Computers chrm. Engng Vol. 22. Suppl.. pp.

S41 l-S416, 1998

8 1998 Elsevier Science Ltd. All rights reserved
Pergamon
PII: SOO98-1354(98)00082-9
Printed in &eat Britain
0098-1354/98 $19.00 + 0.00
Modelling Issues for Control Structure Selection in a Chemical
Process
Yi Cao Diane Rossiter* David W Edwards
Department of Chemical Engineering,
Loughborough University, Loughborough, LEl 1 3TU, UK
Jens Knechtel David Owens
Technische Universitat Berlin
Institut fur Prozess und Anlagentechnik,
D-10623 Berlin, Germany
Centre for Systems and Control Engineering
University of Exeter, Exeter, EX4 4QF, UK
Abstract
A highly-integrated nonlinear dynamic model of the HDA process has been used in the past for control structure
selection studies. The original model consists of many unit operations including three distillation columns. These
columns were originally modelled by Brognaux (1992) using the Shortcut Library models available within SpeedUp.
In this work, the I-IDA case study is revisited but the nonlinear model has been revised to include modified models for
the distillation columns and other unit operations. The modelling issues associated with the development of this more
rigorous HDA model are discussed in the context of its use for control structure selection. 0 1998 Elsevier Science
Ltd. All rights reserved.
Keywords: Modelling, Control structure selection, HDA Process
1 Introduction
The reliability of open loop control structure selection
techniques (Skogestad and Postlethwaite, 1996) in pre-
dicting the most effective structure in closed loop re-
lies strongly on the process model used. If the chosen
model does not represent accurately the inherent prop-
erties of the process then the predictions obtained using
the model are worthless. A question that has been asked
on many occasions is - how rigorous does a model need
to be for the results of the control structure selection
techniques to be a true reflection of how the real pro-
cess would behave? This question is considered here
for a particular case study - the HDA (hydrodealkyla-
tion of toluene) process (Douglas, 1988).
A highly-integrated nonlinear dynamic model of the
HDA process, constructed within SpeedUp (Aspen
Technology, Inc., 1996b), is used for the study. The
SpeedUp model consists of many unit operations in-
cluding three distillation columns. These columns
were originally modelled by Brognaux (1992) using
the Shortcut Library models available within SpeedUp.
Here, however, rigorous models for these columns are
developed.
The complete nonlinear model, including the short-
cut column models, has previously been used for stud-
ies. For example, Cao and Rossiter (1997) determ-
ined the best control structure for the HDA process
under manipulated variable constraints using the SIE
(Single-Input Effectiveness) measure. This work was
further enhanced by Cao et al. (1997b) and Cao et al.
(1997a). They introduced other measures for determin-
To whom all correspondence should be addressed
s411
ing the best choice of manipulated variables and meas-
urable outputs subject to certain constraints on the in-
puts and/or in the presence of disturbances. All these
control structure selection measures are used in open
loop and predict the process closed loop behaviour.
For the analyses, a linearised model of the process is
required. The linear&d model can be determined us-
ing the CD1 (Control Design Interface) of SpeedUp
which generates the A, B, C, D matrices of a state
space linear model for a particular steady state operat-
ing point. Once the linearised model is obtained the
measures for each candidate input (and/or output) of
the control structure are determined and ranked accord-
ingly. The predictions of Cao and Rossiter (1997) and
Cao et al. (3997b) and Cao et al. (1997a) were valid-
ated via closed loop simulation with the nonlinear HDA
SpeedUp model.
In this work, the HDA case study is revisited but the
nonlinear model has been revised to include modified
models for the distillation columns and other unit oper-
ations. The modelling issues associated with the devel-
opment of this more rigorous I-IDA model are discussed
in the context of its use for control structure selection.
The structure of the paper is as follows: after a brief
description of the HDA process in section 2, the model
upgrading for the reactor and distillation columns are
explained in sections 3 and 4. In section 5 model valid-
ation is discussed. Finally, in section 6, the effect of the
model upgrading on the outcome of the control struc-
ture selection is discussed as well as other possibilities
for improving the model.
S412 European Symposium on Computer Aided Process Engineering--X
2 HDA Process
2.1 Flowsheet
The process of hydrodealkylation of toluene (HDA) to
produce benzene involves two reactions: the conversion
of toluene to benzene (1) and the equilibrium between
benzene and diphenyl(2).
Toluene + Hz -+ Benzene + Methane
(1)
2 Benzene G= Diphenyl + Hz
(2)
The chosen structure of the HDA plant is shown in
Figure 1: HDA plant
Figure 1 (Brognaux, 1992). The process is now de-
scribed. Available raw materials are pure toluene at am-
bient temperature and 3.4 atm (50 psi), and a gas feed
including 95% hydrogen and 5% methane at 37.8C
(50F) and 37.4 atm (550 psi). These are mixed with
recycled gas and liquid to form the reactor inlet flow.
There are some recipe constraints on the operation of
the reactor (Douglas, 1988):
?? The pressure has to be 34 atm (500 psi).
?? The temperature has to be in the range from 621C
(1150OF) to 704C (1300F).
?? The ratio of hydrogen on aromatics has to be 5/l
at the inlet.
To satisfy these constraints a furnace is included to heat
the reactor inlet flow. The outlet stream leaving the
reactor contains hydrogen, methane, benzene, toluene
and the unwanted by-product, diphenyl. To save energy
this flow is used to preheat the feed stream via a heat
exchanger hence some energy integration is achieved.
After a cooler, the aromatics are then condensed and the
light gases are flashed off using a flash drum. Through
a splitter a purge stream is removed from the flash va-
pour to prevent the buildup of methane. The remaining
gases are recycled through a compressor. Since not all
the hydrogen and methane can be separated from the
aromatics in the flash drum, a stabilizer is used to re-
move most of the excess. The product benzene is sep-
arated via a distillation column and obtained as the top
product. The remaining two aromatics are separated in
the toluene column and the toluene leaving from the top
of the column is recycled.
2.2 Background for Modelling
The HDA process has been extensively studied using
the hierarchical design approach introduced by Douglas
(1988). In this book, the material balance for the overall
process is calculated using FLOWTRAN. This material
balance is used for validating our dynamic simulation
of the HDA process.
Based on the flowsheet given in Figure 1, Bro-
gnaux (1992) built a dynamical simulation model for
the whole process in SpeedUp (Aspen Technology,
Inc., 1996b). The operation settings of this model were
later upgraded by Wolff (Wolff, 1994) to achieve eco-
nomic potential optimization.
The dynamic model of the plant was based on the
following assumptions:
The units with small time constant (less than 10
seconds) were modelled as instantaneous without
dynamics. This was done for the mixer, feed-
effluent heat exchanger (fehe), furnace, cooler,
purge, and compressor.
The plug flow reactor was modelled as a series of
19 CSTRs.
The computational load was reduced by using the
shortcut dynamic distillation model from the Spee-
dUp library. In this model the distillation columns
are divided into a few lumped tray models thus
resulting in fewer differential equations. Constant
holdups, constant liquid and vapour flows were as-
sumed in the shortcut model. In each lump, cor-
responding to a specific number of trays, constant
relative voiatilities were also assumed.
The light components of H2 and methane were
assumed to be nil in the benzene and toluene
columns.
However, some defects with the above simplified
model have been detected. which are:
The physical properties have been approximated
in the model. These are only valid for a specific
operation condition.
A modelling error exists in Brognaux reactor
model which causes inlet and outlet molar
flowrates of the reactor not to be balanced.
The energy balance is omitted in the shortcut dis-
tillation model. Thus measurements of tray tem-
peratures are not available in Brognaux model.
Hence, some control structure selection issues
cannot be studied using this model.
The physical property issue is solved using the physical
property database provided in Aspen Plus(Aspen Tech-
nology, Inc., 1996a). The other two issues are discussed
in the following sections.
European Symposium on Computer Aided Process Engineering-8 s413
3 Reactor Modelling
The reactor is modelled based on the main reaction in
(1). The reaction is characterized by a single variable,
the conversion, z, defined as:
Moles toluene converted in reactor
x=
Moles toluene fed to reactor
(3)
Another variable, the selectivity, s, is defined to de-
scribe the second reaction:
Molesbenzeneproduced
= Moles toluene converted
(4)
Assuming the reactor inlet is pure toluene, the con-
version and selectivity has a empirical relationship
(Douglas, 1988):
0.0036
The adopted reactor is a Plug Flow Reactor (PFR)
and is modelled by a series of CSTRs which have the
same volumes. Based on the definitions of conversion
and selectivity, the mass balance of a single CSTR is de-
scribed by the following equations: (Brognaux, 1992):
FT = ( 1 - x)FTO
FB = SXFTO +
2s - 1
- - BQ
I - " ,
FU = FDO + x -- FTO +
l-s
2
- FBO
2s
FH = FHO +
1+s
F-FRO - x2 FTO
FAN = FMO +X&I I
d& FTQ - FT
- =
dt V
-- kc&&
c =FTp
T
FM
c =FHp
H
FM
where:
FT
FTO
FB
FBO
FD
FDO
FH
FHO
FM
FMO
F
Cr
CH
k
V
outlet toluene molar flowrate;
inlet toluene molar flowrate;
outlet benzene molar flowrate;
inlet benzene molar flowrate;
outlet diphenyl molar flowrate;
inlet diphenyl molar flowrate;
outlet hydrogen molar flowrate;
inlet hydrogen molar flowrate;
outlet methane molar flowrate;
inlet methane molar flowrate;
total molar flowrate;
toluene concentration;
hydrogen concentration;
reaction rate constant;
reactor volume;
density;
molar weight.
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
In (10) a small error in Brognaux original model
(Brognaux, 1992) is corrected. This error causes the
total molar flowrate in inlet not to be equal to the total
molar flowrate in outlet.
One difficulty in modelling a PFR using a series of
CSTRs is to determine the number of CSTRs and the
volume of each CSTR. The model used by Brognaux
(1992) is a series of 19 CSTRs and each CSTR has a
volume of 470 [ft3]. However, this design gives con-
version value about 0.83 whilst the optimal conversion
value is 0.75 (Douglas, 1988). However due to price
raising of raw materials, recent research shows the op-
timal conversion should be less than 0.7 (Nelson and
Douglas, 1990; Chaudhuri and Diwekar, 1997). Ac-
cording to (5), the higher the conversion the lower the
selectivity and the greater the loss of the valuable raw
materials (toluene) to the by-product diphenyl. Hence.
the reactors conversion needs to be corrected.
For a single CSTR, the relationship between conver-
sion, 2 and volume, V can be derived from (11). At
steady state, dCT/dt = 0. Thus (11) leads to:
P FT
FTO = FT(~ + kVCH_) = -
FM 1-X
(14)
This gives:
xFM
= (1 - z)kCHp
(1.3
For a series of n CSTRs the relation of the total con-
version, xtot of the n-CSTR series and the conversion
ofasingleCSTR,x,,i = l,... ,n,is:
is1
(16)
whilst the total volume of the reactor is
For a PFR, the relationship of reactor volume, V,, and
the conversion, z is given by Douglas (1988, ~526):
vR =: F In P/(1 - x)lM
b
(18)
Douglas also gave a design VR = 4090 [ft3]. To choose
n and Vtot for the CSTR series, (16) and (17) are solved
using SpeedUp for x = 0.75 and inlet data given by
Douglas (1988). As a first guess, equal conversion of
each CSTR is assumed, i.e. x1 = z2 = . . = x,, = l-
(1 - st,,t)(l/). With this assumption, the total reactor
volumes, Vtot for n =l, 2, 5, 10, 12, 19 are shown in
Table I. It is shown that as the number of CSTRs, n,
Table 1: Reactor volume for a series of CSTRs with
same conversion [ft3]
12 CSTRS
Vtot
7% CSTRS
Vtot
1 8676.5 10 4137.2
2 5547.9 12 4091.7
5 4432.3 19 4007.3
increases, the total volume decreases. Brognaux error
s414 European Symposium on Computer Aided Process Engineering-8
Table 2: Reactor volume for a series of CSTRs with
same volume [ft3]
n CSTRs
Vtot
n CSTRs Vtot
PFR 4090 1 10 4140.5
2 5558 12 4096.2
5 4441.5 19 4016.2
Table 3: Outlet of different CSTR series [lbmol/Hr]
(z = 0.75)
Comoonent DOll&lS
1517.41
7l= 10 7L = 12 11 = 19
Hydrogen 1576.96 1517.34 1579.10
Methane 2399.74 2399.63 2399.75 2399.11
Benzene 296.695 297.827 296.943 293.402
Toluene 92.8054 92.9105 92.7953 92.7717
Diphenyl 4.61146 3.99241 4.49225 6.27458
is due to using the same V& for 19 CSTRs as for one
CSTR.
For a series of CSTRs with the same volumes, the
volume of a single CSTR can be obtained by slightly
adjusting the values given in Table 1 to satisfy the given
conversion. The adjusted reactor volumes are shown in
Table 2. Compare Table 2 with Table 1, the total volume
of a series of CSTRs with same volume is slightly lar-
ger than those with same conversion.
To decide an appropriate CSTR number, the outlet
molar flowrates of different CSTR series are compared
with the data provided by Douglas (1988). The results
for n = 10, 12, and 19 are shown in Table 3. The res-
ults in Table 3 show that a series of n = 12 CSTRs pro-
duces the outlet most close to Douglas design. Also the
total volume of the I2-CSTR series in Table 2 is very
close to the value V, = 4090 [ft3] given by Douglas
(1988, p.527).
4 Rigorous Distillation Column
Model
In Brognaux original model, the shortcut models
provided by SpeedUp (Aspen Technology, Inc., 1996b)
were used for the three distillation columns. In these
shortcut models, constant holdup, constant relative
volatility, constant liquid and vapour flows, constant
column pressure and temperature are assumed. Due to
these assumptions, any temperature and pressure dis-
turbances will have no effect on these columns. Also,
the operation of these columns has also little effect on
the whole system (see control structure selection results
in (Cao and Rossiter, 1997)). Obviously, it is not rep-
resentative of the real systems behaviour. To get more
realistic results, a rigorous tray-by-tray column mode1
is considered to replace the shortcut model.
The rigorous column model adopted includes six
parts: stage model, stage with feed model, reboiler
model, total/partial condenser model, condenser drum
model and condenser splitter model. The details of
these models are described in the following sections.
4.1 Stage Model
The stage (tray) model includes: dynamic mass bal-
ance, dynamic energy balance, phase equilibrium, li-
quid and vapour flowrate and overall plate efficiency.
Equations for dynamic mass balance are:
dH
-=Lz+V~-L~-Vo
d(Hif)
(19)
__----Y = Lzxzt + I/lyrz - Lox2 - voyi, (20)
dt
i = l,... ,5
where Lz, LO, VI and VO are liquid inlet and outlet,
and vapour inlet and outlet molar flowrates; zri, x,, Ylz
and yi are liquid inlet and outlet, and vapour inlet and
outlet mole fraction, and H is liquid molar holdup.
Dynamic energy balance equation is:
d(H&)
~ = LrELI + VzEvz - LoEL - VoEv
dt
(21)
where ELI, EL, EVZ and Ev are liquid inlet and outlet,
and vapour inlet and outlet enthalpy.
The enthalpies are calculated by SpeedUp physical
property procedures.
Equations for phase equilibrium and overall plate ef-
ficiency are:
Yei = &Xi,
i = l>... ,5
(22)
&Ye* =2x% (23)
i=l 2=1
yz - yzi = Eo(~ez - yi),
i= l,... ,5
(24)
where K,, i = 1, . . ,5 are distribution coefficients
(K-values) calculated by SpeedUp physical property
procedures, Eo is the overall plate efficiency given in
Douglas book and yei, i = 1, ,5 are internal vari-
ables.
Liquid overflow is governed by a form of the Fran-
cis equation corrected for wall effects (Sinnott, 1993,
p.513):
AhLpL = HML
h, = hL - WH
LOML
-=
CWLhl.
PL
0
(25)
(26)
(27)
where:
A Effective plate area;
hL Liquid level;
pi Liquid density;
ML Liquid molar weight;
h,
Liquid height above weir;
WH Weir height;
WL Weir length;
C Empirical constant in Francis weir equation.
Vapour flow is related to the pressure drop across the
dry plate via a relationship derived for flow through the
European Symposium on Computer Aided Process Engineering-8
s41.5
plate holes (Sinnott, 1993, ~518):
Pvin. -
P,, = AP, + APd
AP, = p&L
(28)
(29)
VIMv
J
Apd
-=
Pv Af
co -
51PV
(30)
where Pvin and P, are inlet and outlet vapour pres-
sures, respectively; AP, and APd are hydrostatic and
dry pressure drops, respectively; M, and pv are va-
pour density and molar weight respectively; Cu is ori-
fice coefficient and Af is the effective free/flow area of
a plate.
4.2 Feed stage
In the feed stage, the equations of mass balance, (19)
and (20) and energy balance, (21) are changed as fol-
lows:
dH
-=F+LI+VI-Lo-V0
dt
(31)
d(Hxi)
- = Fz~+LIxri +V,yri - LOX,
dt
(32)
- VOYi >
i = l,... ,5
d(HEL)
~ = FEF + LIELI + VrEvr
dt
(33)
- LoEL - VoEv
where F, Zi, i = 1,. . ,5 and EF are feed molar
flowrate, molar fractions and enthalpy, respectively.
4.3 Reboiler Model
The reboiler model includes dynamic mass balance, dy-
namic energy balance and phase equilibrium:
H = HLxi +Hvyi
dH
- = LIxri - Vyi - Bx~,
dt
i=l,... ,5
EH = HLEL + HvEv
dEH
-=Q+LIEI-VE-BEL
dt
Yt =Ki Xt , i=l , . . . ,5
(34)
(35)
(36)
(37)
(38)
(39)
i=l
2=1
HvMv
AhL _ H%L +
HLML
Ah=- -

PL
B2 = (Pv - PB + pLhL)K:
(40)
(41)
(42)
(43)
where:
H
HL
Hv
B
EH
Q
A
h
hL
KB
Total holdup;
Liquid holdup;
Vapour holdup;
Bottom stream molar flowrate;
Enthalpy holdup;
Reboiler heat duty;
Reboiler area;
Reboiler height;
Liquid level;
Bottom valve constant.
4.4 Total Condenser
The total condenser model includes mass balance, en-
ergy balance and phase equilibrium:
L = VI
(44)
Xi =Yi,
i= l,... ,5
(45)
Qc = VrEv - LEL
(46)
yei = Kx.
2,
i= l,... ,5 (47)
EYei = 1 (48)
i=l
vf == (PI - P)KL (49)
where:
VI
Q!
J%
EL
2,
YZ
Yea
K,
PI
P
KL
Vapour inlet molar flowrate;
Liquid outlet molar flowrate;
Condenser cooling duty;
Inlet vapour enthalpy;
Outlet liquid enthalpy;
Liquid mole fraction;
Vapour mole fraction;
Internal vapour mole fraction;
Distribution coefficient (K-value):
Inlet pressure;
Outlet pressure:
Vapour inlet line loss constant.
4.5 Partial Condenser
Equations in the partial condenser model are:
I/I=V+L
(50)
VIYIi = Lx, + Vyi,
i=1...,5
(51)
Q,: = VIEI - LEL - VEV
(52)
~1 = KG,
i=l,... ,5
(53)
3
CYz=l
ix1
v; = (PI - P)KL
(54)
(55)
S416 European Symposium on Computer Aided Process Engineering-8
where:
VI Vapour inlet molar flowrate;
L Liquid outlet molar flowrate;
c Vapour outlet (distillate) molar flowrate;
Qc Condenser cooling duty;
El Inlet vapour enthalpy;
Ev Outlet vapour enthalpy;
EL Outlet liquid enthalpy;
Yl%
Vapour inlet mole fraction;
2, Liquid outlet mole fraction;
yZ Vapour outlet mole fraction;
K, Distribution coefficient (K-value);
PI Inlet pressure;
P Outlet pressure;
KL Vapour inlet line loss constant.
4.6 Condenser Drum
The inlet of condenser drum is from the outlet of a
total/partial condenser. The above total/partial con-
denser models are not dynamic, but the condenser drum
model given in this section contains full dynamic rela-
tionships.
!E=L -L
dt I
d(Hx,)
- = LIXIi - LXi
dt
i = l,... ,5
4HE)
dt
= LIE, - LE
(56)
(57)
p = pr + (h + k&L (59)
(60)
where:
H Liquid holdup;
LI Inlet liquid;
L Outlet liquid;
x, Outlet mole fraction;
xii Inlet mole fraction;
E Outlet enthalpy;
Er Inlet enthalpy;
P Outlet pressure;
PI Inlet pressure;
hi Liquid level;
hd Drum elevation from exit point;
A Drum cross-sectional area;
pi Liquid density;
ML Liquid molar weight.
4.7 Reflux splitter
The inlet of a reflux splitter is the outlet of a condenser
drum. The equations for the splitter model are:
F=L+D (61)
L=DR (62)
where:
F Inlet molar flowrate;
L Reflux molar flow;
D Distillate molar flow;
R Reflux ratio
4.8 Column design
The design parameters for the three columns are listed
in Table 4. Part of these data are adopted from Douglas
(1988).
Table 4: Column design for the HDA process
Parameters Stabilizer Benzene Toluene
Number of Trays 21 55 23
Feed Tray 14 25 12
Reflux Ratio 0.2 1.6 0.05
Overall Plate Eff. (Eo) 0.282 0.55 0.34
Diameter [ft] 2 5.1 3
Top pressure [psi] 150 20 15
5 Model validation
Apart from the reactor re-design and adopting rigorous
distillation column model, other changes are: several
pumps and valves are included in the HDA model, all
liquid levels and top pressures of distillation columns
are controlled. All these changes make the model more
realistic than the original model. To validate the new
upgraded HDA model, the mass balance data provided
by Douglas (1988) are used for comparison. The mass
balance of the whole plant is shown in Figure 2.
The data in Figure 2 are quite close to those in
Douglas (1988, p.392). The largest difference of these
two sets data is in the gas recycle flowrate. The reason
for this difference is the different hydrogen/aromatics
ratio setting. In Figure 2, this ratio is set to 5 whilst
in Douglas book it is equal to about 4.6. Since in the
book it is stated a ratio of 5 should be adopted, it ap-
pears there may be an error in Douglas mass balance.
6 Discussion
6.1 Effect on Control
The aim of the model upgrading is to detect the effect
of modelling on control structure selection. After the
model modification, a brief steady state controllability
analysis for the new model has been performed. Com-
parison with previous results, obtained using the ori-
ginal model, shows some differences which are repor-
ted here.
Due to the addition of mass balance control in
the form of pre-included level and pressure con-
trol loops, some manipulated variables are now
not available, such as the liquid valve of the flash
drum.
Some new manipulated variables are introduced
in the new model which can be used for replace-
ment of some old manipulable variables, such as
reboiler heat duty is used to replace reboiler va-
pour ratio.
European Symposium on Computer Aided Process Engineering-8
s417
?? Some new manipulable variables provide more op-
tions for input selection. For example, in the
FEHE (feed-effluent heat exchanger) unit, a by-
pass valve is introduced in the hot flow stream. It
is detected that this valve has the most significant
effect on the flash inlet temperature.
References
Aspen Technology, Inc. (1996a). ASPEN Users
Manual, Aspen Technology Inc., Cambridge, UK.
?? The most significant changes are the relative im-
portance of the three distillation columns. The pre-
vious results showed that only the operation of the
benzene column had a large effect on the whole
plant performance. However, analysis with the
new model shows that all three columns are im-
portant for whole plant operation, at least at steady
state.
Aspen Technology, Inc. (1996b). User Manual for
SpeedUp Release SS-5, Aspen Technology Inc.,
Cambridge, UK.
Brognaux, C. (1992). A case study in operability ana-
lysis: the hda plant, Masters thesis, University of
London.
Cao, Y. and Rossiter, D. (1997). An input pre-screening
technique for control structure selection, Com-
puters Chem. Engng. 21(6): 563-569.
?? The negative effect is that the size of whole dy- Cao, Y., Rossiter, D. and Owens, D. H. (1997a). In-
namic model becomes 10 times larger than the ori- put selection for disturbance rejection under ma-
ginal model. Thus it becomes more difficult to per- nipulated variable constraints, Computers Chem.
form dynamical analysis and dynamic simulation.
Engng. 21: S403-S408.
6.2 Further Model Improvement
Several issues are considered for further model im-
provement, which are:
Cao, Y., Rossiter, D. and Owens, D. H. (1997b).
Screening criteria for input and output selection,
ECC97 CDROM, number 773, Brussels, Belgium.
?? A quench flow from the flash liquid outlet to cool
the reactor effluent to 1150 [OF] needs to be in-
cluded in the model. This modification is also ex-
pected to has some effect on the control structure
selection results.
Chaudhuri, P. D. and Diwekar, U. M. (1997). Syn-
thesis under uncertainty with simulators, Com-
puters Chem. Engng. 21: 733-738.
Douglas, J. (1988). Conceptual Design of Chemical
Processes, McGraw-Hi11 Book Company, New
York.
?? The working point of the model needs to be optim-
ized. However, due to the large size of the model,
it is very difficult to perform optimization in Spee-
dup. A steady state model optimized in Aspen
Plus could be considered. Then the optimized op-
eration point can be transferred to SpeedUp for dy-
namic simulation.
Nelson, D. A. and Douglas, J. M. (1990). A systematic
procedure for retrofitting chemical-plants to oper-
ate utilizing different reaction paths, Ind. & Eng.
Chem. Rex 29: 819-829.
Sinnott, R. K. (1993). Coulson & Richardsons Chem-
ical Engineering, Wumn 6, Pergamon Press, Ox-
ford.
?? Mode1 simplification is also considered to make
it easier to perform dynamic simulation. This
simplification should not affect the control struc-
ture selection results. Some possible approaches
are: reduce the number of components from 5 to
3 for the benzene and toluene columns, drop the
plate efficiency to reduce the number of stages of
the columns. Since the column design given by
Douglas is very conservative, this change will re-
duce the systems overall size dramatically.
Wolff, E. (1994). Studies on Control of I ntegrated
Plunts, PhD thesis, University of Trondheim, the
Norwegian Institute of Technology.
?? According to recent research results in the liter-
ature (Chaudhuri and Diwekar, 1997) the original
flowsheet is also not optimized. For example,
the stabilizer column can be replaced by a second
flash drum and the flash vapour outlet is recycled
(Chaudhuri and Diwekar, 1997).
7 Acknowledgement
This work is funded by the Process Engineering Com-
mittee of EPSRC, UK: Grant No. GR/K74807.
S418 European Symposium on ComputerAided Process Engineering--l
HDA SPEEDUP Rel. 5.5-5
Copyright 1982-1996 Aspen Technology, Inc. All rights reserved
HDA Plant Material Balances
------_-~------_-_-_-------
H2 = 467.875 1 1 To1 = 279.375
CH4 = 24.625 1 1
I 1 H2 = 1543.430 H2 = 192.299
Ben = 1.626 1 ( CH4 = 2289.045 CH4 = 285.197
Tel= 88.616 VV Ben = 29.274 Ben = 3.647
Diph = 0.021 +---+ To1 = 3.346 /--+ +---+ To1 = 0.417
+-----__>I I<----__________+
I<--+
+------__------->
f-+-+ \-A +---+ ypH = 0.399
^
H2 = 2011.305
CH4 = 2313.670
Ben = 30.900
To1 = 371.337
Diph = 0.022
V
+---+ T = 260.803
I
H2 = 1735.729
t--->I
+----+
/ CH4 = 2574.242
f-+-+
I J 7
/ Ben = 32.921
H2 = 1737,373 ) Tel= 3.763
V +---+ CH4 = 2592.204 I Diph = 0.001
+---+ / Ben = 300.230 l
I I
+-+-+ To1 = 92.803 1
+-+-+ / Diph = 4.623 +-+-+
I
+-__--_--------___>I I
I I
T=
V H2/Arom. = 5
+---+ x = 0.750
I I I
+-+-+
I
l T = 1265.893
+------+
99.997 +-+-+
000
I
H2 = 1.644
CH4 = 17.961
Ben = 267.309
H2 = 1.644 To1 = 89.040
CH4 = 17.961 I Diph = 4.623
Ben = 265.000 Ben = 0.682 I
Pur = 99.970 ^
^
I
+-+-+ +-+-+ +-f-+
I
j; ~ <---+
:<__-I
+ T = 100.659
+-+-+
I
+-i-i +-+-+
I I I
+---------+
+------+ T = 373.348
V T = 261.296
Ben = 0.0
+--------________--__--_+
To1 = 0.339 1 FLOWS lb.mol/hr I
Diph = 4.602 1 TEMPERATURE F
I
+------__--------------_+
Figure 2: SpeedUp results of material balance for HDA
process