Upload
alexandre-c
View
227
Download
6
Embed Size (px)
Citation preview
CHAPTER
CASE STUDIES 2121.1 INTRODUCTIONThe three case studies that follow illustrate key issues developed in this book. The first one deals with
the design and simulation of a process for the thermal hydrodealkylation (THDA) of toluene to ben-
zene. This typical petrochemical process has been used in several chapters of this book, particularly in
developing the hierarchical approach of process synthesis. Although relatively simple, this case study is
very helpful for teaching the key aspects of conceptual process design. In this chapter, we will put to-
gether different aspects developed so far with respect to process synthesis and energy integration, com-
plete the steady-state and dynamic simulation with process control implementation and finally discuss
the quality of design from the perspective of flexibility in operation.
The second case study deals with the product and process design. The object is glycerol tert-butylether (GTBE), a combustion enhancer of motor fuels, which adds more value to a low cost by-product
available today in large amounts from biodiesel industries.
The third case study handles an advanced subject in process intensification, the production of bio-
diesel by the esterification of fatty acids with methanol in a reactive divided-wall column (R-DWC).
The above cases are supported by computer simulation in both steady-state and dynamic modes.
21.2 DESIGN AND SIMULATION OF A HDA PLANTBenzene is a valuable base chemical product. A major source is the dealkylation of aromatic feedstock,
namely of toluene, available in large amounts by reforming processes.
The main reaction is the transformation of toluene to benzene in the presence of hydrogen:
Toluene Tð Þ +Hydrogen Hð Þ!Benzene Bð Þ+Methane Mð Þ (21.1)
The reaction is moderately exothermic. At 25 �C, 1 bar and gas-phase reaction, the thermal effect is
�9.98 kcal/mol, while at 35 bar and 630 �C, this raises to�12.1 kcal/mol. The reaction can take place
by a purely thermal (THDA) mechanism or in the presence of suitable catalysts, such as chromium,
molybdenum and platinum catalysts supported on silica or aluminium oxides. The following equation
describes the kinetics of a thermal process (Douglas, 1988):
r1¼ 6:3�1010exp �52,000=RTð ÞcTc1=2H (21.2)
Computer Aided Chemical Engineering. Volume 35. ISSN 1570-7946. http://dx.doi.org/10.1016/B978-0-444-62700-1.00021-8
© 2014 Elsevier B.V. All rights reserved.789
By-products appear too, such as diphenyl, naphthalene and other heavy hydrocarbons. Here, we con-
sider only diphenyl formation by the reversible reaction:
Benzene$Diphenyl + 2Hydrogen (21.3)
In order to cope with the dynamic simulation requirements, we introduce the following overall rate
equation that fits the selectivity data presented in Chapter 7:
r2¼ 3:0�109exp �52,000=RTð Þc2B (21.4)
The reaction rates above are expressed in kmol/m3/s and the concentrations in kmol/m3.
The conditions of the THDA process are temperature 620–700 �C and pressure 30–35 bar. With
catalyst, the temperature may be reduced with 50–60 �C. The technology requires high H2 excess at
the reactor inlet at constant molar ratio hydrogen/toluene of 5/1 to prevent coke formation.
In a first approach, the raw materials are pure toluene and hydrogen with 5% methane.
The purity of the benzene product should be over 99.5%. The study is done for a nominal capacity
of 125 kmol/h toluene, which is equivalent to 11,500 kg/h or to 93,725 tonnes/year for 8150
production hours.
21.2.1 PROCESS SYNTHESISFigure 21.1 presents a simplified flowsheet, as developed in Chapter 7. Fresh rawmaterials, mixed with
recycled toluene and hydrogen, are heated up to the reaction temperature via a feed effluent heat ex-
changer (FEHE) and a furnace. The reaction takes place in an adiabatic plug flow reactor. After heat
exchange with the incoming stream, the outlet mixture is cooled finally at a temperature suitable for
separating gas and liquid streams by a simple flash, usually between 30 and 33 �C. Note that a bypasscold liquid flow is used for quenching the reactor outlet.
The gaseous stream containing hydrogen and methane with traces of benzene and toluene is
recycled to the reaction section. The accumulation of methane in recycle is prevented by using a purge.
This is also the place of eliminating the secondary product, methane, but simultaneously, hydrogen is
eliminated too. The excess of hydrogen in the fresh feed above the stoichiometric requirements and the
loss in purge are linked through the material balance. In addition, the recycle of hydrogen must ensure a
constant ratio hydrogen/hydrocarbon at the reactor inlet. This feature has important consequences on
both design and control of the whole plant.
The liquid stream contains benzene and toluene, as well as heavy impurities, mainly diphenyl, with
traces of dissolved gases. The gaseous components are removed as lights at the top of a stabiliser col-
umn C-1. This is followed by the separation of the benzene product in the distillation column C-2. The
final column C-3 separates on top the unreacted toluene for recycling and, in bottoms, the heavies,
namely, diphenyl.
Note that the separation of the liquid mixture can equally be done in a sequence of two columns,
stabiliser plus benzene top distillation with side stream for toluene recycling and heavies in bottoms.
The existence of two recycles sets an optimum for toluene conversion. Higher conversion decreases
the gas recycle cost but increases the cost of separations due to more by-products and impurities.
A previous optimisation study indicated that the optimal conversion should be around 70%
(Dimian, 2003).
790 CHAPTER 21 CASE STUDIES
21.2.2 DESIGN AND SIMULATIONIn preliminary design, the analysis can focus on the material balance only.
21.2.2.1 Reaction sectionThe reaction rate becomes commercially interesting above 620 �C. Because the adiabatic temperature
rise is about 70 �C, the maximum inlet temperature is considered 640 �C. In Chapter 7, we demonstrate
that using a reasonable large reactor is recommended in order to cope with stability and flexibility in
operation. In this case, a reactor larger than 110 m3 is necessary for toluene feed of 125 kmol/h, con-
version around 70% and gas recycle at 2000 kmol/h. In order to ensure adequate flexibility, we consider
a somewhat larger reactor with the dimensions D¼3 m and L¼20 m and a volume of 141.3 m3.
21.2.2.2 SeparationsThe stabiliser can be simulated as a distillation column with vapour distillate and feed near the top.
Since the distillate amount is very low, setting a small constant reflux flow instead of reflux ratio is
recommended. For the benzene/toluene and toluene/diphenyl separations, the Fenske-Underwood-
Gilliland shortcut method can be used for pre-design. Note that the use of designmode instead of ratingmodelling in preliminary simulation makes the recycles converge easier.
FIGURE 21.1
Simplified flow sheet for the HDA process.
79121.2 DESIGN AND SIMULATION OF A HDA PLANT
21.2.2.3 SimulationThis case study has been simulated with Aspen Plus and Aspen Dynamics version 8.4, but the approach
remains valid for other software. A first issue is the thermodynamic modelling. An equation-of-state
option is appropriate for the whole process, for example, Peng–Robinson or Soave–Redlich–Kwong.
The quench cooling has been simulated by a design specification that keeps the temperature at 620 �Cby manipulating the flow rate of a bypass after the cooler. A pressure drop of 2 bar is assumed for the
whole gas line.
The liquid separation section may be simulated initially as a black box, which allows focusing the
analysis on the reaction section. Rigorous units can be considered after closing the recycles.
An important feature in simulation is that the material balance must be adapted to fulfil the con-
straint of a given ratio hydrogen/toluene at the reactor inlet. This has been done bymeans of a controller
(design specification), where the manipulated variable is the fresh feed of hydrogen.
A debatable point is the specification of purge. Setting purge flow rate or composition is not fea-
sible, because of incompatibility with the overall material balance (see Chapter 3). On the contrary,
setting the split ratio or keeping constant the flow rate of the gas recycle leads always to convergence.
This solution was chosen here for technical reason too, because the compressor is an expensive equip-
ment that should preferably operate at a constant flow rate close to the maximum capacity.
21.2.2.4 ResultsThe objective of the preliminary simulation is getting a good operation point from the material balance
viewpoint. Table 21.1 presents the main streams, while Table 21.3 shows the results for the distillation
columns. The reactor-inlet temperature is at 633 �C, and the gas recycle is at 2000 kmol/h. It can be
seen that due to a large recycle, the reactor-outlet temperature is kept well below 700 �C. The residencetime of 95 s gives a superficial gas velocity of 0.21 m/s. The constraint of a hydrogen/toluene ratio at 5
results in a make-up hydrogen flow of 196.4 kmol/h. The toluene conversion is (173.5�50.5)/
173.5¼0.708. The above hydrogen excess results in a molar fraction of hydrogen in purge of 0.34.
From Table 21.1, it can be seen that the overall yield of benzene product with respect to toluene is
(119.5/123)�100¼95.6%molar but only 81.8%mass. The yield could rise to 96.8%molar if all losses
of benzene in outputs would be recovered. A further increase to about 97.8% can be obtained by re-
ducing the toluene lost in heavies with 1 kmol/h. The rest of yield loss of about 2% is due to the trans-
formation of benzene in diphenyl. Thus, getting a higher overall yield can be achieved by preventing
losses in outputs and working slightly below 70% conversion.
21.2.3 ENERGY INTEGRATIONThe maximum energy saving can be obtained by applying pinch point analysis taking the whole pop-
ulation of streams and heat sources/sinks. This approach followed by Douglas (1988) resulted in the
flow sheet presented in Figure 7.2. In this approach, the reaction and separation systems are tightly
integrated. Thus, the hot reactor-outlet stream is used to drive the reboilers of the distillation columns.
However, this solution raises questions about operability and controllability of the plant. For this rea-
son, we consider another solution, the local energy integration of sections, respectively, reaction and
separations. The excess of heat generated in the reactor can be exported to the utility network and
re-imported for the needs of the distillation section at adequate pressure levels. Moreover, supplemen-
tary low-cost energy can be injected in the reactor heat-integrated loop and recovered to be used as
utility elsewhere.
792 CHAPTER 21 CASE STUDIES
Table 21.1 Preliminary Material Balance
Feed H Feed T Rec. H Rec. T Rin Rout Product Heavies Purge Gas
Temp. (�C) 25 25 31.5 118.4 633.0 670.2 85.9 149.1 30.0 59.1
Pres. (bar) 35.0 35.0 35.0 35.0 34.4 34.0 1.2 1.3 33.0 11.0
Flow (kmol/h) 196.4 125 2000.0 48.2 2369.6 2369.6 119.5 2.5 188.5 10.9
Toluene 0.0 125 1.5 47.0 173.5 50.5 0.6 1.3 0.1 0.0
Hydrogen 186.5 0.0 680.9 0.0 867.5 745.8 Trace 0.0 64.2 0.7
Benzene 0.0 0.0 10.9 1.2 12.1 132.5 118.9 0.0 1.0 0.5
Methane 9.8 0.0 1306.8 0.0 1316.6 1439.6 Trace 0.0 123.2 9.6
Diphenyl 0.0 0.0 <0.001 <0.001 <0.001 1.3 Trace 1.3 Trace Trace
Firstly, we examine the heat balance around the reactor by disregarding the quench. By simulation,
we find that the enthalpy for heating up the reaction mixture from the mixer (40 �C) to reactor inlet
(633 �C) is Qcold¼23.3 MW. On the hot side, the enthalpy difference from the reactor exit
(670 �C) to flash separation (30 �C) is Qhot¼24.9 MW. The amount Qex¼Qcold�Qhot¼1.6MW rep-
resents the excess of energy, mainly due to the exothermic chemical reaction.
