Sie sind auf Seite 1von 12

Int. J.

Production Economics 59 (1999) 53—64

Are push and pull systems really so different?


M.C. Bonney*, Zongmao Zhang, M.A. Head, C.C. Tien, R.J. Barson
Department of Manufacturing Engineering and Operations Management, University of Nottingham, University Park,
Nottingham NG7 2RD, UK

Abstract
A distinction is frequently made between push and pull production planning and control systems. Many people believe
that pull systems are inherently better at reducing stocks because they try to eliminate queues, not provide for them,
whereas push systems encourage queues to cushion operations and to increase work station utilisation but at higher cost.
However, the definitions of push and pull are inconsistent between different researchers. Worse, arguments about
performance are sometimes circular. Thus, if the performance of a pull system is poor then it may be suggested that this is
because the fundamentals of JIT are not being observed, whereas, if the performance of a push system is poor, then that is
a consequence of it being a push system. After defining push and pull systems, this paper examines, by means of
simulation, the effect that push and pull information flows have on system performance, under a variety of conditions. In
particular, the performance of both push and pull information flow systems are considered in conjunction with
high-quality levels, small set-ups and small batches, i.e. the conditions normally associated with JIT continuous
improvement programmes. Similarly, the performance of both push and pull information flow systems are investigated in
the presence of conditions such as large set-up times, which are frequently eliminated as part of a continuous
improvement programme. The question investigated in this study is how system performance is affected by the flow of
control information. The investigation uses models of the material and information flows of push and pull systems to
examine the conditions which affect performance. A production sequence is chosen which consists of ordering materials,
making parts and assembling products which are then despatched to customers. A set of decision rules is used to operate
the systems using different demand and inventory level data.  1999 Published by Elsevier Science B.V. All rights
reserved.

Keywords: Push-pull systems; Material flow; Information flow; Simulation; Petri-nets

1. Introduction system is usually considered to be a typical pull


system. However, closer examination shows that
Many push and pull material control systems most practical systems consist of both push and
have been implemented. MRP is frequently de- pull. For example, a system which operates mainly
scribed as a push system. A kanban operated JIT by local pull control may use a push system to
obtain long lead time items. The Toyota system, the
classical pull system, uses push information flow for
* Corresponding author. Tel.: #44 115 951 4010; fax: #44 the vehicle and pull information flow based on
115 951 4000; e-mail: maurice.bonney@nottingham.ac.uk kanbans to ensure the availability of other parts on

0925-5273/99/$ - see front matter  1999 Published by Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 5 - 5 2 7 3 ( 9 8 ) 0 0 0 9 4 - 2
54 M.C. Bonney et al./Int. J. Production Economics 59 (1999) 53—64

