0263–8762/04/$30.00+0.

00
# 2004 Institution of Chemical Engineers
[Link]=titles=[Link] Trans IChemE, Part A, April 2004
Chemical Engineering Research and Design, 82(A4): 499–512

USING PROCESS SIMULATORS FOR STEADY-STATE AND

DYNAMIC PLANT ANALYSIS
An Industrial Case Study
F. BEZZO1, R. BERNARDI2, G. CREMONESE2, M. FINCO2 and M. BAROLO1,*
1
DIPIC, Dipartimento di Principi e Impianti di Ingegneria Chimica, Università di Padova, Padova, Italy
2
European Vinyls Corporation Italia S.p.A., Porto Marghera, Italy

P
rocess simulation tools are widely adopted for the design and optimization of chemical
processes. However, for quite a long time their use has been confined within research
centres and highly specialized technical groups. This is especially true for dynamic
simulation software, long regarded as a very specific tool requiring considerable expertise. In
this work we intend to demonstrate the benefits that process engineers working on the plant
may receive from an appropriate use of commercial software currently available for steady-state
and dynamic simulation. A case-study concerning the purification section of an industrial plant
for vinyl chloride monomer production will be considered. First of all, a steady-state simulation
will be considered. Primarily, the simulation will allow a better judgement of the plant
operating conditions; then it will be illustrated that sensitivity studies may produce great
benefits in the general economy and productivity of the plant. Secondly, it will be shown how a
dynamic model suitable for practical needs can be derived from the steady-state model. This
model can be used as a powerful tool to assess the performance of the control system in
handling standard operational disturbances as well as abnormal events. Simple improvements
of the control system design will be also simulated and commented on.

Keywords: process simulation; dynamic analysis; distillation control; vinyl chloride; VCM.

INTRODUCTION demanding task and that even the usage of simulation

tools (particularly when dynamic simulation is being consid-
Process simulation is a well-established tool in the process ered) can be rather complicated. Nonetheless, in this paper
industry, being used to study individual unit operations as we would like to demonstrate how steady-state and even
well as multiple interconnected units or even entire plants. dynamic simulations can deliver great benefits at the plant
The latest generation of process simulation software is very level and can be used by plant engineers to better understand
flexible, being able to represent a very wide range of multi- the process behaviour in such a way as to improve the
component, multiphase and reactive systems (Pantelides and steady-state and dynamic performance of the plant. In the
Britt, 1995; Marquardt, 1996). Process simulation is not literature, it is possible to find examples where steady-
only used for steady-state simulation, but dynamic model- state and dynamic modelling have been used to improve
ling is also commonly applied for the design and optimiza- unit operation and control scheme design, even in industrial
tion of chemical processes. In particular, the use of dynamic cases (e.g. Roat et al., 1986; Lausch et al., 1998), but there
models is a powerful tool for the design of control systems is very little information concerning the applicability of
and to verify their effectiveness. process simulation tools, particularly dynamic ones, at the
The use of model-based methods for the support of the very plant level for the management and improvement of the
synthesis, design and operation of chemical processes and process behaviour (apart for some specific examples for
plants has recently gained increasing importance in the daily operator training like in the paper by Bretelle et al., 1994).
work of process engineers, although it is becoming more However, it should be noted that the process industry has
obvious that setting up mathematical models by the use of already started to move towards a more direct and basic
standard flowsheet simulation tools requires an experienced application of modelling tools and, following this, some
designer (Eggersmann et al., 2002). In general, it is true that engineering organizations have set up a number of guide-
the development of unit operation models is always a lines for the use of computer software in the design of
process plants (IChemE CAPE Subject Group, 1999).
*Correspondence to: Professor M. Barolo, DIPIC—Dipartimento di Prin-
cipi e Impianti di Ingegneria Chimica, Università di Padova, via Marzolo 9, The main objective of this paper is to fill a gap that exists
35131 Padova, PD, Italy. in the current scientific literature by illustrating the advan-
E-mail: [Link]@[Link] tages that can derive from using state-of-the-art commercial

499
500 BEZZO et al.

simulation tools at the plant level. To this purpose, a case at a different load than the design one. From the dynamic
study concerning the purification section of an industrial viewpoint, an assessment of the performance of the current
plant for vinyl chloride monomer (VCM) production will be control system was sought, and possible improvements to the
considered. Furthermore, it is our aim to deliver some prac- control configuration were fostered. Moreover, the dynamic
tical guidelines for the set-up and use of steady-state and response of the plant to abnormal operating conditions
dynamic process simulation tools at the plant level. This (such as those arising when one of the oxy-chlorination
may be helpful in outlining critical steps in the model reactors runs out of service) was required. Note that,
development path and in limiting typical reliability issues although these issues are related to the VCM plant under
of simulation results (Kister, 2002). Although the results are study, they are quite general, and most of them could also
related to the simulation of a specific example, we never- apply to other production units. Therefore, this case study
theless believe that the approach that will be described is can be considered a meaningful test bed for a general
quite general, so that the reasoning and conclusions that will approach to the use of process simulators within an
be discussed can be easily extended to the analysis of most industrial setting.
process operations involving continuous separation units. The first column of the train (column C-501) is a 45-real-
tray column fed by three streams (top, mid, and bottom
feed). The feeds come from a set of cracking reactors, where
THE PROCESS
ethylene 1,2-dichloride (EDC) is cracked into VCM and
The steady-state and dynamic simulation of the purifica- hydrogen chloride (HCl). Therefore, the three most impor-
tion section of an industrial plant for the production of VCM tant chemicals in these feeds are HCl, VCM and EDC,
is considered. This plant section basically comprises a train which in all account for 98þ% by weight of the feeds. A
of four distillation columns that purify the VCM (available at careful analysis with the plant engineers revealed that nine

