You are on page 1of 12

Chemical Engineering Science 60 (2005) 4275 4286

www.elsevier.com/locate/ces

Modeling, simulation and control of a methanol synthesis xed-bed reactor


M. Shahrokhi , G.R. Baghmisheh
Department of Chemical & Petroleum Engineering, Sharif University of Technology, P.O. Box 11365-9456, Tehran, Iran

Received 21 June 2004; received in revised form 8 November 2004; accepted 23 December 2004
Available online 18 April 2005

Abstract
In this paper, the dynamic behavior and control of the low pressure methanol synthesis xed bed reactor have been investigated. For
simulation purpose, a heterogeneous one-dimensional model has been developed. First, the reactor simulation is carried out under steady-
state condition and the effects of several parameters such as shell temperature, feed composition (especially CO2 concentration) and recycle
ratio on the methanol productivity and reactor temperature prole are investigated. Using the steady state model and a trained feedforward
neural network that calculates the effectiveness factor, an optimizer which maximizes the reactor yield has been developed. Through the
dynamic simulation, the system open loop response has been obtained and the process dynamic is approximated by a simple model. This
model is used for the PID controller tuning and the performances of xed and adaptive PID controllers are compared for load rejection and
set point tracking. Finally the proposed optimizer is coupled with a controller for online optimization and hot spot temperature protection.
2005 Elsevier Ltd. All rights reserved.

Keywords: Methanol synthesis; Fixed-bed reactor; Dynamic simulation; Adaptive PID; Temperature runaway; Feedforward neural network

1. Introduction design and the most frequently use of these types of reac-
tors, the boundaries of the system and their interactions are
Chemical reactors especially xed-bed catalytic reactors quite complex and can raise complicated problems in de-
that are used in highly exothermic processes are often dif- sign, safe operation, optimization and control. Modeling of
cult to control and it is important to predict the possibility these reactors is a complex task since a system of nonlin-
of hot spot and thermal runaway occurrences (Assaf et al., ear differential equations must be solved and many transport
1996). This is the case when wide uctuations in tempera- and chemical parameters should to be evaluated; in addition
ture extremes occur from relatively minor uctuations in one the diffusion of gas into the solid matrix is hard to model
or more operating variables. Operation in the unstable re- (Parisi and Laborde, 2001). Several authors have studied the
gions may result in poor product and temperature runaway. steady-state modeling of catalytic methanol synthesis reac-
The need for optimal control of chemical reactors has been tor at various level of complexity, but a few studies have been
well recognized since the early 1980s, due to the increasing done on dynamic simulations and control of methanol reac-
cost of raw material and energy (Zhou et al., 1999). Mul- tor. The summary of these studies have been listed below.
titubular xed-bed reactors are industrially used in the low Graaf et al. (1990) modeled low pressure methanol syn-
pressure methanol synthesis from syngas. This catalytic re- thesis using commercial CuZnAl catalyst. They showed
action is highly exothermic and temperature has signicant that commercial size of the catalyst particles exhibit intra-
effect on the reactors yield. In this work dynamic behavior particle diffusion limitations. Vanden Bussche et al. (1993),
and control of a low pressure methanol synthesis xed-bed based on a heterogeneous model, simulated the dynamic
reactor has been considered. Despite of the relatively simple behavior of a xed-bed methanol reactor. Their investiga-
tion includes comparison between different levels of tran-
Corresponding author. Tel.: +98 21 6165419; fax: +98 21 6022853. sient modeling. Lommert et al. (2000) studied mathemat-
E-mail address: shahrokhi@sharif.edu (M. Shahrokhi). ical modeling of internal mass transport limitations in the
0009-2509/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ces.2004.12.051
4276 M. Shahrokhi, G.R. Baghmisheh / Chemical Engineering Science 60 (2005) 4275 4286

methanol synthesis and showed that Thiele modulus con- 3. Development of the reactor model
cept with pseudo-rst-order kinetics are capable to pre-
dict the intra-particle diffusion. Lovik (2001) studied 3.1. Reactor model assumptions
modeling, estimation and optimization of the methanol
synthesis with catalyst deactivation. A rigorous pseudo- For modeling the reactor the following assumptions are
steady state modeling of methanol synthesis loop has made: (1) one-dimensional heterogeneous model is consid-
been considered in his simulation. Velardi et al. (2002) ered, (2) dispersion of mass and energy in axial and radial
investigated the feasibility of carrying out the low pres- direction are negligible, (3) the transient term is taken into
sure methanolsynthesis process in forced unsteady- account for the mass and energy balances, while the pseudo-
state conditions, using a network of three catalytic steady state condition is considered for the solid phase in
xed-bed reactors with periodical change of the inlet calculating the effectiveness factor, (4) the dynamics of the
position. catalyst deactivation is neglected because of its large time
The present paper focuses on the dynamic simulation constant.
and control of the methanol reactor. The paper is orga-
nized as follows. First the process and the related con- 3.2. Reactor mathematical model
trol loops are described. Modeling of reactor and steam
drum is considered next. For solving the equations de- One-dimensional and heterogeneous model has been con-
scribing the system behavior, the numerical methods for sidered for simulating the tubular xed-bed methanol syn-
solving nonlinear algebraic and differential equations thesis. To take into account intra particle mass diffusion, ef-
are discussed. Steady state and dynamic results are pre- fectiveness factor concept (Levenspiel, 2001; Elnashaie and
sented and nally reactor control and optimization are Elshishini, 1993; Wijngarden et al., 1998) has been used.
considered. The mass and energy equations are given below.
Component mass balance:
 
jxi us F 0 jx i
= 0 + i (z)b ri Ar ,
jt F L r jz
2. Process summary
(i = key component (MeOH & CO)). (1)
The process ow sheet of the methanol synthesis loop is Energy balance:
shown in Fig. 1. The catalyst is packed in vertical tubes and
  NR 
surrounded by the boiling water. The reaction heat is trans- jT us C p,g jT 
ferred to the boiling water and steam is produced. Efcient C p,m = + b ((H )k k (z)rk )
jt Lr jz
heat transfer gives small temperature gradient along the re- k=1

actor. Typical operating conditions are 543 K and 80 bar. The 4Ueff
reactor temperature is controlled by controlling the pres- + (T TC ) . (2)
do
sure of the boiling water. Besides the pressure loop a level
controller is used for controlling the water level in steam
For the pressure drop, the following equation has been used
drum.
(Ergun Equation):
  
d 3 1
P = 10 Lr 1.75 + 150
dz NRe
u2s g 1 
. (3)
dp 3

In above equations i is the effectiveness factor which can be


calculated from the following equations that must be solved
with the above equations simultaneously:
j
  j 
De 1 d 2 d Ps

Rg R p  d 
2 2 d  Ts
= b rj , (j = all component), (4)
  NR

Ke 1 d 2 d
 T s = p (Hk )rk , (5)
Fig. 1. Flow sheet of Lurgis synthesis loop and its control strategy (Lovik, Rp2 2 d d
2001). k=1
M. Shahrokhi, G.R. Baghmisheh / Chemical Engineering Science 60 (2005) 4275 4286 4277

 i i
ri dVp De (dPs /dr)r=Rp Table 1
i = = , Kinetic and equilibrium constants
Vp ri |Surface Rg Ts (Rp /3)p ri |Surface
(i = key component (MeOH & CO)). (6) k = A exp(B/Rg T ) A B

The Eqs. (4), (5) are written based on pseudo-steady state ka (bar 1/2 ) 0.499 17197
kb (bar 1 ) 6.62e11 124119
assumption. Regarding this assumption, it should be men-
kb 3453.38
tioned that if the rate equations are linearized and the kd (mol/kg s bar2 ) 1.07 3669
approach proposed by Szukiewicz (2001) is used, the ke (mol/kg s bar) 1.22e10 94765
maximum time constant for the diffusion-reaction equations
(including the transient term) will be about 7 s, while the K eq = 10(A/T B) A B
time constant for the composition change is around 37 s. K1 (bar 2 )
eq
3066 10.592
These observations justify the pseudo-steady state assump- eq
K2 2073 2.029
tion.
The initial and boundary conditions for the bulk phase
and the boundary conditions for the catalyst pellets are given
below. The reaction rates of methanol and carbon monoxide accord-
Boundary conditions for the bulk gas phase: ing to reactions (B) and (C) have been proposed by Vanden
Bussche and Froment (1996) and are given below:
z = 0: xi = 0, i = key component,
   
P = Pinlet , T = TInlet . (7) 1
kd PCO PH2 1 eq (PH2 O PCH3 OH /PH2 PCO2 )
Initial conditions for the bulk gas phase: K1
rB =
3 , (14)
1+kc PH2 O /PH2 +ka PH2 +kb PH2 O
t = 0, z: xi |t=0 = xi |steady state ,
T |t=0 = T |steady state . (8) eq
ke PCO2 1 K2 (PH2 O PCO /PH2 PCO2 )
rC =
. (15)
Boundary conditions for the catalyst pellet: 1 + kc PH2 O /PH2 + ka PH2 + kb PH2 O
d i
 = 0: P |=0 = 0, i = all components, The kinetic parameters are given in Table 1.
d s
d
Ts |=0 = 0, (9)
d
4. Steam drum mathematical model
Dei d i
 = 1: P |=1 = kgi (Psi |=1 P i ),
Rp d  s To develop a steam drum model, the following assump-
Ke d tions are made: (1) in energy balance, pseudo-steady state
Ts |=1 = hf (Ts |=1 T ). (10)
Rp d  condition is considered for vapor phase, (2) energy losses via
surrounding is neglected. The dynamics of this process can
Solving the above equations for calculating the effectiveness be described by the following set of mass and energy bal-
factor is time consuming. One way to reduce the computa- ance equations for liquid and vapor phases (Luyben, 1986;
tion time is using the concept of Thiele modulus. Lommert Ramirez, 1997) and the schematic diagram of the steam
et al. (2000) have linearized the methanol reactions and used drum is shown in Fig. 2.
the concept of Thiele modulus to calculate the effectiveness
factor. The parameters of the linearized rates are adopted
by using the real rates along the reactor. Their approach is
described briey in Appendix A.
The model is completed by the kinetic equations given
by Vanden Bussche and Froment (1996) which is based on
LangmuirHinshelwood and HougenWatson mechanism.
In this kinetic model two independent reactions (hydrogena-
tion of carbon dioxide and the reverse watergas shift re-
action) out of the three following depended reactions are
considered:

(A) CO + 2H2 CH3 OH, (11)

(B) CO2 + 3H2 CH3 OH + H2 O, (12)

(C) CO2 + H2 CO + H2 O. (13) Fig. 2. The schematic diagram of the steam drum.
4278 M. Shahrokhi, G.R. Baghmisheh / Chemical Engineering Science 60 (2005) 4275 4286

Table 2 algebraic equations are integrated by TR-BDF2 method


Methods used for calculating the physicochemical properties (Shampine, 1994). To calculate the effectiveness factor for
Property Method
catalyst pellets, rst, two-point boundary differential equa-
tions for catalyst pellets are discretized by orthogonal col-
Gassolid heat transfer co- ChiltonColburn analogy (Poling location using nite difference (Finlayson, 1980, 1972) and
efcient et al., 2001)
then the resulting nonlinear equations have been solved by
Gas viscosity Lucas (Poling et al., 2001)
Gas conductivity SteilThodos (Perry and Green, GaussNewton and trust-region method (Coleman and Li,
1997) 1996). To decrease the computation time in dynamic simu-
Binary diffusion coefcient, FullerSchettlerGidding (FSG) lation, two other methods are tested. These techniques are:
component i and j (Perry and Green, 1997) (1) articial feedforward neural network (AFFNN) trained
Effective diffusion coef- Wilke equation (Froment and
with the obtained data from exact simulated system (Parisi
cient in pellet, component i Bischoff, 1990)
Overall heat transfer coef- and Laborde, 2001), (2) the adaptive linear kinetic rate
cient gas phase to boiling Lovik, 2001 Thiele Modulus method (ALKTM) described by Lommert
water et al. (2000). Both techniques reduce the computation time
Water saturation pressure KeenanKeyes (Perry and Green, considerably. The AFFNN is faster but ALKTM has better
1997)
accuracy.
Mixed gas heat capacity Ideal gas (Poling et al., 2001)
Mass transfer coefcient Cussler Equation (Perry and Green,
1997)
Gas compressibility factor PengRobinson (Poling et al., 2001) 6. Steady-state simulation results
for reactor inside
Gas compressibility factor SoaveRedlichKwong (Poling 6.1. Model validation
for steam et al., 2001)

The steady-state simulation results are compared with the


data obtained from three industrial reactors. The simula-
Continuity equation for liquid phase: tion results with corresponding industrial data are given in
d(l Vl ) Table 3. As can be seen there is a good agreement between
= m Fm W . (16) simulation results and industrial data.
dt
Through steady-state simulation the following results
Continuity equation for vapor phase: have been obtained: (1) CO2 injection to fresh syngas in-
d(v Vv ) creases the methanol productivity (Table 4), which conrms
= W  v Fv . (17) the results obtained by Skrzypek et al. (1995); (2) increas-
dt
ing the recycle ratio, increases the methanol productivity
Energy balance for liquid phase: (Table 4) and decreases the possibility of hotspot formation
d(l Vl Tc ) at the expense of increasing operation costs. As the nal
Cp,w = m Fm hm hs W + QR , (18) conversion approaches the equilibrium conversion, the ef-
dt
fect of recycle ratio on reactors yield decreases; (3) the
where W, Vv , Fv and Pv are given by the following equations:
methanol productivity and temperature prole in reactor are
W = KMT (P sat Pv ), (19) inuenced by the temperature of cooling water as shown in
Fig. 3(a, b).
z s v R g T c
Pv = , (20) To check the accuracy of the effectiveness factor ob-
MW water tained from Thiele modulus or articial network, these fac-
Vl + Vv = Vt , (21) tors are calculated and compared for simulation of reactor 1

(Table 3) as shown in Fig. 4.
Fv = Kv Pv (Pv P0 ). (22) As can be seen both methods can predict the effective-
The terms KMT and Kv are pseudo-mass transfer and con- ness factor fairly well, but the Thiele modulus method has
trol valve coefcients respectively. The methods used for a higher accuracy.
calculating the physicochemical properties needed for sim-
ulation are given in Table 2. 6.2. Steady-state optimizer

The shell temperature strongly inuences the methanol


5. Numerical method yield and reactor thermal stability. As can be seen from
Fig. 3 there is an optimum temperature that maximizes
The partial-differential equations describing the reactor the methanol conversion for a given feed composition.
dynamics are discretized by nite difference to ordinary- However, the hotspot temperature constraint in reactor
differential equations. The resulting set of differential equa- must be considered for thermal stability. When the feed
tions, steam drum equations and catalyst pellet nonlinear composition is changed, the shell temperature should
M. Shahrokhi, G.R. Baghmisheh / Chemical Engineering Science 60 (2005) 4275 4286 4279

Table 3
Comparison of simulation results and industrial data

Property Simulation result Industrial Simulation result Industrial Simulation result Industrial
for reactor 1 reactor 1 for reactor 2 reactor 2 for reactor 3 reactor 3

MeOH Conversion 46.08 48.85 52.62 57.18 29.84 32.67


CO Conversion 35.17 35.55 38.33 41.42 17.79 18.51
Exit temp. ( C) 249 250 253 255 253 255
Exit P(bar) 79.3 76.1 62.02 61.16 78.1 76.45

Mole%
yCO2 4.87 4.66 2.50 2.18 8.43 8.1
yCO 2.17 2.17 1.70 1.44 2.61 2.38
y H2 57.58 56.92 76.37 75.71 50.12 48.97
yCH4 27.52 27.44 13.05 13.14 31.79 33.05
yH2 O 1.49 1.88 1.36 1.74 1.99 2.48
yCH3 OH 6.37 6.92 5.02 5.49 5.06 5.82
Byproduct 0.01

Table 4 By mixing these streams with different ratios, the reactor


Effects of CO2 injection and increasing recycle ratio on methanol yield feed is generated. In Fig. 5 the feed composition can be
Simulation results for reactor 1 ( Table 3) calculated through F1 /F2 ratio.
As can be seen from Fig. 5 there are two optimal proles.
Type of change Rate of changes Increasing methanol yield
The lower temperature prole is obtained by taking into ac-
CO2 injection rate Increasing CO2 rate 5.2% count the constraint on the temperature. The higher tempera-
by factor 2 ture prole is unconstrained optimal temperature trajectory.
Recycle rate Increasing recycle ra- 4.3% For calculating the optimal shell temperature for a given
tio from 3 to 4
feed composition, an optimizer has been developed. It uses a
steady-state model to calculate the optimal temperature con-
sidering the temperature constraints. In mathematical terms
it can be stated as follows:
be changed accordingly in order to have the maximum
conversion. The optimal temperature prole obtained maxTshell J = FTout yM
out
,
for different feed compositions is shown in Fig. 5. To
change the feed composition, two streams with differ- subject to
ent compositions (F1 : yMeOH = 0, yCO = 0.2052, yH2 =
Tmax  543 K,
0.6853, yCO2 = 0.084, yH2 O = 0.005, yCH4 = 0.0205 & F2 :
yMeOH = 0.0042, yCO = 0.0232, yH2 = 0.6213, yCO2 =
0.0491, yH2 O =0.0042, yCH4 =0.298) have been considered. 505  Tshell ,

Fig. 3. Effect of cooling uid temperature on (a) methanol conversion at reactor outlet and (b) reactor temperature prole along the reactor.
4280 M. Shahrokhi, G.R. Baghmisheh / Chemical Engineering Science 60 (2005) 4275 4286

7. Dynamic simulation results and controller design

7.1. Open loop response

To obtain open loop response of pressure loop, step


changes in different directions with different magnitudes
are applied to steam control valve opening while the level
is being controlled. The results are shown in Fig. 6. As
can be seen the process shows nonlinear behavior. The de-
gree of nonlinearity depends on deviations from operating
conditions.

7.2. Ofine identication

Fig. 4. Comparison of effectiveness factors obtained from three different In order to model the pressure dynamics, maximum
methods. 20% deviation in manipulated variable is considered and
sequences of step changes with different magnitudes are
applied to process input and the inputoutput data are col-
where FTout and yM out are reactor outlet owrate and methanol
lected. Two different models, rst and second order, are
mole fraction respectively. The rst constraint is considered considered. Using the ofine least squares method, the pa-
to avoid hotspot in the reactor and the second constraint rameters of corresponding discrete models are identied.
is due to required quality of the produced vapor (Lovik, Having the sample time and the discrete models, the param-
2001). In the above objective function only methanol yield eters of the corresponding continuous models are calculated
has been considered and penalty on generated steam in shell (Seborg et al., 1989). These models are given in Table 5.
side is set to zero. The optimization has been carried out us-
ing Golden Section Search method (MATLAB optimization
7.3. PID controller tuning and adaptive PID controller
toolbox (fminbnd)). In this method the two interior points
within the total interval are so selected that the interval elim-
The internal model control (IMC) technique has been used
inated on one iteration will be of the same proportion to the
for PID controller tuning. The PID controller formula and
total interval. For more details the readers are referred to
the controller parameters are as follows:
Edgar and Himmelblau (1988). To handle the constraints,
 
rst a relatively high shell temperature is chosen and the 1 D s
Gc (s) = Kc 1 + + .
optimum shell temperature is calculated by using Golden I s N s + 1
Section Search method. If the upper bound on reacting gas
temperature is satised the search is completed and the pro- For the rst order model:
gram terminates, otherwise a smaller shell temperature is 
Kc = , I = . (23)
chosen and the above calculations will be repeated. K p c

Fig. 5. Optimal shell temperature for different feed compositions.


M. Shahrokhi, G.R. Baghmisheh / Chemical Engineering Science 60 (2005) 4275 4286 4281

performance. To compare the performances of xed PID


with adaptive PID, an adaptive PID controller based on
IMC techniques has been designed. For online identica-
tion recursive least squares with variable forgetting factor
(Fortescue et al., 1981) is used. To describe the system dy-
namics the following ARMA model is selected.

y(k) b0 + b1 q 1
G(q) = = .
u(k) 1 + a1 q 1 + a2 q 2

The system parameters are updated as follows:

P (k 1)(k)
Fig. 6. Open loop responses of shell pressure due to step changes in (k) = (k 1) + e(k),
steam owrate.
 + T (k)P (k 1)(k)

P (k)

1 P (k1)(k)T (k)P (k1)
Table 5 = P (k1) . (25)
 +T (k)P (k1)(k)
Ofine identication results for pressure control loop

Process parameters First order model Second order model


Where (k), (k) and e(k) are given below:
kp 2.9286e 3 mPa/s l 2.9286e 3 mPa/s l
1 (s) 170 210
2 (s) 70
T (k) = (a1 (k), a2 (k), b0 (k), b1 (k)), (26)
3 (s) 40
T (k) = (y(k 1), y(k 2), u(k), u(k 1)), (27)
kp kp (3 s + 1)
Gp1 = Gp2 =
1 s + 1 (1 s + 1)(2 s + 1)
e(k) = y(k) y(k)
= y(k) T (k)(k 1). (28)

 is the forgetting factor and computed from the following


equation:
Table 6
PID controller parameters
min , 0 < min ,

Controller N (s) D (s) I (s) Kc < 1,
(k) = max e2 (k)

1
PI (Pressure loop) 0 170 5.801e2 ,
PI (Level loop) 0 300 72 (1 +  (k)P (k 1)(k))
T
PID (Pressure loop) 40 21.25 159 8.195e2 (29)

where the is the tuning parameter. To solve the above


equations some initial values should be considered for (k)
and P (k).
For second order model:
The controller transfer function is
1 + 2 3
Kc = , I = 1 + 2 3 ,
K p c 1a 1 + a1 z1 + a2 z2
1 2 D(z) =  , (30)
D = 3 , N = 3 , b1 b2
1 + 2 3
(24) (1 z1 ) 1 + z1
b1
where c is a tuning parameter. This parameter is obtained
by minimizing the integral of absolute value of the error for where a, is a tuning parameter and is given below:
set point tracking. For level loop a PI controller is considered
and its parameters are obtained by the same procedure. The a = exp(2.3Ts /Tr ), Tr = 0.9 settling time,
PID parameters are given in Table 6. Ts = sampling time. (31)
The performances of PI and PID controllers for set
point tracking are compared and shown in Fig. 7. The The performances of xed and adaptive PID for set point
simulation results indicate that PID controller has a better tracking and load rejection are compared in Figs. 8, 9. For
4282 M. Shahrokhi, G.R. Baghmisheh / Chemical Engineering Science 60 (2005) 4275 4286

Fig. 7. Comparison of PI & PID performances for pressure set point tracking, (a) shell pressure vs. time, (b) methanol owrate vs. time, (c) steam
owrate vs. time.

set point tracking several step changes are applied to the To evaluate the performance of the control sys-
shell pressure set point as shown in Fig. 8a and for load tem coupled with the developed optimizer, two dif-
rejection the feed composition has been changed according ferent loads are applied to feed composition at t =
to gures given in Table 7. 395 and t = 1000 s. The results are shown in
As can be seen the performance of adaptive PID is much Fig. 11.
better than xed PID in both set point tracking and load The simulation results show that the optimizer has
rejection. It should be also noted that the maximum reactor increased the methanol yield after loads are applied to
temperature overshoot is higher for the xed PID controller the feed composition. As can be seen from Fig. 11(b),
compared to adaptive PID. the second load has caused the reactor temperature to
exceed the maximum allowable temperature for a cer-
tain period of time, but it has been decreased when
the optimizer has provided the new pressure set point.
8. Online optimization
As can be seen the optimizer provides the maximum
methanol productivity while satisfying the temperature
When the feed composition is changed the set point
constraints.
of pressure loop should be changed accordingly in or-
der to have the maximum methanol yield. To calculate
the pressure set point corresponding to feed composition
an optimizer has been developed. The optimizer receives 9. Conclusions
the feed specications in each sample time and if detects
changes, calculates the new pressure set point (and cor- In this article using conservation laws, the dynamic
responding shell temperature) by using the steady-state behavior of a methanol reactor has been simulated.
model. For simulation it is assumed that the feed compo- To reduce the computation time the effectiveness fac-
sition analyzer has 200 s delay. Fig. 10 shows the above tor is estimated through AFFNN and ALKTM. ALKTM
procedure. provides more accurate results but AFFNN is faster.
M. Shahrokhi, G.R. Baghmisheh / Chemical Engineering Science 60 (2005) 4275 4286 4283

Fig. 8. Comparison of xed & adaptive PID performances for pressure set point tracking, (a) shell pressure vs. time, (b) methanol owrate vs. time, (c)
steam owrate vs. time.

Fig. 9. Comparison of xed & adaptive PID performances for feed composition changes according to Table (6). (a) Shell pressure vs. time, (b) reactor
maximum temperature vs. time, (c) steam owrate vs. time.
4284 M. Shahrokhi, G.R. Baghmisheh / Chemical Engineering Science 60 (2005) 4275 4286

Table 7 Steady-state simulation result has a good agreement with


Load applied to feed composition industrial data. Using IMC technique and system model
Mole Initial feed Feed composition Feed composition obtained from open loop response, xed PID controller
fraction composition at t = 399 s at t = 1000 s is tuned. Performances of xed and adaptive PID con-
yCO2 8.39 6.83 8.4 trollers are compared for load rejection and set point
yCO 20.48 20.48 20.48 tracking. To increase the reactor yield an optimizer has
yH2 68.40 0.72 66.5 been developed which can be coupled with the control
yCH4 2.54 0.5 4.43 system. Simulation results indicate that using the opti-
yH2 O 0.19 1.019 0.19
mizer has increased the reactor productivity and thermal
yCH3 OH 0 0 0
stability.

Notation

Ar reactor section area, m2


C p,g specic heat of uid, kJ/m3 K
C p,m mean specic heat of bed (C p,g + s (1
)Cp,s ), kJ/m3 K
Cp,s specic heat of solid, kJ/kg K
Cp,w specic heat of water, kJ/kg K
dp , do catalyst and reactor outside diameter, m
Dei effective diffusion coefcient of compo-
Fig. 10. Control system coupled with an optimizer for optimizing the nent i in the pellet, m/s
methanol yield due to changes in feed composition. F0 inlet mole owrate, mol/s

Fig. 11. The performance of control system coupled with the proposed optimizer when two subsequent changes are applied to feed composition (a) shell
pressure vs. time, (b) reactor maximum temperature vs. time, (c, d) methanol owrate vs. time.
M. Shahrokhi, G.R. Baghmisheh / Chemical Engineering Science 60 (2005) 4275 4286 4285

Fm , Fv volume owrate of makeup water, vapor, Appendix A


m3 /s
Gc transfer function of pressure loop or con- The procedure for calculating the effectiveness factor pro-
or p
troller posed by Lommert et al. (2000) is given below. The lin-
hf gas inside heat transfer coefcient, earized methanol reaction rates are as follows:
kW/m2 K
rMethanol
= kMethanol
eq
(CH2 CMethanol /kMethanol ), (A.1)
hm mass enthalpy of makeup, J/kg
eq
hs mass enthalpy of steam, J/kg rCO = kCO (CH2 CCO /kCO ), (A.2)
ka,b,c,d reaction rate parameters eq eq
kgi mass transfer coefcient from gas to pellet where KMethanol and KCO are dened as follows:
component i  
eq CMethanol
ke effective pellet conductivity, kW/m K kMethanol = , (A.3)
CH2
Lr reactor length, m eq
 
NRe Reynolds number based on particle eq CCO
NR number of reactions kCO = . (A.4)
CH2 eq
P , PInlet , P sat bulk, reactor inlet and saturation pressure,
kPa Having reaction the rates of methanol & CO which can be
P0 steam valve discharge pressure, kPa obtained from Eqs. (14) and (15), the pseudo-kinetic con-

Psi partial pressure inside the pellet, compo- stants KMethanol and KCO are determined from equations
nent i, kPa (A.1) and (A.2). Thiele modulus is dened as
QR heat generated by reaction, W 
 eq
R p k (k + 1)
r radial coordinate in catalyst, m j
M =  j j , (j = Methanol, CO). (A.5)
rB,C j eq
reaction rate, mol/kg s 3 De kj
ri,j or k reaction rate for component i, j or k,
mol/kg s From the Thiele modulus the effectiveness factor can be cal-
Rg gas constant, kPa m3 /mol K culated from the following equation (Froment and Bischoff,
Rp catalyst radius, m 1990):
t time, s  j j j

3Sh M cosh M sinh M
T , TInlet , Tc , Ts temperature (reactor inside, reactor inlet, j = j j j j
, (A.6)
cooling water, solid), K M 2 M cosh M + (Sh 1) sinh M
us supercial gas velocity, m/s where
Ueff effective heat transfer coefcient,
Kgi Rp
kW/m2 K Sh = j
. (A.7)
Vv,l,p,t volume (vapor, liquid, particle, total), m3 De
W water boiling rate, kg/s
xi conversion of component i References
yi mole fraction of component i
z dimensionless reactor length Assaf, E.M., Giordano, R.C., Nascimento, C.A.O., 1996. Thermal runaway
zs compressibility factor for steam of ethylene oxidation reactors: prevision through neural network.
Chemical Engineering Science 51, 31073112.
Coleman, T.F., Li, Y., 1996. An interior trust region approach for nonlinear
minimization subject to bounds. SIAM Journal on Optimization 6, 418
Greek letters
445.
Edgar, T.F., Himmelblau, D.M., 1988. Optimization of Chemical Process.
(H )i heat of reaction, reaction i, J/mol McGraw-Hill, New York, pp. 171175.
Elnashaie, S.S.E.H., Elshishini, S.S., 1993. Modeling, Simulation and
 void fraction in catalyst bed Optimization of Industrial Fixed Bed Catalytic Reactors. Gordon &
 dimensionless reactor radius Breach, London, pp. 204260.
i effectiveness factor of component i Finlayson, B.A., 1972. The Mathematical of Weighted Residuals and

g viscosity of gas phase, Pa s Variation Principles: With Fluid Mechanics Heat and Mass Transfer.
Academic Press, London, pp. 97147.
b , p , g bed, particle and gas density, kg/m3 Finlayson, B.A., 1980. Nonlinear Analysis in Chemical Engineering.
m , v , l mass density of water for makeup, vapor, McGraw-Hill, New York, pp. 60171.
liquid, kg/m3 Fortescue, T.R., Kershenbaum, L.S., Ydstie, B.E., 1981. Implementation
1 , 2 , 3. ,  process time constants,s of self-tuning regulator with variable forgetting factor. Automatica 17,
831835.
I , D , N. , c controller tuning parameters Froment, G.F., Bischoff, K.B., 1990. Chemical Reactor Analysis and
j
M Thiele modulus for component j Design, second ed. Wiley, New York, pp. 162175.
4286 M. Shahrokhi, G.R. Baghmisheh / Chemical Engineering Science 60 (2005) 4275 4286

Graaf, G.H., Scholtens, H., Stamhuis, E.J., Beenackers, A.A.C.M., 1990. Shampine, L.F., 1994. Numerical Solution of Ordinary Differential
Intra-particle diffusion limitation in low-pressure methanol synthesis. Equations. Chapman & Hall, New York.
Chemical Engineering Science 45, 773783. Skrzypek, J., Lachowska, M., Grzesik, M., Sloczynski, J., Nowak, P.,
Levenspiel, O., 2001. Chemical Reactor Engineering, third ed. Wiley, 1995. Thermodynamic and kinetic of low-pressure methanol synthesis.
New York, pp. 376391. Chemical Engineering Journal 58, 101108.
Lommert, B.J., Graaf, G.H., Beenackers, A.A.C.M., 2000. Mathematical Szukiewicz, M.K., 2001. Approximate model for diffusion and reaction
modeling of internal mass transport limitations in methanol synthesis. in the porous catalyst with mass-transfer resistances. A.I.Ch.E. Journal
Chemical Engineering Science 55, 55895598. 47, 21312135.
Lovik, I., 2001. Modeling, estimation and optimization of the Vanden Bussche, K.M., Froment, G.F., 1996. A steady-state kinetic model
methanol synthesis with catalyst deactivation. Ph.D. Thesis, Norwegian for methanol synthesis and the water gas shift reaction on commercial
University of Science and Technology. Cu/ZnO/Al2O3 catalyst. Journal of Catalysis 161, 110.
Luyben, W.L., 1986. Process Modeling, Simulation and Control for Vanden Bussche, K.M., Neophytides, S.N., Zolotarskii, I.A., Froment,
Chemical Engineering. McGraw-Hill, New York, pp. 5462. G.F., 1993. Modeling and simulation of the reversed ow operation
Parisi, D.R., Laborde, M.A., 2001. Modeling steady-state heterogeneous of the xed-bed reactor for methanol synthesis. Chemical Engineering
gassolid reactor using feedforward neural networks. Computer and Science 48, 33353345.
Chemical Engineering 25, 12411250. Velardi, S., Barresi, A.A., Antonello, A., 2002. Methanol synthesis in a
Perry, R.H., Green, D.W., 1997. Perrys Chemical Engineering Handbook. forced unsteady-state reactor network. Chemical Engineering Science
seventh ed. McGraw-Hill, New York. 57, 29953004.
Poling, B.E., Prausnitz, J.M., OConnell, J.P., 2001. The Properties of Wijngarden, R.J., Kronberg, A., Westerberg, K.R., 1998. Industrial
Gases & Liquids. fth ed. McGraw-Hill, New York. Catalysis, Optimization and Process. McGraw-Hill, New York, pp.
Ramirez, W.F., 1997. Computational Method for Process Simulation, 113140.
second ed. Butterworth-Heinemann, London, pp. 217226. Zhou, X.G., Liu, L.H., Yuan, W.K., 1999. Optimization and control of
Seborg, D.E., Edgar, T.F., Mellichamp, D.A., 1989. Process Dynamics wallcooled xed bed reactor. Chemical Engineering Science 54,
and Control. Wiley, New York, pp. 580581. 27392744.

You might also like