the assembly track. Similarly, a MRP system may completion of a process, the job proceeds to
use local progress chasing to pull items through the subsequent processes on the designated process
manufacturing process. route. When all processes are completed the job
Our study investigated whether there was any exits from the system [1].
difference in performance as a direct result of using
push or pull information flows. A brief summary of Pull
some published material on push and pull systems Activities at the process station are triggered by
and their performance is presented. From this it is depleted output kanban stock at the process sta-
clear that some inequitable comparisons have been tions. Each depleted kanban stock constitutes
made. This immediately raises the question of how a queue unit at the station. Before a job can be
push and pull should be defined. This paper pro- loaded a check is made to ensure that the preced-
poses definitions related to information flows and ence constraint is satisfied; that is, there must be
then models several examples of each system. The sufficient inventory in the output kanban stock
performance of each system is then derived and of the upstream processes of that job. If so,
compared. a draw is made from the output kanban stock.
Should this cause the output kanban stock to fall
below the re-order level, the job is queued at that
2. Some definitions of push and pull systems station [1].
It has already been observed that most manufac- Venkatesh [2] described push and pull as opera-
turing control systems are hybrid, i.e. a mixture of tional paradigms “In a push system, a preceding
both push and pull. However, this observation is machine produces parts without waiting for a re-
not very meaningful if the terms push and pull have quest from the succeeding machine. On the other
been left deliberately vague. In order to clarify the hand, in a pull system a preceding machine pro-
terms, we examined papers which discussed the duces parts only after it receives a request from the
performance of push and pull systems. The papers succeeding machine”.
are broadly of two kinds: firstly those by authors Toni et al [3] attempt a classification of push and
who have tried to investigate push and pull systems pull applied to manufacturing systems. They con-
by simulation and, secondly those by authors who sider three sub-systems: inventory systems; manu-
have looked more descriptively at manufacturing facturing priority, material picking and moving
systems. Broadly, the former investigate well-de- sub-system; production planning sub-system.
fined manufacturing sub-systems whilst the latter Push and pull are distinguished by the source of
have a wider perspective, perhaps referring to the the information. In their classification they use the
whole enterprise. The simulation modelling exam- terms push and pull as in Goddard and Brooks [4];
ples refer to push and pull to identify the type of push means to take action in anticipation of a need,
system, but necessarily then go on to define in detail pull means to take action on request. They say the
the rules of operation and other parameters which logic has been classified by Da Villa [5] as
describe the system modelled. This makes it very push"look back and pull"look ahead. In re-
difficult to compare directly the results of these spect of inventory management techniques, push
studies. Views of some of the authors are now means launching an order based on previous con-
described: sumption, conversely pull means to take action on
Lee [1] described procedures for push and pull request and the logic on which an order is launched
as follows: is a look ahead one based on requirements. Using
these definitions, they show re-order point to be
Push
a push inventory management technique and ma-
Jobs on entry into the system are queued at the terial requirements planning to be a pull inventory
first required process. Queue priority is resolved management technique. They go on to say that it is
according to the selected scheduling rule. On not rare to find definitions completely opposite to
M.C. Bonney et al. /Int. J. Production Economics 59 (1999) 53—64 55

those given. They say the misunderstanding comes 3. Definitions used in this investigation
from missing the distinction between inventory
management sub-systems and priority assignment In this paper the definitions of push and pull
sub-systems. Similarly, they classify production systems are based on the information flows used for
planning techniques according to push and pull control. Simulation models of systems with these
logic with e.g a master production schedule being information flows were constructed and the perfor-
push if based on forecasting and pull if based on mance of each was then investigated under care-
orders. fully defined conditions.
Villa and Watanabe [6] present descriptive char- A pull system is one in which the control informa-
acterisations of push and pull, e.g. commenting that tion flow is in the opposite direction to the material
“existing approaches in Japan and the West to flow.
managing production orders are clearly different. For a single-stock location the situation is as
The former aims at assuring lean production, flow- shown in Fig. 1 i.e. the standard inventory control
ing without waste, where each stage of the manu- situation. A replenishment order is generated when
facturing process pulls the supply of items and items are taken from stock. For multiple stock
resources according to needs. The latter forces locations replenishment orders cascade through the
management to plan for fat production against the stock locations as shown in Fig. 2. With our defini-
risk of stock-out, so the loading conditions at each tion, examples of pull systems are ROP and ROC
stage in the manufacturing process have to be systems, the base stock system and also various
pushed by suitable plans”. kanban control systems.
Kenworthy and Little [7] present the following A push system is one in which the control
comments: “MRP II and kanban are frequently information flow is in the same direction as the
regarded as competing techniques, with proponents material flow.
of kanban disparagingly referring to MRP II as In a push system, instructions can be derived
a push technique, implying that materials are from one of the following: customers orders, a mas-
pushed into production faster than they emerge as ter schedule which consists of customers orders, or
finished goods, thus leading to high work in pro- a master schedule based on forecast demand. Each
gress. Kanban in contrast, is described as a pull of these will usually have a due date associated with
technique because the kanban signal or card pulls it but that date could be “as soon as possible”.
material forward only when it is required by the Either forward or backward scheduling (or both)
next stage of production”. may be used for planning, but control is forward
BS 5192 [8] defines push and pull as follows: against the targets so derived, ie the aim is that
Push system: System of ordering in which orders items will reach a particular location by a time
are issued for completion by specified due-dates, determined by central or local control (e.g. schedule
based on estimated lead times. or priority rules).
Pull system: System of ordering in which a fixed The material and information flow of an MRP
stock is held of every item and orders are issued for system is shown in Fig. 3. In MRP, the gross
the immediate replacement of any items which are
removed from stock. Examples of pull systems are
base stock control and kanban.
These definitions and the description in the pre-
vious section show that there is little consistency in
how the terms are defined. Sometimes the emphasis
is placed on inventory replenishment, sometimes
on lead times, sometimes on scheduling methods
sometimes on lean production and sometimes on
the source of information on which the logic is
based. Fig. 1. Material and information flow for single-stock location.
56 M.C. Bonney et al./Int. J. Production Economics 59 (1999) 53—64

Fig. 2. Material and information flow for multiple-stock location pull system.

