You are on page 1of 21

Ind. Eng. Chem. Res.

2009, 48, 10941–10961 10941

Application and Evaluation of Three Methodologies for Plantwide Control of the


Styrene Monomer Plant
Suraj Vasudevan, G. P. Rangaiah,* N. V. S. N. Murthy Konda,† and Wee Hwa Tay
Department of Chemical and Biomolecular Engineering, National UniVersity of Singapore, Engineering DriVe
4, Singapore 117576

Plantwide control (PWC) methodologies have gained significant importance given the high and increasing
degree of integration in chemical processes due to material recycle, energy integration, and stringent product
quality control, all of which though economically favorable, pose tough challenges to smooth plant operation.
As part of the continuing search for more effective PWC system design methods, an integrated framework of
heuristics and simulation was proposed [Konda et al. Ind. Eng. Chem. Res. 2005, 44, 8300-8313]. The basic
idea behind this development is to make effective use of rigorous process simulators to aid in decision-
making during the development of the heuristic-based PWC structure. Konda and co-workers have successfully
applied the procedure to the toluene hydrodealkylation process. Though the integrated framework is promising,
there is still a need to test its applicability to other complex industrial processes. The present contribution
considers the development of a PWC structure for the styrene monomer plant using the integrated framework.
In addition, in order to gauge its effectiveness in comparison to the other plantwide control methods, two
more methods are considered in this study. First, the heuristics procedure of Luyben and co-workers [Luyben
et al. Plant-Wide Process Control; McGraw-Hill: New York, 1998], which is a popular heuristics-based
methodology, is also applied to the same flowsheet, and is considered as the base case for performance
assessment. Second, the self-optimizing control procedure [Skogestad, S. Comput. Chem. Eng. 2004, 28,
219-234] is also used in order to have a more comprehensive analysis of the effectiveness of the integrated
framework. An analysis of the results indicates that while all the procedures give stable control structures,
the integrated framework and self-optimizing control procedures give more robust control structures than the
heuristics procedure. This is the first study to develop simulation models and complete PWC structures for
the styrene plant, together with a detailed analysis of the relative performance of the resulting structures in
order to evaluate the different PWC methodologies.

1. Introduction the reactor-separator-recycle (RSR) process. Only a few


studies have considered other processes like the vinyl acetate
Plantwide control (PWC) refers to the design of a control monomer (VAM) plant. Hence, there is still a necessity to further
system for an entire chemical plant consisting of many intercon-
study other highly integrated and complex processes in order
nected unit operations. A typical industrial process comprises
to better understand the PWC problems and evaluate the PWC
a complex flowsheet that includes many units, recycle streams,
methodologies.
and energy integration. These factors combined with the
chemical component inventories lead to more interactions, and One important process listed in Table 2 that has not been
hence, the need for a perspective beyond the individual units. studied much in the previous PWC works is the styrene
Although Buckley4 presented the first study on PWC methodol- monomer plant. So far, Turkay et al.49 have considered
ogy in 1964, PWC has been actively studied mainly in the past generation of a regulatory control structure for the styrene
15 years. During the period from the early 1990s, process control monomer plant using mixed-integer linear programming (MILP)
researchers have developed many systematic PWC methodolo- techniques. However, they only utilized a steady-state model
gies and applied them to typical chemical processes. These and simulation of the plant to obtain the gain information and
methodologies can be classified based on the approach used study the performance of the corresponding control loops. They
to develop the PWC structure as heuristic-based, mathemat- stated that the dynamic performance of the selected control
ics-based, optimization-based, and mixed approaches.5,6 The structure configurations for each individual unit needs to be
approach-based classification and main features of PWC checked before actual implementation. Zhu and Henson50 have
methodologies proposed since the year 2000 are summarized applied their PWC strategy based on integration of linear and
in Table 1. nonlinear model predictive control (MPC) to the styrene process
The complete, realistic processes that have been considered flowsheet. The plant is decomposed according to the nonlinearity
in various PWC studies are listed in Table 2 together with the of the individual unit operations, and sequential MPC algorithms
relevant references. It can be seen that the various PWC design are applied to the individual linear and nonlinear subsystems.
methodologies developed have been mainly evaluated on the However, they only presented the MPC solution to be applied
Tennessee Eastman (TE) process presented by Downs and to the decomposed styrene plant, and not the actual application
Vogel22 and the toluene hydrodealkylation (HDA) plant, besides to the styrene plant. Tarafder et al.52 have considered the
multiobjective optimization of the styrene process. Hence, the
* To whom correspondence should be addressed. E-mail: chegpr@
styrene monomer plant has been chosen in this study to apply
nus.edu.sg.

Current association: Postdoctoral Research Associate, Chemical some of the recently developed PWC methodologies and
Engineering Department, Imperial College London, UK, SW7 2AZ. evaluate their effectiveness.
10.1021/ie900022h CCC: $40.75  2009 American Chemical Society
Published on Web 10/13/2009
10942 Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009

Table 1. Approach-Based Classification and Features of PWC System Methodologies Proposed Since the Year 2000 in Chronological Order
ref(s) classification main features
3,7,8
Skogestad mixed In this self-optimizing control methodology, control system design is divided
into three layers based on time scale: local optimization, supervisory
control and regulatory control.
Jorgensen and Jorgensen9 mixed The control structure selection problem is formulated as an MILP problem
employing cost coefficients.
10
Zhu et al. optimization hybrid strategy integrating linear and nonlinear MPC
Groenendijk et al.11 and Dimian et al.12 mathematical combination of steady-state and dynamic controllability analysis for
evaluating the dynamic inventory of impurities
Robinson et al.13 mixed design of a decentralized PWC system using an optimal control-based
approach
14
Kookos and Perkins mixed mixed-integer nonlinear programming problem to minimize overall
interaction and sensitivity of the closed-loop system to disturbances
15
Wang and McAvoy mixed MILP problem in each of the three stages of the control system synthesis:
control of safety, production, and remaining process variables
16 17
Chen and McAvoy and Chen et al. mixed This hierarchical method was based on linear dynamic process models and
optimal static output feedback controllers. Later, Chen et al.17 extended it
to processes with multiple steady states.
Vasbinder and Hoo18 mixed decision-based approach, where the plant is decomposed into smaller
modules using a modified analytical hierarchical process
19 20
Cao and Saha and Cao and Kariwala mathematical This is an improved and more efficient algorithm of the “branch and bound
(BAB)” method for control structure screening. Later, Cao and Kariwala20
presented a bidirectional BAB algorithm for efficient handling of
large-scale processes.
Konda et al.1 heuristics integrated framework of simulation and heuristics which uses steady-state
and dynamic simulation to take or support the decisions taken by heuristics
21
Baldea et al. mixed controller design procedure integrating self-optimizing control with singular
perturbation analysis.

Table 2. Processes Studied by Researchers in PWC


process ref(s)
Tennessee Eastman (TE) plant Downs and Vogel, McAvoy and Ye,23 Price et al.,24 Lyman and
22

Georgakis,25 Ye et al.,26 Ricker and Lee,27 Banerjee and Arkun,28 McAvoy


et al.,29 Ricker,30 Luyben et al.,2,31 McAvoy,32 Kookos and Perkins,14
Wang and McAvoy,15 Chen et al.,17 Tian and Hoo33
hydrodealkylation (HDA) plant Stephanopoulos,34 Ponton and Laing,35 Fonyo,36 Ng and Stephanopoulos,37
Cao and Rossiter,38 Luyben et al.,2,31 Kookos and Perkins,14 Luyben,39
Herrmann et al.,40 Qiu et al.,41 Bildea and Dimian,42 Vasbinder et al.,43
Konda et al.,1,44 Araujo et al.,45,46 Bouton and Luyben47
vinyl acetate monomer (VAM) plant Luyben et al.,2,31 Chen and McAvoy,16 Olsen et al.48
vinyl chloride monomer (VCM) plant Groenendijk et al.,11 Dimian et al.12
styrene monomer plant Turkay et al.,49 Zhu and Henson50
dimethyl ether (DME) plant Vasbinder and Hoo18
tert-amyl methyl ether (TAME) process Al-Arfaj and Luyben51

Though every methodology proposed has its own advantages, example, which has been used to develop viable control
it has some limitations. The mathematical and optimization- structures for the HDA, TE, VAM, TAME, and a few other
based methodologies, though rigorous, are not easy to formulate processes.2,39,47,51 With this motivation, the need arises to choose
and require extensive computations especially for complex a suitable process that is complex enough and highly integrated
chemical processes. In addition, the solution might be affected to make it a suitable test-bed for the application of the integrated
by the model assumptions and details. Hence, heuristic-based framework. Further, the process should be relatively new to the
methodologies are found to be attractive as they are easier to PWC areasi.e. it has not been studied by PWC researchers,
understand and implement. One of the most popular heuristic- unlike the commonly considered TE plant or the HDA process.
based methods to date is that of Luyben et al.2 They proposed As mentioned earlier, one such process that has received little
a comprehensive nine-step procedure that ranks control and attention in the past is the styrene plant.
operational objectives based on their importance. However, one Styrene is one of the top ten bulk petrochemicals in the
major disadvantage of the heuristic-based methods is the over- world.55 More than 85% of styrene is produced by the direct
reliance on experience. Hence, to overcome this disadvantage dehydrogenation of ethyl benzene (EB).56 The presence of heat-
of heuristic-based methods, Konda et al.1 proposed an integrated integrated adiabatic plug-flow reactors (PFRs) with highly
framework of heuristics and simulation that makes effective use endothermic vapor-phase reactions, a tall vacuum distillation
of steady-state and dynamic simulation to improve the accuracy column performing the very difficult separation of styrene and
of the control decisions taken at each step. They have success- EB, and a material recycle stream makes the styrene process
fully applied the procedure to the HDA process. Subsequently, an ideal, challenging process for PWC study. Operational
Konda and Rangaiah53 have proven that their framework leads experience indicates that the EB dehydrogenation reactors, the
to a viable control structure that is more robust than some of high-purity vacuum column, and the high-purity EB recycle
the previously developed control structures for the HDA plant. column introduce significant nonlinearity to the process.50
Another recent and promising method is the self-optimizing Hence, the styrene monomer plant is chosen for the current study
procedure of Skogestad.3 So far, this method has been success- in order to test the applicability of the integrated framework of
fully applied to the HDA plant45,46 and the ammonia synthesis Konda et al.1 In fact, this is the first study to simulate and
process.54 develop a complete PWC structure for the styrene plant. Just
Though the procedure of Konda et al.1 has been shown to be developing the PWC structure for the styrene plant using the
promising, there is still a need to test its applicability to other integrated framework is of little use, unless the resulting
complex industrial processes. So far, it has only been tested on structure is compared with other control structures. Due to the
the HDA process as opposed to Luyben’s procedure, for lack of the complete control structures for the styrene plant in
Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009 10943
the literature, we chose to apply the procedure of Luyben et then recycled back. As the boiling points of styrene and EB
al.2 to the styrene plant, to develop the base case control structure are very close, the first column requires a very large number of
for comparison. The self-optimizing procedure of Skogestad3 stages for effective separation, typically 70 to 100.56 Separation
is also applied to the same flowsheet for a comprehensive and of distillate streams from the recycle column into toluene and
detailed comparison of different methodologies to identify their benzene is not considered in this study as its effect on the PWC
relative merits. design is expected to be minimal.
The rest of the paper is organized as follows: the next section 2.2. Steady-State and Dynamic Simulation. It is highly
gives a description of the styrene monomer process together important to select the most suitable fluid package for realistic
with details of the development of the simulation model in simulation. The Peng-Robinson equation of state is chosen,
HYSYS. Sections 3, 4, and 5 discuss the application of the as it is very reliable for predicting the properties of hydrocarbon
heuristics procedure of Luyben et al.,2 the integrated framework components over a wide range of conditions and is appropriate
of heuristics and simulation of Konda et al.,1 and the self- for the components in the styrene production process. The
optimizing control procedure of Skogestad,3 respectively, to the steady-state model of the styrene process is developed in
styrene plant. The three procedures are briefly reviewed in the HYSYS, and the flowsheet is optimized. The optimal conversion
respective sections. Section 6 presents the results of the per- is 66.61%. The distillation columns are modeled by rigorous
formance analysis of the resulting control structures. Finally, tray-by-tray calculations. Preliminary estimates of the number
conclusions are presented in section 7. of trays and feed tray location for each column in the process
are obtained by using the shortcut column in HYSYS and then
2. Overview and Simulation of the Styrene Process refined using the rigorous calculations. The final steady-state
flowsheet developed is given in Figure 1.
2.1. Styrene Process. Styrene is usually produced by the
A proper control system needs to be placed before the
vapor-phase adiabatic dehydrogenation of EB in two consecu-
dynamic simulation. This is important because inventories
tive PFRs using potassium-promoted iron oxide catalyst.57 The
(levels and pressures) may go to unstable levels if left
overall reaction is endothermic, and requires high temperature
uncontrolled. This is where PWC procedures are put to use to
and low pressure. Steam is added to the reactor for better
develop the control structure. Before switching to the dynamic
temperature control, to lower the partial pressure of EB (and
mode, a systematic procedure involving plumbing and equip-
thus shift equilibrium) and to prevent coking of catalyst.
ment sizing needs to be followed.39 Plumbing involves installing
Temperature control is very crucial due to many side reactions
pumps, compressors, and control valves at the correct locations
in the reactor; the typical operating range is 600-655 °C.57 The
in the flowsheet. Sizing information for all equipment must be
reactor pressure is between 1.4 and 2.4 atm. The molar ratio of
provided because the dynamic response of process units depends
steam to EB in the feed entering the reactor should be between
on the size of the equipment. The final dynamic simulation
12 and 17.58
model of the styrene plant consists of 1292 nonlinear, highly
In the styrene process, fresh EB and a part of the low-pressure
coupled algebraic and differential equations. Interested readers
steam (LPS) are initially mixed and then preheated in a feed-
are welcome to contact the corresponding author for the HYSYS
effluent heat exchanger (FEHE) using the reactant effluent
simulation files.
stream. The remaining LPS is superheated in a furnace to a
2.3. Anticipated Disturbances. It is important to consider
higher temperature between 700 and 850 °C57 and then mixed
a varied range of anticipated disturbances in order to assess the
with the preheated mixture to attain a temperature of around
performance of the control structures developed. This is because
650 °C. It is then fed to the two adiabatic PFRs, in series with
every disturbance has a unique impact on the process, and it is
a heater in between, for the production of styrene. The six main
not necessary that a control structure works well in handling
reactions that occur in the reactors are as follows:59
all the disturbances. In this study, six important disturbances
C6H5CH2CH3 T C6H5CHCH2 + H2 (1) have been studied together with two dual disturbances (that is,
a combination of feed rate and feed composition disturbances).
C6H5CH2CH3 f C6H6 + C2H4 (2) These are listed in Table 3, together with their respective
magnitudes. Reactor catalyst deactivation has been simulated
C6H5CH2CH3 + H2 f C6H5CH3 + CH4 (3) as disturbances d5 and d6. In addition, a set-point tracking study
has been conducted for the three-phase separator liquid level
2H2O + C2H4 f 2CO + 4H2 (4) (d7). This is because the set-point change of the organic phase
level controller in the three-phase separator is an important
H2O + CH4 f CO + 3H2 (5) plantwide disturbance that impacts the separation section, which
in turn affects the reactor section through the plant recycle.
H2O + CO f CO2 + H2 (6)
3. Application of Luyben’s Heuristic Procedure
The reactor effluent is cooled in the FEHE and further cooled
in a cooler before being sent to the three-phase separator, where Luyben et al.2,31 have proposed a heuristics procedure which
the light gases are removed as the light products and water is decomposes the PWC design problem into various levels. The
removed as the heavy product. The intermediate organic layer basic idea is to develop the simplest control system that achieves
is sent to a set of distillation columns for styrene separation the desired objectives. This procedure is applied to the styrene
from the other components. In the present study, the Monsanto process of Figure 1 to develop the decentralized control
approach is used for this separation, whereby styrene is separated structure. There is a certain degree of subjectivity on the part
first.56 In the first column (i.e., product column), operating under of the authors when applying this procedure as it is purely
vacuum to prevent styrene polymerization, styrene is removed heuristic-based. Luyben et al.2 recommend the use of quantita-
as the bottom product, and the top product is sent to a second tive measures from linear control theory (e.g., steady-state
column (i.e., recycle column) to separate the unreacted EB from process gains), as and when required to assess relationships
the two byproducts, toluene and benzene. The unreacted EB is between the controlled and manipulated variables. This work
10944 Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009