Figure 21.2 describes the heat integration around the adiabatic chemical reactor. The initial mixture
enters a FEHE, where heat exchange with the hot outlet stream from reactor takes place. Further, the
stream goes to a furnace, being heated up to the temperature at which the reaction starts. Note that
installing a heat exchanger for setting the temperature at the inlet of the adiabatic reactor is compulsory
because of dynamic stability reasons, as it can be demonstrated theoretically. From energy balance
viewpoint, the duties of the FEHE and furnace are linked by the relation
Q FEHEð Þ+Q furnaceð Þ¼Qcold (21.5)
The repartition of duties between the two items can be set by technological reasons. Thus, one can
operate with a large FEHE and a smaller furnace, or the opposite (Luyben et al., 1999; Bildea and
Dimian, 1998). In the first case, energy saving is favoured, and in the second one, the process dynamics.
The furnace can be also used to put value on available low-value fuel.
On the hot side, we can install a heat exchanger for recovering high-temperature energy by steam
generation (SG), the FEHE and a cooler for completing the heat balance. The three items are linked by
the relation
Q SGð Þ +Q FEHEð Þ+Q coolerð Þ¼Qhot (21.6)
It can be seen that the sum of duties of steam generator and cooler is determined by the furnace:
Q SGð Þ+Q coolerð Þ¼Q furnaceð Þ+Qex (21.7)
In addition, the heat exchangers must respect the condition of a minimum temperature difference
between hot and cold streams.
FIGURE 21.2
Heat integration around the adiabatic HDA reactor.
794 CHAPTER 21 CASE STUDIES
The above relations make possible a number of alternatives. Without SG, the excess of energy in
furnace generated by reaction is lost in the cooler. Note that the direct analysis by simulation of the
above heat-integrated loop is possible in Aspen Plus by using the two-stream block HeatX. The
short-cut option can be used in design or rating mode. Easy convergence is obtained when specifying
the duty.
The above scheme was studied theoretically by Bildea and Dimian, (1998) from the viewpoint of
integrating design and control. They considered as disturbances the reactor input flow, SG duty and
overall heat transfer coefficient of FEHE because of fouling. It was found that the best alternative
was with moderate furnace, small steam generator and high-efficiency FEHE.
Table 21.2 presents the characteristics of heat exchangers implied in simulation. Note that relatively
high overall heat transfer coefficients can be assigned because of the high conductivity of the gas mix-
ture (methane and hydrogen) and higher pressure.
Next, we examine the heat integration of the distillation section. Table 21.3 presents some charac-
teristics of the three columns. It can be seen that thermal coupling of (C-2) and (C-3) would be possible
by raising the pressure in the last. A saving of about 30% in energy could be achieved, but the disad-
vantages are higher reboiler temperature and change in utilities’ profile. A much better solution is
assembling them in a single side-stream column, as mentioned earlier.
Table 21.2 Characteristics of Heat Exchangers Around the Heat-Integrated HDA Reactor
Heat Exchanger Duty (MW) Temperature (�C) LMTD (�C) U (W/m2K) Area
a(m
2)
FEHE 16.5 81 523
40!474
45 150 2500
Furnace 5.0 522!633 250 50 500
Steam generator
Steam 40 bar
6.5 671!523
250!251
110 50 1000
Cooler 2.0 81!30
40 20
22 200 500
aRounded to modules.
Table 21.3 Simulation Results for the Separation Section
C-1 C-2 C-3
Nstages 7 27 7
R/D R¼500 kg/h 1.3 0.6
Pc (bar) 11 1.2 1.1
Tc (�C) 50 85.9 113.5
Tr (�C) 194 123.9 149
Qc (kW) �76 �2150 �710Qr (kW) 1460 2150 710
79521.2 DESIGN AND SIMULATION OF A HDA PLANT
21.2.4 PLANTWIDE CONTROLFigure 21.3 illustrates the flowsheet of the HDA plant with control implementation. The assigned
plantwide control objectives and suitable control loops are as follows:
1. Production: in a first approach, toluene feed is set on flow control (self-regulation). This is
the simplest solution, which is expected to work because the plant was designed for high
toluene conversion, XTol¼0.7, which is achieved by means of a large reactor. Alternatively, the
toluene may be fed as make-up in a buffer vessel (e.g. the reflux drum of the recycle column), from
where a constant flow rate is sent to the reactor (Bildea et al., 2004). This solution may
accommodate large-throughput variations, but the production is set indirectly by modifying the set
point of the recycle.
2. Hydrogen make-up: hydrogen/toluene ratio kept on setpoint with the fresh hydrogen feed.
3. Reaction system: reactor-inlet temperature with the furnace duty.
4. Gas recycle: keep constant the recycle flow rate close to the maximum compressor capacity.
5. Gas loop pressure: keep constant the gas pressure after flash by manipulating the vapour flow.
Reactor
Flash
Purge
FEHE Furnace
Steam generator
Quench
Compressor
Sta
bilis
er
(FT)
(FH)
(1)
(22)
(D)
(P)
(RH)
Cooler
TC
LC
PC
LC
TC
LC FC
LC
TC
LC FC
LC
(B)
(M)
TCLC
PC
TC
TC
FC
Pro
duct
col
umn
Rec
ycle
col
umn
H2/T
FC
Production(2)
(3) (4)
(5)
(6)
(7)(8)
(20)
(10)
(23)
(24)
(26)
(27)
(RT)
x
(11)
PC
PC
FIGURE 21.3
Process control implementation of the HDA plant.
796 CHAPTER 21 CASE STUDIES
The control of units may follow the standard control structures applicable for stand-alone units.
The HDA plant has been decomposed in several parts for an easier control implementation in Aspen
Dynamics™ as follows:
1. Heat-integrated reaction loop
2. Separation section
3. Make-up of reactants
21.2.4.1 Control of the heat-integrated reaction loopThe following controllers can be considered:
• TC1: reactor-inlet temperature with furnace duty
• TC2: reactor-outlet temperature with quench flow
21.2.4.2 Control of the separation sectionThe control of the separation section is developed unit by unit by applying standard control schemes.
Stabiliser. The control objective is the stripping of gases from liquid but with minimum loss of
benzene. A suitable control structure is as follows:
• Reboiler level with bottom product
• Reflux drum level with reflux flow rate
• Column pressure with vapour distillate
• Composition (inferential control) with reboiler duty
Benzene column. The purity of distillate has to be over 99.5%. One-point control of purity gives good
results if the reflux flow rate is high enough. Setting the reflux ratio allows good behaviour at large
variations in throughput. The control configuration is as follows:
• Level in reflux drum with distillate
• Level in reboiler with bottoms
• Condenser pressure with cooling water
• Constant reflux ratio
• One-point quality control: inferential concentration measurement (sensitive temperature in the
rectifying section) with reboiler duty
Recycle column. This column operates more as stripping. One-point quality control is sufficient:
• Level in reflux drum with distillate
• Level in reboiler with bottom product
• Condenser pressure with cooling water
• One-point quality control: inferential concentration measurement (sensitive temperature in the
rectifying section) with reboiler duty
21.2.4.3 ResultsIn this exercise, the toluene feed is set on flow control, giving a direct measure of the achievable pro-
duction. Hydrogen is on make-up control to keep constant hydrogen/toluene ratio at the reactor inlet.
The purge flow is let free. The proposed control structure is of self-regulation type.
79721.2 DESIGN AND SIMULATION OF A HDA PLANT
Figure 21.4 presents the flexibility of the plant, in terms of benzene production rate and product
purity, for a nominal toluene feed rate of 125 kmol/h with increase/decrease of 25 kmol/h (20%),
keeping the same parameters for controllers. The reactor-inlet temperature was increased from 633 to
640 �C. It may be seen that the plant reacts smoothly at these large variations. The benzene production
increases/decreases proportionally. At higher production, the product purity increases slightly, while by
decreasing the throughput, the opposite happens. When the reactor-inlet temperature was 633 �C, themaximum increase in the production rate was only 8% but still important, demonstrating that the reactor
volume alone ensures already a good flexibility in operation by self-regulation. However, acting on
reaction temperature allows getting a much stronger effect, as reported also by Luyben et al. (1999).
21.2.5 SUGGESTIONS FOR FOLLOW-UP PROJECTS1. Examine solutions ensuring better efficiency of raw materials, such as the use of membranes for
recovering methane as a valuable by-product.
2. Examine the process for simultaneous production of benzene and xylenes by toluene
disproportionation.
3. Examine other dealkylation processes for benzene production, namely, by the catalytic conversion
of aromatic streams.
4. Study the production of BTX from low-value refinery streams with recent technologies.
21.3 DESIGN AND ROBUST OPTIMISATION OF A PROCESS FOR GTBEMANUFACTURING21.3.1 INTRODUCTIONIn Chapter 12, GTBEs were identified as valuable fuel additives. Based on literature data, the reaction
of glycerol with isobutene catalysed by p-toluene sulphonic acid appeared as a promising process. The
reaction leads to a mixture of mono-, di- and tri-tert-butyl glycerol ethers (ME,DE and TE), fromwhich
ME is recycled, DE and TE being the process products:
0.99
0.992
0.994
0.996
0.998
1
100
110
120
130
140
150
160
0 1 2 3 4 5
Pur
ity
Flo
w r
ate
(km
ol/h
)
Time (h)
Flow rate
Purity
0.99
0.992
0.994
0.996
0.998
1
80
90
100
110
120
130
0 1 2 3 4 5
Pur
ity
Flo
w r
ate
(km
ol/h
)
Time (h)
Flow rate
Purity
FIGURE 21.4
Dynamic response of the product flow rate and purity to the increase and decrease of fresh toluene feed rate.
798 CHAPTER 21 CASE STUDIES
G+ i-B$MEME+ i-B$DEDE+ i-B$ TE
(21.8)
Several alternative processes for the production of GTBE were presented in Figure 12.14. The reaction
takes place at moderate pressure, using p-toluene sulphonic acid as homogeneous catalyst or Amberlyst
15 as heterogeneous catalyst. Isobutene is separated from the reaction mixture as vapour in a flash or a
striping column. Unreacted glycerol and ME are separated from DE and TE mixture by extraction with
glycerol and recycled. Final purification of high ethers is achieved by vacuum distillation.
The important drawback of the processes described in Chapter 12 is the use of high-purity iso-
butene as a raw material, which is expensive and therefore seldom available as feedstock for glyc-
erol etherification process. Therefore, the design of the GTBE plant must take into account the
variability of this raw material. A key unit of the etherification plant is the extraction column, where
glycerol is used for ME separation by extraction. However, the binary interaction parameters of the
liquid activity model used to calculate the liquid–liquid equilibria are usually affected by experimen-
tal errors. Another uncertainty is related to the kinetic data used for designing the reactor, which
may be affected by modelling assumptions, experimental errors and catalyst deactivation. Last
but not least, the plant must process variable amounts of glycerol, as delivered by the biodiesel pro-
duction plant.