Fig. 3. Material and information flow for push systems.

requirements calculation is derived level by level by product. In this paper it is assumed that the manu-
backward scheduling from product to material. facturing processes and the material flow for each
Nevertheless, the control information is from raw product are known. Pre-planning also includes set-
material to product, so that items reach the re- ting up the system and priming the system with
quired state at the right time. With the definitions initial stocks so that a kanban control system can
given, make to order systems and MRP are each work.
push systems. Production Planning: This is the process of decid-
ing what to make and over what planning horizon.
Plans include the master schedule and make-and-
4. Manufacturing control buy instructions for parts, etc. The plans can be
based on real orders or forecast demand and the
Manufacturing control systems have the follow- planning process may use formal or informal deci-
ing essential elements: materials flow, the opera- sion rules. Priorities are chosen either as part of the
tions performed, the resources used, information planning procedure or by means of separate dis-
flow, the decision rules for planning and control patching control routines. Planning can use either
and, finally, the performance measures used. Differ- or both forward-and-backward scheduling, e.g. as
ences between (and possibly within) push and pull in critical path methods.
systems performance can arise not only from differ- Action: This is the set of activities to be under-
ences in the information flows but from the decision taken in order to achieve the plans. In the context
rules, the initial conditions (stocks), the demand of this paper, the activities are the planning and the
characteristics to which the system responds and manufacturing activities.
the environment within which the system works. Progressing: This is the process of re-setting
The following steps are involved in manufactur- priorities in response to unexpected and usually
ing control: unwanted occurrences such as scrap, machine
Pre-planning: This is the process of determining breakdown, etc. Progressing is normally a direct
the operations to be performed to make the part or pull operation but may arise from recalculating
M.C. Bonney et al. /Int. J. Production Economics 59 (1999) 53—64 57

priorities of work waiting in a queue. In this 5. Scope of the study


investigation it is assumed that no progressing is
required. Simulation was chosen to investigate some as-
Forward Scheduling: This usually occurs from pects of the relative performance of push and pull
a given starting time, e.g. now and it may be as- information flow systems. The aim was that the
sumed that one should produce requirements as models should be relatively simple but exhibit some
soon as possible. If resources are limited then it will aspects of complexity. The systems chosen could
be necessary to set priorities. If coordination is potentially make two products for each of which
required, e.g. having components available for as- material had to be ordered, parts made and prod-
sembly, any mishap (causing lateness, poor quality, ucts assembled. As far as possible, the only differ-
etc.) will require progressing the work. ence between the models was the information flow.
Backward Scheduling: The assumption is that The push and pull models are broadly those repre-
backward scheduling occurs from a desired com- sented in Petri-net form in Figs. 4 and 5 which
pletion time. Typical backward scheduling correspond to the planned and reactive systems
methods are incorporated in infinite capacity described in Bonney et al. [9]. Two symbols are
(MRP) or finite capacity (MRP II), material re- used; (*) to represent a place and (") to represent
quirements planning systems. a transition. A place may be marked, in this case to
The emphasis of push systems is on planning, i.e. represent an order or stock being available. If all
looking ahead to determine what should happen, input places to a transition are marked, it can fire,
whereas the emphasis of pull systems is on pre- i.e. an activity takes place. In these diagrams, ob-
planning, e.g. production starts with pre-planned tain material, make parts and assemble are timed
levels of stock and demand is smoothed. If the transitions representing the time it takes to process
manufacturing lead time is long relative to the the activity. Referring to Figs. 4 and 5, orders are
ordering lead time, the system either has to use generated in the bottom left hand corner. In the
push information flow or have high safety stocks, push system, (Fig. 4) this has the effect of marking
for example, at a finished product warehouse. With the places representing despatch notes and
short manufacturing lead times, the system used authorising the raising of an order set which will
can be either push or pull. Table 1 shows some eventually mark the places representing material
system types, their planning methods and perfor- orders, parts orders and assembly orders. If prod-
mance measures. uct stock is available then, as both that and the

Table 1
System types and their planning methods

System type Inputs Planning action Performance measures

Push forward Customers orders and/or The orders are the Master Schedule. Choose Completion time, resource
forecast demand Priorities to finish as soon as possible utilisation, queue lengths, number
of orders rejected, MWT etc.,
stocks, cash flows
Push back Customer orders with Create a Master Schedule
delivery dates and/or
forecast demand
Derive scheduled start and end times for each Also performance against plans
operation by backward scheduling
Pull forward Customer orders plus Create a set of prioritised orders As above
planned stock at each
location
Local action maintains planned stocks
58 M.C. Bonney et al./Int. J. Production Economics 59 (1999) 53—64

Fig. 4. UNISON Petri-net model of the push system.

Fig. 5. UNISON Petri-net model of the pull system.


M.C. Bonney et al. /Int. J. Production Economics 59 (1999) 53—64 59

despatch notes places are marked, the despatch push systems. A kanban was released in the pull
transition fires and products are despatched. In the system, or 1 batch of parts issued to the push
pull system, (Fig. 5) despatch again occurs when system, if the product stock was less than or equal
product stock is available but now a replenishment to the specified control quantity in the control
assembly kanban is released and so on back rules. The control rules thereby allow the simula-
through the system to release material kanbans. tion model to represent safety stock in both the
Minor changes are needed to deal with specific push and pull systems. The meaning attached to
decision rules, e.g. to allow back-ordering or, alter- issuing a batch of parts through the push system is
natively, to lose orders not satisfied immediately. that a despatch note is created and also that an
A range of different situations was investigated. order is released which will take that order require-
These included: make to order, make to forecast, ment through the system separately. As soon as
back-ordering if stock is not available, orders lost if product stock is available to meet the needs of the
stock is not available, variability in demand, varia- despatch note, the product will be despatched. As
bility in processing times, various batching rules, had been anticipated, the results obtained from
different master scheduling rules, varying set up using the pull and push systems in these highly
times, etc. One difficulty was how to make equi- constrained situations were the same.
table comparisons. To achieve this, the batch size in
the push system was chosen initially to be the same
as the kanban controlled container size (henceforth 6. Discussion of the experiments
called the kanban size or just the kanban) in the
pull system. The initial stocks in each system were Experiments were performed to see how the dif-
chosen to be the same and at the same locations. ferent information flow systems responded to
The generated load arises as a set of “orders” to greater internal and external (order size) variability
which the system is asked to respond. The orders and to changes in other variables. In each experi-
input to each system were generated using the same ment 120 orders were released. Each order corre-
random number seeds. Likewise, the seeds of the sponded to a demand for approximately 20
random numbers used to generate the transition products and so the total demand, the sum of the
times for material ordering, make parts and prod- individual order quantities, is approximately 2400.
uct assembly, though different from each other Each experiment is outlined and the results dis-
were, respectively, the same in the equivalent push cussed. In Tables 3 and 4 the terms “mat”, “part”
and pull systems. and “pd” are the number of kanbans for material
A preliminary experiment was performed to test stock, part stock and product stock in the pull system
how the two systems responded to demand for and the equivalent initial stocks in the push system.
a single product. The orders placed on the system
were generated by sampling from a uniform distri-
bution in the range of 15—25 (mean of 20 units). One 6.1. Experiments 1, 2 and 3 — demand not satisfied
order of variable size, but with a mean of 20 units, immediately is lost
was generated per hour. The processing times at
each transition were chosen to be constant, respec- The experiments are summarised in Table 2a
tively, 1, 3 and 1 h for the material order, make and the results are shown in Table 3.
parts and assemble transitions. The kanban size in In the first two experiments variability was intro-
pull and the batch size in push, were each set at 20 duced into the time taken for each transition. In the
items. The initial stocks were chosen to be 1 batch first experiment, the same seeds were used so that
or 1 kanban of material, 3 batches or 3 kanbans of the same variability was introduced into each
parts and 1 batch or 1 kanban of products. When transition. However, the performance of each sys-
orders arrived, demand not satisfied immediately tem was affected because of randomness in the
from stock was lost. A range of control rules was processing time. The results were that, if the stocks
used to control production in both the pull and were the same, more demand was satisfied using
60 M.C. Bonney et al./Int. J. Production Economics 59 (1999) 53—64

Table 2

(a) Simulations performed when back orders not allowed

Experiment Range of order size Push batch size Pull kanban size Special rules

1 15—25 20 20 Same variability in each transition


2 15—25 20 20 Different variability in each transition
3 5—35 20 20 Greater order size variability

(b) Simulations performed when back orders allowed

Experiment Range of order size Push batch size Pull kanban size Special rules

4 15—25 20 20 As expmt 2 but demand backlogged


5 15—25 160 20 Assume MPS"1 day
6 Two products 160 20 Assume MPS"1 day
(a) 15—25 (p"0.7)
(b) 15—25 (p"0.3)

The processing times in hours for 1 batch of 20 were obtained by sampling from a uniform distribution with ranges: 0.9—1.1, 2.7—3.3 and
0.9—1.1 for material ordering, make parts and assemble.

Table 3
Results of experiments 1—3

Experiment Outputs Experiment 1 Experiment 2 Experiment 3


variables
Control Rules (PdAv)

)5 25 )5 25 )5 35

mat"1 Satisfied demand 1960 2291 1920 2240 1560 2120


Pull part"3 Rejected demand 425 94 465 145 795 235
pd"1 Final stock 20 89 40 40 40 40
mat"1 Satisfied demand 1960 2320 1920 2280 1560 2160
Push part"3 Rejected demand 425 65 465 105 795 195
pd"1 Final stock 20 40 40 40 40 40
mat"1 Satisfied demand 1960 2320 1920 2280 1560 2180
Pull part"4 Rejected demand 425 65 465 105 795 175
pd"1 Final stock 20 40 40 40 40 40
mat"1 Satisfied demand 1560 2180
Push part"4 Rejected demand 795 175
pd"1 Final stock 40 40
mat"2 Satisfied demand 1960 2300
Pull part"3 Rejected demand 425 85
pd"1 Final stock 20 40

a push information flow system. When the number performance, but it still did not match the results of
of part kanbans was increased the performance push.
of the two systems became the same. Increasing In experiment 2, different random numbers were
the number of material kanbans improved the used to introduce different variability into the
M.C. Bonney et al. /Int. J. Production Economics 59 (1999) 53—64 61

processing time at each operation. As in experiment but restricted stock levels the mean waiting time for
1, similar results were obtained if the number of demand to be satisfied was longer for pull systems
part kanbans was increased. With 3 batches of part than for push but as stock increased, the mean
stock in each system, the push system performed waiting times became the same. This is shown in
better in the sense that more demand was satisfied. Table 4.
In experiment 3, greater order size variability Experiment 5 introduced a simple master pro-
was introduced (uniform 5,35). Randomness of the duction schedule into the push system. This sched-
transition times was as in experiment 2. Although ule was perfect in the sense that it smoothed the
many more orders were rejected by both systems, actual daily requirements over the planning hor-
the push system satisfied more than the pull. If the izon of 1 week. The total order quantity (TOQ)
number of kanbans for a part was increased to 4 in over 1 week is an aggregate of the requirements
the pull system, then the results were found to be represented by the orders for that week. The daily
the same as when the part stock was equal to production requirement was B"TOQ/5 which
4 batches in push. was then made in one batch. Because the average
order size was 20, an 8 h day assumed and one
order received per hour, the batch size B was ap-
6.2. Experiments 4 to 6 — demand backlogged proximately 160. The results of this rule were then
compared with a pull system which, as before, oper-
The experiments are summarised in Table 2b ated with a kanban size of 20. The larger batch size
and the results shown in Tables 4—6. In the first understandably increased work in process stocks
three experiments demand not satisfied immedi- and also increased the manufacturing lead time.
ately from stock was lost. A more realistic model is This is indicated in Table 5 where it will be seen
to backlog unsatisfied demand and despatch it that the push system with larger batch size resulted
when products become available. With this rule, in a longer mean waiting time (MWT) for orders to
both the kanban (pull) and make to order (push) be satisfied because of the later delivery.
systems could potentially have delays in satisfying Whereas experiments 1—5 dealt with a single
demand. Experiment 4 found that under equivalent product, experiment 6 considered the effect of

Table 4
Results of experiment 4

Experiment Outputs Experiment 4


variables Control Rules
(PdAv) )5
Initial Stock
(PdAv)"20

mat"1 Total demand 2385


Pull part"3 Mean waiting time 3.678
pd"1 Final stock 12
mat"1 Total demand 2385
Push part"3 Mean waiting time 2.030
pd"1 Final stock 12
mat"1 Total demand 2385
Pull part"4 Mean waiting time 1.159
pd"1 Final stock 12
mat"1 Total demand 2385
Push part"4 Mean waiting time 1.159
pd"1 Final stock 12
62 M.C. Bonney et al./Int. J. Production Economics 59 (1999) 53—64

Table 5
Results of experiment 5

Experiment 5

Pull Push

Control rules (PdAv)

)R R R R MPS MPS MPS MPS

Initial stock 20 80 160 170 20 80 160 170


Total demand 2385 2385 2385 2385 2385 2385 2385 2385
Mean waiting time 3.678 0.849 0 0 4.334 1.876 0.192 0.114
Final stock 35 95 175 185 30 90 170 180

Table 6
Results of experiment 6

Outputs Type Experiment 6


of part Initial stock: a"112, b"48

Set up time

0 0.05 0.10 0.15 0.20

Total demand a 1717 1717 1717 1717 1717


b 650 650 650 650 650
Pull Mean waiting time a 0 0.091 0.664 1.870 3.636
b 0 0.015 0.271 1.676 3.716
Final stock a 115 115 115 115 115
b 78 78 78 78 78

Total demand a 1717 1717 1717 1717 1717


b 650 650 650 650 650
Push Mean waiting time a 1.417 1.615 1.852 2.125 2.431
b 0.120 0.168 0.267 0.422 0.597
Final stock a 125 125 125 125 125
b 53 53 53 53 53

producing two products. The situation considered bigger the set-up times, the poorer became the
was where 70% of orders were for product “a” and performance of both systems. However, because the
30% of the orders were for product “b”. Again the batch size chosen in the pull system incurred more
MPS generated 1 days work but now this was in set-ups, the performance of the pull system pro-
the form of two batches, one for “a” and one for gressively deteriorated relative to the push system
“b”. Set up time was incurred whenever there was as set-up time increased. This illustrates the obvi-
a change from making “a” to making “b” or vice ous point that reducing set up times is important if
versa. When set up times were small, the push one wishes to use small batches. This is illustrated
information flow system had a longer MWT. The in Table 6.
M.C. Bonney et al. /Int. J. Production Economics 59 (1999) 53—64 63

7. Observations and conclusions ROP/ROC pull systems have not been discussed at
all, although their performance could, by changing
Different results were obtained for the push and the control rules and the kanban size, use the same
pull systems. Provided that the batch sizes were the simulation model. Initially we compared a single
same and orders were released order by order, then product make to order push system which started
push information systems performed better, operating with the same (base) stock as the pull
whether measured by demand rejected when no system (which issued from stock and generated
backlogging occurred or MWT when demand was kanbans for replacement orders). The stocks/batch
backlogged. A related result was that push systems sizes, etc., in each system were the same. Then,
sometimes required lower stocks to obtain equiva- provided that information processing times were
lent performance. This probably arose because in the same, which they were because of the constant
a push system a replenishment order is released processing time at each transition, the performance
immediately whereas, in the pull system, there may of both systems was identical. However, delays in
be a delay when waiting for the kanbans to cascade relaying the pull information can occur when
back through the system. transition times are variable, and this could affect
The experiments with the kanban pull system the performance of pull relative to push.
and the make to order push system used the same If more than one product is made, then schedul-
initial stocks so as to provide equivalence. How- ing is required. Pull systems potentially have a de-
ever, some differences are immediately apparent lay associated with the time taken to make the
between the two systems With zero planned stock, kanban of parts and therefore push information
the make to order system would always be able to flow systems will in general be better.
satisfy an order even though on occasions it would Further investigations of multi-product situ-
be late, whereas, with a zero number of kanbans, ations, with different demand patterns and different
a pull system would not work at all. If a pull system methods of choosing a master production schedule,
runs out of stock then a delay occurs equal to the are needed. When forecasting in a multi-product
processing time remaining on the previous replen- situation with long lead time items, the problem
ishment kanban. Over the complete set of manufac- becomes more complex. First, there is the choice of
turing operations, a delay can occur which is not master production schedule which allows a trade
greater than the sum of the individual processing off between stocks and schedule stability. Second,
times of 1 kanban at each operation. With 1 kan- both push and pull systems allow the possibility of
ban at each location, the replenishment time would aggregation, which may reduce the manufacturing
usually not be more than the processing time of costs, increase capacity or just increase stocks.
1 batch at the last operation, but potentially this We have provided some insight into the necessity
could increase. At the extreme, for example, in start for reducing set-up times in pull. We believe the
up condition, the replenishment time could be the same effort to reduce set-up times should be applied
sum of the processing times at each operation. As to push. If so, push appears to be less constrained
the number of product kanbans increases, the prob- than pull, e.g. in response to individual orders or to
ability of satisfying an order, obviously, tends to wide demand variability. The implication of the
one. This suggests that one should be able to calcu- analysis in this paper is that, if lean production
late the number of orders satisfied as a function of methods are used, then push systems should per-
the number of kanbans. If this is so then one could form at least as well as pull systems provided batch
calculate the number of kanbans which would sizes are not allowed to grow and the resources are
minimise the sum of the stock holding costs and not overloaded. In addition to the improvements
lost order costs, a somewhat alien approach for pull made to manufacturing by the lean production
systems. philosophy, there is also an opportunity to improve
The investigation has touched on the problems performance further by using scheduling methods.
of MRP, the classic push system, at which many Finally, the local information processing and vis-
criticisms are levelled. On the other hand, the ibility of most practical pull systems probably has
64 M.C. Bonney et al./Int. J. Production Economics 59 (1999) 53—64

an important effect. This was not included in the many helpful suggestions for improvements to the
models and it may be that this is the reason that the original paper.
simulation results are inconsistent with conven-
tional wisdom. On the other hand, although less
References
commonly used, visual methods of local scheduling
are possible with push. [1] L.C. Lee, A comparative study of the push and pull pro-
It is believed that the comparisons that have ductions systems, International Journal of Operations and
been made in this paper provide some insight into Production Management 9 (4) (1989) 5—18.
the effect of information flows and, in particular, [2] K. Venkatesh, M.-C. Zhou, M. Kaighobadi, R. Caudill,
A Petri net approach to investigating push and pull para-
indicate that it may be possible to obtain similar digms in flexible factory automated systems, International
performance improvements with push systems as Journal of Production Research 34 (3) (1996) 595—620.
have already been produced by pull systems. As the [3] A. Toni De, M. Caputo, A. Vinelli, Production manage-
results are a consequence of using similar condi- ment techniques: push — pull classification and application
tions, no doubt the same concerted effort will be conditions, International Journal of Operations and Pro-
duction Management 8 (2) (1988) 35—51.
required using push control systems as has been [4] W.E. Goddard, R.B. Brooks, Just-in-time: a goal for MRP
applied to successful pull systems. Perhaps the II, Readings in Zero Inventory, Conference Proceedings
question is how to achieve the same commitment APICS, 1984.
with push systems as has been achieved with the [5] F. Da Villa, Programmazione e controllo della produzione,
best pull systems? Etas Libri, 1985, p. 39.
[6] A. Villa, T. Watanabe, Production management: beyond
the dichotomy between ‘push’ and ‘pull’, Computer Integ-
rated Manufacturing Systems 6 (1) (1993) 53—63.
[7] J. Kenworthy, D. Little, When push comes to shove is
Acknowledgements MRP II infinite push or finite pull? BPICS Control (1995)
31—32.
[8] BS 5192: Giude to production Control Part I: 1993, British
The authors gratefully acknowledge the British
Standards Institution, London.
Council support of Zhang Zongmao as a visiting [9] M.C. Bonney, N. Huang, M.A. Head, C.C. Tien, R.J.
scholar to the University of Nottingham. Thanks Barson, Inventory and Enterprise Integration, Interna-
are also extended to the two referees who provided tional Journal of Production Economics 45 (1996) 91—99.

Das könnte Ihnen auch gefallen