Figure 1. Steady-state flowsheet of the styrene monomer process.

Table 3. Anticipated Disturbances in the Styrene Process (c) The temperature at both the reactor inlets should be
no. disturbance magnitude 650 °C to minimize the side reactions.
(d) The temperature of the three-phase separator should
d1 production rate -5%
d2 +5% be at or below 40 °C in order to maximize organic
d3 -20% product recovery.
d4 feed composition -2% (e) The first distillation column should operate under
d5 catalyst deactivation -5% (PFR1)
(pre-exponential factor of main reaction) vacuum due to the risk of styrene polymerization in
d6 -5% (PFR2) the column bottom section. Hence, the column
d7 phase separator liquid percent level +5% operating pressure is set at 0.1 bar.60
d8 dual disturbances d1 and d4
d9 d2 and d4 Level 2: Determine Control Degrees of Freedom (CDOF).
The CDOF for the styrene plant is determined using the
uses the aid of the steady-state simulation model to calculate restraining number procedure of Konda et al.61 to be 22.62
these measures. The step-by-step application of this procedure Level 3: Establish Energy Management System. Energy
is discussed in this section; each step is briefly described prior disturbances should not be allowed to propagate throughout the
to its application. process. This can be done by transferring them to the plant utility
system. The control system must ensure that exothermic heats
Level 1: Establish Control Objectives. The steady-state
of reaction are removed from the process. In the current
design of the process is assessed and the dynamic control
application, the main reversible EB dehydrogenation (reaction
objectives are formulated. The objectives typically include 1) is highly endothermic and is favored at high temperatures.
reaction and separation yields, product quality specifications, The dealkylation reaction producing toluene (reaction 3) is
environmental restrictions, and safety concerns. The PWC slightly exothermic, but it occurs to a considerably lower extent.
objectives for the styrene plant are as follows. Hence, there is no serious issue of removing exothermic heat
(1) Production capacity: at least 100 000 tonnes/y (more than of reactions. The inlet temperature of both the reactors needs
110 kmol/h of styrene). to be controlled at the desired values. While the most direct
(2) Product quality: ca. 99.7%, which is an industry require- manipulator for the second reactor inlet temperature is the
ment.52 intermediate heater duty, two options are available for the first
(3) Process constraints: reactor inlet temperature, namely, steam split ratio and furnace
(a) The split ratio of the fresh steam is 0.18. This means duty. The latter is chosen based on steady-state gain analysis.
18% of the fresh steam is mixed with the total EB Level 4: Set Production Rate. The process variables that
(feed plus recycle), while the rest is sent to the dominate the reactor productivity are typically reactor conditions
furnace for superheating. such as temperature, pressure, reactant concentrations, and hold-
(b) The ratio of steam to EB molar flow rates at the ups. The most dominant variable should be used to set
entrance of the first reactor is set at 15. throughput. The most suitable manipulator to control the selected
Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009 10945
Table 4. Downs Drill Table Indicating Component Material Balance
component input generation (+) outputa (-) consumption (-) accumulation () inventory) controlled by
EB fresh feed 0 0 eqs 1-3 total EB flow control
steam fresh feed 0 water eqs 4-6 steam/EB ratio control
styrene 0 eq 1 SM 0 styrene column composition control
benzene fresh feed eq 2 tol/ben 0 recycle column composition control
toluene 0 eq 3 tol/ben 0
hydrogen 0 eqs 1 and 4-6 lights eq 3 three-phase separator pressure control
ethylene 0 eq 2 lights eq 4
methane 0 eq 3 lights eq 5
carbon dioxide 0 eq 6 lights 0
carbon monoxide 0 eqs 4 and 5 lights eq 6
a
Refer to Figure 1 for the stream names.

variable is then identified. For the styrene plant, there is no of the aqueous and organic phases are controlled using the
constraint to set production via either supply or demand. From respective outflows. The condenser and reboiler levels of the
the dehydrogenation reaction kinetics, reaction temperature, product column are controlled using distillate flow and bottoms
pressure, and EB concentration seem to be the dominant flow as the manipulators, respectively. The condenser and
variables determining the reactor productivity. Of these, the most reboiler levels of the recycle column are controlled by manipu-
direct handles for changing production rate are temperature and lating reflux flow and bottoms flow, respectively.
EB concentration. However, temperature cannot be selected as Level 7: Check Component Balances. It is important to
the reactor is operating close to the upper limit of 655 °C. And, analyze how each component enters, leaves, and is produced/
the reactor inlet composition is fixed due to the constraint consumed in the process in order to ensure that the overall
on the steam-to-EB ratio. Hence, the total flow (fresh feed plus material balance for each and every component is satisfied at
recycle) of EB is used for setting the production rate. The most steady state. The analysis can be done by preparing a “Downs
suitable manipulator to control total EB flow is the fresh feed Drill” table that lists each chemical component, its input, output,
flow of EB. consumption, and generation. Accordingly, Table 4 is prepared
Level 5: Product Quality, Safety, Operational and to check whether all the component balances have been satisfied.
Environmental Constraints. The best manipulators should be While all the control loops listed have already been developed
selected to achieve tight control of the product quality, safety, in the previous levels, the composition control for the distillate
and environmentally important variables. The selection should stream of the recycle column is configured now. This is achieved
be made such that the dynamic relationships feature small time by manipulating the distillate flow.
constants and large steady-state gains between the controlled Level 8: Control of Individual Unit Operations. The control
and manipulated variables. The main product in this case is loops necessary to operate each of the individual unit operations
styrene coming from the bottoms stream of the product column. are established. The reflux in the product column is flow-
To achieve the product quality control, EB impurity composition controlled. The vapor flow rate through the product column
of the bottoms stream from the product column is controlled overhead compressor is controlled using compressor duty. The
using the reboiler duty. bottoms composition of the recycle column is controlled by
The important operational constraints on the two reactor inlet manipulating reboiler duty.
temperatures have already been handled in level 3. A ratio Level 9: Optimize Economics or Improve Dynamic
controller is installed to maintain the ratio of steam to total EB Controllability. The remaining CDOFs should be used in the
of the first reactor inlet stream at the desired optimum value by best possible manner either to optimize steady-state economics
adjusting the flow rate of fresh steam. The operating pressure or to improve dynamic performance. The set-points of some
of the vacuum/product column has to be controlled. Condenser controllers can also be adjusted. The basic regulatory control
duty is chosen as the manipulator over the vent stream flow structure for the styrene plant comprising 20 control loops has
rate and overhead compressor duty due to the larger steady- been established and is named HS. The remaining two CDOFs
state gain. The temperature of the three-phase separator is are still unused. However, there is no particular necessity to
controlled using the cooling water flow of the cooler preceding establish any additional control loops. Economics and plant
the separator. Finally, the split ratio of the fresh steam is performance can be further optimized by adjusting the set-points
maintained by controlling the split steam flow. of some controllers. As the total EB flow is used to set
Level 6: Fix the Flow in Recycle Loops and Select the production, the set-point of both the reactor temperature
Best Manipulated Variables for Inventory Control. Flow in controllers can be selected to optimize styrene yield. The product
every recycle loop is to be fixed. Then, the best available column reflux flow is set based on steady-state optimization
manipulated variables are chosen to control the liquid and gas with the aim of minimizing column energy consumption and
inventories based on Richardson’s rule, which states that an loss of component recovery. The set-points for the steam-to-
inventory variable should be controlled with the manipulated EB ratio controller, split steam flow, and the product column
variable that exerts the largest effect on it within that unit. First, vent flow are also set at their optimal values.
the recycle flow of EB has to be fixed. This has already been The final regulatory control structure HS is summarized in
settled in level 4 when the total EB flow was controlled to set Table 5. The flowsheet with the controllers installed is given in
the production rate. Three pressures must be controlled, of which Figure 2. The process and manipulated variables, and the tuning
the product column pressure has been settled in the previous parameters for all the controllers are given in Table 5. The
level. The three-phase separator and recycle column operating control valves are all designed for 50% valve opening under
pressures are controlled using the respective most direct initial steady-state conditions. All the flow and pressure loops
manipulators (i.e., lights flow and condenser duty respectively). are tuned based on Luyben’s tuning rules.39 The recycle column
Six levels must be now controlled. The level control scheme condenser pressure loop is further fine-tuned. Most of the level
for the three-phase separator is quite straightforward; the levels loops are proportional-only as recommended by Luyben.39 The
10946 Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009

