, 
, - a School of Industrial Engineering and Statistics, Universidad del Valle, Cali, Colombia
- b Industrial Management, School of Engineering, University of Seville, Spain
Abstract
Some authors have developed important advances in order to integrate and improve the performance of the supply chain (SC), by sharing information through communication technologies with a decentralized approach. An alternative to improve the performance in an SC may be to introduce a CONWIP supply chain (CONWIP SC) policy with a centralized approach in the supply chain management. In this paper we review the literature to present the benefits of the CONWIP system in different production environments and discuss the possible utilization of this system to manage the entire supply chain. A characterization of the CONWIP-based approach to supply chain management is presented, and a simulation model to explore and evaluate the advantages of this strategy, in comparison with a fully integrated supply chain (FI SC) is developed. We carry on this analysis for different levels of demand volatility and material flow constraints along the chain, and we optimize parameters to ensure the best performance of each control supply chain policy in each scenario. Our research shows that CONWIP SC policy may offer advantages and improved performances in global metrics while both policies offer similar service levels. Main advantages are smaller average of orders placed, less impact of demand variability on the ordering policy, shorter average finished goods inventory and work in process levels and potential inventory cost, substantial comparative funds savings to run the business model, and easier control of inventories.
Keywords
- Conwip;
- Supply chain management;
- Centralized control;
- Simulation;
- System dynamics
1. Introduction
CONWIP supply chain (CONWIP SC), is an approach by which we attempt to improve the supply chain (SC) performance, through an extension of the closed production control system CONWIP–CONstant work in process (Spearman et al., 1990).
CONWIP is a “long-pull” production technique, generalized from a kanban system, which focuses on controlling WIP by controlling order releases to the shop floor (similarly to the old input output control techniques). In CONWIP systems cards are assigned to the whole production line (Spearman et al., 1990). When beginning the production, all available cards are located at the beginning of the line (on a bulletin board). When orders arrive, and there are enough available cards in the system, the necessary cards are attached to the order, and together they proceed through the production line. When the order is processed completely in the line, and leaves the final station, the card is dropped off and released back to the beginning of the line. No order can enter the line without its corresponding card, i.e. if orders arrive and no free card exists, the orders accumulate as backorders, from where they will be discharged as cards are released. Intermediate buffers are established between two consecutive stations, driven by a first-come first served (FCFS) discipline. The detailed flow control mechanism of CONWIP is extensively discussed by Hopp and Spearman (1996).
This is the mechanism by means of which the work-in-progress (WIP) stays constant (whenever the demand is above the capacity). To design a pattern of the CONWIP system, two fundamental questions should be analyzed, the administration of the backlog and the computing of the number of cards. Keeping in mind the time when information about demand at the final buffer is forwarded and the path the information takes, Fig. 1shows the operative difference between pull and CONWIP control systems.
- Fig. 1. Pull and CONWIP system diagrams.
CONWIP SC is defined in this paper as a production–distribution system, in which the production line of each firm has a similarity to a “work center” being a part of a “global line” of supply. The set of cards mentioned in the description of CONWIP system, extends now to a virtual center of control that governs the SC and manages the parts flow and the inventories along the chain. When orders arrive at the final node, the production orders and required materials are released to the first node considering its production capacity constraints. There is a unique and centralized control of the backorders of the SC. Thus, the centralized information control through Internet type of tools, is critical in this context.
In the following section, we first review the main contributions relating to supply chain integration issues. InSection 3 we present a review of the CONWIP system and then discuss the CONWIP SC. In Section 4, we characterize a CONWIP SC model and compare it with a model for a FI SC. In Section 5, CONWIP SC simulation model outputs are discussed to validate the main variables behavior. In Section 6, certain metrics are selected and their results are obtained and discussed to evaluate and compare CONWIP SC vs. FISC policy. Finally we present conclusion in Section 7.
