Printer Friendly

A modified MRP for a production system with the coexistence of MRP and kanbans.


During the last decade just-in-time (JIT) manufacturing has become more and widely accepted in U.S. industry. Before the introduction of JIT production in the U.S., the materials requirement planning (MRP) system was the predominant methodology of production management. Since JIT and MRP are two independently developed approaches for managing production and inventory, the two systems are commonly viewed as two mutually exclusive systems.

Actually MRP and JIT systems can coexist. An obvious case of the coexistence of MRP and JIT takes place when an MRP system is converted into a JIT system. Since a JIT system cannot be fully implemented in a very short period of time, an ongoing process of converting an MRP system into a JIT system generally results in a coexistence of both MRP and JIT systems.

Even in a company fully applying a JIT system, some material ordering or part manufacturing processes may function better under MRP than JIT. The coexistence of JIT and MRP could be advantageous in managing a production system. For example, a company that orders a certain part from overseas may have a better control over the part if they can "lot-size" the future requirement of this part. As another example, a cutting machine tool requiring manufactured parts to be grouped together on metal plates (and the grouping of parts may require a long lead time) may function better with an MRP "forward looking" lot-sizing procedure.

The generally accepted guideline of MRP and JIT applications is that JIT is suitable for a repetitive production environment, and MRP is useful for a production system with lumpy demand. In a repetitive production environment, where JIT becomes practical, there may exist some parts having lumpy demand. Such parts with lumpy demand can be better controlled by a forward-looking MRP system, while parts with repetitive demand can be better managed by a JIT pull system.

A kanban pull system (See Monden (1983)) will be assumed to exist in a JIT system throughout this paper. With the coexistence of an MRP and JIT system, the production environment is obviously different from that of a pure MRP system.

This article addresses a modified MRP procedure for a production system with the coexistence of a push-type MRP system and a pull-type kanban system. Relevant research work is first introduced, and a modified MRP approach is then presented. Finally, empirical results from a simulation study are given.

This research study results in the recommendation of a modified MRP system to perform part explosion using a new lot-sizing approach when MRP and kanban systems are both employed in a production system.


It has been reported in a number of studies (Jain (1986), Kanmarkar (1986), and Vaughn (1988)) that MRP and kanban are complementary in strengths and weaknesses and it has been suggested that an integration of MRP and kanban systems can strengthen both systems.

The MRP system is essentially an overall planning tool that can virtually compute material requirements with any kind of demand patterns. A weakness of an MRP system is that it normally results in excessive work-in-process (WIP). This, in contrast, is the strength of a kanban system. The requirement for a kanban system, however, is that the demand needs to be nearly constant. A kanban system does not deal with lumpy demands well because its system parameters are generally set to satisfy a certain given demand rate.

A number of efforts, which will be described later, have been made attempting to integrate MRP and kanban systems. Such efforts were not intended to deal with a system with the coexistence of MRP and JIT. Rather, such integrations of MRP and kanban were developed to use MRP as a overall planning tool being "superimposed" on a kanban system which manages the shop floor so the production system can be enhanced.

Karmarkar (1986) suggested a framework for an integration of MRP and kanban systems in which the MRP's function is to make adjustments for the numbers of kanbans. He suggested that MRP part explosion could be eliminated. Work releases at all stages of production are assumed to be controlled by kanbans in this framework.

Groenevelt and Karmarkar (1988) developed a production control system based on the kanban system. The system adds a "push" element to the kanban approach by adjusting the numbers of cards (kanbans) in the system as a function of changes in the demand. This approach, as they suggested, can be regarded as an instance of a hybrid pull-push system because of the capability in adjusting the numbers of cards according to the demand.

Hall (1986) reported an integrated MRP/kanban system, Syncrho-MRP, developed by the Yamaha Corporation of Japan. In this reported system, MRP orders and kanbans are both used as production triggering signals. The two signals, MRP work orders and kanbans, are synchronized at the shop floor level by requiring both signals be received before an actual production run can be started.

DeLeersnyder et al. (1989) based their methodology on Yamaha's Synchro-MRP but had only a partial set of stations following both the MRP and kanban signals. It is worthwhile to mention that their simulation result indicated that a hybrid system of this kind outperformed a kanban system in the WIP level.