Table 5. Controllers with their Parameters for HS Control as both the reflux to feed and reflux ratios must change in order
Structure for the Styrene Plant to maintain the product purity in the presence of feed composi-
controller tion disturbances. Hence, the distillate to reflux ratio is
controlled manipulated parameters [Kc (%/%), controlled, which in turn is manipulated by the stage 9
variable variable Ti (min)]
temperature controller. Inferential composition control can be
reaction section used in the rectifying section of the recycle column as there
total EB flow EB feed flow (TPMa) 0.5, 0.3 are significant changes in temperature from stage to stage.
steam/EB ratio at PFR1 inlet steam feed flow 0.36, 0.035
split steam flow split steam flow 0.5, 0.3
PFR1 inlet T furnace duty 0.11, 0.088 4. Application of the Integrated Framework
PFR2 inlet T intermediate heater duty 0.54, 0.087
phase sep T cooling water flow 0.13, 0.14 A simulation-based heuristic methodology to develop a PWC
phase sep P lights flow 2, 10 structure has been recently proposed by Konda et al.1 The main
phase sep liquid % level organic flow 18.8, 0.45 advantage of this integrated framework of simulation and
phase sep aqueous % level water flow 1.31, 0.12
heuristics is that the use of nonlinear simulation tools such as
product column HYSYS makes it easier to apply the more difficult heuristics,
condenser P condenser duty 2, 10 which require experience and process understanding for their
condenser level distillate flow 0.8
reboiler level bottoms flow 0.5
effective usage. In each level of the eight-level improved
reflux flow flow specification heuristic methodology, nonlinear steady-state and dynamic
bottoms EB composition reboiler duty 0.12, 108 simulation models of the plant are used to take the decision or
vent flow compressor duty 0.5, 0.3 support the decision suggested by heuristics. This procedure is
recycle column applied to the styrene process (Figure 1) in order to develop a
condenser P condenser duty 1.5, 30 decentralized control structure. As was done in the previous
condenser level reflux flow 1.2 section, each step is briefly described prior to its application.
reboiler level bottoms flow 2
top EB composition distillate flow 0.43, 73.5 Level 1.1: Define PWC Objectives. PWC objectives should
bottoms toluene composition reboiler duty 6.54, 1.05 be formulated from the operational requirements of the plant.
a
TPM: throughput manipulator.
Typically, the control objectives comprise product quality and
production rate, plant stability requirements, process and equip-
exceptions are the level loops in the three-phase separator, which ment constraints, safety requirements, and environmental regula-
are implemented as proportional integral (PI) controllers and tions. Steady-state simulation models can be used to check
tightly tuned. The reason for this is that the levels in the three- whether the process is operating at stable steady-state or not.
phase separator are important variables that greatly affect process For the styrene plant, the PWC objectives are the same as those
variables in the separation section, which in turn affects the listed in level 1 of Luyben’s procedure in section 3.
process variables in the reaction section. The composition, Level 1.2: Determine CDOF. The CDOF of the process
temperature, and ratio control loops are tuned as PI controllers flowsheet is determined using the restraining number method
using the autotuner in HYSYS. Some of these controllers are of Konda et al.61 For the styrene plant, the CDOF ) 22.62
further fine-tuned to reduce undesirable oscillations. The entire Level 2.1: Identify and Analyze Plantwide Disturbances.
plant with the controllers installed is simulated for 100 min Plant disturbances have a considerable effect on the control
without any disturbance. This base case is then used for the structure selection and controller tuning. Hence, steady-state
performance assessment in section 6. simulation models can be used to try expected disturbances and
An alternative heuristics control structure was proposed by observe how their effect is propagated throughout the plant. The
Prof. William L. Luyben during the course of review of this important plantwide disturbances in the styrene process are (5%
manuscript; we provided him details of the styrene process and -20% variation in the fresh EB feed rate, (10% variation
model for this purpose. This control structure, summarized in in the steam feed rate, and -2% variation in EB feed
Appendix A, is named LS and is also included in the composition. Using the steady-state simulation model, it is
performance assessment in section 6. The main difference observed that (5% (-20%) variation in the EB feed flow causes
between HS and LS is the implementation of several feed- a (5% (-20%) variation in the separation section flows. The
forward elements in the latter, such as the control of the reflux -2% variation in the EB feed composition leads to -2%
to feed ratio in the product column as opposed to the control of variation in the styrene flow rate. The steam flow rate variation
reflux flow alone in HS. Also, application of the heuristics has very little effect on the separation section flow rates (<1%).
procedure by different people leads to different control structures Thus, both the upstream (reaction) and separation sections of
as heuristics are not always definitive and are hence subject to the styrene plant are observed to be equally sensitive for majority
the user’s choice. In particular, the column composition control of the disturbances.
schemes are significantly different in LS. In the product column, Level 2.2: Set Performance and Tuning Criteria. At this
the EB impurity on stage 59 is controlled instead of bottoms preliminary stage, settling time is chosen as the performance
EB impurity as direct control of a high purity product is quite criterion. While preliminary tuning of flow, level, and pressure
difficult due to high nonlinearity. Temperature cannot be used controllers can be done based on guidelines available in
to infer composition as the boiling points of EB and styrene Luyben,39 the built-in autotuning tools in dynamic simulators
are very close. Next, the reflux to feed ratio is controlled as can be used to obtain good initial controller settings for
this scheme is found to handle changes in feed composition composition and temperature loops. From the disturbance
better than a constant reflux ratio scheme. As for the recycle analysis in the previous level, it is concluded that any change
column, the toluene impurity on stage 29 is controlled instead in the EB feed flow produces the same amount of change in
of bottoms toluene impurity following the same reasoning given the separation section flow rates. Hence, no additional concerns
for the product column. Temperature cannot be used in this case regarding the tuning criteria arise.
as the temperature profile in the stripping section is very flat. Level 3.1: Production Rate Manipulator Selection. This
Next, a second composition controller is required in this column step involves the identification of the primary process path from
Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009 10947

Figure 2. Flowsheet with controllers for HS control structure.

the main raw material to the main product. Implicit variables steam. Second, ratio control is employed to maintain the steam
(e.g., reactor operating conditions) on this path are preferred to EB ratio at the first reactor inlet by manipulating the fresh
over explicit variables (fixed-feed followed by on-demand steam flow rate. Third, there are two possible manipulators for
options) as the throughput manipulator (TPM). The steady-state controlling the inlet temperature of the first reactor, namely,
simulation model comes in handy to make the choice on steam split ratio or the duty of the furnace used for steam
the primary process path as well as the TPM. The process superheating. The reactor section is separately simulated at
variable with the maximum steady-state gain will be the primary steady-state to study the percentage changes in the manipulated
choice for TPM. From the steady-state simulation, the steady- variable required to maintain the inlet temperature of the first
state gain of EB to styrene is observed to be greater than that reactor at its set-point for both the possible manipulators. It is
of steam to styrene. Hence, the primary process path is chosen observed that only a small change in the furnace duty is required
as EB to styrene. As reactor conversion is an optimization to maintain the inlet temperature at the set-point. Hence, the
decision and hence cannot be used for production rate manipu- furnace duty is chosen as the manipulator. As for the inlet
lation, the next best alternative for TPM is the fresh feed flow temperature of the second reactor, the duty of the intermediate
of EB. heater preceding it is the best manipulator. Next, the three-phase
Level 3.2: Product Quality Manipulator Selection. This separator temperature is controlled using the cooling water flow
step involves the selection of the manipulated variable for of the cooler preceding the three-phase separator. Finally, the
product quality. The unit producing the product stream can be manipulator for the product column condenser pressure is
simulated separately (steady-state and, if required, dynamic) for condenser duty due to the larger steady-state gain.
making this selection. On the basis of the relative gain analysis Level 4.2: Selection of Manipulators for Less Severe
(RGA)63 results for the product column, reboiler duty is chosen Controlled Variables. This step deals with the level and
as the product quality manipulator. pressure control loops. Levels along the primary process path
Level 4.1: Selection of Manipulators for More Severe should be self-consistent.1 Dynamic simulation can be used to
Controlled Variables. This step deals with process constraints aid in taking the decisions based on heuristics. In the case of
such as equipment and operating constraints, safety concerns, any conflict, decisions supported by simulation must be chosen
and process stability issues. The best manipulators for meeting as heuristics may not always be valid. For the current case, as
severe process constraints can be selected with the use of the reflux and boil up ratios for both the columns are consider-
dynamic simulation. As stated in section 3, the process ably less than 10, there are no severe restrictions on the choice
constraints in the styrene plant are the steam split ratio, steam of the level control variables.64 The levels in the primary process
to EB ratio at the first reactor inlet, stream temperatures at both path should be controlled in the direction of flow in order to
the reactor inlets, temperature of the three-phase separator, and have a self-consistent structure for the selected TPM (namely,
operating pressure of the product column. First, the split ratio fresh EB feed flow rate). On the basis of this heuristic, the
of the fresh steam is maintained by flow control of the split reboiler level for the product column is controlled by the bottoms
10948 Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009

Table 6. Settling Time, Change in Per-Pass EB Conversion, and DDS for Control Structure IF with and without Recyclea
change in conversion
settling time (min) (final - initial, %) DDS (kmol)
no. disturbance magnitude IF(o) IF(r) IF(o) IF(r) IF(o) IF(r)
d1 production rate -5% 110 340 +0.38 +0.69 14 23
d2 +5% 125 350 -0.37 -0.65 16 24
d3 -20% 300 505 +1.66 +3.10 58 87
d4 feed composition -2% 0 155 +0.19 +0.32 11 14
a
IF(o) refers to control structure IF without the recycle and IF(r) refers to control structure IF with the recycle loop closed.

flow. The other levels, namely the liquid and the aqueous levels
in the three-phase separator are also controlled in the direction
of flow.
The levels in side paths should be controlled such that the
disturbances are directed away from the primary process path.
On the basis of this heuristic, the condenser level in the product
column should be controlled by the distillate flow, and the
condenser and reboiler levels in the recycle column should be
controlled by the distillate flow and reboiler duty, respectively.
The reboiler level in the recycle column should not be controlled
by bottoms flow as it backpropagates the disturbances to the Figure 3. EB accumulation profile in the presence of disturbance d1 for
primary process path. However, this heuristic could not be control structure IF without and with recycle.
followed for the recycle column for the following reason. From
the RGA results of the recycle column, the manipulators for In that case, the control structure developed has to be modified
the condenser and reboiler levels are the reflux and bottom flows, using heuristics to aid in making the decision.
respectively. To summarize the results for the two columns, So far, dynamic simulation has been run with the recycle
the controlled variable-manipulated variable pairings decided stream unconnected. In this step, the closed-loop dynamic
in this step are condenser level-distillate flow and reboiler level- simulation is run, both with and without recycle, and analyzed
bottoms flow for the product column and condenser level-reflux for the important plantwide disturbances, namely d1, d2, d3,
flow and reboiler level-bottoms flow for the recycle column. and d4. The performance indicators considered are the process
Finally, the operating pressures of the recycle column and settling time and the recently developed dynamic disturbance
the phase separator are appropriately controlled by manipulating sensitivity (DDS) measure of Konda and Rangaiah53 (a brief
condenser duty and lights flow, respectively. introduction on DDS is given in Appendix B). The resulting
Level 5: Control of Unit Operations. This step should be settling times, change in the overall EB conversion, and the
considered before checking component material balances as DDS values are presented in Table 6 for both with and without
some component inventory loops can be implicitly taken care the recycle stream connected. It can be observed that the
of at this stage, which will make the analysis in the next level presence of (connected) recycle increases the settling time to
easier. The individual unit operations can be simulated to decide nearly twice/thrice that of the case without recycle. The EB
on the composition and temperature loops, and also for controller inventory profiles shown in Figure 3 clearly illustrate that, while
tuning. The composition and temperature loops are set up and the general trends are similar, the structure with recycle exhibits
tuned using the autotuning facility in HYSYS. On the basis of a slower response. Next, compared to the plant without recycle,
the results obtained in the earlier levels, LV and DV configura- the change in conversion is twice as much in the presence of
tions are employed for the product and recycle columns recycle for all the four disturbances (Table 6). DDS values are
respectively. In addition, the vapor flow rate through the product also significantly larger, indicating poorer performance.
column overhead compressor is controlled using the compressor Due to the aforesaid effects of the liquid recycle, the control
duty. system needs to be improved further. From the above analysis,
Level 6: Check Component Material Balances. It must be it is suspected that the larger change in conversion in the
ensured that the accumulation of each component in the overall presence of disturbances is the root cause of the poorer
process is zero. Accumulation tables can be prepared in the performance. This is because the deviation of conversion from
simulation models to check that the accumulation of all the the optimum value has a profound effect on the separation
components is zero while the plant without the recycle loops is section and recycle flows, and the system takes a longer time
in operation. For the current case, the dynamic simulation for to settle down to the new steady-state. Hence, it is advisable to
the whole plant without the recycle loop closed is run with the control the conversion at the optimum value. The three potential
accumulation tables of all the components and units prepared manipulators for controlling the EB conversion are the temper-
using the spreadsheet in HYSYS. These results reveal that ature, pressure, and composition at the first reactor inlet. The
accumulation is negligible, and thus, the component inventory additional constraint of maintaining the steam to EB ratio at
is regulated well. the reactor inlet and the relatively lesser impact of pressure on
Level 7: Effects Due to Integration. Unlike Luyben’s the reactions taking place leaves reactor inlet temperature
heuristic procedure, this analysis is done only after all the above as the most promising manipulator for conversion. It is in turn
levels have been handled. The severity of the recycle dynamics maintained by the furnace duty (cascade loop).
can be studied by simulating the process for anticipated The conversion controller is installed and the simulation is
disturbances, both with and without the recycle stream.1 The run for the same plantwide disturbances. From the results in
recycle dynamics can be taken to be severe if the process with Table 7, it can be seen that controlling the conversion overcomes
recycle shows considerably slower or even unstable dynamics. all the mentioned problems. The process settles faster, conver-
Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009 10949
a
Table 7. Settling Time and DDS for Control Structure IF without and with Conversion Controller
change in conversion
settling time (minutes) (final - initial, %) DDS (kmol)
no. disturbance magnitude IF(a) IF(b) IF(a) IF(b) IF(a) IF(b)
d1 production rate -5% 340 290 +0.69 0 23 19
d2 +5% 350 250 -0.65 0 24 23
d3 -20% 505 440 +3.10 0 87 74
d4 feed composition -2% 155 110 +0.32 0 14 13
a
IF(a) refers to control structure IF without conversion controller and IF(b) refers to control structure IF with the conversion controller implemented.