32.5% by weight from the ethylene dichloride cracking further components are important for the evaluation of the
furnaces) to a polymer grade (99.99þ%). The annual pro- plant performance (mainly because these components act as
duction is
250,000 tons of polymer-grade VCM. Figure 1 tracers that can aid understanding the general efficiency of
illustrates the four distillation columns of the purification the VCM production and separation processes), and for this
section. Several issues were open for investigation before the reason these components were included in the simulation.
simulation work was started. One quite general issue Therefore, a 12-component feed was eventually considered
was related to the improvement of plant understanding. A (Table 1).
somewhat cascaded one was the improvement of the plant The gaseous HCl stripped from the top of C-501
operating conditions. A further issue is related to the evalua- (
99.8 wt%) is sent to the oxy-chlorination reactors for EDC
tion of the approach to equipment hydraulic limits; this production, while the column bottoms are fed to column
issue was quite severe, since the train is currently operated C-502 (70 trays), where the main separation between VCM

Figure 1. Simplified process flow diagram.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
 STEADY STATE AND DYNAMIC PLANT ANALYSIS 501
Table 1. Components
considered in this
is carried out by non-expert users, who may otherwise be
simulation. tempted to unwisely modify the unit models. Besides, note
that the training of plant engineers is more effective if
Components approached in terms of unit blocks without moving into
Hydrogen chloride the complicated field of advanced process modelling.
Vinyl chloride Indeed, the modelling activity should not go beyond the
Ethylene 1,1-dichloride connection of different unit operations and the specifications
Ethylene 1,2-dichloride of their operating conditions. From this point of view; the
1,3-Butadiene
Benzene interface should allow a clear representation of unit opera-
Ethyl chloride tions, in terms of model characteristics and input=output
Methyl chloride declarations. The process flow sheet should be visualized as
Acetylene the engineers see it in the real plant. This will greatly
Tetrachloroethylene
Trichloroethylene
facilitate the understanding of the simulation process as
Carbon tetrachloride well as the critical evaluation of the results deriving from it.
The last important issue concerns the dynamic simulation.
This should be easy to run and to derive from the steady-
state model. Ideally, the same tool should be able to perform
(distillate) and EDC (bottoms) takes place. The purified VCM both the steady-state and the dynamic simulation without
is fed to the top tray of C-504 (30 trays), where polymer- changing the simulation environment. Once the modelling
grade VCM is eventually obtained as the bottom product; specifications have been fulfilled, the simulator should be
the distillate of this column is recycled back to C-501. easily switched from a steady-state to a dynamic representa-
Column C-503 is a 65-real-tray column used to purify tion of the events.
EDC from several byproducts (particularly trichloroethylene) For the case study under investigation, the company
before further purification and recycling to the cracking involved in the VCM project owned an Aspen Plus1 v.11.1
furnaces for VCM production. Only columns C-501, C-502 and an Aspen Dynamics1 licence. Both simulators
and C-504 will be considered in this paper. performed reasonably well and satisfied most of the above
requirements. However, we point out that we cannot exclude
(actually, we are quite confident) that other commercial
simulators may have performed equally well.
PROCESS SIMULATION TOOLS
A large variety of tools is available for steady-state and
dynamic simulation. The description of the specific char-
acteristics of these tools is beyond the scope of this paper. STEADY-STATE ANALYSIS
An exhaustive review of the features of state-of-the-art
The Thermodynamic Model
simulators can be found, for example, in the papers by
Biegler (1989), Pantelides and Britt (l995) and Marquardt As is well known, the accuracy of a distillation column
(1996). After noting that a fundamental requirement is that model is greatly affected by the description of the system
any process simulation tool must rely on a robust solver phase equilibria. The choice of a suitable thermodynamic
capable of handling dynamic and steady-state sets of model- model depends on the nature of the components to be sepa-
ling equations, we intend to point out a number of char- rated and on the plant operating conditions. For the case
acteristics that a process simulator should comply with when under study, most of the component pairs exhibit quite
used at a process plant level. strong liquid-phase interactions. Furthermore, column C-501
First of all the simulator should be easy to use. Although operates at a pressure above 10 bar, which calls for the use
the actual modelling task may be carried out by a modelling of an equation-of-state (EOS) approach for the calculation
expert, it is important that the simulation can be run even of vapour–liquid equilibria. The rest of the train operates at
without a detailed knowledge of the system equations. 5 bar or less, which makes the use of a g–j approach more
Furthermore, the implementation of minor changes in the sound. Confirming the indications of Pilavachi et al. (2000),
unit operation models (e.g. variations in operating condi- we would like to stress that a key issue that must be
tions and connectivity between units) and the performance considered for accurate thermodynamic modelling is the
of sensitivity studies should always be a relatively simple calculation of the pure-component vapour pressures. Indeed,
task. From this point of view, a simulator adopting a for some components we experienced several difficulties in
standard library of unit operation models is advisable. achieving a correct evaluation of vapour pressure by using
User-defined models may be more powerful, but sometimes an EOS [namely, the Soave–Redlich–Kwong (SRK) EOS
display a lack of standardization that may complicate their was selected], although this was not quite unexpected since
usage; also they are often short of proper documentation. In it is well known that the SRK EOS may prove defective for
general, at least as a first attempt, it is good practice to adapt highly polar components. The wrong estimation of vapour
somewhat the process description by using standard library pressures may lead to significant errors in the description
models rather than setting up an ad hoc model. Although it of binary vapour–liquid equilibria. Consider, for instance,
is widely recognized that the use of a library of ‘non- the EDC–benzene and EDC–trichloroethylene binaries:
transparent’ models will cause a lack of modelling flexibility Figure 2 compares the SRK predictions with those of an
and will not adequately support the solution of more NRTL model whose interaction parameters had been
involved problems (Marquardt, 1996), our experience previously adjusted on available data. It is clear that the
suggests that this is usually an advantage when a simulation inaccurate estimation of EDC vapour pressure leads to large

