Está en la página 1de 68

Accepted Manuscript

Title: Integration of process design and controller design for


chemical processes using model-based methodology
Authors: Mohd Kamaruddin Abd Hamid, G urkan Sin, Raqul
Gani
PII: S0098-1354(10)00029-3
DOI: doi:10.1016/j.compchemeng.2010.01.016
Reference: CACE 3966
To appear in: Computers and Chemical Engineering
Received date: 3-9-2009
Revised date: 7-1-2010
Accepted date: 21-1-2010
Please cite this article as: Hamid, M. K. A., Sin, G., & Gani, R.,
Integration of process design and controller design for chemical processes
using model-based methodology, Computers and Chemical Engineering (2008),
doi:10.1016/j.compchemeng.2010.01.016
This is a PDF le of an unedited manuscript that has been accepted for publication.
As a service to our customers we are providing this early version of the manuscript.
The manuscript will undergo copyediting, typesetting, and reviewof the resulting proof
before it is published in its nal form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that
apply to the journal pertain.
Page 1 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
1
Integration of process design and controller design for chemical 1
processes using model-based methodology 2
3
Mohd Kamaruddin Abd Hamid, Grkan Sin
-
and Rafiqul Gani 4
5
Computer Aided Process-Product Engineering Center (CAPEC), Department of Chemical and 6
Biochemical Engineering, Technical University of Denmark, DK-2800, Kgs. Lyngby, 7
Denmark. 8
9
10
Abstract 11
In this paper, a novel systematic model-based methodology for performing integrated 12
process design and controller design (IPDC) for chemical processes is presented. The 13
methodology uses a decomposition method to solve the IPDC typically formulated as a 14
mathematical programming (optimization with constraints) problem. Accordingly the 15
optimization problem is decomposed into four sub-problems (i) pre-analysis, (ii) design 16
analysis, (iii) controller design analysis, and (iv) final selection and verification, which are 17
relatively easier to solve. The methodology makes use of thermodynamic-process insights and 18
the reverse design approach to arrive at the final process design-controller design decisions. 19
The developed methodology is illustrated through the design of: (a) a single reactor, (b) a 20
single separator, and (c) a reactor-separator-recycle system and shown to provide effective 21
solutions that satisfy design, control and cost criteria. The advantage of the proposed 22

-
Corresponding author. Tel.: +45 45 252 806; Fax: +45 45 932 906; Email: gsi@kt.dtu.dk
*Manuscript
Page 2 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
2
methodology is that it is systematic, makes use of thermodynamic-process knowledge and 23
provides valuable insights to the solution of IPDC problems in chemical engineering practice. 24
25
Keywords: Model-based methodology; process design, controller design; decomposition 26
method; graphical method, integration. 27
28
29
1. Introduction 30
31
Traditionally, process design and controller design are two separate problems that are 32
dealt with sequentially. The process is designed first to achieve the design objectives, and 33
then, the operability and control aspects are analyzed and resolved to obtain the controller 34
design. This traditional sequential approach is often inadequate since many process control 35
challenges arise because of poor design of the process and may lead to overdesign of the 36
process, dynamic constraint violations, and may not guarantee robust performance (Malcom et 37
al., 2007). Another drawback has to do with how process design decisions influence the 38
controllability of the process. To assure that design decisions give the optimum economic and 39
the best control performance, controller design issues need to be considered simultaneously 40
with the process design issues. The research area of combining process design and controller 41
design considerations is referred here as integrated process design and controller design 42
(IPDC). One way to achieve IPDC is to identify variables together with their target values that 43
have roles in process design (where the optimal values of a set of design variables are obtained 44
to match specification on a set of process variables) and controller design (where the same set 45
of design variables serve as the actuators or manipulated variables and the same set of process 46
Page 3 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
3
variables become the controlled variables). Also, the optimal design values become the set- 47
points for the controlled and manipulated variables. Using model analysis, controllability 48
issues are incorporated to pair the identified actuators with the corresponding controlled 49
variables. The integrated design problem is therefore reduced to identifying the dual purpose 50
design-actuator variables, the process-controlled variables, their sensitivities, their target- 51
setpoint values, and their pairing. 52
The importance of an integrated process-controller design approach, considering 53
operability together with the economic issues, has been widely recognized (Allgor and Barton, 54
1999; Bansal et al., 2000; Bansal et al., 2003; Kookos and Perkins, 2001; Luyben, 2004; 55
Meeuse and Grievink, 2004; Patel et al., 2008; Ricardez Sandoval et al., 2008; Schweiger and 56
Floudas, 1997). The objective has been to obtain a profitable and operable process, and control 57
structure in a systematic manner. The IPDC has advantage over the traditional-sequential 58
method because the controllability issues are resolved together with the optimal process 59
design issues. Meeuse and Grievink (2004) used the Thermodynamic Controllability 60
Assessment (TCA) technique to incorporate controllability issues into the design problem. The 61
IPDC problem, however, involved multi-criteria optimization and needed trade-off between 62
conflicting design and control objectives. For example, the process design issues point to 63
design of smaller process units in order to minimize the capital and operating costs, while, 64
process control issues point to larger process units in order to smooth out disturbances 65
(Luyben, 2004). 66
A number of methodologies have been proposed for solving IPDC problems (Sakizlis et 67
al., 2004; Seferlis and Georgiadis, 2004). In these methodologies, a mixed-integer non-linear 68
optimization problem (MINLP) is formulated and solved with standard MINLP solvers. The 69
continuous variables are associated with design variables (flow rates, heat duties) and process 70
Page 4 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
4
variables (temperatures, pressures, compositions), while binary (decision) variables deal with 71
flowsheet structure and controller structure. When an MINLP problem represents an IPDC, the 72
process model considers only steady state conditions, while a MIDO (mixed-integer dynamic 73
optimization) problem represents an IPDC where steady state as well as dynamic behaviour 74
are considered. 75
A number of algorithms have been developed to solve the MIDO problem. From an 76
optimization point of view, the solution approaches for MIDO problems can be divided into 77
simultaneous and sequential methods, where the original MIDO problem is reformulated into a 78
mixed-integer nonlinear program (MINLP) problem (Sakizlis et al., 2004). The former 79
method, also called complete discretization approach, transforms the original MIDO problem 80
into a finite dimensional nonlinear program (NLP) by discretization of the state and control 81
variables. Avraam et al. (1999), Flores-Tlacuahuac and Biegler (2007) and Mohideen et al. 82
(1996) applied this complete discretization approach and solved the resulting MINLP problem 83
using outer approximation (OA) and generalized Benders decomposition (GBD) frameworks. 84
However, this method typically generates a very large number of variables and equations, 85
yielding large NLPs that may be difficult to solve reliably (Exler et al., 2008; Patel et al., 86
2008), depending on the complexity of the process models. 87
As regards the sequential method, also called control vector parameterization approach, 88
only control variables are discretized. The MIDO algorithm is decomposed into a sequence of 89
primal problems (nonconvex DOs) and relaxed master problems (Bansal et al., 2003; 90
Mohideen et al., 1997; Schweiger and Floudas, 1997; Sharif et al., 1998). Because of 91
nonconvexity of the constraints in DO problems, such solution methods are possibly excluding 92
large portions of the feasible region within which an optimal solution may occur, leading to 93
the suboptimal solutions (Chachuat et al., 2005). 94
Page 5 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
5
In order to overcome convergence to the suboptimal solution in DO or MIDO problems, 95
stochastic and deterministic global optimization (GO) methods have also been proposed. 96
Regarding stochastic GO methods, a number of works have shown that the region of global 97
solutions can be located with relative efficiency (Banga et al., 2003; Moles et al., 2003; Sendin 98
et al., 2004), but they tend to be computationally expensive and have difficulties with highly 99
constrained problems. Most importantly, their major drawback is that global optimality cannot 100
be guaranteed. While deterministic GO methods can guarantee that the optimal performance 101
has been found (Esposito & Floudas; 2000), however their applicability is limited only to 102
problems with medium complexity (Moles et al., 2003). 103
The objective of this paper is to present an alternative systematic model-based IPDC 104
approach that is simple to apply, easy to visualize and efficient to solve. Here, the IPDC 105
problem is solved by the so-called reverse approach (reverse design algorithm) by 106
decomposing it into four sequential hierarchical sub-problems: (i) pre-analysis, (ii) design 107
analysis, (iii) controller design analysis, and (iv) final selection and verification (Hamid and 108
Gani, 2008). Using thermodynamic and process insights, a bounded search space is first 109
identified. This feasible solution space is further reduced to satisfy the process design and 110
controller design constraints in sub-problems 2 and 3, respectively, until in the final sub- 111
problem all feasible candidates are ordered according to the defined performance criteria 112
(objective function). The final selected design is verified through rigorous simulation. In the 113
pre-analysis sub-problem, the concepts of attainable region (AR) and driving force (DF) are 114
used to locate the optimal process-controller design solution (see section 2.4) in terms of 115
optimal condition of operation from design and control viewpoints. While other optimization 116
methods may or may not be able to find the optimal solution, depending on the performance of 117
Page 6 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
6
their search algorithms and computational demand, the use of AR and DF concepts is simple 118
and able to find at least near-optimal designs (if not optimal) to IPDC problems. 119
This paper is organized as follows. First the new model-based IPDC methodology 120
together with the decomposition into sub-problems and the methods used within the sub- 121
problems are introduced in Section 2. Then, in Section 3, the application of the IPDC 122
methodology in solving process design-controller design problems related to of a single 123
reactor, a single separator, and a reactor-separator-recycle system are presented and discussed. 124
Finally, future perspectives and conclusions are presented. 125
126
127
2. The I PDC Methodology 128
129
2.1 Problem formulation 130
131
The IPDC problem is typically formulated as a generic optimization problem in which a 132
performance objective in terms of design, control and cost is optimized subject to a set of 133
constraints: process (dynamic and steady state), constitutive (thermodynamic states) and 134
conditional (process-control specifications) 135
136

