See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/229014592

Scheduling of Multipurpose Batch Processes with Multiple Batches of the

Products

Article in Hungarian Journal of Industry and Chemistry · January 2002

CITATIONS READS

11 150

4 authors:

Tibor Holczinger Jessenia Romero

University of Pannonia, Veszprém University of Cuenca
24 PUBLICATIONS 640 CITATIONS 88 PUBLICATIONS 873 CITATIONS

SEE PROFILE SEE PROFILE

Luis Puigjaner Ferenc Friedler

Universitat Politècnica de Catalunya Széchenyi István University, Gyor
605 PUBLICATIONS 9,667 CITATIONS 223 PUBLICATIONS 4,124 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Luis Puigjaner on 22 January 2014.

The user has requested enhancement of the downloaded file.

 HUNGARIAN JOURNAL
OF INDUSTRIAL CHEMISTRY
VESZPRÉM
Vol. 30. pp. 305 - 312 (2002)

SCHEDULING OF MULTIPURPOSE BATCH PROCESSES WITH MULTIPLE

BATCHES OF THE PRODUCTS

T. HOLCZINGER, J. ROMERO1, L. PUIGJANER1 and F. FRIEDLER

(Department of Computer Science, University of Veszprém,

Veszprém, Egyetem u. 10, H-8200, HUNGARY
1
Chemical Engineering Department, Universitat Politècnica de Catalunya,
E.T.S.E.I.B., Diagonal 647, E-08028 Barcelona, SPAIN)

Received: December 12, 2002

Optimal scheduling of batch processes becomes excessively complex if large number of batches of the products has to be
generated. In most cases, however, the search space of the optimization procedure can be drastically reduced by
eliminating the redundant solutions through additional combinatorial constraints. The proposed approach that generates
these additional constraints is described here for a graph-theoretical method for batch process scheduling, nevertheless, it
can be conveniently embedded into other methods developed for optimal scheduling.

Keywords: multipurpose batch plants, multiple batches, optimal scheduling, S-graph, search space reduction

Introduction A major question in any scheduling algorithm deals

with the time problem representation. Kondili et al. [5]
present a formulation to the modeling of a series of
In multipurpose plants, a variety of production resources
scheduling problems that arises in batch plants, namely
(raw materials, equipment, utilities, manpower) are
the State Task Network (STN). This representation is
shared by a number of processing operations that
based on a uniform time discretization and on the
manufacture several products. The types of operations
assumption that events only happen at the boundaries of
involved can vary from continuous to batch, a fact that
these intervals, which implies the generation of a large
given the inherent flexibility of such types of plants,
number of integer variables in problems of industrial
leads to complex scheduling problems.
relevance. The computational effort has been reduced by
The problem of short-term scheduling of batch plants
reformulating the allocation constants [6] or by heuristic
seeks to determine the optimal strategy for satisfying the
decomposition [7]. These improvements have
production demand of a variety of products at specific
encouraged recent work toward the development of
dates and/or at the end of a given production horizon.
efficient methods of comparable generality based on the
The short-term scheduling is specially relevant for
continuous-time representation, which was first
flexible production networks (multipurpose plants),
introduced by Sahinidis and Grossmann [8].
where the production of individual batches, even for the
The Resource-Task-Network (RTN) representation
same product, does not follow the same pattern but must
proposed by Pantelides [9] was the basis for a
be specified according to an overall performance index
continuous time formulation by Zang and Sargent [10].
and is subject to capacity and time constraints. It
Later, Shilling and Pantelides [11] presented a simpler
involves the allocation of equipment and resources to
RTN formulation. However, the resulting MILP
orders, the sequences of these orders and the route
problem is still very cumbersome to solve.
determination of the material flows through the plant.
Otherwise, an approach that has been traditionally
Short-term scheduling of multipurpose batch plants
used is based on a graph representation combined with a
has received considerable attention over the past two
branch and bound method. These techniques for the case
decades. Many diverse approaches, mathematical
of scheduling, known as edge finding methods [12-14]
formulations and solution algorithms have been
have proved to be very effective for solving special
proposed. Recent reviews can be found in [1-4].
types of job shop scheduling problems. An extensive

Correspondence concerning this article should be addressed to F. Friedler ([email protected])