All GTBE production processes involve a reaction unit followed by a separation section and
two recycle streams containing isobutene and glycerol–ME mixture, respectively. Such reaction–
separation–recycle systems exhibit a non-linear behaviour (Bildea et al., 2000). In particular, small
changes of the feed rate, reaction kinetics or separation performance can lead to very large increase
in the recycle flow rate (snowball effect, Luyben, 1994). Therefore, the operating conditions must
be chosen and the plantwide control must be designed such that robustness with respect to disturbances
or uncertain design parameters is ensured.
This chapter will present the design and optimisation of the glycerol etherification plant taking into
account robustness with respect to the following disturbances and uncertain design parameters
(Table 21.4):
– Glycerol feed rate. The plant must be able to process variable amounts of glycerol, as delivered by an
upstream biodiesel production plant.
Table 21.4 Uncertain Parameters and Their Range of Variation
Parameter Nominal Value, p(0) Maximum Change, Dp
Glycerol feed rate, FG,0 (kmol/h) 2.25 0.5
Butane fraction in feed, yB,0 0.1 0.1
Catalyst activity, ’ 1 0.5
Reaction temperature, T (K) 363 10
Fraction of DE recovered in the extract stream, a 0.15 0.05
Fraction of TE recovered in the extract stream, b 0.15 0.05
79921.3 DESIGN AND ROBUST OPTIMISATION OF A PROCESS
– Raw material purity. High-purity isobutene is expensive and therefore seldom available as feedstock
for glycerol etherification process. The design of the GTBE plant should take into account the
presence of inert species (modelled here as isobutane) and therefore the uncertainty regarding the
purity of the raw material.
– Kinetic data used for designing the reactor, which may be affected by modelling assumptions,
experimental errors and catalyst deactivation.
– Reaction temperature.
– Errors affecting the binary interaction parameters, which lead to uncertainty of the calculated phase
equilibria (in particular, the fraction of DE and TE recovered in the extract stream of the extraction
column).
21.3.2 PLANT DESIGNA nominal capacity of 2.25 kmol/h processed glycerol, assumed to be the by-product of a
15,000 tonnes/year biodiesel plant, is considered. The raw materials are pure glycerol and isobutene
with significant amounts of an inert of similar volatility (10% isobutane, the raw material for isobutene
production).
The process flowsheet is presented in Figure 21.5 combines ideas taken from the previous works.
The reaction takes place in a CSTR. The reaction temperature and pressure are set to 363 K and
Reactor
EX
i-Butene
Glycerol
i-Butene / inert recycle
Glycerol, ethers and catalyst
0
1b
2
3c
Di-etherTri-ether
MIX-IB
C1 C2
0
4a
4b
Purge
1a
3a 3b
3d
FC
x
FC
r1
LC
2a
YCyB,4
Catalystmake-up
FIGURE 21.5
Flowsheet of the glycerol etherification plant (Vlad et al., 2013).
800 CHAPTER 21 CASE STUDIES
1.4 MPa, respectively, when the reaction mixture is liquid. The immiscibility of high ethers in glycerol
is exploited for separating the reactants from products, by using the fresh glycerol as an extraction sol-
vent (column EX). The raffinate stream ‘2a’ contains isobutene, inert,DE, TE and small quantities ofGandME and is sent to separation. Column C1 separates the isobutene and the isobutane. The column has
a total condenser and is operated at 0.5 MPa in order to keep the isobutene as liquid. This avoids the
disadvantage of using a compressor for recycling purpose. The entire amount of isobutene and isobu-
tane present in column feed is recovered in stream ‘3c’. A purge (stream ‘4b’) is necessary to remove
the inert from the plant. The stream ‘3d’ is mixed with fresh isobutene and fed to the reactor. Column
C2 separatesDE and TE from glycerol andME. The column has a total condenser and is operated under
vacuum (0.02 MPa) to avoid high temperature in the bottom of the column. The distillate ‘4a’ is the
plant product. The bottom stream ‘3b’, containing mainly G and ME, is mixed with the glycerol-rich
phase from the extraction column and fed to the reactor.
The plant was simulated using Aspen Plus. The properties of glycerol, isobutene and butane are
available from Aspen Plus database. The ethers’ properties were estimated by group contribution
methods. The behaviour of the liquid phase was described by the NRTL activity model. The interaction
parameters of pairs involving ethers and glycerol or isobutene were taken from Behr and Obendorf
(2003). The other interaction parameters were estimated using UNIFAC–Dortmund modified method.
The kinetic expressions used to calculate the reaction rates are presented in Table 21.5.
The economics of the process was investigated by generating several design alternatives, a selection
being presented in Table 21.6. The reactor volume V and the ratio between the reactor-inlet flow rates
r1¼F1b/F1a were considered as design decisions. The glycerol feed rate, which determines the plant
throughput, was fixed to 2.25 kmol/h. The purge composition was set to yB,4¼0.5. After setting these
values, the plant was designed using common heuristic rules, such as 99.5% recoveries in separation
units, R¼1.2 Rmin and N¼2 Nmin.
The capital cost includes the costs of reactor, extraction and distillation columns and heat ex-
changers, calculated by well-known relationships (Dimian, 2003). An M&S index of 1476.7 (year
2010) and a payback period of 3 years were considered. The utilities cost includes costs of cooling
water and steam. It should be remarked that the etherification plant can be integrated with a plant per-
forming the dehydrogenation of isobutane to isobutene. Because a typical dehydrogenation plant has
Table 21.5 Kinetic Parameters for Etherification of Glycerol with Isobutene Catalysed
by p-Toluene sulphonic Acid (Behr and Obendorf, 2003)
Reaction Pre-exponential Factors Activation Energies (kJ/kmol)
G + i-B��! ��k1
k�1ME k0,1¼3.04�108 L/min/mol 74.04
r1¼k1 �CG �Ci-B�k�1 �CME k0,�1¼3.69�1013 min�1 111.78
ME + i-B��! ��k2
k�2DE k0,2¼1.70�1011 L/min/mol 92.80
r2¼k2 �CME �Ci-B�k�2 �CDE k0,�1¼8.54�1014 min�1 118.06
DE+ i-B��! ��k3
k�3TE k0,3¼2.26�1010 L/min/mol 92.56
r3¼k3 �CDE �Ci-B�k�3 �CTE k0,�3¼6.35�1015 min�1 125.13
80121.3 DESIGN AND ROBUST OPTIMISATION OF A PROCESS
Table 21.6 Economic Evaluation of Several Design Alternatives
Design (1)
(2) (Base
Case) (3) (4) (5) (6) (7)
Design decisions
Reactor volume (m3) 2 4 4 10 10 15 15
r1¼F1b/F1a 1.01 0.6 2.12 0.76 1.94 0.59 1.90
Design results
Feed streams
Glycerol, FG,0 (kmol/h) 2.25 2.25 2.25 2.25 2.25 2.25 2.25
i-Butene (90% purity), FIB,0
(kmol/h)
5.91 6.09 6.12 6 6.2 5.98 6.22
Product (D+T)
Flow rate (kmol/h) 2.25 2.25 2.25 2.25 2.25 2.25 2.25
xDE 0.89 0.831 0.815 0.865 0.785 0.873 0.781
xTE 0.10 0.161 0.176 0.13 0.207 0.122 0.212
xME (103) 1 1 1 1 1 1 1
xG (103) 1.71 1.49 1.96 1.14 2.8 1.3 2.8
Purge
Flow rate (kmol/h) 1.17 1.18 1.22 1.2 1.2 1.2 1.24
Reactor
Installed cost (103 US$) 99.43 153.1 153.1 270.7 270.7 348.4 348.4
Extraction column (two trays)
Tray volume (m3) 0.473 0.44 0.20 0.37 0.32 0.42 0.3
Installed cost (103 US$) 47.34 45.64 29.16 41.10 38.10 44.18 37.10
C1
Number of trays 10 10 9 10 10 10 10
Reflux ratio 0.17 0.13 0.09 0.28 0.14 0.2 0.15
Diameter (m) 0.17 0.11 0.14 0.16 0.11 0.13 0.10
Reboiler duty (kW) 105.6 96.44 112.46 98.8 104.76 97.88 104.05
Condenser duty (kW) 34.5 17.23 39.43 15.26 24.09 12.97 21.84
Installed cost (103 US$) 105.98 87.58 107.95 91.20 95.23 86.83 92.84
C2
Number of trays 16 16 13 16 16 17 17
Reflux ratio 4.1 3.9 4.8 3.1 3.2 4.04 3.2
Diameter (m) 0.44 0.47 0.55 0.45 0.46 0.49 0.46
Reboiler duty (kW) 119.75 147.17 221.6 124.23 127.3 167.37 125.8
Condenser duty (kW) 190.73 214.13 298.4 192.9 199.8 236.18 198.15
Installed cost (103 US$) 205.95 226.54 272.92 209.58 213.96 245.11 215.71
Utilities’ cost (103 US$/
year)
70.43 75.41 104.64 68.82 72.04 81.93 71.25
TAC (103 US$/year) 238.07 259.36 304.39 284.95 287.45 336.63 311.89
802 CHAPTER 21 CASE STUDIES
large capacity (e.g. 300,000 tonnes/year in CATOFIN process), the cost of processing the purge stream
does not have an important contribution to the economy of the overall process (etherification+dehy-
drogenation). For this reason, the cost of the fresh C4 stream was not included in the total annual cost
(TAC), and the influence of the purge composition (which was set to 0.5) was neglected.
The smallest TAC (238.07�103 US$/year) was obtained for a base case employing a 2 m3 reactor,
two-stage extraction column and the ratio between reactor-inlet streams (r1¼F1b/F1a, related to
reactants ratio) set to r1¼1.01, the cost of utilities and vacuum distillation column C2 having the main
contributions. Decreasing the concentration of inert in the purge leads to a slightly smaller TAC.
However, this is achieved at the expense of a larger purge. The TAC increases for larger reactors,
because the more costly reactor is not compensated by a cheaper separation. However, the following
section will show that the design alternatives with small reactors have high sensitivity of the recycle
streams with respect to disturbances and uncertain parameters. This issue should be included in the
optimisation problem.
Although Aspen Plus offers facilities for solving constrained-optimisation problems, the construc-
tive non-linear dynamics technique that will be used as solution method is computationally intensive
and requires Jacobian matrices that are difficult to obtain from the Aspen Plus simulation. Therefore,
the following steps were taken:
– Derive a simple cost model to be used as objective function.
– Derive a simple reactor–separation–recycle model that can be easily solved in Matlab. This model
will be used to evaluate the recycle sensitivity with respect to operating conditions and design
parameter uncertainty.
– Formulate the constrained-optimisation problem.
– Use the constructive non-linear dynamics technique to find the optimal values of the design
decisions (solution of the optimisation problem).
– Develop rigorous steady-state simulation (Aspen Plus) for economic evaluation.
– Use rigorous dynamic simulation (Aspen Dynamics) to prove the robustness in the worst-case
scenario.
21.3.3 ROBUST OPTIMISATION21.3.3.1 The objective functionAs first step in solving the optimisation problem, the dependence of the TAC versus the design param-
eters V and r1 was correlated by the following objective function:
TAC 105US$� �¼ 2:42 + 1:79�10�3�V2 + 0:28�r21�0:04�V�r1 + 0:067�V�0:306�r1 (21.9)
The parity plot is presented in Figure 21.6. Each point represents one design alternative obtained for
certain values of the design parameters (V, r1). The TAC model agrees very well with the values
obtained by detailed calculations.