Table 8. Controllers with their Parameters for IF Control Structure procedure, the control system is divided into three layers based
for the Styrene Plant on time scale: local optimization (h), supervisory control (min),
controller parameters and regulatory control (s). The layers are linked by the controlled
controlled variable manipulated variable [Kc (%/%), Ti (min)] variables, whereby the set-points computed by the upper layer
reaction section are implemented by the lower layer. The implementation of this
styrene production rate EB feed flow (TPMa) 0.5, 0.3 procedure is divided into two main parts: top-down analysis
steam/EB ratio at PFR1 inlet steam feed flow 0.36, 0.035 (levels 1-4 below) to identify degrees of freedom and primary
split steam flow split steam flow 0.5, 0.3 controlled (self-optimizing) variables and bottom-up design
EB conversion PFR1 inlet T set point 0.1, 0.5
PFR1 inlet T (cascade) furnace duty 0.11, 0.088
(levels 5-8) to determine secondary controlled variables and
PFR2 inlet T intermediate heater duty 0.54, 0.087 control system structure. This procedure is applied to the styrene
phase sep T cooling water flow 0.13, 0.14 process (Figure 1) in order to develop the PWC structure. As
phase sep P lights flow 2, 10 was done in the previous sections, each step is briefly described
phase sep liquid % level organic flow 18.8, 0.45
phase sep aqueous % level water flow 1.31, 0.12 prior to its application.
product column
Level 1: Definition of Operational Objectives. This step
deals with identifying the operational constraints and also a
condenser P condenser duty 2, 10
condenser level distillate flow 2 scalar cost function J to be optimized. For the styrene plant,
reboiler level bottoms flow 1.5 the following profit function is to be maximized:
top styrene composition reflux flow 0.5, 27.3
bottoms EB composition reboiler duty 0.23, 54
vent flow compressor duty 0.5, 0.3
(-J) ) (pSMFSM + pben/tolFben/tol + pfuelQfuel) -
recycle column (pEBFEB + pLPSFLPS + (7)
condenser P condenser duty 2, 20 pfuelQheaters + psteamQsteam + pcwQcw + ppowWpow)
condenser level reflux flow 2
reboiler level bottoms flow 2
top EB composition distillate flow 0.43, 73.5 subject to the following constraints:
bottoms toluene composition reboiler duty 6.54, 1.05 1. EB feed rate ) 151.3 kmol/h.
a
TPM: throughput manipulator. 2. Feed steam split ratio for mixing with the total EB stream:
0.1 e split ratio e 1.0.
sion is maintained at its set-point and DDS values are also lower 3. Furnace exit temperature e855 °C.
for all disturbances (Table 7). Thus, the inclusion of the 4. First reactor inlet steam-to-oil ratio: 7 e SOR e 20.
conversion controller results in a balanced control structure 5. First PFR temperature: 600 °C e PFR1 T e 655 °C.
where the effects of the disturbances are reduced and distributed. 6. Second PFR temperature: 600 °C e PFR2 T e 655 °C.
Level 8: Enhance Control System Performance, if 7. Cooler outlet temperature ) 40 °C.
Possible. Further modifications, if needed, can be made to 8. Product column operating pressure ) 0.1 bar.
enhance the control system performance. Two more CDOFs 9. Minimum production rate g110 kmol/h of styrene
are still left, but the control structure developed is adequate. monomer.
Hence, further modifications are not required. 10. Styrene product purity g0.997.
The final decentralized control structure IF is summarized in In eq 7, pSM, pben/tol, pfuel, pEB, pLPS, psteam, pcw and ppow refer
Table 8. The flowsheet with the controllers installed is given in to the prices of styrene monomer product, benzene/toluene
Figure 4. The controlled and manipulated variables along with product, fuel (includes both fuel to the furnace and heater as
the parameter values for all the controllers are given in Table well as the lights stream from the three-phase separator), EB
8. The control structure comprises 21 control loops, which feed, LPS feed, steam to the reboilers, cooling water to the
includes a ratio controller and a cascade controller. The control cooler and condensers, and power to the pumps and compressor,
valves are all designed for 50% valve opening under steady- respectively. It is assumed that the lights stream from the three-
state conditions. The controller tuning procedure is similar to phase separator is sold as fuel. Refer to Table 9 for the unit
that of HS in section 3. The tuning parameters for the reaction price of raw materials and utilities. FSM, Fben/tol, Qfuel, FEB, FLPS,
section are the same as that of HS (Table 5) whereas they are Qheaters, Qsteam, Qcw, and Wpow refer to the flows of styrene
slightly different for some loops in the separation section. The product, benzene/toluene product, fuel value of the lights stream
entire plant with the controllers installed is simulated for 100 from the three-phase separator, EB feed, LPS feed, fuel to the
min without any disturbance. This base case is then used for furnace and heaters, steam to the reboilers, cooling water to
the performance assessment in section 6. the cooler and condensers, and power to the pumps and
compressor, respectively.
Level 2: Manipulated Variables and Degrees of
5. Application of Self-Optimizing Procedure
Freedom (DOF). The steady-state and dynamic DOF need to
Skogestad3 presented a systematic approach based on self- be identified. For the styrene plant, as mentioned earlier, the
optimizing control for PWC of chemical plants. In this CDOF is determined to be 22 using the restraining number
10950 Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009

Figure 4. Flowsheet with controllers for IF control structure.

Table 9. Unit Prices of Raw Materials, Products, and Utilities an optimization of the flowsheet needs to be carried out with
chemical/fuel/utility price unit remarks respect to steady-state DOFs for various disturbances, in order
ethyl benzene 0.4 US $/kg Chemical Market Reporter
to identify active constraints in the plant. Next, a local linear
(April 2001) analysis is carried out to identify the remaining controlled
styrene monomer 0.8 US $/kg variables. This is followed by evaluation of the loss for the
LPS 0.013 US $/kg
benzene 0.349 US $/kg promising sets of controlled variables, by adopting the constant
toluene 0.322 US $/kg set-point policy in the presence of disturbances and/or imple-
benzene/toluene 0.325 US $/kg based on stream composition
and pure chemical and mentation errors.
separation costs
electricity 1.68 × 10-5 US $/kJ Turton et al.65 (based on 2001 Level 3.1: Optimization. Optimization of the Distillation
rates) Columns. As listed in Table 11, there are five steady-state DOFs
cooling water 3.54 × 10-7 US $/kJ
fuel 6.00 × 10-6 US $/kJ for the distillation train. In order to avoid product give-away, it
steam utility 7.78 × 10-6 US $/kJ is always optimal to have the most valuable product as an active
constraint. For the styrene plant, the main product is styrene
Table 10. Steady-State Degrees of Freedom for the Styrene Plant
and hence its purity requirement of 99.7% is an active constraint.
process unit steady-state DOF For the other distillation products, the optimal operating point
external feed streams (EB and LPS) 1×2)2 is determined by a trade-off between maximizing the recovery
LPS splitter 1×1)1 of the valuable components and minimizing the energy con-
heat exchangers (2 heaters and 1 cooler) 1×3)3
heat exchangers in the two columns 2×2)4 sumption in the columns. The relation between reboiler duty
product column overhead compressor 1×1)1 and the respective mole fraction of the valuable component in
total steady-state DOF 11
each column is plotted (not shown for the sake of brevity). Note
that one of the DOFs in the product column (overhead vapor
procedure of Konda et al.61 Out of these, there are 11 steady-
flow) is utilized to reduce energy consumption in the recycle
state DOFs for the styrene plant (Table 10). Tables 11 and 12
list the steady-state and dynamic DOFs and the selected column. As mentioned in Araujo et al.,45 when the mole fraction
controlled variables for the styrene process, respectively. of an impurity component is sufficiently small, the amount of
Manipulated variables U1-U6 in Table 11 are the steady-state energy required to reduce the impurity rises exponentially.
DOFs for the reaction section, and U7-U11, for the separation Hence, a good trade-off is achieved at the edge of the “flat”
section. The controlled variables in Table 12 are selected based region in the graph. Accordingly, the optimum values for the
on the analysis of the process flowsheet. remaining four steady-state DOFs (that is, product compositions)
Level 3: Primary Controlled Variables. The implementation are determined and are listed in Table 13 together with the
of the procedure at this level is divided into three steps. First, already identified active constraint.
Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009 10951
Table 11. Manipulated Variables Considered in this Work for the Styrene Plant
manipulated variables with steady-state effect only manipulated variables with dynamic effect only
U1 EB feed flow U12 three-phase separator vapor flow (pressure control)
U2 LPS feed flow U13 three-phase separator organic flow (level control)
U3 feed LPS split ratio U14 three-phase separator aqueous flow (level control)
U4 furnace heat duty U15 product column condenser duty (pressure control)
U5 intermediate heater duty U16 product column distillate flow (level control)
U6 cooler cooling water flow U17 product column bottoms flow (level control)
U7 product column overhead compressor duty U18 recycle column condenser duty (pressure control)
U8 product column reflux flow U19 recycle column distillate flow (level control)
U9 product column reboiler duty U20 recycle column bottoms flow (level control)
U10 recycle column reflux flow U21 PFR1 exit flow (pressure control)
U11 recycle column reboiler duty U22 PFR2 exit flow (pressure control)

Table 12. Candidate Controlled Variables Selected in this Work for the Styrene Planta
no. candidate controlled variable no. candidate controlled variable
Y1 furnace inlet temperature Y30 recycle column distillate compositions
Y2 furnace outlet temperature Y31 recycle column bottoms compositions
Y3 FEHE tube-side inlet temperature Y32 PFR1 inlet compositions
Y4 FEHE tube-side outlet temperature Y33 furnace inlet pressure
Y5 FEHE shell-side inlet/PFR2 outlet temperature Y34 furnace outlet pressure
Y6 FEHE shell-side outlet/cooler inlet temperature Y35 FEHE tube-side inlet pressure
Y7 PFR1 inlet temperature Y36 FEHE tube-side outlet pressure
Y8 PFR1 outlet temperature Y37 FEHE shell-side inlet/PFR2 outlet pressure
Y9 PFR2 inlet temperature Y38 FEHE shell-side outlet/cooler inlet pressure
Y10 cooler outlet/3-phase separator temperature Y39 PFR1 inlet pressure
Y11 product column intermediate stage temperature Y40 PFR1 outlet pressure
Y12 recycle column intermediate stage temperature Y41 PFR2 inlet pressure
Y13 EB feed flow Y42 3-phase separator pressure
Y14 LP steam feed flow Y43 product column overhead vapor flow
Y15 total EB flow Y44 product column pressure
Y16 mixer (MIX2) outlet flow Y45 recycle column pressure
Y17 steam flow to furnace Y46 furnace heat duty
Y18 PFR2 reactor exit flow Y47 intermediate heater duty
Y19 3-phase separator vapor flow Y48 cooler heat duty
Y20 3-phase separator aqueous flow Y49 SOR at PFR1 inlet
Y21 3-phase separator organic/feed to product column flow Y50 overall eb conversion across both reactors
Y22 recycle column bottoms/recycle flow Y51 selectivity to styrene across both reactors
Y23 styrene monomer/product column bottoms flow Y52 3-phase separator liquid level
Y24 3-phase separator vapor compositions Y53 3-phase separator aqueous level
Y25 3-phase separator aqueous compositions Y54 product column condenser level
Y26 3-phase separator organic/feed to product column compositions Y55 product column reboiler level
Y27 product column vent compositions Y56 recycle column condenser level
Y28 product column distillate compositions Y57 recycle column reboiler level
Y29 product column bottoms compositions
a
Stream compositions are numbered singularly for conciseness. Individual components are numbered by adding a suffix to the stream composition
number (e.g.: Y26a for feed to product column EB composition).

Table 13. Self-Optimizing Specifications for the Distillation Columns degrees of freedom is now 6 - 3 ) 3. In the next step, a local
column specification value linear analysis is carried out to identify the promising sets of
three controlled variables.
product column mole fraction of styrene in distillate stream (D1) 0.01
(T-1) Level 3.2: Local Linear Analysis. With the identification
mole fraction of EB in bottoms stream (B1) 0.003 of eight active constraints, three unconstrained DOFs in the form
overhead vapor flow rate (Vent) 1.5 kmol/h
recycle column mole fraction of EB in distillate stream (D2) 0.01 of LPS feed rate, reactor intermediate heater duty, and LPS split
(T-2) ratio are still left. Subtracting the 8 controlled variables
mole fraction of toluene in bottoms stream (B2) 0.01
consumed by the active constraints from the 57 candidate
Optimization of the Entire Process (Reactors and controlled variables listed in Table 12, the number of possible
Recycle). Optimization with respect to the remaining six steady- sets of controlled variables is still very large (49C3 ) 18 424).
state DOFs needs to be carried out. However, in order to have Hence, similar to the analysis in Araujo et al.,45 a two-stage
a fair comparison with the two control structures developed selection procedure is employed to identify the remaining
earlier in the previous sections, the original operating point is controlled variables. The first step involves a local linear analysis
still used to carry out the subsequent steps in the development to shortlist the sets of candidate controlled variables. Here, the
of the self-optimizing control structure. On the basis of the branch-and-bound algorithm procedure laid out in the work of
chosen operating point, three more active constraints in the Araujo et al.45 for maximizing the minimum singular value of
reaction section are now identified in addition to the active S1G(Juu)-1/2 is used to identify the most promising sets of co-
constraints listed in Table 13. These are: (1) flow rate of EB ntrolled variables. Refer to ref 45 for details on the calculation
feed stream ) 151.3 kmol/h, (2) inlet temperature of the first of S1 (diagonal matrix of inverse of total spans), G (steady-
PFR ) 650 °C (chosen due to its importance to the reactions state gain matrix from the unconstrained degrees of freedom to
in this process), and (3) three-phase separator/cooler outlet the controlled variables), and Juu (Hessian of the cost function
temperature ) 40 °C. The remaining number of unconstrained with respect to the unconstrained degrees of freedom).
10952 Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009

