Imperial College London
ISE-2 Surprise 97 Project
Stochastic Modelling of Manufacturing Systems -
Role of Performance Modelling in Automated Manufacturing Systems
Written by Nick Chapman
Ever since the first production line was implemented by Henry Ford in 1913 to produce an
engine in 84 stages, their use as a method of manufacture has become the dominant approach
for mass production. The emergence of high performance automated manufacturing systems
(AMSs), has lead to the need for methods of modelling these types of system in order to
maximise throughput, flexibility and competitiveness. This article intends to provide
an insight into the area of performance modelling and the methods used in achieving their
optimisation.
An AMS is a complex network of processing, inspecting and buffering nodes connected by
system of transportation mechanisms. For an AMS to be considered viable as capital outlay
it must be flexible in the product that it can produce, so it should be able to last for
several different product life cycles. It must also have a level of fault tolerance to
avoid costly breaks in the production line in the case of sub system failure. It is also
desirable that the line is capable of increasing or decreasing output with the rise and
fall of demand.
All of these specifications show the complexity of decision making in the field of AMSs and
the need for concise and accurate modelling methods.
AMSs belong to the domain of discrete event dynamic systems (DEDS) in which the evolution
of the system depends on the complex interactions of various discrete events such as the
arrival of raw materials, departure of finished goods, failure of equipment etc. The state
of DEDS changes only at these discrete points in time instead of continuously. Over the
last decade several models have come about to describe DEDS and these can be grouped into
two distinct areas.
- Qualitative models are concerned with the logical aspects of system evolution
such as controllability, stability and the existence of deadlocks in system operation,
etc. This category also includes Petri Nets, extended state machines and finitely recursive
processes.
- Quantitative models are concerned with the quantitative system performance in terms
of throughput and lead time. This category also includes discrete event simulation,
min-max algebra, Markov Chains, stochastic Petri nets, queues, and queuing networks.
Quantitative models are a general term including performance modelling which is the area
of interest to this article.
Within the life cycle of an AMS various decisions are made concerning implementation,
design and operation of the system. The role of performance modelling is to assist in
these decisions in an affective way as possible. Typical decisions at the planing stage
include number and type of machines, number of material handling devices, number of
buffers, size of pallet pool and number of fixtures, best possible layout, tool storage
capacity, evaluate candidate AMS configurations, part type selection, machine grouping,
batching and balancing decisions, and scheduling policies.
During the operational phase of an AMS performance modelling can be used to assist
decisions about how to cope with in the event of a breakdown, removal or addition of
resources and parts, optimal scheduling in the event of machine failure or sudden changes
in the product or its demand and in the avoidance of unstable situations such as
deadlocks.
Performance modelling is also used in the design stage of the system. It is used
in decisions such as whether to use central versus local storage, push production
versus pull production, shared versus distributed resources, the effect of flexibility,
etc. Performance predictions obtained using faithful models can be used to convince
customers or investors and also give the designer another perspective on the design
enabling better designs.
The performance of an AMS can be measured by a set of generic measures. These are :
manufacturing lead time, work in progress, throughput, machine utilisation, capacity,
flexibility, performance, and quality. Using performance measuring these values can
be evaluated and used to compare AMS performances. The measures are quite dependant on
each other and each is important in its own right. Some typical questions that are asked
and can be answered using performance modelling are:
- What is the probability that a particular product can be delivered before the deadline?
- What is the minimum number of working machines required so that the average throughput of finished parts just
exceeds the targeted production?
- How many fixtures/pallets are to be used in order to increase the average
machine utilisation beyond 80%?
- Does the given AMS configuration have enough capacity to deliver the required amounts
of products in the set deadlines?
- Which of the candidate layouts offers the best flexibility to part-mix changes?
- What is the minimum number of resources in the system (machines, transporters, buffers)
that would ensure that the probability of producing at least 100 parts in a shift of 8
hours, in the presence of unscheduled downtime, exceeds 95%?
- What is the effect of machine blocking on throughput and manufacturing lead time?
Should we add one more buffer?
- What are the potential bottleneck resources and congestion points in the system?
- Are there deadlocks in the system? What is the mean time to deadlock?
- How do throughput and lead time change when we have one less machine? One more
machine? One more transporter? Some more buffers?
- Is there a spare capacity in the system to undertake some low-priority jobs?
Performance evaluation falls into two broad categories; performance measurement and
performance modelling. Performance measurement is primarily for existing systems and
is most commonly used to monitor key systems to enable re-configuration. Performance
modelling can be split itself in to two sub types; simulation modelling and analytical
modelling. Simulation is the original method of modelling and is still commonly used,
there are many computer based programs design solely for manufacturing simulations.
Analytical modelling is becoming more widely used and is a strong alternative to
simulation, there are many accepted models for manufacturing purposes such as stochastic
Petri nets, Markov chains and queuing theory. Both systems are commonly used and each
has its advantages and disadvantages which are two deep to be explored within this article.
Performance modelling has become a very important part of automated manufacturing system
design and is equally important for maintaining the system at its peak of ability.
The manufacturing methods in use by companies has change dramatically in recent years
with the use of advance robotics and computer control to optimise production, this in
tern has lead to reduced prices and higher quality of product. The production lines can
only get better with more modelling and investment and this is best achieved with the use
of performance modelling.
Bibliography
Stochastic Models of Manufacturing Systems,
John A Buzacott,
George Shanthikumar
Usefulness: 6
Readability: 5
Comments: Designed for mainly for post-graduates and specialist level. A strong knowledge of Statistics is require to appreciate fully the book.
Performance Modelling of Automated Manufacturing Systems,
N.Viswanadham,
Y.Narahari
Usefulness: 6
Readability: 5
Comments: A good introductory book with a reasonable level of previous knowledge required.
Last Updated
20th May, 1997