2. Relevant aspects regarding supply chain integration
The issues involved in supply chain integration have been studied in the literature from various perspectives. The benefits of the integration of information flows in a supply chain has been analyzed for a capacitated two-echelon SC by Gavirneni et al. (1999). In Chen et al. (1998) the authors have studied the importance of having access to accurate demand information for the SC upstream members. The benefits of integrating the SC and diminishing the demand oscillation transmission along the chain (bullwhip effect) has been explored by Wikner et al. (1991), Towill et al. (1992) and Chen et al. (1999). There is a wide consensus that information systems integration is essential (Cooper and Ellram, 1993; Houlihan, 1985; Stevens, 1989; Cooper et al., 1997; Bowersox, 1972). In the same sense, authors agree that the integration among partners in a supply chain is feasible thanks to new information and communication technologies. Nowadays, Internet-based collaboration tools can certainly contribute to the achievement of high collaboration and synchronization levels between SC partners (Mentzer, 2001). Tools enable information sharing (including the information needed for a faster product design and the information required to track the materials flow along the chain), collaboration for a common forecast, collaboration for a common planning, and automated financial transactions.
Crespo et al. (2001) studied SC potential improvements by using Internet collaboration tools, and presented a model showing that the collaborative demand forecasting and the visibility of the inventories along the chain may allow much better levels of SC effectiveness (throughput), and efficiency (service level), counteracting significantly, at the same time, the bullwhip effect along the chain. This model certainly focuses the integration between nodes which perform, in effect, in decentralized form, sharing relevant information in order to facilitate the coordinated production planning on the whole, and gain global performance.
3. CONWIP in a production system versus CONWIP in a SC
In the following paragraphs we present a review about the most important studies carried out up to now showing the advantages of CONWIP control systems over other control systems.
The literature review shows how CONWIP system has some superior characteristics over the other pull systems. Among others, we highlight the following:
- •
- It is simpler in the sense that only a single card count setting is required instead of a card count for each workstation (Hopp and Spearman, 1996).
- •
- It can accommodate a changing part mix, due to its use of line-specific cards and a work backlog (Hopp and Spearman, 1996).
- •
- It can accommodate a floating (mix-dependent) bottleneck, due to the natural tendency of WIP to accumulate in front of the slowest machine (Hopp and Spearman, 1996).
- •
- It introduces less operator stress due to a more flexible pacing protocol (Hopp and Spearman, 1996).
- •
- In a flow line that produces a single part type, Spearman and Zazanis (1992) showed that CONWIP produces a higher mean throughput than kanban. In the same scenario, Muckstadt and Tayur (1995a) and Muckstadt and Tayur (1995b)) showed that CONWIP produces a less variable throughput and a lower maximal inventory than kanban.
- •
- Although some researches have shown that a kanban system would have lower WIP levels than a CONWIP system with the same throughput under some certain conditions (Gstettner and Kuhn, 1996), most researches have pointed out that CONWIP would result in lower WIP levels than kanban system with the same throughput in most cases (Spearman et al., 1990; Spearman and Zazanis, 1992). In other words, CONWIP system yield larger throughput than kanban system for the same number of containers (maximal inventory) (Tayur (1992) and Tayur (1993)), even for systems with yield losses (Muckstadt and Tayur, 1995a). This is due to the existing global control (of orders, backorders, and inventory) in a CONWIP system.
- •
- It copes with flow shop operations with large set-up times and permit a large product mix (Spearman and Zazanis, 1992).
- •
- According to the simulation study carried out by Roderick et al. (1992), CONWIP system is superior to other production control systems with respect to due dates and cycle times.
- •
- Bonvik and Gershwin (1996), found that a simple CONWIP control policy outperforms kanban with respect to average inventory levels, when subject to the same requirements on throughput and service level (fill rate). They also concluded that, when the system operates close to capacity, the hybrid control combining CONWIP and kanban improves the inventory levels further.
- •
- Huang et al. (1998) found that the CONWIP production control system is very efficient for the production and inventories control of semi-continuous manufacturing. According to them, it can greatly reduce the WIP, decrease the average inventory and average inventory costs, and guarantee a higher throughput rate and facility utilization.
In comparison with push systems:
- •
- Herer and Masin (1997), built a model in which they highlighted the main advantages of CONWIP over MRP systems. They affirm that the difference between the MRP and CONWIP production control systems lies in the way inventory is handled, because in MRP-ruled manufacturing systems, the amount of inventory managed in the system is theoretically unlimited (see Wight, 1981, p. 34). This difference results in long lead times of MRP systems, poor service levels, and large WIP and finished good inventories (Chase and Aquilano, 1985).