= =
=
m
i
n
j
j j i
w P J
1 1
,
(1) 137
s.t. 138
Process (dynamic and/or steady state) constraints 139
( ) t Y f dt d , , , , , d x u x = (2) 140
Page 7 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
7
Constitutive (thermodynamic) constraints 141
x v, = ) ( 0
1
g (3) 142
Conditional (process-control) constraints 143
( ) x u, 0
1
h = (4) 144
( ) d x u , , h
2
0 s (5) 145
Y CS u x + = (6) 146
147
In the above equations, x is the set of process (controlled) variables; usually 148
temperatures, pressures and compositions. u is the set of design (manipulated) variables. d is 149
the set of disturbance variables, is the set of constitutive variables (physical properties, 150
reaction rates), v is the set of chemical system variables (molecular structure, reaction 151
stoichiometry, etc.) and t is the independent variable (usually time). The performance 152
function, Eq. (1) includes design, control and cost, where i indicates the category of the 153
objective function term and j indicates a specific term of each category.
j
w is the weight 154
factor assigned to each objective term
j i
P
,
(i = 1, 3; j = 1, 2). 155
Eq. (2) represents a generic dynamic process model from which the steady state model is 156
obtained by setting 0 = dt dx . Eq. (3) represents constitutive equations which relate the 157
constitutive variables to the process and chemical system variables. Eqs. (4) (5) represent 158
sets of equality and inequality constraints (such as product purity, chemical ratio in a specific 159
stream) that must be satisfied for feasible operation - they can be linear or non-linear. In Eq. 160
(6), Y is the set of binary decision variables for the controller structure selection (corresponds 161
to whether a controlled variable is paired with a particular manipulated variable or not). 162
Different optimization scenarios can be generated as follows: 163
Page 8 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
8
164
- Maximize
j
P
, 1
to achieve process design objectives. Here, maximize
1 , 1
P the 165
performance criteria for reactor design and maximize
2 , 1
P the performance criteria for 166
separator design. 167
- Minimize-maximize
j
P
, 2
to achieve the control objectives. Here,
1 , 2
P is minimized by 168
minimizing ( d x d d ) the sensitivity of controlled variables x with respect to disturbances 169
d, and
2 , 2
P is maximized by maximizing ( x u d d ) the sensitivity of manipulated 170
variables u with respect to controlled variables x for the best controller structure 171
(controlled-manipulated pairing). 172
- Minimize
j
P
, 3
to achieve the economic objectives. Here,
1 , 3
P is minimized by 173
minimizing the capital cost and
2 , 3
P is minimized by minimizing the operating costs. 174
175
The multi-objective function in Eq. (1) is reformulated as, 176
177
2 , 1 ) 1 ( ) 1 (
, 3 , 3 2 , 2 2 , 2 1 , 2 1 , 2 , 1 , 1
= + + + = j P w P w P w P w J
j j j j
(7) 178
179
180
2.2 Decomposition-based solution strategy 181
182
In most of IPDC problems, the feasible solutions to the problems may lie in a relatively 183
small portion of the search space due to the large number of constraints involved. The ability 184
to solve such problems depends the effectiveness of the method of solution in identifying and 185
Page 9 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
9
locating the feasible solutions (one of these is the optimal solution). Hence, one approach to 186
solve this IPDC problem is to apply a decomposition method as illustrated in Fig. 1. The basic 187
idea here is that in optimization problems with constraints, the search space is defined by the 188
constraints within which all feasible solutions lie and the objective function helps to identify 189
one or more of the optimal solutions. In the simultaneous approach, all the constraint 190
equations are solved together with the objective function to determine the values of the 191
optimization variables (design-manipulated and decision variables) that satisfy the constraints 192
and lead to the optimal objective function value. In the decomposition-based approach 193
(Karunanithi et al., 2005) the constraint equations are solved in a pre-determined sequence 194
such that after every sequential sub-problem, the search space for feasible solutions is reduced 195
and a sub-set of design-manipulated and/or decision variables are fixed. When all the 196
constraints are satisfied, it remains to calculate the objective function for all the identified 197
feasible solutions to locate the optimal. 198
The IPDC problem is decomposed into four hierarchical stages: (1) pre-analysis, (2) 199
design analysis, (3) controller design analysis, and (4) final selection and verification. As 200
shown in Fig. 1, the set of constraint equations in the IPDC problem is decomposed into four 201
sub-problems which correspond to four hierarchical stages (see Figs. 1 2). In this way, the 202
solution of the decomposed set of sub-problems is equivalent to that of the original problem. 203
As each sub-problem is being solved, a large portion of the infeasible solution of the search 204
space is identified and eliminated, thereby leading to a final sub-problem that is significantly 205
smaller, which can be solved more easily. Therefore, while the sub-problem complexity may 206
increase with every subsequent stage, the number of feasible solutions is reduced at every 207
stage, as illustrated in Fig. 2. 208
Page 10 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
10
Stage 1: Pre-analysis. The objective of this stage is to define the operational window and 209
set the targets for the design-controller solution. First, all x and u are analyzed and the 210
important ones with respect to the multi-objective function, Eq. (7) are shortlisted. The 211
operational window is defined in terms of x and u (note that d is known). Choice is made 212
for x based on thermodynamic-process insights and Eq. (3) (also defines the optimal solution 213
targets). Then Eqs. (4) (5) are solved (for u ) to establish the operational window. For each 214
reactor task, an attainable region (AR) is drawn and the location of the maximum in the AR is 215
selected as the reactor design target. This point gives the highest selectivity of the reaction 216
product with respect to the limiting and/or a selected reactant. Similarly, for each separation 217
task, the design target is selected at the highest driving force (DF). Note that, both plots of AR 218
and DF have a well defined maximum. 219
Stage 2: Design analysis. The search space within the operational window identified in 220
stage 1 is further reduced in this stage. The objective is to validate the targets defined in stage 221
1 by finding acceptable values (candidates) of x and u by considering Eq. (2) the steady 222
state process model. If the acceptable values cannot be found or the solution is located outside 223
the operational window, then a new target is selected and the procedure is repeated until a 224
suitable match is found. 225
Stage 3: Controller design analysis. The search space is further reduced by considering 226
now the feasibility of the process control. This sub-problem considers the process model 227
constraints, Eq. (2) (dynamic and steady state forms) to evaluate the controllability 228
performance of feasible candidates, and Eq. (6) for the selection of the controller structure. In 229
this respect, two criteria are analyzed: (a) sensitivity ( d x d d ) of controlled variables x with 230
respect to disturbances d, which should be low, and (b) sensitivity ( x u d d ) of manipulated 231
variables u with respect to controlled variables x , which should be high. Lower value of 232
Page 11 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
11
d x d d means the process has lower sensitivity with respect to disturbances, hence the process 233
is more robust in maintaining its controlled variables against disturbances. On the other hand, 234
higher value of x u d d will determine the best pair of the controlled-manipulated variables (to 235
satisfy Eq. (6)) and also the optimal control action. It is assumed by this methodology that the 236
best set-point values of the controller are actually those already defined as design targets. It 237
should be noted that, the objective of this stage is not to find the optimal value of controller 238
parameters or type of controller, but to generate the feasible controller structures. 239
Stage 4: Final selection and verification. The final stage is to select the best candidates 240
by analyzing the value of the multi-objective function, Eq. (7). The best candidate in terms of 241
the multi-objective function will be verified using rigorous simulations or by performing 242
experiments. It should be noted that, the rigorous simulation will be easy because very good 243
estimates of x and u are obtained from stages 1 3. For controller performance verification 244
is made through open or closed loop simulations. For closed loop simulation, the standard 245
Cohen-Coon tuning method (Cohen and Coon, 1953) or any other tuning methods can be used 246
to determine the value of controller parameters. 247
248
2.3 The algorithm of decomposition-based methodology 249
250
Stage 1: Pre-analysis 251
a. Variables analysis 252
Analyze all x and u , and shortlist the important ones with respect to the multi- 253
objective functions, Eq. (7). 254
b. Operational window identification 255
Define the operational window in terms of x and u variables by solving Eqs. (4) (5). 256
Page 12 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
12
c. Design-control target identification 257
Draw AR and DF diagrams using Eq. (3) and identify design-control target by locating 258
the maximum points on the AR and DF diagrams. 259
260
Stage 2: Design analysis 261
Calculate the acceptable values (candidates) of x and u variables using steady state 262
process model of Eq. (2). 263
a. For reactor design: at the maximum point of AR, identify the corresponding value of 264
concentrations. Then find all other values of design (manipulated) and process 265
variables i.e., volume, flow rates. 266
b. For separator design: at the maximum point of DF and given desired product 267
composition, then find all other value of design (manipulated) and process variables 268
i.e., feed stage, reflux ratio, reboil ratio, reboiler and condenser duties. 269
270
Stage 3: Controller design analysis 271
a. Sensitivity analysis 272
Calculate d x d d using Eq. (2) to determine the process sensitivity with respect to 273
disturbances. 274
b. Controller structure selection 275
Calculate x u d d using Eq. (2) to determine the best pair of the controlled-manipulated 276
variables to satisfy Eq. (6). 277
278
Stage 4: Final selection and verification 279
a. Final selection: verification of design 280
Page 13 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
13
Evaluate the multi-objective function for the feasible candidates using Eq. (7) to select 281
the optimal. 282
b. Dynamic rigorous simulations: verification of controller performance 283
Perform open or closed loop rigorous simulations. Solve Eqs. (2) (5). 284
285
2.4 Defining design targets 286
287
The AR concept is used in this methodology to find the optimal (design target) values of 288
the process variables for any reaction system. Glasser et al., (1987, 1990) considered a reactor 289
as a system where the only processes occurring are reaction and mixing. For given kinetics 290
and given feeds, it might be possible to find the set of outputs from all possible reactor 291
systems. They have also shown that once the AR is found the optimization of the problem is 292
straightforward. If one knows the AR, one can then search all over the entire region (often the 293
boundary) to find the output conditions that maximize an objective function. In this paper, the 294
AR-concept is used to determine the maximum of the objective function (
1 , 1
P ) (for reactor 295
design) in terms of selectivity or maximum concentration of the reaction product. 296
Similarly, the DF concept is used in this methodology to find the optimal (design target) 297
values of the process variables for separation systems. Gani and Bek-Pedersen, (2000) 298
proposed a design method based on identification of the largest DF, defined as the difference 299
in composition of a component i between the vapor phase and the liquid phase, which is 300
caused by the difference in the volatilities of component i and all other components in the 301
system. This DF is calculated for a binary mixture or a binary pair of key components of a 302
multi component mixture. As the DF approaches zero, separation of the corresponding key 303
component i from the mixture becomes more difficult. On the other hand, as the DF 304
Page 14 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
14
approaches a maximum, separation becomes easier and the energy necessary to maintain the 305
two-phase system is a minimum. Therefore, if the separation design is based on maximizing 306
the DF, it naturally leads to a highly energy efficient design and the optimal objective function 307
value (
2 , 1
P ). 308
For each reactor design problem, the AR is drawn and the location of the maximum in the 309
AR is selected as the reactor design target. Similarly, for each separation design problem, the 310
DF is drawn and the design target is selected at the highest DF. From a process design point of 311
view, at these targets give the highest selectivity of the product with respect to limiting and/or 312
selected reactant for a reactor, and the lowest energy required for the separation. From a 313
controller design point of view, at these targets the controllability of the process is best 314
satisfied. At these targets, the value of d x d d is minimum and the value of x u d d is 315
maximum. According to Russel et al. (2002), the value of d x d d will determine process 316
sensitivity and flexibility with respect to disturbances. If d x d d is small, the process 317
sensitivity is low and process flexibility is high. This means that, at these targets the process is 318
more robust in maintaining its controlled variables at optimal set points in the presence of the 319
disturbances. On the other hand, the maximum value of x u d d will determine the best pair of 320
the controlled-manipulated variables and also the optimal control action. At these targets, the 321
best controller structure can be selected with the optimal control action. Therefore, by locating 322
the maximum point of the AR and DF as design targets, insights can be gained in terms of 323
controllability, and the optimal solution of the IPDC problems can be obtained in a systematic 324
manner. 325
326
327
Page 15 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
15
3. Applications 328
329
In this section, the solution of the IPDC problems through the proposed decomposition 330
methodology is presented for the design of: (i) a single reactor system, (ii) a single separator 331
system, and (iii) a reactor-separator-recycle system, involving the ethylene glycol production 332
process. 333
334
3.1 Application to a single reactor design 335
336
In this section we consider a simple case study that is related to IPDC problem. We 337
consider the following situation. In a continuous stirred tank reactor (CSTR), the product 338
ethylene glycol (EG) is to be produced from ethylene oxide (EO) and water (W). The 339
production of EG involves an isothermal, irreversible liquid phase reactions and can be 340
represented as follows: 341
342
O
k1
+ H
2
O HO
OH
(8) 343
HO
OH
k
2
OH HO
O
O
+
(9) 344
k
3
OH HO
O
O
+
O HO
O OH
(10) 345
346
where, EO and W react to produce EG in Eq. (8). Eqs. (9) - (10) are the side-reactions where 347
EG reacts with EO to produce diethylene glycol (DEG), and DEG reacts with the remaining 348
Page 16 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
16
EO to produce triethylene glycol (TEG), respectively. The production of further glycols is 349
comparatively small and is therefore neglected. 350
351
) 10583 163 . 30 exp( 238 . 5
1
T k = ;
1 2
1 . 2 k k = ;
1 3
2 . 2 k k = (11) 352
353
EO and W are considered to be premixed at the same ratio of 1:1 (for justification, see section 354
3.1.2), and other component concentrations are zero in the given feed. The kinetic data in Eq. 355
(11) for the above reactions are taken from Parker and Prados (1964). The objective is then to 356
determine the design-control solution in which the multi-objective function, Eq. (7) is optimal. 357
A schematic of the process is depicted in Fig. 4. 358
359
3.1.1 Problem formulation 360
361
The IPDC problem for the EG production process described above is defined in terms of a 362
performance objective (Eq. (7)) and the three sets of constraints (process, constitutive and 363
conditional). 364
365
- Process constraints: 366
NC , i VR C F C F dt dC
i i , i , i ,
1
2 2 1 1 2
= = (12) 367
( )
R R i
Q H VR H F H F dt dH A =
2 2 2 1 1 1 2
(13) 368

