. Weinberg, Ironmaking Steelmaking
[44] Q. Yuan, B. Zhao, S. P. Vanka, B. G. Thomas, Steel Res. Int. 2005,
1974, 1, 90.
76, 33.
[7] R. B. Mahapatra, J. K. Brimacombe, I. V. Samarasekera, N. Walker,
[45] Ansys Fluent Theory Guide, Release 16.2. Ansys Inc., Canonsburg,
E. A. Paterson, J. D. Young, Metall. Trans. B 1991, 22B, 861.
PA, [Link]: 2016.
[8] C. A. M. Pinheiro, I. V. Samarasekera, J. K. Brimacombe,
[46] B. E. Launder, D. B. Spalding, Comput. Methods Appl. Mech. Eng.
B. N. Walker, Ironmaking Steelmaking 2000, 27, 37.
1974, 13, 269.
[9] J.-K. Park, B. G. Thomas, I. V. Samarasekera, U.-S. Yoon, Metall.
[47] R. Chaudhary, C. Ji, B. G. Thomas, S. P. Vanka, Metall. Mater. Trans.
Mater. Trans. B 2002, 33B, 437.
B 2011, 42, 987.
[10] B. Petrus, K. Zheng, X. Zhou, B. G. Thomas, J. Bentsman, Metall.
[48] C. Kratzsch, K. Timmel, S. Eckert, R. Schwarze, Steel Res. Int. 2015,
Mater. Trans. B 2011, 42, 87.
86, 400.
[11] M. L. S. Zappulla, B. G. Thomas, L. C. Hibbeler, Metall. Mater. Trans.
[49] R. Liu, PhD Thesis, University of Illinois at Urbana-Champaign,
A 2017, 48A, 3777.
2014.
[12] J. Yang, Z. Xie, J. Ning, W. H. Liu, Z. P. Ji, Metall. Mater. Trans. B
[50] R. Liu, B. G. Thomas, L. Kalra, T. Bhattacharya, A. Dasgupta, Iron
2014, 45, 1545.
Steel Technol. 2014, 11, 87.
[13] Y. Meng, B. G. Thomas, ISIJ Int. 2006, 46, 660.
[51] K. Jin, S. P. Vanka, R. K. Agarwal, B. G. Thomas, Int. J. Comput. Fluid
[14] J. Sengupta, M. Trinh, D. Currey, B. G. Thomas, in AISTech 2009
D 2017, 31, 36.
Steelmaking Conf. Proc., Louis, MO 2009.
[52] Q. Yuan, B. G. Thomas, S. P. Vanka, Metall. Mater. Trans. B 2004,
[15] J. X. Song, Z. Z. Cai, F. Y. Piao, M. Y. Zhu, J. Iron Steel Res. Int. 2014,
35, 685.
21, 1.
[16] X. D. Wang, L. W. Kong, M. Yao, X. B. Zhang, Metall. Mater. Trans. B [53] Q. Yuan, B. G. Thomas, S. P. Vanka, Metall. Mater. Trans. B 2004,
2017, 48, 357. 35, 703.
[17] J. Lee, H. Han, K. Oh, J. Yoon, ISIJ Int. 1999, 39, 435. [54] R. Singh, B. G. Thomas, S. P. Vanka, Metall. Mater. Trans. B 2014,
[18] X. D. Liu, M. Y. Zhu, ISIJ Int. 2006, 46, 1652. 45, 1098.
[19] F. M. Du, X. D. Wang, Y. Liu, T. Y. Li, M. Yao, J. Iron Steel Res. Int. [55] B. K. Li, Z. Q. Liu, F. S. Qi, F. Wang, G. D. Xu, Acta Metall. Sin. 2012,
2016, 23, 83. 48, 23.
[20] X. Liu, M. Zhu, ISIJ. Int. 2006, 46, 1652. [56] Z. Q. Liu, B. K. Li, M. F. Jiang, L. Zhang, G. D. Xu, Acta Metall. Sin.
[21] V. K. De Barcellos, V. L. D. Gschwenter, H. Kytonen, C. A. Dos 2013, 49, 513.
Santos, J. A. Spim, S. Louhenkilpi, J. Miettinen, Steel Res. Int. 2010, [57] Z. Liu, B. Li, M. Jiang, Metall. Mater. Trans. B 2014, 45, 675.