The results presented in Figure 21.6, together with the TAC cost model, indicate that from an eco-
nomic point of view, plants built around small reactors and low isobutene/glycerol ratio are better.
However, as we will show in this section, they have limited flexibility due to high sensitivity of the
internal flow rates with respect to production rate or uncertain parameters.
80321.3 DESIGN AND ROBUST OPTIMISATION OF A PROCESS
21.3.3.2 Reactor–separation–recycle modelThe notation used in this section will follow Figure 21.5. Fk, j will denote the molar flow rate of spe-
cies k in stream j, while 0, 1, 2, 3 and 4 will refer to fresh feed, reactor-inlet, reactor-outlet, and re-
cycle and product streams, respectively. For example, FG,1 stands for the flow rate of glycerol at the
reactor inlet.
The model of the plant considers a CSTR operated at a fixed temperature (90 ºC). The reactions
described earlier were considered, while the formation of isobutene oligomers and glycerol decompo-
sition were neglected because these reactions were not observed in homogeneous catalysis (Behr and
Obendorf, 2003). It is assumed that isobutene, glycerol and ME are recovered from the reactor-outlet
stream and recycled. Parameters a and b in the next equations account for the fractions of DE and
TE from the reactor outlet, which are found in the extract stream ‘3a’ and recycled. V, n and g are
the reaction volume, the matrix of stoichiometric coefficients and the purge fraction, respectively.
The factor ’ multiplying the reaction rates was introduced to account for uncertainty in the kinetic
parameters that can arise due to experimental errors or due to catalyst deactivation:
Fk,2� Fk,1 +V�Xi¼1,3
uk, i�’�ri !
¼ 0, k¼G, i-B,B,ME,DE,TE (21.10)
FG,1� FG,0 +FG,2ð Þ¼ 0, FG,4¼ 0 (21.11)
Fi-B,1� Fi-B,0 + 1� gð Þ�Fi-B,2ð Þ¼ 0, Fi-B,4� g�Fi-B,2¼ 0 (21.12)
2
2.5
3
3.5
2 2.5 3 3.5
TAC (105 $/year)
TA
C_m
od
el
(10
5 $
/ye
ar)
(2, 0.66)
(2, 1.01)
(3.71, 1.41)(4, 1.60)
(10, 1.35)
(4, 0.88)
(4, 0.61)
(6, 0.62)
(10, 0.76)
(10, 1.94)(4, 2.12)
(15, 1.90)(15, 1.20)
(15, 0.59)
(16.1, 1.49)
FIGURE 21.6
Parity plot of the TAC model (Vlad et al., 2013).
804 CHAPTER 21 CASE STUDIES
FB,1� FB,0 + 1� gð Þ�FB,2ð Þ¼ 0, FB,4� g�FB,2¼ 0 (21.13)
FME,1�FME,2¼ 0, FME,4¼ 0 (21.14)
FDE,1�a�FDE,2¼ 0, FDE,4� 1�að Þ�FDE,2¼ 0 (21.15)
FTE,1�b�FDE,2¼ 0, FTE,4� 1�bð Þ�FTE,2¼ 0 (21.16)
Fi-B,0
FB,0¼ 1�yB,0� �
yB,0(21.17)
FB,4�yB,4 Fi-B,4 +FB,4ð Þ¼ 0 (21.18)
Fi-B,1 +FB,1ð Þ� r1� FG,1 +FME,1 +FDE,1 +FTE,1ð Þ¼ 0 (21.19)
The concentrations necessary to calculate the reaction rates are obtained assuming ideal mixing:
ck,2¼ Fk,2XkVm,k�Fk,2
, k¼G, i-B,ME,DE,TE,B (21.20)
The values used for the molar volume Vm of glycerol, isobutene, ME, DE, TE and butane are (in m3/
kmol) 0.072, 0.094, 0.155, 0.143, 0.18 and 0.094, respectively.
The model of the glycerol etherification plant consists of 21 equations. After specifying the reactor
volume V, the feed composition yB,0 and the separation performance a and b, there are 24 model vari-
ables: 21 species flow rates, the purge composition yB,4, the purge fraction g and the ratio between
reactor-inlet flow rates r1. Therefore, three additional specifications are needed. The control structurepresented in Figure 21.5 fixes the glycerol feed rate FG,0, the purge composition yB,4 and the ratio r1.
21.3.3.3 Problem formulationReaction–separation–recycle systems can show high sensitivity of recycle flow rate with respect to
changes of the feed rate, reaction kinetics or separation performance (snowball effect; Luyben, 1994).
Figure 21.7 presents the dependence of recycle flow rates F3c and F1a (Figure 21.5) versus the glycerol
feed rate FG,0, obtained from the simplified model, for different values of the kinetic uncertainty ’ and
fixed values of the reactor volume, flow rates ratio and purge composition (V¼4 m3, r1¼1 and
yB,4¼0.5).
At the nominal operating point (FG,0¼2.25 kmol/h, ’¼1), the recycle flow rates are
F3c¼4.0 kmol/h and F1a¼6.5 kmol/h. These values were used to size the distillation column C-1
(reboiler and condenser duties, diameter, trays type, etc.) and the extraction column EX (stage
hold-up). The equipment has a limited flexibility, being able to withstand flow rate variations within
certain limits. For example, we assume that the capacities of the distillation column C-1 and extraction
column EX are limited to F3cmax¼10 kmol/h distillate and F1a
max¼10 kmol/h extract, respectively. These
feasibility limits should not be exceeded during operation, even if there are disturbances or the design
parameters are uncertain:
80521.3 DESIGN AND ROBUST OPTIMISATION OF A PROCESS
F3c�Fmax3c (21.21)
F1a�Fmax1a (21.22)
Results from Figure 21.7 show that the distillation column C1 easily withstands a 50% kinetic uncer-
tainty or a 0.5 kmol/h increase of the glycerol feed rate, when F3c<F3cmax. However, the combination of
these disturbances cannot be tolerated. The situation is worse when the glycerol–ether recycle F1a is
considered (Figure 21.7, right), as the operational limit is reached when only one disturbance (’¼0.5)
or a combination of smaller disturbances (’¼0.75, FG,0¼2.75 kmol/h) is introduced.
In the following, we will address the problem of finding the optimal design (minimum TAC) for
which the feasibility conditions are fulfilled even when the plant is disturbed or the design parameters
are uncertain.
The following short notation is introduced. The next equations represent the process model, and the
feasibility constraints, where x, p and q are the unknowns, the uncertain and the design parameters,
respectively. The uncertain parameters p are presented in Table 21.4. The reactor volume and the ratio
of reactor-inlet flows are chosen as decision variables q¼ (V, r1). The uncertain parameters p are de-
fined in terms of their nominal value p(0), the maximum expected change Dp and the scaled change d.The optimisation problem aims to find the optimal design defined by the objective function chosen as
the TAC relationship, achieving feasibility that is robust with respect to disturbances and uncertain
parameters:
f x, p, qð Þ¼ 0 (21.23)
~f x, p, qð Þ� 0 (21.24)
p¼ p 0ð Þ +Dp�d, d2 �1,1ð Þ (21.25)
minq
h x, p, qð Þ (21.26)
0
5
10
15
20
0.75
1.0
V = 4 m3
r1 = 1yB,4 = 0.5
0
5
10
15
20
1.75 2 2.25 2.5 2.75 3 1.75 2 2.25 2.5 2.75 3
FG,0 (kmol/h)FG,0 (kmol/h)
F3
c (
km
ol/
h)
F1
a (
km
ol/
h)
j = 0.5j = 0.5
0.75
1.0
V = 4 m3
r1 = 1yB,4 = 0.5
FIGURE 21.7
Dependence of recycle flow rates versus the glycerol feed rate.
806 CHAPTER 21 CASE STUDIES
21.3.3.4 Solution methodThe constructive non-linear dynamics method (Monnigmann and Marquardt, 2003; Marquardt and
Monnigmann, 2005) allows finding the economical optimum while ensuring that the feasibility bound-
ary is not crossed when design parameters change or are uncertain. By solving the extended system
together with the additional relationships, the method calculates the shortest distance r from the nom-
inal operating conditions d(0)¼0 to the feasibility constraint and the direction u along which this dis-
tance is measured:
f Tx~fT
x
h i�v¼ 0 (21.27)
u� f Td~fT
d
h i�v¼ 0 (21.28)
vTv�1¼ 0 (21.29)
d 0ð Þ � d +r� u
uj jj j� �
¼ 0 (21.30)
In these equations, f x, ~f x are Jacobian matrices with respect to unknowns x, f d, ~f d are Jacobian
matrices with respect to uncertain parameters d and v is an auxiliary vector having the same dimension
as x.The uncertainty is overestimated by a ball of radius
ffiffiffiffiffinpp
, where np is the number of uncertain pa-
rameters. Regardless of the actual values of the uncertain parameters in the robustness box, the fulfil-
ment of the feasibility constraint is guaranteed by enforcing the distance r to be larger than the radius ofthe ball:
r� ffiffiffiffiffinpp
(21.31)
The above equations define the mathematical formulation of the robust optimisation problem.
21.3.4 RESULTS21.3.4.1 Robust, optimal designTaking into account the uncertainties presented in Table 21.4, the result of the robust optimisation
is V¼16.1 m3, r1¼1.49. Compared to the base case, the reactor is four times larger and requires a higher
excess of co-reactant. The final design has a TAC of 3.21 �105 US$/year, which is almost 50% more ex-
pensive than the cheapest alternative from Table 21.6. This is the price that has to be paid to achieve
robustness.
At the optimum, both constraints were active. Using these specifications, the full plant was designed
and simulated in Aspen Plus. Details are given in Table 21.7, while Table 21.8 presents the stream
report. Relaxing the constraint on F1a to F1amax¼12.5 kmol/h, it becomes inactive, the optimal design
being V¼15.0 m3, r1¼1.2, for a TAC of 3.14 �105 US$/year.We emphasise the effectiveness of using a simple reactor–separation–recycle model of the plant
during the robust optimisation procedure. Application of the method to a rigorous model of a whole
80721.3 DESIGN AND ROBUST OPTIMISATION OF A PROCESS
chemical plant (such as glycerol etherification) is very computationally demanding, due to the large
number of equations involved. However, although the separation units are the origin of most equations
(mass and energy balance and physical properties), their input–output behaviour is much simpler be-
cause they comply with the given performance specifications. In this context, using the simple reactor–
separation–recycle model and replacing the cost calculation by a cost model are very effective, allow-
ing solution of the optimisation problem in a short time.
21.3.4.2 Dynamics and controlFor a rigorous check of the optimal process robustness, a dynamic model was developed using Aspen
Dynamics. Figure 21.8 shows the full flowsheet, together with the control structure (Vlad
et al., 2013).
Figure 21.9 presents results of dynamic simulation (Vlad et al., 2013). The simulation starts from
the nominal steady state, which is maintained for the first 2 h. Then, large disturbances were intro-
duced. At time t¼2 h, the glycerol feed rate was increased from 2.25 to 2.75 kmol/h. The reactor-inlet
Table 21.7 Economic Evaluation of the Optimal, Robust Design
Design Optimal, Robust Design
Reactor volume (m3) 16.1
r1¼F1b/F1a 1.49
Reactor
Installed cost (103 US$) 364.13
Extraction column (two trays)
Tray volume (m3) 0.3
Installed cost (103 US$) 36.42
C1
Number of trays 10
Reflux ratio 0.18
Diameter (m) 0.12
Reboiler duty (kW) 100.95
Condenser duty (kW) 17.58
Installed cost (103 US$) 90.46
C2
Number of trays 17
Reflux ratio 3.4
Diameter (m) 0.47
Reboiler duty (kW) 136.3
Condenser duty (kW) 207.02
Installed cost (103 US$) 223.46
Utilities’ cost (103 US$/year) 73.95
TAC (103 US$/year) 321.01
808 CHAPTER 21 CASE STUDIES
Table 21.8 Stream Report
Stream Name G0 IB0 1a 1b 2 2a 3a 3b 3c 4a 4b
Flow (kmol/h) 2.25 6.15 5.22 7.78 8.08 5.52 4.81 0.40 2.85 2.25 1.23
Flow (kg/h) 207.5 346.8 598.8 439.6 1038.4 697.4 548.5 50.3 163.0 481.7 70.3
Temp. (K) 363 293 377.5 297.2 363 346.1 375.8 479.9 312.6 431 312.6
Pressure (MPa) 1.4 1.4 1.4 1.4 1.4 0.55 1.4 0.02 0.5 0.02 0.5
Flow (kmol/h)
G 2.25 0 2.99 Trace 0.74 0.19 2.80 0.19 Trace 0.005 Trace
IB 0 5.5 0.12 6.35 1.55 1.43 0.12 Trace 1.42 0.004 0.61
B 0 0.61 0.12 1.43 1.55 1.43 0.12 Trace 1.42 Trace 0.61
ME 0 0 1.74 Trace 1.74 0.21 1.53 0.20 Trace 0.002 Trace
DE 0 0 0.19 Trace 2.02 1.83 0.18 0.009 Trace 1.82 Trace
TE 0 0 0.04 0 0.46 0.42 0.04 Trace Trace 0.42 Trace
LC
i-Butene
FC
LCLC
Glycerol
IB0
1b
Glycerol and ethers recycle
FC
x r1
FG,0
TC
i-Butene recycle
CC
FC
PC
TC
LC
TC
PC
LC
4a
LC
TC
4b3d
3c
G0
3a
1a
Purge
GTBE
EX C1C2
3b
2a
2
FIGURE 21.8
Plant-wide control structure.
flow rate F1a increases, more isobutene is withdrawn from the vessel V to keep reactants ratio at the
specified value r1¼1.49, and more fresh isobutene is fed to the process (FIB,0). Because the dynamics
of the plant throughput is slow, the theoretical value F4a¼2.75 kmol/h ethers is reached after 10 h of
operation. A new steady state is established. At time t¼20 h, the reaction temperature is decreased
from 90 to 80 ºC. After a short transient period, a new steady state is reached. A rather big disturbance
is introduced at time t¼40 h, when all the pre-exponential factors in kinetic expression are reduced to
half, to simulate the reduction of the catalyst activity. The flow rates F3c (C1 distillate) and F1a (fresh
glycerol+ recycled glycerol and ethers) settle to higher values. The concentrations of glycerol andMEin the product stream 4a increase, but the control system is able to bring them to the specified values.
Finally, at t¼60 h, the concentration of inert in feed is increased from yB,0¼0.1 to yB,0¼0.2. This is
compensated by larger feed and purge (FIB,0 and F4b, respectively) with small influence on other pro-
cess variables. We note that the fraction of DE and TE in the extract stream drop from the initial values
of �0.15 to around 0.1, which has a favourable effect on robustness. Therefore, during the entire sim-
ulation, the recycle streams F3c and F1a stay well below the specified feasibility boundaries.
From the study presented in this section, we conclude that the plant achieves stable operation both at
the nominal operating point and when large disturbances are introduced. The concentration of glycerol
in the product streams is less than 0.3%mol, corresponding to 0.15%wt. This means that when 5%wt.
GTBE mixture is added to biodiesel, the final glycerol contents is below the 0.02% specification of the
ASTM 6571 standard.
21.3.5 CONCLUSIONSThis section presented the optimal design of glycerol etherification plant. The considered plant has a
processing capacity of 2.25 kmol/h glycerol and uses isobutene with large amounts of inert impurity
(10% molar) as raw material. Compared to the base case, the optimal solution requires a much larger
reaction volume and higher isobutene/glycerol reactor-inlet ratio to ensure robust feasibility with re-
spect to uncertainties. Based on the experience gained by solving this case study, we recommend the
following efficient procedure for robust, optimal design of complex chemical plants:
0
2
4
6
8
10
t (h) t (h)
F (
km
ol/h
)
F4b
F4a
F3c
FIB,0
F1a
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 0 10 20 30 40 50 60 70 80
x
0
0.001
0.002
0.003
0.004
0.005
x
xD,4a
xT,4a
yB,4b
xG,4a
xM,4a
FIGURE 21.9
Dynamic simulation results.
81121.3 DESIGN AND ROBUST OPTIMISATION OF A PROCESS
1. Starting from the plant flowsheet, recognise the design decisions and generate several design
alternatives.
2a. Identify the feasibility constraints, uncertain design and operating parameters.
2b. Perform the economic evaluation of the alternatives and develop a simple model relating the
economic objective to the design decisions.
2c. Develop a simplified reactor–separation–recycle model of the plant.
3. Apply a robust optimisation method (e.g. constructive non-linear dynamics) using the models 2a,
2b and 2c.
4. Check the robustness by rigorous dynamic simulation.
21.3.6 SUGGESTIONS FOR FOLLOW-UP PROJECTS1. Assess the feasibility of glycerol (G) etherification with tert-butyl alcohol (TBA) as a co-reactant
and ion-exchange resins as catalyst, according to the following reactions (the kinetic constants are
given in Table 21.9):
Main reactions:
G+TBA$ME +H2O
ME+ TBA$DE+H2O
DE+ TBA$TE+H2O
Secondary reaction (dehydration of TBA to i-butene):
TBA! i-B +H2O
2. Assess the feasibility of glycerol (G) etherification with isobutene and TBA as a co-reactant, using
catalytic distillation as process intensification technique.
21.4 DESIGN AND SIMULATION OF A REACTIVE DWC FOR BIODIESELPRODUCTIONProcess intensification technologies for biodiesel production play an important role in the quest of
driving the investment and operating costs to lower values. Among them, reactive separations and
dividing-wall column (DWC) technology stand out as very promising technologies to achieve these
Table 21.9 Equilibrium Constants and Rate Constants (Kiatkittipong et al., 2011)
Equilibrium Constant Rate Constant (mol/s/kg)
Keq1¼exp(2.581�754.8/T) k1¼exp(17,342�6835/T)
Keq2¼exp(1.228�942.1/T) k2¼exp(26,953�10,382/T)
Keq3¼exp(1.779�2212/T) k3¼exp(26,953�10,382/T)
– k4¼exp(23.358�12,480/T)
812 CHAPTER 21 CASE STUDIES
goals (Kiss, 2013, 2014a,b). Moreover, the cost margin of biodiesel (a mixture of fatty esters) can be
much improved when low-cost rawmaterials (e.g. waste vegetable oil or animal fat) with high free fatty
acid (FFA) content are used instead of virgin oils. The following case study briefly illustrates the op-
timal design and simulation of an R-DWC for biodiesel production.
Conventional biodiesel processes are plagued by the drawbacks of using homogeneous catalysts
and the high-energy requirements. Reactive separation processes such as reactive distillation and re-
active absorption were proposed for the biodiesel production (Kiss and Bildea, 2012). Most of these
processes make use of solid acid/base catalysts, thus eliminating all conventional catalyst-related op-
erations, improving efficiency and reducing the energy requirements for the biodiesel production. The
problem of these reactive separation systems is that a stoichiometric ratio of reactants is required in
order to allow complete conversion of the fatty raw materials and production of two high-purity prod-
ucts (e.g. water by-product as top distillate and FAME as bottom product). However, maintaining the
stoichiometric ratio in practice can be difficult especially when the composition of the fatty acids feed
is not continuously monitored or not known. To solve this problem and still maintain all the advantages
of RD, we developed a novel design based on a reactive dividing-wall column (R-DWC). This allows
the use of a slight excess of methanol (�15% or higher) while still delivering pure products – methanol
as top distillate that can be recycled, water by-product as side stream and FAME as bottom product.
21.4.1 PROCESS OPTIMISATION STRATEGYSimulated annealing (SA) is used hereafter as an optimisation strategy, but othermethods are also possible.
SA mimics the thermodynamic process of cooling of molten metals to attain the lowest free energy state.
Starting with an initial solution, the algorithm performs a stochastic partial search of the space defined for
decision variables. In minimisation problems, uphill moves are occasionally accepted with a probability
controlled by the parameter called annealing temperature (TSA). The probability of acceptance of uphill
moves decreases as TSA decreases. At high TSA, the search is almost random, while at low TSA, the searchbecomes selective where good moves are favoured. The core of this algorithm is the Metropolis criterion
that is used to accept or reject uphill movements with an acceptance probability given by
M TSAð Þ¼ min 1,exp�DfTSA
� �� (21.32)
where Df is the change in objective function value from the current point to the new point.
The objective function is evaluated at the trial point, and its value is compared to the objective value
at the starting/current point. The Metropolis criterion is used to accept or reject the trial point. If this
trial point is accepted, the algorithm continues the search using that point. Otherwise, another trial point
is generated within the neighbourhood of the starting/current point. A fall in TSA is imposed upon the
system using a proper cooling schedule. Thus, as TSA declines, uphill moves are less likely to be ac-
cepted and SA focuses on the most promising area for optimisation. These iterative steps are performed
until the specified stopping criterion is satisfied. The random numbers can be uniformly distributed in
the interval [0, 1]. If the condition rand<M(TSA) is fulfilled, then the trial point is being accepted. Oth-erwise, the starting/current point is used to start the next step. The temperature TSA can be considered acontrol parameter. The initial temperature (Ti) is related to the standard deviation of the random per-
turbation, and the final temperature (Tf) with the order of magnitude of the desired accuracy gives the
location of the optimum.
81321.4 DESIGN AND SIMULATION OF A REACTIVE DWC
In order to optimise the complex R-DWC, we use the SA implementation in Matlab. The SA pa-
rameters were tuned using several short tests in order to improve the efficiency of the stochastic
method, while the initial point of SA was created randomly in the feasible region. The values of
the key parameters used in the SA are annealing function (Boltzmann), re-annealing interval (100),
temperature update (linear) and initial temperature (100).
Specifically, for process design of complex separation schemes, the minimisation of the heat duty of
the distillation column is the optimisation target – since up to 80% of the TACs are associated with
energy requirements even for complex distillation columns. Consequently, the use of the heat duty
is always a good approximation of the TAC. In the R-DWC, the optimisation problem for the mini-
misation of the reboiler heat duty is defined as
Min Qð Þ¼ f T,NR,NDWC,D,FSIDESTR,V,NSIDESTR,NRECYCLE,rV,rLð Þsubject to y
!m� x
!m
(21.33)
where T is the temperature in the FFA heater, NR is the number of reactive stages, NDWC is the total
number of stages in the DWC,D is the distillate rate, FSIDESTR is the side-stream product flow rate, V is
the boilup ratio, NSIDESTR and NRECYCLE are the side-stream product and recycle location, rL and rV are
the liquid and vapour split, while ym and xm are vectors of obtained and required purities for the mproducts, respectively.
The optimal design of the R-DWC was found by using SA as an optimisation method implemented
in Matlab and coupled with Aspen Plus simulations. Figure 21.10 shows the connection of MathWorks
Matlab with AspenTech Aspen Plus via MS Excel, including the flow of data between these programs
(Kiss et al., 2012). During the evolution of the SA, the vector values of decision variables (Vx) are sent
from Matlab to Microsoft Excel using DDE (dynamic data exchange) by COM technology. These
values are attributed in Excel to the corresponding process variables (Vp) and then sent to Aspen Plus
by a similar interface. Note that using the COM technology, it is possible to add code such that the
applications behave as an Object Linking and Embedding (OLE) automation server. After running
the rigorous simulation, Aspen Plus returns toMS Excel the vector of results (Vr). Finally, Excel returns
the objective function (FOB) value to Matlab for the SA procedure.
21.4.2 OPTIMAL PROCESS DESIGNThe integrated RD process was designed according to the process synthesis methods for reactive sep-
arations, as described in specialised monographs (Sundmacher and Kienle, 2003; Schmidt-Traub and
Gorak, 2006; Luyben and Yu, 2008). Rigorous simulations embedding experimental results were
performed using Aspen Plus. The RD column was simulated using the rigorous RADFRAC unit
Vx
FOB
Vp
Vr
Microsoft ExcelMathWorks Matlab
Optimization Toolbox (SA)AspenTech Aspen Plus
FIGURE 21.10
Connection of MathWorks Matlab with AspenTech Aspen Plus via MS Excel.
814 CHAPTER 21 CASE STUDIES
and explicitly considering three-phase balances. Phase splitting must be accounted for, as free water
phase can deactivate the solid acid catalyst. Nevertheless, as later revealed by the composition profiles,
the molar fraction of water in the liquid phase does not exceed 0.1 on the reactive stages – so catalyst
deactivation does not occur here, under these process conditions.
The physical properties required for the simulation and the binary interaction parameters for the
methanol–water and acid–ester pairs were available in the Aspen Plus database of pure components,
while the other interaction parameters were estimated using the UNIFAC–Dortmund modified group
contribution method. Similar results are obtained using other state-of-the-art estimation methods, such
as UNIFAC andUNIFAC–Lyngbymodified. The fatty components were conveniently lumped into one
fatty acid and its fatty ester according to R-COOH+CH3OH$R-COO-CH3+H2O. Dodecanoic (lau-
ric) acid/ester was selected as lumped component due to the availability of experimental results and
VLLE parameters for this system (Kiss and Bildea, 2012). The assumption of lumping components is
very reasonable since fatty acids and their corresponding fatty esters have similar properties. In this
work, sulphated zirconia is considered as solid acid catalyst, since kinetic data for the esterification with
methanol are also available from previous work.
Figure 21.11 shows the proposed flowsheet (Kiss et al., 2012). The conceptual design of the process
is based on an R-DWC that integrates the reaction and separation steps into a single operating unit. By
combining reaction and separation, one can shift the reaction equilibrium towards products formation
by continuous removal of reaction products, instead of using an excess of reactant. Since methanol and
water are much more volatile than the fatty ester and acid, these will separate easily as top distillate and
side stream. High conversion of the reactants is achieved, with the productivity of the RD unit exceed-
ing 20 kg fatty ester/kg catalyst/h and the purity specifications over 99.9%wt for the final biodiesel
product (FAME stream).
ACID F-ACID
ALCO1
F-ALCO
SIDESTR
WATER
RECYCLE
BTM FAME
TOPALC
ALCO
HEX1
HEX2
DEC
COOLER
DWC
B2
NL
NV rV
rL
NR
N
N-1
1
2 MeOH >90%wt
>99.0%wt
>99.9%wt1250 kg/h
MeOH >97%wtH2O <3% wt
MeOH : Acid ratio > 1.15
MIX
FIGURE 21.11
Flow sheet and topology of reactive DWC for FAME production: N, number of stages; NR, number of reactive
stages; NL, liquid split stage; NV, vapour split stage.
81521.4 DESIGN AND SIMULATION OF A REACTIVE DWC
Figure 21.12 shows the liquid–vapour composition, as well as the temperature profile along the
R-DWC (Kiss et al., 2012). Similar profiles were obtained also for the alternative designs. The con-
centration of the fatty acid and methanol is high at the top of the column, while the fatty ester concen-
tration increases from the top to bottom. Therefore, the water accumulated in the middle zone is
removed as side stream, while FAME is delivered as high-purity bottom product.
The tray arrangements and the most important design variables for the base case and the optimal
R-DWC structures are given in Table 21.10 (Kiss et al., 2012). Remarkably, energy savings of up to
25% can be achieved, just by properly tuning the design of the R-DWC. It can be seen that the total
number of stages in the optimised DWCs is higher in comparison to the base case, and consequently,
the number of reactive stages is increased. An important point is that the optimised designs also comply
with all the restrictions on the purities and recoveries of the products. Consequently, an important issue
is reviewing the feasibility to obtain the purity of 99.9%wt for FAME in all these reactive schemes. The
results shown in Table 21.10 clearly indicate that it is indeed feasible to obtain biodiesel of very high
purity in all schemes.
The improved alternatives (S2–S7) were generated by the SA algorithm previously described. The
SAmethod is stochastically by nature, so it has twowell-known advantages: It can be readily connected
to highly sophisticated simulators – such as Aspen Plus – and it converges towards a global optimum as
computing time approaches infinity. In practice, it finds the global optimality very efficiently. The evo-
lution course of the best function value for the SA is conveniently shown in Figure 21.13, while
Table 21.10 provides the best solution (S7) along with other five configurations (S2–S6) evaluated
every 1000 iterations. It can be seen that a great number of the function values are jumped off as
the optimisation method reduces the best function value.
Notably, the design with the greatest savings in energy requirements (Case S7) shows the largest
total number of stages and reactive stages as compared to the base case. It can be seen that the total
number of stages and reactive stages are 140% and 162.5% higher than those of the base case (Case S1).
These results are used to conduct a study on the total annual cost of operating these systems in order to
see the impact of the equipment cost and utilities’ cost on the total annual cost. In this way, the designer
could make decisions about the potential energy savings that the equipment can have, without a high
increment in the total annual cost due to the increase in the number of stages.
0
4
8
12
16
X (-) Y (-)
Tra
y
MeOH
MeOH
Acid H2O
FAME
FAME
0
4
8
12
16
Tra
y
MeOHH2O
Acid FAME
MeOH
H2O
0
4
8
12
160 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 300 350 400 450 500 550 600
Temperature (K)
Tra
y Main column side
RD side
FIGURE 21.12
L–V molar composition and temperature profiles along the R-DWC (base case).
816 CHAPTER 21 CASE STUDIES
Table 21.10 Design Parameters for the Reactive DWC: Base Case Versus Designs Found by SA
S1 S2 S3 S4 S5 S6 S7
Column topology
Number of stages, N 15 20 39 33 34 35 36
Number of reactive stages, NR 8 11 24 20 21 19 21
Liquid split stage, NL 3 5 8 7 7 8 8
Vapour split stage, NV 12 16 32 27 28 27 29
Side-draw stage, N1 8 12 30 18 18 17 13
Organic phase-return stage, N2 3 9 28 7 25 25 26
Feed streams
Acid flow rate (kmol/h) 5.83 5.83 5.83 5.83 5.83 5.83 5.83
Temperature HEX1 (K) 418.1 385.6 373.2 410.4 428.1 423.2 418.9
ALCO flow rate (kmol/h) 6.8 6.997 6.997 6.997 6.997 6.997 6.997
Temperature HEX2 (K) 340.6 338.7 338.7 338.7 338.7 338.7 338.7
Specifications
Distillate rate (kg/h) 33 35.34 39.06 34.71 37.12 37.45 37.12
Boilup ratio (kg/kg) 5 4.85 4.89 4.43 3.96 3.75 3.39
Side-draw flow rate (kg/h) 107 118.23 107.63 113.49 122.74 120.15 122.04
Liquid flow rate, L1 (kmol/h) 2.6 3.14 2.23 1.32 1.37 1.50 1.29
Vapour flow rate, V1 (kmol/h) 2.6 3.25 3.96 1.53 1.90 1.94 2.14
Product streams
Water flow rate (kmol/h) 5.70 5.82 5.66 5.82 5.75 5.70 5.67
Water purity (%wt) 99.10 97.96 98.68 97.34 98.34 97.73 96.99
FAME flow rate (kmol/h) 5.83 5.83 5.83 5.83 5.83 5.83 5.83
FAME purity (%wt) 100.00 100.00 100.00 99.97 99.94 99.96 99.96
Recycle stream
TOPALC flow rate (kmol/h) 1.10 1.10 1.12 1.08 1.12 1.10 1.07
TOPALC purity (%wt) 94.00 99.65 92.28 99.69 96.36 93.73 92.39
Continued
Table 21.10 Design Parameters for the Reactive DWC: Base Case Versus Designs Found by SA—cont’d
S1 S2 S3 S4 S5 S6 S7
Energy requirements
HEX1 duty (kW) 95.30 68.27 58.34 88.66 103.84 99.57 95.88
HEX2 duty (kW) 9.69 9.54 9.54 9.54 9.54 9.54 9.54
Reboiler duty (kW) 405.59 396.48 399.20 361.64 323.39 306.23 277.20
Condenser duty (kW) �278.11 �246.60 �239.43 �232.21 �209.11 �187.79 �155.18Cooler duty (kW) �205.57 �200.71 �200.71 �200.72 �200.72 �200.72 �200.72Total heating duty (kW) 510.58 474.29 467.08 459.84 436.77 415.34 382.62
Total cooling duty (kW) �483.68 �447.31 �440.14 �432.93 �409.84 �388.51 �355.89Key performance indicators
Energy requirements (kw h/tonne FAME) 408.46 379.43 373.67 367.87 349.42 332.27 306.10
Energy savings (%) 0.00 7.11 8.52 9.94 14.46 18.65 25.06
Total CO2 emission (tonnes/year) 1405 1291 1277 1238 1165 1105 1012
Economic evaluation
Annual operating cost (k$/year) 149.79 147.85 136.61 135.04 123.38 117.24 107.71
Capital cost (k$) 171.60 214.17 395.36 303.21 331.79 324.52 357.06
Total annual cost (k$/year) 184.11 190.68 215.68 195.68 189.74 182.15 179.12
Note: S1, base case design; S2–S6, intermediate designs found by SA; S7, optimal design.
21.4.3 RE-OPTIMISED PROCESS DESIGNSignificant savings in the operating and capital investment cost can be achieved by combining RD and
DWC technologies, but the high degree of integration can raise more difficulties in process control.
Therefore, having an optimal design is not always sufficient since the problem of dynamics and process
control is just as important as the optimal process design. At a first glance, the control of an R-DWC
may not appear to be an issue as several efficient control structures are available for DWC (Kiss, 2013).
Nonetheless, DWC units are designed for ternary separations, and, therefore, the already proposed con-
trol structures are applicable to multi-component separations without reaction. Moreover, in case of
reactive distillation, the performance of the control structure strongly depends on the process design
and on the type of chemical reactions that occur in the column.
The R-DWC unit studied here is designed for a quaternary system – two products and two reactants,
with one of the reactants in excess. A critical aspect of the process is to ensure the full conversion of the
FFA by having an excess of methanol and, therefore, avoiding that unreacted FFA becomes an impurity
in the bottom stream. However, the excess methanol can become an impurity in the side stream. The
reason is that the prefractionator of the R-DWC is operated with multiple feeds (with lightest compo-
nent fed at the lower part) and therefore unable to perform the sharp split between the light and the
heavy components of the system like as is typically the case in a standard DWC.
Rigorous dynamic simulations were performed in order to understand the dynamic behaviour of the
process. The initial results of the dynamic simulations pointed out that under different scenarios, the
excess methanol becomes an impurity in the side stream. This problem arises from the fact that meth-
anol is the light key component of the system and, moreover, is fed at the bottom of the column pre-
fractionator. An effective solution to overcome this problem is to feed the alcohol stream as vapour
instead of saturated liquid and to increase the acid inlet temperature. Therefore, the design is
re-optimised and an efficient control strategy is proposed.
200
250
300
350
400
450
500
550
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10,000
Number of iterations
DW
C h
eat d
uty
(kW
)
S2 S3 S4 S5 S6 S7
FIGURE 21.13
Evolution of best function value during the course of SA optimisation.
81921.4 DESIGN AND SIMULATION OF A REACTIVE DWC
Figure 21.14 (left) shows the updated R-DWC design, where FAME is produced as pure bottom
product and water as side stream, while the methanol excess is recovered as top distillate and recycled
(Ignat and Kiss, 2013). The alcohol-to-acid ratio plays a crucial role in the design and operation of the
integrated R-DWC unit. All products can be obtained at high purity only in a narrow optimal range of
the alcohol-to-acid ratio. At smaller ratio, methanol is fully converted but the excess of FFA becomes
an impurity in the FAME stream, while at high ratio, the FFA are fully converted and the excess of
methanol becomes an impurity in the side stream (Kiss and Bildea, 2012).
The initial optimal R-DWC design was modified and optimised in terms of minimal energy require-
ments, this time using the sequential quadratic programming (SQP) method from Aspen Plus. Based on
a solid theoretical and computational foundation, the SQP method has become the most successful
method for solving non-linearly constrained-optimisation problems. The SQP optimisation method
and the effective sensitivity analysis tool from Aspen Plus were employed in the R-DWC optimisation
procedure illustrated in Figure 21.14 (right) (Ignat and Kiss, 2013). The objective of the optimisation is
to minimise the total reboiler duty required, as follows:
Min Qð Þ¼ f TFFA,NT,NRT,NF,NSS,FSS,NDWS,NDWC,RxR,V,RR,rV,rLð Þsubject to y
!m� x
!m
(21.34)
Methanol
Water
F-ACID
rL
rV
1
1 7
18
22
19
6
FAME
(biodiesel)
4
12
RD
CDWC
16
DEC
Recycle-ACID
Side stream
ACID
F-ALCOALCO
R-DWC
9
InitialisationTFFA, NT, NRT, NF, NSS, FSS, NDWS, NDWC
NT, NRT, NF, NSS, NDWC/DWS, V, RR, RxR
V, RR, RxR, rL, rV
Change
Optimal designConverged R-DWC profiles
and stage requirements
No
Adjust rL and rV
min QR
min N(RR+1)
Yes
No
Yes
FIGURE 21.14
Reactive DWC used for the FAME synthesis from free fatty acids and methanol (left). Procedure for the optimal
design of a reactive (DWC) (right).
820 CHAPTER 21 CASE STUDIES
where the optimisation parameters used here are temperature of the inlet FFA stream (TFFA), totalnumber of stages (NT), total number of reactive stages (NRT), feed location (NF) of each reactant,
side-stream product flow rate (FSS), side-draw stage (NSS), wall size (NDWS) and location (NDWC),
alcohol-to-acid reactants ratio (RxR), boilup rate (V), reflux ratio (RR) and liquid and vapour split
(rL and rV), while ym and xm are the vectors of the obtained and required purities for the m products.
Note that in order to determine the optimal ratio between the energy cost and the number of stages, an
additional objective function was used, Min NT (RR+1), which approximates very well the minimum
of the TAC of a conventional distillation column (Ignat and Kiss, 2013).
Figure 21.15 plots the temperature and composition profiles along the R-DWC unit, while the key
parameters of the optimal design are presented in Table 21.11 and the mass balance is provided
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Stage (-)
Stage (-)
Mas
s fr
actio
n (-
)
FAME
Water
FFA
Methanol
40
60
80
100
120
140
160
180
200
220
240
260
280
0 2 4 6 8 10 12 14 16 18 20 22
0 2 4 6 8 10 12 14 16 18 20 22
Tem
pera
ture
(�C
)
Prefractionator (PF)
Main column
FIGURE 21.15
Temperature and composition profiles along the reactive DWC.
82121.4 DESIGN AND SIMULATION OF A REACTIVE DWC
in Table 21.12 (Ignat and Kiss, 2013). Compared to the optimal design by SA presented earlier, there
are some key differences in the optimal R-DWC design that must be noted. In addition to the change
of the feed streams’ thermal state, a key finding of this study is that it is imperative to use a vapour feed
of alcohol in order to reach the product specifications. Moreover, while requiring a 39% fewer number
of stages and 57% fewer reactive stages, water and methanol are obtained as high-purity products
(99.8%), with just a minor trade-off – only a 1.5% increase in the total heat duty.
21.4.4 DYNAMICS AND PROCESS CONTROLThe economic benefits predicted by the optimal design should also be achievable in the practical oper-
ation of the plant. Therefore, the controllability of the process is just as important as the savings in the
capital and operating costs. As composition analysers are rather expensive and require more mainte-
nance, their use should be minimal or avoided altogether. In practice, inferential temperature measure-
ments are preferred over direct composition measurements. For the selection of the trays to control the
Table 21.11 Design Parameters of an Optimal Reactive (DWC)
Design Parameters Value Unit
Flow rate of acid feed stream 1168.2 kg/h
Flow rate of ALCO feed stream 198.1 kg/h
Temperature of acid feed stream 160 �CTemperature of ALCO feed stream 69 �CPressure of feed stream 1.2 bar
Operating pressure 1 bar
Column diameter 1.1 m
Number of stages prefractionator side 12 –
Number of reactive stages prefractionator side 9 –
Total number of stages DWC 22 –
Acid feed stage prefractionator 4 –
ALCO feed stage prefractionator 9 –
Side-stream withdrawal stage 16 –
Organic phase-return stage 7 –
Wall position (from/to stage) 7–18 –
Liquid split ratio (rL) 0.07 kg/kg
Vapour split ratio (rV) 0.26 kg/kg
Methanol product purity 99.80/99.60 %wt/%mol
Water product purity 99.80/99.90 %wt/%mol
FAME product purity 99.99/99.99 %wt/%mol
HEX 1 duty 108 kW
HEX 2 duty 69 kW
Reboiler duty 212 kW
Condenser duty �154 kW
Total heating duty 389 kW
822 CHAPTER 21 CASE STUDIES
temperature, several methods based on steady-state calculations are available – namely, the slope cri-
terion, invariant temperature criterion, sensitivity criterion and singular value decomposition (SVD)
method. Hereafter, the sensitivity analysis and SVD criteria were used. The steady-state gain matrix
(Kp) is expressed (decomposed) as a product of three matrices: a U matrix, a diagonal matrix s and a
VTmatrix. The highest value in each column of theUmatrix indicates the tray on which the temperature
should be controlled:
Kp¼UsVT (21.35)
A small change is made in one of the manipulated variables (V, L, S and RxR) while keeping the othersconstant. The steady-state gains for each tray are calculated by dividing the change in each tray tem-
perature by the change in the manipulated variable. For a DWC, there are two steady-state gain ma-
trices, one for the main column and one for the prefractionator. Each matrix is then decomposed using
the SVD function available in Matlab, and the U vectors are plotted against tray number:
U, S, V½ ¼ SVD x, 0ð Þ (21.36)
Table 21.12 Mass Balance of a 10 ktpy Biodiesel Process Based on Reactive DWC
FAME
F-
ACID
F-
ALCO
Recycle-
ACID
Side
Stream Methanol Water
Temperature
(�C)275.21 160.00 68.90 103.34 103.34 64.25 103.34
Mass flow (kg/h)
Methanol Trace 0.00 198.10 Trace 0.21 11.03 0.21
Acid <0.001 1168.20 0.00 Trace Trace Trace Trace
Water Trace 0.00 0.00 0.12 105.15 0.02 105.04
Ester-M 1249.8 0.00 0.00 15.63 15.63 Trace Trace
Mass fraction
Methanol Trace 0.0000 1.0000 168 ppm 0.0018 0.9980 0.0020
Acid 300 ppb 1.0000 0.0000 125 ppm 6 ppm Trace 116 ppb
Water Trace 0.0000 0.0000 0.0074 0.8690 0.0020 0.9980
Ester-M 1.0000 0.0000 0.0000 0.9924 0.1292 Trace 26 ppm
Mole flow (kmol/h)
Methanol Trace 0.00 6.18 Trace 0.01 0.34 0.01
Acid Trace 5.83 0.00 Trace Trace Trace Trace
Water Trace 0.00 0.00 0.01 5.84 0.001 5.83
Ester-M 5.83 0.00 0.00 0.07 0.07 Trace Trace
Mole fraction
Methanol Trace 0.0000 1.0000 0.0010 0.0011 0.9964 0.0011
Acid 381 ppb 1.0000 0.0000 124 ppm 581 ppb Trace 10 ppb
Water Trace 0.0000 0.0000 0.0815 0.9865 0.0036 0.9989
Ester-M 1.0000 0.0000 0.0000 0.9175 0.0123 Trace 2 ppm
82321.4 DESIGN AND SIMULATION OF A REACTIVE DWC
Figure 21.16 provides the results of the SVD analysis, clearly showing that the most sensitive stages
are 11, 13 and 17 of the main column, for changes in the reflux rate, side-stream rate and reboiler heat
duty (Ignat and Kiss, 2013). However, stages 11 and 13 are too close to be independent – and such
situation can lead to a less effective dynamic response of the control structure. For the change in
reflux rate, a smaller peak can be observed at stage 7. Therefore, the temperature on stage 7 will
be controlled by manipulating the reflux rate. Note that the results for the SVD analysis are presented
only for the main column without taking into consideration the prefractionator side (the steady-state
gain matrix for the prefractionator). Since the inlet feed stream temperature is fixed and the liquid
split is not used as a manipulated variable, there are no more variables available that can be effec-
tively used as manipulated variables to control at least one of the temperatures in the prefractionator
side of the column.
In order to limit the amount of excess methanol that may pass below the partition wall of the main
column and therefore become an impurity in the side stream, an extra composition controller that
–0.8–0.6–0.4–0.2
00.20.40.60.8
11.21.41.6
0 2 4 8 10 12 14 16 18 20 22Stage
Gai
ns
DT/DSS
DT/DQR
DT/DRR
–1
–0.8
–0.6
–0.4
–0.2
0
0.2
0.4
0 2 4 86 10 12 14 16 18 20 22
Stage
U
U3 U1
U20.6
6
FIGURE 21.16
Results of the SVD analysis: plots of the steady-state gains and U vectors.
824 CHAPTER 21 CASE STUDIES
measures the methanol mass fraction in the water stream is added to set the alcohol-to-acid ratio.
Figure 21.17 illustrates the final proposed control structure, while Table 21.13 gives the tuning param-
eters of the PID controllers – where the control loops were tuned by a simple version of the direct syn-
thesis method (Ignat and Kiss, 2013).
The level of the reflux drum, the reboiler and the decanter can be controlled by the following ma-
nipulated variables: D (distillate), B (bottoms), S (side stream) and Rec (recycle stream). The compo-
sition of the free product streams is controlled by the remaining variables: L (liquid reflux), S (side
stream) and V (vapour boilup). An extra control loop is needed in order to set the alcohol-to-acid ratio
(RxR). The proposed control structure is somewhat similar to the DB/LSV control structure that was
reported as the best option for a DWC, in previous studies (Kiss, 2013). However, there are some key
differences such as the inability to use the additional optimisation loop that manipulates the liquid split
to control the heavy component composition at the top of prefractionator and implicitly achieve mini-
misation of the energy requirements.
In the case of standard DWC, any heavy component going out the top of the wall will appear also in
the liquid flowing down in the main column, thereby affecting the purity of the side stream (S). How-
ever, this is not the case here, since the water by-product is removed as a side stream and separated in a
Water
F-ACID
rL
rV
1
1 7
18
22
19
6
4
12
RD
C
DWC
16
DEC
Recycle-ACID
Side stream
ACID
F-ALCOALCO
RC
Methanol
LC
TC
LC
TCFAME
CC
9
17
TC13
LC
LC
FIGURE 21.17
Control structure of the R-DWC unit, based on level control (4�), inferential temperature control (3�),composition (1�) and reactants ratio (1�) control.
82521.4 DESIGN AND SIMULATION OF A REACTIVE DWC
decanter from which the organic phase (heavy component of the system) is recycled back to the col-
umn. Therefore, the liquid split is not effective as a manipulated variable for controlling the heavy
impurity at the top of the prefractionator. Actually, in this study, the situation is the other way around.
Any light component going down below the bottom of the dividing wall will become an impurity in the
side stream since the prefractionator is not operated at the preferred split between the light and heavy
key components like in the case of a standard DWC. Since the side stream is collected as a liquid prod-
uct, it means that any small amounts of light impurity in the vapour phase will become an impurity in
the side stream. Because the location of the wall fixes how the vapour flow splits between the two sides
of the column, the vapour split (rV) variable is not adjustable during operation for control purposes
(Kiss, 2013). Therefore, the vapour split ratio is not used as a manipulated variable in the dynamic
simulations presented hereafter.
Figures 21.18 and 21.19 present the results of the dynamic simulations for industrially relevant dis-
turbances such as production rate changes and modifications in the vapour split ratio (Ignat and Kiss,
2013). Remarkably, the mass fractions of all components are returning to their set point within reason-
able short times and low overshooting, thus proving that the system can successfully reject the distur-
bances. As the methanol stream is fed as vapour – and thus ensures most of the vapour amount required
on the prefractionator side of the column – the vapour split ratio has in this case only a weak influence
on the purities of the products. Note that no disturbances are expected in the liquid split (rL) since thisvariable is set (manipulated) in open loop.
The performance of the control system was also tested for a deactivating catalyst. The investigated
scenario assumed a decrease of the pre-exponential factor from 250 (at t<2 h) to only 200 kmol �m3/
(kg2 � s) (at time t¼10 h). The control system succeeds to keep the throughput and product quality
unchanged. Figure 21.20 presents the temperature profiles, while Figure 21.21 shows the composition
profiles in the column for different catalyst activities at different times: t¼1 h and t¼10 h (Ignat and
Kiss, 2013). As one can notice, the lower catalyst activity is compensated by a slight increase of the
temperature, while the product purities are kept at the desired values. In other words, the control system
compensates the lower catalyst activity by a higher temperature inside the column, with the result of
maintaining a similar reaction rate.
Table 21.13 Tuning Parameters of the PID Controllers
Controlled
Variable
Manipulated
Variable
Gain, P(%/%)
Int. Time, I(min)
Drv. Time, D(min)
Control
Direction
T7_DWC L 0.4 30 0 +
T13_DWC S 1 20 0 +
T17_DWC V 1 20 0 –
xmethanol rL 0.6 40 0 –
DWC_TankLevel D 1 60 0 +
DWC_ReboilerLevel B 1 60 0 +
Dec_OrganicPhaseLevel Rec 1 60 0 +
Dec_AquaPhaseLevel Water 1 60 0 +
826 CHAPTER 21 CASE STUDIES
This case study illustrated how to tackle the optimal design, dynamics and control of an inte-
grated R-DWC for biodiesel synthesis from FFAs and methanol. As the first optimal design had
poor dynamic performance, a key finding was that it is imperative to use a vapour feed of alcohol
in order to reach the product specifications. The re-optimised design proved to work well also in
transient (dynamic) regime. SVD was effectively used to find the sensitive trays for inferential tem-
perature control. The control structure proposed for a two-reactant and two-product RD system suc-
cessfully demonstrates the excellent performance of the system in the case of industrially relevant
disturbances such as production rate changes and catalyst deactivation. However, in case of differ-
ent types of reactions (e.g. two reactants and one product), other control structures have to be
considered.
0.9965
0.9970
0.9975
0.9980
0.9985
0.9990
0.9995
1.0000
0 1 2 3 4 5 6 7 8 9 10Time (h)
Mass
fra
ctio
n (
–)
+10% Production rate change
FAME
Water
Methanol
0.9965
0.9970
0.9975
0.9980
0.9985
0.9990
0.9995
1.0000
0 1 2 3 4 5 6 7 8 9 10
Time (h)
Mass
fra
ctio
n (
–)
–10% Production rate change
FAME
Water
Methanol
FIGURE 21.18
Results of the dynamic simulations at production rate changes of 10%.
82721.4 DESIGN AND SIMULATION OF A REACTIVE DWC
0.9965
0.9970
0.9975
0.9980
0.9985
0.9990
0.9995
1.0000
0 1 2 3 4 5 6 7 8 9 10Time (h)
Mas
s fr
actio
n (–
)
+10% Vapour split ratio disturbance
FAME
Water
Methanol
0.9965
0.9970
0.9975
0.9980
0.9985
0.9990
0.9995
1.0000
0 1 2 3 4 5 6 7 8 9 10
Time (h)
Mas
s fr
actio
n (–
)
–10% Vapour split ratio disturbance
FAME
Water
Methanol
FIGURE 21.19
Results of the dynamic simulations at vapour split ratio disturbance of 10%.
406080
100120140160180200220240260280
0 2 4 6 8 10 12 14 16 18 20 22
Stage (–)
Tem
pera
ture
(°C
)
Prefractionator (PF)
Main column
FIGURE 21.20
Change of the temperature profile during catalyst deactivation.
21.4.5 SUGGESTIONS FOR FOLLOW-UP PROJECTS1. Design an optimal intensified process for bioethanol dehydration based on extractive distillation
taking place in a single (DWC).
2. Design a bioethanol dehydration process based on azeotropic distillation in a DWC, and examine
the differences compared to the extractive DWC alternative.
3. Design an optimal reactive distillation or R-DWC for other types of equilibrium-limited reactions
(esterification, etherification, condensation, alkylation, etc.).
REFERENCESBehr, A., Obendorf, L., 2003. Development of a process for acid-catalyzed etherification of glycerine and isobu-
tene forming glycerine tertiary butyl ethers. Eng. Life Sci. 2, 185–189.
Bildea, C.S., Dimian, A.C., 1998. Stability and multiplicity approach to the design of heat-integrated PFR.
AIChE J. 44, 703–2712.
Bildea, C.S., Dimian, A.C., Iedema, P.D., 2000. Nonlinear behavior of reactor-separator-recycle systems. Comput.
Chem. Eng. 24, 209–215.
Bildea, C.S., Cruz, S., Dimian, A.C., Iedema, P., 2004. Design of tubular reactors in recycle systems. Comput.
Chem. Eng. 28, 63–72.
Dimian, A.C., 2003. Integrated Design and Simulation of Chemical Processes. Elsevier, Amsterdam.
Douglas, J.M., 1988. Conceptual Design of Chemical Processes. McGraw-Hill.
Ignat, R.M., Kiss, A.A., 2013. Optimal design, dynamics and control of a reactive DWC for biodiesel production.
Chem. Eng. Res. Des. 91, 1760–1767.
Kiatkittipong, W., Intaracharoen, P., Laosiripojana, N., Chaisuk, C., Praserthdam, P., Assabumrungrat, S., 2011.
Glycerol ethers synthesis from glycerol etherification with tert-butyl alcohol in reactive distillation. Comput.
Chem. Eng. 35, 2034–2043.
Kiss, A.A., 2013. Advanced Distillation Technologies – Design, Control and Applications. Wiley.
Kiss, A.A., 2014a. Process Intensification Technologies for Biodiesel Production – Reactive Separation Processes.
Springer.
Kiss, A.A., 2014b. Distillation technology – still young and full of breakthrough opportunities. J. Chem. Technol.
Biotechnol. 89, 479–498.
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0 2 4 6 8 10 12 14 16 18 20 22
Mol
e fr
actio
n (–
)
FAMEWater
FFA
Methanol
t = 10 h
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0 2 4 6 8 10 12 14 16 18 20 22
Stage (–) Stage (–)
Mol
e fr
actio
n (–
)
FAME
Water
FFA
Methanol
t = 1 h
FIGURE 21.21
Change of the composition profile during catalyst deactivation.
829REFERENCES
Kiss, A.A., Bildea, C.S., 2012. A review on biodiesel production by integrated reactive separation technologies.
J. Chem. Technol. Biotechnol. 87, 861–879.
Kiss, A.A., Segovia-Hernandez, J.G., Bildea, C.S., Miranda-Galindo, E.Y., Hernandez, S., 2012. Reactive DWC
leading the way to FAME and fortune. Fuel 95, 352–359.
Luyben, W.L., 1994. Snowball effects in reactor-separator processes with recycle. Ind. Eng. Chem. Res.
33, 299–305.
Luyben, W.L., Yu, C.C., 2008. Reactive Distillation Design and Control. Wiley-AIChE.
Luyben, W.L., Tyreus, B.D., Luyben, M.L., 1999. Plantwide Process Control. McGraw-Hill.
Marquardt, W., Monnigmann, M., 2005. Constructive nonlinear dynamics in process systems engineering. Com-
put. Chem. Eng. 29, 1265–1275.
Monnigmann, M., Marquardt, W., 2003. Steady-state process optimization with guaranteed robust stability and
feasibility. AIChE J. 49, 3110–3126.
Schmidt-Traub, H., Gorak, A., 2006. Integrated Reaction and Separation Operations. Springer.
Sundmacher, K., Kienle, A., 2003. Reactive Distillation: Status and Future Directions. Wiley-VCH.
Vlad, E., Bildea, C.S., Bozga, G., 2013. Robust, optimal design of glycerol etherification process. Chem. Eng.
Technol. 36 (2), 251–258.
830 CHAPTER 21 CASE STUDIES