R
c
out out , c out , c
c
in in , c in , c
c
out
Q H F H F dt dH + = (14) 369
370
Page 17 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
17
Eq. (12) is the mass balance for the reactor for component i (there are i = 1, NC equations, 371
where NC is the total number of components), Eqs. (13) (14) represent the energy balances 372
for the reactor and jacket, respectively. Assuming that the volume and density are constant for 373
both reactor and jacket,
2 1
F F F = = and
out , c in , c c
F F F = = . The steady state process is 374
obtained by setting the right hand side of Eqs. (12) (14) equal to zero. 375
376
- Constitutive constraints: 377
{ } NC , i C T k R
i , i
1 0
2
= = (15) 378
2 1 0 , j T H
j j
= | = (16a) 379
out , in j T H
j c
c
j
= | = 0 (16b) 380
381
Eqs. (15) (16) represent the phenomena models for the reaction rate and enthalpies (reactor 382
and jacket), respectively. The reactor temperature, T is assumed to be equal to the reactor 383
effluent stream temperature, i.e.,
2
T T = . 384
385
- Conditional (process-control specifications) constraints: 386
Sizing equations 387
t = F V 0 (17) 388
( ) V . V V
R
1 0 30 + > (18a) 389
( ) V . V V
R
1 0 3 + s (18b) 390
( )

=
>
NC
i
*
i i
opt
P x P P
1
(19) 391
Page 18 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
18


< <
i
b
i i
i
m
i i
T x K T T x ) ( (20) 392
Controller structure selection (Eq. (6)) 393
Eqs. (17) (19) are the sizing equations for a single reactor. Eq. (17) represents the 394
reactor volume as a function of the flow residence time. Eqs. (18a) (18b) represent the real 395
reactor volume,
R
V by summing the reaction volume, V with the head space, where the 396
headspace is calculated from 10% of the reaction volume. The acceptable value of
R
V for a 397
jacketed reactor is 30 3 s s
R
V m
2
(as defined in Table 6.2 of Sinnott (2005) as a relation 398
between capacity and cost for estimation of purchased equipment costs). The reactor optimal 399
pressure is calculated by analyzing the vapour pressure for all components at the optimal 400
operating temperature using Eq. (19). The optimal pressure
opt
P that is greater than the 401
operating pressure P is selected in order to have all components in the liquid phase. The 402
allowable operating temperature is calculated using Eq. (20) where,
i
x is the mole fraction of 403
component i, and
m
i
T and
b
i
T are the melting and boiling points, respectively, of component i. 404
The initial conditions of the process are given in Table 1. 405
406
3.1.2 Decomposition-based solution strategy 407
408
The summary of the decomposition-based solution strategy for this problem is tabulated in 409
Table 2. It can be seen that the constraints in the problem are decomposed into four sub- 410
problems which correspond to the four hierarchical stages. In this way, the solution of the 411
decomposed set of sub-problems is equal to that of the original problem. 412
413
Page 19 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
19
Stage 1: Pre-analysis 414
415
a. Variables analysis 416
All design/manipulated and process/controlled variables are tabulated in Table 3. From 417
Table 3, the important design/manipulated and process/controlled variables are identified and 418
tabulated in Table 4. Design/manipulated variables ] , [
c m
F V = u are selected since they are 419
unknown variables and their values are directly related to the capital and operating costs. 420
Process/controlled variables ] , , , [
2 , 2 , 2 , EG W EO m
C C C T = x , on the other hand, are the important 421
intensive variables that need to be monitored and controlled. 422
423
b. Operational window identification 424
Operational window is identified based on reactor volume for
m
u and operating 425
temperature constraints for
m
x . For a single reactor, its volume should satisfy the sizing and 426
costing constraints as defined in Eqs. (18a) (18b). The temperature range is defined between 427
the minimum melting point and maximum boiling point of components, Eq. (20). Therefore, 428
the operational window (feasible solutions) within which the optimal solution is likely to exist, 429
is given by 30 ) ( 3
3
s s m V and 562 ) ( 161 s s K T . 430
431
c. Design-control target identification 432
The AR is drawn from the feed points using Eqs. (21a) (21d), which are derived from 433
Eq. (15). Detailed derivation can be obtained from the authors. 434
Page 20 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
20

