Analysis of delayed product differentiation under pull type policies

Delayed product differentiation (DPD) increases manufacturers' competitiveness in the market by enabling them to more quickly respond to changes in customers' demands. DPD has also been shown to require less Work-in-Process (WIP) than a non-DPD setup in some cases. Previous research was mainly focused on the level of semi-finished and/or finished good inventory under a base-stock policy. The control of WIP inventory was not considered. DPD may also improve response times under pull inventory control schemes, in which the amount of WIP is controlled directly. These systems can be modeled as closed queueing networks in which a fixed number of kanbans circulate as customers among each set of one or more processing stages.; In this study, we first developed models to analyze the performance of simple kanban and CONstant-WIP (CONWIP) controlled systems and set the number of kanbans to achieve a specified performance level. The models help us better understand the behavior of pull systems. The performance evaluation method uses nonlinear programming (NLP) models to bound the throughput for fixed number of kanbans or minimize the number of kanbans necessary to achieve a specified throughput. The model shows how random supplies and demands prevent equilibrium from occurring in a single-stage kanbans system.; We studied a model for a system of two products with unlimited supply and demand using three CONWIP loops to represent the common processes and the differentiated processes for each product. The same system after DPD has more common processes and fewer differentiated processes. The NLP model can determine numbers of kanbans for each loop to achieve specified throughput targets. Because the throughput bounds are not as tight as desired, we developed a heuristic algorithm that starts from the NLP solution and adjusts the kanbans using simulation to evaluate the performance. A comparison of the result of the heuristic algorithm for the systems with and without DPD indicates that DPD reduces the amount of WIP necessary to achieve a specified throughput. Furthermore, we show how models of systems with similar structure can be generalized.