306

computational study of the problem was also performed 3 7

8 A
by Applegate and Cooke [15], in which the authors 9 S3
1 6 2
develop heuristics for finding feasible schedules, cutting S1 E1 9
4 17
planes for obtaining lower bounds and a specialized S4 9 B
0
branch and bound method. Very recently, a new graph 0
representation called S-graph, appropriate for 5 6 16 7 8
combinatorial algorithms, has been introduced [16, 17]. E1 S7 10 C
S5 14
Once all process tasks are represented in a recipe-graph,
an appropriate search strategy permits to generate the S- Fig.1 S-graph representation of task sequence 2-6 for
graph of the optimal schedule effectively, i.e., a drastic equipment unit E1 with NIS policy
reduction of computation time can be achieved
compared to mathematical programming solution the introduction of an additional node or nodes for the
techniques. In this paper, the properties of scheduling "unlimited" storage unit or units transforms the policy of
problems are further exploited by embedding an the problem to NIS. In the following, only the NIS policy
additional acceleration tool in the solver, which results will be considered.
in an increased computational efficiency needed in the Formally, directed graph G can be given as a pair
solution of large scheduling and re-scheduling problems.
( N , A) where N is a finite set, the set of nodes, and A is
a set of pairs of nodes identifying the arcs of the graph
S-graph representation (i.e., A ⊆ N × N ). In an S-graph, two classes of arcs, the
so-called recipe-arcs and schedule-arcs are specified.
Therefore, an S-graph is given in the form of
The present work is based on the S-graph representation
G ( N , A1 , A2 ) , where N, A1, and A2 denote the sets of
[16] and a corresponding general framework [17] that
can solve different types of production scheduling nodes, recipe-arcs, and schedule-arcs, respectively. It is
problems. In this framework, a node (so-called task supposed that A1 ⊆ N × N , A2 ⊆ N × N , and
node) is assigned to each task given in the recipe for A1 ∩ A2 = ∅ ; furthermore, a nonnegative value, c(i, j ) ,
producing a single batch of a product or products. An assigned to each arc, denotes the weight of the arc (i, j ) .
additional node (so-called product node) is introduced
In practice, if an arc is established from node i to node j,
for each product for technical reasons. It is supposed
i.e., (i, j ) ∈ A1 ∪ A2 , then it is supposed that the task
that there is at least one equipment unit to perform each
task; set Si denotes the set of those equipment units that corresponding to node j cannot start its activity earlier
can perform the task represented by task node i. The than c(i, j ) time after the task corresponding to node i
processing orders of the tasks are given by weighted started.
arcs (so-called recipe-arcs) of the graph. The processing
time of a task may be different for different equipment
units. In this case, the weight of the recipe-arc is the Multiple batches of a product
minimum of the processing times of the plausible
equipment units. In this representation, the value A simple way of describing the multipurpose scheduling
assigned to an arc expresses a lower bound for the problems with more than one batch of the products is by
difference of the starting times of the two related tasks. considering each batch as an individual product. Even
The resultant graph is the recipe-graph for the single though the basic algorithm published in [17] can be used
batch production. For generating multiple batches of the directly for solving this simplistic model, it is not
products, the appropriate part of this S-graph is repeated necessarily efficient enough if the number of batches of
according to the number of batches to get the recipe- the same product is excessively large. Embedding
graph of multiple batches. additional tools into the basic algorithm, however, may
An S-graph can represent both non-intermediate result in a sufficient acceleration for solving large-scale
storage (NIS) and unlimited intermediate storage (UIS) problems. First, a simple example will illustrate the
policies appropriately. For the former, let τj denote the source of inefficiency to be overcome by the embedded
set of those tasks that follow task j according to the tools.
recipe. If equipment unit Ei is assigned to task j and
consecutively to task k, then, a zero weighted arc (or an
arc whose weight is equal to the length of the Example 1
changeover time if applicable), called schedule-arc, is
established from each element of τj to k. For example, 2-
6 is the task sequence of equipment unit E1 shown in Suppose that two batches of product A and one batch of
Fig.1. In this example, E1 is assigned to tasks 2 and 6 product B are to be produced. Product A is produced in
and zero weighted arcs are established from the tasks three consecutive steps that can be performed by any
that follow task 2 (τj={3,4}) to task 6 (consecutive task element of the sets of equipment units S1, S2, and S3,
of task 2). If UIS policy is applied in any part of the respectively. Product B is produced in two consecutive
scheduling problem or globally for the whole problem, steps, where the first step and second step can be
performed by any equipment unit given in sets S4 and
 307
1,
1,1 1,2 … 1,m
(m-1)

2,
2,1 2,2 … (m-1) 2,m

3,
3,1 3,2 … (m-1) 3,m

Fig.2 Recipe-graph of Example 1

…
…

…
…
E1 1 4 (n-1), (n-1), (n-1), (n-1),
Product A
…
1 2 (m-1) m
E2 2 7 5 Product B

E3 3 8 6 n,
n,1 n,2 … (m-1) n,m

E1 4 1
E2 5 7 2 Fig.5 Recipe-graph extended by auxiliary-arcs if n batches of
the product are to be produced
E3 6 8 3

Fig.3 Two technically identical schedules

Fig.6 Recipe-graph extended by auxiliary-arcs in case of

Fig.4 Recipe-graph of Example 1 extended by an auxiliary-arc complex recipe for 2 batches
expressing the order of the two batches of product A

the product excludes the unnecessary permutations of

S5, respectively. The recipe-graph of this example is the batches from consideration. This ordering is
given in Fig.2. represented by arcs in the S-graph as Fig.5 illustrates.
It is assumed that equipment units E1, E2, and E3 Naturally, the optimal solution of the scheduling
are in sets S1, S2∩S4, and S3∩S5, respectively. Two problem with the additional constraints is also an
technically identical schedules are shown in Fig.3; only optimal solution of the original problem. If more than
one of them should be considered, even though both of one initial tasks appear in the recipe of one product, the
them are generated by the basic algorithm. redundancy can be similarly treated (see, e.g., Fig.6).
As far as the example illustrates, the source of The auxiliary-arcs exclude all but one permutations
inefficiency is the possibility of generating different of the batches from the search space of the scheduling
orders of technically identical batches. This redundancy algorithm; nevertheless, further reductions can be
in the set of solutions can, however, be eliminated by achieved in certain cases.
additional constraints. For example, if it is supposed that
the activity denoted by task node 4 cannot start before
the starting time of the activity denoted by task node 1, Single equipment unit for a task
redundancy appearing in Fig.3 is excluded. This
constraint can be conveniently represented by an
Assume that one equipment unit is available to perform
additional arc from node 1 to 4 in the S-graph (see
a task of a product. In this case, the ordering of the
Fig.4). Note that, the additional arc or arcs to exclude
starting times of the task nodes of the second task of the
redundant solutions are not considered as recipe-arcs or
product must be the same as the ordering of the starting
schedule-arcs, they will be cited as auxiliary-arcs and
times of the task nodes of the first task. Similarly, the
illustrated by dotted lines.
ordering of the starting times of the task nodes of the
In general, if more than two batches of a product are
third task is the same as that of the task nodes of the
to be produced, the redundancy is even more serious.
second task, etc. The resultant auxiliary-arcs are shown
For instance, n factorial, practically identical solutions
in Fig.7. Since NIS policy is considered, further
can be generated for n batches of a product. An ordering
sharpening of the constraints can be achieved.
on the starting time of the task nodes of the first task of
308

… 1,
1,1 1,2 1,m
(m-1)

2,
2,1 2,2 … 2,m
(m-1)

3,
3,1 3,2 … (m-1) 3,m

…
…

…
…

(n-1), (n-1), … (n-1), (n-1), Fig.9 Recipe-graph with transformed auxiliary-arcs in case of
1 2 (m-1) m
a complex recipe for two batches: one equipment unit is
n,
available for each task
n,1 n,2 … (m-1) n,m

… 1,
Fig.7 Recipe-graph extended by auxiliary-arcs if one 1,1 1,2 (m-1)
1,m
equipment unit is available to perform a task of a product
2,
2,1 2,2 … 2,m
(m-1)
… 1,
1,1 1,2 1,m
(m-1)
3,
3,1 3,2 … (m-1) 3,m
2,
2,1 2,2 … 2,m
(m-1)

…
…

…
…
3,
3,1 3,2 … (m-1) 3,m

(n-1), (n-1), … (n-1), (n-1),

1 2 (m-1) m
…
…

…
…

n,
n,1 n,2 … (m-1) n,m
(n-1), (n-1), … (n-1), (n-1),
1 2 (m-1) m
Fig.10 Recipe-graph with auxiliary-arcs: multiple equipment
n, units are available for a task, their processing times are
n,1 n,2 … n,m
(m-1) identical

Fig.8 Recipe-graph with transformed auxiliary-arcs if one

equipment unit is available for each task equipment unit may start the operation of a task earlier
than the corresponding task is finished in a batch started
earlier. Nevertheless, if the processing times of all
An equipment unit assigned to task node (i, j) shown equipment units available for a task are identical, there
in Fig.7 (i-th batch, j-th task) can be reassigned to task is an optimal solution with the property that the order of
node (i+1, j) when the activity represented by task node the starting times of the corresponding task nodes of a
(i, j) is performed and the produced material is task is the same as the starting times of the task nodes of
transferred to the equipment unit assigned to node (i, their previous task. This results in auxiliary-arcs, e.g.,
j+1). Consequently, the starting point (i, j) of an shown in Fig.10 and in Fig.11.
auxiliary-arc can be moved one step forward to (i, j+1) In general, the combination of cases given in the
as shown in Fig.8. In addition to the exclusion of the previous and this sections occurs in practice.
redundant permutations of the batches, this additional Furthermore, some tasks in the recipe that can be
transformation of auxiliary-arcs sharpens the bound in performed by multiple equipment units may have
the procedure of generating the optimal solution different processing times. In this case, the initial task
resulting in additional acceleration. nodes can be chained with zero weighted arcs as shown
A complex recipe can be treated similarly, however, in Figs.5 and 6, nevertheless, consecutive tasks can be
if more than one task follow a task in the recipe, an chained only if the following two additional properties
auxiliary-arc is established from each of the task nodes are satisfied.
of the consecutive tasks (see, e.g., Fig.9).
(P1) The task can be performed by one equipment
unit or the task can be performed by multiple
Optional equipment units for a task with identical equipment units with identical processing
processing time times.
(P2) All the preceding tasks satisfy property (P1).
If at least two equipment units are available to perform a If a task can be performed by a single equipment
task, the statement given in the previous section and unit, the related auxiliary-arc is transformed as shown in
illustrated in Figs.8 and 9 is not valid since a concurrent Figs.8 and 9.
 309
Input (recipe,
number of batches)

1 Generation
of the recipe-graph

2 Extension of the
recipe-graph with auxiliary-arcs

3 Scheduling algorithm
based on S-graph

Fig.11 Recipe-graph with auxiliary-arcs for a complex recipe: Solution(s)

multiple equipment units are available for a task; their
processing times are identical
Fig.14 Solution procedure

1 2 3 4 5 16 6 2 9 3 7
S1 S2 S3 S4 S5 1 13 A
E1 E3 E4

6 7 8 9 10
S3 17 4 9 5 15 6 17
S1 S2 S4 S4 14 B
E2 E3 E4

11 12 13 14 15
S2 S3 S4 18 7 8 8 14 9 16
S1 S4 15 C
E4 E1 E2

Fig.12 Recipe-graph of Example 2 10 7 11 11 12 4

E1 16 D
E2 E3

1 2 3 4 5 Fig.15 Recipe-graph of Example 3 for four batches

16
S1 S2 S3 S4 S5

6 7 8 9 10 Fig.14). The algorithm has been realized in C++, its

S3 17
S1 S2 S4 S4
demonstration version is available at
11 12 13 14 15
http://www.dcs.vein.hu/demo/hjic together with
18
S1 S2 S3 S4 S4 illustrative examples. Although the acceleration tool of
the present work has been developed for the S-graph
Fig.13 Recipe-graph of Example 2 with auxiliary-arcs framework, it is not specific to S-graphs and, therefore,
can be fitted to other scheduling algorithms.
Example 2
Example 3
For the recipe-graph shown in Fig.12, suppose that
S1={E1, E2}, S2={E3}, S3={E4, E5}, S4={E1}, and This example is introduced by Sanmartí et al. [17]; four
S5={E2}. Furthermore, the processing time for the first equipment units, E1, E2, E3, and E4, are available to
task, i.e., the processing times related to task nodes 1, 6, generate four products, A, B, C, and D. The recipes of
and 11, is independent of the selection of E1 or E2 while the products are given in Fig.15. The basic algorithm
the processing time of the third task depends on the and its acceleration have been compared by solving this
selection of the equipment unit. The first, the second, the example with different number of batches per products;
fourth, and the fifth tasks satisfy property (P1); the results are shown in Table 1.
moreover, the first, the second, and the third tasks
satisfy property (P2). Therefore, the task nodes of the
first task is to be chained by auxiliary-arcs and the task Example 4
nodes of the second task is to be chained by transformed
auxiliary-arcs (see Fig.13).
This example is introduced by Voudouris and
Grossmann [18]. Five equipment units (stages), E1, E2,
Program realizations and applications E3, E4, and E5, are available to generate four products,
A, B, C, and D. The recipes of the products are given in
Fig.16.
The basic framework algorithm that has been published The basic algorithm and its acceleration have been
in [17] has two main steps: the generation of the recipe- compared by solving this example with different number
graph and the branch-and-bound type scheduling of batches per products, the results are shown in Table 2
algorithm based on S-graph (see steps 1 and 3 in (solved by PC-Pentium 667 MHz). For more than eleven
Fig.14). The present work accelerates the algorithm for batches the basic algorithm did not arrive at the optimal
multiple batches by an additional step, step 2 (see solution in reasonable time (in several hours).
310
Table 1 Comparison of the algorithms in generating an optimal solution

Number of batches Basic algorithm Accelerated algorithm Acceleration

[Sanmartí et al. [17]] [present work] ratio
A B C D CPU time (s)* CPU time (s)*
1 1 1 1 0.17 0.17 1
2 1 1 1 2.42 1.26 1.92
2 2 1 1 11.54 3.02 3.82
2 2 2 1 142.11 20.76 6.85
2 2 2 2 3019.86 81.72 36.95
3 2 2 2 N/A 329.5 N/A
3 3 2 2 N/A 1169.04 N/A
3 3 3 2 N/A 6577.11 N/A

*PC-Pentium 667 MHz

Table 2 Comparison of the algorithms in generating an optimal solution

Number of batches Base algorithm Accelerated algorithm Acceleration

[Sanmartí et al. [17]] [present work] ratio
A B C D CPU time (s)* CPU time (s)*
1 1 1 1 0.17 0.17 1
2 1 1 1 0.55 0.45 1.22
2 2 1 1 0.55 0.22 2.5
2 2 2 1 15.71 3.63 4.32
2 2 2 2 72.22 6.59 10.96
3 2 2 2 309.06 12.32 23.2
3 3 2 2 72.83 6.71 10.85**
3 3 3 2 26827.26 72.56 369.73
3 3 3 3 N/A 233.61 N/A
4 3 3 3 N/A 390.08 N/A
4 4 3 3 N/A 161.59 N/A
4 4 4 3 N/A 2530.26 N/A
4 4 4 4 N/A 7579.39 N/A

*PC-Pentium 667 MHz

**Instance solved by Voudouris and Grossmann [18]

1 8 2 5 3 3
13 A
E1 E4 E5

4 7 5 3 6 4
14 B
E1 E3 E5

7 6 8 9 9 3
15 C
E2 E4 E5

10 4 11 6 12 4
16 D
E2 E3 E5 Fig.17 Recipe-graph of Example 5 for two batches

Fig.16 Recipe-graph of Example 4 for four batches

Voudouris and Grossmann [18] published the result
of the case 3, 3, 2, and 2 batches of products A, B, C,
Note that, the running time of the algorithm is not and D, respectively with 293 second running time using
necessarily monotonic with the number of batches, as GAMS 2.25/Sciconic 2.11 on an IBM/R6000/Power
shown in Table 2, since the number of batches may 530 workstation (see the highlighted row of Table 2 for
affect the structure of the problem that may make a comparison).
smaller problem more complex in some cases.
 311
Table 3 Comparison of the algorithms in generating an optimal solution

Number of batches Base algorithm Accelerated algorithm Acceleration

[Sanmartí et al. [17]] [present work] ratio
A B CPU time (s)* CPU time (s)*
1 1 0.06 0.06 1
2 1 0.16 0.11 1.45
2 2 0.38 0.27 1.4
3 2 1.26 0.49 2.57
3 3 6.76 0.93 7.27
4 3 46.09 1.59 28.99
4 4 386.51 2.96 130.58
5 4 3632.72 5.11 710.9
5 5 41667.74 9.94 4191.93
6 5 N/A 17.81 N/A
6 6 N/A 36.74 N/A
7 6 N/A 68.01 N/A
7 7 N/A 144.46 N/A
8 7 N/A 273.48 N/A
8 8 N/A 565.68 N/A

*PC-Pentium 667 MHz

Example 5 REFERENCES

Two products (product A and B) are to be produced 1. PINTO J. M. and GROSSMANN I. E.: Ind. Eng. Chem.
according to recipe given in Fig.17. Si (i=1,2,..., 9) Res., 1995, 34, 3037
denotes the set of those equipment units that can 2. SHAH N.: Single-and Multisite Planning and
perform task i. The sets are specified as S1={E1}, Scheduling: Current Status and Future Challenges,
S2={E2}, S3={E3}, S4={E4}, S5={E1}, S6={E2}, Foundations of Computer-Aided Process
S7={E3}, S8={E4}, and S9={E5}. Table 3 shows the Operations. AIChE Symposium Series No.320.
result for different number of batches. (Eds. Joseph F. Pekny and Gary E. Blau), American
Institute of Chemical Engineers (AIChE), New
York, 94, 75, 1998
Concluding remarks 3. PEKNY J. and REKLAITIS G. V.: Towards the
Convergence of Theory and Practice: A Technology
By eliminating redundant solutions, additional Guide for Scheduling /Planning Methodology,
combinatorial constraints may drastically reduce the Foundations of Computer-Aided Process
search space of optimal scheduling of batch processes Operations. AIChE Symposium Series No.320.
for generating multiple batches of the products. This (Eds. Joseph F. Pekny and Gary E. Blau), American
type of scheduling problems can be solved more Institute of Chemical Engineers (AIChE), New
effectively with these additional constraints as its York, 94, 91, 1998
applications show for the S-graph based scheduling 4. PUIGJANER L.: Computers Chem. Engng., 1999,
methodology. 23S, S929
5. KONDILI E. C., PANTELIDES C. C. and SARGENT R.
W. H.: A General Algorithm for Scheduling of
Acknowledgement Batch Operations, Proc. 3rd Intl. Symp. On Process
Systems Engng., Sydney, Australia, 62-75, 1988
6. SHAH N., PANTELIDES C. C. and SARGENT R. W.
This project has been financially supported in part by H.: Computers Chem. Engng., 1993, 17, 229
the Hungarian National Science Foundation OTKA T- 7. ELKAMEL A.: Scheduling of Process Operations
029-309 and by the MCyT (Project No. OCCASSION: using Mathematical Programming Techniques, PhD
DPI2002-00856). Thesis, Purdue University, 1993
8. SAHINIDIS N. V., GROSSMANN I. E., FORNARI R. E.
and CHATHRATHI M.: Computer Chem. Engng.,
1991, 15, 255
9. PANTELIDES C. C.: Unified Frameworks for Optimal
Process Planning and Scheduling, Proceedings of
 312

the Second Conference on Foundations of Computer 14. CORMEN T. H., LEISERSON C. E. and RIVEST R. L.:
Aided Operations (FOCAPOII), 235-274, 1994 Introduction to algorithms, The MIT Press, 1997
10. ZANG, X and SARGENT R. W. H.: Computers Chem. 15. APPLEGATE D. and COOKE W.: ORSA Journal of
Engng., 1996, S20, S1287-S1292 Computing, Spring, 1991, 3(2), 149-156
11. SCHILLING G. and PANTELIDES C. C.: Computers 16. SANMARTÍ E., FRIEDLER F. and PUIGJANER L.:
Chem. Engng., 1996, S20, S1221 Computers chem. Engng., 1998, 22, S847-S850
12. ADAMS J., BALAS E. and ZAWACK D.: Management 17. SANMARTÍ E., HOLCZINGER T., PUIGJANER L. and,
Science, 1998, 34, 391-401 FRIEDLER F.: AIChE Journal, 2002, 48(11), 2557-
13. CARLIER J. and PINSON E.: Management Science, 2570
1989, 35, 164-176 18. VOUDOURIS V. T. and GROSSMANN I. E.:
Computers chem. Engng., 1994, 20(11), 1335-1360

View publication stats