- •
- Huang et al. (1998) assure that (a) CONWIP is superior to push systems when the production system runs under the highest possible throughput rates, and (b) CONWIP also appears to alleviate a problem found in many push systems called “overtime vicious circle”.1Hopp and Spearman (1996) add: (a) the WIP level is directly observable, while the release rate in a push system must be set with respect to (unobservable) capacity, (b) it requires less WIP on average to attain the same throughput, (c) it is more robust to errors in control parameters and (d) it facilities working ahead of a schedule when favorable circumstances permit it.
CONWIP, however, may have its own disadvantages:
- •
- CONWIP does not always generate the smallest total number of trips between stages (Yang, 2000).
- •
- CONWIP may also require a larger storage space between alternate stages than kanban type flow lines, because all full containers (with attached withdrawal cards) may accumulate between any pair of alternate stages (Yang, 2000).
- •
- Graves et al. (1995), assures a serious drawback of CONWIP is that it does not consider the impact that a bottleneck work center may have on the performance of a manufacturing line.
If we concentrate now in supply chain management (SCM) policies, according to the different level of SC integration, the SC uses several push and pull elements in the movement of parts along it (Crespo Marquez et al., 2001). For instance, collaborative forecasting may introduce a push effect in the early stages of the SC, while replenishments pulses could be, at the same time, moving materials among partners, producing pull-type local effects. Since there were found relevant advantages of CONWIP versus other pull and push production systems, we will try to enlarge the underlying philosophy in the CONWIP production environment searching to materialize some of those advantages in a SCM context. For instance, could we expect to maintain less WIP levels for the same fill rate of the SC when using CONWIP SC? In which scenarios is that possible?
Compared with the FI SC (extending our literature review about CONWIP systems), we conclude that the following advantages could be capitalized by introducing CONWIP elements into SCM. The intermediate nodes would:
- •
- not need to establish ordering policies, thus we may avoid the amplification of the variance of the demand signal along the chain, only a central entity would authorize the release of work on the basis of the system status, defined by the total number of cards “attached” to orders;
- •
- not need to establish policies to fix safety inventories, reducing holding and ordering costs and inventory cash requirements, even more, using less space;
- •
- not need to control backlog, due to centralized supply chain management;
- •
- make simpler operative decisions (under normal conditions, in a centralized supply chain, firms would simply process and send the whole received parts);
- •
- make easier control of materials flow and WIP, since parts are transferred between partners by means of a push effect;
- •
- not need to carry out individual forecasts, nor do they need to know the levels of inventories of the other nodes; and
- •
- be in a position to promote teamwork, the development of products and processes, and to realize prominent cooperative agreements in order to achieve virtual financial integration.
However, in the light of existing management paradigms, managers could see disadvantages and prefix obstacles to implementing a centralized control policy (with constant work in process in the SC), for example:
- •
- In a general sense, the SC is constituted by different firms with different goals, probably in conflict.
- •
- Information about demand would not be shared between nodes, but centralized, so that the firms would not manage individual forecasts, and ordering policy would be global, diminishing the internal control in the firms.
- •
- Limiting the total number of parts allowed into the SC at the same time, may prevent a particular firm use its full capacity.
- •
- Centralized management may impose conditions on the financial policies, as well as the productive processes and policies of the firms. Furthermore, difficulties (e.g. bottleneck, quality defects, cost overrun, etc.) in one firm, easily may affect all the chain.
- •
- Not all firms are willing to share financial and operative risks.
- •
- CONWIP could not be well suited for the case of complex SC networks.
4. Modeling a CONWIP SC vs. a FI SC
Before proceeding with the model development and discussion, we first describe the notations and definition of the main variables for CONWIP SC model
In order to compare performance of the two SCM alternatives, we will model both systems according to the following procedure:
- (1)
- Formalization of the FI SC model according to Crespo et al. (2001), in which it is assumed that (a) partners make individual ordering decisions, implying that a decentralized control exists in the chain, (b) the orders in the supply chain are visible in real time, and processed accordingly and (c) partners gain access to additional information that they do not control, and use it in their planning process (e.g. forecasts of final nodes, on-hand inventories and pipelines of the other supply chain members, backlog, etc.).
- (2)
- Comparison of the equations for the FI SC model to the ones for the CONWIP SC model, indicating those which are different to their homologous in the CONWIP SC model (see Table 1).
- Table 1. Model equations in FI SC
-
- a
- Equation for this variables differs in CONWIP SC model (see equations in following paragraphs).
- Full-size table
- (3)
- Equations of the CONWIP SC model will be presented.
- (4)
- Behavior patterns for some CONWIP SC model variables will be provided for their validation.
- (5)
- The simulation study will be carried out, to compare performance of both systems for a series of designed scenarios. We will make sure that for each particular scenario in the study, both systems are operating in their best-possible conditions (i.e. with the best-possible value for their parameters, measuring criteria are presented in Section 6). This will be done via a multi-parametric optimization for each particular case.
In Fig. 2, the connection between the main CONWIP SC information flow variables (thin lines) and the material flow variables (thick lines) can be clearly appreciated.
- Fig. 2. Basic influence diagram of CONWIP SC for the variables in node [i] and node [n].
Now we will explain how the equations are obtained for the CONWIP SC model.
4.1. Incoming orders
The incoming orders variable (Dt) takes the generated values for the orders placed by the final customers (OPtcust), this is
(1)
4.2. Shipments
It is assumed that the incoming orders to the final node (Dt), are immediately shipped to final customers as they are received. Backlog are considered to be lost sales. Therefore, the desired shipments DSt from the last node are equal to the orders received in each period, as follows:
(2)
Likewise, the desired production orders to release to first node of the supply chain, becomes equivalent to incoming orders:
(3)
The shipments from the node i in week t (Sti), depends on the available total FGI (TYti) and the maximal load of parts to be processed in the node i+1 during its cycle time (MLPi+1/Li+1). When delivering finished goods, inventory constraints may appear in the node n, reducing the amount of units shipped to customers, Stn (seeFig. 2). The shipment rate to final customers from final node, is the minimum among the available total FGI TYtn and the desired shipments to final customers DSt. The expression for the supply flow is as follows:
(4)
For all nodes, the available total FGI is calculated as follows:
(5)
4.3. Inventory, materials flow and WIP
The FGI Yti in the buffer of the node i, diminishes according to the units transferred to the next node i+1 or to final customers, and increases according to the rate of processed parts that arrive (initial conditions are assumed known):
(6)
(7)
(8a)
When assuming unlimited supply of parts to the first node and without variability in the market of external suppliers to the supply chain, Iti is equivalent to order placed by the first node OPt, calculated by the control central system of the CS, this is
(8b)
Therefore, the WIP in the pipeline is defined (assuming its initial conditions are also known) as
(9)
4.4. Backlog, procurement orders and available production cards
When the inventory constraints limit the supply of finished goods from the last node of the SC to final customers, a portion of the orders cannot be fulfilled. These orders will stay as backlog Bt (in the central control of SC), and they are considered lost sales. Contrary to FI SC, in CONWIP SC backlog is generated only in final node, from incoming orders Dt and finally served orders DOt, in the following way (supposing initial conditions are known):
(10)
(11)
When there are not enough available production cards to release production orders, the flow of orders to the entire supply chain is limited, then, a portion of the final customer orders cannot be released to production. Those orders will stay as production backorders PBt (production orders not released on time), and they are computed as following equation:
(12)
(13)
The available production cards APCt are calculated as follows:
(14)
The total number of cards (TNPC), is calculated to maximize the throughput while the total inventories are minimized in the entire chain (see Section 6). This is done by using direct search techniques applied to each particular scenario.
4.5. Financial flows and the working capital
Financial flow variables are estimated extending part of the one-stage model presented in Lineys (1980) to a multiple-stage model of the supply chain. We now present the relationship between the main financial variables (related to cash requirements) for each node, using the following notation:
For the purpose of this paper, the main variables relationships are as follows:
(15)
(16)
(17)
(18)
(19)
5. Validation of the behavior patterns for the main CONWIP SC model variables
In order to validate the behavior patterns of the main variables of CONWIP SC policy model, the following example similar to the beer game (Sterman, 1984) was built, this time with variable demand. The SC consists of four nodes (Fig. 3) in series: a supplier, a first (1st) manufacturer, a second (2nd) manufacturer, and the distribution and retail channels. The behavior pattern results are mainly presented for the global SC in order to study the impact of the centralized control and limited total inventory levels on performance measurers, while we treat the SC like a single cell. Customer demand is generated as a number from a normal distribution with mean assumed to be 4 units/week and standard deviation (SD) 4 units. Unit per container (UCi) value is 1 and selected smoothing factor (α) is 0.2. The total number of production cards is 68. Initial conditions for WIP and FGI in firms are:
- Fig. 3. Sample SC selected for variable behavior validation.
The values adopted for the other operative parameters are included in Fig. 3.
Fig. 4 shows the behavior of the orders placed by the factory and shipment to end-customers (throughput). In the scenario selected for this validation, the maximum load constraint does not limit the ordering policy, however the available production cards (Fig. 5) could do it (e.g. weeks 8th. and 11th.). Anyway, the ordering policy allows releasing enough part amounts to supplier once they move to the FGI buffer of retailer on time for serving demand.
- Fig. 4. Input and throughput rates.
- Fig. 5. Production cards and total inventories.
Fig. 4 and Fig. 5 also shows how, (a) during the first three weeks it is not necessary to place orders due to established initial inventory amounts, (b) the condition TNPC=APC+ total (WIP+FGI) is completed, and (c) starting from the moment in which WIP+FGI in the chain reaches the steady state, as new orders arrive, sufficient number of cards is always available for releasing the necessary orders, in order to produce and meet all demand.
Fig. 6 shows the behavior of WIP+FGI per node. Inventories move between the firms in sequential form along the chain during the 52 weeks. This is a consequence of the existent push effect between firms of the SC (CONWIP systems characteristic) and the great parts load capacity settled in this example for all firms.
- Fig. 6. Total WIP+FGI per node.
According to previous results obtained with this supply chain model, we have observed how it reproduces the main characteristics of a CONWIP control policy (according to (Spearman et al., 1990)).2
6. Simulation and comparison of SCM policies
In this section we present a series of results comparing CONWIP SC with FI SC, and for a set of performance metrics. These results are presented for both SCM policies and for the entire supply chain, consisting of four nodes (factory, distributor, wholesaler and retailer) as described by Sterman (1989), in the article for the beer game. We remind the reader the assumption made in this paper is in the following sense: the retailer does not hold any backlog (due to the fact that end-customers will not wait in any scenario).
Keeping in mind that this study is carried out for a supply chain in which the final node serves end-customers, we set enough initial amount of WIP and FGI in each firm in order to ensure that the SC does not lose the initial sales, even for those scenarios where the production capacity constraint (MLPi/Li) may limit the ordering policies in some periods.
The simulation runs are for 52 weeks. In Table 2, we present the four demand situations considered in the study, and the initial values for WIP and FGI in all nodes.
- Table 2. Specified parameters for final demand and initial inventory conditions
Additionally, each of these situations is simulated varying the maximal load of parts to be processed in the firms (in all simulations we assume that we have the same value MLP for all firms), from 30 to 10 units. In summary, for each combination of demand standard deviation (SD) and maximal load of parts (MLP), we will present the results of each SCM policy and for different performance metrics (see Fig. 7).
- Fig. 7. Dimensions of the simulation study carried out for each SCM policy.
A key issue in this study is that FI SC and CONWIP SC are compared under their optimal operating conditions for each particular scenario. These operating conditions are based on the achievement of two main goals settled in this paper: maximizing throughput and minimizing total inventories in the chain under production capacity constraints (determined by the allowed maximal load of parts to be processed in a cycle time in firms). To search the parameters values for each policy producing its best performance in each scenario, we have used a direct-search numerical optimization technique which does not need to evaluate the gradient, and which is very suitable for the analysis of dynamics of complex nonlinear control systems. This technique is the modified Powell method (Powell, 1964), which is well known among direct-search methods, to obtain a very fast convergence.3
Table 3 shows the equivalent variables to optimize and the parameters to search for each policy optimization, and also shows the constraints of the system, equal for both cases. We have assigned weights to balance the numerical sizes of the variables to optimize, with the criteria of ensuring the same and high fill rates for both policies in the different scenarios. Later, the other operative and financial performance metrics will be compared for each case. In the next paragraph we explain and justify the selection of the optimization criteria.
- Table 3. Optimization criteria and CSM policy
Throughput is defined as shipment to end-customers. Desired production orders and WIP+FGI are the constituents of the base (target) stock in each node in case of FI SC policy. Similarly, available production cards and total WIP+FGI are the constituents of the total number of production cards in the chain for the CONWIP SC policy. In this way, for FI SC policy, we make sure that we minimize the base stock, and for CONWIP SC policy, we ensure that we minimize the total number of production cards, which are the main variables in the underlying management philosophy of each policy. Table 4 presents the values for parameters considered constant in all scenarios. The values obtained for search parameters are shown inFig. 8 and Table 5.
- Table 4. Values for constant parameters in all simulations
- Fig. 8. Computed values for TNPC.
- Table 5. Computed values for SS, βS and βSL
Regarding CONWIP SC policy, notice that for common small values of the maximal load of parts to firms, the trend of the optimal total number of production cards is increasing. The reasons why model computes high values for TNPC are:
- 1.
- The initial conditions value of WIPs and FGIs in firms, impact on computing the total number of production cards.
- 2.
- In some periods, the ordered quantities may be shorter than the end-customer demand because the production capacity constraint of the factory delimits the ordering policy. When this happens, the chain will be able to serve demand only if there is enough accumulated inventory.
In consequence, in this scenarios the trend of the total WIP+FGI will be decreasing with time, and the trend of the available production card will be increasing.
In the same sense, for the smaller values of demand SD, TNPC does not take the smallest values since the chain needs to accumulate enough inventories, specially when the available production cards delimits the ordering policy. In these cases, the maximal production capacity of the factory does not prevent holding high inventories (included FGI), avoiding lost sales.
For FI SC policy, notice that the general trend in values obtained for SS is generally the same than for TNPC, with similar justification.
The following set of graphs shows the global comparative results for order, material and financial flow variables, and some measures of the chain financial performance.
Fig. 9 shows how SCM policies obtain similar high results for service level (defined as fill rate, i.e. fraction of final customer incoming orders that is served by the SC) as it was expected.
- Fig. 9. Results for the service level.
Results in Fig. 10 show that as allowed maximal load of parts to factory decreases, the average orders placed by factory diminishes in both policies. Once again, this is due to the subsequent limits imposed by the maximal production capacity on the respective ordering policies. However as demand SD decreases, the behavior of the orders placed is analogous to the one of the total number of production cards and the safety stock. On the other hand, results indicate that, in all scenarios the CONWIP SC policy need to place smaller average orders than the FI SC policy while they offer similar service levels, but this difference between them is bigger in scenarios with relative low maximal load and low demand SD. This result is because CONWIP SC policy orders exclusively on the strength of the forecast while FI SC attempt to order to maintain a base stock (bigger than the forecast). This is a key issue for this result.
- Fig. 10. Results for the average orders placed by the factory.
Fig. 11 illustrates how when the allowed maximal load of parts to the factory decreases, the standard deviation of orders placed by factory diminishes in both policies. In the same sense, the smaller demand SD does not imply smaller values of orders placed variability. Notice that for the small values of the maximal load, FI SC presents smaller variability than CONWIP SC policy. Contrary to what happens in the CONWIP SC policy, the procurement/production rate constraints many times delimit the quantities ordered by factory in the FI SC policy. This fact controls variability of the orders placed too. Anyway, CONWIP SC policy is less impacted overall by variability in demand.
- Fig. 11. Results for deviation standard of orders placed by factory.
The possible bullwhip effect is substantially mitigated in CONWIP SC policy. Reasons for that are :
- (a)
- centralized forecast and inventory management, avoiding a node to depend on another node,
- (b)
- straight line of node's information to central control.
In summary, both policies order whenever end-customers remove goods from the FGI of retailer, but parts release by CONWIP SC ordering policy are better (exactly) adjusted to the resulting mean (4.2 units in Fig. 10) of demand when capacity and/or cards limitations appear.
Benefits achieved with the CONWIP SC policy can be clearly appreciated (Fig. 12) in terms of less funds required to run that business model, specially during demand SD decreases, which represent substantial comparative financial savings. This is the consequence of the smallest inventory cash requirements need, once the sales outstanding periods are all assumed to be equal (2 weeks).
- Fig. 12. Results for normalized cash requirements.
Fig. 13 shows one of the most significant differences between the two SCM policies. CONWIP SC policy reaches less average global inventory levels in all scenarios. This result is the evidence of an expected advantage of CONWIP SC policy over FI SC policy, this is, largest inventory efficiency giving similar service levels.
- Fig. 13. Results for average global FGI+WIP.
Finally, with regard to cost of administration of inventories (holding + stock out penalties), CONWIP SC policy is more robust to incur in less costs in all situations (Fig. 14). These results demonstrate more uniform materials flow, easier inventory control and better synchronization between firms.
- Fig. 14. Results for average inventory cost in the chain.
7. Conclusions
In this paper we evaluated the performance of the CONWIP control policy managing a supply chain in a variable environment, and compared that performance to the fully integrated SC one. Based upon the obtained results, it appears that CONWIP SC policy represents a centralized control along the chain that may offer advantages in performance, compared to the decentralized FI SC. A CONWIP SC policy provides the following advantages over FI SC policy, when they offer similar service levels:
- 1.
- Easier control on WIP, since flow materials and parts is centrally controlled and limited.
- 2.
- CONWIP SC policy need to place smaller average orders than FI SC specially in scenarios with relative low maximal load and low demand SD.
- 3.
- Generally, the ordering policy in FI SC policy is more vulnerable as variability demand appear and load capacity in nodes is not very small. In CONWIP SC policy, centralized manage of the demand and inventories mitigate the amplification of demand, allowing that the ordered quantities are adjusted to real flow material needs.
- 4.
- By CONWIP SC policy the supply chain obtain substantial comparative financial saves regard to cash requirements, specially during demand SD decreases.
- 5.
- CONWIP SC policy offers potential shorter average FGI+WIP levels (larger efficiency) leading the supply chain to manage shorter average inventory costs.
In summary, although we may find some resistance to change in the firms acting as partners of the SC (see comments in Section 3), exploring the utilization of CONWIP for SCM purposes could be a source of potential benefits in the near future.
Acknowledgements
This research has been funded by the Spanish Ministry of Science and Technology, Project DPI 2001-3110, besides FEDER funds.
References
- Spearman et al. (1990)
- CONWIPa pull alternative to kanban
- International Journal of Production Research, 28 (1990), pp. 879–894
- |
- Hopp and Spearman (1996)
- Factory PhysicsIrwin, Chicago, USA (1996)
- Gavirneni et al. (1999)
- Value of information in capacitated supply chains
- Management Science, 45 (1) (1999), pp. 16–24
- |
- Chen et al. (1998)
- Chen, F., Drezner, Z., Ryan, J.K., Simchy-Levy, D., 1998. The Bullwhip effect: managerial insights on the impact of forecasting and information on variability in a supply chain. In: Quantitative Models for Supply Chain Management, International Series in Operations Research & Management Science, Vol. 17. Kluwer Academic Publishers, Boston, pp. 417–439, ISBN 0-7923-8344-3.
- Wikner et al. (1991)
- Smoothing supply chain dynamics
- International Journal of Production Economics, 22 (1991), pp. 231–248
- | |
- Towill et al. (1992)
- Industrial dynamics simulation models in the design of supply chains
- International Journal of Physical Distribution and Logistics Management, 22 (1) (1992), pp. 3–13
- Cooper and Ellram (1993)
- Characteristics of supply chain management and the implications for purchasing and logistics strategy
- The International Journal of Logistics Management, 4 (2) (1993), pp. 13–24
- Houlihan (1985)
- International supply chain management
- International Journal of Physical distribution and Materials Management, 15 (1) (1985), pp. 22–38
- Stevens (1989)
- Integrating the supply chain
- International Journal of Physical Distribution and Materials Management, 19 (1) (1989), pp. 3–8
- Cooper et al. (1997)
- Supply chain managementmore than a new name for logistics
- The International Journal of Logistics Management, 8 (1) (1997), pp. 1–14
- Bowersox (1972)
- Planning physical distribution operations with dynamic simulation
- Journal of Marketing, 36 (1) (1972), pp. 17–28
- Mentzer (2001)
- Supply Chain ManagementSage Publications, Beverly Ailb, CA (2001)
- Crespo et al. (2001)
- Benefits of the internet for the supply chain management
- A characterization and simulation study. International Journal of Agile Manufacturing, 4 (2) (2001), pp. 25–42
- Spearman and Zazanis (1992)