1 1 1 2
1
0
0
.
C
C
.
C
C
C
C
W
W
W
W
W
EG

|
|
.
|

\
|

|
|
.
|

\
|
= (21a) 435

2 . 1 2 . 2
1 1 . 2
0
0

|
|
.
|

\
|
(
(

|
|
.
|

\
|
|
|
.
|

\
|
=
W
W
W
W
W
EG
W
DEG
C
C
C
C
C
C
C
C
(21b) 436

(
(

|
|
.
|

\
|
|
|
.
|

\
|
= 1 2 2
0
W
W
W
DEG
W
TEG
C
C
C
C
.
C
C
(21c) 437

(
(

|
|
.
|

\
|
(

|
|
.
|

\
|
+
|
|
.
|

\
|
+ +
|
|
.
|

\
|

|
|
.
|

\
|
= 1 2 2 1 2 1
0 0 0
W
W
W
DEG
W
EG
W
W
W
EO
W
EO
C
C
C
C
.
C
C
.
C
C
C
C
C
C
(21d) 438
439
Solving Eqs. (29a) (29d) for specified values of
W
C with
0
W
C = 1.00 kmol/m
3
and
0
EO
C = 440
1.00 kmol/m
3
, values for
EG
C ,
DEG
C ,
TEG
C and
EO
C are calculated. Then, the AR is created 441
by plotting the concentration of
EG
C with respect to concentration of
W
C as shown in Fig. 5. 442
The location of the maximum point in the AR (Point A) is selected as the reactor design target. 443
It can easily be seen from Fig. 5 that a maximum of 0.1667 kmol/m
3
of
EG
C can be achieved 444
using a CSTR with effluent of 0.59 kmol/m
3
of
W
C . The calculation is repeated for different 445
ratios of initial concentration of EO and W of 1:2, 1:10, and 1:20. It was found that by 446
increasing ratio of
W
C in the feed, concentration of
EG
C is also increasing. This is because by 447
adding more
W
C , the side reactions are suppressed and make the main reaction more active, 448
thus more
EG
C is produced. However, the normalized value of
EG
C with respect to
0
W
C is still 449
the same as shown in Fig. 5 for all ratios. Besides, it was found that there is an operation 450
Page 21 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
21
constraint of
W
C for all ratios as shown in Fig. 5. For ratio of 1:1, the range of operation with 451
respect to
W
C was 0.54
W
C (kmol/m
3
) 1.0. When
W
C 0.54,
EO
C was all exhausted, 452
thereby, turning off the operation. For other ratios, the operation ranges of
W
C were 0.72 453
W
C (kmol/m
3
) 1.0 for ratio 1:2, 0.92
W
C (kmol/m
3
) 1.0 for ratio 1:10, and 0.96 454
W
C (kmol/m
3
) 1.0 for ratio 1:20. For ratio higher than 1:1, the maximum point (point A) was 455
located outside the operation range (see Fig. 5). The initial design of the reactor is made at the 456
maximum point of AR for
W EO
C C : of 1:1. 457
458
Stage 2: Design analysis 459
460
In this stage, the search space defined in Stage 1 is further reduced using design analysis. 461
The established target (Point A) in Fig. 6(a) is now matched by finding the acceptable values 462
(candidates) of the design/manipulated and process/controlled variables. If feasible values 463
cannot be obtained or the variable values are lying outside of the operational window, a new 464
target is selected and variables are recalculated until satisfactory matching is obtained. At 465
Point A, the allowable operating temperature is calculated using Eq. (30). The feasible 466
solution search space for temperature is now reduced to 406 ) ( 251 s s K T from 467
562 ) ( 161 s s K T . At this range, a feasible pressure range of 8 . 5 ) ( 0 . 1 s s atm P is predicted 468
using Eq. (19). 469
With this new range, the feasible solution range for the volume 470
(11.78<
R
V (m
3
)<1.082x10
8
) are calculated using Eq. (17). However, the upper limit of the 471
volume is more than what defined in Stage 1. Therefore, volume that is more than 30 m
3
and 472
its corresponding temperature are eliminated. For that reason, the search space for temperature 473
Page 22 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
22
is further reduced to 406 ) ( 394 s s K T . After Stage 2, the region of the feasible solutions is 474
now between 406 ) ( 394 s s K T and 89 . 26 ) ( 78 . 11
3
s s m V with feasible pressure of 475
8 . 5 ) ( 5 . 4 s s atm P . Within the feasible solutions for temperature 406 ) ( 394 s s K T , different 476
feasible candidates can be enumerated. For illustration purposes, only four feasible candidates 477
are considered with the scale of temperature decreasing by 4K. Candidates of 478
design/manipulated and process/controlled variables for stage 2 are tabulated in Table 5. In 479
principle, if the design is repeated for higher amounts of
W
C and fixed
EO
C , the pressure 480
would decrease but the size parameters would increase. 481
482
Stage 3: Control analysis 483
484
a. Sensitivity analysis 485
The search space is further reduced by considering feasibility of the process control. The 486
feasible candidates from stage 2 are evaluated in terms of controllability performance. The 487
process sensitivity is analyzed by calculating the derivative of the controlled variables with 488
respect to disturbances. In this case,
W
C and
1
T are potential sources of disturbance in the 489
reactor feed while
EG
C is the controlled variable which needs to be maintained at its optimal 490
value (set point). Fig. 6(b) shows plots of derivative of
EG
C with respect to
W
C and feed 491
temperature
1
T . It can be seen that the derivative values are smaller at the maximum AR point 492
(point A). Smaller value of derivative to disturbances means process sensitivity is lower, 493
hence process is more robust with respect to feed concentration and temperature variations. As 494
shown in Fig. 6(b), the value of ( )( ) 0 ~ =
W EG W EG
dC dT dT dC dC dC and 495
( )( ) 0
1 1
~ = dT dT dT dC dT dC
EG EG
, thus from a control perspective, concentration and 496
Page 23 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
23
temperature control are feasible. However, since value of
1
dT dC
EG
is smaller than 497
W EG
dC dC , thereby, temperature control should have better performance than concentration 498
control. By selecting temperature as a controlled variable rather than
EG
C at the highest AR 499
point, the controller performance should be the best. At this point, any big changes to the 500
temperature will result in smaller changes in the
EG
C (see Fig. 6(b)). Therefore, by 501
maintaining (controlled) temperature at its optimal value (set point) at the highest AR point, 502
EG
C can more easily be controlled. 503
504
b. Control structure selection 505
Next, the controller structure is selected by calculating the derivative value of the 506
manipulated variable u with respect to the controlled variable x . Since there is only one 507
actuator (
c
F ) available for controlled variable ( T ), therefore, T can be controlled by 508
manipulating
c
F . The derivative value of dT dF
c
is calculated and plotted in Fig. 6(c). It can 509
be seen that value of dT dF
c
at the maximum AR point is higher. The big value of dT dF
c
510
means the process gain is high (Russel et al., 2002). Suppose that disturbance move our 511
controlled variable away from its optimal set point. If the process has a high gain, then the 512
controlled variable is very sensitive to the changes in the manipulated variable and the 513
controller should make small action to correct the error. Conversely, if the process has a small 514
gain, then the controller needs to make large action to correct the same error (high control 515
cost). 516
0 ~
|
|
.
|

\
|
|
.
|

\
|
|
|
.
|

\
|
=
c
W
W
EG
c
EG
dF
dT
dT
dC
dC
dC
dF
dC
(22) 517
Page 24 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
24
From Eq. (22), since 0 ~
c EG
dF dC as shown in Fig. 6(c), it makes sense to control 518
reactor temperature by manipulating
c
F in order to maintain
EG
C at its optimal value (set- 519
point). Therefore, the concentration-to-temperature cascade control is proposed. In this 520
structure, the concentration
EG
C controller is the primary (master or outer loop) controller, 521
while the reactor temperature controller is the secondary (slave or inner loop) controller. This 522
is effective because the reactor temperature controller is less sensitive than concentration 523
controller (see Fig. 6(b)). An inner-loop disturbance, such as feed temperature, will be 524
sensed by the reactor temperature before it has a significant effect on the concentration 525
EG
C . This inner-loop controller then adjusts the manipulated variable before a substantial 526
effect on the primary output has occurred. With this control structure, the robust performance 527
of a controller in order to maintain desired product
EG
C at its optimal set point in the presence 528
of disturbance can be assured. Thus, the proposed controller structure is as follows: 529
530
Primary controlled variable :
EG
C 531
Secondary controlled variable : T 532
Manipulated variable :
c
F 533
Primary setpoint : 0.1667 kmol/m
3
534
Secondary set point : 406 K 535
536
The proposed control structure for an ethylene glycol process is shown in Fig. 7. 537
538
Stage 4: Final selection and verification 539
540
Page 25 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
25
a. Final selection: Verification of design 541
The multi-objective function Eq. (7) is calculated by summing up each term of the 542
objective function value using equal weights. This is given in Table 6.
s ,
P
1 1
corresponds to the 543
scaled value of the concentration of EG.
s ,
P
1 2
and
s ,
P
2 2
are the scaled value of
1
dT dC
EG
and 544
dT dF
c
, represent process sensitivity and process gain, respectively. Whereas,
s ,
P
1 3
and
s ,
P
2 3
545
are the scaled value of reactor volume and cooling water flow rate, respectively, which 546
represent capital and operating costs. Since all candidates in Table 6 are at the maximum point 547
of AR (point A), values for
s ,
P
1 1
,
s ,
P
1 2
and
s ,
P
2 2
are the same. It can be seen that, value of 548
J for Candidate 1 is higher than other candidates. Therefore, it is verified that Candidate 1 is 549
the optimal solution to integrated design and control of ethylene glycol reaction process which 550
satisfies the design, control and cost criteria. It should also be noted that a qualitative analysis 551
( J highest for point A) is sufficient for the purpose of controller structure selection. 552
553
b. Open loop dynamic simulation: Verification of controller performance 554
As explained in the section 2.4, when a reactor is designed corresponding to the 555
maximum point of the AR (point A), the controllability of the system is also best satisfied. 556
This is verified by selecting two but sub-optimal points in the AR (see Fig. 6(a)). From a 557
design point of view, they are not feasible since points B and C generate lower EG 558
concentrations. From control point of view, the derivative values of the desired product
EG
C 559
with respect to disturbances (
W
C and
1
T ) at Point A is smaller than those at points B and C, as 560
shown in Fig. 6(b). This in turn means that any changes in
W
C and
1
T will give smaller 561
changes in
EG
C at Point A compared to points B or C. 562
Page 26 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
26
In order to further verify the controllability aspects, a disturbance (+10% step change in 563
feed temperature
1
T ) moves reactor temperature T away from its set points (points A, B, C). 564
According to Fig. 6(b), any changes in the
1
T at points B and C will easily move the desired 565
product
EG
C away from its steady state value in a big scale and as a result, it will be more 566
difficult to maintain the
EG
C at these points than at Point A. 567
Fig. 8 shows the open-loop output response of T and
EG
C when +10% step changes in 568
feed temperature
1
T is applied at points A, B, and C. One observes that the effect of 569
disturbance to the
EG
C is negligible at Point A, whereas for points B and C are quite 570
significant (see Fig. 8(a)). This means that, process sensitivity at Point A is lower than other 571
points. As a result, Point A offers better robustness in maintaining its desired product 572
concentration
EG
C against disturbance. Therefore, it can be verified (albeit empirically) that, 573
designing a reactor at the maximum point of AR leads to a process with lower sensitivity with 574
respect to disturbance. 575
As a summary, the results demonstrate the potential use of the decomposition method in 576
solving a simple IPDC problems particularly its ability to reduce the dimension of the feasible 577
solutions and locate the optimal solution. It was confirmed that in each subsequent stage the 578
search space for temperature and volume are reduced until in the final stage only a small 579
number of the remaining feasible candidates are evaluated. It was also confirmed that 580
designing a reactor at the maximum point of the AR leads to a process with lower sensitivity 581
with respect to disturbance. All in all, this application demonstrates that the developed 582
methodology is viable and effective tool in solving IPDC problems for a single reactor system. 583
584
585
Page 27 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
27
3.2 Application to a single separator design 586
587
The application of the decomposition-based methodology is illustrated for the separation 588
system of an ethylene glycol process. We consider the following situation. The effluent stream 589
from the reactor in the previous case study is now fed to a distillation column where it is split 590
into two streams of specified purity - bottom product (stream B with mainly EG, DEG and 591
TEG) and distillate product (stream D containing 99.5% of unreacted W and 100% EO). The 592
objective is then to determine the design-control solution in which the multi-objective function 593
Eq. (7) is optimal. The process is operated at a nominal operating point as specified in Table 7. 594
A schematic of the process is depicted in Fig. 9. 595
596
3.2.1 Detailed formulation of the problem 597
The IPDC problem consists of a performance objective function (Eq. (7)) and a set of 598
constraints: process (dynamic and steady state), constitutive (thermodynamic states) and 599
conditional (process-control specifications). 600
601
- Process constraints: 602
We assume potential feeds on all of the stages and adopt the following set notation. The 603
number of stages in the column is assumed to be N inclusive of both the reboiler and 604
condenser, with stages numbered from the bottom. The set STAGES := {1, ..., N) will denote 605
the numbered stages and index, j subscripted to a quantity associated with stage, j. The set 606
COMP denotes the components in the column. The superscripts l and v refer to the quantities 607
associated with the liquid and vapor phases, respectively. 608
Total mass balance on each stage 609
Page 28 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
28