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
502 BEZZO et al.

Figure 2. EDC-benzene (a) and EDC-trichloroethylene (b) binaries: comparison between SRK and NRTL vapour–liquid equilibria calculations.

errors in the phase-equilibria calculations, and these errors and (b) for columns C-501 and C-502, respectively (trays are
cannot be compensated for by optimally selecting the EOS numbered from the bottom up). It can be seen that the
binary interaction parameters. calculated temperatures match quite satisfactorily the plant
The calculation of the attractive parameter through an ones (the temperature controller setpoints were used as
a-extended approach (Schwartzentruber et al., 1989) even- design specifications for the steady-state simulation).
tually allowed a very accurate calculation of vapour pres- Discrepancies at the feed trays are usually tolerated since
sures. The binary interaction parameters were obtained either the location of temperature probes on these trays may not
from the process simulator database, or from regression of be optimal, so that a tray temperature measurement may be
proprietary experimental data (however, note that some data affected by the feed temperature. Note that, since almost
had to be discarded due to clear inconsistencies in the pure HCl is found in the plant at the top of C-501, the
experimental vapour pressures). For some binaries, neither temperature there is dictated by the HCl vapour pressure
databank properties nor experimental data were available. only; this in turn confirms the accuracy of the HCl vapour
For these binaries, the EOS interaction parameters were pressure calculations by the SRK EOS.
determined by fitting ‘pseudo-experimental’ data generated A further check on the accuracy of the model can be (and
with a predictive Soave–Redlich–Kwong equation of state actually was) carried out by comparing the reboiler and
(Holderbaum and Gmehling, 1991). The use of predictive condenser duties predicted by the model with those calcu-
methods for the estimation of the SRK EOS parameters also lated on the basis of available plant data. Taking into
persuaded us of the unsuitability of adopting a single account the inaccuracies that invariably affect industrial
thermodynamic model for the process simulation. On the measurements, a match of the enthalpy balances within
one hand, an EOS approach was required for a consistent 10% is generally enough, and ensures that the internal
description of high pressure equilibria; on the other, it would flows calculated by the simulator are in agreement with
be rather unwise to discharge a solid thermodynamic the actual ones (an issue that is quite important for correct
approach (NRTL) based on experimental data. evaluation of tray hydraulics). In some cases, the condenser
On the basis of the experience that we have gained in the duty cannot be calculated directly from plant data; to
simulation of this and several other industrial plants with circumvent this problem, the actual (if available) and calcu-
different process simulators, we estimate that the time that is lated reflux flows can be compared directly to get an indirect
required to perform the thermodynamic modelling accounts assessment of the enthalpy balance matching (this was
for about 40–60% of the whole time dedicated to a steady- actually done for columns C-501 and C-502).
state simulation project. We would like to stress that, for the During the model validation phase, the interaction with
description of the system thermodynamics, the support of an plant operators is of paramount importance. Operators can
expert is highly advisable. guide the engineers through the analysis of process opera-
tion data sheets; they can highlight which plant measure-
ments can be trusted and which cannot; knowing the
Model Validation and Simulation Results operators’ ‘tricks’ to make the plant perform better can be
In the following, some of the results concerning the very useful to fine-tune the model. For example, by talking
steady-state simulation are briefly presented. After verifying to some operators we became aware that, to reduce the
the closure of heat and mass balances,* the first issue that contamination of purified VCM by 1,3-butadiene, from time
needs to be considered is the assessment of the accuracy of to time they drew off this impurity from one certain tray of
the simulation model. This can be tested in different ways. column C-502. This information was extremely useful to
One way is to evaluate the accuracy of reproduction of check the ‘tuning’ of the model; in fact, a peak in the liquid-
the column internal temperature profiles, at least for those phase 1,3-butadiene concentration was indeed found by
columns where the temperature gradient between the top and simulation on that tray of C-502 (Figure 4), which indirectly
the bottom is significant. This test is shown in Figure 3(a) confirmed the model reliability.
In general, our experience reveals that steady state model
* This is a good practice even if no error messages are reported by the validation may roughly take 20–30% of the total time
simulator, as pointed out by Le et al. (2000). allotted to the steady-state simulation project.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
 STEADY STATE AND DYNAMIC PLANT ANALYSIS 503

Figure 3. Internal temperature profiles in columns C-501 (a) and C-502 (b); the feed trays are indicated in brackets.