- Push and pull production systemsissue and comparison
- Operations Research, 40 (3) (1992), pp. 521–532
- |
- Muckstadt and Tayur (1995a)
- A comparison of alternative kanban control mechanismsI, background and structural results
- IIE Transactions, 27 (1) (1995), pp. 140–150
- Muckstadt and Tayur (1995b)
- A comparison of alternative kanban control mechanismsII, experimental results
- IIE Transactions, 27 (1) (1995), pp. 151–161
- Gstettner and Kuhn (1996)
- Analysis of production systems kanban and conwip
- International Journal of Production Research, 34 (1996), pp. 3253–3273
- Tayur (1992)
- Properties of serial kanban systems
- Queuing Systems, 12 (1992), pp. 297–318
- Tayur (1993)
- Structural properties and heuristic for kanban controlled serial lines
- Management Science, 39 (1993), pp. 1347–1368
- Roderick et al. (1992)
- A comparison of order release strategies in production control systems
- International Journal of Production Research, 30 (11) (1992), pp. 2559–2572
- Bonvik and Gershwin (1996)
- Beyond KanbanCreating and Analyzing Lean Shop Floor Control PoliciesOperations Research Center, MIT, Cambridge, MA (1996)
- Huang et al. (1998)
- Simulation study of CONWIP for a cold rolling plant
- International Journal of Production Economics, 54 (2) (1998), pp. 257–266
- Herer and Masin (1997)
- Mathematical programming formulation of CONWIP based production lines; and relationships to MRP
- International Journal of Production Research, 35 (1997), pp. 1067–1076
- Wight (1981)
- MRP IIUnlocking America’ Productivity PotentialCBI Publishing Co, Boston, MA (1981)
- Chase and Aquilano (1985)
- Production and Operations Management, 4th EditionIrwin, Home-wood, I (1985)
- Yang (2000)
- Managing a flow line with single-kanban, dual-kanban or CONWIP
- Production and Operations Management, 9 (4) (2000), pp. 349–366
- Graves et al. (1995)
- Literature review of material flow control mechanisms
- Production Planning and Control, 6 (5) (1995), pp. 395–403
- Chen et al. (1999)
- The impact of exponential smoothing forecast on the bullwhip effectWorking Paper, Northwestern University (1999)
- Sanders (1994)
- Forecasting practices en United-States corporations
- Survey results. Interfaces, 24 (2) (1994), pp. 92–100
- Tversky and Kahneman (1974)
- Judgement under uncertainty
- Heuristics and biases. Science, 185 (1974), pp. 1124–1131
- Sterman (1989)
- Modeling managerial behaviormisperceptions of feedback in a dynamic decision making experiment
- Management Science, 35 (3) (1989), pp. 321–339
- Lineys (1980)
- Lineys, J.M., 1980. Corporate Planning and Policy Design: a system dynamics approach, MIT Press/Wright-Allen series in system dynamics. MIT Press, Cambridge, MA.
- Sterman (1984)
- Instructions for running the beer distribution Game (D-3679)Sloan School of Management. MIT, Cambridge, MA (1984)
- Powell (1964)
- An efficient method for finding the minimum of a function of several variables without calculating derivatives
- Computer Journal, 7 (2) (1964), pp. 155–162
- Powell (1968)
- On the calculation of orthogonal vectors
- Computer Journal, 11 (2) (1968), pp. 302–304
PDF (1132 K)
- Corresponding author. Tel.: 34-954-487215; fax: 34-954-486112
- 1
- This concept has to do with real life steady state cycles in a plant, as a consequence of capacity computation, randomness in job arrivals creating the bottleneck process to starve, and later increasing WIP and cycle time, which leads normally to the “one time” authorization of overtime until WIP and Cycle time decreases again (see Hopp and Spearman, 1996, pp. 284–285).
- 2
- The validation of the FI SC model is performed in Crespo et al. (2001).
- 3
- The basic idea behind Powell's method (Powell, 1964) is to break the N dimensional minimization down into N separate 1D minimization problems. Then, for each 1D problem a binary search is implemented to find the local minimum within a given range. Furthermore, on subsequent iterations an estimate is made of the best directions to use for the 1D searches. Some problems, however, are not always assured of optimal solutions because the direction vectors are not always linearly independent. To overcome this, the method was revised (Powell, 1968) by introducing new criteria for formation of linearly independent direction vectors; this revised method is called “The Modified Powell Method”.