1 1 1 2
1
F L V L
dt
dM
+ = (23) 610

j j j j j
j
F L V L V
dt
dM
+ + =
+ 1 1
{ } N , \ STAGES j 1 e (24) 611

N N N N
N
F L D V V
dt
dM
+ =
1
(25) 612
where M
j
, L
j
, V
j
, F
j
are the holdup, liquid flowrate, vapor flowrate and feed rate on the jth 613
stage, respectively. 614
Component balance on each stage 615
For each component COMP i e , we have: 616

1 1 1 1 1 1 2 2
1
, i , i , i , i
, i
z F x L y V x L
dt
dM
+ = (26) 617

j , i j j , i j j , i j j , i j j , i j
j , i
z F x L y V x L y V
dt
dM
+ + =
+ + 1 1 1 1
{ } N , \ STAGES j 1 e (27) 618
N , i N N , i N N , i N , i N N , i N
N , i
z F x L Dx y V y V
dt
dM
+ =
1 1
(28) 619
where M
i,j
, z
i,j
, x
j,i
, y
i,j
represent the hold-up, feed, liquid and vapor composition of component 620
i on the jth stage, respectively. 621
Energy balance on each stage 622
In the following ( )
j j j j
T , y , x U , ( )
j j
l
j
T , x h and ( )
j j
v
j
T , y h define the stage holdup internal 623
energy and the specific heat content of liquid and vapor emanating from stage j. These are 624
functions of composition of the mixture and stage temperature. 625
r
f l v l
Q h F h L h V h L
dt
dU
+ + =
1 1 1 1 1 1 2 2
1
(29) 626
Page 29 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
29
f
j j
l
j j
v
j j
l
j j
v
j j
j
h F h L h V h L h V
dt
dU
+ + =
+ + 1 1 1 1
{ } N , \ STAGES j 1 e (30) 627
c
f
N N
l
N N
l
N
v
N N
v
N N
N
Q h F h L Dh h V h V
dt
dU
+ =
1 1
(31) 628
with
f
j
h representing the specific enthalpy of the feed stream to stage j and Q
r
and Q
c
are the 629
reboiler and condenser heat duties, respectively. 630
631
- Constitutive constraints: 632
For each stage STAGES j e 633
634

j i j i i
x y FD
, ,
= (32) 635

( ) 1 1 o +
o
=
jk , i j , i
j , i jk , i
j , i
x
x
y COMP i e (33) 636

k , j
j , i
jk , i
K
K
= o (34) 637
( )
j j i j , i
P , T K K = (35) 638
639
- Conditional constraints: 640
Product quality, 05 . 0 s
W
x (36) 641
Controller structure selection, Eq. (6) 642
643
3.2.2 Decomposition-based solution strategy 644
645
Page 30 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
30
The summary of the decomposition-based solution strategy is tabulated in Table 8. It can be 646
seen that the constraints in the problem are decomposed into four sub-problems which 647
correspond to the four hierarchical stages. In this way, the solution of the decomposed set of 648
sub-problems is equal to that of the original problem. 649
650
Stage 1: Pre-analysis 651
a. Variables analysis 652
All design and control variables involve in this process are tabulated in Table 9. From 653
Table 9, the important variables are identified and tabulated in Table 10. Design/manipulated 654
variables ] , , , , , , [
c r F m
Q Q D B RB RR N = u are selected since they are unknown variables and 655
have a potential to be manipulated variables except for
F
N . Beside that, values of
r
Q and
c
Q 656
are directly proportional to the operating cost. On the other hand, process/controlled variables 657
] , , , , , [
T EG W B EG W m
T y y T x x = x are important since they are potential candidates to be 658
controlled for the bottom and top composition. 659
660
b. Operational window identification 661
The operational window is identified based on bottom and top products purity. Since 662
desired product is recovered in the bottom, for that reason, its quality should be monitored and 663
controlled. On the other hand, since most of the unreacted reactants are recovered at the top, 664
its purity will not be monitored and controlled because it is going to be recycled back to the 665
reactor. In order to satisfy product quality, the bottom water composition
W
x should be less 666
than 0.05. 667
668
Page 31 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
31
c. Design-control target identification 669
The step-by-step algorithm for a simple distillation column proposed by Gani and Bek- 670
Pedersen (2000) are implemented here. The DF diagram for a W-EG (the key-components of 671
binary pair) system at P = 5.8 atm is drawn as shown in Fig. 10(a). DF is a measure of the 672
relative ease of separation. The larger the driving force, the easier the separation is. In this 673
graphical method, the target for the optimal process-controller design solution for distillation 674
is identified at the maximum point of the DF (point D). In Fig. 10(a) also, two other points 675
which are not at the maximum are identified as candidate alternative designs. From a process 676
design point of view, they are not optimal since at points E and F the value of driving force is 677
smaller hence separation at this point is more difficult. Therefore, from a design perspective 678
point D is the optimal solution for distillation (this claim will be tested/verified in stage 4). 679
680
Stage 2: Design analysis 681
682
The established targets (points D, E and F) in Fig. 10(a) are now matched by finding the 683
acceptable values of the design/manipulated variables (e.g. feed stage, reflux ratio, etc.). The 684
values of the design variables are determined graphically as shown in Fig. 11. Table 11 685
summarizes the results with respect to design/manipulated variables at three different design 686
alternatives. With the values of N ,
F
N , RR, product purity, and feed condition are specified, 687
the design of a distillation column can be verified through rigorous simulation. Results of the 688
steady-state simulation at different design alternatives are tabulated in Table 12. It can be 689
noted that design at the maximum point of DF (Point D) corresponds to the minimum with 690
respect to energy consumption than other points, as also confirmed by Gani and Bek-Pedersen 691
(2000). 692
Page 32 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
32
693
Stage 3: Control analysis 694
695
a. Sensitivity analysis 696
In this stage, the controllability of the selected design candidates (column designs D, E, 697
F) is analyzed. In this respect, two criteria are considered: (a) process sensitivity with respect 698
to disturbances, which should be low and (b) sensitivity of manipulated variable with respect 699
to controlled variable, which should be high. The process sensitivity is analyzed by calculating 700
the derivative of the controlled variables with respect to disturbances. In this case, bottom 701
composition of W (
W
x ) is the most important variable that need to be monitored and 702
controlled whereas, feed composition of W (
W
z ) and feed temperature are potential sources of 703
disturbance. Fig. 10(b) shows plots of derivative of DF with respect to composition of W and 704
temperature. It can be seen that derivative values are smaller at the maximum point of DF. 705
Hence, at this point the process is more robust in maintaining its product purity against feed 706
composition and temperature variations. As shown in Fig. 10(b), the value of 707
( )( ) 0 ~ =
i B B i i i
dx dT dT dFD dx dFD and ( )( ) 0 ~ = dT dT dT dFD dT dFD
B B i i
, thus from a 708
control perspective, composition and temperature control are feasible. However, since value of 709
dT dFD
i
is smaller than
i i
dx dFD , thereby, temperature control has better performance than 710
composition control. By selecting bottom temperature as a controlled variable rather than
W
x 711
at the highest DF point, the controller performance will be the best. At this point, any big 712
changes to the bottom temperature will result in smaller changes in the
W
x (see Fig. 10(b)). 713
Therefore, by maintaining (controlled) bottom temperature at its optimal value (set point) at 714
the highest DF point,
W
x can more easily be controlled. 715
Page 33 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
33
716
b. Control structure selection 717
Next the controller structure is selected by calculating the derivative value of the 718
manipulated variable u with respect to the controlled variable x . Since column bottom level, 719
b
h and condenser level,
d
h are controlled by manipulating distillate flow rate, D and bottom 720
flow rate, B , respectively, there are two available manipulated variables (vapour boilup, V 721
and reflux rate, L) to be paired with bottom and distillate composition of W (
W
x and
W
y ). 722
Since V is directly related to
W
x and L is directly related to
W
y , it is possible to pair
W
x V 723
and
W
y L . For bottom composition controller, the derivative value of
B
dT dV is calculated 724
and plotted in Fig. 10(c). It can be seen that value of
B
dT dV at the maximum DF point is 725
slightly higher at column design D and other designs. Therefore, control action at column 726
design D is better than in column designs E and F. Since 0 ~ dV dx
W
as shown in Fig. 10(c) 727
and from Eq. (37), it makes sense to control bottom temperature by manipulating V , in order 728
to maintain
W
x at its optimal value (set point). 729
0 ~ |
.
|

