Lot streaming in a two-stage assembly system and a hybrid flow shop

In this dissertation, we investigate the use of lot streaming in a two-stage assembly system and a two-stage hybrid flow shop in order to improve system performance. Lot streaming accelerates the flow of a production lot through a production process by splitting it into sublots, and then, processing these sublots in an overlapping fashion over the machines, thereby reducing work-in-process and cycle-time. Traditionally, lot streaming has been applied to problems in various flow shop machine configurations. It has also been applied to machine environments of job shop, open shop, and parallel machines. Its application to assembly system is relatively new.

The two-stage assembly system that we consider consists of multiple suppliers at Stage 1 with each supplier producing one type of a subassembly (or a component), and one or more assembly locations at Stage 2, where the subassemblies are then put together. Lot-attached and sublot-attached setup time and cost are encountered on the machines at both the stages, and sublot-attached time and cost are encountered for the transfer of sublots from Stage 1 to Stage 2. Mass customization is an example of such a system in which the final assembly of a product is postponed to capture specific customer demands. Dell Computer constitutes a real-life example of this system. A customer picks his/her computer processor, memory, storage, and other equipment, on Dell's web site. Dell's supply chain is configured to obtain subassemblies from suppliers (stage 1), and then, to assemble the requisite systems in different market areas (stage 2). This enables a reduction in operating cost while improving responsiveness to customers. The problem that we address is as follows: Given a maximum number of sublots of each lot, determine the number of sublots to use (assuming equal sublot sizes), and also, the sequence in which to process the lots, in order to minimize two criteria, namely, makespan, total cost. We propose two column generation-based methods that rely on different decomposition schemes. The results of our computational investigation conducted by using randomly generated data sets reveal that the proposed column generation methods obtain solutions in a few seconds of CPU time while the direct solution by CPLEX of a mixed integer programming model of the problem requires much larger CPU times.

For the hybrid flow shop lot streaming problem, the machine configuration that we consider consists of one machine at Stage 1 and two machines at Stage 2 (designated as 1+2 system). A single lot is to be processed in the system, and the objective is to minimize the makespan. A removal time is associated with each sublot at Stage 1. We present a mixed integer programming model for this problem to determine optimal number of sublots and sublot sizes. First, we consider the case of a given number of sublots for which we develop closed-form expressions to obtain optimal, continuous sublot sizes. Then, we consider determination of optimal number of sublots in addition to their sizes. We develop an upper bound on optimal number of sublots, and use a simple search procedure in conjunction with the closed-form expressions for sublot sizes to obtain an optimal solution. We also consider the problem of determining integer sublot sizes, and propose a heuristic method that directly solves the mixed integer programming model after having fixed values of appropriate variables. The results of our numerical experimentation reveal the efficacy of the proposed method to obtain optimal, continuous sublot sizes, and also, that of the proposed heuristic method to obtain integer sublot sizes, which are within 0.2% of optimal solutions for the testbed of data used, each obtained within a few seconds of CPU time.

The last problem that we address is an extension of the single-lot lot streaming problem for a $1+2$ hybrid flow shop considered above to the case of multiple lots, where each lot contains items of a unique product type. We consider two objectives: minimize makespan, and minimize the sum of the completion times for all the lots. The consideration of multiple lots introduces a complicating issue of sequencing the lots. We use the results derived for the single-lot problem and develop effective heuristic methods for this problem. The results of our computational investigation on the use of different heuristic methods reveal their efficacy in solving this problem.