In the above-mentioned literature, only Karmarkar's (1986) framework is indirectly related to the modification of MRP to deal with parts managed by kanbans in a production system. Karmarkar's framework in which part explosion is eliminated suggests a simplified MRP computation which can be viewed as a modified MRP method of dealing with the coexistence of MRP and kanban systems. Such a method, however, did not consider different pull methods and various kanban container sizes possibly existing in a JIT system.


The following terms are first defined: an MRP part is one that is to be produced only if an order is released from an MRP system; a kanban part is one that is to be produced only if a production kanban order is issued.

Imagine a multi-stage production system in which some production stages are run by a pull system and other stages by a push system. The question that can be asked about such a multi-stage system is: how can an MRP system tie together such different production stages? Since a regular MRP approach does not address kanban parts, a modified MRP methodology needs to be developed.

An MRP record for a part begins from gross requirement data, and the projected usage of a part gives rise to the gross requirement data for the part. However, if an MRP part is used by a kanban-part production process (i.e., an MRP part is fed into a kanban part), how would one project the usage of this MRP part? This question will be addressed in this section . It is noted that, in general, an MRP component that is fed into a kanban part requires the gross requirement data in an MRP system based on the order release pattern of the kanban part, but not vice versa.

When a kanban system is in place, it can generally be a "one-bin" or "multi-bin" system. A one-bin system is essentially a reorder-point system in which a reorder quantity is triggered each time. A multi-bin system generally can be triggered if a containerful of a certain part is used up.

To link the MRP and kanban parts in an MRP system, the lot sizes used in the kanban system should first be discussed. When the production of a part is triggered by one or more production kanban cards, the quantity of a multiple of the kanban container sizes is produced. Therefore, a kanban part has a lot size as a multiple of its kanban container size (or bin size).

Another subject which should be discussed is the timing for the production of a lot to be triggered in a kanban system. A pull system has an important characteristic, that is, the production of a part is generally triggered after the part is used or withdrawn. It is, therefore, a "backward looking" production triggering mechanism. This is different from a "forward looking" production triggering of an MRP system in which production is planned based on future requirements.


The modified MRP system (called "method 1" later) that is to be presented here is different from a regular MRP system in the treatment of kanban parts. Since kanban parts are not reordered until parts are withdrawn, accumulation of demand generally determines order releases. In the modified MRP system proposed here, an order release of a kanban part is to be entered in its MRP file whenever the gross requirement accumulates to a container size in a multi-bin kanban system or the projected available quantity reaches the reorder point in a one-bin kanban system. This is because when a container is emptied in a multi-bin system, production to replenish the container is triggered. Also, in a one-bin system, when a reorder point of a bin is reached, production to replenish the bin is triggered.

After this treatment, an order receipt is then projected through a lead-time delay, and the on-hand inventory is updated accordingly. Part explosion is performed in this modified MRP approach based on order releases of part files as in a regular MRP.

Since a kanban part is triggered for production only by a shop floor kanban, the modified MRP only reflects the order status of a kanban part and does not affect the actual production of the kanban part.

This modified MRP can be illustrated by a two-level explosion in Table 1. The first level represents an independent demand. From level 1 to level 2, the factor of usage is 1. Other levels are ignored in this example. To simplify the example, terms such as beginning inventory (begin inv), projected available (pro avail), and scheduled receipt (sch rec) are not listed if they are not needed for computing the order release, as in the first subtable. Only gross requirement (gross req) and planned order release (plan rec) are needed in the first subtable. [TABULAR DATA 1 OMITTED]

In this example, as can be seen in the first subtable, a container size of 8 and a usage of 9 give rise to an order release of 8 units in period 1. This continues until period 5, where the total residue exceeds 8 and the order release is increased to 16. The remainder of the first subtable is calculated in the same manner. The second subtable's order release is based on the projected available quantities. That is, every time the projected available quantity reaches the reorder point of 16, a 32 is entered in the order release row. The arrows in Figure 1 indicate the determining factors by which planned order releases are determined.

It is noted that such a modified MRP determines order releases directly from gross requirements or projected available. In a regular MRP procedure, the net requirement is first calculated and lot-size decisions are made based on net requirements to determine scheduled receipts. Order releases are then determined based on a lead-time offset of scheduled receipts. Part explosion in this modified MRP is performed by carrying order releases into lower-level part gross requirements, as in a regular MRP.

Since an order release of the modified MRP method reflects parts pulled in a kanban system according to the gross demand, the modified MRP should reflect actual demand closely. When an MRP part requires the dependent demand data for making lot-size decisions, MRP gives necessary gross requirement data resulting from part explosion. This modified MRP should provide the necessary linkage for MRP and kanban parts in a production system.


