Advanced Information Processing in Automatic Control (AIPAC '89)

527 Pages·1990·47.215 MB·English

by  R. Husson

IFAC SYMPOSIA SERIES
Editor-in-Chief
JANOS GERTLER, Department of Electrical Engineering,
George Mason University, Fairfax, Virginia 22030, USA
JOHNSON et al.: Adaptive Systems in Control and Signal Processing (1990, No. 1)
ISIDORI: Nonlinear Control Systems Design (1990, No. 2)
AMOUROUX 8c EL J AI: Control of Distributed Parameter Systems (1990, No. 3)
CHRISTODOULAKIS: Dynamic Modelling and Control of National Economies (1990, No. 4)
HUSSON: Advanced Information Processing in Automatic Control (1990, No. 5)
NISHIMURA: Automatic Control in Aerospace (1990, No. 6)
RIJNSDORP et al: Dynamics and Control of Chemical Reactors, Distillation Columns and Batch Processes
(DYCORD '89) (1990, No. 7)
UHI AHN: Power Systems and Power Plant Control (1990, No. 8)
REINISCH 8c THOMA: Large Scale Systems: Theory and Applications (1990, No. 9)
KOPPEL: Automation in Mining, Mineral and Metal Processing (1990, No. 10)
BAOSHENG HU: Analysis, Design and Evaluation of Man-Machine Systems (1990, No. 11)
PERRIN: Control, Computers, Communications in Transportation (1990, No. 12)
PUENTE & NEMES: Information Control Problems in Manufacturing Technology (1990, No. 13)
NISHIKAWA & KAYA: Energy Systems, Management and Economics (1990, No. 14)
DE CARLI: Low Cost Automation: Components, Instruments, Techniques and Applications (1990, No. 15)
KOPACEK, MORITZ Sc GENSER: Skill Based Automated Production (1990, No. 16)
COBELLI & MARIAN I: Modelling and Control in Biomedical Systems (1989, No. 1)
MACLEOD 8c HEHER: Software for Computer Control (SOCOCO '88) (1989, No. 2)
RANTA: Analysis, Design and Evaluation of Man-Machine Systems (1989, No. 3)
MLADENOV: Distributed Intelligence Systems: Methods and Applications (1989, No. 4)
LINKENS 8c ATHERTON: Trends in Control and Measurement Education (1989, No. 5)
KUMMEL: Adaptive Control of Chemical Processes (1989, No. 6)
CHEN ZHEN-YU: Computer Aided Design in Control Systems (1989, No. 7)
CHEN HAN-FU: Identification and System Parameter Estimation (1989, No. 8)
CALVAER: Power Systems, Modelling and Control Applications (1989, No. 9)
REMBOLD: Robot Control (SYROCO '88) (1989, No. 10)
JELLALI: Systems Analysis Applied to Management of Water Resources (1989, No. 11)
Other IFAC Publications
AUTOMATICA
the journal of IFAC, the International Federation of Automatic Control
Editor-in-Chief: G. S. Axelby, 211 Coronet Drive, North Linthicum,
Maryland 21090, USA
IFAC WORKSHOP SERIES
Editor-in-Chief: Pieter Eykhoff, University of Technology, NL-5600 MB Eindhoven,
The Netherlands
Full list of IFAC Publications appears at the end of this volume
NOTICE TO READERS
If your library is not already a standing/continuation order customer or subscriber to this series, may we recommend that you place a standing/
continuation or subscription order to receive immediately upon publication all new volumes. Should you find that these volumes no longer serve your
needs your order can be cancelled at any time without notice.
Copies of all previously published volumes are available. A fully descriptive catalogue will be gladly sent on request.
ROBERT MAXWELL
Publisher
ADVANCED INFORMATION PROCESSING
IN AUTOMATIC CONTROL
Selected Papers from the IFACIIMACSIIFORS Symposium (AIPAC '89)
Nancy, France, 3—5 July 1989
Edited by
R. HUSSON
Centre de Recherche en Automatique de Nancy,
France
Published for the
INTERNATIONAL FEDERATION OF AUTOMATIC CONTROL
by
PERGAMON PRESS
Member of Maxwell Macmillan Pergamon Publishing Corporation
OXFORD · NEW YORK · BEIJING · FRANKFURT
SÄO PAULO · SYDNEY · TOKYO · TORONTO
U.K. Pergamon Press pic, Headington Hill Hall, Oxford OX3 OBW, England
U.S.A. Pergamon Press, Inc., Maxwell House, Fairview Park, Elmsford, New York 10523, U.S.A.