Table 14. List of Anticipated Disturbances for Self-Optimizing Level 4: Production Rate Manipulator. The location of the
Control TPM needs to be identified next. The optimal location follows
disturbances/set-point changes nominal value new value change from the steady-state optimization in the previous level. The
D1 EB feed flow (high) 151.3 158.87 +5% choice of the TPM affects the way inventory is controlled as it
D2 EB feed flow (low) 151.3 143.74 -5% is desired to have a self-consistent control structure. There are
D3 benzene impurity in EB feed (high) 0.003 0.008 +0.005 two modes of operation: mode I (fixed throughput) and mode
D4 pre-exponential factor for main 4427 4206 -5%
reaction in PFR1 (low)a II (maximum throughput). While the former occurs when the
D5 pre-exponential factor for main 4427 4206 -5% feed or product rate is given or limited, the latter occurs when
reaction in both PFRs (low)a
D6 PFR1 inlet pressure (high) 2.40 bar 2.52 bar +5% the market conditions are such that it is optimal to maximize
D7 PFR2 inlet pressure (high) 1.80 bar 1.89 bar +5% the throughput. For the styrene plant, the analysis in the previous
D8 PFR1 inlet temperature (high) 650.0 °C 655.0 °C +5 °C three levels has been done by assuming a fixed feed rate (that
D9 PFR2 inlet temperature (high) 650.0 °C 655.0 °C +5 °C
D10 styrene product purity 0.997 0.999 +0.002 is, mode I of operation). In order to identify the TPM for mode
requirement (high) II, the bottleneck in the process needs to be determined by
D11 product column distillate 0.015 0.035 +0.02 identifying the maximum achievable feed rate in the presence
styrene impurity (high)
D12 recycle column distillate 0.01 0.03 +0.02 of disturbances.3 The TPM needs to be placed at the bottleneck
EB impurity (high) in order to maximize the flow through it.
D13 recycle column bottoms 0.01 0.03 +0.02
toluene impurity (high) The HYSYS steady-state model is utilized to locate the
D14 D4 + D8 650.0 °C 655.0 °C +5 °C bottleneck by increasing the external stream (EB feed) flow rate
D15 D5 + D8 + D9 650.0 °C 655.0 °C +5 °C
and determining the location in the plant that first reaches its
a
These disturbances simulate deactivation of catalyst. maximum capacity. The maximum capacity considered for each
The list of anticipated disturbances considered in the analysis equipment is listed in Table 19. Figure 5 clearly illustrates that
is given in Table 14. This list is slightly different from that as the fresh EB flow rate is increased, the operating profit of
given in Table 3 earlier, as it has been prepared to conform to the plant increases almost linearly. The linear baseline in the
the disturbances considered in the work of Araujo et al.,45 who figure is the projection of the profit if the plant operated at the
considered set-point changes also as disturbances. The effect same profit-to-feed ratio as the original point. As the EB flow
of these disturbances on 23 selected process variables is rate is increased, the operation reaches its ceiling at the EB feed
illustrated in Table 15. These process variables are selected as flow rate of 182.4 kmol/h when the reactor intermediate heater
they are expected to lead to small losses. In this table, “optimal duty (E-3) reaches its maximum duty. This location is thus
variation” refers to the largest deviation of the concerned process identified as the bottleneck in the plant.
variable at the final steady state from the nominal value in the
presence of disturbances. The implementation error is assumed With the identification of the bottleneck, the next question is
to be 5% of the nominal value except for the compositions, for the location of the TPM. The reactor intermediate heater duty
which it is fixed at 0.0001. “Total span” is the sum of the optimal is unfortunately an important manipulated variable for stabilizing
variation and implementation error. the process, due to its usage for the control of the second reactor
Next, values of S1, G, and Juu are obtained using the HYSYS temperature. Using any other manipulator may lead to significant
steady-state model. Singular value decomposition is then deviations in the EB conversion and styrene selectivity from
performed using MATLAB, and the maximum values of the optimal point. Hence, intermediate heater duty cannot be
σ(S1G(Juu)-1/2) obtained is listed in Table 16 for the top 10 used as the TPM in this process. Also, the additional concern
candidate sets of controlled variables. Note that the top 10 sets of allowing for interchangeability between modes I and II leads
include the furnace outlet temperature as one of the controlled to the selection of the EB feed flow as the TPM for both the
variables. Hence, it is selected as a primary controlled variable, modes. Anyway, in order to maximize the throughput in mode
and the next step considers the selection of the remaining two II, a cascade control scheme is implemented with the intermedi-
controlled variables. ate heater duty manipulating the set-point of the EB feed flow
Level 3.3: Nonlinear Analysis (Evaluation of the Loss). controller. This duty controller keeps the heater duty at a certain
Next, the loss for the top 10 promising sets of controlled value (back-off) below the maximum value in order to avoid
variables from Table 16 is evaluated by keeping a constant set- saturation of the duty in the presence of disturbances. Using
point policy in the presence of disturbances and implementation the steady-state HYSYS model this value is determined to be
errors. This is done using the HYSYS steady-state model with 1.75 × 107 kJ/h (a 6.37% back-off), and the corresponding EB
the aid of the “Adjust” module. The results are summarized in feed rate is 176.8 kmol/h.
Table 17. The three individual losses listed in this table are the
Level 5.1: Regulatory Control Layer. The purpose of this
average for six disturbances, seven set-point changes, and two
control layer is to stabilize the plant using low-complexity
implementation errors, respectively, and the overall loss is the
average for all these individual losses. Though the actual order controllers such that the plant does not drift too far away from
of ranking of the top 10 sets from the local linear analysis and the optimal operating point, and the supervisory layer can then
from the evaluation of the loss slightly differs, the topmost set handle the effect of disturbances on the primary outputs. The
is found to be the same in both cases. In addition, it is observed main structural issue is the selection of the secondary controlled
that one of the selected controlled variables is the inlet variables in the regulatory layer and their pairing with suitable
composition to the first reactor and the other is the product manipulated variables that do not saturate. While the primary
composition from the reaction section (Table 17). Logically, controlled variables are selected based on economic concerns,
these are the rightful candidates that would lead to optimal the secondary controlled variables are selected with stable and
control of the reaction section. This reinforced the confidence smooth plant operation in mind. The guidelines given by
in both the local analysis and the loss evaluation. To summarize Skogestad3 are employed to select the appropriate secondary
the findings in this level, the final list of the selected primary controlled variables and the corresponding manipulators for the
controlled variables is given in Table 18. styrene process.
Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009 10953
Table 15. Candidate Controlled Variables with Small Losses in the Local Linear Analysis
process variable nominal value optimal variation implementation error total span
furnace outlet temperature Y2 801.84 15.22 40.09 55.31
PFR1 inlet temperature Y7 650 5 32.5 37.5
PFR1 outlet temperature Y8 588.65 4.29 29.43 33.72
PFR2 inlet temperature Y9 650 5 32.5 37.5
PFR2 outlet temperature Y5 617.87 4.56 30.9 35.45
cooler outlet/three-phase Y10 40 0 2 2
separator temperature
EB feed flow Y13 151.3 7.57 7.57 15.14
LPS feed flow Y14 3371.13 0 168.56 168.56
three-phase separator vapor flow Y19 264.5 12.78 13.23 26.01
feed to T-1 (EB) composition Y26a 0.34 0.02 0.0001 0.02
feed to T-1 (SM) composition Y26b 0.56 0.013 0.0001 0.014
T-1 vent flow Y27 0.0034 0.0068 0.0001 0.0069
T-1 distillate (SM) composition Y28b 0.01 0.02 0.0001 0.02
T-1 bottoms (EB) composition Y29b 0.003 0.002 0.0001 0.0021
T-2 distillate (EB) composition Y30a 0.01 0.02 0.0001 0.02
T-2 bottoms (toluene) composition Y31d 0.01 0.02 0.0001 0.02
PFR1 inlet (EB) composition Y32a 0.063 0.0042 0.0001 0.0043
PFR1 inlet (steam) composition Y32b 0.94 0.0043 0.0001 0.0044
PFR1 inlet pressure Y39 240 12 12 24
PFR2 inlet pressure Y41 180 9 9 18
steam-to-oil ratio at PFR1 inlet Y49 15 1.14 1.5 2.64
overall EB conversion Y50 65.64 2.03 1 3.03
overall styrene selectivity Y51 84.21 1.45 1 2.45

Table 16. Minimum Singular Values for the Top 10 Candidate Sets of Unconstrained Controlled Variables
rank controlled variable 1 controlled variable 2 controlled variable 3 σ(S1G(Juu)-1/2)
1 furnace outlet temperature (Y2) feed to T-1 styrene composition (Y26b) PFR1 inlet EB composition (Y32a) 0.0985
2 furnace outlet temperature (Y2) feed to T-1 styrene composition (Y26b) PFR1 inlet steam composition (Y32b) 0.0979
3 furnace outlet temperature (Y2) LPS feed flow (Y14) feed to T-1 styrene composition (Y26b) 0.0914
4 furnace outlet temperature (Y2) feed to T-1 EB composition (Y26a) PFR1 inlet EB composition (Y32a) 0.0854
5 furnace outlet temperature (Y2) feed to T-1 EB composition (Y26a) PFR1 inlet steam composition (Y32b) 0.0848
6 furnace outlet temperature (Y2) LPS feed flow (Y14) feed to T-1 EB composition (Y26a) 0.0769
7 furnace outlet temperature (Y2) PFR1 inlet EB composition (Y32a) overall EB conversion (Y50) 0.0627
8 furnace outlet temperature (Y2) PFR1 inlet steam composition (Y32b) overall EB conversion (Y50) 0.0624
9 furnace outlet temperature (Y2) feed to T-1 styrene composition (Y26b) steam-to-oil ratio at PFR1 inlet (Y49) 0.0614
10 furnace outlet temperature (Y2) LPS feed flow (Y14) overall EB conversion (Y50) 0.0531

Table 17. Average Losses for the Top 10 Candidate Sets from the Local Linear Analysis
average loss (US $/h)
rank controlled variable 1 controlled variable 2 disturbances set-point changes implementation overall

1 feed to T-1 styrene composition (Y26b) PFR1 inlet EB composition (Y32a) 33.01 8.22 –0.89 16.92
2 feed to T-1 styrene composition (Y26b) PFR1 inlet steam composition (Y32b) 33.55 11.94 0.34 19.04
3 LPS feed flow (Y14) feed to T-1 styrene composition (26b) 38.73 10.66 22.46 23.46
4 feed to T-1 EB composition (Y26a) PFR1 inlet EB composition (Y32a) 34.20 9.20 –1.05 17.83
5 feed to T-1 EB composition (Y26a) PFR1 inlet steam composition (Y32b) 34.64 11.17 0.19 19.09
6 LPS feed flow (Y14) feed to T-1 EB composition (Y26a) 35.22 8.94 23.26 21.36
7 PFR1 inlet EB composition (Y32a) overall EB conversion (Y50) 34.05 8.92 –3.38 17.33
8 PFR1 inlet steam composition (Y32b) overall EB conversion (Y50) 34.66 10.83 –1.38 18.73
9 feed to T-1 styrene composition (Y26b) steam-to-oil ratio at PFR1 inlet (Y49) 33.54 9.10 31.94 21.92
10 LPS feed flow (Y14) overall EB conversion (Y50) 35.45 9.52 28.04 22.36

Table 18. Selected Primary Controlled Variables Table 19. Equipment Capacity Limits for the Styrene Plant
no. process variable equipment variable nominal duty (kJ/h) maximum (kJ/h)
1 EB feed flow Y13 furnace (E-1) Q1 6.509 × 10 7
9.764 × 107 (+50%)
2 furnace outlet temperature Y2 heater (E-3) Q3 1.245 × 107 1.867 × 107 (+50%)
3 PFR1 inlet temperature Y7 cooler (E-4) Q4 2.320 × 108 3.481 × 108 (+50%)
4 PFR1 inlet EB composition Y32a compressor (K-1) Q5 1.588 × 104 1.905 × 104 (+20%)
5 three-phase separator temperature Y10 T-1 condenser QC1 2.464 × 107 3.696 × 107 (+50%)
6 feed to T-1 styrene composition Y26b T-1 reboiler QR1 2.625 × 107 3.938 × 107 (+50%)
7 T-1 distillate styrene composition Y28b T-2 condenser QC2 2.648 × 106 3.972 × 106 (+50%)
T-2 reboiler QR2 4.526 × 106 6.789 × 106 (+50%)
8 T-1 bottoms EB composition Y29b
9 T-1 overhead vent flow Y43
10 T-2 distillate EB composition Y30a later for the furnace outlet temperature too in order to avoid temperature
11 T-2 bottoms toluene composition Y30d drift. The two concerned manipulators available are the steam split
ratio and furnace duty. RGA analysis indicates that the control of the
Stabilization of Unstable Modes (Including Liquid first reactor inlet temperature should be by manipulating the furnace
Levels). In the reaction section, the two reactor temperatures need duty.
to be controlled to stabilize the reactor operations. While the control Next, levels in both the reaction and separation sections need
of the second reactor inlet temperature is straightforward (by manipu- to be stabilized. In the reaction section, the aqueous and organic
lating the intermediate heater duty), the control of the first reactor inlet levels in the three-phase separator need to be controlled. The
temperature needs RGA analysis, as a manipulator needs to be selected manipulators for these are straightforward, that is, the respective
10954 Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009

However, based on the work of Riggs,66 inferential composition


control would not be effective in the product column due to
the very close boiling points of EB and styrene, which leads to
a relative volatility of less than 2. For the recycle column, the
procedure presented in Araujo et al.46 is used to identify the
appropriate stage for temperature control. They suggested
the control of the stage temperature where the temperature is
most sensitive to the reboiler duty and the temperature gradient
is significant. Accordingly, stage 6 is selected for implementing
the inferential composition control.
Avoiding Drift: Flow Control. Flow control is employed
Figure 5. Variation of the plant operating profit with EB feed flow.
to regulate both the reactants in the process. As the recycle EB
phase outflows. In the separation section, the condenser and flow rate is comparable with the EB feed flow rate, the total
reboiler levels in both the columns need to be controlled. For EB flow is controlled by manipulating the EB feed flow, which
both of the columns, the regular LV configuration (i.e., is the TPM. The LPS feed flow is controlled by manipulating
condenser level is controlled by distillate flow, and reboiler level the valve in the feed line (VLV-2 in Figure 1).
is controlled by bottoms flow) is employed. Level 5.2: Possible Intermediate Regulatory Control
Avoiding Drift: Pressure Control. In most processes, Layer. Some of the self-optimizing controlled variables selected
pressure dynamics are relatively fast. Hence, pressure drift can in level 3 are compositions whose dynamics are slow and
be avoided by control of pressure at appropriate locations in unreliable. Therefore, the purpose of the intermediate control
the plant. First, pressure needs to be controlled somewhere in layer is to achieve indirect control of the primary controlled
the reaction section. Dehydrogenation of EB (reaction 1) is variables and, thus, ensure near optimal plant operation in case
favored at low pressures. This leads to the plant design any of the primary control loops fails. The focus at this level is
containing a minimal amount of valves in the reaction section on the reaction section; for the styrene plant, two of the primary
in order to avoid excessive pressure drops that would inevitably controlled variables are compositions. Hence, a possible inter-
lead to build-up of pressure upstream. As a result, only one mediate layer could be implemented to achieve an acceptable
pressure control loop can be employed in the reaction section control of the plant at the intermediate time scale. All the
(for the three-phase separator). The two other CDOFs related available CDOFs have been employed in the regulatory control
to the manipulation of the effluent streams from the reactors layer except the two reactor exit flow rates, the reason for this
are not exercised. The three-phase separator pressure is con- has already been given in level 5.1. Therefore, any controller
trolled by manipulating the lights flow from the vessel. As for employed in the intermediate layer will be manipulating the
the distillation section, the operating pressures of both the set-points of existing controllers in the regulatory layer. As some
columns are controlled by manipulating the respective condenser of the primary controlled variables are already controlled in the
duties. regulatory control layer, CDOFs associated with these variables
Avoiding Drift: Temperature Control. Temperature mea- are discounted from the analysis for the intermediate regulatory
surements are again usually fast and reliable and, hence, can layer. This leaves two CDOFs in the reaction section, namely,
be controlled in the regulatory layer to avoid drift. For the set-points for the LPS feed flow control and the second reactor
reaction section, temperature control loops are placed at two inlet temperature control. This is the same number as the number
more locations. They are the furnace outlet temperature and the of composition variables that are primary controlled variables.
three-phase separator temperature (both of which are primary A local linear analysis similar to that described in level 3 is
controlled variables). Control of the three-phase separator then performed for the potential candidate sets. The results for
temperature is straightforward and can be achieved by manipu- the top six candidate sets are given in Table 20. As the top-
lating the cooling water flow to the cooler. As for the furnace ranked candidate set comprises the secondary controlled vari-
outlet temperature, as mentioned earlier RGA analysis indicates ables that have been already handled by the regulatory layer, it
that the steam split ratio is the better manipulator. is concluded that there is no additional economic benefit of
As for the employment of inferential composition control for introducing the intermediate regulatory control layer.
the distillation columns, the main aim of this study is to evaluate Level 6: Supervisory Control Layer. The purpose of this
the three different methodologies mainly based on the perfor- layer is to keep the primary controlled variables at optimal set-
mance of the resulting control systems. It is believed that the points by using as manipulators the set-points of the controllers
use of inferential composition control would not make a big in the regulatory control layer and any unused CDOFs. The
impact on the performance of the PWC system. Nevertheless, structural issue here is the selection of the controlled variable-
separation of the control structure in the self-optimizing control manipulated variable pairing. For the styrene plant, the TPM
procedure into a fast (regulatory) and slow (supervisory) layer, has been chosen as the feed flow of EB in both the modes.
with the set-points of the regulatory layer given by the While it is fixed at 151.3 kmol/h in mode I, it is adjusted in
supervisory layer, leads us to consider the possibility of using order to utilize the maximum capacity of the plant bottleneck
temperature control in the columns for controlling compositions. (with some back-off) in mode II. With the development of the
Table 20. Top Six Candidate Control Sets from the Local Analysis for the Intermediate Regulatory Layer
rank controlled variable 1 controlled variable 2 σ(S1G(Juu)-1/2)
1 PFR2 inlet temperature (Y9) LPS feed flow (Y14) 0.0338
2 LPS feed flow (Y14) 3-phase separator vapor flow (Y19) 0.0330
3 PFR2 outlet temperature (Y5) LPS feed flow (Y14) 0.0256
4 PFR2 outlet temperature (Y5) 3-phase separator vapor flow (Y19) 0.0199
5 PFR2 inlet temperature (Y9) PFR2 outlet temperature (Y5) 0.0128
6 PFR2 inlet temperature (Y9) 3-phase separator vapor flow (Y19) 0.0115
Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009 10955
regulatory control layer, six primary controlled variables, which Table 21. Controllers with their Parameters for SOC Structure for
are basically compositions, are still left to be controlled. Four the Styrene Plant
of the compositions are active constraints, while the remaining controller parameters
controlled variable manipulated variable [Kc (%/%), Ti (min)]
two are the unconstrained (self-optimizing) controlled variables
identified from the local linear analysis and the evaluation of regulatory layer
the loss. In addition, for mode II, there is an additional controlled
reactor section
variable in the form of the intermediate heater duty as mentioned
total EB flow EB feed flow (TPMa) 0.5, 0.3
in level 4. steam feed flow (cascade) steam feed flow 0.5, 0.3
The control loops in the reaction section are first settled. The furnace outlet T split steam flow 0.57, 0.12
two unconstrained variables selected in level 3 are the first PFR1 inlet T furnace duty 0.11, 0.089
PFR2 inlet T (cascade) intermediate heater duty 0.55, 0.087
reactor inlet EB composition and the product column feed intermediate heater duty total EB flow SP 0.35, 0.6
styrene composition. The manipulators available are the LPS (only in mode II)
feed flow and the set-point of the second reactor inlet temper- phase sep T cooling water flow 0.13, 0.14
phase sep P lights flow 2, 10
ature controller, as these are the CDOFs in the reaction section phase sep liquid % level organic flow 18.8, 0.45
that have not been involved in the control of a primary controlled phase sep aqueous % level water flow 1.31, 0.12
variable. RGA analysis suggests the following intuitive pairing: product column
first reactor inlet EB composition with LPS feed flow and
condenser P condenser duty 2, 10
product column feed styrene composition with the set-point of condenser level distillate flow 2
the second reactor inlet temperature controller. In addition, for reboiler level bottoms flow 2.2
mode II, it has already been decided in level 4 to control the vent flow compressor duty 0.5, 0.3
duty of the intermediate heater duty by manipulating the set- recycle column
point of the total EB flow controller. condenser P condenser duty 2, 10
The control loops in the separation section are next finalized. condenser level distillate flow 2
reboiler level bottoms flow 2
This involves the control of four exit stream compositions in stage 6 T (cascade) reflux flow 0.45, 4.47
the two columns, which have been identified as active constraints
supervisory layer
in level 3. As already discussed in level 5.1, the LV configu-
ration is employed in both the product and recycle columns. reactor section
This means that, in the product column, the distillate styrene PFR1 inlet EB composition steam feed flow SP 0.24, 0.02
composition is controlled by the reflux flow and the bottoms product column feed PFR2 inlet T SP 0.28, 20.3
styrene composition
EB composition is controlled by the reboiler duty. As for the
product column
recycle column, the distillate EB composition is controlled by
manipulating the set-point of the stage 6 temperature controller top styrene composition reflux flow 0.5, 27.3
bottoms EB composition reboiler duty 0.23, 54
and the bottoms toluene composition is controlled by the reboiler
recycle column
duty. In addition to these loops, the product column overhead
vent flow is controlled using the compressor duty. top EB composition recycle column stage 6 T SP 0.033, 73.5
bottoms toluene composition reboiler duty 6.54, 1.05
Level 7: Possible Optimization Layer. The main issue here
a
is whether real-time optimization is required. For the styrene TPM: throughput manipulator.
plant, as the current self-optimizing control design already
addresses the disturbances with acceptable loss, it is decided the previous sections. HS is the base-case control structure
that online optimization is not required. developed by applying the nine-step heuristics procedure of
Level 8: Validation. This last step involves running the Luyben et al.,2 IF is the control structure developed using the
nonlinear dynamic simulation of the plant with the developed integrated framework of heuristics and simulation,1 and SOC
control structure. This is discussed in the next section when is the control structure developed using the self-optimizing
doing the performance analysis. control procedure.3 There are four main differences among these
Both the regulatory and the supervisory control layers have control structures. First, throughput manipulation is achieved
now been developed. The final control structure summarized by changing EB feed flow in IF, while it is achieved by changing
in Table 21 comprises 17 control loops in the regulatory control total EB flow (feed plus recycle) in HS and SOC. Second, the
layer, together with 6 (7 for mode II) control loops in the recycle flow is explicitly fixed in HS; this is achieved by
supervisory control layer. The flowsheet with the controllers controlling the total feed flow. Third, the furnace outlet
installed is given in Figure 6. This control structure utilizes 20 temperature, PFR1 inlet composition, and product column feed
CDOFs; the reason for the nonutilization of the remaining 2 compositions are controlled only in SOC as they are the
CDOFs has already been given in level 5. The controlled and unconstrained controlled variables in this procedure. Finally,
manipulated variables along with the parameter values for all IF utilizes the conversion controller as a result of the detailed
the controllers are given in Table 21. The control valves are all analysis of the effects of recycle. In addition to these differences,
designed for 50% valve opening under steady-state conditions. the SOC control structure is divided into two layers (supervisory
The controller tuning procedure is similar to that discussed in and regulatory layers) based on time-scale with some controllers
section 3 for HS. The entire plant with the controllers installed in the supervisory layer providing the set-points for the related
is simulated for 100 min without any disturbance. This base controllers in the regulatory layer. The performance of all these
case is then used for the performance assessment in the next control structures together with the LS control structure proposed
section. by Luyben (Appendix A) is evaluated using the DDS perfor-
mance measure of Konda and Rangaiah.53 Besides DDS, the
6. Evaluation of the Control Systems process settling time and profiles of styrene production rate are
also used to evaluate the performance.
Three different PWC methodologies have been applied to All the control structures are subjected to the nine chosen
develop control structures for the styrene monomer process in disturbances listed in Table 3, and the resulting performance is
10956 Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009

Figure 6. Flowsheet with controllers for SOC control structure.

Table 22. Settling Time and DDS Values for Control Structures: HS, IF, SOC, and LSa
settling time (min) DDS (kmol)
no. disturbance magnitude HS LS IF SOC HS LS IF SOC
d1 production rate -5% 480 175 290 175 46 38 19 20
d2 +5% 300 185 250 205 51 38 23 20
d3 -20% 830 190 440 290 179 174 74 75
d4 feed composition -2% 0 0 110 0 27 13 13 14
d5 catalyst deactivation -5% (PFR1) 155 45 0 0 14 6 8 6
d6 -5% (PFR2) 0 0 0 0 18 5 4 5
d7 phase separator liquid % level +5% 155 85 140 75 10 15 11 10
d8 dual disturbances d1 and d4 475 185 300 195 59 61 30 30
d9 d2 and d4 245 170 155 185 52 39 15 24
a
The lowest value in each case is shown in bold.

quantified in terms of settling time and DDS. The criterion used the former performs better in terms of DDS. The smaller values
for computing the settling time is that the plant production rate of DDS for IF and SOC illustrate that the plant with either of
must be within (1% of the final steady-state value after the these structures is relatively less sensitive to disturbances, which
settling time. Thus, even though settling time is 0 for some cases, means that they can handle disturbances well.
other process variables particularly compositions might still not The reason for the differences in performance of the four
be within (1% of their final values. The results are presented control structures is investigated by analyzing the process
in Table 22. Some of the disturbances (d5 and d6) have response to disturbances in some important sections of the plant.
relatively lesser impact on the process, which is evident from Table 23 shows the percentage change in the separation section
the very low values of DDS. The performance of the control feed flow, recycle flow, reboiler duties and vapor boil up rates
structures also differs significantly. All of the four PWC in the presence of feed flow disturbances. While the product
structures provide stable control of the plant for all the column feed flow responds similarly for all disturbances, the
disturbances. There are no control valve saturation problems remaining variables (especially the ones shown with gray
encountered. background) respond significantly different from the expected.
From Table 22, it can be seen that both LS and SOC generally In general, SOC shows the most balanced response throughout
settle faster than the other structures for most disturbances. The the plant, whereas HS responds unevenly in different sections
only instance where IF settles faster is for d9. On the other hand, of the plant. The product column is comparatively less affected
a look at the DDS values in Table 22 leads to the conclusion compared to the recycle column in the case of HS. As for LS,
that the performance of IF and SOC is generally superior to the recycle flow changes more than expected for all the
that of LS and HS for most disturbances. Between LS and HS, disturbances. In the case of IF, the process variables in the two
Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009 10957
Table 23. Percentage Change in Important Variables in the Distillation Section for Control Structures: HS, LS, IF and SOC in the Presence of
Feed Rate Disturbancesa
Percentage Change [(Final - Initial) × 100/Initial]
d1 d2 d3
quantity HS LS IF SOC HS LS IF SOC HS LS IF SOC
Feed to T-1 –5.1 –5.1 –5.0 –4.9 5.1 5.0 4.9 4.9 –20 –20 –20 –20
recycle flow –7.3 –6.8 –5.2 –4.3 9.5 6.9 5.0 4.3 –27 –27 –20 –18
reboiler duty T-1 –1.5 –5.4 –5.7 –5.0 1.8 5.4 5.2 5.4 –5.8 –21 –21 –21
T-2 –6.9 –6.4 –6.6 –5.5 8.1 6.7 6.4 5.8 –26 –25 –25 –22
vapor boil-up T-1 –1.6 –5.6 –6.0 –5.2 1.9 5.6 5.4 5.7 –6.0 –22 –21 –21
T-2 –6.9 –6.6 –6.7 –5.7 7.8 6.7 6.6 6.0 –26 –25 –25 –23
a
The variables that respond significantly different than expected are shown in bold.