Improving Plant Performance

Once the steady-state model has been validated, it can be
used to improve the plant performance. As mentioned
above, the plant is currently run at different operating
conditions than the design ones. Therefore, it was very
important to estimate how close to the limiting conditions
(flooding and=or weeping) the columns were operated.
Although most of the results that we have achieved cannot
be disclosed for confidentiality reasons, some of them can
nevertheless be mentioned and can be of general interest.
Figure 5 shows the profile of the tray flooding factor for
column C-501 as calculated by the process simulator.
Although it appears that flooding is not a problem for this
column, it is also noticeable that the column section between Figure 5. Profile of the flooding factor for column C-501.
the top feed (tray 34) and the mid feed (tray 26) is quite
‘unloaded’ (the flooding factor is lower than 20%). This is
indirect evidence that weeping may occur in that section (it Further analyses were considered to assess the process
should be noted that the selected simulator cannot perform performance under different operating conditions. In parti-
weeping calculations directly). Further investigation with cular, a study concerning C-502 operating conditions was
the tray vendor (Koch–Glitsch) confirmed this conjecture, carried out in the attempt to reduce the column’s consider-
and locking some tray valves in that column section was able energy demand. A sensitivity analysis was performed
found to be a viable way to increase the vapour velocity and by varying the reflux ratio. It appeared that the reflux ratio
counteract the problem. Tray hydraulics can significantly can be slightly decreased with respect to the present value,
affect the performance of distillation columns, and it is not and this was shown to reduce the energy consumption by an
uncommon for some process equipment malfunctioning to amount equivalent to a saving of
50,000 euros per year.
be related to incorrect hydraulic behaviour. Process simula-
tion may help in detecting potential issues which, however,
should be further investigated (e.g. by consulting vendors DYNAMIC ANALYSIS
and company experts). Steady-state simulation may help answer several issues
concerning plant operating conditions and may give helpful
hints about the control system design and response. Some
considerations on control loop performance may be
retrieved by steady-state indicators, such as the steady-
state relative gain array (RGA). However, it has long been
recognized that such measures do not always provide reli-
able information to guide the control system designer
(McAvoy, 1983). A better understanding of the controller
performances may be obtained by using dynamic controll-
ability and resiliency measures (Seider et al., 1999). Unfor-
tunately, these measures are not automatically generated by
available commercial simulators, and their estimation is in
general a rather demanding task requiring a deep knowledge
of the process models. Therefore, plant engineers are
advised to assess the process resiliency to external distur-
bances and the control system performance by direct use of
Figure 4. Profile of 1,3-butadine mass fraction in column C-502. dynamic simulations.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
504 BEZZO et al.

