1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
"1961 and all that: The Influence of Jay Forrester and John Burbidge on
the Design of Modern Manufacturing Systems"
Denis R Towill
Cardiff Industrial System Dynamics Group
School of Electrical, Electronic and Systems Engineering
University of Wales College of Cardiff
PO Box 917, Cardiff, CF2 1XH, Wales, UK
Abstract
Modem fi ing systems are expected to respond rapidly, effectively and efficiently to changes
in the marketplace. Simultaneously there is the drive to achieve world class customer service levels
coupled with minimum reasonable inventory (MRI). We thus have the classic conflict of interests
between marketing, production and materials ing wants the plete product
range available off-the-shelf; production is still all-too-often looking to fe in i
batch quantities so as to achieve ies of scale; and i is trying to
storage and distribution costs which in turn requires that a total systems MRI policy be adopted.
As reviewed in the paper the ground rules for effective manufacturing system design were
co-incidentally established in 1961. Jay Forrester showed that medium period demand amplification
was a system dynamics phenomenon which could be tackled by reducing and eliminating delays and
the proper design of feedback loops. In parallel, via his "rules to avoid bankruptcy" and “laws of
manufacturing systems", John Burbidge showed that short period demand amplification was due to
multi-phased, multi-period ordering policies.
Some thirty years later the work of both pioneers (which is conveniently summarised in the frequency
domain) is still ignored at their peril by manufacturing industry. As an aid to fruitful exploitation of
these ideas we describe their application to the analysis and design of a real world automotive spares
supply chain and to a multi-product "to-make" ordering model which drives an MRP system to ensure
that customer service level targets are met. Input-Output block diagrams were found to be
particularly useful diagnostic tools for these projects.
Production and Operations Management, page 105
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
1961 and all That: The Influence of Jay Forrester and John Burbidge on
the Design of Modern Manufacturing Systems
1. INTRODUCTION
ane has been an interesting reaction to a recent paper by Towill [1992] which suggests that
any probl still d in facturing systems op can be at least partially
solved by implementing ideas which were well documented some thirty years ago. This feedback
has been triggered not only by the drawing together of the dol of
Industrial Dynamics (Forrester, 1961) and Material Flow Control (Burbidge, 1961) but by the
realisation that they were indeed cognate in two very important respects. Firstly, they cover a
wide frequency spectrum of activities; secondly they cover strategic, tactical, and operational
aspects of manufacturing system design.
A problem typically encountered in the automotive industry is that JIT shop floor controls seem
to be working well, yet the orders placed on the shop-floor via the MRP II system appear to
fluctuate excessively compared to perceived demand in the marketplace. Under such
circumstances the Production Director may argue with some justification, "Volatility of this
magnitude costs me a great deal of ‘money due to unnecessary changes in capacity and work-in-
process. I believe this situation is partially due to poor systems design. Can you please
substantiate this view, and if true put a ballpark estimate on the effect?" Subsequent investigation
requires the formation of a multi-disciplinary Task Force [Parnaby 1993] which may also be
subsequently charged with system re-design and implementation once the case for change is
proven.
2. "THE FORRESTER EFFECT" AND "THE BURBIDGE EFFECT"
It is customarily the case when the information available to the Task Force is analysed that the
problem is seen to be twofold as shown in time series form in Fig.1. Thus there is usually both an
unnecessary amplification affecting the medium-term behaviour (the Forrester Effect) and
ry short-term fl ions (the Burbidge Effect). No identification technique
used to date has reliably separated the two effects: nor is this necessary since the Task Force will
be charged with tracking down all significant causes using the manifold tools at their disposal. It
is enough at the analysis stage that the Task Force is fully aware of both phenomena and is
continuously on the look-out for them.
What Fig.1 does do is to prioritise the most fruitful areas for immediate follow-up. This depends
on the local situation and in particular who is the "product champion" responsible for both
funding and driving the re-engineering process. Thus in terms of Fig.2 we may well be
constrained to act within the individual business but for preference need to urgently create
circumstances in which we can re-engineer the business interfaces with customer and/or supplier.
In simplest terms the Forrester route may be regarded as dynamic simulation driven strategic re-
design, whereas the Burbidge approach is to avoid many potential operating pitfalls by good
tactical application of sound industrial engineering principles in order to assure better local
control of material flow.
3. MATERIAL FLOW CONTROL
Although the ideas of material flow control date back to at least 1961, two later publications put
these ideas even more succintly and it is these which we shall quote in detail. Firstly, there are the
five rules for avoiding bankruptcy (Burbidge, 1983), which state:
RULE 1:Only make products which you can quickly despatch and invoice to customers.
RULE 2:Only make in one period those
Pp you need for in the next period.
RULE 3: Minimise the material throughput time.
RULE 4:Use the shortest planning period, i.e. the smallest run quantity which can be managed
efficiently.
RULE 5:Only take deliveries from suppliers in small batches when needed for processing or
assembly.
Production and Operations Management, page 106
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
These were subsequently enriched and supplemented by the six "Laws of Manufacturing Systems"
(Burbidge 1984) as follows:
*The Law of Gestalt. *The Law of Material Flow.
*The Law of Prescience. *The Law of Industrial Dynamics.
*The Ordering Cycle Law. *The Law of Connectance.
se ¢ 4¢~goosooccc
Be nut-phase ort
mul orderint
oes (he Burbidge alec}
IMPLIED
CAPACITY
SWING
’ (a) Effect of poor Jotvaton flow plus
multi-phase orderin:
FACTORY
ORDERS:
+
;
(b) Demand amplification reduced via
synchronised ordering
i Eo
Acceptable demand 592
fluctuation Les
256
FACTORY
ORDERS.
(a) Demand amplification further reduced
via good information flow
Figure 1. Typical avoidable demand amplification
experienced by a manufacturing process
These are briefly defined in TABLE |. Taken together, these Rules and Laws integrate into a
prescriptive eee engineering methodology many of the slogans for improved international
by These include Ingersoll Engineers,
(Small, 1983, At ha Integrate, and Automate"), and the Boston Consultancy Group (Stalk
and Hout, 1990, "How Time Based Competition is Re-shaping Global Markets").
Production and Operations Management, page 107
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Delays
Inventories
Decision
les
Uncertainties
Distributor
Rules
Uncertainties
individual
Business
Issues
Delays
Inventories
Rules
Uncertainties
Components
Supplier
Flow of Goods
Flow of Orders C——>
Flow of Money L$—>
Supply Chain Issues
Figure 2. Individual business versus supply chain issues
The reduction or better still, elimination of time delays is well known to system dynamicists as a
preferred route to achieving better dynamic behaviour. This is clearly a dual area of concern with
material flow control as Burbidge readily acknowledges. So we see many good reasons why in the
internal scenario part of Fig.2 management consultants tackle process flow time reduction as the
top priority in a new assignment. it not only makes good sense both from the Forrester and
Burbidge viewpoints, but much perfc can be deli d inside the business
boundaries under the direct control of the “problem owner". The Consultant can thus make
significant immediate savings and leave until a little later the difficult problem of initiating the
cultural change usually needed to improve information flows across boundaries [Towill, 1992].
Production and Operations Management, page 108
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
TABLE I: THE SIX BURBIDGE LAWS OF MANUFACTURING SYSTEMS.
1, THE LAW OF GESTALT.
"The whole is not the sum of its parts" and by extension "a set of sub-optimum
solutions can never produce a true optimum solution". An holistic approach requires
that all system design stages should be multi-disciplinary and start with corporate
planning followed by a total material flow system covering purchasing, manufacture,
and distribution.
2. THE. LAW OF MATERIAL FLOW
"The ic ofa ‘ing system is it ly proportional to the complexity of|
its material flow system". So we should design simple material flow systems based on
product organisation and then restrict freedom of production planning choice to
maintain this simple system.
3. THE LAW OF PRESCIENCE.
"It is not given to human beings to foretell the future". Most manufacturing systems
ignore this Law by basing production on long term forecasts of future sales. Because of
frequent and unavoidable changes in the programme there will be high stock-holding
and obsolescent costs. So we should regulate material flow via a series of short term
programmes which are then much more likely to cover actual needs.
4. THE LAW OF INDUSTRIAL DYNAMICS.
"If demand for goods is transmitted along a series of inventories using stock control
(i.e. level triggeredjordering, then the amplitude of the demand variation will increase
with each transfer". There is also a Murphy's Corollary; "If the system design is such
that there is any possible way that demand can amplify then this effect will inevitably be
triggered during normal operations".
5. THE ORDERING CYCLE LAW.
"If the various components made in a factory are ordered and made to different time
cycles, they will g high and unpr ble variations in both stocks
and load as the many contributing component stock cycles drift in and out of phase".
Too many factories still use multi-cycle ordering systems based on EBQ theory.
6. THE LAW OF CONNECTANCE.
"A given direction of change in the value of any manufacturing system variable ve
induce, or be induced by a given direction of change i in at least one other variable".
first attempt to codify practical causal ps has p d “
hypotheses concerning these changes.
The reduction or better still, elimination of time delays is well known to system dynamicists as a
preferred route to achieving better dynamic behaviour. This is clearly a dual area of concern with
material flow control as Burbidge readily acknowledges. So we see many good reasons why in the
internal scenario part of Fig.2 management consultants tackle process flow time reduction as the
top priority in a new assignment. It not only makes good sense both from the Forrester and
Burbidge viewpoints, but much performance improvement can be delivered inside the business
boundaries under the direct control of the "problem owner". The Consultant can thus make
significant immediate savings and leave until a little later the difficult problem of initiating the
cultural change usually needed to improve information flows across boundaries [Towill, 1992].
4. THE PIVOTAL ROLE OF FEEDBACK
As was seen from Fig.2, no business, and no manufacturing system is an island but is in fact part
of a complex production-distribution system known as a Supply Chain. Thus the explanation for
demand volatility sought by the Production Director in Section 1 may be found to be caused by
internal forces; external forces; or by both acting together or in opposition in whatever way is
required to produce the unwanted demand amplification.
If we refer to Fig.1 then there is a good chance that any observed high-frequency amplification
can be traced to a "Burbidge" origin, typically due to multi-cycle ordering. This is often
ised both in its and in its effect until the evidence is produced in the form of
Production and Operations Management, page 109
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
vast swings in inventory way beyond that necessary to maintain realistic safety stocks. Indeed it
is still our experience that stock control dynamics is very poorly understood and even more badly
applied. At least by adopting the "reduce process flow times as top priority" strategy the
consultant knows there is a chance that waste will be reduced in proportion to the improvement
achieved [Stalk and Hout, 1990].
What is still lost by this parochial approach is the further benefit to be obtained by proper
dynamic design. As we would expect, this is where Forrester [1961] comes to the fore with his
statement on the pivotal role of feedback within system modelling and simulation as typified by;
"Systems of information feedback contro] are fundamental to all life and human endeavour; a
feedback control system exists whenever the environment causes a decision which in turn affects
the original environment. In business, orders and inventory levels lead to manufacturing
decisions which satisfy orders and correct inventories. Feedback theory explains how decisions,
delays and predictions can produce either good control or dramatically unstable operation. It
relates sales promotion to production swings, purchasing and pricing policies to inventory
fluctuations and typical life cycles to the need for research. Typical manufacturing and
distribution practices can generate the types of business disturbances which are often blamed on
conditions outside the company. Thus random, meaningless sales fluctuations can be converted
into annual, seasonal production cycles by the action of feedback. Advertising and price discount
policies of an industry can similarly create two and three year "rogue" sales cycles".
So we would expect to find many potential traps for the unwary in manufacturing system design,
particularly in the supply chain scenario where for reasons which will become apparent later the
Law of Industrial Dynamics seeks centre stage with the Laws of Gestalt and Material Flow.
5. THE SUPPLY CHAIN CHAIN SCENARIO.
Each stage within a supply chain is known as an echelon, and embraces the following
characteristics;
*
*
perceived demand for products which may be firm orders or just forecasts.
a "production" or "added value" process (at least as perceived by the particular echelon
concerned).
performance data (which may be stale and/or sistortes)
disturbances due to breakdowns, faulty parts
lags for i ion, material, aad cash flows.
points in the chain where information is brought together for management decision making.
formal or informal algorithms for issuing instructions both internally and externally.
Crucially, there is iated with perceived demand, with the quality of information
available, and with the value Sof transmission lags. It is therefore not surprising to find that many
supply chains exhibit the dynamic behavior already met in Fig.] with medium period
amplification present. This phenomenon is almost certainly due to "The Forrester Effect" and
Fig. 3 shows an example of a causal loop diagram which explains how this can happen due to the
interaction between the decision makers in | in an supply chain
[Towill, 1988].
Also shown is an Input-Output Diagram which summarises the sources of amplification noted by
the Task Force conducting the study. Note that at the analysis stage the synthesis techniques to
be used later are not pre-selected but depend on what is di: d during the i igation [Weill
and Veit, 1989].
eR RE
An obvious possibility for improving supply chain performance in this scenario is to combine the
manufacturing and distribution functions. This procedure, which had already been proposed in
general terms by Jay Forrester was implemented to good effect in this particular supply chain.
However, an even more spectacular example is due to Toyota [Towill 1992]. That company
found that the distribution process was taking longer than needed for car manufacture. Not only
does this build up excessive total system stocks, but the Distributor masks the true market
demand. So Toyota promptly bought out the Distributor and solved two problems for the price
of one.
Production and Operations Management, page 110
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Retailer Distril 's 's
Orders Buyer Perception of
Buying Policy
Hetier Distributor
Deliveries Orders
Distributor \
Distributor Perception of Factor
Stocks Factory Performance Scheduler
Distributor Factory Factory
Deliveries Backlog Orders
Facioy ———— Production
(a) Caual loop iption of ion between il and
situated within an automotive products supply chain
Multiple-Cycle Ordering
—eee
Multi-Phased Ordering
reef]
EBQ
——————
Uncertainties Manufacturing § Avoidable Demand
System |
Se Amplification
Excess Delays
|
Feedforward Logic
—_—_
Feedback Logic
—— ++
Burbridge
Effect
Forrester
Effect
(b) Input-output diagram of influences on avoidable demand amplification
Figure 3. Sources of demand amplification in an
automotive products supply chain
6. SUPPLY STRATEGY AS A SYSTEM DYNAMICS PROBLEM
Burbidge in his 1961 paper also demonstrated graphically that supply strategy may be succintly
described via system dynamics. As shown in Fig.4, by using a frequency domain representation of
demand it is possible to schematically demonstrate the industrial meaning of such strategies as
Deliver from Stock i.e. "make what think we can sell", to the other extreme of Keep the
Production and Operations Management, page 111
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Factory Busy at all Costs i.e. "Sell what we make". The first policy is strongly evident, for
example, in the components industries, whilst the second is common in capital intensive
industries such as rolling of steel products.
STRATEGY PROCESS INVENTORY DEMAND:
FIXED PROGESS \ L\ [\
RATE
OR
FIXED INVENTORY ——————_
LEVEL
oR
coMPROMIsE | /— \ S| \_S
DESIGN
(a) Trade-off Between Process Flow Rates
and InventoryLevels
B: —
Od
Zz
<
6
zZ
<
r
s)
ca
oa
a
5
no
(b) Equivalent Parts of Supply Chain
Frequency Response
Figure 4. Relating process flow rates and
inventory levels to the filter characteristics of a
supply chain (Towill and Del Vecchio, 1994)
An advantage of Fig.4 is that it enables both individual manufacturing systems, and quite
complicated supply chains to be studied using filter theory, since this allows us to design individual
echelons which are well damped, or if this is not possible, to avoid unnecessary amplification by
ranging the resonant frequencies over a limited part of the spectrum [Towill and del Vecchio,
1994]. Whilst the approach is limited to linear and quasi-linear systems, there is much evidence
Production and Operations Management, page 112
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
to suggest that this is what the system user actually prefers, i.e. simple robust controls [Monniot
et al, 1987]. Unavoidable non-linearities are then allowed for on an individual basis. For
example if capacity limitation is a problem then the best "linear" design under this circumstance
is readily established using the root-sq locus technique, or better still, the Bode plot equivalent
[Towill, 1981].
Where the dynamics approach has a uniquely important contribution to make is in the optimum
design of systems which meet agreed customer service levels. An example is the design of an
adaptive ordering system which will achieve, say, 97.5% "off-the-shelf" deliveries to the
customer across a 400 product range with minimum reasonable inventory (MRI) to achieve this
goal [Cheema, 1994]. The way in which this links in with the existing industrial MRP system is
shown in Fig.5. The purpose of this dynamic "to-make" ordering model is to drive the MRP
system so as to achieve target customer service levels despite inevitable variability in
manufacturing lead times. Hence we incorporate an adaptive feedback loop based on current shop
floor performance of individual orders thus updating the pipeline control scheme advocated by
Lyneis [1980].
It should be noted that the concept of MRI as a target for system design is gaining ground
[Grunwald and Fortuin, 1992]. This, by definition is a dynamic concept which can only be
at ily by Tt was duced as a backlash to JIT pundits talking as
though zero inventory is ievable simul ly with i
under all operating circumstances. It cannot, as with all dynamic Sime there is a trade-off to
be evaluated and a judgement to be made from amongst many competing alternatives.
7. SYSTEM DESIGN LESSONS FROM 1961
The Japanese appear to find little use for simulation in the design of manufacturing systems.
Whether this is true or not in practice is an icult to establish with any degree of certainty. What
we do know is that manual simul to blish the principles of good
system design provided we keep things simple! There is a Corollary to Ashby's Law of Requisite
Variety which says that complex controls are only needed for complex systems. The absence of
computer based design techniques should therefore not be construed as meaning there is a lack of
insight into effective systems synthesis.
Is it not the case that the work of Jay Forrester and Jack Burbidge is urging us to "keep it simple"
in the manufacturing systems scenario? Let us consider the evidence; both strenuously urge the
reduction of all process flow times, and Burbidge has given us a host of tools to achieve this goal
including Group Technology and Material Flow Concepts. Yet thirty years on we still have
management consultants anxious to implement lead time reduction programmes for
manufacturing industry and offering many persuasive reasons for wanting our patronage.
Burbidge pointed out that stock control systems triggered by pre-determined re-order levels are
already centuries old. They treat each component individually and generate extremely variable
multi-cycle, multi-phase flow leading to highly fluctuating inventories and stop-go plant
utilisation. These systems are based on the belief that increasing the stock will increase service.
Perhaps because to our ancestors winter stocks of food meant survival, we may be programmed at
birth to associate high stocks with peace of mind. Instead we need to see JIT operations in the
Japanese style as a manifestation of lead time reduction coupled with simple localised production
control systems, i.e. design simplicity in.
Jay Forrester showed that feedback systems had to be designed with extreme care. But this is no
recipe to make them unnecessarily complex. Keeping delays of material flow, information flow,
and cash flow to a minimum will help enormously. Also we need to ensure that information is
not "stale" or "noisy", and the system has good filtering properties. Also esoteric games should
not take place in which decision-makers try and double-guess what is really happening further
downstream in the supply chain to the detriment of all echelons conducting business therein.
Production and Operations Management, page 113
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
Sales A
Data on present demand Department ‘
Forecast future demand
YY
7 Present customer service levels by part No.
De e
Ba
8 Bes
ete fe
Bee || 335
Sa8|| 28>
e32\|| Be2
B|| 888
xu yy Orders
placed on MRP
acy System
System Dynamics by par NO |
"To-Make' Model! ¥ .
Scheduling
See 4p 44> oftware
AEE
3 Produetion
23|| 35 schedule by
3 || 85 parte
|| s-us
gil é
3 Shop Floor
= Pi ‘a
|
3 Processes
£
s
“—j ‘Free Stock’ » End
in Store ? Customer
SEE Material flows <=> information flows
Figure 5. Block diagram representation of a multi-product
factory incorporating a ‘to-make' system dynamics model
to ensure achieving target customer service levels
8. CONCLUSIONS
So the ‘Purpose of "1961 and all that" to show that in at least two respects we have long had
tools for cturing systems design has been achieved. These tools
have been well demonstrated in industrial practice by the Japanese via a process of evolution, and
by well organised Western companies. With a thirty year knowledge lead time we really have no
excuses left!
Production and Operations Management, page 4
1994 INTERNATIONAL SYSTEM DYNAMICS CONFERENCE
REFERENCES
Burbidge, J.L. "The New Approach to Production", Prod. Engnr., 1961, 40(12), 769-784.
Burbidge, J.L. "Five Golden Rules to Avoid Bankruptcy", Prod. Engnr., 1983, 62(10), 13-14.
Burbidge, J.L. "Automated Production Control with a Simulation Capability", Proc. IFIP Conf.
WG 5-7, Copenhagen, 1984.
Cheema, P. PAD Thesis, 1994, Cardiff Industrial Systems Dynamics Group, University of Wales
Cardiff.
Forrester, Jay W. Industrial Dynamics, MIT Press, Cambridge, Mass., 1961.
Grunwald, H.J. and Fortuin, L. "Many Steps Towards Zero Inventory", European Journal of
Operations Research, 59 (1992), 359-369.
Lyneis, J.M. Corporate Planning and Design: a System Dynamics Approach. MIT Press,
Cambridge, Mass., 1980.
Monniot, J.P., Rhodes, D.J., Towill, D.R., Waterlow, J.C. "A Study of Computer Aided
Production management in UK Batch Manufacturing". Int. Jnl. Operat. Prod. Man 7(2), 1987, 2-
57.
Parnaby, J. "Business Process Systems Engineering", Paper 1, presented at the BPR Conference,
London, November 1993 published by Armstrong Information Ltd.
Stalk, G.H. and Hour, T.M. Competing Against Time: How Time Based Competition is reshaping
Global Markets, Free Press, New York, 1990.
Towill, D.R. "Coefficient Plane Models for Control System Analysis and Design", Research
Studies Press/John Wiley, Chichester, 1981.
Towill, D.R. "Supply Chain Dy ics - the Chain Engineering Challenge of the Mid-1990's",
Proc. Inst. Mech.Engnrs., 206 (1992), 233-245.
Towill, D.R. "The Dynamic Analysis Approach to Manufacturing Systems Design", The
University of Wales Review, 3 Spring 1988, 3-14.
Towill, D.R. and del Vecchio, A. "The Application of Filter Theory to the Study of Supply Chain
Dynamics", Int. Jnl. Prod. Plan and Control, 1994, 5(1), 82-96.
Weill, H.B. and Veit, K.P. "Corporate Strategic Thinking: the Role of System Dynamics", in
Milling, P.M. and Zahn, E. (Eds), Computer-Based Management of Complex Systems, Springer-
Verlag, Heidelberg, 1989.
Production and Operations Management, page 115
