20
INTEGRATING MRP WITH KANBAN/PULL SYSTEMS by Uday S. Karmarkar Working Paper Series No. QM8615 June 1986 Center for Manufacturing and Operations Management The Graduate School of Management The University of Rochester

INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

  • Upload
    others

  • View
    4

  • Download
    0

Embed Size (px)

Citation preview

Page 1: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

by

Uday S. Karmarkar

Working Paper Series No. QM8615

June 1986

Center for Manufacturing and Operations Management The Graduate School of Management

The University of Rochester

Page 2: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

..

Page 3: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

Uday S. Karmarkar

Center for Manufacturing and Operations Management Graduate School of Management

University of Rochester

ABSTRACT

Push manufacturing systems like MRP and pull systems like Kanban, are

sometimes thought of as representing antithetical approaches to controlling

production. However. these approaches involve principles that are not

incompatible. A hybrid system is described that combines MRP at the

plant-wide level with a Kanban approach at the loca] cell leve]. The

calculations required in the MRP system are simplified. but the ability to

utilize advance information about demand is retained. Simultaneously, the

reactive abilities and the appealing incentive structure of the Kanban system

are incorporated in the hybrid system. This approach is appropriate for a

repetitive batch manufacturing environment. with part fabrication. subassembly

and assembly stages.

Page 4: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS
Page 5: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

Uday S. Karmarkar Center for Manufacturing and Operations Management

Graduate School of Management University of Rochester

.. Introduction

Push systems of .anufacturing control, such as Material Requirement

Planning (MRP) were originally seen as improving on pull systems of the Order

Point, Order Quantity type. Unlike the latter, MRP s~stems take actual

demands into account, and trigger production of "dependent" demand items in

the required quantities, and at the required times as estimated by the lead

time offsetting procedure. While this approach sounds suspiciously like a JIT

system, it does not appear to work that way in practice. A problem with MRP

logic is that the dependence of production lead times on shop conditions,

batch sizing and release timing is not recognized. Instead, lead times are

taken to be given exogenously as fixed para.eters. This inability of MRP to

relate lead times to capacity loading and order release policies can lead to

poor on-time, work-in-process and lead time performance.

Despite the seeming inefficiencies inherent in pull systems, due to their

lack of use of information about demands, the Japanese Kanban system appears

to function effectively 1n certain production environ.ents. While, pull

systems are not, from a definitional point of view "just-in-ti.e", they are so

effective as to be identified closely with JIT methods. At least one reason

for the success of these systems is that they create an informational and

organizational environment, where the role of production lead times becomes

obvious, and where there are incentives to control and reduce lead times.

- 1 ­

Page 6: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

Descriptions of the working of Kanban systems have been given by Hall

(1983), Monden (1983) and Schoenberger (1983), who also describe other aspects

of Japanese manufacturing strategy and JIT techniques. Karmarkar (1986)

discusses the complementarity between Kanban systems, and other aspects of

manufacturing strategy, and relates the Kanban approach to other pull control

techniques. The dynamics of Kanban systems have been discussed by Groenevelt

and Karmarkar (1986), Huang,Rees and Taylor (1983), Karaarkar and Kekre

(1986), and Kimura and Terada (1981). The comparative effectiveness of push

and pull systems has been discussed by Kekre (1985), and Karmarkar and

Shivdasani (1986) correlate the use of push and pull .anufacturing control

techniques with the characteristics of the physical process being controlled.

Finally, Karmarkar (1986) discusses the evolution of push and pUll systems.

and argues that key features of both are captured in hybrid control systems

that combine aspects of the two approaches. Groenevelt and Karmarkar (1986a,

1986b) develop techniques for forecast-driven Kanban systems that adapt

dynamically to changing demand conditions, and describe an application.

In this paper we discuss the design and operation of a hybrid control

system for a batch .anufacturing environment. The overall facility is

controlled by an MRP approach, which computes requireMents at the item and

subassembly level from a master schedule. Locally, the facility is organized

into .anufacturing cells which are each operated using Kanban .ethods. The

difference with the conventional approach is that the MRP syste. provides each

cell with future production requirements for that cell. This information is

used to change the parameters of the Kanban control scheme. The basic

approach is to control the card count within the Kanban system on the basis of

- 2 ­

Page 7: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

the requirements generated by the MRP system. An extension of this approach

is to also control the production batch sizes in the cell based on the

information provided by the MRP system.

Requirement Triggered Kanbans

The cards (Kanbans) in a Kanban pull system play the role of work orders.

That is, the release of a card due to withdrawal from inventory, authorizes

production of a replenishment batch. The quantity to be produced is also

indicated on the card. The mechanism of the release timing process -- by the

physical withdrawal of output inventories -- is what categorizes a Kanban

system as a pull system. This pull release Mechanism does not use any

information about future demands. and production is not triggered for any

particular future requirement. As a result. even if information on impending

variations in demand is available well in advance. pull systems in their

pristine form cannot incorporate it in their operation.

By contrast. an MRP system directly releases work orders by computing net

requirements for each item. by combining requirements into production or

release lot sizes, and then by determining the release time for these planned

orders by offsetting the order based on production lead time data. The timing

of production thus depends directly on the lead times assumed by the MRP

system. If these lead times are not accurate. the resulting effect on

production timing can cause excess inventories or missed due dates, depending

on the nature of the errors.

In the hybrid system, MRP is still used to compute gross requirements for

each item and subasseMbly, but not net requirements. These gross requirements• are used to set the number of cards in each cell. However, they do not

- 3 ­

Page 8: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

directly determine the timing of production; that is still left to the local

control of each cell. As a result, the total material inventories in the cell

are controlled by the cell superv i ~('J. who determines when a card ",i 11

•actually be released for production.

Let us suppose for a moment that the MRP system had precise knowledge of

what the lead times would be at the time that each unit was produced. This

lead time would include queue time at each work center or cell. The total

quantity associated with cards and in finished inventory in a cell at any time

must equal the production required from the cell in the next lead time

interval. To see this note that a card just released to the cell will be

required after a lead time interval. Furthermore, any card released less than

a lead time interval in the past would not as yet be removed from the cell.

This concept is analogous to the base stock system (see Karmarkar. 1986;

Groenevelt and Karmarkar, 1986 for a discussion).

Now if the lead time at each work center for each order or lot were

actually known exactly, the finished inventory in the system would be zero,

the cards in the system would correspond to demand over the next lead time

interval, and items would be produced for finished inventory just as they were

demanded. However, in practice such clairvoyance has not been present in MRP

implementations. Of course, the underlying problem is that lead times are a

aoving target, constantly changing with shop load, job mix and with random

events. As a result, we can assume that the lead times in the MRP system, are

as usual inflated to cover the variability and ignorance about actual lead

times. A card will be released to a work center at a time that is a safe

estimate of a lead time away from the actual requirement for the finished

- 4 ­

Page 9: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

material. This gives the cell supervisor or mana2er some leeway to decide on

the actual time that a card authorized by the MRP system can be physically

released to the floor.

To summarize, the conceptual working of the system is as follows: master

schedules are the input to MRP computations that generate gross requirements

by item. offset by estimated lead times. The gross requirements determine the

cards released to authorize production of each item to the corresponding cell

or work center. The total number of cards in a work center at any given time.

whether in queue, in process or in finished inventory, are equal to the gross

requirements for that work center over the lead time interval for the work

center, as estimated from the lead time information available to the MRP

system. While that is the "authorized" quantity of cards, the Dlanageror

supervisor of the cell may choose to use a smaller number of cards to be

active at any given time. Of course, if this can be done consistently, it

implies that the lead times used in the MRP computations are too large.

This entire approach is most appropriate for repetitive batch

.anufacturing environments with parts and raw material purchasing, parts

fabrication, subassembly and assembly production stages. The cell

organization with pull control can be employed in both fabrication and

subasseDlbly stages. The approach is not appropriate for a custom or

en2ineered product environ.ent or for a custo.er order oriented system. In

the latter case, the identification of production lots with particular

customer orders can only be done at the final assembly stage. Customer orders

cannot be identified with particular production batches at other stages very

precisely, since there is not specific batch or production work order to bp

associated with an end item order. Note that kanban cards are "generic" and

- 5 ­

Page 10: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

do not have any special identification that would allow pegging to customer

orders. Of course, a system could be devised that would generate cards with

specific pegging information associated with them. However, the added costs

for information processing and implementation should be recognized. The major

purpose of pegging is to permit expediting of particular orders. The pull ,

system approach is predicated on making the production system so reactive and

reliable that expediting is not required. Or rather, the detection of

specific order related problems is replaced by the rapid identification of

underlying process problems.

The detailed operation of the system involves the precise rules for

generation and removal of cards. Cards are generated by the MRP system, but

may also be added locally. In particular buffer inventories and compensation

for yield, scrap and rework can be handled through local control of cards.

These details and modifications are discussed in a later section.

The Benefits from Hybrid Systems

The rationale behind the hybrid approach is that the useful features of

both push and pull systems can be melded into one system. Broadly, the push

features permit the effective use of detailed knowledge about demand

variations. The pull features provide the appropriate incentives at the cell

level, and .ake use of the detailed information available at the local level,

without trying to collect that information in a centralized system. There are

several specifically identifiable advantages to this hybrid scheme.

Local Inventory Manage.ent. A major role of a conventional MRP system is

inventory management. The quantities (and perhaps location) of on-hand •

inventories at all production stages are controlled by the system. Such an

- 6 ­

Page 11: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

approach removes responsibility of controlling and maintaining inventories

from the local cell level. In the hybrid system, inventories in a cell.

whether of raw material or finished parts. are the responsibility of that

cell. As a result, there is no need to impose discipline about inventory

.anagement from "above" or to monitor inventory levels and locations. It is

in the cell supervisor's interest to see that inventory levels and'locations

are known and controlled. It should be noted that there are situations where

inventory tracking is required because lot identities must be maintained. For

example, in batch chemical production. each batch may have its own composition

which has to be recorded and the batch located. Centralization of this lot

tracking requirement may be the most efficient approach. especially since the

information may be required at separate points in the plant and process.

Reduction of Shop Floor Data Collection. Since material and inventory

manage.ent on the floor is decentralized. data on inventory levels and

locations does not have to be collected and centrally maintained.

Furthermore. individual orders do not have to be tracked. since the

responsibility of supplying a succeeding cell promptly. lies with the cell

.anager. Recall that this system is thus most appropriate for a repetitive

batch flow environment. rather than for engineered products. customized

products or customer order oriented production.

It should be noted that while current shop status information does not

have to be collected. shop performance over a given period does have to be

.easured. Among the measurements that should be made are cell lead times,

average work-in-process levels, scrap and rework levels, backorder and

shortage incidence. and setup and processing times. Many of these• measurements can be conveniently done using the card system itself.

- 7 ­

Page 12: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

Reduction of MRP Computations. Since inventory management is a local

function. and since the incremental quantities produced are at the discretion

of the cell. Certain computations become unnecessary. In particular. netting

of requirements is not required; in effect it is done locally by determining

which of the authorized cards are already covered by available material.

Purthermore. pegging is not required and the associated data management

overhead can be eliminated. Even if pegging is to be implemented, MRP

generated cards can be pegged directly to end item orders or the end item

production schedule. Stage by stage pegging is not possible. and should not

be necessary.

MRP computations in this system are thus reduced to simply converting end

item production schedules to gross requirements for each item. offsetting the

gross requirements by total lead time for that item. and then computing the

corresponding card count or equivalently a card generation schedule. Note

that this can be done without a level by level BOM explosion. and a single

level BOM is adequate.

Reactive Ability. Since the cell level is controlled with a pull

approach, the manufacturing system is able to react to unplanned events by

local adaptation. Por example. in a pure push system, the loss of part

inventory due to damage would require the generation of a new order from the

MRP system. The loss would only be discovered at a cycle count point or when

the inventory was found to be unavailable for use. In the pull framework.

inventory responsibility lies with each cell. and there are incentives to

continually monitor inventories. A loss of inventory would tend to be noted

- 8 ­

Page 13: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

immediately, and replacement initiated. Note that this causes unplanned use

of incoming material and the card generation scheme must be modified to

account for the additional production.

Local Incentives for Performance Improvement. An important feature of

the Kanban environment, is that there are incentives at the local level for

lead time and WIP reduction, and quick response to contingencies. These

incentives are lost in an MRP or push environment, because leadtimes are

planned by the system. and allowance is made for leadtime variability in the

release policy. As a result. there is no benefit and no reward for reducing

production lead times, or equivalently, work-in-process. In this hybrid

system, the cell level incentive structure can be maintained by assigning

responsibility for inventory levels to the cell, and by clarifying that the

cards released by the MRP system are authorization to produce and not

production releases in the usual sense. The cell manager has the option of

holding a card as late as possible, to reduce the importation of raw material

into the cell.

In addition to the incentives, it is also important that the supervisor

of the cell has the authority to make the decisions on layout, equipment

choice, labor assignment. engineering improvements and so on, that can effect

setup reduction, lead time reduction and flexibility improvements in the cell.

Perforaance Measurement. The cell organization that is maintained in the

systea requires the measurement of manufacturing performance at the cell

level. Lead time monitoring is especially important. Production lead time in

the cell is measured as the average time taken from the time that a card is

posted in the cell to the time that finished material corresponding to that

card is available for use by succeeding cells. The use of cards generated for

- 9 ­

Page 14: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

each cell simplifies this measurement task. Furthermore, as lead times are

reduced. there is immediate feedback to the central MRP system. since the

release time of cards by the system can be compared to the time that the cards

are actually released to the cell.

Implementation and Modifications

One possible realization of the card generation will be described here.

It is important to note that special features of particular environments may

require modification of the basic process. Some of these modifications will

be described later in this section. It will be assumed that the plant

includes some part fabrication. subassembly operations, and final assembly of

end items.

Given a master schedule for final items, the MRP computations are

simplified to computation of gross requirements only. As described in the

previous section. inventory management and netting is not done at the plant

level but at the local, cell level. The pattern of gross requirements

determines the card count in the cell: the quantity of material on order in a

cell at any point is equal to the requirements on the cell in a period equal

to the production lead time for the cell. The gross requirements for each

item are suamed from the current time to a horizon equal to the lead time for

the cell. This number is then converted into the nu.ber of cards released to

the cell. by dividing by the batch size associated with each card. A rounding

procedure is used to give an integer number of cards and the difference

between actual and rounded values is either carried forward or subtracted from

the next requirement as appropriate.

- 10 ­

Page 15: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

..

This computation gives the total number of cards active in the cell at

any time. This number will change with each period so that cards have to be

added or· subtracted accordingly. An alternative way to look at this is that

the change in the number of cards in each period is obtained by subtracting

cards for the gross requirement in the current period, and adding cards for

the gross requirement in the period that is one lead time away.

It is instructive to compare this scheme with the approach used in rate

oriented production as in the automotive industry. Typically there too,

requirements are not netted; rather the cumulative requirement for each item

is computed from gross requirement. Production then "chases" the cumulative

requirement. The problem with this approach is that the current load on a

shop is not apparent from the system. In the hybrid system proposed here.

this load is precisely the gross requirement (demand) over the succeeding lead

time; which is communicated as the "open" number of cards or equivalently the

quantity "on order".

The scheme described thus far has the advantage of simplicity in

coaputations. However, it has certain limitations. First of all it assumes

that production lead times for each cell are known to the central MRP system.

In general. these lead times cannot be accurately known centrally because they

change with shop loading. Of course, unlike the usual MRP procedure, these

lead tiaes determine card quantities and not actual production timing.

Nevertheless, if lead times change substantially over time, the number of

cards generated may not be correct. Secondly. the central system assumes that

there is no uncertainty in production. If in fact, there is the potential for

quality problems, yield and rework. late material deliveries and so on. the

- 11 ­

Page 16: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

number of cards authorized at any time lIay not be correct. Certain

modifications to the basic system can ameliorate some of these problems. We

describe two such 1I0difications.

Consider the lead time issue: the problem here is that if lead times at

sOlie point. are lIuch longer than "norllal" due to a high load on the shop. the

number of cards generated will be sllaller than necessary. In a conventional

MRP system. the COII.on response to this variability in lead times is to

inflate planned lead times sufficiently to cover the variations.

Unfortunately, this means that most of the tille, the lead time allowed is .uch

longer than is necessary and as a result. work-in-process and finished

inventories are increased. What is more, because the lead time data is

maintained in a centralized data base, there is no incentive to correct

inflated figures. Lead times once increased, tend to stay high. In the

hybrid system. this situation can be avoided to a considerable extent. First

of all, even if lead times in the MRP system are inflated. each cell has the

option of not putting authorized cards into production immediately. Since

releasing cards to production raises inventories in the cell, there are

incentives to delay production as far as possible. In addition, the basic

card generation system can be modified by allowing cell managers to generate

extra cards. For example, if a cell lIanager realizes that lead times are

likely to increase because of increased load on the cell or because of lIachine

failure, absenteeism or other losses of production capacity. the lIanager can

recognize that the number of cards needed in the cell is larger to cover the

longer lead time. The lIanager should have the ability to generate such cards

and it is useful to have the extra cards (electronic or otherwise)

..

~.

- 12 ­

Page 17: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

distinguished by color or some other method. Tracking the lead time and time

in finished inventory experienced by these cards will measure the

effectiveness of this system.

In the preceding example, there is no change in the total cumulative

quantity produced. The extra cards are produced to account for lead time

variability, and create a larger pool of work-in-progress (on order) to allow

for the increase in lead times due to special conditions that are foreseen by

a cell manager. The same local card generation scheme may also be used by a

cell manager to create safety or buffer stocks to protect against unforeseen

machine failure or unforeseen short term variations in demand. The important

point to note here is that these buffer or safety stocks are locally created

in response to local problems. Furthermore, since the manager bears the costs

of the added inventories, there is an incentive to reduce these stocks as much

as possible, and still maintain service levels. This .echanism is very

different from centrally determined safety stocks and safety times, where

there is the continuing risk of institutionalizing errors in estimation.

Now consider the very different problem of accounting for imperfect

yields, quality problems, and inventory losses. The issue here is that unlike

the former case, here total production in the cell, and in all parts of the

production system supplying the cell, must be increased to account for the

lost production. In effect, the total quantity demanded over the current lead

time interval, for all supplying cells is increased. There are two ways in

which this situation can be handled. One is to simply depend on safety stocks

held in each cell against downstream production loss problems. This is not

the best approach, because the point of protection is removed from the point

at which the problem occurred and managers who have no control over the

- 13 ­

Page 18: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

proble~ are being penalized by having to hold inventories. An alternative

approach is to require that extra production to cover losses should require

generation of extra cards. These cards would be produced for all suppliers of

the cell in which the loss occurred, and would be correspondingly marked. Of

course, new purchase orders to account for lost .aterial must be generated,

and this system would also create these new orders. The drawback of this

system is that it imposes a high administrative cost on the system. It would

be most efficient to produce the extra cards centrally, on notification from

the cell. Such a procedure would allow efficient (net change) processing of

the extra ~aterial, and would simplify maintaining data on yield and quality

performance.

Summary

This paper has described a hybrid push/pull system embodying elements of

both kinds of control system. An MRP system without netting is used for

material planning and for generating purchase orders. The manufacturing

process is organized as cells which are operated by a Kanban system, using

cards or some other signalling or triggering device. Unlike the typical

Kanban system, the number of cards in the system is varied dynamically. Cards

are generated by the MRP system, based on the gross requirements for each

cell, over a cell production lead time interval. Thus, the cell is able to

react to changing demand and load conditions by making use of advance

information about planned orders. This is the "push" part of the process. At

the same time, production activity is triggered by the cards and the

incentives in the system are local as in a pull system. In particular,

- 14 ­

Page 19: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

inventory is locally "owned" and managed, and safety or buffer stocks are set

locally. This hybrid approach is suitable for repetitive batch production

where lot identity does not have to be maintained .

- 15 ­

Page 20: INTEGRATING MRP WITH KANBAN/PULL SYSTEMS

REFERENCES

Groenevelt. H., and Karmarkar. U. S. (1986a), "Flexible Dynamic Kanban Systems: A Case Study". Graduate School of Management, Working Paper No. QM • University of Rochester, Rochester, New York.

Groenevelt, H.• and Karmarkar. U. S. (1986b), "Forecast Driven Kanban Systems", Graduate School of Manage.ent, Working Paper No. QM University of Rochester, Rochester, New York.

Hall, W. R. (1983), "Zero Inventories", Dow Jones, Irwin, Illinois.

Huang, P. Y., Rees, L. P. and Taylor, B. W. (1983), "A Si.ulation Analysis of the Japanese Just-In-Time Technique (with Kanbans) for a Multiline, Multistage Production System", Decision Sciences, Vol 14, pp. 326-344.

Karmarkar, U. S. (1986a), "Push, Pull and Hybrid Control Schemes", Graduate School of Management. Working Paper No. QM8614, University of Rochester, Rochester, New York.

Karmarkar, U. S. (1986b), "Kanban Systems", Graduate School of Management, Working Paper No. QM8612, University of Rochester, Rochester, New York.

Karmarkar, U. S., and Kekre, S. (1986), "Batching Policy in Kanban Systems", Graduate School of Management, Working Paper No. QM ,University of Rochester, Rochester, New York.

Kar.arkar, U. S. and Shivdasani, I. M. (1986), "Alternatives for Batch Manufacturing Control", Graduate School of Management, Working Paper No. M8613, University of Rochester, Rochester, New York.

Kekre, S. (1984), "Management of Job Shops", Ph.D. Thesis, Graduate School of Management, University of Rochester, Rochester, New York.

Ki.ura, O. and Terada, H. (1981). "Design and Analysis of Pull System, a Method of Multi-Stage Production Control", International Journal of Production Research, Vol. 19, No.3, pp. 241-253.

Monden, Y. (1983), Toyota Production System, Industrial Engineering and Management Press. Norcross, Georgia.

Schonberger, R. J. (1983). "Applications of Single-Card and Dual-Card Kanban" , Interfaces, Vol. 13, No.4, pp. 56-67.

• •

- 16 ­