\
|
|
|
.
|

\
|
|
|
.
|

\
|
=
dV
dT
dT
dFD
dFD
dx
dV
dx
B
B
i
i
W W
(37) 730
Therefore, the composition-to-temperature cascade control is proposed. In this structure, the 731
composition
W
x controller is the primary controller, while the bottom column temperature 732
controller is the secondary controller. With this control structure, the robust performance of a 733
controller in order to maintain desired bottom product purity at its optimal set point in the 734
presence of disturbance can be assured. Similarly the control structure for the distillate 735
composition control is also identified. The proposed control structure for an ethylene glycol 736
Page 34 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
34
separation system is shown in Fig. 12 and the proposed controller structures are shown in 737
Table 13. 738
739
Stage 4: Final selection and verification 740
741
a. Final selection: Verification of design 742
The multi-objective function, Eq, (7) is calculated by summing up each term of the 743
objective function value using equal weights. This is given in Table 14.
s ,
P
1 1
corresponds to 744
the scaled value of DF.
s ,
P
1 2
and
s ,
P
2 2
are the scaled value of dT dFD
i
and
B
dT dV , 745
represent process sensitivity and process gain, respectively. Whereas,
s ,
P
1 3
and
s ,
P
2 3
are the 746
scaled value of
r
Q and
c
Q , respectively, which represent the operating cost. It can be seen 747
that, the value of J for column D is higher than other designs. Therefore, it is verified that 748
column design D is the optimal solution to integrated design and control of ethylene glycol 749
separation process which satisfies design, control and cost criteria. Noted also that a 750
qualitative analysis ( J is highest at point D) is sufficient for the purposed of controller 751
structure selection. 752
753
b. Open loop dynamic simulation: Verification of controller performance 754
In order to further verify the controllability aspects, a disturbances (5K step changes in 755
feed temperature) moves
B
T away from its set points (points D, E, F). According to Fig. 10(b), 756
any changes in the
B
T at points E and F will easily move
W
x away from its set point in a big 757
scale and consequently, it will be more difficult to maintain the
W
x at these points than at 758
Point D. Fig. 13 shows open-loop dynamic responses of bottom
B
T ,
W
x and
EG
x for all 759
Page 35 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
35
column designs when 5K step changes in feed temperature is applied. One can clearly see 760
that the response of
W
x manages to maintain to its set point at column design A in the 761
presence of disturbance whereas for column design F is most sensitive (see Fig. 13). This 762
means that, process sensitivity with respect to disturbance at column design D is lower than 763
other designs. As a result, column design D offers better robustness in maintaining its desired 764
composition
W
x against disturbance. Therefore, it can be verified that, designing a distillation 765
column at the maximum point of DF leads to a process with better robustness with respect to 766
disturbance. 767
As a summary, the results reveal the potential use of the decomposition and DF methods 768
in solving IPDC problem of a single distillation column. It was confirmed that designing a 769
distillation column at the maximum point of the DF leads to a process with lower energy 770
required and more robust in maintaining its product purity than any other points. In general, 771
this application has shown that the proposed methodology is viable and provides valuable 772
insights to the solution of the IPDC problem for a single separator system. 773
774
3.3 Application to a reactor-separator-recycle design 775
776
This section demonstrates the use of decomposition methodology in solving integrated 777
design and control of a RSR system as illustrated in Fig. 14. We consider the following 778
situation. The effluent stream from the CSTR (reactor case study) is fed to the distillation 779
column (distillation case study) where it is split into two streams of specified purity. The 780
reactant-rich stream Y is recycled back to the reactor, to increase the process economy when 781
the conversion in the reactor is low. The objective here is to solve sub-problems 1 2 of the 782
general IPDC problem that is, to only identify a feasible window of operation within which 783
Page 36 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
36
the so-called snowball effect will not appear and the reaction product composition will he 784
high. 785
As Luyben and Floudas (1994) have shown, the RSR system shown in Fig. 14 exhibits 786
the snowball effect when a small disturbance in the fresh feed rate causes a very large 787
disturbance to the recycle flow rate. However, according to Kiss et al. (2007), the control 788
problems created by snowball effects can be avoided through reactor (volume) design. 789
Consequently, instead of managing the snowball effect using some control strategy, it is 790
possible to avoid it through an appropriate reactor design. Therefore, it is important to define 791
the feasible range of operation with respect to manipulated (design) and controlled (process) 792
variables where the snowball effect can be avoided. 793
For sub-problems (stages 1 2) we only need the process model and Eq. (4), i.e., the set of 794
conditional constraints. Eq. (4) is derived for the RSR system under the following 795
assumptions: 796
797
A
0
. Steady-state condition using a CSTR, 798
A
1
. Complete recovery of EO recycled back to the reactor ( 1 = o
S , Y
), 799
A
3
. No recycle of EG, DEG and TEG ( 0
, , ,
= = =
S Y S Y S Y
c o _ ), 800
A
4
. Equimolar feed flowrate of reactants (
F , W F , EO
F F = ), 801
A
5
. Isothermal reaction in CSTR. 802
803
Through manipulation of the mass balance equations, the following set of conditional 804
constraints are obtained in terms of dimensionless variables ,
S , Y
| , Da and variables
EO
m , 805
W
f . The detailed derivation for these equations can be obtained from the authors. 806
Page 37 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
37
807
( ) ( )( )
1 3 2 1 1
1 0 O | =
W EO S , Y
f m (38) 808
809
( )( ) ( ) ( )( ) ( ) | |
1 2 1 3 2 1 1 2
1 1 2 1 0 | O | =
W S , Y EO S , Y
f . m (39) 810
811
( ) ( ) ( ) ( ) | |
2 1 3 2 3 2 1 2 3
1 2 2 2 0 O = . . m
EO
(40) 812
with 813

( ) ( ) ( ) | |
2
3 2 1
1 + | +
= O
S , Y W EO
f m
Da
814
815
In this IPDC problem, we want to identify the feasible range of operation in terms of 816
dimensionless design variable (
F EO F EO
F VC k Da
,
2
, 1
= ) and within which the highest 817
composition of product EG ) (
,S EG
z can be obtained and the snowball effect can be eliminated. 818
Eqs. (38) (40) can be written in compact form as, 819
820
0 = f | | u , (41) 821
where 822
| |
W EO S Y
f m Da , , ,
,
| = u 823
Vector u represents the set of design variables. Once the vector u has been determined, Eq. 824
(41) is solved for and using Eqs. (42) (47) (representing the steady state process model) 825
the values of the important process variables are obtained. 826
( )
3 2
1
1
+
|
|
.
|

\
|
|

+ =
S , Y
W
EO
f
m S (42) 827
Page 38 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
38

( )
( ) ( ) ( )
3 2 1
3 2 1
1 + | +
+ +
=
S , Y W EO
EO
S , EO
f m
m
z (43) 828

( ) ( )
( ) ( ) ( )
3 2 1
1
1
1
+ | +
|
=
S , Y W EO
S , Y W
S , W
f m
f
z (44) 829