Karmarkar (1986) stated that no part explosion will be necessary in the MRP calculation of his proposed MRP/JIT hybrid system. Karmarkar's method, therefore, implicitly suggests that gross requirements are "dropped down" to the gross requirements rows of its dependent items (this method will be called "method 2" later on).

A leading U.S. tractor manufacturer is dealing with the coexistence of MRP and kanban parts routinely. The factory has both kanban pull and MRP push systems for various parts. In this factory involving thousands of parts, its MRP system uses a lot-for-lot lot sizing to deal with kanban parts. In the meantime, the periodic-order-quantity (POQ) lot-sizing method is used for other regular MRP parts. This method of treating kanban parts in a production system will be referred to as "method 3."

To better describe the two existing methods, two examples are given in the Tables 2 and 3. The examples are based on the same parts used in Table 1. It can be asserted that there is no evidence supporting the effectiveness of any of the three previously mentioned methods. In order to determine the effectiveness of these three different methods, a computer simulation is conducted here. To objective of the simulation study is to assess the effectiveness and robustness of the three methods under a lead-time uncertainty. [TABULAR DATA 2 AND 3 OMITTED]


In order to make a comparison of the three modified MRP approaches, a production model representing a combination of MRP and kanban parts is assumed. This model is given in Figure 1. Relevant data are also included in the figure.

The simulation based on this model assumes that levels 1, 2, and 3 contain kanban parts, and levels 4 and 5 contain MRP parts. A master schedule (plotted in Figure 2) is input into the simulation model for 60 periods. The estimated lead times, and numbers of kanbans and container sizes (or recorder points and recorder levels) for kanban parts are included in Figure 1. It is assumed that level 2 part uses a multi-bin kanban system and the level 3 parts use one-bin systems. (The rationale for this assumption is that the parts in level 3 can be thought of as small component parts and the part in level 2 can be thought of as a subassembly. It is more reasonable to assume a one-bin system for small component parts than for a subassembly.) The lot-sizing methods and estimated lead times for MRP parts are also specified in Figure 1.

Using the GPSS-H simulation language, a factory model is simulated with seven work stations each dealing with an individual part stated in Figure 1. The raw material is ordered at work station 7 and the finished product is made at work station 1.

Independent of the simulation program, the three different modified MRP methods are run using the master schedule as the input data and applying part explosion to carry the master schedule to the lower levels. Since levels 1, 2, and 3 have kanban parts, the three modified MRP methods use different ways of dealing with parts in level 1, 2, and 3. At levels 4 and 5, part 5 and the raw materials are treated by regular lot sizing (POQ is used here). Three sets of MRP work order releases of part 5 and the raw material for the next 60 periods are generated based on the three modified MRP methods. These there different sets of MRP work orders for station 6 and 7 are then input into the simulation program for issuing production of part 5 and the raw material at three separate simulation runs.

The simulation is run for 60 periods in order to see the effectiveness of stations 6 and 7, where MRP lot-sizing is applied. The lot-sizing method used for stations 6 and 7 is the POQ method. In order to better understand the effect of the three methods for treating kanban parts, POQ cycle times of four and eight periods, and cycle times of two and four periods are used in two separate simulation runs, respectively, for each modified MRP approach.

Lead times are assumed to follow a uniform distribution. For a kanban part, a lead time range is assumed to be 25% plus and minus the stated lead time. For an MRP part, a lead time range is assumed to be 25% to 0% below the stated lead time.

The simulation outcome is evaluated based on the inventory of shortages at stations 6 and 7. Conceivably, lower inventory and shortages represent a better production system. Although a shortage at the lower (i.e., stations 6 and 7) stages does not necessarily affect the final assembly schedule, with succeeding stations controlled by kanbans this shortage is more likely to result in shortages at the succeeding assembly stage.

The following figures give the simulation results. These figures are intended for comparing the overall inventory levels and shortage occurrences. Shortages were recorded whenever a retrieval attempt is made by a succeeding station and this retrieving demand cannot be met.