Although the basis for the development of a dynamic the pilot tray is located in the middle of the temperature
model is the steady-state one, a number of parameters need break, and this allows easy detection of any losses of VCM
to be specified in order to run a dynamic simulation. In fact, from the bottom. Moreover, the column clearly appears to be
the actual size of most of the column and ancillary equip- non-optimally fed. In fact, this is a sound choice from the
ment must be assigned. Particular care must be taken in the control standpoint: the pinch point just below the feed
calculation of hold-ups, which greatly affect the dynamic protects the top (VCM) from the composition disturbances
response of the plant. Usually, a balance must be struck travelling in the stripping section, and allows to use a single
between the level of detail included in the model and the TC to control the purity of both products. Since the reflux
model speed of running. In fact, although available simula- rate is constant, the distillate flow is more stable, which in
tors allow to include several detailed calculations (i.e. turn stabilizes the operation of column C-504.
pressure losses due to friction, heat exchanger dynamics,
pump and valve characteristic equations, etc.), too detailed a
Assumptions for a Dynamic Simulation
model may result impractical (or even impossible) to run in
a reasonable time at the plant level. Furthermore, if complex The dynamic model was derived from the steady-state
models may ensure a very good description of all process model described above. No change was made in the descrip-
phenomena, this may nevertheless raise convergence issues. tion of the system thermodynamics. However, as mentioned
Because the scope is the definition of an easy-to-use model before, some further assumptions are required for a dynamic
that can be used by plant engineers to check the process representation of distillation columns. Ancillary equipment
behaviour and to improve process operation and control, (reboilers, condensers) can be modelled either by adopting a
numerical stability and ease of manipulating the model are very detailed geometry or by considering the essential heat
requirements as important as the model capability of captur- exchanger information (type of exchanger, duty, etc.). It was
ing the fundamental process dynamics. verified that the use of very detailed models may produce
A critical step is the definition of the tuning constants of severe convergence issues when the entire set of columns is
control loops. In the plant under study, all the control loops run simultaneously (at least in the sequential modular
were centred on standard PID controllers, but the unit mode). Besides, the computational burden may become so
dimensions of gains and of integral and derivative times heavy as to discourage a frequent use of the simulation
provided by the operators had to be double-checked for tools. This strongly suggested that the use of simplified
correct reproduction in the simulator (potential errors may dynamic models capable of describing the fundamental
arise from incorrect evaluation of transmitter spans and dynamics as well as capturing the steady-state column
valve gains). profiles should be preferred to more rigorous modelling
The control configurations for the three columns are approaches that may generate numerical issues and hinder a
illustrated in Figure 6. Product compositions in column plant engineer from correctly running the simulation. In
C-501 are indirectly controlled by means of two temperature particular, in the VCM plant being examined the following
controllers (TJCA-501 and TICA-502 at trays 14 and 38) assumptions were made:
acting on the reboiler duty and the reflux rate, respectively.
 condenser holdups are negligible compared to reflux
The condenser duty is used to control the pressure in the
drum holdups;
large reflux vessel. This choice makes the vapour distillate
 reboiler holdups are incorporated in column sumps;
available for manipulation by a different controller, namely
 pipe holdups are negligible.
PICA-102, whose setpoint is driven by the HCl demand at
the oxy-chlorination reactors. We will comment on this Some tests were run to quantify the effect of the type of
control scheme further on. The sump level is controlled by model being adopted. For example, a step disturbance on
manipulating the bottoms flowrate. flowrate of all feed streams was delivered. This test demon-
As far as colunm C-502 is concerned, it is worth noting strated that the time responses of the most important process
that there is only one temperature controller (TRC-503) variables are not significantly affected by the accuracy in the
placed close to the bottom of the column (tray 5). The description of the ancillary equipment.
desired reflux rate is enforced by means of a flow controller. The ‘tuning’ of the model may take over 40% of the
No composition control is considered for column C-504; overall time of a dynamic simulation project (if it is assumed
stable working conditions in column C-504 are ensured by that the dynamic model may be easily retrieved from
minimizing the upsets to the feed and keeping the tempera- previous steady-state description).
ture of steam to the reboiler constant. Note that the reflux
drum level controller in C-502 is loosely tuned, which
Assessing the Control Performance
guarantees that the feed flow disturbances to C-504 are
kept to a minimum. A typical issue concerning the process under investigation
derives from the variability in the conversion of the EDC
cracking reactions. A change of EDC conversion determines
Information from Steady-State Analysis
a change in the flowrate and composition of all the feeds
Some preliminary information concerning the appropri- entering the distillation train. Therefore, the response of the
ateness of the control system design may be inferred by control system to a disturbance in the feed flow and
steady-state analysis. For instance, consider column C-502. composition was studied, assuming a 2% step change in
From the analysis of the steady-state temperature profile in conversion (a reasonable value, as suggested by plant
this column (Figure 3b), one can derive useful indications engineers).
about the control performance of this column. The choice of The tuning of all controllers was set as in the plant.
the pilot tray for temperature control appears adequate since The step disturbance enters column C-501 at time t ¼ 1 h.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
 STEADY STATE AND DYNAMIC PLANT ANALYSIS 505

Figure 6. Basic control configuration for columns C-501, C-502 and C-504.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
506 BEZZO et al.

Figure 7. C-501 control system response to a step decrease of EDC conversion. Temperature profiles at trays 38 (a) and 14 (c), and related manipulated
variables: reflux rate (b) and reboiler duty (d).

Figure 7 illustrates the effect of the disturbance (a decrease the calculated responses closely matched the responses that
in conversion) on the controlled and manipulated variables they experienced in ‘real-life’ plant operation.
concerning controllers TICA-501 and TICA-502 (see It can be seen that there is no significant interaction
Figure 6). Temperature of tray 38 takes about 40 min to be between the composition control loops of column C-501 (this
restored to the set point (Figure 7a), with an undershoot of was also inferred from the calculation of the steady-state

2 C. At the column bottom, control appears to be easier. RGA diagonal element, which was found to be equal to 1.4).
Controller TICA-501 always maintains the temperature of tray Although the tuning seems good, the lack of interaction
14 within 0.5 C from the setpoint and adjusts the transient suggested the possibility of improving the control response
trajectory in about 20 min (Figure 7c). Both the manipulated by tuning the composition controllers slightly more tightly.
variables (reflux rate and reboiler duty) present smooth Note that, since the column is controlled according to the
profiles (Figures 7b and d). Discussion with the plant opera- LV control configuration, the control performance is almost
tors was important during this phase to get a confirmation that independent of the tuning of level controllers. This allowed

Figure 8. Composition profiles of VCM in the distillate (a) and HCl in the bottoms (b) of column C-501 after a step decrease of EDC conversion.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
 STEADY STATE AND DYNAMIC PLANT ANALYSIS 507

Handling Abnormal Events

In unusual circumstances (e.g. maintenance) the plant
capacity is switched from nominal to minimum: during this
procedure, the feed to the cracking furnaces and the conver-
sion through the furnaces change dramatically. Presently, these
operations are managed through manual control of all units,
while the automatic control system is disconnected. First of
all, the model was tested to obtain a clear indication of why
the current control system fails, and to find indications for the
development of an alternative control configuration.
This very severe disturbance was simulated by emulating
the actual procedure being activated when the furnace
capacity is decreased. The capacity reduction goes through
Figure 9. C-502 control system responses to a step decrease of EDC four steps: during each step (which lasts 4 h), a ramp
conversion: temperature profile at the pilot tray (5).
decrease is applied to the furnace feeds and temperatures,
which affect conversion. Figure 11 illustrates the procedure
concerning the five furnaces (A–E).
us to maintain loose bottom level control in C-501, which in The simulation confirmed the plant experience. It was
turn makes the operation of C-502 more stable. verified that the current control system is not able to prevent
Figure 8 shows dynamic behaviour of key impurity the loss of HCl from the bottom of C-501 (Figure 12a,
compositions in the distillate and bottom products. VCM dashed line), which eventually leads to a substantial increase
variations are not significant, whereas we can observe that of C-504 pressure (Figure 12b, dashed line) and contamina-
the hydrogen chloride concentration in the bottom is rather tion of the VCM product.
sensitive to a decrease of conversion in the cracking This can be explained by observing the evolution in the
furnaces, although the HCl level is still well below the temperature profile (Figure 13) after plant capacity has been
critical threshold (
100 ppm). reduced to the minimum. The plot shows that a pinch point
Nevertheless, it appears that, while temperature at tray 38 appears in the lower section of the column, which prevents
is a good indicator of VCM concentration in the top product, the separation of HCl from the bottoms.
the same is not true for temperature at tray 14. This may The dynamic model can be used very easily to quickly
motivate looking for a different controller configuration in test alternative control configurations, which can potentially
C-501. fix the problem related to the HCl loss. In order to remove
The composition disturbance takes 5–6 min to reach the pinch point and enforce the HCl separation in the bottom
column C-502 through the C-501 bottoms. The effects of section of C-501, the control of a temperature difference,
the disturbance are rather easily controlled by TRC-503, as rather than of a single temperature, was evaluated. The idea
illustrated in Figure 9. In particular, there are no significant of using multiple temperature measurements (instead of
variations in the top and bottom composition, as shown in a single temperature) to control the composition profile
Figure 10. inside a column is not new (Luyben, 1972). In practice,
Column C-504 has no composition control: only total when the difference between two tray temperatures is used
mass balance closure is enforced in this column by means of as the variable to be controlled, the controller tries to keep
level controllers. Simulation confirmed that no further constant the fractionation between those two trays, rather
regulation is required, since an effective control on columns than on a single tray as is done in conventional temperature
C-501 and C-502 coupled to the large number of trays of control. In the VCM plant, the best results were obtained by
C-504 are sufficient to guarantee the desired purity of the considering the difference of temperature between tray 10
bottom product. and the column sump (both these temperatures are measured

Figure 10. Column C-502: VCM concentration in the bottom (a) and HCl concentration in the distillate (b) after a step decreases of the EDC conversion.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
508 BEZZO et al.

Figure 11. Operating procedure for the switch to minimum capacity of

cracking furnaces. Furnace temperature is indicated in terms of ‘TRC’ Figure 13. C-501 temperature profile at nominal steady-state conditions
and during the transition to minimum plant capacity (current control
configuration).

in the plant). We indicate this difference with DT10, while

the new controller is named DT10C. Therefore, this control- The disturbance propagates to columns C-502 and C-504
ler tries to force the lower section of the column to achieve a without affecting the purity of products. Although the feed
prescribed fractionation between the two selected trays, and to column C-502 is drastically reduced (Figure 16a), the
this in turn prevents the pinch point to settle in this column concentrations of key components are kept under control
section. (Figure 16b). Controller DT10C was also tested for the
The new control configuration leads to substantial control of C-501 at nominal plant capacity. Figure 17 shows
improvements of the control performance, as shown by the the response to a disturbance after a decrease of cracking
full lines of Figure 12: controller DT10C is capable of conversion of the key component concentrations in the
keeping the concentration of HCl within the design speci- distillate and bottoms products in column C-501.
fications and C-504 pressure does not change significantly. It can be seen that controller DT10C is capable of keeping
Note that controller DT10C was tuned with the same both HCl and VCM compositions under control. The results
parameters as controller TICA-501. Figure 14(a) illustrates proved the helpfulness of dynamic simulation at the plant
how the temperature of tray 38 (top pilot tray) varies during level. In fact, it was possible to detect an easy-to-implement
this procedure, while the variation of the controlled variable alternative control scheme that is simultaneously capable of:
DT10 is reported in Figure 14(b): the undershoots are within  automating the switchover to minimum plant capacity;

1 C. It is also noteworthy that the controllers’ settling time  keeping the process production within specifications at
are about 5 h, i.e. no more than the time window over which nominal plant capacity.
the severe disturbance is applied.
Finally, Figure 15 compares the steady-state temperature The simulations were carried out by the plant engineers
profile at minimum capacity to the profile at nominal and were also exploited as a training approach for a correct
operating conditions. Controller DT10C maintains the and helpful use of simulation tools. Note that the simulation
shape of the profile in the bottom of the column, although time was always reasonably small, as was required by
shifted towards higher temperatures (which is not surprising, operators who may need frequent and quick answers to
since the column feeds are richer in EDC). address process behaviour issues.

Figure 12. Transition to minimum plant capacity: HCl in C-501 bottoms (a) and pressure at C-504 top (b), with conventional and modified control
configurations.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
 STEADY STATE AND DYNAMIC PLANT ANALYSIS 509

Figure 14. Transition to minimum plant capacity. Trajectories of controlled variables in the modified C-501 control configuration: temperature at tray 38 (a)
and DT10 (b).

 assess the effectiveness of present safety procedures and

collect data to improve them;
 estimate the plant dynamics to foresee the time available
for intervention before a breakdown appears.
In this case, the dynamic model was used to check the
plant response to varying operating conditions in the three
oxy-chlorination reactors down the line. Figure 18 illustrates
in detail the control configuration that was previously
commented on.
In order to understand how the control system works, let
us consider the chain of events that follow a decrease in
the demand of gaseous HCl to the oxy-chlorination reac-
Figure 15. C-501 temperature profile at nominal steady-state conditions and
tors. This is obtained by decreasing (one of) the setpoints
after the reduction to minimum capacity (modified control configuration). of flow controllers FIC-109=118=127 (Figure 18), which
raises the pressure in the upstream line. This increase in
pressure triggers controller PICA-102 to close the valve
Hazard Analysis and flow less HCl to the reactors. Pressure in tank D-501
increases, and PRCA-501 increases the duty to condenser
Hazard analysis is especially important for plant engi-
E-502 to allow for a reduction of gaseous HCl in the tank.
neers to detect potential hazard in the operation of the
The large size of D-501 guarantees that the additional
plant, and to define an effective response to safety problems.
liquid holdup can be effectively stored without a level
However, it is clear that the experimental data required for
controller, even if the reflux demand to C-501 (set by the
a safety analysis are rarely available and cannot be deliber-
temperature controller TICA-502) does not change. In
ately produced in the plant. Therefore, dynamic simulation
effect, tank D-501 decouples the production specifications
is quite a unique tool in order to predict the plant response to
of column C-501 from those of the oxy-chlorination
abnormal events. Therefore it can be exploited to:
reactors. Nonetheless, it is clear that this ‘mass imbalance’
 verify the plant resiliency to heavy disturbances; between the two process sections will cause an increase in

Figure 16. Transition to minimum plant capacity: feed flow profile (a) and key component composition trajectories (b).

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
510 BEZZO et al.

Figure 17. Comparison between present and modified control configuration: trajectories of bottom HCl (a) and top VCM (b) concentrations after a reduction
of conversion in the cracking furnaces.

Figure 18. C-501 detailed control configuration.

Figure 19. Blockage (solid line) and reduction (dashed line) of the load to one oxy-chlorination reactor: D-501 pressure (a) and C-501 condenser duty (b)
trajectories.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
 STEADY STATE AND DYNAMIC PLANT ANALYSIS 511

the D-501 level and may lead to overfill the tank if the  quickly evaluate the potential benefits and pitfalls of
feed to C-501 is not decreased in the long term. modified plant designs.
The first case under examination concerns the (very
The main benefits of dynamic modelling were:
improbable) blockage of one of the oxy-chlorination reac-
tors. The simulation predicts that the control system is not  the assessment of the control system performance;
able to handle such an event, although the large capacity of  the definition of an alternative control configuration for
tank D-501 allows for some compensation of the distur- automating some specific operating procedures;
bance before the pressure starts increasing in column C-501  a hazard analysis to evaluate the plant dynamics and the
and running out of control (Figure 19a, solid line). The control system response in the case of abnormal events.
disturbance is given at time t ¼ 0 and after very few minutes
condenser E-502 has already reached its condensing limit Finally, we would like to point out that this case study has
(Figure 19b, solid line; note that the duty has negative sign). served as a helpful tool in defining a set of guidelines for
After
30 min the column pressure becomes critical. This is plant engineers when using process simulation tools. It was
the estimated available time for intervention. stressed that a solid thermodynamic modelling is essential to
The second case study considers a severe reduction of the any reliable description of distillation columns (and, more
load to one of the reactors. The reactor feed is reduced from generally, of equilibrium-based unit operations): this affects
10,600 to 8700 kg h1 (8%) through a 3 min ramp. both steady-state and dynamic simulation. Additionally, it
Although this is a remarkable disturbance, note that the was shown that any good modelling practice requires close
existing control system is capable of compensating for it and interaction with plant operators in order to validate the
the column pressure settles back to steady state in about model and to detect phenomena that may give light to
20 min (Figure 19a, dashed line). The condenser hits its aspects of practical importance for a meaningful grasp of
condensing limit for a short while, after which it reaches a the process behaviour. Last but not least, it was demon-
new steady-state (Figure 19b, dashed line). It was verified strated that a sensible simplification of models can be the
that there are no significant variations in the key impurity only approach for a dynamic simulation software, whenever
concentrations. this is intended for plant engineer usage.
Once again, the dynamic simulation software proved to be Steady-state process simulators are a really mature, robust
a very helpful tool in the comprehension of plant behaviour. and reliable product. One current limitation of dynamic
In this case, plant engineers benefited from simulation results process simulators for plantwide control applications is
to improve the process safety standards. Note that these that, to date, process simulators offer no tools either for
simulations can be carried out in a very short time, and automatically generating alternative control structures for
directly in the plant offices, where they can be discussed with a given process, or for simply ranking among different
plant engineers and operators. In fact, dynamic simulation ‘user-provided’ control schemes according to steady-state
can be very useful for operator training, especially in such and dynamic controllability and resiliency indices. We
cases (as the one just considered) for which the execution of believe that, while the second feature (ranking among
the real event in the plant is certainly impractical. different control structures) could in principle be embedded
rather easily in a process simulator (even in a steady-state
one) using state-of-the-art control theory and software
engineering, the first feature (automatic generation of
CONCLUDING REMARKS ‘sound’ alternative control structures) may be somewhat
harder to achieve in a short time. Steady-state and dynamic
Process simulation tools are commonly used in the simulation tools are anyway extremely useful in assisting the
process industry, and computers are now essential in the plant engineers when decisions are to be taken quickly at the
design and operation of process plants. However, it appears plant level to improve the performance of process equip-
that, especially in the scientific literature, process simulation ment. Note that this should not imply that experience and
is still regarded as a highly specialized tool for highly knowledge are somehow unnecessary to correctly use simu-
specialized users. Reality in the process industry is quite lation tools. A qualified engineer must be ultimately respon-
different, as it can be observed that the usage of steady-state sible for all design and operation decisions. However, as this
and even dynamic simulation tools is spreading at every work has demonstrated, computer-aided simulation tools do
level. This paper has demonstrated how process simulation not exclusively belong to specialized research centres, and
tools can be effectively used at the plant level to evaluate unquestionable advantages may be obtained by means of
process performance and control system effectiveness, and sensible use of these tools on the production site.
improve them. An industrial case study was exploited to
identify a set of useful analyses that may be carried out at
the plant level and to suggest some guidelines for a correct
approach to simulation tools. In particular, steady-state REFERENCES
simulation allowed to: Biegler, L.T., 1989, Chemical process simulation, Chem Eng Prog, 85(10):
50–61.
 obtain a plant representation consistent with the actual Bretelle, D., Chua, E.S. and Macchietto, S., 1994, Simulation and on-line
one, and ready to be used directly by the plant engineers scheduling of PVC plant for operator training, Comput Chem Eng, 18:
as a part of the daily routine; S547–S551.
 assess the performance of process equipment; Eggersmann, M., Von Wedel, L. and Marquardt, W., 2000, The management
and reuse of models in industrial design processes (in German), Chem
 evaluate the production capacity of the plant section being Ingr Tech, 74: 1068–1078.
considered; Holderbaum, T. and Gmehling, J., 1991, PSRK: a group-contribution
 improve the process operation; equation of state based on UNIFAC, Fluid Phase Equil, 141: 251–265.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512
512 BEZZO et al.

IChemE CAPE Subject Group, 1999, Good Practice Guidelines—the Use of Roat, S.D., Downs, J.J., Vogel, E.F. and Doss, J.E., 1986, Chemical Process
Computers by Chemical Engineers, Version 1.0 (IChemE, Rugby, UK). Control—CPCIII, Morari, M. and McAvoy, T.J. (eds) (CACHE=Elsevier,
Kister, H.Z., 2002, Can we believe the simulation results? Chem Eng Prog, Austin, TX, USA).
98(10): 52–58. Schwartzentruber, J., Renon, H. and Watanasiri, S., 1989, Development of
Lausch, H.R., Wozny, G., Wutkewicz, M. and Wendeler, H., 1998, Plant- a new cubic equation of state for phase equilibrium calculations, Fluid
wide control of an industrial processes, Trans IChemE, Part A, Chem Eng Phase Equil, 52: 127–134.
Res Des, 76: 185–192. Seider, W.D., Seader, J.D. and Lewin, D.R., 1999, Process Design
Le, N.D., Sel, B.J. and Edwards, V.H., 2000, Doublecheck your process Principles. Synthesis, Analysis, and Evaluation (Wiley, New York, USA).
simulations, Chem Eng Prog, 96(5): 51–58.
Luyben, W.L., 1972, Profile position control of distillation columns with
sharp temperature profiles. AIChE J, 18: 238–240.
Marquardt, W., 1996, Trends in computer-aided process modeling, Comput
Chem Eng, 20: 591–609.
McAvoy, T.J., 1983, Interaction Analysis (Instrument Society of America, ACKNOWLEDGEMENTS
Research Triangle Park, USA).
Pantelides, C.C. and Britt, H.I., 1995, Multipurpose process modelling The authors gratefully acknowledge the support of Ing. C. Favro for the
environments, Proceedings of FOCAPD ’94, Biegler, L.T., et al. (eds) study-state model development. F.B. and M.B. would like to thank E.V.C.
(CACHE, Austin, TX, USA). Italia S.p.A for allowing them to publish the results presented in this paper.
Pilavachi, P.A., Schenk, M., Bek-Pedersen, E. and Gani, R., 2000, Design
and analysis of separation by distillation—role of property models, Chem The manuscript was received 28 February 2003 and accepted for
Eng Res Des, 78: 217–230. publication after revision 20 October 2003.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2004, 82(A4): 499–512