Figure 7. Overall absolute accumulation transient for control structures:


IF and LS in the presence of disturbance d3.

columns (reboiler duty and boil-up rates) show a slightly


unbalanced response.
Thus, the snowball effect is more profound in HS and LS, as
indicated by the relatively larger percentage change in the
recycle. This in turn can be attributed to the fixing of the flow
in the recycle loop to avoid the snowball effect. This does not
totally eliminate the snowball effect, but just transfers it from
one part of the plant to another (in the case of styrene plant, to
the recycle column). Larsson et al.67 also made similar observa- Figure 8. Styrene production rate transients for control structures HS, IF,
tions in the case of a simple RSR process. They show that the SOC, and LS in the presence of disturbances: d2 (upper plot) and d3 (lower
structure with the recycle flow fixed results in instability even plot).
for a small increase in the feed rate, as the snowball effect is
transferred to the distillation column. They conclude that the an unbalanced response and greater snowball effect. While HS
rule of fixing the flow in the recycle loop has a limited basis. has been developed based on the PWC methodology and case
Similar observations are also made by Konda and Rangaiah53 studies presented in refs 2 and 39, LS has been proposed by
in the case of the HDA process. On the other hand, control of Luyben during the course of this manuscript’s review as
reactor and product column inlet compositions in SOC and the mentioned earlier. One possible reason for this difference in
presence of conversion controller in IF are the main reasons performance could be the use of feed-forward control effects
for the more balanced nature of these structures. In other words, in LS.
the impact of disturbances among different sections is more One final interesting result is the slower settling time of IF
equally distributed in SOC and IF. in response to d4, whereas the settling time is 0 for all the other
As mentioned earlier, the results in Table 22 present a structures. This is probably due to definition of settling time
conflicting picture in the case of LS. While the settling times based on production rate and fixing the EB feed flow in IF,
are generally the smallest with LS, the DDS values indicate whereas it is allowed to vary in the other structures. For d4,
relatively poorer performance. This can be attributed to this results in a 2% decrease in EB feed flow in IF as opposed
the greater amplitude of response during the initial transient to 0.75%, 0.14%, and 1% decrease in the case of HS, SOC,
state in the case of LS compared to IF and SOC. The overall and LS, respectively. The greater decrease in EB feed flow
absolute accumulation profiles shown in Figure 7 for control causes slower process response in the case of IF.
structures IF and LS in the presence of disturbance d3 clearly Figure 8 shows the styrene production rate transients due to
illustrate this point. As has been discussed by Konda and disturbances d2 and d3 for all the four control structures. While
Rangaiah,53 the DDS is again proven to be a better indicator of the production rate stabilizes at about the same time for d2 in
control system performance compared to settling time as it both IF and SOC, the transients for IF are smoother. LS shows
considers the magnitude of the process response during the the fastest settling time for both d2 and d3, but the production
transient state. rate shows larger variations during the transient state compared
Another aspect of the results presented is the relatively to IF and SOC. This again explains the reason for the larger
superior performance of LS compared to HS. This can be values of DDS for LS in spite of faster settling time. Thus, DDS
attributed to the difference in the response of the process is a better measure than process settling time as it also considers
variables in different sections of the plant (Table 23). HS shows the behavior of the process variables during the transient state.
10958 Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009

In general, SOC shows the best performance in terms of both Table A1. Controllers with their Parameters for LS Control
the speed of response and the magnitude of variations in the Structure for the Styrene Plant.
process variables. This is manifested in the relatively faster controller parameters
settling times (compared to IF) and the low values of DDS for controlled variable manipulated variable [Kc (%/%), Ti (min)]
most disturbances. reaction section
Taking all the aforementioned results into consideration, it total EB flow EB feed flow (TPMa) 0.5, 0.3
total EB/LP2 ratio LP2 flow 0.36, 0.035
can be concluded that, while all the PWC systems are generally total EB/LP1 ratio LP1 flow 0.1, 0.018
robust and stable in the presence of disturbances, the best control PFR1 inlet T Q1/total EB SP 0.21, 22
structure developed in this study for the styrene plant is IF as Q1/total EB (cascade) furnace duty 0.1, 5
PFR2 inlet T intermediate heater duty 0.96, 24
its performance is slightly better (as indicated by the DDS phase sep T cooling water flow 0.13, 0.14
values) for most disturbances and SOC is the second best. In phase sep P lights flow 2, 10
fact, the two are comparable (the total DDS is 197 for IF phase sep liquid % level organic flow 18.8, 0.45
phase sep aqueous % level water flow 1.31, 0.12
compared to 204 for SOC). Note that the focus of the
product column
performance assessment in this work has been on dynamic
performance. The operating profit (eq 7), which depends mostly condenser P condenser duty 2, 10
condenser level distillate flow 1.2
on the steady-state operation, was not considered. SOC is reboiler level bottoms flow 2.2
expected to perform best economically due to the selection of reflux/feed flow reflux flow SP 1, 5
self-optimizing controlled variables that achieve small steady- reflux flow (cascade) reflux flow 0.5, 0.3
stage 59 EB composition reboiler duty 0.23, 54
state economic losses for the expected disturbances. This vent/feed flow vent flow SP 0.47, 0.025
however does not rule out the possibility of the other control vent flow (cascade) compressor duty 0.5, 0.3
structures performing equally well economically. recycle column
In terms of the application of the three methods, our condenser P condenser duty 1.8, 30
condenser level reflux flow 1.2
experience shows that the integrated framework is relatively reboiler level bottoms flow 2
simple and straightforward to apply, as heuristics/guidelines and stage 9 T distillate/reflux flow SP 6.7, 41
simulation tools are used together at each level to aid in the distillate/reflux flow (cascade) distillate flow 0.42, 0.026
stage 29 toluene composition reboiler duty 0.13, 36
analysis. The only difficulty seems to be in the analysis of the
a
effects of recycle in step 7. It is believed that this step can be TPM: throughput manipulator.
simplified by the introduction of some clear-cut guidelines and/
or mathematical tools. On the other hand, application of Appendix A
Luyben’s heuristic procedure poses a few difficulties. First, the During the course of review of this publication, an alternative
selection of manipulated variables for the distillation columns control structure was proposed by Prof. William L. Luyben for
requires considerable effort and engineering judgment as RGA the styrene plant. This control structure is summarized in Table
analysis is not used in the selection. Second and more A1 together with the tuning parameters. The tuning parameters
importantly, selecting the inventory control configuration for for all the major loops like temperature and composition are
the recycle column is a difficult task. It is not clear when to use the same as those given by Prof. Luyben.
the manipulation of fresh makeup feed to control the column
level. Luyben39 suggests this alternative when the recycle flow Appendix B
is large compared to the feed flow; but, the question is how
DDS is a new dynamic performance index recently developed
large. In the current case, the recycle flow is around 48% of
by Konda and Rangaiah.53 The basic idea is to make effective
the fresh feed and around 33% of the total feed (that is fresh
use of rigorous process simulators in the performance assessment
feed plus recycle). As for the self-optimizing control, though
of PWC systems. DDS makes use of the strong correlation
its application is straightforward, many more steps and extensive between the overall control system performance and the sum
computations, especially in the steps that deal with the local of the individual component accumulations. When a process is
linear analysis and the evaluation of the loss, are involved. affected by a disturbance, all the process variables go though
different transients and ultimately reach a steady-state only when
7. Conclusions the overall accumulation of all the components becomes zero
or negligible. The time integral of sum of the absolute
In this study, steady-state and dynamic HYSYS simulation accumulation of all the components in the entire plant is a
models have been developed for the styrene monomer plant. measure of the PWC structure performance, and so DDS is thus
Three PWC methodologies, namely, heuristics procedure,2 defined as
integrated framework of heuristics and simulation,1 and self-
optimizing control procedure,3 are then applied to develop three DDS )
alternative control structures for this plant. The control structures
∫ ∑
ts
( (absolute accumulation of component i)) dt
developed together with the one proposed by Luyben are then t)0
i)1 to n
analyzed for their performance in the presence of expected
disturbances. It is found that all the PWC methodologies give where ts is the time taken for the process to reach steady-state
viable control structures that stabilize the styrene plant for all and n is total number of components. Obviously, a smaller value
the disturbances tested. In particular, the integrated framework of DDS indicates better control. Within the context of this paper,
and self-optimizing control structures perform better and are DDS is computed for the entire plant but not for the individual
more or less equally robust. The control structure proposed by units. The latter is however more appropriate to study the
Luyben is also viable, though it performs poorer for the more performance of individual units for troubleshooting a given
important feed rate disturbances. control structure, which is beyond the scope of this work.
Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009 10959
DDS is a dynamic and not a steady-state performance by ignoring the concentration of components that are present
measure. It is more realistic and reliable as it includes level in small or negligible quantities; the calculated DDS will still
and pressure dynamics, which would not be possible in steady- be reasonably accurate. Furthermore, the computation procedure
state procedures. It is relatively simple to compute DDS in a can be easily automated as it remains the same for a given
process simulator. This computation can be further simplified process flowsheet irrespective of its control structure.

Table C1. Control Structure Decisions for HS Control Structure at Each Level
level selected controlled variables selected manipulated variables reason for selection
3 PFR1 T furnace duty steady-state gain analysis
PFR2 T intermediate heater duty most direct manipulator
4 total EB feed flow fresh EB feed flow most direct manipulator
5 T-1 bottoms EB composition reboiler duty best manipulator based on heuristics
steam/EB ratio at PFR1 inlet steam feed flow
T-1 pressure condenser duty
V-1 temperature cooling water flow
split steam flow split steam flow
6 V-1 pressure lights flow Richardson’s rule
T-2 pressure condenser duty
V-1 liquid level organic flow
V-1 aqueous level water flow
T-1 condenser level distillate flow
T-1 reboiler level bottoms flow
T-2 condenser level reflux flow
T-2 reboiler level bottoms flow
7 T-2 distillate EB composition distillate flow follows from selection made for recycle column condenser level in level 6
8 T-1 reflux T-1 reflux heuristics
overhead vent flow compressor duty most direct manipulator
T-2 bottoms toluene composition reboiler duty follows from selection made for recycle column reboiler level in level 6

Table C2. Control Structure Decisions for IF Control Structure at Each Level
level selected controlled variables selected manipulated variables reason for selection
3.1 EB fresh feed flow EB fresh feed flow heuristics
3.2 T-1 bottoms EB composition reboiler duty RGA analysis
4.1 split steam flow split steam flow best manipulator based on simulation
steam/total EB ratio at PFR1 inlet steam feed flow
PFR1 T furnace duty
PFR2 T intermediate heater duty
V-1 T cooling water flow
T-1 pressure condenser duty
4.2 V-1 pressure lights flow most direct manipulator
T-2 pressure condenser duty
V-1 liquid level organic flow
V-1 aqueous level water flow
T-1 condenser level distillate flow heuristics with simulation
T-1 reboiler level bottoms flow
T-2 condenser level reflux flow
T-2 reboiler level bottoms flow
5 T-1 distillate styrene composition reflux flow follows from selections made in level 4.2
T-2 distillate EB composition distillate flow
T-2 bottoms toluene composition reboiler duty
vent flow compressor duty simulation
7 EB conversion PFR1 T set-point most promising manipulator

Table C3. Control Structure Decisions for SOC Control Structure at Each Level
level selected controlled variables selected manipulated variables reason for selection
5.1 PFR1 T furnace duty RGA analysis
PFR2 T intermediate heater duty most direct manipulator
V-1 liquid level organic flow most direct manipulator
V-1 aqueous level water flow most direct manipulator
T-1 condenser level distillate flow RGA analysis
T-1 reboiler level bottoms flow RGA analysis
T-2 condenser level distillate flow RGA analysis
T-2 reboiler level bottoms flow RGA analysis
V-1 pressure lights flow most direct manipulator
T-1 pressure condenser duty most direct manipulator
T-2 pressure condenser duty most direct manipulator
furnace outlet T steam split ratio set-point RGA analysis
V-1 T cooling water flow most direct manipulator
T-2 stage 6 T reflux flow RGA analysis
total EB feed flow EB fresh feed flow heuristics
steam feed flow steam feed flow most direct manipulator
6 PFR1 inlet EB composition steam feed flow set-point RGA analysis
T-1 feed styrene composition PFR2 T set-point RGA analysis
T-1 distillate styrene composition reflux flow RGA analysis
T-1 bottoms EB composition reboiler duty RGA analysis
T-2 distillate EB composition T-2 stage 6 T set-point RGA analysis
T-2 bottoms toluene composition reboiler duty RGA analysis
vent flow compressor duty most direct manipulator
10960 Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009

Appendix C (27) Ricker, N. L.; Lee, J. H. Non-Linear Model-Predictive Control of


the Tennessee Eastman Challenge Process. Comput. Chem. Eng. 1995, 19,
This section details the control structure decisions for the HS, 961–981.
IF, and SOC control structures at each level in Tables C1- (28) Banerjee, A.; Arkun, Y. Control Configuration Design Applied to
the Tennessee Eastman Plant-Wide Control Problem. Comput. Chem. Eng.
C3. 1995, 19, 453–480.
(29) McAvoy, T. J.; Ye, N.; Gang, C. Nonlinear Inferential Parallel
Cascade Control. Ind. Eng. Chem. Res. 1996, 35, 130–137.
Literature Cited (30) Ricker, N. L. Decentralized Control of The Tennessee Eastman
Challenge Problem. J. Process Control 1996, 6, 205–221.
(1) Konda, N. V. S. N. M.; Rangaiah, G. P.; Krishnaswamy, P. R. Plant-
(31) Luyben, M. L.; Tyreus, B. D.; Luyben, W. L. Plant-Wide Control
Wide Control of Industrial Processes: An Integrated Framework of
Design Procedure. AIChE J. 1997, 43, 3161–3174.
Simulation and Heuristics. Ind. Eng. Chem. Res. 2005, 44, 8300–8313.
(32) McAvoy, T. J. Synthesis of Plant-Wide Control Systems Using
(2) Luyben, W. L.; Tyreus, B. D.; Luyben, M. L. Plant-Wide Process
Optimization. Ind. Eng. Chem. Res. 1999, 38, 2984–2994.
Control; McGraw-Hill: New York, 1998.
(3) Skogestad, S. Control Structure Design for Complete Chemical (33) Tian, Z. H.; Hoo, K. A. Multiple Model-Based Control of the
Plants. Comput. Chem. Eng. 2004, 28, 219–234. Tennessee Eastman Process. Ind. Eng. Chem. Res. 2005, 44, 3187–3202.
(4) Buckley, P. S. Techniques of Process Control; Wiley: New York, (34) Stephanopoulos, G. Chemical Process Control; Prentice Hall: New
1964. Jersey, 1984.
(5) Larsson, T.; Skogestad, S. Plant-Wide Control - A Review and a (35) Ponton, J. W.; Laing, D. M. A Hierarchical Approach to the Design
New Design Procedure. Model., Ident. Control 2000, 21 (4), 209–240. of Process Control Systems. Trans. Inst. Chem. Eng. 1993, 71, 181–188.
(6) Konda, N. V. S. N. M.; Rangaiah, G. P.; Krishnaswamy, P. R. (36) Fonyo, Z. Design Modifications and Proper Plant-Wide Control.
Systematic Classification of Plant-Wide Control System Design Procedures Comput. Chem. Eng. 1994, 18, S483–S492.
with Application to the HDA Process. Presented at AIChE Annual Meeting, (37) Ng, C. S.; Stephanopoulos, G. Synthesis of Control Systems for
San Fransisco, Nov 2003. Chemical Plants. Comput. Chem. Eng. 1996, 20, S999–S1004.
(7) Skogestad, S. Plant-Wide Control: The Search for the Self- (38) Cao, Y.; Rossister, D. An Input Pre-Screening Technique for
Optimizing Control Structure. J. Process Control 2000, 10, 487–507. Control Structure Selection. Comput. Chem. Eng. 1997, 21, 563–569.
(8) Skogestad, S. Self-Optimizing Control: The Missing Link Between (39) Luyben, W. L. Plant-Wide Dynamic Simulators in Chemical
Steady-State Optimization and Control. Comput. Chem. Eng. 2000, 24, 569– Processing and Control; Marcel Dekker: New York, 2002.
575. (40) Herrmann, G.; Spurgeon, S. K.; Edwards, C. A Model-Based Sliding
(9) Jorgensen, J. B.; Jorgensen, S. B. Towards Automatic Decentralized Mode Control Methodology Applied to the HDA Plant. J. Process Control
Control Structure Selection. Comput. Chem. Eng. 2000, 24, 841–846. 2003, 13, 129–138.
(10) Zhu, G. Y.; Henson, M. A.; Ogunnaike, B. A. A Hybrid Model (41) Qiu, Q. F.; Rangaiah, G. P.; Krishnaswamy, P. R. Application of
Predictive Control Strategy for Non-Linear Plant-Wide Control. J. Process a Plant-Wide Control Design to the HDA Process. Comput. Chem. Eng.
Control 2000, 10, 449–458. 2003, 27, 73–94.
(11) Groenendijk, A. J.; Dimian, A. C.; Iedema, P. D. Systems Approach (42) Bildea, C. S.; Dimian, A. C. Fixing Flow Rates in Recycle Systems:
for Evaluating Dynamics and Plant-Wide Control of Complex Plants. AIChE Luyben’s Rule Revisited. Ind. Eng. Chem. Res. 2003, 42, 4578–4585.
J. 2000, 46, 133–145. (43) Vasbinder, E. M.; Hoo, K. A.; Mann, U. Synthesis of Plant-Wide
(12) Dimian, A. C.; Groenendijk, A. J.; Iedema, P. D. Recycle Interaction Control Structures using a Decision-Based Methodology. In The Integration
Effects on the Control of Impurities in a Complex Plant. Ind. Eng. Chem. of Process Design and Control; Seferlis, P., Georgiadis, M. C., Eds.;
Res. 2001, 40, 5784–5794. Elsevier: New York, 2004; pp 375-400.
(13) Robinson, D.; Chen, R.; McAvoy, T.; Schnelle, P. D. An Optimal (44) Konda, N. V. S. N. M.; Rangaiah, G. P.; Lim, D. K. H. Optimal
Control Based Approach to Designing Plant-Wide Control System Archi- Process Design and Effective Plant-Wide Control of Industrial Processes
tectures. J. Process Control. 2001, 11, 223–236. by a Simulation-Based Heuristic Approach. Ind. Eng. Chem. Res. 2006,
(14) Kookos, I. K.; Perkins, J. D. Heuristic-Based Mathematical 45, 5955–5970.
Programming Framework for Control Structure Selection. Ind. Eng. Chem. (45) Araujo, A. C. B.; Govatsmark, M.; Skogestad, S. Application of
Res. 2001, 40, 2079–2088. Plant-Wide Control to the HDA Process. I - Steady-State Optimization and
(15) Wang, P.; McAvoy, T. Synthesis of Plant-Wide Control Systems Self-Optimizing Control. Control Eng. Pract. 2007, 15, 1222–1237.
Using a Dynamic Model and Optimization. Ind. Eng. Chem. Res. 2001, (46) Araujo, A. C. B.; Hori, E. S.; Skogestad, S. Application of Plant-
40, 5732–5742. Wide Control to the HDA Process. II - Regulatory Control. Ind. Eng. Chem.
(16) Chen, R.; McAvoy, T. Plant-Wide Control System Design: Res. 2007, 46, 5159–5174.
Methodology and Application to a Vinyl Acetate Process. Ind. Eng. Chem. (47) Bouton, G. R.; Luyben, W. L. Optimum Economic Design and
Res. 2003, 42, 4753–4771. Control of a Gas Permeation Membrane Coupled with the HDA Process.
(17) Chen, R.; McAvoy, T.; Zafiriou, E. Plant-Wide Control System Ind. Eng. Chem. Res. 2008, 47, 1221–1237.
Design: Extension to Multiple-Forcing and Multiple-Steady-State Operation. (48) Olsen, D. G.; Svrcek, W. Y.; Young, B. R. Plant-Wide Control
Ind. Eng. Chem. Res. 2004, 43, 3685–3694. Study of a Vinyl Acetate Monomer Process Design. Chem. Eng. Commun.
(18) Vasbinder, E. M.; Hoo, K. A. Decision-Based Approach to Plant- 2005, 192, 1243–1257.
Wide Control Structure Synthesis. Ind. Eng. Chem. Res. 2003, 42, 4586– (49) Turkay, M.; Gurkan, T.; Ozgen, C. Synthesis of Regulatory Control
4598. Structures for a Styrene Plant. Comput. Chem. Eng. 1993, 17, 601–608.
(19) Cao, Y.; Saha, P. Improved Branch and Bound Method for Control (50) Zhu, G. Y.; Henson, M. A. Model Predictive Control of Inter-
Structure Screening. Chem. Eng. Sci. 2005, 60, 1555–1564. connected Linear and Non-Linear Processes. Ind. Eng. Chem. Res. 2002,
(20) Cao, Y.; Kariwala, V. Bidirectional Branch and Bound for 41, 801–816.
Controlled Variable Selection. Part 1. Principles and Minimum Singular (51) Al-Arfaj, M. A.; Luyben, W. L. Plant-Wide Control for TAME
Value Criterion. Comput. Chem. Eng. 2008, 32, 2306–2319. Production Using Reactive Distillation. AIChE J. 2004, 50, 1462–1473.
(21) Baldea, M.; Araujo, A.; Skogestad, S.; Daoutidis, P. Dynamic (52) Tarafder, A.; Rangaiah, G. P.; Ray, A. K. Multi-objective Optimiza-
Considerations in the Synthesis of Self-Optimizing Control Structures. tion of an Industrial Styrene Monomer Manufacturing Process. Chem. Eng.
AIChE J. 2008, 54, 1830–1841. Sci. 2005, 60, 347–363.
(22) Downs, J. J.; Vogel, E. F. A Plant-Wide Industrial Process Control (53) Konda, N. V. S. N. M.; Rangaiah, G. P. Performance Assessment
Problem. Comput. Chem. Eng. 1993, 17, 245–255. of Plant-Wide Control Systems of Industrial Processes. Ind. Eng. Chem.
(23) McAvoy, T. J.; Ye, N. Base Control for the Tennessee Eastman Res. 2007, 46, 1220–1231.
Problem. Comput. Chem. Eng. 1994, 18, 383–413. (54) Araujo, A. C. B.; Skogestad, S. Control Structure Design for the
(24) Price, R. M.; Lyman, P. R.; Georgakis, C. Throughput Manipulation Ammonia Synthesis Process. Comput. Chem. Eng. 2008, 32, 2920–2932.
in Plant-Wide Control Structures. Ind. Eng. Chem. Res. 1994, 33, 1197– (55) Meili, A. Optimize Styrene Distillation. Hydrocarbon Process.
1207. 1998, 77 (12), 85–89.
(25) Lyman, P. R.; Georgakis, C. Plant-Wide Control of the Tennessee (56) Denis, H. J.; Castor, W. M. Styrene. In Ullmann’s Encyclopedia
Eastman Problem. Comput. Chem. Eng. 1995, 19, 321–331. of Industrial Chemistry, A25; Elvers, B., Hawkin, S., Russey, W., Eds.;
(26) Ye, N.; McAvoy, T. J.; Kosanovich, K. A.; Piovoso, M. J. Optimal John Wiley and Sons: New York, 1992; pp 325-335.
Averaging Level Control for the Tennessee Eastman Problem. Can. J. Chem. (57) Woodle, G. B. Styrene. In Encyclopedia of Chemical Processing;
Eng. 1995, 73, 234–240. Lee, S., Eds.; Taylor and Francis: New York, 2006; 2859-2869.
Ind. Eng. Chem. Res., Vol. 48, No. 24, 2009 10961
(58) Sheel, J. C. P.; Crowe, C. M. Simulation and Optimization of an (64) Fruehauf, P. S.; Mahoney, D. P. Distillation Column Control Design
Existing Ethyl Benzene Dehydrogenation Reactor. Can. J. Chem. Eng. 1969, Using Steady-State Models: Usefulness and Limitations. ISA Trans. 1993,
47, 183–187. 32 (2), 157–175.
(59) Elnashaie, S. S. E. H.; Elshishini, S. S. Modelling, Simulation and (65) Turton, R.; Bailie, R. C.; Whiting, W. B.; Shaeiwitz, J. A. Analysis,
Optimization of Industrial Fixed Bed Catalytic Reactors. In Topics in Synthesis, and Design of Chemical Processes; Prentice Hall: New Jersey,
Chemical Engineering; Gordon and Breach Science Publishers: New York, 2003.
1993; Vol. 7. (66) Riggs, J. B. Improve Distillation Column Control. Chem. Eng. Prog.
(60) Lee, C. H.; Hubbell, O. S. Styrene. In Encyclopedia of Chemical 1998, 94 (10), 31–47.
Processing and Design; McKetta, J., Weismantel, G. E., Eds.; Wiley: New
(67) Larsson, T.; Govatsmark, M. S.; Skogestad, S.; Yu, C. C. Control
York, 1982; Vol. 55, pp 197-217.
Structure Selection for Reactor, Separator, and Recycle Processes. Ind. Eng.
(61) Konda, N. V. S. N. M.; Rangaiah, G. P.; Krishnaswamy, P. R. A
Chem. Res. 2003, 42, 1225–1234.
Simple and Effective Procedure for Control Degrees of Freedom. Chem.
Eng. Sci. 2006, 61, 1184–1194.
(62) Vasudevan, S.; Konda, N. V. S. N. M.; Rangaiah, G. P. Control ReceiVed for reView January 6, 2009
Degrees of Freedom Using the Restraining Number: Further Evaluation. ReVised manuscript receiVed September 21, 2009
Asia-Pacific J. Chem. Eng. 2008, 3, 638–647. Accepted September 25, 2009
(63) Svrcek, W. Y.; Mahoney, D. P.; Young, B. R. A Real-Time
Approach to Process Control; John Wiley and Sons: New York, 2000. IE900022H

You might also like