With POQ cycle times of four and eight periods, at stations 6 and 7 respectively, the weighted total inventory levels at stations 6 and 7 are plotted in Figure 3 for all the three modified MRP methods. The weight factors for averaging inventory and shortage are one third at station 6 and one fourth at station 7. The weighted shortage levels at stations 6 and 7 are also plotted for method 3. Methods 1 and 2 do not have any shortages. In Figure 3, it can be seen that the shortage performance of method 3 turns out to be the worst. Method 3 has the highest overall inventory in Figure 2. The overall inventory levels for methods 1 and 2 are very close, as can be seen in Figure 3.

Figure 4 contains the weighted total inventory and shortage levels for POQ cycle times of two and four periods at stations 6 and 7, respectively. This figure showed that method 3 is the most ineffective one under both the inventory and shortage criteria. The weighted total inventory level of method 1 is slightly less than that of method 2.

The overall average inventory levels of method 1 and 2 in Figure 3 represent only 2.42 and 2.65 periods of demands, respectively. Also, there are no shortages for either methods 1 or 2. It is reasonable to say that methods 1 and 2 performed effectively.

Although shortages and inventory are reported above only for stations 6 and 7, it is conceivable that shortages at stations 6 and 7 should result in shortages at the succeeding stations (stations 1 through 5). The simulation result indicated that methods 1 and 2 had no shortage throughout the 60 periods at stations 1 through 5, but method 3 had shortages at stations 1 to 5 quite frequently in the simulation study. Inventory levels are not reported for stations 1 to 5 because inventory is capped by the numbers of kanban cards or reorder quantities at these stations.

Timing is critical to production management. In this simulation study methods 1 and 2 are two methods representing efforts to reflect the timing of kanban orders, but this is not the case for method 3. Method 3 is a conventional MRP lot-sizing method which represents a "forward-looking approach and there is no direct attempt in this method to reflect the timing of kanban orders. The simulation results showed that methods 1 and 2 performed quite well in inventory and shortages, but method 3 performed quite poorly, particularly in shortages.


The question of how an MRP system can deal with the coexistence of MRP and kanban ordered parts is a continuing one and needs to be addressed. This paper examined three modified MRP methods. These modified MRP methods include two that have been developed by others previously (one by Karmarkar and the other by a major tractor manufacturer), and the modified MRP method that is proposed as an alternative in this paper.

These three methods were all analyzed by a GPSS simulation study to compare their effectivenesses in the area of inventory and shortages. This analysis suggests that the proposed modified MRP methods and Karmarkar's method were effective in inventory and shortages. The third method had occasional shortages and higher inventory peaks that limited its effectiveness.

The modified MRP and Karmarkar's method both show that there are effective ways to allow the MRP and kanban systems of ordering to coexist. This coexistence can then further allow these two production systems to complement rather than compete with each other.


DeLeersnyder, J.-L., T.J. Hodgson, R.E. King, P.J. O'Grady, and A. Savva. "Integrating Kanban Type Pull Systems and MRP Type Push System: Insights from a Markovian Model." Forthcoming in IIE Transactions. Groenevelt, H., and U.S. Karmarkar. "A Dynamic Kanban System Case Study." Production and Inventory Management Journal, vol. 29, no. 2, 2nd quarter, 1988, 46-51. Hall, R.W. "Synchro MRP: Combining Kanban and MRP, the Yamaha PYMAC System--Production Planning and Control in Japan." Driving the Productivity Machine: Production Planning and Control in Japan. Falls Church, VA: American Production and Inventory Control Society, 1986, 43-56 Jain, A.K. "Can MRP II and JIT Coexist?" Proceedings of 1986 International Industrial Engineering Conference, Dallas, TX: Institute of Industrial Engineers, 1986. Karmarkar, U.S. "Integrating MRP with Kanban Pull Systems." Working Paper no. QM8615, University of Rochester, Rochester, NY, 1986. Monden, Y. Toyota Production System. Norcross, GA: Industrial Engineering and Management Press, 1983. Vaughn, O. "Are MRP and JIT Compatible?" The Journal of Applied Manufacturing Systems, vol. 1, no. 1, Fall 1988, 17-22.

Fong-Yuen Ding Ming-Nang Yuen North Dakota State University, Fargo, ND 58105
COPYRIGHT 1991 American Production and Inventory Control Society Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1991 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Title Annotation:material requirements planning
Author:Fong-Yuen Ding; Ming-Nang Yuen
Publication:Journal of Operations Management
Date:Apr 1, 1991
Previous Article:The capacitated dynamic lot-sizing problem with startup and reservation costs: a forward algorithm solution.
Next Article:Measurement of manufacturing flexibility: a value based approach.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters