You are on page 1of 7

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

net/publication/272237879

The Design and Scheduling of Multipurpose Batch Plants.

Conference Paper · September 1992

CITATIONS READS

0 120

1 author:

Roger William Herbert Sargent


Imperial College London
214 PUBLICATIONS   5,445 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Archiving old papers. View project

Modeling of Continous Separation Process View project

All content following this page was uploaded by Roger William Herbert Sargent on 14 February 2015.

The user has requested enhancement of the downloaded file.


The Design and Scheduling of Multipurpose Batch Plants

by

N. Shah, C.C. Pantelides* and R.W.H. Sargent,


Centre for Process Systems Engineering,
Imperial College, London SW7 2BY

ABSTRACT
In multipurpose batch plants, the production rates are strongly dependent not only
on the level of the available resources, but also on the efficiency of resource utilisa-
tion. Consequently, plant scheduling must be considered simultaneously with the plant
design problem.
This paper considers the design of multipurpose batch plants operating in a cyclic
mode in which the same pattern of operations is repeated at a constant frequency over
long time periods.
The design problem involves the simultaneous determination of the required lev-
els of processing equipment, storage capacity, and utility provisions, as well as a
detailed cyclic operating schedule that fully exploits the inherent flexibility of batch
plants in utilising such resources.
This problem is formulated rigorously as a mixed integer linear program (MILP)
that is solved by a branch-and-bound method, enhanced by exploiting the problem
structure.

1 Introduction
The recently renewed interest in the batch mode of operation has been docu-
mented in a number of articles (Parakrama, 1985, Rippin, 1991, Reklaitis, 1991). In
this paper, we consider plants operating in a periodic mode, with the same pattern of
operations being repeated at constant time intervals (the "cycle time") over a long
time horizon of interest (a "campaign"). This may be contrasted with the short-term
case, where the horizon of interest is short, and schedules do not necessarily follow
any regular pattern.
Previous work in this area (e.g. Loonkar and Robinson, 1970, Yeh and Reklaitis,
1987) assumed simple product recipes and operating procedures which did not take
account of many complications commonly occurring in practice, such as the sharing of
intermediates among different processing steps, the existence of material recycles, the
re-use of equipment during product manufacture etc. The design problem was hierarch-
ically divided into "synthesis" and "sizing" problems, where the synthesis problem
determines the equipment structure (possibly including storage units) while the sizing
sub-problem determines the equipment sizes based on rather restrictive assumptions
regarding the equipment utilisation. This two-level approach, together with the restric-
tions outlined above, implies that solution optimality cannot be guaranteed.
Birewar and Grossmann (1990) solved the synthesis and sizing problems simul-
taneously. Like most other authors, they assumed that each processing step in the
recipe is associated with a unique equipment type, but they also allow merging of con-
secutive steps in the recipe if they are performed in the same vessel. More recently,
Wellons and Reklaitis (1991) and Papageorgaki and Reklaitis (1991) allowed more
flexibility in the parallel utilisation of multiple equipment items for the same process-
ing step. However, no item of equipment is allowed to be used for more than one pro-
cessing step in the same campaign.
In addition to the various restrictions in equipment utilisation, all the work out-
lined above is based on simple product recipes. The work described in this paper is
based on the State Task Network (STN) representation of Kondili et al. (1992). STNs
allow the description of general process recipes as networks of processing steps
("tasks") transforming one or more types of materials ("states") into one or more
other types. Our approach considers simultaneously the design and scheduling

Author to whom all correspondence should he addressed.


problems, allowing complete flexibility in the selection and utilisation of processing
equipment, storage capacity and utilities.

2 Problem Definition and Formulation


To incorporate scheduling in the design problem, we use the periodic scheduling
formulation described by Shah (1992). This is based on establishing an optimal period-
ic operating schedule for a fixed cycle time T. The scheduling problem involves deter-
mining the utilisation of resources (processing equipment, utilities and storage) over
the cycle and the flow of material through the plant.
A discrete time representation is adopted in which the cycle time is divided into a
number of time periods of equal length_ The problem is formulated mathematically as
a mixed integer linear program (MILP), with the following variables characterising
resource utilisation:
• writ 1 if unit j starts processing task i at the beginning of time period t;
0 otherwise.
• Bur = amount of material which starts undergoing task i in unit j at the
beginning of time period t.
• Sst amount of material stored in state s over time period t.
• Uut = total demand for utility u over time period t.
The constraints that must be satisfied during solution include allocation con-
straints to ensure that each item of equipment processing at most one task at any time,
capacity constraints on storage, batchsizes and utility utilisation, material balance con-
straints and cleaning constraints. A key advantage of the formulation is that the con-
straints need only be satisfied over one cycle for them to be valid for the entire cam-
paign, which comprises a sequence of such cycles. This is achieved by enforcing zero
nett accumulation of intermediates over any single cycle, and ensuring that tasks not
completed during a cycle "wrap around" to the start of the cycle. A detailed account
of this formulation is given by Shah (1992).
The design problem is simply a periodic scheduling problem with additional
degrees of freedom, relating to resource existence and capacity, and may be defined as:

Given:
(i) the state-task network description of the recipe;
(ii) the required production rates for each product;
(iii) the characteristics and maximum level of the resources (processing units, storage
vessels, utilities) that should be considered for incorporation in the design;
(iv) maximum and minimum bounds on the cycle time, Tmax and Tmin,
Determine:
(i) the types and levels of the resources to be incorporated in the design;
(ii) the cycle time of operation, T;
(iii) a cyclic operating schedule, i.e. the allocation of equipment to tasks, the batch-
sizes, the task timings and the material flows over the duration of a cycle,
so as to optimise a certain objective function, such as minimisation of total capital
cost.

2.1 Resource Characterisation


Processing units may be available in continuous sizes (normally bounded by
lower and upper limits) or in a number of discrete sizes. In order to accommodate both
these cases, a more general characterisation is used, where each processing unit j is
assumed to belong to one of a number of different "types". The k th type for a given
unit j is characterised by a size range defined by minimum and maximum capacities,
VI'''. and yr. The capital cost of a unit j of type k may depend on its capacity V.
'
and is assumed to be of the form
Kjak ± Kjk Vi
The above form maintains the linearity of the formulation, but even nonlinear cost-
capacity relations may be approximated by a series of linear expressions by splitting
each type into a number of sub-types of narrower capacity ranges.
Similarly, dedicated storage capacity for a state s may be provided by a number
of vessels of different types. The k th vessel type for state s is characterised by
minimum and maximum capacities Kr and Vsiknal, while the cost of a vessel is a
linear function of its size with coefficients 4, and Ksik respectively.
Finally, each utility u may be provided by a number of sources of different types.
We delimit the capacity range for source type k of utility u by Um and yr
respec-
tively, with corresponding cost coefficients Kutik and K.

2.2 Design Variables


In addition to the scheduling variables described above, the following design vari-
ables must be introduced to characterise the selection and capacity of the various pro-
duction resources to be incorporated in the plant:
• E. = 1 if processing unit j is included in the design; 0 otherwise.
• tlik = 1 if processing unit j is of the k th type; 0 otherwise.
• = capacity of processing unit j.
• Eisnk = 1 if vessel n used for storing material in state s is of type k; 0 otherwise.
• Vin = capacity of vessel n used for storing material in state s .
• Cs = the total storage capacity dedicated to state s.
• Eunk = 1 if utility source n providing utility u is of type k; 0 otherwise.
• iir,„ = the capacity of the n th source of utility u .
• Uf" = total available capacity for utility u .

2.3 Constraints
In this section, we introduce only those constraints that are related directly to the
design problem, and in particular the selection and sizing of processing equipment,
storage capacity and utility sources. These constraints must be satisfied simultaneously
with those describing the utilisation of the above resources as considered by the
optimal periodic scheduling problem (Shah, 1992).
We start by considering the processing equipment. A unit, if it is included in the
design, can be of at most one type. Mathematically, this can be expressed as
Eik = Ei tfj (1)
k

The variables Ei are used to modify the equipment allocation constraints in the
scheduling formulation to state that a unit j can only start processing a task i at any
time t (i.e. Wifi=1), if it is actually incorporated in the design (i.e. if Ei=1).
The volume of processing unit j is limited by the minimum and maximum capa-
cities of the type selected for this unit:
Eik ilifen Vj C E Eik Virx V j (2)
k k

The batchsize of a task i in a unit j at any time t is constrained by the unit capa-
city if the unit is allocated to it. Otherwise it must be zero:
(1)YlinV-W-•
ij <
— B••
zit <— OP-
ti "V-W-•
ut V i, jE Ki , t=1..T (3)
where Ki is the set of units that are suitable for performing task i, and (1),Tin and (I)Iyax
are given minimum and maximum capacity utilisation factors respectively. Constraints
(3) involve the bilinear terms V.W.111• ' but these may be reformulated in terms of linear
constraints because of the binary nature of Wiit .
Turning now to the storage capacity that is dedicated for the storage of material
in any particular state s, we note that it is simply the sum of the capacities of the indi-
vidual vessels used for this purpose:
C, = vs, s (4)
n

The capacity of the n th storage vessel depends on the type chosen for this vessel:
E„kVlin 5_ Vs, 5_ Esnk Vs`knax V S ,n (5)
k k

Of course, any storage vessel incorporated in the design can only be of one type:
Esrik 5 I V s ,n (6)
k
The amount of a state s stored in dedicated storage at any time is constrained by the
total storage available:
0 Sst < Cs V s , t (7)
Finally, the total availability of utility u may arise from a number of different
sources and is given by
tium a x = I vun
Vu
n
while the capacities of the individual sources depend on their type:
E„k Vunklin 5_. Vw, Eunk Virklax V it ,n
k k
As for storage vessels, we also have
Eunk u ,n (10)
k
while the total amount of utility u being used at any time t during the cycle cannot
exceed the maximum utility availability:
0 < [Jut < Uum" (11)
The simplest design objective is the minimisation of the capital cost C of the
plant. This will involve contributions from the cost of processing equipment, storage
capacity, and utility sources:
Cal Eik (11 + Kik Vi ) + X I E„k + Ksik Vs, ) + I E„„k (K3 + Kulk V„)
jk snk unk
The above expression involves some bilinear terms (e.g., Eik Vi), but, since Eik is
a binary variable, these products can always be reformulated exactly in linear form.
Other objective functions can easily be formulated. For instance, the nett present
value (NPV) of the plant involves a weighted sum of the capital cost C and the nett
value-added per cycle.
The above constitutes the formulation for a fixed cycle time. If it is desired to
determine the optimal cycle time between given lower and upper bounds T" and
Tmax, then this can be achieved by carrying out a search over all T in this range.
The resulting MILP may involve large numbers of binary variables. A number of
techniques, described by Shah (1992), are utilised to improve the computational
efficiency of a modified branch-and-bound algorithm used for the solution of this prob-
lem.

3 An Example
This example is based on a hydrolubes blending process. The recipe, illustrated as a
STN in figure 1, involves the manufacture of three different products from a number
of different feedstocks. The processing time for each task, and, for tasks with multiple
inputs, the proportions of each input, are also shown in this diagram. For instance, task
Mixing] has a processing time of 6 hours, and involves mixing 62% FeedB with 38%
Intl.
As can be seen from figure 1, the recipe involves a reaction step, a blending step, and
three different mixing steps. The objective of the exercise is to design a plant of
minimum capital cost that will permit production rates of 1.20 te/hr for product Prod1,
1.30 te/hr for Prod2 and 1.23 te/hr for Prod3. Due to current working practices, the
plant will operate periodically with a fixed cycle time of one week (168 hours).
The reactor, being a specialised unit, must be manufactured to the required size.
The latter is allowed to be in the range 30-50 te, with a variable cost of 0.15 units/te.
Reaction 0.999 ElendingB
Read
(24h) •
Prod (oh)

0.001

Figure 1: STN for Example Problem


Blenders are available "off-the-shelf" in discrete sizes of 10, 15, 20, and 25 te, with
corresponding costs of 1.5, 1.6, 1.8, and 2.0 units. Similarly, mixers can be purchased
in sizes of 20, 24, 28 and 32 te, with corresponding costs of 1.8, 2.0, 2.2, and 2.4
units.
Due to the relatively high cost involved, only one reactor vessel is to be con-
sidered for incorporation in the design. However, the latter may comprise up to two
blenders and three mixers. Each mixer can, in principle, be used for carrying out one
or more of the three different mixing tasks during any one cycle. However, before
switching between different mixing tasks, the mixer must be cleaned for 3 hours.
In addition to the processing equipment, the level of intermediate storage capacity
for the two intermediates ReactProd and Intl must be determined. Such capacity can
be procured at essentially any required level in the range 10-50 te, at a cost of 0.025
units/te.
The above problem was solved to within 2.5% of optimality leading to a plant
design with a capital cost of 10.92 cost units. The plant comprises only one processing
unit of each type: a 39.3 te reactor, a 15 te blender, and a 28 te mixer. Storage capaci-
ties of 39.3 te and 10.0 te are included for intermediates ReactProd and Intl respec-
tively.
The Gantt chart summarising the utilisation of the processing equipment over one
cycle is shown in figure 2. The number above each horizontal segment denotes the
task for which the unit is being used (see key in right margin of figure). The number
below the segment is the corresponding batchsize. Ignoring plant start-up and shut-
down, the operating plan over a long campaign can be obtained by the juxtaposition of
a number of such cyclic schedules. It must be noted that this schedule is likely to be
only one of a number of different schedules of equal merit.
It can be seen that the same mixer is used to perform all three different mixing
tasks at different times during the cycle, with cleaning operations (shown as shaded
boxes) being inserted whenever appropriate. Such exploitation of the flexibility of the
equipment is particularly important: earlier design algorithms for batch plants insist
that no equipment item be used for more than one task during a cycle and are there-
fore intrinsically incapable of arriving at such economical designs.
Figure 3 shows the utilisation of storage capacity for the two intermediates over a
cycle (solid line) against the capacity incorporated in the design (dotted line). It can be
seen that, as expected, the capacity availability is limiting at several times during the
cycle — otherwise a cheaper design could have been obtained by reducing the level of
such storage availability.

4 Conclusions
The optimal design of multipurpose batch plants that operate in a periodic mode has
been considered. The proposed approach is based on a simultaneous consideration of
the selection and sizing of all processing equipment, intermediate capacity and utility
provisions, as well as the detailed scheduling of the plant operation. This leads to more
economical plant designs as all plant resources are utilised in the best possible way.
Eck'lament Batch 5checlie Hydrokbes Design Cycle Tine :16Ei Key to Taks
-.
LREACT1CN

Reactor 1 1
I 1 1 I 1 1 1 1 1
4 393 393
4 393 393 393
39.3 37.7 28LEITNG9

Bet!, .1 2i
i1.2+21 1 1111F1111 11,1 F112 1 2 1 20 1
2222222222,2 1 2 1 2 1 2 1 12. 3.144`..G1
10 13 15 10 5 15 10 15 15 15 15 11 10 8 8 5 13 15 15 14 15 1

41y1>=2
Mier .515.3 1 30/313131313.5,015151515151! 4 4 4 4 4 4 4 4
28 11 27 26 19 20 28 28 28 26 28 28 27 28 28 10 19 28 26 28 28 25 28 28 28

5.140.1G3

. . . . . . . . . .
0 15 30 45 60 _/-rTip> 90 105 120 135 150 16..Ta8

Figure 2: Gantt Chart for Example Problem


Storage Requirements for State ReeetPred
39

0
au ra 7113 12U

Storage Reeeiremsrits for State EMI


10

.d J6 4,S Sir it sltir— Ida- teru da du te


rtgure i: 3wrage capacity Utilisation for Example Problem

The limiting feature of the formulation presented is that it considers only single
campaigns involving a number of identical cycles. The design of plants operating in a
multiple campaign mode is currently being investigated.

References
Birewar, D.B. and I.E. Grossmann, "Simultaneous Synthesis, Sizing, and Scheduling
of Multiproduct Batch Plants", Ind. Eng. Chem. Res., 29, 2242-2251 (1990).
Kondili, E, C.C. Pantelides and R.W.H. Sargent, "A General Algorithm for Short-
Term Scheduling of Batch Operations — Part I : Mathematical Formulation", submit-
ted to Comput. Chem. Engng. (1992).
Loonkar, R.Y. and J.D Robinson, "Minimisation of Capital Investment for Batch
Processes", Ind. Eng. Chem. Process Des_ Dev., 9, 625-629 (1970).
Papageorgaki, S. and G.V. Reklaitis, "Optimal Design of Multipurpose Batch Plants",
Ind. Eng_ Chem. Res., 29, 2054-2062 (1990).
Parakrama, R., "Improving Batch Processes", The Chemical Engineer, September,
24-25 (1985).
Reklaitis, G.V., "Perspectives on Scheduling and Planning of Process Operations",
presented at Fourth Intl. Symp. on Proc. Systems Eng_ (1991).
Rippin, D.W.T., "Batch Process Planning", Chem. Engng., May, 101-107 (1991).
Shah, N., "Efficient Scheduling, Planning and Design of Multipurpose Batch
Processes", PhD thesis, University of London (1992).
Wellons, M.0 and G.V. Reklaitis, "Scheduling of Multipurpose Batch Chemical
Plants. 1. Formation of Single-Product Campaigns". Ind. Eng. Chem. Res., 30, 671-
688 (1991).

View publication stats

You might also like