Professional Documents
Culture Documents
, A. Jahanmiri
School of Chemical and Petroleum Engineering, Shiraz University, Mollasadra Avenue, Shiraz, Fars 713481154, Iran
a r t i c l e i n f o
Article history:
Received 26 December 2009
Received in revised form 2 September 2010
Accepted 29 November 2010
Available online 4 December 2010
Keywords:
Fixed-bed reactor
Heterogeneous model
DME reactor control
a b s t r a c t
Dimethyl ether (DME) as a clean fuel has attracted the interest of many researchers from both industrial
communities and academia. The commercially proven process for large scale production of dimethyl
ether consists of catalytic dehydration of methanol in an adiabatic xed-bed reactor. In this study, the
industrial reactor of DME synthesis with the accompanying feed preheater has been simulated and con-
trolled in dynamic conditions. The proposed model, consisting of a set of algebraic and partial differential
equations, is based on a heterogeneous one-dimensional unsteady state formulation. To verify the pro-
posed model, the simulation results have been compared to available data from an industrial reactor
at steady state conditions. A good agreement has been found between the simulation and plant data. A
sensitivity analysis has been carried out to evaluate the inuence of different possible disturbances on
the process. Also, the controllability of the process has been investigated through dynamic simulation
of the process under a conventional feedback PID controller. The responses of the system to disturbance
and setpoint changes have shown that the control structure can maintain the process at the desired
conditions with an appropriate dynamic behavior.
2010 Elsevier B.V. All rights reserved.
1. Introduction
Dimethyl ether as a basic chemical product is mainly consumed
inindustries as a solvent and propellant invarious aerosol products
[1]. The similarity of some physical properties of DME, liqueed
petroleum gases and diesel fuel has presented DME as a clean fuel
that can be used in diesel engines, power generation, and other
purposes such as a substitute for LNG [2]. It can be produced from
a variety of feed-stock such as natural gas, crude oil, residual oil,
coal, waste products and bio-mass [3]. Although the commercially
proven technology for DME production is through dehydration of
pure methanol in an adiabatic xed-bed reactor, many researchers
are working on the direct conversion of syngas to DME over a dual
heterogeneous catalyst [4].
DME can be produced in packed bed or slurry reactors. Although
slurry reactors have the advantage of better heat removal capa-
bility by coil or jacket and good productivity and good interface
contacting due to small solids particle size, they suffer from signif-
icant mass transfer resistance due to the use of inert liquid phases
such as waxy liquid [5]. At present, DME is commercially produced
through methanol dehydration using acidic porous catalysts in adi-
Corresponding author. Tel.: +98 711 2303071; fax: +98 711 6287294.
E-mail addresses: eslamlo@shirazu.ac.ir, reslamlou@gmail.com
(R. Eslamloueyan).
abatic packed bed reactors [6]. Due to simplicity and lower costs,
adiabatic xed bed reactors can be the rst choice for catalytic pro-
cesses having lowor intermediate heat of reactions [7]. However, in
case of highly endothermic or exothermic reactions, the problems
of reaction extinguishing or catalyst sintering may happen [8]. The
uidized bed reactors have been suggested by some researchers
as a perfect reactor for DME synthesis [9,10]. Fluidized bed reac-
tors showed the better heat removal characteristics due to freely
moving catalyst particles in the bed. However, collision between
catalyst particles and the reactor wall causes loss of catalyst.
Simulation of packed bed reactors is not an easy task, since sys-
tems of nonlinear partial differential equations have to be solved
along with nonlinear algebraic relationships [11]. Modeling and
simulation of xed-bed reactors have been done for many indus-
trial andpilot scales reactors. For instance, Shahrokhi et al. modeled
and simulated the dynamic behavior of a xed bed reactor for
methanol production from syngas and proposed an optimizer to
maximization of the production rate [12]. Jahanmiri and Eslam-
loueyan modeled and simulated the isothermal xed-bed reactor
for methanol productioninindustrial scale[13]. Theycalculatedthe
optimal temperature prole of the shell side to maximize methanol
production.
Fixed-bed DME synthesis reactors have been investigated
through a number of research studies. Lee et al. modeled and ana-
lyzed DME production fromsyngas in a xed-bed reactor at steady
state condition [14]. Nasehi et al. simulated an industrial adiabatic
0255-2701/$ see front matter 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.cep.2010.11.013
86 M. Farsi et al. / Chemical Engineering and Processing 50 (2011) 8594
Table 1
The reaction kinetics and equilibrium constants.
Parameter k =A(i)exp
B(i)/RT
A(i) B(i)
k
0
3.7exp
10
1,05,000
K
H
2
O
0.84exp
1
41,100
K
CH
3
OH
7.9exp
4
7050
xed-bed reactor for DME production from methanol dehydration
at steady state condition, and showed that the difference between
one and two dimensional models is negligible [15]. Omata et al.
studied DME production from syngas in a temperature gradient
reactor for overcoming both the equilibrium limit of the reaction
at hightemperature andlowactivity of the catalyst at lowtempera-
ture [16]. Then, theyoptimizedthe reactor for higher COconversion
by a combination of genetic algorithmand neural networks. Fazlol-
lahnejad et al. investigated methanol dehydration in a bench scale
adiabatic packed reactor [17]. The reactor was packed with 1.5mm
-Al
2
O
3
pellets as catalyst and operated at atmospheric pressure.
They investigated the effects of weight hourly space velocity and
temperature on methanol conversion.
Although many research studies have been carried out on the
dynamic simulation and control of xed bed reactors, we could not
nd any work in this eld for DME synthesis from methanol dehy-
dration. Fixed-bed catalytic reactors are often difcult to control
due tononlinearity, process deadtime andpositive feedbackof feed
preheating by the reactor efuent stream[18]. Jorgensenpresented
a good reviewfor dynamic modeling and control of xed bed reac-
tors [19]. Aguilar et al. developedarobust PIDcontrol lawinorder to
control the outlet temperature of a uid catalytic cracking reactor
[20]. The controller is synthesized using an inputoutput lineariz-
ing control lawcoupled to a proportional-derivative reduced order
observer to infer on-line the unknown heat of reaction. This control
strategywas robust against model uncertainties, noisytemperature
measurements and set point changes.
In this paper, we develop an unsteady heterogeneous model
for dynamic simulation of an industrial DME synthesis reactor and
designa conventional feedbackcontrol systemtomaintainthe inlet
stream temperature at the desired value.
2. Reaction kinetics
The reaction equation of DME synthesis from dehydrogenation
of methanol is as the following.
2CH
3
OH CH
3
OCH
3
+ H
2
O, H
298
= 23.4 (kJ/mol) (1)
In this work, the rate expressions have been selected from the
kinetic model proposed by Ber ci c et al. [21].
r =
k K
2
CH
3
OH
((C
CH
3
OH
(C
C
2
H
6
O
C
H
2
O
)/K
eq
))
(1 +2(K
CH
3
OH
C
CH
3
OH
)
.5
+K
H
2
O
C
H
2
O
)
4
(2)
where,
ln(K
eq
) = 086 log T +
3138
T
+1.33 10
3
T 1.23
10
5
T
2
+3.5 10
10
T (3)
k = A(i) exp
B(i)/RT
(4)
C
i
is the molar concentration of component i in mol/m
3
. k
1
,
K
CH
3
OH
and K
H
2
O
are the reaction rate constant and the adsorption
equilibrium constant for methanol and water vapor, respectively,
which are tabulated in Table 1. Commercially, -Al
2
O
3
catalyst is
used in methanol dehydration reaction. Also, the characteristics of
Table 2
Equipment and catalyst parameters.
Parameter Value
Reactor
Catalyst density (kgm
3
) 2050
Catalyst void fraction 0.2
Specic surface area of catalyst pellet 673
Catalyst diameter (m) 0.317510
2
Ratio of void fraction to tortuosity of catalyst particle 0.066
Reactor Length (m) 8
Reactor diameter (m) 4
Bed void fraction 0.5
Catalyst density (kgm
3
) 2010
Density of bed (kgm
3
) 1005
Heat exchanger
Tube side diameter (m) 1.90510
2
Tube thickness (m) 2.1110
3
Shell diameter (m) 0.615
Overall heat transfer coefcient (Wm
1
K
1
) 325
Heat exchanger length (m) 4.57
the catalyst pellet and the reactor design specications are sum-
marized in Table 2.
3. Process description
DMEproductionloopconsists of anadiabatic packedbedreactor
as well as a shell and tube heat exchanger as depicted in Fig. 1.
The inlet feed to the reactor is preheated in the heat exchanger
through heat transfer to the outlet products from the reactor. The
reactor product is transferred to DME purication unit, and DME is
separated fromwater and methanol. Also, the unreacted methanol
is recycled to the reactor.
The developed model of this process includes modeling equa-
tions for both the reactor and the feed preheater.
3.1. Reactor model
In this study, a one dimensional heterogeneous model has been
considered for steady state simulation of the process. The basic
structure of the model is composed of heat and mass conservation
equations coupled with thermodynamic and kinetic relations as
well as auxiliary correlations for prediction of physical properties.
In the dynamic modeling of the reactor the following assumptions
have been considered.
- No heat loss to the surrounding from the reactor wall
- Plug ow pattern and negligible concentration and temperature
variations in radial direction
- Uniform temperature within each catalyst pellet (Biot num-
ber =0.0125)
Biot number is a dimensionless number dened in conduction heat
transfer, and gives a simple index of the ratio of the heat transfer
Fig. 1. Process ow diagram of DME production.
M. Farsi et al. / Chemical Engineering and Processing 50 (2011) 8594 87
Table 3
Feed specication of the industrial DME reactor.
Parameter Value
T (
C) 260
P (bar) 18.2
Mass ow rate (kg/s) 50.81
Mass fraction (%)
Methanol 94%
Water 1%
DME 5%
resistances inside and at the surface of a body. For the Biot numbers
smaller than0.1, the heat conductioninside the body is muchfaster
than the heat convection away from the surface, and the tempera-
ture gradient is negligible inside of the body. To justify the rst two
assumptions, it should be notied the industrial reactor considered
inthis study have insulationthat causes the heat loss fromthe reac-
tor wall to the environment negligible in comparison to the heat
release in the reactor, and also because of high Reynolds number
(about 10,000), the assumption of plug owis reasonable. Further-
more, Nasehi et al. simulated the reactor at steady state condition
by using a two-dimensional heterogeneous model with considera-
tion of the heat loss from the insulation to environment [22]. The
results of this simulation showed that the temperature difference
between the reactor center and the reactor wall is at most 4
C. So,
the heat loss has a negligible inuence on the simulation results in
one-dimensional plug ow model.
The mass and energy balances for the gas and solid phase are
expressed by the following equations:
Gas phase:
C
i
t
= u
s
C
i
z
k
gi
a
v
(C
i
C
s
is
) (5)
g
c
pg
T
t
= u
s
g
c
p
T
z
+h
f
a
v
(T
s
s
T) (6)
Solid phase:
(1 )
C
is
t
= k
gi
a
v
(C
i
C
s
is
) +
i
Bs
r
i
(7)
(1 )
Bs
c
ps
T
s
t
= h
f
a
v
(T
s
s
T)
i
Bs
(H)r
i
(8)
The pressure drop in the bed is calculated by Ergun equation
[23]:
dp
l
= 150
(1 )
2
u
c
d
2
3
+1.75
(1 ) u
2
c
d
3
(9)
The initial and boundary conditions for these equations are
given below. Also, the feed specication of the industrial DME reac-
tor is reported in Table 3.
Boundaryconditions : C
Z=0
= C
0
, T
Z=0
= T
0
, P
Z=0
= P
0
Initial conditions : C
i
Z,t=0
= C
ss
i
, T
Z,t=0
= T
ss
i
3.2. Heat exchanger model
There are no phase changes in the heat exchanger and this
unit is small in compare with the reactor unit. The shell and tube
heat exchanger used for preheating feed stream is modeled at the
dynamic condition by the following equations:
dT
hot
dt
= u
g
dT
hot
dz
D
i
A
hot
C
p
U(T
hot
T
cold
) (10)
dT
cold
dt
= u
g
dT
cold
dz
+
D
o
A
cold
C
p
U(T
hot
T
cold
) (11)
The overall heat transfer coefcient in the heat exchanger is
calculated from
1
U
=
1
h
i
+
A
i
ln(D
o
/D
i
)
2LK
w
+
A
i
A
o
1
h
o
(12)
In Eq. (12), the heat transfer coefcient in the shell side (h
o
) and
the tube side (h
i
) are calculated using the following equations [24]:
h
o
D
o
k
f
= 0.33(Re)
0.6
_
C
p
k
_
1/3
(13)
h
i
D
i
k
= 0.023(Re)
0.8
_
C
p
k
_
1/3
_
w
_
0.14
(14)
3.3. Auxiliary equation
Auxiliary equations are used for prediction of the model param-
eters. In these equations
i
is the effectiveness factor which is
dened as actual reaction rate per particle to theoretical reaction
rate based on the external pellet surface concentration, which is
obtained from dusty gas model calculations [25]. It can be calcu-
lated from the following equation:
i
=
_
r
i
dV
p
V
p
. r
i
surface
(15)
In the heterogeneous model the physical properties of chemi-
cal species and overall mass and heat transfer coefcients between
catalyst solid phase and gas phase must be estimated. The over-
all mass transfer coefcient between solid and gas phase has been
obtained from the correlation proposed by Cussler [26].
kg
i
= 10
3
(1.17Re
0.42
Sc
i
0.67
u
g
) (16)
In Eq. (16), the Sc is calculated for components along the reac-
tor. Mass transfer diffusion coefcient for each component in the
mixture has been estimated by:
D
im
=
1 y
i
(y
i
/D
ij
)
(17)
Binary mass diffusion coefcient has been determined by the
following equation [20]:
D
ij
=
10
7
T
3/2
_
(1/M
i
) +(1/M
j
)
P(
3/2
ci
+
3/2
cj
)
2
(18)
The overall heat transfer coefcient betweensolidandgas phase
(h
f
) has been predicted from [27]:
h
f
C
p
_
C
p
K
_
2/3
=
0.458
B
_
ud
p
_
0.407
(19)
The overall heat capacity inEq. (19) depends onthe temperature
and composition, and is calculated using Eqs. (20) and (21):
C
p(i)
= a +bT +cT
2
+dT
3
(20)
C
p
=
y
i
C
p(i)
(21)
Table 4 presents the values of a, b, c and d in Eq. (20).
4. Numerical solution
To solve the set of nonlinear partial differential equations (PDE)
obtained fromdynamic modeling, the reactor length is divided into
equal discrete intervals, and by using nite difference method the
PDEs are converted into a set of ordinary differential equations
88 M. Farsi et al. / Chemical Engineering and Processing 50 (2011) 8594
Table 4
Parameters of heat capacity coloration.
a b c 10
5
d10
9
DME 0.18 17.02 5.23 1.92
CH
3
OH 21.15 0.07 2.59 0.29
H
2
O 32.24 0.002 1.06 3.6
(ODE) in time domain. The set of equations are solved by 4th order
RungeKutta [28].
Before carrying out dynamic simulation, the stationary con-
dition of the system should be obtained through solving the
governing steady-state equations. The aim of performing steady
state simulation of the DME reactor is to determine the con-
centration and temperature proles along the reactor at normal
operation. These steady state proles are used as the initial con-
ditions of the unsteady state PDEs. After rearranging the modeling
equations for the steady state condition, a set of ordinary differen-
tial equation is obtained. This set of equations was solved by the
method of 4th order RungeKutta.
5. Control structure
The main objective of the control system of DME reactor is to
maintain the operating condition at the optimumdesigned state in
spite of probable disturbances. Since DME is produced through an
exothermic reaction in a xed bed adiabatic reactor, the control of
feed temperature is crucial for keeping the conversion at its maxi-
mum level. The proposed control structure for a typical industrial
DME reactor is showninFig. 2. The reactor feedstreamis preheated
in a shell and tube heat exchanger with the hot reactor efuent. A
bypass streamaroundthis heat exchanger is manipulatedtocontrol
the reactor feedtemperature. Inthis study, a conventional feedback
PID algorithm carries out the load rejection and set-point tracking
tasks.
6. Results and discussion
6.1. Steady state simulation
In this section, steady state simulation results of DME reac-
tor have been presented. Assuming an adiabatic tubular reactor
with plug ow pattern, the reactor has been simulated based on
a one dimensional heterogeneous model. Modeling validation is
done through comparing the simulation results with the industrial
Table 5
Comparison of the steady state simulation results of the reactor with plant data.
Simulation
result
Plant data Absolute
error
DME exit mole ow rate (kmol/h) 2457 2506 1.95%
MeOH exit mole ow rate (kmol/h) 940.6 937.7 0.31%
Outlet temperature (K) 652.2 644 1.27%
0 1 2 3 4 5 6 7 8
0
1000
2000
3000
4000
5000
6000
Length (m)
M
o
l
e
F
l
o
w
(
k
m
o
l
h
r
-
1
)
DME
MEOH
Water
Fig. 3. The proles of mole owrates of DME, Methanol andwater along the reactor.
reactor data provided by Petrochemical Research and technology
Company which is a subsidiary of National Iranian Petrochemical
Company. Table 5 indicates the absolute errors of the steady-state
simulation results in comparison to available data of the industrial
reactor. This table shows that the proposed model has been able to
predict the exit concentrations and temperature very well.
Fig. 3illustrates the molar owrate of DME, methanol andwater
along the reactor length at steady state condition.
Themolefractionof DMEalongthereactor insolidandgas phase
predicted by the heterogeneous model has been compared in Fig. 4.
Near the end of the reactor, where the reaction equilibrium is pre-
vailed, the difference between gas and solid phase concentrations
is vanished.
Fig. 2. Schematic of the control structure.
M. Farsi et al. / Chemical Engineering and Processing 50 (2011) 8594 89
0 1 2 3 4 5 6 7 8
0.05
0.14
0.23
0.32
0.41
0.5
Length (m)
M
o
l
e
F
r
a
c
t
i
o
n
Gas Phase
Solid Phase
Fig. 4. DME mole fraction in solid and gas phase.
Also, the predicted temperature prole along the reactor
obtained from the simulation is shown in Fig. 5. The temperature
approaches to equilibrium temperature due to heat generation by
reaction, and remains constant.
The calculatedaverage heat transfer coefcient betweengas and
catalyst phase (h
f
) in Eq. (19) is about 1170Wm
2
K
1
. Also, the
average estimated viscosity and molar heat capacity of the reac-
tionmixture are about 1.610
7
kgm
1
s
1
and69.5Wmol
1
K
1
,
respectively. The effective diffusion coefcients of components in
the mixture and mass transfer coefcients for components are
calculated along the reactor. The average effective diffusion and
average mass transfer coefcients of DME, MeOH and water vapor
in the mixture are presented in Table 6.
6.2. Open loop dynamic simulation
The open loop behavior of the reactor has been considered in
two levels. In the rst level the behavior of the reactor without its
preheater has been studied. The second level investigates both of
the reactor and the related preheater as a heat integrated process.
0 1 2 3 4 5 6 7 8
520
555
590
625
660
Length (m)
T
e
m
p
e
r
a
t
u
r
e
(
K
)
Fig. 5. Predicted temperature prole along the reactor.
Table 6
Calculated average mass transfer and effective diffusion coefcients in the reactor.
D
DME-mixture
(ms
1
) 0.173710
6
D
MeOH-mixture
(ms
1
) 0.174510
6
D
water-mixture
(ms
1
) 0.268510
6
kDME (m
2
s
1
) 0.0036
k
MeOH
(m
2
s
1
) 0.0036
kwater (m
2
s
1
) 0.005
0
10
20
30
40
50
0
2
4
6
8
520
560
600
640
680
Time (sec)
Length (m)
T
e
m
p
e
r
a
t
u
r
e
(
K
)
Fig. 6. Dynamic temperature prole along the reactor for 10
C
increase in the feed temperature. In Fig. 6, dynamic variation of
the temperature along the reactor length has been illustrated in
a three-dimensional diagram. Fig. 7 represents time variations of
DME concentration along the reactor bed.
As shown in Fig. 8, the response of the outlet temperature of
the reactor for this disturbance has a time delay about 15s, and it
reaches to the new steady state point in 50s.
0
10
20
30
40
50
0
2
4
6
8
0
0.1
0.2
0.3
0.4
0.5
Time (sec)
Length (m)
D
M
E
m
o
l
e
f
r
a
c
t
i
o
n
Fig. 7. Dynamic DME mole fraction prole along the reactor for 10
C step change
in the inlet stream to the reactor.
90 M. Farsi et al. / Chemical Engineering and Processing 50 (2011) 8594
0 10 20 30 40 50
651
653
655
657
659
661
Time (sec)
T
e
m
p
e
r
a
t
u
r
e
(
K
)
Fig. 8. The outlet temperature prole from the reactor to 10