PEOPLE'S REPUBLIC Pergamon Press, Room 4037, Qianmen Hotel, Beijing, People's Republic of China
OF CHINA
FEDERAL REPUBLIC Pergamon Press GmbH, Hammerweg 6, D-6242 Kronberg, Federal Republic of Germany
OF GERMANY
BRAZIL Pergamon Editora Ltda, Rua Eça de Queiros, 346, CEP 04011, Paraiso, Sâo Paulo, Brazil
AUSTRALIA Pergamon Press Australia Pty Ltd., P.O. Box 544, Potts Point, N.S.W. 2011, Australia
JAPAN Pergamon Press, 5th Floor, Matsuoka Central Building, 1-7-1 Nishishinjuku, Shinjuku-ku,
Tokyo 160, Japan
CANADA Pergamon Press Canada Ltd., Suite No. 271, 253 College Street, Toronto, Ontario, Canada M5T 1R5
Copyright © 1990 IF AC
All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in
any form or by any means: electronic, electrostatic, magnetic tape, mechanical, photocopying, recording or other-
wise, without permission in writing from the copyright holders.
First edition 1990
Library of Congress Cataloguing in Publication data
Advanced information processing in automatic control (AIPAC'89):
selected papers from the IFAC/IMACS/IFORS Symposium, Nancy, France,
3-5 July 1989/edited by R. Husson.—1st ed.
p. cm.—(IFAC symposia series; 1990, no. 5)
Proceedings of the IFAC Symposium on Advanced Information
Processing in Automatic Control.
I. Automatic control—Congresses. I. Husson, R. (Raoul), 1901-
II. IFAC Symposium on Advanced Information Processing in
Automatic Control (1989: Nancy, France) III. International
Federation of Automatic Control. IV. International Association for
Mathematics and Computers in Simulation. V. International
Federation of Operation Research Societies. VI. Series.
TJ212.2.A39 1990 629.8—dc20 90-6923
British Library Cataloging in Publication Data
Advanced Information Processing in Automatic Control
Conference (1989: Nancy, France)
Advanced Information Processing in Automatic Control
(AIPAC '89)
1. Industries. Process control. Automatic control systems
I. Title II. Husson, R. III. International Federation of
Automatic Control IV. Series
670,427
ISBN 0-08-037034-9
These proceedings were reproduced by means of the photo-offset process using the manuscripts supplied by the
authors of the different papers. The manuscripts have been typed using different typewriters and typefaces. The
lay-out, figures and tables of some papers did not agree completely with the standard requirements: consequently
the reproduction does not display complete uniformity. To ensure rapid publication this discrepancy could not be
changed: nor could the English be checked completely. Therefore, the readers are asked to excuse any deficiencies
of this publication which may be due to the above mentioned reasons.
The Editor
Printed in Great Britain by BPCC Wheatons Ltd, Exeter
IFAC SYMPOSIUM ON ADVANCED INFORMATION
PROCESSING IN AUTOMATIC CONTROL
Sponsored by
International Federation of Automatic Control (IFAC)
Technical Committees Theory and Systems Engineering
Co-Sponsors
International Association for Mathematics and Computers Simulation
(IMACS)
International Federation of Operation Research Societies (IFORS)
Association française pour la Cybernétique Economique et Technique
(AFCET)
Centre National de la Recherche Scientifique (CNRS)
Institut National de Recherche en Informatique et en Automatique (INRIA)
International Programme Committee
M. G. Singh, UK (Chairman) N. Kheir, USA
P. Borne, France (Vice-chairman) R. E. King, Greece
M. Thoma, FRG (Vice-chairman) P. De Larminat, France
G. A. Almasy, Hungary L. Motus, USSR
M. Aubrun, France M. Najim, France
M. Basseville, France J. Nevins, USA
A. Costes, France A. Niemi, Finland
J. Debaigt, France J. O'Shea, Canada
B. Dubuisson, France H. Prade, France
P. Elzer, FRG J. Ragot, France
P. Eykoff, Netherlands J. Robert, France
A. J. Fossard, France M. G. Rodd, UK
C. Foulard, France J. Romagnoli, Argentina
T. Futagami, Japan D. D. Siljak, USA
L. T. Grujic, Yugoslavia M. Staroswiecki, France
R. Hanus, Belgium S. Tzafestas, Greece
R. Husson, France P. Uronen, Finland
G. Jourdain, France
National Organizing Committee
R. Husson (Chairman) B. Maudinas
J. Ragot (Vice-chairman) M. Musset
T. Cecchin M. Poinsignon
G. Davoust A. Rachid
B. Heit A. Richard
C. Iung M. Robert
M. Lamotte D. Sauter
A. Mailfert E. Yvroud
PREFACE
The papers in these proceedings are the results of the selection made by the
International Programme Committee for the Advanced Information Processing in
Automatic Control (AIPAC'89) Symposium held in Nancy, France, 3-5 July 1989.