81, 461. [58] S. M. Cho, B. G. Thomas, S. H. Kim, Metall. Mater. Trans. B 2016,
[22] Y. Meng, B. G. Thomas, Metall. Mater. Trans. B 2003, 34, 685. 47, 3080.
[23] L. C. Hibbeler, M. M. C. See, J. Iwasaki, K. E. Swartz, R. J. O’malley, [59] K. Jin, S. P. Vanka, B. G. Thomas, Metall. Mater. Trans. B 2017,
B. G. Thomas, Appl. Math. Model. 2016, 40, 8530. 48, 162.
[24] B. Petrus, D. Hammon, M. Miller, B. Williams, A. Zewe, Z. Chen, [60] K. Jin, S. P. Vanka, B. G. Thomas, Metall. Mater. Trans. B 2017,
J. Bentsman, B. G. Thomas, Iron Steel Technol. 2015, 12, 58. (under revision) in press.
[25] M. J. Long, D. F. Chen, Q. X. Wang, D. H. Luo, Z. W. Han, Q. Liu, [61] Z. Liu, B. Li, Metall. Mater. Trans. B 2017, 48, 1833.
W. X. Gao, Ironmaking Steelmaking 2012, 39, 370. [62] H. Bai, B. G. Thomas, Metall. Mater. Trans. B 2001, 32, 253.
[26] J. Yang, Z. Xie, Z. P. Ji, H. J. Meng, ISIJ Int. 2014, 54, 328. [63] H. Bai, B. G. Thomas, Metall. Mater. Trans. B 2001, 32, 269.

steel research int. 2017, 1700312 1700312 (19 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

[64] H. Bai, B. G. Thomas, Metall. Mater. Trans. B 2001, 32, 707. Continuous Casting Consortium, University of Illinois, August 19,
[65] R. Chaudhary, G. G. Lee, B. G. Thomas, S. M. Cho, S. H. Kim, [Link] Models of Continuous Casting of Steel Slabs”,
O. D. Kwon, Metall. Mater. Trans. B 2011, 42, 300. BG Thomas, ed., Annual Report to Continuous Casting Consor-
[66] R. Liu, B. G. Thomas, J. Sengupta, S. D. Chung, M. Trinh, ISIJ Int. tium, University of Illinois, August 19, 2015
2014, 54, 2314. [101] R. Liu, B. G. Thomas, Metall. Mater. Trans. B 2015, 46, 388.
[67] L. L. Zhang, D. F. Chen, M. J. Long, H. B. Chen, Y. W. Huang, [102] J. Herbertson, Q. L. He, P. J. Flint, R. B. Mahapatra, in Steelmaking
Z. H. Dong, Metals-Basel 2016, 6, 104. Conf. Proc., Iron Steel Society (AIST), Washington, D.C. 1991,
[68] B. G. Thomas, in Making, Shaping and Treating of Steel: Continuous p. 171.
Casting, 11 edn., AISE Steel Foundation, Pittsburgh, PA 2003, [103] P. H. Dauby, Rev. Metall. 2012, 109, 113.
p. 14.1. [104] J. Kubota, K. Okimoto, A. Shirayama, H. Murakami, in Mold
[69] N. Bessho, R. Yoda, H. Yamasaki, T. Fujii, T. Nozaki, S. Takatori, ISIJ Operation for Quality and Productivity, Iron Steel Soc (AIST), 1991.
Int. 1991, 31, 40. [105] S. Feldbauer, A. Cramb, in PTD Conf. Proc., Iron Steel Soc. (AIST),
[70] B. G. Thomas, X. Huang, R. C. Sussman, Metall. Mater. Trans. B Warrendale, PA 1995, p. 327.
1994, 25, 527. [106] S. M. Lo, Application of Population Balance to CFD Modeling of
[71] D. Creech, M.S. Thesis, University of Illinois at Urbana-Champaign, Bubbly Flow via the MUSIG Model, AEAT-1096, AEA Technology,
Urbana, IL, 1999. Canonsburg, PA 1996.
[72] B. Li, T. Okane, T. Umeda, Metall. Mater. Trans. B 2000, 31, 1491. [107] Y.-S. Hwang, P.-R. Cha, H.-S. Nam, K.-H. Moon, J.-K. Yoon, ISIJ Int.
[73] T. Shi, M.S. Thesis, University of Illinois at Urbana-Champaign, 1997, 37, 659.
Urbana, IL, 2001. [108] A. Theodorakakos, G. Bergeles, Metall. Trans. B 1998, 29, 1321.
[74] Q. Yuan, T. Shi, S. P. Vanka, B. G. Thomas, in Computational [109] G. A. Panaras, A. Theodorakakos, G. Bergeles, Metall. Mater. Trans.
Modeling of Materials, Minerals and Metals Processing, TMS, B 1998, 29, 1117.
Warrendale, PA 2001, p. 491. [110] B. Rietow, B. G. Thomas, in AISTech 2008, Assoc. Iron Steel Tech.,
[75] T. Toh, H. Hasegawa, H. Harada, ISIJ Int. 2001, 41, 1245. Warrendale, PA Pittsburgh, PA 2008.
[76] R. Sánchez-Perez, L. García-Demedices, J. P. Ramos, M. Díaz-Cruz, [111] P. Zhao, Q. Li, S. B. Kuang, Z. S. Zou, Metall. Mater. Trans. B 2017,
R. D. Morales, Metall. Mater. Trans. B 2004, 35, 85. 48, 456.
[77] C. Pfeiler, M. Wu, A. Ludwig, Metall. Mater. Trans. A 2005, 413, 115. [112] Y. Wang, L. Zhang, ISIJ Int. 2010, 50, 1777.
[78] V. Singh, S. K. Dash, J. S. Sunitha, S. K. Ajmani, A. K. Das, ISIJ Int. [113] P. Ramírez-L opez, L. G. Demedices, O. Dávila, R. Sánchez-Perez,
2006, 46, 210. R. D. Morales, Metall. Mater. Trans. B 2005, 36, 787.
[79] C. Pfeiler, B. G. Thomas, M. Wu, A. Ludwig, A. Kharicha, Steel Res. [114] I. Calderon-Ramos, R. D. Morales, Metall. Mater. Trans. B 2016,
Int. 2008, 79, 599. 47, 1866.
[80] Z. Q. Liu, B. K. Li, M. F. Jiang, F. Tsukihashi, ISIJ Int. 2013, 53, 484. [115] P. Zhao, Q. Li, S. B. Kuang, Z. S. Zou, High Temp. Mater. Proc.-Isr.
[81] C. L. Liu, Z. G. Luo, T. Zhang, S. Deng, N. Wang, Z. S. Zou, J. Iron 2017, 36, 551.
Steel Res. Int. 2014, 21, 403. [116] Z. Q. Liu, Z. B. Sun, B. K. Li, Metall. Mater. Trans. B 2017, 48, 1248.
[82] Z. Q. Liu, F. S. Qi, B. K. Li, M. F. Jiang, J. Iron Steel Res. Int. 2014, 21, [117] K. Cukierski, B. G. Thomas, Metall. Mater. Trans. B 2008, 39, 94.
1081. [118] Y. F. Wang, L. F. Zhang, Metall. Mater. Trans. B 2011, 42, 1319.
[83] S. M. Cho, S. H. Kim, B. G. Thomas, ISIJ Int. 2014, 54, 845. [119] X. C. Miao, K. Timmel, D. Lucas, Z. M. Ren, S. Eckert, G. Gerbeth,
[84] P. E. Ramirez-Lopez, P. N. Jalali, J. Björkvall, U. Sjöström, Metall. Mater. Trans. B 2012, 43, 954.
C. Nilsson, ISIJ Int. 2014, 54, 342. [120] R. Chaudhary, B. G. Thomas, S. P. Vanka, Metall. Mater. Trans. B
[85] Z. Liu, F. Qi, B. Li, M. Jiang, Metall. Mater. Trans. B 2015, 46, 933. 2012, 43, 532.
[86] Z. Liu, L. Li, F. Qi, B. Li, M. Jiang, F. Tsukihashi, Metall. Mater. Trans. [121] R. Singh, B. G. Thomas, S. P. Vanka, Metall. Mater. Trans. B 2013,
B 2015, 46, 406. 44, 1201.
[87] L. M. Li, Z. Q. Liu, B. K. Li, J. Iron Steel Res. Int. 2015, 22, 30. [122] S. M. Cho, S. H. Kim, B. G. Thomas, ISIJ Int. 2014, 54, 855.
[88] Z. Q. Liu, F. S. Qi, B. K. Li, S. C. P. Cheung, Int. J. Multiphase Flow [123] Z. Q. Liu, L. M. Li, B. K. Li, JOM 2016, 68, 2180.
2016, 79, 190. [124] K. Jin, S. P. Vanka, B. G. Thomas, TMS Annual Meeting, Nashville,
[89] H. Bai, B. G. Thomas, Metall. Mater. Trans. B 2001, 32, 1143. TN, 2016, p. 159.
[90] K. G. Rackers, B. G. Thomas, in Continuous Casting Vol. 10 Tundish [125] E. Takeuchi, M. Zeze, T. Toh, S. Mizoguchi, Magnetohydrodynamics
Operations, Iron Steel Society, Warrendale, PA 2003, p. 264. Proc. Met, Szekely, J., Ed. TMS, Warrendale, PA 1991.
[91] S. Z. Wu, J. M. Zhang, Z. Z. Li, J. Iron Steel Res. Int. 2010, 17, 6. [126] A. Maurya, P. K. Jha, Appl. Math. Model. 2017, 48, 736.
[92] E. Gutierrez, S. Garcia-Hernandez, J. D. J. Barreto, Steel Res. Int. [127] Q. Fang, H. W. Ni, B. Wang, H. Zhang, F. Ye, Metals-Basel 2017, 7,
2016, 87, 1406. 72.
[93] J. Knoepke, M. Hubbard, J. Kelly, R. Kittridge, J. Lucas, in [128] K. Timmel, S. Eckert, G. Gerbeth, Metall. Mater. Trans. B 2011,
Steelmaking Conf. Proc., Iron Steel Society, Chicago, IL 1994, p. 381. 42, 68.
[94] B. G. Thomas, A. Dennisov, H. Bai, in 80th Steelmaking Conference [129] B. G. Thomas, R. Singh, S. P. Vanka, K. Timmel, S. Eckert,
Proc., Iron and Steel Society (AIST), Warrendale, PA 1997, p. 375. G. Gerbeth, J. Manuf. Sci. Prod. 2015, 15, 93.
[95] M. Long, X. Zuo, L. Zhang, D. Chen, ISIJ Int. 2010, 50, 712. [130] K. Jin, B. G. Thomas, X. M. Ruan, Metall. Mater. Trans. B 2016,
[96] L. C. Hibbeler, B. G. Thomas, R. C. Schimmel, H. H. Visser, in 47, 548.
ECCC2014, ASMET, Leoben, Austria 2014, p. 675. [131] X. Huang, B. G. Thomas, F. M. Najjar, Metall. Trans. B 1992, 23, 339.
[97] R. B. Tuttle, J. D. Smith, K. D. Peaslee, Metall. Mater. Trans. B 2007, [132] B. Zhao, B. G. Thomas, S. P. Vanka, R. J. O’malley, Metall. Mater.
38, 101. Trans. B 2005, 36B, 801.
[98] L. Wang, C. Beckermann, Metall. Mater. Trans. B 2006, 37, 571. [133] J. Ni, C. Beckermann, Metall. Trans. B 1991, 22, 349.
[99] F. M. White, Fluid Mechanics, 7 edn., McGraw Hill, International [134] C. Y. Wang, C. Beckermann, Metall. Mater. Trans. A 1996, 27, 2754.
2011. [135] V. R. Voller, C. Prakash, Int. J. Heat Mass Tranfer 1987, 30, 1709.
[100] H. Yang, B. G. Thomas, “Mathematical Models of Continuous [136] L. L. Zhang, D. F. Chen, H. B. Chen, M. J. Long, X. Xie, Ironmaking
Casting of Steel Slabs”, BG Thomas, ed., Annual Report to Steelmaking 2017, 44, 193.

steel research int. 2017, 1700312 1700312 (20 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
[Link] [Link]

[137] Q. Q. Wang, L. F. Zhang, JOM 2016, 68, 2170. [168] P. F. Kozlowski, B. G. Thomas, J. A. Azzi, W. Hao, Metall. Trans. A
[138] M. J. Long, D. F. Chen, L. F. Zhang, Y. Zhao, Q. Liu, Met. Int. 2011, 1992, 23, 903.
16, 19. [169] S. Koric, B. G. Thomas, J. Manuf. Proc. Technol. 2008, 197, 408.
[139] Y. F. Wang, A. P. Dong, L. F. Zhang, Steel Res. Int. 2011, 82, 428. [170] M. R. Ridolfi, Metall. Mater. Trans. B 2014, 45, 1425.
[140] B. G. Thomas, Q. Yuan, S. Mahmood, R. Liu, R. Chaudhary, Metall. [171] J. H. Weiner, B. A. Boley, J. Mech. Phys. Solids 1963, 11, 145.
Mater. Trans. B 2014, 45, 22. [172] M. Rowan, B. G. Thomas, R. Pierer, C. Bernhard, Metall. Mater.
[141] Z. Q. Liu, L. M. Li, B. K. Li, M. F. Jiang, JOM 2014, 66, 1184. Trans. B 2011, 42, 837.
[142] L. Zhang, Y. F. Wang, JOM 2012, 64, 1063. [173] C. L. Zhang, M. Bellet, M. Bobadilla, H. F. Shen, B. C. Liu, Metall.
[143] L. Zhang, J. Aoki, B. G. Thomas, Metall. Mater. Trans. B 2006, Mater. Trans. A 2010, 41A, 2304.
37, 361. [174] B. G. Thomas, A. Moitra, R. McDavid, in Continuous Casting, Iron
[144] Y. G. Xu, M. Ersson, P. G. Jonsson, ISIJ Int. 2016, 56, 1982. Steel Soc, (AIST), Warrendale, PA 1997, p. 337.
[145] P. E. Ramirez-Lopez, P. D. Lee, K. C. Mills, ISIJ Int. 2010, 50, 425. [175] S. Kumar, J. A. Meech, I. V. Samarasekera, J. K. Brimacombe,
[146] P. E. Ramirez-Lopez, P. D. Lee, K. C. Mills, B. Santillana, ISIJ Int. V. Rakocevic, Ironmaking Steelmaking 1999, 26, 269.
2010, 50, 1797. [176] G. Poltarak, S. Ferro, C. Cicutti, Steel Res. Int. 2017, 88. https//doi.
[147] P. E. Ramirez-Lopez, K. C. Mills, P. D. Lee, B. Santillana, Metall. org/10.1002/srin.201600223
Mater. Trans. B 2012, 43, 109. [177] M. Bellet, A. Heinrich, ISIJ Int. 2004, 44, 1686.
[148] P. D. Lee, P. E. Ramirez-Lopez, K. C. Mills, B. Santillana, Ironmaking [178] N. Triolet, M. Bobadilla, M. Bellet, L. Avedian, P. Mabelly, Rev.
Steelmaking 2012, 39, 244. Metall. 2005, 102, 343.
[149] X. Yan, A. Jonayat, B. G. Thomas, in TMS Annual Meeting, Frontiers [179] S. Koric, B. G. Thomas, V. R. Voller, Numer. Heat Transfer
in Solidification, TMS, Nashville, TN 2016, p. 181. B-Fundam. 2010, 57, 396.
[150] J. Sengupta, B. G. Thomas, H. J. Shin, G. G. Lee, S. H. Kim, Metall. [180] J. D. Lee, C. H. Yim, ISIJ Int. 2000, 40, 765.
Mater. Trans. A 2006, 37A, 1597. [181] K. Toishi, Y. Miki, ISIJ Int. 2016, 56, 1764.
[151] B. G. Thomas, H. Zhu, in Proceedings of Internat. Symposia on [182] K. Liu, Q. S. Sun, J. Q. Zhang, C. Wang, Metall. Res. Technol. 2016,
Advanced Materials & Tech. for 21st Century, Ohnaka, I., Stefanescu, 113. https//[Link]/10.1051/metal/2016012
D., Eds. TMS, Warrendale, PA: Honolulu, HI, 1996, pp 197–208. [183] K. Liu, C. Wang, G. L. Liu, N. Ding, Q. S. Sun, Z. H. Tian, High Temp.
[152] B. G. Thomas, J. T. Parkman, in Solidification 1998, 1998, 509. Mater. Proc.-Isr. 2017, 36, 359.
[152] B. G. Thomas, J. T. Parkman, in Solidification 1998, 1998, 509. [184] A. Yamanaka, K. Nakajima, K. Okamura, Ironmaking Steelmaking
[153] C. Ojeda, J. Sengupta, B. G. Thomas, J. Barco, J. Arana, Luis in 1995, 22, 508.
AISTech 2006 Steelmaking Conf. Proc., AIST, Warrendale, PA 2006, p. [185] Y. M. Won, T. J. Yeo, D. J. Seol, K. H. Oh, Metall. Mater. Trans. B
1017. 2000, 31, 779.
[154] J. Sengupta, C. Ojeda, B. G. Thomas, Int. J. Cast Met. Res. 2009, 22, 8. [186] M. Bellet, O. Cerri, M. Bobadilla, Y. Chastel, Metall. Mater. Trans. A
[155] X. J. Zuo, R. G. Lin, N. Wang, J. Yang, X. N. Meng, M. Y. Zhu, Steel 2009, 40A, 2705.
Res. Int. 2016, 87, 413. [187] T. Koshikawa, M. Bellet, C. A. Gandin, H. Yamamura, M. Bobadilla,
[156] T. O’conner, J. Dantzig, Metall. Mater. Trans. B. 1994, 25B, 443. Metall. Mater. Trans. A 2016, 47A, 4053.
[157] I. V. Samarasekera, D. L. Anderson, J. K. Brimacombe, Metall. Trans. [188] M. Bellet, G. H. Qiu, J. M. Carpreau, J. Mater. Process. Technol. 2016,
B 1982, 13, 91. 230, 143.
[158] I. V. Samarasekera, J. K. Brimacombe, Ironmaking Steelmaking 1982, [189] B. G. Thomas, M. Bellet, in ASM Handbook, ASM International,
9, 1. Materials Park, OH 2008, p. 449.
[159] B. G. Thomas, G. Li, A. Moitra, D. Habing, I&SM (ISS Trans.) 1998, [190] Y. M. Won, B. G. Thomas, Metall. Mater. Trans. A 2001, 32, 1755.
25, 125. [191] D. L. You, C. Bernhard, G. Wieser, S. Michelic, Steel Res. Int. 2016,
[160] L. C. Hibbeler, S. Koric, K. Zhu, B. G. Thomas, C. Spangler, Iron Steel 87, 840.
Technol. 2008, 6, 60. [192] C. Beckermann, Int. Mater. Rev. 2002, 47, 243.
[161] J.-K. Park, B. G. Thomas, I. V. Samarasekera, U.-S. Yoon, Metall. [193] F. Mayer, M. Wu, A. Ludwig, Steel Res. Int. 2010, 81, 660.
Mater. Trans. B 2002, 33B, 425. [194] M. H. Wu, J. Domitner, A. Ludwig, Metall. Mater. Trans. A 2012, 43A,
[162] L. C. Hibbeler, B. G. Thomas, R. C. Schimmel, G. Abbel, Metall. 945.
Mater. Trans. B 2012, 43, 1156. [195] J. Domitner, M. H. Wu, A. Kharicha, A. Ludwig, B. Kaufmann,
[163] Abaqus, Standard User Manuals v6.4. Abaqus, Inc.: Providence, RI J. Reiter, T. Schaden, Metall. Mater. Trans. A 2014, 45A, 1415.
2004. [196] S. Luo, M. Y. Zhu, S. Louhenkilpi, ISIJ Int. 2012, 52, 823.
[164] C. Li, B. G. Thomas, Metall. Mater. Trans. B 2004, 35, 1151. [197] Y. Zhao, R. S. Qin, D. F. Chen, X. M. Wan, Y. Li, M. T. Ma, Steel Res.
[165] L. C. Hibbeler, B. G. Thomas, B. Santillana, A. Hamoen, Int. 2015, 86, 1490.
A. Kamperman, Metall. Ital. 2009, 29. [198] D. J. Seol, K. H. Oh, J. W. Cho, J. E. Lee, U. S. Yoon, Acta Mater. 2002,
[166] V. D. Fachinotti, S. Le Corre, N. Triolet, M. Bobadilla, M. Bellet, 50, 2259.
Int. J. Numer. Methods Eng. 2006, 67, 1341. [199] B. Bottger, M. Apel, B. Santillana, D. G. Eskin, Metall. Mater. Trans.
[167] C. L. Zhang, M. Bellet, M. Bobadilla, H. F. Shen, B. C. Liu, Acta A 2013, 44A, 3765.
Metall. Sin. 2010, 46, 1206. [200] M. Yoshiya, M. Watanabe, K. Nakajima, N. Ueshima,
K. Hashimoto, T. Nagira, H. Yasuda, Mater. Trans. 2015, 56, 1467.

steel research int. 2017, 1700312 1700312 (21 of 21) © 2017 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim