You are on page 1of 7

6572 Ind. Eng. Chem. Res.

2002, 41, 6572-6578

PROCESS DESIGN AND CONTROL

Optimal Temperature Control of a Propane Thermal Cracking


Reactor
Mohammad Shahrokhi* and Ali Nejati
Department of Chemical Engineering, Sharif University of Technology, P.O. Box 11365-9456, Azadi Avenue,
Tehran, Iran

In this paper, using a suitable model for propane thermal cracking and based on a cost function,
the optimal temperature profile is obtained. To achieve the optimal temperature trajectory, the
furnace is divided into several zones and proportional-integral-derivative controllers are used
to control the temperature of each zone. According to temperature measurements available,
two cases are considered. In the first case, it is assumed that the furnace wall temperatures at
different points are measured along with the reactor outlet gas temperature. In the second case,
we assume that the reacting gas temperatures inside the reactor at different points are available.
Simulation results indicate that for the first case the optimal temperature profile can be achieved
if an accurate model is available. In the case of model mismatch, a method has been proposed
which results in a near optimal profile. For the second case, two different strategies for set
point tracking are proposed and their performances are compared through simulation. The
controller performance for load rejection is also evaluated.

Introduction dynamic model is developed. Using this dynamic model,


the performances of proportional-integral-derivative
Hydrocarbon thermal cracking is the most important (PID) controllers for controlling the reactor gas temper-
process in the production of olefins. To produce olefins, ature are evaluated through computer simulation. In
feed is introduced into a furnace where thermal cracking the next section, the static model is discussed.
is carried out. The feed of a thermal cracking furnace
can be a variety of compounds such as ethane, propane, Static Model
butane, isobutane, naphtha, and gas oil. Operating
conditions should be such that to produce a maximum Modeling of a propane thermal cracking reactor
yield of the desired material. The main parameters requires information about the reaction mechanisms
affecting the products are the feed composition, reactor and transport processes such as mass, energy, and
gas temperature, and steam ratio. The effect of the feed momentum transfer in the reactor. For simulation
composition for ethylene production is discussed by Lee purposes, the kinetic model developed by Sundaram and
and Aitani.1 Froment8 is used. The individual reactions are given in
As mentioned above, one of the main parameters Table 1.
affecting the product composition is the reactor gas A one-dimensional plug-flow model is used to simulate
temperature. Increasing the temperature will increase the thermal cracking reactor. Because the reactor
the olefin yield and also the rate of coke deposition, diameter compared to the tube length is small (dt/L )
which is undesired. The mechanism of coke formation 10-3) and the gas flow is turbulent, radial temperature
is studied by Kopinke et al.2,3 Coke formation can be variation is neglected. The mass balances for each
decreased by adding some additives.4,5 Suppression of species and energy and pressure drop equations are
coke formation is studied by Grace Chan et al.6 To given by
achieve the maximum yield of a desired olefin and
reducing the expenses due to reactor shutdown for dFi πdt2
decoking, an optimal temperature profile should be
dz
)
4
∑j sijrj (1)
applied along the reactor.

[ ]
In this paper, using the static model proposed by
Sundaram and Froment7,8 and based on an objective dTG 1 πdt2
function proposed by Towfighi et al.,9 an optimal tem- ) Qπdt + ∑i ri(-∆Hi) (2)
perature profile for producing ethylene and propylene
dz ∑
FiCpi 4

( ) ( )
from propane is obtained. When mass and energy
balances and a pressure drop equation are applied, a d 1 1 1 dTG
+ + Fr
dPt dz Mm Mm TG dz
) (3)
* To whom correspondence should be addressed. E-mail: dz 1 Pt
shahrokhi@sina.sharif.ac.ir. Fax: 0119821-6022853. Tel: -
MmPt RG2RT
0119821-6005819. G

10.1021/ie0106783 CCC: $22.00 © 2002 American Chemical Society


Published on Web 11/12/2002
Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002 6573

Table 1. Reaction Scheme for Propane Cracking


no. reaction rate expression K0i, s-1 or L/mol‚s Ei, kJ/mol
1 C3H8 f C2H4 + CH4 r1 ) k1CC3H8 4.692 × 1010 211.7
2 C3H8 T C3H6 + H2 r2 ) k2(CC3H8 - CC3H6CH2/kc2) 5.888 × 1010 214.6
3 C3H8 + C2H4 f C2H6 + C3H6 r3 ) k3CC3H8CC2H4 2.536 × 1013 247.1
4 2C3H6 f 3C2H4 r4 ) k4CC3H6 1.514 × 1011 233.5
5 2C3H6 f 0.5C6 + 3CH4 r5 ) k5CC3H6 1.423 × 109 190.2
6 C3H6 T C2H2 + CH4 r6 ) k6(CC3H6 - CC2H2CCH4/kc6) 3.794 × 1011 248.5
7 C3H6 + C2H6 f C4H8 + CH4 r7 ) k7CC3H6CC2H6 5.553 × 1014 251.1
8 C2H6 T C2H4 + H2 r8 ) k8(CC2H6 - CC2H4CH2/kc8) 4.652 × 1013 272.8
9 C2H4 + C2H2 f C4H6 r9 ) k9CC2H4CC2H2 1.026 × 1012 172.6
10 C4H8 f C6 r10 ) k10CC4H8 6.960 × 107 143.6
11 C3H6 f 3C + 3H2 r11 ) k11CC3H6 ) rc 5.820 × 1014 307.8

Table 2. Characteristics of the Propane Cracking periodically. In the next section, an objective function
Reactor for obtaining an optimal temperature profile will be
length 100 m considered.
diameter 0.108 m
radius of the bend 0.178 m Objective Function
length of the bend 0.553 m
thickness of the wall 0.008 m An objective function which combines the favorable
inlet temperature 600 °C higher product yield and the negative effect of the coking
inlet pressure 3.0 atm abs
propane flow rate 45.99 kmol/h rate is given by Towfighi et al.9 In what follows, their
density of the coke 1.6 kg/L approach is explained briefly. They have introduced the
coke conductivity 0.00154 kcal/m‚s‚K following cost function:
dilution factor 0.406 kg of steam/kg of propane
profit ) income - cost (7)
where j represents the species, i the reaction number,
and sij the stoichiometric coefficients. income ) ethylene production rate ×
The friction factor is computed from ethylene price × net yearly operating time (8)

Fr ) 0.092Re-0.2/dt (4) or

for the straight portion of the tubes and from income ) FC2H4PrC2H4(t0 - niti) (9)

Fr ) 0.092Re-0.2/dt + ζ/πRb (5) where t0 is the yearly operating time, ti is the time
required for decoking, and ni is the number of reactor
for the bends, where ζ is given by shutdowns. The labor and utility costs necessary for

( )( )
decoking operations are expressed as a fraction, λ, of
Λ dt the term representing the benefit lost due to interrup-
ζ ) 0.7 + × 0.35 0.051 + 0.19 (6) tions.
90° Re
The angle of the bend, Λ, is 180° in the present case. cost ) λFC2H4PrC2H4niti (10)
Characteristics of the industrial reactor, used for
simulation, are given in Table 2. According to rate Combining eqs 8-10 yields
equation (11), coke will be produced as reactions go on.
The coke deposition on the wall limits the heat transfer profit ) FC2H4PrC2H4[t0 - niti(1 + λ)] (11)
to the reactor, increases the pressure drop, and even
plugs the reactor in extreme cases. The pressure drop
along the above industrial reactor due to coke formation The frequency of decoking is proportional to the average
is obtained through simulation and is shown in Figure rate of coke formation, which, in turn, is a function of
1. To burn off the coke, the reactor has to be shut down the temperature profile along the reactor. If for a given
reference temperature profile the frequency of decoking
and the average rate of coke formation are denoted by
nir and rjcr, respectively, then we have

ni rjc
) (12)
nir rjcr

where rjc is the average rate of coke formation for a


specific temperature profile and is given by

rjc )
1
L
∫0Lrc(z) dz (13)

Combining eqs 12 and 13 yields

nir
Figure 1. Pressure drop along the reactor vs the coke thickness
at the outlet.
ni )
Lrjcr
∫0Lrc(z) dz (14)
6574 Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002

Figure 2. Optimal temperature profiles based on three objective


functions.

Substituting eq 14 into eq 11 gives


Figure 3. Reactor tube cross section.

profit ) FC2H4PrC2H4t0[1 - β(1 + λ) ∫0 rc(z) dz]


L
(15) the heat flux profile along the reactor is obtained from
the energy balance equation. The furnace wall temper-
where ature profile can be obtained by back calculation. It is
assumed that the dominant heat-transfer mechanism
nirti from the furnace wall to the reactor is radiation, and
β) (16) heat-transfer rate is given by
t0rjcrL

Increasing the reactor gas temperature will increase the q ) AwF12σ(TFW4 - TW4) (19)
olefin yield and consequently the income. On the other
hand, increasing the temperature will also increase the where Aw is the furnace wall area and F12 is the view
rate of coke deposition, which results in higher cost. factor, which is assumed to be 1. TFW and TW are the
When the above objective function is maximized, using furnace wall and reactor outside wall temperatures,
a dynamic programming technique,10 the optimal tem- respectively. The heat-transfer coefficient inside the
perature profile along the reactor is obtained. tube is calculated from the following correlation:
Depending on the desirable products (ethylene, pro-
pylene, ethylene + propylene), the optimal temperature hdt
) 0.023Re0.8Pr0.4 (20)
profiles are obtained by simulations and are shown in k
Figure 2. The industrial temperature profile is also
shown which is close to the case where ethylene and The tube cross section is shown in Figure 3. As can be
propylene are desirable products. seen because of coke formation, there are three thermal
resistances to heat transfer from the outside wall to the
Dynamic Model gas inside the tube. If the overall heat-transfer coef-
ficient and reactor inside heat-transfer area are denoted
To evaluate the performance of a control scheme, it by U and Ac, the heat-transfer rate is given by
is necessary to have a dynamic model of the process.
By application of conservation laws, such a model can q ) AcU(TW - TG) (21)
be obtained. The unsteady mass and energy balances
yield the following equations: where
2 2
∂Fi πdt πdt ∂Ci 1 Ac ln(R3/R2) Ac ln(R2/R1) Ac
-
∂z
+
4
∑j sij rj )
4 ∂t
(17) U
)
2πkwL
+
2πkcL
+
2πR1Lh
(22)

[ ]
∂TG πdt 2 Having q and TG, the furnace wall temperature can be

1 calculated using eqs 19 and 21.
- + Qπdt + ri(-∆Hi) )
∂z ∑FiCpi 4 For control purposes, the furnace is divided into 10

( )
thermal zones separated with radiation shields to avoid
πdt2 ∂ CTTG the direct effect of each zone on the other zones. Using
(18)
4 ∂t FT the industrial gas temperature profile, the furnace wall
temperature profile is calculated and shown in Figure
For simulating the system, eqs 3, 17, and 18 must be 4. The desired furnace wall temperature profile can be
solved simultaneously. It is difficult to measure the gas approximated by 10 step functions as shown in Figure
temperature inside the reactor because of coke deposi- 4. If this temperature profile is applied to the furnace
tion. If the desired furnace wall temperature corre- wall, the resulting reactor gas temperature will be as
sponding to the desired reactor temperature is known, shown by the solid line in Figure 4. As can be seen, this
it would be much easier to measure and control the temperature profile is very close to the desired temper-
furnace wall temperature than the reactor temperature. ature. It is clear that the difference between the above
Having the gas temperature profile inside the reactor, profiles decreases as the number of zones increases.
Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002 6575

Figure 4. Gas and furnace wall temperature profiles. Figure 5. Ultimate temperature profiles controlling the furnace
wall temperature.
Electrical heaters are used to provide necessary heat,
and it is assumed that the transfer function between
the electrical current of the heating element and the
related zone temperature is given by
TFWn K exp(-τds)
) G(s) ) n ) 1, 2, ..., 10 (23)
In τs + 1
For simulation the following parameters are used:
K ) 100, τ ) 5 min, τd ) 2 min
Furnace Wall Temperature Control
It was shown that based on a cost function an optimal
temperature profile for the reactor gas temperature can
be obtained. Having this profile by back calculation, the
corresponding furnace wall temperature is obtained.
Figure 6. Gas outlet temperature vs time.
Discretizing this profile provides the temperature set
point for each zone. For temperature control of each
zone, a PI controller can be used. On the basis of
minimizing the integral of the absolute value of the
error (IAE), all PI controllers are tuned. Using the
heaters’ model and the following equations, controllers
parameters are calculated11

Kc ) ()
a1 τd
K τ
b1
(24)

τ
τI ) (25)
a2 + b2(τd/τ)
where
a1 ) 0.758, b1 ) -0.861, a2 ) 1.02, b2 ) -0.323
Figure 7. Ultimate temperature profiles controlling the furnace
It is assumed that the wall temperature of each zone is wall temperature with model mismatch.
measured. The final profile of the reactor gas temper-
ature after changing from the industrial profile to the furnace wall temperature is controlled instead of the
desired optimal profile is shown in Figure 5. As can be reactor gas temperature and no reactor temperature
seen, the final gas temperature profile is very close to measurement is available. In the case of modeling error,
the optimal profile obtained, if the wall temperature the optimal profile cannot be reached. To show this, the
varied continuously. Variation of the gas outlet tem- previous run is simulated with a 40% error in the heat-
perature is shown in Figure 6. It should be noted that transfer coefficient. The result is shown in Figure 7. To
because of coke formation the overall heat-transfer reduce the discrepancy between optimal and actual
coefficient decreases and therefore the wall set-point profiles, it is proposed to use a virtual furnace wall set
temperature should be updated to keep the optimal point as described below. It is assumed that the inlet
temperature inside the reactor. and outlet gas temperatures are measured. Through the
following procedure, the virtual set-point profile for the
Furnace Wall Temperature Control with Model furnace wall is obtained.
Mismatch 1. Calculate the difference between the gas temper-
The control strategy described in the previous section ature and the optimal temperature at the reactor outlet
is successful if there is no model mismatch, because the denoted by e(L).
6576 Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002

Figure 8. Ultimate temperature profiles controlling the furnace Figure 10. Reactor gas temperature transient response for
wall temperature using the proposed correction scheme. activating controllers simultaneously.

Figure 9. Gas outlet temperature using the proposed correction Figure 11. Reactor outlet gas temperature vs time.
scheme.
tion. However, in this section we assume that the gas
2. Find the error at the end of each zone temperature measurements at the end of each zone are
available. In this case the set point of each controller is
(n∆l
0.5

L )
e(n∆l) ) e(L) n ) 1, ..., 9 the optimal temperature at the outlet of the correspond-
ing zone. To evaluate the performance of the control
where ∆l denotes the length of each zone and n the zone system, the reactor temperature is changed from the
number. industrial profile to the optimal profile. The reactor gas
3. Add the optimal temperature at each zone outlet temperature profiles after 9, 15, 30, 45, and 60 min are
to the corresponding error to obtain the temperature shown in Figure 10. The variation of the reactor outlet
estimate at the end of each zone. temperature versus time is shown by the dashed line
4. Having temperature estimates at zone outlets, in Figure 11. As can be seen, the temperature overshoot
obtain the estimated temperature profile by any stan- is high (about 35 °C). The large overshoot can be
dard curve-fitting method. explained as follows. When the set-point profile is
5. The virtual optimal temperature set-point profile changed, all controllers become active and temperatures
is the mirror image of the estimated reactor tempera- of all zones start changing. The temperature change of
ture with respect to the optimal temperature profile. each zone will affect the next zone. This interaction
Using the virtual optimal profile, find the virtual deteriorates the controllers’ performances. To get a
furnace wall set-point profile by back calculation. better transient response, it is proposed to activate the
6. Apply the virtual set point obtained in step 5, and controller of the first zone and freeze the output of the
measure the new steady reactor outlet temperature. other controllers when the set point is changed. When
7. Step 1-6 is repeated until e(L) becomes less than the temperature error in the first zone becomes less
a predetermined threshold. than a predetermined value, the second controller is
To show the effectiveness of the proposed scheme, it activated and so on. Depending on the predetermined
is applied to the case with a 40% error in the heat- error threshold, it is possible that several consecutive
transfer coefficient, and the result is shown in Figure controllers become activated simultaneously. If a small
8. The variation of the reactor outlet temperature is threshold is chosen, the overshoot is low but the
shown in Figure 9. As can be seen, the final temperature response will be sluggish. On the other hand, if a large
profile is very close to the optimal profile. threshold is selected, the response becomes fast but
overshoot increases. Therefore, a suitable error thresh-
Reactor Temperature Control Using Gas old should be obtained by trial and error through
Temperature Measurements simulation. The results for the same set-point change
As mentioned before, it is difficult to measure the gas but activation of the controllers sequentially are shown
temperature inside the reactor because of coke deposi- in Figure 12. For this run the error threshold is set to
Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002 6577

Figure 12. Reactor gas temperature transient response for


Figure 15. Gas outlet temperature vs time for feed flow rate load
activating controllers sequentially.
change.

feed rate. The effect of load changes on the reactor outlet


temperature is shown in Figure 15. As can be seen, the
control system has been able to handle the load changes
satisfactorily. It should be noted that, for load change,
all controllers are activated simultaneously and activat-
ing controllers sequentially is only recommended for set-
point tracking.

Conclusions
In this paper optimal temperature control of a pro-
pane thermal cracking reactor is studied. On the basis
of conservation laws, a dynamic model for the process
is developed. Using an objective function, the optimal
temperature profile is obtained, and by back calculation,
Figure 13. Manipulated variable at the reactor outlet vs time. the corresponding furnace wall temperature trajectory
is calculated. Reactor gas temperature control by con-
trolling the furnace wall temperature is considered.
Model mismatch has been taken into account, and a
scheme for improving controller performance is pro-
posed. Reactor temperature control using gas temper-
ature measurement is studied, and two different ways
for activating the feedback controllers are proposed for
set-point tracking. Simulation results indicate that
activating controllers sequentially is preferable. Feed-
forward control can also be used if an accurate model
is available. Finally, the controller performance for load
rejection is evaluated through simulation.

Nomenclature
Ac ) area, m2
Figure 14. Ultimate temperature profiles using feed forward with Ci ) concentration of the ith component, mol/L
model mismatch. CT ) total concentration, mol/L
Cp ) heat capacity, kJ/kmol‚K
15 °C. Variation of the reactor outlet temperature versus dt ) tube diameter, m
time is shown by the solid line in Figure 11. As can be Ei ) activation energy, kJ/mol
seen, the temperature overshoot is considerably smaller Fi ) molar flow rate of hydrocarbons, kmol/s
FT ) total molar flow rate, kmol/s
than that in the previous case. Variations of manipu-
G ) total mass flow rate, kg/m2‚s
lated variables for two cases at the last zone are shown
h ) heat-transfer coefficient, W/m2‚s
in Figure 13. Feed-forward strategy can also be used if I ) electrical current, A
an accurate model is available. However, in the case of Kc ) equilibrium constant, mol/L
model mismatch, the optimal temperature profile cannot K ) specific reaction rate coefficient, L/mol‚s
be tracked. The ultimate temperature profile using feed- k ) gas thermal conductivity, W/m‚s‚K
forward control for a 40% error in the heat-transfer kw ) reactor thermal conductivity, W/m‚s‚K
coefficient is shown in Figure 14. As can be seen, the kc ) coke thermal conductivity, W/m‚s‚K
deviation from the optimal profile is considerable. L ) reactor length, m
To evaluate the controller performance for load Mm ) mean molecular weight, kg/kmol
changes, a pulse with a magnitude of 20% of the feed ni ) number of reactors shut down
flow rate and a duration of 100 min is introduced to the ∆H ) heat of reaction, kJ/kmol
6578 Ind. Eng. Chem. Res., Vol. 41, No. 25, 2002

Pt ) total pressure, atm abs (3) Kopinke, F. D.; Zimmermann, G.; Froment, G. F.; Reyniers,
Pr ) Prandtl number G. C. Relative Rates of Coke Formation from Hydrocarbon in
PrC2H4 ) ethylene price, $/kmol Steam Cracking of Naphtha. 3. Aromatic Hydrocarbons. Ind. Eng.
Chem. Res. 1993, 32, 2620.
Q ) heat flux, kJ/m2‚s
(4) Tran, T.; Senkan, S. M. Coke Formation in the Pyrolysis
r ) reaction rate, kmol/m3‚s and Oxidative Pyrolysis of Methane and Methyl Chloride. Ind.
Rb ) radius of the bend, m Eng. Chem. Res. 1994, 33, 32.
R ) gas constant, atm‚m3/kmol‚K (5) Kumar, S. Triethyl Phosphite Additive-Based Fouling In-
Re ) Reynolds number hibition Studies. Ind. Eng. Chem. Res. 1999, 38, 1364.
TG ) reactor gas temperature, °C (6) Grace Chan, K. Y.; Inal, F.; Senkan, S. Suppression of Coke
t0 ) yearly operating time, days Formation in the Steam Cracking of Alkanes: Ethane and
Propane. Ind. Eng. Chem. Res. 1998, 37, 901.
ti ) time required for decoking, days
(7) Sundaram, K. M.; Froment, G. F. Modeling of Thermal
U ) overall heat-transfer coefficient, W/m2‚s Cracking KineticssIII. Radical Mechanisms for the Pyrolysis of
z ) reactor length coordinate, m Simple Paraffins, Olefins, and Their Mixtures. Ind. Eng. Chem.
Fundam. 1978, 17, 174.
Greek Symbols (8) Sundaram, K. M.; Froment, G. F. Kinetics of Coke Deposi-
R ) constant in pressure drop equation, atm/Pa tion in the Thermal Cracking of Propane. Chem. Eng. Sci. 1979,
ζ ) friction factor in the bend 34, 635.
(9) Towfighi, J.; Karimaei, M.; Karimzadeh, R. Application of
Subscript Dynamic Programming Method in Optimal Control of Light
Hydrocarbons Pyrolysis Reactors. Sci. Iran. 1998, 5, 148.
FW ) furnace wall (10) Luus, R. Piecewise Linear Continuous Optimal Control By
Iterative Dynamic Programming. Ind. Eng. Chem. Res. 1993, 32,
Literature Cited 859.
(1) Lee, A. K. K.; Aitani, A. M. Saudi Ethylene Plants Move (11) Smith, C. A.; Corripio, A. B. Principles and Practice of
Toward More Feed Flexibility. Oil Gas J. 1990, 88, 60. Automatic Process Control; John Wiley & Sons: New York, 1997.
(2) Kopinke, F. D.; Zimmermann, G.; Froment, G. F.; Reyniers, Received for review August 14, 2001
G. C. Relative Rates of Coke Formation from Hydrocarbon in Revised manuscript received July 3, 2002
Steam Cracking of Naphtha. 2. Paraffins, Naphthenes, Mono-, Di-, Accepted August 9, 2002
and Cycloolefins, and Acetylenes. Ind. Eng. Chem. Res. 1993, 32,
56. IE0106783

You might also like