Lot Streaming in Two-Stage Flow Shops and Assembly Systems
The research work presented in this dissertation relates to lot streaming in two-stage flow shops and assembly shops. Lot streaming refers to the process of splitting a production lot into sublots, and then, processing the sublots on different machines simultaneously in an overlapping manner. Such a strategy allows finished material at each stage to be transferred downstream sooner than if production and transfer batches were restricted to be the same size. In the case when each sublot consists of just one item, a single-piece-flow is obtained. Such a continuous flow is a key element of the Toyota Production System. However, single-piece-flow increases the number of transfers and the total transportation cost (time). As a result, it may not be economically justifiable in many cases, and therefore, material may have to be transferred in batches (called transfer batches, or sublots). Lot streaming addresses the problems of determining optimal sublot sizes for use in various machine environments and optimizes different performance measures.Given this relationship between lot streaming and the Toyota Production System, lot streaming can be considered a generalization of lean principles.
In this dissertation, we first provide a comprehensive review of the existing literature related to lot streaming. We show that two-stage flow shop problems have been studied more frequently than other machine environments. In particular, single-lot two-machine flow shops have been very well researched and efficient solution techniques have been discovered for a large variety of problems.
While two-stage flow shop lot streaming problems have been studied extensively, we find that the existing literature assumes that production rates at each stage remain constant. Such an assumption is not valid when processing rates change, for example, due to learning. Learning here, refers to the improvements in processing rates achieved due to experience gained from processing units. We consider the case when the phenomenon of learning affects processing and setup times in a two-stage flow shop processing a single lot, and when, sublot-attached setup times exist. The decrease in unit-processing time, or sublot-attached setup time, is given by Wright's learning curve. We find closed-form expressions or simple search techniques to obtain optimal sublot sizes that minimize the makespan when the effect of learning reduces processing times, sublot-attached setup times, or, both. Then, we provide a general method to transform a large family of scheduling problems related to lot streaming in the presence of learning, to their equivalent counterparts that are not influenced by learning. This transformation is valid for all integrable learning functions (including the Wright's learning curve). As a result, a large variety of new problems involving learning can be solved using existing solution techniques.
We then consider lot streaming in stochastic environments in the context of sourcing material. Such problems have been well studied in the literature related to lot streaming for cost-based objective functions when demand is continuous, and when processing times are deterministic, or, for material sourcing problems when the time required to procure a lot is stochastic but is independent of the lot size. We extend this study to the case when the time required to produce a given quantity of products is stochastic and dependent on the number of units produced. We consider the case when two sublots are used, and also compare the performance of lot streaming to the case when each sublot is sourced from an independent supplier.
Next, we address a new problem related to lot streaming in a two-stage assembly shop, where we minimize a weighted sum of material handling costs and makespan. We consider the case when several suppliers provide material to a single manufacturer, who then assembles units from different suppliers into a single item. We assume deterministic, but not necessarily constant, lead times for each supplier, who may use lot streaming to provide material to the manufacturer. Lead times are defined as the length of the time interval between a supplier beginning to process material and the time when the first sublot is delivered to the manufacturer; Subsequent sublots must be transported early enough so that the manufacturer is not starved of material. The supplier may reduce this lead time by using lot streaming, but at an increased material handling cost. The decrease in lead time is also affected by other factors such as lot attached/detached setup times, transportation times etc. We allow these factors to be different for each supplier, and each lot processed by the same supplier. We refer to this problem as the Assembly Lot Streaming Problem (ALSP). We show that the ALSP can be solved using two steps. The first step consists of solution to several two-stage, single-lot, flow shop, makespan minimization problems. The solution to these problems generate prospective sublot sizes. Solution methods outlined in the existing literature can be used to complete this step. The second step obtains optimal number of sublots and production sequence. For a given production sequence, this step can be executed in polynomial-time; otherwise, the second step problem is NP-hard and integer programming formulations and decomposition-based methodologies are investigated for their solution. We make very limited assumptions regarding the handling cost and the relationship between the supplier lead time and number of sublots used. As a result, our solution methodology has a wide scope.