( ) ( ) ( )
3 2 1
2 1
1 + | +

=
S , Y W EO
S , EG
f m
z (45) 830

( ) ( ) ( )
3 2 1
3 2
1 + | +

=
S , Y W EO
S , DEG
f m
z (46) 831

( ) ( ) ( )
3 2 1
3
1 + | +

=
S , Y W EO
S , TEG
f m
z (47) 832
833
The responses of the product composition of EG (
S EG
z
,
) and the reactor effluent flow 834
rate, S are plotted in Fig. 15 (a b). In Fig. 15(a), it can be observed that the maximum value 835
of
S EG
z
,
is within the range of 10 3 < < Da . But, when 5 < Da , the S increases significantly 836
indicating a possible snowball effect, as shown in Fig. 15(b). In order to avoid the snowball 837
effect, the system should be operated at a higher value of Da (for example 4 > Da ) (see Fig. 838
15(b)). Therefore, for the maximum values for the production of EG and also to eliminate the 839
snowball effect, the feasible range for Da is identified within 10 5 s s Da and 5 . 5 = Da . 840
Once the feasible range of Da has been established, design-control targets identified earlier at 841
the maximum points of AR and DF, for reactor and separator designs, respectively are used to 842
determine the remaining design variables and controller structure design. 843
As a summary, the results demonstrate the potential use of the decomposition-based 844
method in solving IPDC problem of RSR systems. It was confirmed that by applying the 845
developed methodology, the nonlinearity such as the snowball effect in this process is avoided 846
while maintaining higher productivity and controllable process. All in all, this application 847
Page 39 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
39
demonstrates that the developed methodology has advantages that it is systematic, makes used 848
of thermodynamic-process knowledge and provides valuable insights to the solution of IPDC 849
problems for RSR systems. 850
851
852
4. Future perspectives 853
854
An important issue with model-based process design and controller design is the effect of 855
uncertainties such as those related to the operating conditions (i.e. feed flowrates and 856
concentrations, catalyst activity etc.), model parameters (i.e. heat transfer coefficients, kinetic 857
constants, etc.) and the costs or prices of the materials. It is possible that an optimal design 858
under nominal conditions would show poor operability performances under uncertainties. 859
IPDC under uncertainty has been discussed by others, which have shown that it is important to 860
develop an optimal process for the entire range of uncertainties to ensure robust operability 861
(Bansal et al., 2000; Malcom et al., 2007; Moon et al., 2009; Ricardez Sandoval et al., 2008). 862
However, adding the uncertainties significantly increases the complexity of these problems 863
and leads to massive optimization models (Sahinidis, 2004). Therefore, as future perspectives 864
the effect of uncertainties will be incorporated during the analysis (sub-problems 1 3) to 865
ensure robust operability of the optimal designed process. 866
867
868
5. Conclusions 869
870
Page 40 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
40
This paper presents a novel systematic model-based methodology for solving IPDC 871
problems in chemical processes. The main idea is to decompose the complexity of the IPDC 872
problem by following four hierarchical stages (sub-problems) (i) pre-analysis, (ii) design 873
analysis, (iii) controller design analysis, and (iv) final selection and verification, which are 874
relatively easier to solve. The developed methodology incorporates thermodynamic-process 875
insights to determine a priori, the optimal values of the process variables and then through 876
them, all other design and decision variables are obtained. The AR and DF concepts have been 877
used to define the design targets and then matching these targets through a decomposed search 878
technique. The application of this methodology has been illustrated with the help of three case 879
studies. In the first case study, an optimal solution was found with respect to design, control 880
and cost criteria of a single reactor system for EG production. In the second case study, the 881
design-control problem of a single separator system was addressed. Finally, an optimal 882
solution was identified with respect to design, control and cost of a reactor-separator-recycle 883
system in the third case study, where the designed system is able to produce higher 884
productivity without experiencing nonlinearity problem. In general, all results from three case 885
studies indicate the viability and effectiveness of the developed methodology. The 886
methodology has advantages that it is systematic, makes use of thermodynamic-process 887
knowledge and provides valuable insights to the solution of IPDC problem. 888
889
Acknowledgements 890
891
The financial support for this PhD project provided by the Malaysian Ministry of Higher 892
Education (MoHE) and Universiti Teknologi Malaysia (UTM) is gratefully acknowledged. 893
894
Page 41 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
41
Nomenclature 895
B Bottom flowrate 896
0
EO
C ,
F , EO
C Feed concentration of Ethylene Oxide 897
0
W
C Feed concentration of Water 898
DEG
C Concentration of Diethylene Glycol 899
EG
C Concentration of Ethylene Glycol 900
EO
C Concentration of Ethylene Oxide 901
TEG
C Concentration of Triethylene Glycol 902
W
C Concentration of Water 903
pc p
C C , Heat capacity for component and coolant 904
D Distillate flowrate 905
Da Damkhler number 906
d Set of disturbance variables 907
i
FD Driving force 908
c
F Coolant flowrate 909
j
F Feed flowrate on the jth stage 910
F , EO
F Ethylene Oxide feed flowrate 911
F , W
F Water feed flowrate 912
W
f Dimensionless Water feed flowrate 913
R
H A Heat of reaction 914
j
H Reactor enthalpy of stream j 915
Page 42 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
42
c
j
H Jacket enthalpy of stream j 916
l
j
h Specific heat content of liquid emanating from stage j 917
v
j
h Specific heat content of vapor emanating from stage j 918
J Objective function 919
i
k Reaction kinetic of reaction i 920
j , i
K Equilibrium constant of component i on the jth stage 921
j
L Liquid flowrate on the jth stage 922
j , i
M Holdup of component i on the jth stage 923
j
M Holdup on the jth stage 924
EO
m Dimensionless Ethylene Oxide mixer flowrate 925
N No. of stage 926
F
N Feed stage 927
opt
P Optimal pressure 928
*
i
P Partial pressure of component i 929
P Pressure 930
j
P
, 1
Design objective term 931
j
P
, 2
Control objective term 932
j
P
, 3
Economic objective term 933
c
Q Condenser duty 934
r
Q Reboiler duty 935
R
Q Heat transfer between the jacket and the reactor 936
Page 43 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
43
i
r Reaction rate of component i 937
i
R Net reaction rate of reaction i 938
min
, RB RB Real reboil ratio, minimum reboil ratio 939
min
, RR RR Real reflux ratio, minimum reflux ratio 940
S Reactor effluent flowrate 941
t Time 942
j
T Temperature of stream j 943
c co
T T , Coolant temperature (input and output) 944
b
i
m
i
T T , Melting and boiling point of component i 945
i
U Holdup internal energy on the jth stage 946
u Set of design/manipulated variables 947
v Set of chemical system variables 948
V Reactor volume 949
j
V Vapor flowrate on the jth stage 950
R
V Real reactor volume 951
j
w Weight factor assigned to each objective term 952
x Set of process/controlled variables 953
j , i
x Liquid mole fraction for component i on the jth stage 954
Y Binary decision variables 955
j , i
y Vapor mole fraction for component i on the jth stage 956
j , i
z Feed composition for component i on the jth stage 957
Greek symbols 958
Page 44 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
44
jk , i
o Relative volatility of component i 959
S , Y
o Recovery of Ethylene oxide at stream Y w.r.t. stream S 960
S Y ,
| Recovery of Water at stream Y w.r.t. stream S 961
S , Y
_ Recovery of Ethylene Glycol at stream Y w.r.t. stream S 962
S , Y
o Recovery of Diethylene Glycol at stream Y w.r.t. stream S 963
S , Y
c Recovery of Triethylene Glycol at stream Y w.r.t. stream S 964
Set of constitutive variables 965
i
Dimensionless extent of reaction of component i 966
c
, Density for component and coolant 967
R
t Reaction residence time 968
969
970
References 971
Allgor, R. J., & Barton, P. I. (1999). Mixed-integer dynamic optimization I: problem 972
formulation. Computers and Chemical Engineering, 23 (4-5), 567-584. 973
Avraam, M., Shah, N., & Pantelides, C. C. (1999). A decomposition algorithm for the 974
optimization of hybrid dynamic processes. Computers and Chemical Engineering, 975
23(Suppl), S451-S454. 976
Banga, J. R., Moles, C. G., & Alonso, A. A. (2003). Global optimization of bioprocess using 977
stochastic and hybrid methods. In: C. A. Floudas & P. M. Pardalos (Eds.), Frontier in 978
global optimization, nonconvex optimization and its applications, Kluver Academic 979
Publishers, 45-70. 980
Page 45 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
45
Bansal, V., Perkins, J. D., Pistikopoulos, E. N., Ross, R., & Van Schijndel, J. M. G. (2000). 981
Simultaneous design and control optimization under uncertainty. Computers and 982
Chemical Engineering, 24, 261-266. 983
Bansal, V., Sakizlis, V., Ross, R., Perkins, J. D., & Pistikopoulos, E. N. (2003). New 984
algorithms for mixed-integer dynamic optimization. Computers and Chemical 985
Engineering, 27, 647-668. 986
Chachuat, B., Singer, A. B., and Barton, P. I. (2005). Global mixed-integer dynamic 987
optimization. AIChE Journal, 51(8), 2235-2253. 988
Cohen, G. H., and Coon, G. A. (1953). Theoretical considerations of retarded control. Trans. 989
ASME, 75, 827-834. 990
Esposito, W. R., Floudas, C. A. (2000). Deterministic global optimization in nonlinear optimal 991
control problems. Journal of Global Optimization, 17, 245-255. 992
Exler, O., Antelo, L. T., Egea, J. A., Alonso, A. A., & Banga, J. R. (2008). A Tabu search- 993
based algorithm for mixed-integer nonlinear problems and its application to integrated 994
process and control system design. Computers and Chemical Engineering, 32, 1877- 995
1891. 996
Flores-Tlacuahuac, A., and Biegler, L. t. (2007). Simulatenous mixed-integer dynamic 997
optimization for integrated design and control. Computers and Chemical Engineering, 998
31, 588-600. 999
Gani, R., & Bek-Pedersen, E. (2000). A simple new algorithm for distillation column design. 1000
AIChE Journal, 46(6), 1271-1274. 1001
Glasser, D., Hildebrandt, D., & Crowe, C. (1987). A geometric approach to steady flow 1002
reactors: The attainable region and optimization in concentration space. Industrial and 1003
Engineering Chemistry Research, 26(9), 1803-1810. 1004
Page 46 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
46
Glasser, D., Hildebrandt, D., & Crowe, C. (1990). Geometry of the attainable region generated 1005
by reaction and mixing: With and without constraints. Industrial and Engineering 1006
Chemistry Research, 29(1), 49-58. 1007
Hamid, M. K. A., & Gani, R. (2008). A model-based methodology for simultaneous process 1008
design and control for chemical processes. In: Proceedings of the FOCAPO 2008, 1009
Massachusetts, USA, 205-208. 1010
Karunanithi, A. P. T., Achenie, L. E. K., and Gani, R (2005). A new decomposition-based 1011
computer-aided molecular/mixture design methodology for the design of optimal solvents 1012
and solvent mixtures, Industrial and Engineering Chemistry Research, 44, 4785-4797. 1013
Kiss, A. A., Bildea, C. S., and Domian, A. C. (2007). Design and control of recycle systems 1014
by non-linear analysis, Computers and Chemical Engineering, 31, 601-611. 1015
Kookos, I. K., & Perkins, J. D. (2001). An algorithm for simultaneous process design and 1016
control. Industrial and Engineering Chemistry Research, 40, 4079-4088. 1017
Luyben, W. L. (2004). The need for simultaneous design education, in: Seferlis, P. and 1018
Georgiadis, M. C. (Eds.). The integration of process design and control. Amsterdam: 1019
Elsevier B. V., 10-41. 1020
Luyben, M. L., & Floudas, C. A. (1994). Analyzing the interaction of design and control 2. 1021
reactor-separator-recycle system. Computers and Chemical Engineering, 18(10), 971- 1022
993. 1023
Malcom, A., Polam, J., Zhang, L., Ogunnaike, B. A., & Linninger, A. A. (2007). Integrating 1024
system design and control using dynamic flexibility analysis. AIChE Journal, 53(8), 1025
2048-2061. 1026
Page 47 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
47
Meeuse, F. M., & Grievink, J. (2004). Thermodynamic controllability assessment in process 1027
synthesis, in: Seferlis, P. and Georgiadis, M. C. (Eds.). The integration of process design 1028
and control. Amsterdam: Elsevier B. V., 146-167. 1029
Mohideen, M., Perkins, J. D., & Pistikopoulos, E. N. (1996). Optimal design of dynamic 1030
systems under uncertainty. AIChE Journal, 42(8), 2251-2272. 1031
Mohideen, M., Perkins, J. D., & Pistikopoulos, E. N. (1997). Towards an efficient numerical 1032
procedure for mixed integer optimal control. Computers and& Chemical Engineering, 1033
21(Suppl), S457-S462. 1034
Moles, C. G., Gutierrez, G., Alonso, A. A., & Banga, J. R. (2003). Integrated process design 1035
and control via global optimization: A wastewater treatment plant case study. Chemical 1036
Engineering Research and Design, 81, 507-517. 1037
Moon, J., Kim, S., Ruiz, G. J., and Linninger, A. A. (2009). Integrated design and control 1038
under uncertainty algorithms and applications. In: M. M. El-Hawagi & A. A. Linninger, 1039
Design for energy and the environment, CRC Press, 659-668. 1040
Parker, W. A., & Prados, J. W. (1964). Analog computer design of an ethylene glycol system. 1041
Chemical Engineering Progress, 60(6), 74-78. 1042
Patel, J., Uygun, K., & Huang, Y. (2008). A path constrained method for integration of 1043
process design and control. Computers and Chemical Engineering, 32, 1373-1384. 1044
Ricardez Sandoval, L. A., Budman, H. M., & Douglas, P. L. (2008). Simultaneous design and 1045
control of process under uncertainty. Journal of Process Control, 18, 735-752. 1046
Russel, B. M., Henriksen, J. P., Jrgensen, S. B., & Gani, R. (2002). Integration of design and 1047
control through model analysis. Computers and Chemical Engineering, 26, 213-225. 1048
Sahinidis, N. V. (2004). Optimization under uncertainty: state-of-the-art and opportunities. 1049
Computers and Chemical Engineering, 28, 971-983. 1050
Page 48 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
48
Sakizlis, V., Perkins. J. D., & Pistikopoulos, E. N. (2004). Recent advances in optimization- 1051
based simultaneous process design and control design. Computers and Chemical 1052
Engineering, 28, 2069-2086. 1053
Schweiger, C. A., & Floudas, C. A. (1997). Interaction of design and control: optimization 1054
with dynamic models, in: W. Hager & P. Pardalos (Eds.), Optimal Control: Theory, 1055
Algorithms and Applications, Kluver Academic Publishers, Gainesville, USA, 388-435. 1056
Seferlis, P., & Georgiadis, M. C. (2004). The integration of process design and control. 1057
Amsterdam: Elsevier B. V. 1058
Sendin, O. H., Moles, C. G., Alonso, A. A., & Banga, J. R. (2004). Multiobjective integrated 1059
design and control using stochastic global optimization methods. In: P. Seferlis & M. 1060
Georgiadis (Eds.), The integration of process design and control. Amsterdam: Elsevier B. 1061
V., 555-581. 1062
Sharif, M., Shah, N., & Pantelides, C. C. (1998). On the design of multicomponent batch 1063
distillation columns. Computers and Chemical Engineering, 22 (Suppl), S69-S76. 1064
Sinnott, R. K. (2005). Chemical Enginering, Volume 6, Fourth edition, Chemical Engineering 1065
Design, Elsevier Butterworth-Heinemann. 1066
1067
1068
1069
1070
1071
1072
1073
1074
Page 49 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
49
List of Figures 1075
1076
Fig. 1. Decomposition method for IPDC problem (Hamid and Gani, 2008). 1077
1078
Fig. 2. The number of feasible solution is reduced to satisfy constraints at every sub- 1079
problems. 1080
Page 50 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
50
1081
Fig. 3. Determination of optimal solution of design-control for a reactor using AR diagram 1082
(left) and a separator using DF diagram (right). 1083
1084
Fig. 4. CSTR for an ethylene glycol production. 1085
Page 51 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
51
1086
Fig. 5. Normalized plot of the desired product concentration
EG
C and
EO
C with respect to 1087
W
C for different
W EO
C C : . 1088
Page 52 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
52
1089
Fig. 6. (a) AR diagram for the desired product concentration
EG
C with respect to
W
C for 1090
W EO
C C : of 1:1, (b) Corresponding derivatives of
EG
C with respect to
W
C and T , (c) 1091
Corresponding derivative of
c
F with respect to T and derivative of
EG
C with respect to
c
F . 1092
Page 53 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
53
1093
Fig. 7. Proposed reactor control structure for an ethylene glycol process. 1094
Page 54 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
54
1095
1096
Fig. 8. Dynamic open-loop responses of: (a) desired product concentration
EG
C , (b) reactor 1097
temperature, T to +10% step change in feed temperature,
1
T for different alternative reactor 1098
designs. 1099
Page 55 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
55
1100
Fig. 9. Distillation column for an ethylene glycol process. 1101
Page 56 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
56
1102
Fig. 10. (a) DF diagram for the separation of water-ethylene glycol by distillation, (b) 1103
Corresponding derivatives of the DF with respect to composition and temperature, (c) 1104
Corresponding derivative of vapour boilup, V with respect to temperature and derivative of 1105
composition of water with respect to vapour boilup, V . 1106
Page 57 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
57
1107
1108
1109
Fig. 11. Driving force diagram with illustration of the distillation design parameters at (a) 1110
point D; (b) point E; and (c) point F. 1111
Page 58 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
58
1112
Fig. 12. Proposed distillation control structure for an ethylene glycol process. 1113
1114
Page 59 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
59
1115
1116
Fig. 13. Open-loop dynamic responses of bottom T
B
, x
W
, and x
EG
when +5K step change (top) 1117
and -5K step change (bottom) in feed temperature (406K) are applied for all columns. 1118
Page 60 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
60
1119
Fig. 14. RSR flowsheet of an ethylene glycol process. 1120
Page 61 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
61
1121
Fig. 15. Product composition of EG (a) and reactor outlet flowrate S (b) as a function of Da 1122
number. 1123
Page 62 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
62
Table 1. Initial feed conditions of the EG reaction process. 1124
Variable Value Description
F 1,000 m
3
/h Feed flow rate
1 , EO
C 1 kmol/m
3
Concentration of EO in feed
1 , W
C
1 kmol/m
3
Concentration of W in feed
1125
Table 2. Mathematical equations and decomposition-based solution for a single reactor design 1126
Case Study Mathematical
Equations
Decomposition Method Corresponding
Variables
Single Reactor Multi-objective
Function
Eq. (7)

Process Constraints
Eqs. (12), (13), (14)

Constitutive Constraints
Eqs. (15), (16a),
(16b)
Eqs. (21a), (21b),
(21c), (21d)

Conditional constraints

Sizing equations
Eqs. (17), (18a),
(18b), (19), (20)

Controller structure
selection
Eq. (6)
Stage 1: Pre-analysis

a. Variable analysis

b. Operational
window
Eqs. (18a), (18b),
(20)

c. Design-control
target
Eqs. (21a), (21b),
(21c), (21d)

Stage 2: Design analysis
Eqs. (17), (19),
(20), (21), (22),
(30)

Eqs. (12 14) in
steady state, (15),
(16a), (16b)

Stage 3: Controller
design analysis
Eqs. (12), (13), (14)
Eqs. (15), (16a),
(16b), Eq. (6)

Stage 4: Final selection
and verification
Eq. (7)






V
min
VV
max
,
T
min
TT
max




C
EG,2
, C
W,2
, C
DEG,w
,
C
TEG,2
, C
EO,2




T, V, P


F
c
, T
c
, T, C
EG,2
, C
W,2
,
C
DEG,2
, C
TEG,2
, C
EO,2





dC
EG
/dC
W
, dC
EG
/dT
1

dF
c
/dT, dC
EG
/dF
c




C
EG
, dC
EG
/dT
1
,
dF
c
/dT, V, F
c

1127
Page 63 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
63
Table 3. List of all design/manipulated and process/controlled variables of the EG reaction 1128
process. 1129
Design/manipulated variables ( u )
V ,
c
F , F
Process/controlled variables ( x )
2
T
, out , c
T
, 2 , EO
C ,
2 , W
C ,
2 , EG
C ,
2 , DEG
C ,
2 , TEG
C
1130
Table 4. List of important design/manipulated and process/controlled variables of the EG 1131
reaction process. 1132
Design/manipulated variables (
m
u ) V ,
c
F
Process/controlled variables (
m
x )
2
T
, 2 , EO
C ,
2 , W
C ,
2 , EG
C
1133
Table 5. Candidates of design/manipulated-process/controlled variables for Stage 2 of the EG 1134
reaction process. 1135
Candidates
W
C
(kmol/m
3
)
EG
C
(kmol/m
3
)
T
(K)
P
(atm)
V
(m
3
)
c
F
(m
3
/h)
1 0.59 0.1667 406 5.8 11.78 1388.31
2 0.59 0.1667 402 5.3 15.76 1388.22
3 0.59 0.1667 398 4.9 20.53 1388.26
4 0.59 0.1667 394 4.5 26.89 1388.34
1136
1137
1138
1139
Page 64 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
64
Table 6. Objective function calculation at different operating points of the EG reaction 1140
process. 1141
Candidate
s ,
P
1 1

s ,
P
1 2

s ,
P
2 2

s ,
P
1 3

s ,
P
2 3
J
1 1.00 1.00 1.00 0.44 0.96 6.34
2 1.00 1.00 1.00 0.59 0.96 5.74
3 1.00 1.00 1.00 0.76 0.98 5.33
4 1.00 1.00 1.00 1.00 1.00 5.00
1142
Table 7. Nominal operating point of the EG separation process. 1143
Variable Value Description
F 1,667 m
3
/h Feed flow rate
T 406 K Feed temperature
P 5.8 atm Feed pressure
EO
z 0.1856 Composition of EO in feed
W
z 0.4886 Composition of W in feed
EG
z 0.1358 Composition of EG in feed
DEG
z 0.0770 Composition of DEG in feed
TEG
z
0.1130 Composition of TEG in feed
N 10 No. of stages
1144
1145
1146
Page 65 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
65
Table 8. Mathematical equations and decomposition-based solution for a single separator 1147
design 1148
Case Study Mathematical
Equations
Decomposition Method Corresponding
Variables
Single
Separator
Multi-objective
Function
Eq. (7)

Process Constraints
Eqs. (23) - (31)

Constitutive Constraints
Eqs. (32) - (35),

Conditional constraints

Product quality
Eq. (36)

Controller structure
selection
Eq. (6)
Stage 1: Pre-analysis

a. Variable analysis

b. Operational
window
Eq. (36)

c. Design-control
target
Eqs. (32)

Stage 2: Design analysis
Step by step
algorithm for a
simple distillation
design (Gani &
Bek-Pedersen,
2000)

Eqs. (23 31) in
steady state
Eqs. (32 - 35)

Stage 3: Controller
design analysis
Eqs. (21) - (31)
Eq. (6)

Stage 4: Final selection
and verification
Eq. (7)






x
w
0.05



FD = y
W
x
W
, x
W





N, N
F
, RR, RB




B, D, Q
r
, Q
c
, x
EG
, x
W
,
y
EG
, y
W
, T
B
, T
T





dFD/dx
W
, dFD/dT
dV/dT
B
, dx
W
/dV



FD, dFD/dT, dV/dT
B
,
Q
r
, Q
c

1149
1150
1151
Page 66 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
66
Table 9. List of all design/manipulated and process/controlled variables of the EG separation 1152
process. 1153
Design/manipulated variables ( u )
F
N , RR , RB , B , D,
r
Q ,
c
Q ,
Process/controlled variables ( x )
EO
x ,
W
x ,
EG
x ,
DEG
x ,
TEG
x ,
EO
y ,
W
y ,
EG
y ,
DEG
y ,
TEG
y ,
B
T ,
T
T
1154
Table 10. List of important design/manipulated and process/controlled variables of the EG 1155
separation process. 1156
Design/manipulated variables (
m
u )
F
N , RR , RB , B , D,
r
Q ,
c
Q
Process/controlled variables (
m
x )
W
x ,
EG
x ,
B
T ,
W
y ,
EG
y ,
T
T
1157
Table 11. Value of design/manipulated variables of at different operating points of the EG 1158
separation process. 1159
Point N
F
N
min
RR
min
RB RR RB
D 10 8 0.2255 1.2146 0.2705 1.4575
E 10 7 0.4713 1.1103 0.5655 1.3324
F 10 9 0.1177 1.4358 0.1412 1.7230
1160
1161
1162
1163
1164
Page 67 of 67
A
c
c
e
p
t
e
d

M
a
n
u
s
c
r
i
p
t
CACE 5658 - Revised
67
Table 12. Steady-state simulation results at different design alternatives of the EG separation 1165
process. 1166
Point N
F
N RR
D
(kmol/h)
B
(kmol/h)
T
T

(K)
B
T

(K)
r
Q
x10
3
(kJ/h)
c
Q
x10
3
(kJ/h)
D 10 8 0.271 827.90 400.00 341.72 530.00 26.54 -47.35
E 10 7 0.566 827.90 400.00 341.72 534.88 28.12 -50.22
F 10 9 0.141 827.90 400.00 341.72 535.74 27.06 -48.14
1167
Table 13. Proposed control structure for an ethylene glycol separation system. 1168
Bottom Distillate
Primary controlled variable
W
x
W
y
Secondary controlled variable
B
T
D
T
Manipulated variable V L
Primary set point 0.03 0.77
Secondary set point 530 K 341 K
1169
Table 14. Objective function calculation at different design alternatives of the EG separation 1170
process. 1171
Point
s ,
P
1 1

s ,
P
1 2

s ,
P
2 2

s ,
P
1 3

s ,
P
2 3
J
D 1.00 0.01 1.00 0.94 0.94 113.75
E 0.91 0.49 0.19 1.00 1.00 5.14
F 0.93 1.00 0.09 0.96 0.96 4.10
1172

También podría gustarte