Setting lead times and due dates in stochastic assembly systems using MRP

This research addresses production planning issues for electronic assembly systems using MRP with uncertain supply processes. We make use of information available in MRP to develop more accurate transient models of the supply process than the steady state models typically used in the current research literature. The first problem we research is a practical method for setting safety lead times for purchased components in assembly systems. We develop a two-stage production model in which suppliers are uncapacitated with stochastic lead times. Then we develop a combinatorial optimization method that takes advantage of structural properties to produce an optimal integer solution of safety lead times. Our experimental results reveal some interesting policy issues including that there are conditions when safety lead time is useful even with perfect suppliers.; We also propose a new model for quoting due dates in a make-to-order environment where customers request due dates. The model incorporates inventory costs, fill rate issues, and service level issues. In particular, we consider order delay costs that measure the intangible cost of quoting due dates greater than requested. We utilize a two-stage production model that assumes that production is constrained primarily by an uncertain procurement process. This simplification results in a newsboy-like formulation that enables us to obtain a simple approximately optimal due-date setting policy that is well-suited to MRP environments. The model also yields several interesting policy conclusions such as a result that shows our policy often yields shorter lead times than an analogous lead time minimization formulation of the problem.; Finally, we develop a methodology for modeling supplier lead time and lateness distributions for purchase orders in an MRP system. Our approach relies on data readily available in MRP. Moreover, because historical data is generally limited due to short product life cycles, we make use of linear regression models that predict purchase order earliness and tardiness based on order attributes. We empirically compare our lead time modeling method to two intuitive methods and demonstrate that our method is more accurate. In a simulation test, we show that improved lead time model accuracy yields improved service level and inventory performance.