The Symposium was sponsored by IFAC and its sister organizations IMACS and
IFORS and by three French scientific organizations (AFCET, CNRS, INRIA).
More than 200 participants coming from research centers, industrial companies
and universities gathered in Nancy to attend the symposium.
The main purpose of AIPAC'89 Symposium consisted in presenting the state of
the art in the field of fault diagnosis and system's reliability. But another
major theme was concerned with the use of intelligent approaches, such as
Expert system and other Knowledge Based approaches, to handle the vast
quantities of data that the large scale systems generate.
Both fault detection and diagnosis, associated to Artificial Intelligence are
used in large fields of application.
Most of the applications deal with manufacturing and continuous systems, but
other fields which have been gathered under five main titles are covered too.
The four tutorials provide an overview of the underlying concerns expressed in
the contributed papers to the Symposium.
R. HUSSON
Symposium Chairman
M. G. SINGH
Chairman of the IPC
vu
Copyright © IFAC Advanced Information SURVEY PAPERS
Processing in Automatic Control, Nancy, France
1989
AI AND COMPLEX SYSTEMS TECHNIQUES IN
MANUFACTURING MANAGEMENT AND
CONTROL
M. G. Singh and K. S. Hindi
Decision Technologies Group, Computation Department,
University of Manchester, Institute of Science and Technology (UM 1ST),
Manchester M60 1QD, UK
Abstract. AI and complex systems techniques are of value for many different aspects of manu-
facturing management and control. In this paper, we address two specific classes of manufacturing
environments, namely multi-line semi-continuous batch manufacture (which might typically be found
in the manufacture of detergents) and a class of flexible machining and assembly systems for dis-
crete parts manufacture (which might typically be found in the manufacture of aircraft systems).
In the context of production management and control in these environments, we highlight the need
for decision aids at various levels (e.g. for short-term scheduling, medium term planning, market-
ing/manufacturing interface). In each case, we discuss the key issues involved and the use of AI and
complex systems techniques for resolving these issues in relation to the design of specific decision aids.
The framework described is part of work carried out for a leading manufacturer of detergents and
within a major current project in collaboration with Jaguar Cars, Unilever and Logica. A number
of key conclusions arising from this work are also described.
Keywords. AI; complex systems theory; production planning an and control; flexible machining
and assembley systems; multi-line semi-continuous batch manufacture; decision aids.
Introduction
• The cost of data collection is relatively modest, since
individual items are aggregated into families. A ma-
Production management involves taking complex decisions
jor information system would be required, if detailed
based on choosing from among a large number of alter-
one-level planning is adopted instead, as this would
natives by trading off partially conflicting objectives in the
require demand, productivity and cost data and fore-
presence of complex technological and marketing constraints.
casts for, perhaps, thousands of items.
The complexity involved should make model-based systems
invaluable in supporting decision taking. However, the scale • The data used is more accurate, since aggregate fore-
of modern manufacturing systems is large indeed. More- casts will normally have less variance. Moreover, since
over, implementation of flexible manufacturing technologies a smaller number of forecasts is required, more effort
may impose various constraints on the availability of tools can be expended on improving the quality of the in-
and pallets, limit the size of work-in-progress buffers and re- put information as well as on using more sophisticated
strict material handling operations between work-stations. estimation models.
The decision problems are, thus, very large and very com-
plicated, making it necessary to resort to an hierarchical • Understanding of the results of the planning, as well
approach [1], whereby decision models with varying granu- as the planning process itself, is made easier.
larity can be considered at the different levels of the hierar-
chy. The relationships among the various levels are estab- Moreover, the resulting hierarchical production plan-
lished by aggregation / disaggregation, while the uncertain- ning and control system facilitates interface with strategic
ties present in the planning process can be reflected in the planning by clearly delineating the required inputs from the
time scales considered at each level. Thus at the aggregate higher level. This is particularly true in the case of the cru-
production level, multi-period models are considered. The cial marketing / manufacturing interface.
resulting plans are then disaggregated for a much smaller In addition to employing mathematical models for solv-
number of time periods at the intermediate level, while at ing some of the decision problems within the framework of
the lowest level, disaggregation is carried out for the next a hierarchical production planning system, there is scope,
time period only. The whole process can then repeated on and need for, employing KBS approaches. Broadly, these
a rolling horizon basis. seem to be most effective in one of three types of situations
[3]
• The first arises when the experience, expertise and gut
In addition, the overall planning system can be multi- feeling of expert managers play a paramount role due
pass by providing the possibility of carrying out several total to the uncertainties involved. Examples are problems
or partial planing iterations in any one planning exercise. at the marketing / manufacturing interface.
The advantages of the basic underlying methodology of
aggregation / disaggregation [2] can be summarised as fol- • The second type of situations arises when mathemat-
lows: ical modelling techniques and algorithmic approaches
1
2 M. G. Singh and Κ. S. Hindi
can not be envisaged, but where heuristics, based on a terial handling systems. Figure 1 shows the topology of a
combination of rules and domain knowledge, are likely typical plant with two manufacturing units, four storage si-
to provide efficacious solutions. Several examples of los, three material handling systems and four packing lines.
this type of situations are normally present:
1. If the aggregate schedule proves infeasible when
disaggregated to provide detailed production sched
ules, then there is, a nee,d for modifying the lead
times and resource parameters of the aggregate
model. In all probability, the task can be accom-
plished effectively by a knowledge based system
encoding the knowledge of expert dispatchers.
2. If a number of production facilities with markedly
different characteristics are present, then the prob-
lem of job allocation becomes significant. Here
again, the task can be accomplished by a knowl-
edge based system built around the local knowl-
edge of what jobs with which characteristics are
M « manufacturing unit
best suited to what facility. MHS - material handling system
3. a knowledge based system can be used for solving
coordination problems which may arise at the in- Figure \ . Typical layout of a two-stage plant.
terface between different levels of the production
hierarchy.
There is usually a small number of manufacturing units
4. assessing the detailed plans and schedules pro- (two to four), which can make from one to ten products, one
duced by the overall planning system may re- at a time. After production, the products are either stored
veal problems with machine utilisation, conges- temporarily before packing or directly fed to the packing
tion and contention for resources which necessi- lines through the material handling system.
tate tuning the lead times and resource param- Generally, for units dedicated to the manufacture of a
eters to carry out a further planning pass, ei- single product, cleaning is not necessary. But in the case
ther partially or overall by repeating the plan- of general purpose units, which can make more than one
ning process as a whole. This crucial and diffi- product, cleaning is important, mainly to guard against
cult coordination task may be best carried out undesirable interaction between products, such as colour
not automatically, but by experienced produc- interaction and cross contamination. Cleaning time is of
tion engineers aided by a knowledge based deci- the order of one hour. Thus long runs are very desirable to
sion support system. minimise the setup costs of the manufacturing units, and
also, perhaps more importantly, to keep the packing lines
• The third type of situations arises when a problem can packing with minimum change-over.
be mathematically cast into an optimisation model
and algorithmic solutions can be shown to work in
principle, but, due to problem dimension, are likely In a given plant, there could be one or the other of two
to be impracticable1 computationally. The problem of possible modes of processing : instantaneous and delayed.
detailed scheduling is a distinguished example. Here, In Instantaneous processing, raw materials react instanta-
the number of variables and the number of sets of neously to form the final product. There is, thus, an inflow
possible decisions is so large that optimisation based of raw materials and an outflow of final product without
scheduling systems are at a serious disadvantage com- delay. There is also an upper bound and a lower bound
putationally. Remedying this is a very active area of on the rate with which a manufacturing unit produces a
contemporary research. Nevertheless, it seems that given product and the amount of production is limited by
one of the most promising avenues to explore is to the maximum capacity of the unit.
employ knowledge based systems techniques for carry- In delayed processing, production is in discrete amounts
ing out, possibly constraint-directed or goal-directed, (batches), and it takes the manufacturing unit a certain
search of the solution space, while solving the local time to process a batch. There are a lower and an upper
problems which arise during the search by optimisation- limit on the amount of product manufactured by batch;
based models. the upper limit being determined by the capacity of the
unit, the lower limit by operational consideration, though
In the following, the above considerations are discussed sometimes processing is carried out at full capacity only.
more concretely in the context of two different manufactur- In both cases, when changing over from one product to
ing environments. another, thorough cleaning is needed to avoid cross contam-
ination. Also, in instantaneous processing, when a unit is
Production Planning and Scheduling for a Class of changed over to a new product, it takes a period of the or-
Two-stage Semi-Continuous Manufacturing Systems der of one hour, to reach the full production rate and start
up rates are sometimes significantly slow.
The manufacturing systems considered here are two- Intermediate storage consists of silos that are either ded-
stage : upstream, there are parallel manufacturing units icated to one product or flexible and capable of storing a
where base products are manufactured; downstream, there number of products, one at a time.
is a number of parallel packing lines where the base prod- The silos are fed from the manufacturing units and in
ucts are packed in different formats. Between these two turn feed the material handling systems through pipes or
processes, there is an intermediate storage system which conveyers. Not all silos are connected to all manufacturing
feeds the packing lines, generally through a number of ma- units or all material handling systems, and there may be
AI and Complex Systems Techniques 3
restrictions on the number of simultaneously active connec- ol resources over the whole aggregated period.
tions either side. There are also maximum flow constraints Aggregation of items is based on grouping items into
on the connections, but usually, these do not cause bottle- families such that the members of a family have identical
necks. or similar production and inventory costs and productiv-
There can be from five to ten packing lines that can pack ity factors, share common major setups and have small (or
one or more formats (sizes). Some are dedicated to pack- negligible) minor setups among them.
ing the different formats of one product, while others can Aggregation of resources allows the detailed variables
pack different products in different formats. A combination corresponding to the allocation of resources to different
size/product is called an item. Packing rates depend on tasks to be aggregated so that the reduced number of the ag-
both item and line. Although it will be assumed throughout gregated capacities can be taken into account at the medium-
that a line packs at its constant nominal rate once switched term scheduling level.
on for an item, start up rates are generally slower and a line The resulting aggregate medium-term scheduling prob-
may take significant time to reach full efficiency. lem is essentially a capacitated lot-size scheduling problem.
There are two types of packing line change-overs: minor Such problems are difficult to solve, but the current problem
and major. The former occur between items of the same has been modelled as a mixed integer programming problem
size, are sequence independent and do not usually require and solved successfully by exploiting certain of its structural
intensive labour. The latter occur between sizes, may be se- properties. Two solution approaches have been developed;
quence dependent, require intensive labour by skilled fitters one [5] is based on using a combination of Lagrangian relax-
and take longer. ation and perturbation search techniques while the other [4]
is based on hybrid Lagrangian / linear programming relax-
Packing lines are operated by operators whose number
ation within the framework of a branch and bound search.
depends on the line and the item it is packing. The num-
ber of operators in the packing room is usually limited, The solution of the medium-term scheduling problem
and this limits the number of simultaneously active pack- provides cost-effective short-term demands for each aggre-
ing lines. Further constraints are imposed by the incoming gated planning period separately. Thus, detailed schedul-
connections and the number of products that can be packed ing, taking into account detailed production conditions, can
then be carried out separately for each aggregated period.
at the same time in the packing room may be restricted by
the number of material handling systems.
The short term scheduling horizon is usually of one or Short-Term Scheduling
two weeks, divided into a number of smaller time periods.
By the end of the horizon, the item lots dictated by a higher Due to the complexity involved, a monolithic mathemat-
ical formulation would not be fruitful. Indeed, the resource
level planning system have to be produced. The objective is
constraints, the intermediate storage and the operational
to minimise production costs subject to various structural,
constraints, particularly the minimum run length require-
operational and resource constraints.
ments, make the problem over-constrained. Thus the use
of heuristic algorithms is essential. However, for a two-
Aggregate Medium-Term Planning
stage, multi-product manufacturing system with interme-
diate storage, like that considered here, a one pass heuristic
The above description of the production system serves
is impracticable, since a feasible schedule for one stage may
to throw into relief the complexity of the interlocked de-
impose on the other stage demands that are impossible to
cision problems involved. A realistic way of handling this
meet. This is due to the fact that there are, generally, sev-
complexity is by structuring a production scheduling sys-
eral products sharing the same manufacturing facility as
tem in an hierarchical fashion [5], as shown in Figure 2.
well as to the difference in the operating speed between the
packing lines and the manufacturing units. Hence the need
Detailed plant database is for a multi-pass strategy, whereby the problem is solved
iteratively until a sufficiently good solution is obtained[6, 7].
Aggregation of the
production scheduling
model The problem of scheduling the packing lines is solved
Aggregate model Medium-term forecasts, first. The lots demanded by the medium-term planning sys-
i inventory status tem are scheduled for packing during the given horizon, so
that a cost function consisting of a combination of change-
Adaptation | AAggggrreeggaattee mmiidd--tteerrmm
sscchheedduulliinngg over and packing costs is minimised. The schedule obtained
1
imposes a set of demands, period by period, on the manu-
Lot sizes
facturing units. These are then scheduled to satisfy these
DDeettaaiilleedd ((ddiissaaggggrreeggaattee)) demands, minimising a cost function consisting of setup and
sscchheedduulliinngg production costs.
Both subproblems are solved using enumerative pro-
Detailed schedules
cedures. If there are no feasible solutions at the manu-
Fig. 2. Hierarchical structure of the production scheduling system.
facturing units level, the whole problem is reconsidered,
by employing a knowledge-based coordination device which
At the upper level, an aggregate medium-term schedul- chooses the variables to alter and how to alter them. The
ing model is created and solved. This model can be ob- process is repeated until a satisfactory schedule is obtained.
tained from the detailed scheduling problem by aggregation
of periods, items and resource constraints. Aggregation of Coordination
a number of periods into one aggregated period results in
creating aggregate variables, such as total production of an At the heart of the multi-pass heuristic for short-term
item in the aggregated period, or inventory level of an item scheduling is the knowleclge-Dasecl system entrusted with
at the end of the aggregated period. Also the groups of coordinating the schedule of the packing lines with that of
constraints corresponding to a number of periods are sub- the manufacturing units. The elements of the coordina-
stituted by aggregate constraints accounting for availability tion strategy are based on two possible ways of alleviating
4 M. G. Singh and Κ. S. Hindi
bottlenecks: reconsidering the initial inventory for one or Each component corresponds to a job consisting of a
more products or altering the demand at the manufactur- number of operations performed on various machines. Each
ing units level by altering the quantity packed at a certain operation contributes to the manufacturing lead time by
time period. adding setup and processing times, transfer time from the
The first possibility, i.e. adding directly a positive incre-
upstream machine and waiting time (if the corresponding
ment to the initial inventory and solving the manufactur-
machine is busy). The most varying times are the waiting
ing units subproblem again, is viable. This is so because
times which depend on such factors as the actual workload
the overall problem is solved on a rolling horizon basis,
imposed by other jobs on the work-centres, priority disci-
which means that the previous schedule of the manufac-
plines, etc. These factors are difficult or even impossible
turing units is known and hence the maximum quantity of
to consider in detail at the aggregate scheduling level; they
initial inventory that can be called upon.
may be reflected only in an aggregate way, by considering
As for the second possibility, the packing lines pack at not actual but average aggregate waiting times. It is as-
fixed nominal rates, which implies that during any time sumed that aggregate waiting times are variable depending
period, a line will either pack a fixed quantity or be idle. upon the average machine utilisation during a given pe-
Therefore, one way of altering the total quantity of a given riod. It is acknowledged that average waiting times depend
product packed at a given period, in order to decrease the on the level of utilisation. If the machines are loaded to
demands on the manufacturing units, is to reduce the num- full capacity or are overloaded during certain periods, the
ber of items, belonging to that product, that can be packed corresponding waiting times become long. Conversely, if
during the period concerned; thus, forcing one or more a machine has a workload which utilises, say, 75 percent
packing lines to be idle or pack another product. of its capacity, the resulting average queuing time is rela-
Moreover, if the number of products to be manufactured tively small and more predictable. Therefore, it is to be
at a given time period exceeds the number of manufacturing expected that average lead times can be reasonably esti-
mated only if appropriate capacity-constrained production
units, then the maximum number of products that can be
scheduling problems are solved. ' On the other hand, pre-
packed at this time period has to be reduced.
diction of lead times is a prerequisite for formulating suffi-
Infeasibility can also be alleviated either by extending
ciently accurate aggregate production scheduling problems.
the planning horizon at the manufacturing units or at the
Thus the decision problems at the different levels of the de-
packing lines level, which in effect means introducing some
cision hierarchy should be embedded in a multi-pass, hybrid
overtime.
decision support framework, where various large-scale sys-
Finally it is worth noting, that the lot sizes required by
tem techniques, discrete optimisation algorithms and intel-
the medium-term planning system may render the overall
ligent inference rules are combined in a hybrid, hierarchical
detailed scheduling problem infeasible, in which case it is
knowledge-based system, like that shown in Figure 4.
necessary to resort to one of two measures: either altering
some of the lot sizes, or modifying the parameters of the
medium-term scheduling problem and resubmitting it to
the medium-term scheduling system. The choice of which aggregate. capacitated
multistage scheduling
measure to adopt in any given case and implementing the
chosen measure is entrusted to a knowledge-based system. lots and latest finish
dates of families of items
The coordination functions are currently carried out by Γ
the user of the system in an interactive fashion, pending the
Ι disaggregation of the families I
completion of a knowledge-based system that would auto- / \
mate these tasks. lots and the latest finish dates
ell 1 of the. root jobs
Production Planning and Scheduling for a Class of
aggregate multi-period aggregate multi- period
Flexible Machining and Assembly Systems job-shop scheduling job-shop scheduling
1
The production systems, under consideration here, are aimed detailed s±eq uencing detailed sequencing
in each period in each period
at producing a set of end products which have a flat com-
ponent structure, as illustrated in Figure 3. End items can
tuning the lead times
be grouped into families of items having similar component machine utilization and resource
structures, similar productivity factors and inventory costs, parameters
and sharing common major setups. The differences among
Figure Hierarchical structure of the production planning and scheduling.
the members of a family, as measured by the number of dif-
ferent components, non-common setups, etc., are relatively
Aggregate Capacitated Lot-Size Planning
small.
The production of an item can be represented as a set of
parallel bunches of jobs. Each bunch of jobs corresponds to
a set of components at different levels of the hierarchy which
form a final component at the upper level. Final compo-
nents used for the final assembly are called root components.
A bunch of jobs corresponds to a pattern of operations per-
formed for different components on different machines in
order to complete the root component. If the completion
time of one root component is postponed by an hour, the
completion time of the end item is also postponed. On the
Figure^. An example of the flat assembly structure of an end item. other hand, an hour saved on a subset of the root compo-
AI and Complex Systems Techniques 5
nents does not necessarily reduce the completion time of (i) the detailed schedule is feasible and has acceptable
the end item. Thus completion times of the root compo- performance measures such as flow times of jobs, or
nents should be synchronised in order to permit efficient workloads of bottleneck machines,
and timely final assembly of the end item. This is a rela-
tively easy task in view of the flat assembly structure. The (ii) the detailed schedule is unacceptable due to violation
difficult and important problem, however, is to schedule the of available resources in some periods and/or a per-
production of the large number of end items competing for formance measure having a poor value.
resources making full use of the existence of sets of similar
In the second case, cell scheduling can be improved by
items.
two-level job shop scheduling at cell level, see Figure 4.
This is in marked contrast with the crucial concern in
multi-stage systems having a deep assembly structure; see
Coordination
[8]. In such systems, the manufacture of an item requires a
certain number of components and the item itself is in turn
Since it is proposed that the decision problems at the
a component of a single parent item and so on until the end
upper and cell levels be embedded in a multi-pass, hierarchi-
item is produced. Since the number of stages is relatively
cal decision support framework, the need for coordination
large, the major problem is to determine the replenishment
strategies arises. This coordination can be effected through
quantities for all component items to satisfy the external
a feedback process carried out by planning engineers, who
demand for the end item during each time period.
by studying the detailed job-shop schedules and machine
Due to the large dimension of the problems involved, the
utilisation data can perform the crucial coordination task
objective of planning the production of the large number of
of tuning lead times and resource parameters; see Figure 4.
end items competing for resources in production systems
However, it is proposed that future work be directed at
with flat assembly product structure, while making full use
developing various coordination strategies within a knowl-
of the existence of sets of similar items, can be achieved only
edge based system framework, utilising both the knowledge
by adopting an aggregation / disaggregation approach.
of expert planning engineers and some automatic analysis
At the planning level, it is unrealistic to consider de-
of the detailed schedules and utilisation data. In particular,
tailed capacity constraints, such as total availability of an
the following can be taken into account:
individual machine during a period. THese constraints can
be aggregated to represent functionally similar resources
• Scheduling at cell level may be performed in the course
such as groups of lathes, grinders, hones, groups of tools,
of solving the aggregate upper level problem. It is not
etc. Furthermore, items can be aggregated into families,
necessary to hold the disaggregation until the aggre-
each consisting of similar items. These two measures com-
gate schedule satisfies the aggregate constraints.
bined facilitate aggregating production and inventory vari-
ables leading to a significant reduction in problem size, • In the aggregate models it is necessary to predict the
which can then be formulated as a mixed-integer program- lead times and the parameters of the resource con-
ming Problem. However, although reduced in size, the straints. These parameters1 may be inaccurately pre-
model is still hard to solve by using existing MILP codes dicted, hence the schedules can be infeasible. How-
which utilise linear programming relaxations, but can be
ever, the solutions may provide important informa-
solved efficiently by specialised procedures [1].
tion which can be fed to the upper levels and used for
The success of the overall planning system depends cru- tuning lead times and resource utilisation parameters
cially on the effectiveness of the aggregation / disaggrega- of the aggregated models.
tion procedures adopted. The problems can be cogently
modelled and efficiently solved within a mathematical pro- Conclusions
gramming framework [2].
The work described in this paper serves to indicate the
following conclusions:
Scheduling at Cell Level
• Hierarchical production planning can be an effective
Aggregate scheduling provides production lot sizes and
framework for the management and control of ad-
the latest finish dates of the end items. Requirements for
vanced, complex manufacturing systems.
end items must be further translated into requirements for
detailed scheduling of jobs which correspond to the root • Within such a framework, decision models with vary-
components of the end items. Short-term detailed schedul- ing granularity can be considered at the different lev-
ing can be carried out for each cell separately, with the els of the hierarchy. The relationships among the var-
aim of processing all jobs so that work-in-progress is re- ious levels are established by aggregation / disaggre-
duced and available resources are utilised efficiently. The gation, while the uncertainties present in the planning
performance criterion can be based upon flow or upon com- process can be reflected in the time scales considered
pletion times of jobs. The detailed constraints, such as at each level. The planing can be carried out on a
limited buffers between machines, material handling limita- rolling horizon basis.
tions, no-wait requirements, can also be taken into account
at this detailed level. Depending on the structure and the • Within the proposed framework, there is ample scope
size of the resulting detailed sequencing subproblems, either for creatively combining various complex systems tech-
heuristic priority-based dispatching rules, artificial intelli- niques, drawn from, for example, operations research
gence or optimisation-based heuristics or, for smaller prob- and control theory, with AI techniques. The latter
lems, optimal branch-and-bound scheduling algorithms can are likely to prove particularly useful for carrying out
be used. coordination tasks.
At the upper level, the resource and precedence con-
straints can only be reflected in an aggregate manner. Thus • Due to the complexity of the problems involved, hu-
in the course of translating the aggregate schedule into a man planners and schedulers tend to resort to over-
detailed schedule, two cases may occur: simplification and reduction. It is, therefore, believed