Beruflich Dokumente
Kultur Dokumente
Abstract: This paper deals with the model based optimization of tubular reactors for the
production of LDPE. Due to the high complexity, solving an optimization problem of an
industrial application is not straightforward. Often researchers seek the shortest way to reach
their final goal by going directly to the development of a high-complexity model and optimize
this with respect to a certain objective. Although this approach seems the fastest way to success,
it can be a bumpy road with a lot of dead ends. Therefore, a divide and conquer strategy is
adopted, i.e., first develop a conceptual low-complexity model, set up the optimization problem
and then use the obtained knowledge during the optimization of more complex models. The aim
of this paper is to give the results of the three steps which have to be accomplished in order to
achieve this subgoal. First, the multizone process of LDPE production is modelled as a sequence
of conceptual modules which simulate the steady-state characteristics of one reaction and cooling
zone. Then, this model is fitted to industrial data such that it quantitatively describes the real
process. Finally, a multiple objective design optimization problem is formulated, i.e., where
along the reactor and which amount of initiator has to be injected to maximize the profit at
different economic situations.
434
DYCOPS 2010
Leuven, Belgium, July 5-7, 2010
Fig. 2. Schematic view of a high pressure multizone jacketed tubular reactor for the production of LDPE.
subject to: When assuming (i) there is no axial dispersion, (ii) steady-
state conditions, and (iii) an Arrhenius law dependence
dx of the reaction rate on the temperature, the reactor is
= f [x(), u(), ] (3)
d described by the following first-order differential equations
0 = I[x(0)] (4) with respect to the spatial coordinate z [m]:
0 = S[x(f )] (5) dx1 k0 Ci,in
= (1 x1 )(1 x3 )e 1+x2 (7)
0 C(x(), u(), ) (6) dz v
The vector f represents the dynamic system equations dx2 k0 Ci,in
= (1 x1 )(1 x3 )e 1+x2 + (u x2 ) (8)
on the interval [0, f ], in this case the model equations dz v v
developed in step 1, with consistent initial and terminal dx3 k0
= (1 x1 )(1 x3 )e 1+x2 (9)
boundary conditions given by the vectors I and S. The dz v
vector C indicates path and/or terminal inequality con- subject to the initial conditions:
straints on the states and the controls.
A deterministic multiple shooting method is employed to x1 (0) = 0 (10)
tackle the optimal control problems and to generate the x2 (0) = 0 (11)
resulting Pareto front.
x3 (0) = 0 (12)
3. MODELLING with x1 = (Cin C)/Cin , the dimensionless monomer
concentration C [mole L1 ], x2 = (T Tin )/Tin , the
The LDPE production process is carried out in a multizone dimensionless reactor temperature T [K], x3 = (Ci,in
jacketed tubular reactor with the main monomer feed at Ci )/Ci,in , the dimensionless initiator concentration Ci
the inlet and where the initiators, most often mixtures of [mole L1 ], and u = (Tw Tin )/Tin , the dimensionless
peroxides, are fed to the reactor through lateral injections. jacket temperature Tw [K], respectively. v [m/s] represents
A peak, following the initiator injections, consists out of the fluid superficial velocity, while k0 [s1 ] is the kinetic
a short reaction zones with a high heat generation and a constant. The constants , , and are defined as follows:
cooling zone where the reactor is mainly used as a heat H Cm,in E
exchanger, in order to reach appropriate temperatures for = =
the next initiator addition. In other words, the model for cp Tin RTin
the complete tubular reactor (Fig. 2) is built up as a series 4h Cm,in
= =
of modules, which describe the working principle of one cp d
peak. A 1-peak module is composed of the mass balance with H [J kmole1 ] the heat of reaction (H < 0
of the monomer and initiator and the energy balance along for an exothermic reaction), and [kg m3 ], Cp [J
the length of the reactor. kg1 K1 ], E [J mole1 ], R [J mole1 K1 ], h [W
m2 K1 ], and d [m], the fluid density, the specific
3.1 One-peak module
heat, the kinetic constant, the activation energy, the ideal
gas constant, the heat transfer coefficient, and the reactor
Every 1-peak module represents a classic tubular reactor diameter, respectively. [-] represents the average chain
with one initiator injection point at the inlet and with a length of the produced polymer.
surrounding cooling jacket present to remove the heat of
reaction. In the tubular reactor the supercritical single-
3.2 Multi-peak model
phase ethylene-polyethylene mixture allows the reaction
to take place as a free-radical initiated polymerization.
Instead of incorporating the full complex polymerization The multi-peak model is nothing more than a sequence
mechanism existing of several initiation, propagation and of the 1-peak modules, with different model parameters
termination reactions, the model describes the polymeriza- fitted on industrial data (obtained in Section 4). Every
tion as a single exothermic, irreversible, first-order reaction time a 1-peak module is added to the model an extra state
of the monomer M and an initiator radical I . Assuming variable is defined, i.e., an extra dimensionless initiator
that is the average chain length of the produced polymer concentration because at the beginning of each peak a
P , the simplified reaction looks as follows: different type of initiator can be injected. It also makes the
model more robust in case a mixture is added containing
0 k
M + I P. previous used initiators.
435
DYCOPS 2010
Leuven, Belgium, July 5-7, 2010
0.4 0.7
0.35 0.6
0.3
0.5
Temperature []
Conversion []
0.25
0.4
0.2
0.3
0.15
0.2
0.1 industrial data industrial data
8 model parameters 8 model parameters
0.05 0.1
11 model parameters 11 model parameters
16 model parameters 16 model parameters
0 0
Reactor length [m] Reactor length [m]
Fig. 3. Parameter estimation of three case studies on the industrial data of 4-peak reactor: conversion (left) and
temperature (right).
436
DYCOPS 2010
Leuven, Belgium, July 5-7, 2010
values at the end of each peak. Thus by not using the 5.2 4-peak multiple objective optimization
knowledge gathered in the previous section, not only
incorrect cooling profiles were obtained, but this also The degrees of freedom are (i) the inlet concentrations of
resulted into incorrect -values. Both the second and third the initiator at each of the four injection points as (ii)
case, with 11 and 16 parameters respectively, describe the length of each peak, i.e., the placement of the initiator
accurately the industrial data (except for the temperature injectors along the reactor.
in the first reaction zone, as predicted in Section 4.2). The resulting Pareto front of the 4-peak reactor and the
Based on the MSE values the third case still has a minor corresponding optimal temperature and conversion profiles
advantage on the second case, but note that 50 % more are depicted in Figure 4. For A-values lower than 0.3, i.e.,
unknown parameters have to be estimated. Therefore, in where the produced LDPE is worth a lot more than the
line with the conceptual approach of this research the needed investment cost, identical optimal designs of a 4-
model with 11 parameters is chosen. peak reactor are found. If the design is not limited by
the investment cost (reactor length), maximal conversion
5. MULTIPLE OBJECTIVE OPTIMIZATION of a multipeak reactor is obtained by pushing every peak
to the temperature bounds, i.e. a maximum concentration
First, in Section 5.1 the MOO, based on equation (1) to of initiator is injected such that the reactor temperature
(6), is formulated. Then, in Section 5.2 an overview of just remains beneath the upper bound and the length of a
the optimization results for the 4-peak LDPE reactor is peak is maximized in order to cool down until the mixture
given, followed in Section 5.3 by the results for a range of reaches the jacket temperature. For increasing A-values,
multizone reactors going from 1-peak till 6-peak reactors. the investment cost becomes more important and the
corresponding peak lengths decrease. By shortening the
5.1 Mathematical formulation cooling zone, the reactor temperature is higher at the next
injection point, hence, a smaller amount of initiator can be
Model. The model developed in Section 3.2 with the injected in order to avoid reactor temperature violation
obtained model parameters from Section 4.3 is used. The and the conversion decreases. Around A = 0.6 the four
control variable, the jacket temperature, is assumed to be peaks are reduced to three peaks because the length
constant along the reactor. between two injections becomes to small to sufficiently
cool the mixture. As the A-value reaches 0.9, the reactor
Constraints. The inlet and outlet temperatures are fixed is reduced to one peak. Further increase of the A-value
due to the required conditions of the up- and down stream leads to useless designs, i.e., the reactor is too expensive
processes. To ensure a safe process operation the reactor to build.
temperature is bounded along the reactor. At the end
of the reactor all initiator has to be converted in order
to avoid further polymerization in downstream processes. 0.5
Due to practical reasons the total reactor length is limited
Conversion []
0.4
to 15000 meters, which is still longer than for real LDPE
0.3
reactors. A
0.2
Objectives. The goal is on the one hand to maximize the 0.1
return or in other words to maximize the conversion x1 at
0
the end of the reactor Jconversion . On the other hand the Reactor length [m]
investment cost Jinvest has to remain minimal. This can for
example be quantified by minimizing the reactor length L.
0.5
The multiple conflicting objectives are reformulated to a
Conversion
A 0.3
J = (1 A) Jconversion + Jinvest (14)
K 0.2
A = 0.3
with trade-off coefficient A [0, 1] and a scaling factor K. 0.1 A = 0.6
For each trade-off value A, the solution is known to be A = 0.9
0
Pareto optimal. The trade-off value can easily be linked Reactor length [m]
to real a economic meaning. For instance, when K = 0.4
pconversion,ref
pinvest,ref (conversion/meter tube) is the ratio of the ref-
Temperature []
the actual conversion and investment prices. Varying A Reactor length [m]
from 0 to 1 not only covers the entire economic spectrum,
i.e., from freely available reactor tubes to worthless re- Fig. 4. 4-peak reactor: Pareto front (top) with correspond-
action products, but also yields an approximation of the ing optimal conversion (middle) and temperature pro-
Pareto front. files (bottom).
437
DYCOPS 2010
Leuven, Belgium, July 5-7, 2010
5.3 Multi-peak multiple objective optimization return, i.e., maximize the produced LDPE (conversion)
while minimizing the investment cost. The two conflicting
0,9
costs are reformulated to a single objective by a weighted
sum, where trade-off coefficient A can be linked to the
0,8 market situation. For high A-values the LDPE is worthless
0,7 compared to the investment and small reactor with only
one peak is designed. For decreasing A-values, the market
conversion []
0,6
price of LDPE increases and the optimizer automatically
0,5 adds extra peaks in order to increase the conversion. The
0,4 Pareto fronts obtained for the multipeak models up to
six peaks already helps the designer in comparing the
0,3
2 peak (dis)advantages of adding an extra injector versus varying
0,2 3 peak the initiator amount and prolonging the peak lengths.
4 peak
0.1 5 peak
6 peak ACKNOWLEDGEMENTS
0,0
reactor length [m]
Work supported in part by SABIC (The Netherlands) and Projects
OT/09/025/TBA and EF/05/006 (Center-of-Excellence Optimiza-
Fig. 5. Comparison of the Pareto fronts of the multi-peak tion in Engineering) of the Research Council of the Katholieke
reactors up to six injections points. Universiteit Leuven, Project KP/09/005 (SCORES4CHEM) of the
Industrial Research Council of the Katholieke Universiteit Leuven,
In Section 5.2 the Pareto front is obtained for a reactor and the Belgian Program on Interuniversity Poles of Attraction, initi-
with at most four initiator injections. To get a more global ated by the Belgian Federal Science Policy Office. J. Van Impe holds
overview, the optimization is performed with all multi- the chair Safety Engineering sponsored by the Belgian chemistry
peak reactor models up to six injections. The resulting and life sciences federation essenscia. The scientific responsibility is
Pareto fronts are depicted in Fig. 5. Despite the use of assumed by its authors.
different multi-peak models, the results of the optimization
for high A-values are practically the same, i.e., although REFERENCES
multiple peaks are possible, the length of all peaks but Agrawal, N. and Han, C. (1975). Analysis of the high pressure
one is set to zero due to the high investment cost of polyethylene tubular reactor with axial mixing. AIChE Journal,
the tubes. As the A-value decreases, the produced LDPE 21(3), 449465.
becomes more valuable and the advantage of additional Asteasuain, M., Tonelli, S., Brandolin, A., and Bandoni, J. (2001).
peaks towards higher conversion becomes clear. It is easy Dynamic simulation and optimisation of tubular polymerisation
reactors in gPROMS. Computers and Chemical Engineering, 25,
to deduct from Fig. 5 which multi-peak reactor is preferred
509515.
at which market situation and what the additional profit is Bokis, C., Ramanathan, S., Franjione, J., Buchelli, A., Call, M.,
of adding an extra peak. Although no values are given due and Brown, A. (2002). Physical properties, reactor modeling and
to confidentiality reasons, it should be noted that for the polymerization kinetics in the low-density polyethylene tubular
current market situation the resulting optimized reactor reactor process. Industrial and Engineering Chemistry Research,
design based on the conceptual model does not differ much 41(5), 10171030.
from the real-life reactor. Brandolin, A., Lacunza, M., Ugrin, P., and Capiati, N. (1996). High
pressure polymerization of ethylene. An improved mathematical
model for industrial tubular reactors. Polymer Reaction Engi-
6. CONCLUSION neering, 4(4), 193241.
Brandolin, A., Valles, E., and Farber, J. (1991). High pressure
In this paper the first part of the divide and conquer tubular reactors for ethylene polymerization optimization aspects.
strategy, i.e., the conceptual modelling and optimization Polymer Engineering and Science, 31(5), 381390.
of LDPE tubular reactors, is described in order to obtain Hafele, M., Kienle, A., Boll, M., Schmidt, C.U., and Schwibach,
some generic insight on the application and to use the M. (2005). Dynamic simulation of a tubular reactor for the
profiles as good initialization for the optimization of the production of low-density polyethylene using adaptive method of
lines. Journal of Computational and Applied Mathematics, 183,
more complex models. First, the LDPE multizone high- 288300.
pressure tubular reactor is modelled as a series of steady- Hafele, M., Kienle, A., Boll, M., Schmidt, C.U., and Schwibach, M.
state 1-peak modules. A 1-peak module is composed of (2006). Modeling and analysis of a plant for the production of low
the mass balance of the monomer and initiator and the density polyethylene. Computers and Chemical Engineering.
energy balance along the length of the reactor. Simulations Kiparissides, C. (1996). Polymerization reactor modeling: a review of
of this conceptual model proved that it was capable of recent developments and future directions. Chemical Engineering
qualitatively describing the typical profiles of the peaks. Science, 51(10), 16371659.
In a second step the model parameters are estimated on Yao, F., Lohi, A., Upreti, S., and Dhib, R. (2004). Modeling,
real industrial data. Despite the conceptual approach, the simulation and optimal control of ethylene polymerization in non-
isothermal, high-pressure tubular reactors. International Journal
simulation results of the multi-peak model with obtained
of Chemical Reactor Engineering, 2(2).
parameter values fit almost accurately the data. Only Yoon, B. and Rhee, H. (1985). A study of the high-pressure polyethy-
in the beginning of the first peak a deviation between lene tubular reactor. Chemical Engineering Communications, 34,
simulation and data is detected, but this is due to the 253265.
assumption of only injecting one initiator instead of a mix- Zhou, W., Marshall, E., and Oshinowo, L. (2001). Modeling LDPE
ture. Finally, with the obtained model a multi-objective tubular and autoclave reactors. Industrial and Engineering
optimization problem is formulated in order to maximize Chemistry Research, 40, 55335542.
438