Characteristics of optimal buffer allocation in unbalanced serial production lines

Graduation date: 1999For many decades, production and assembly lines have played an important role in industrial manufacturing systems. In particular, they have proved to be an efficient way to organize the production of high volume products in many industries. In their effort to improve the efficiency of production lines, researchers encountered two major design problems: The first problem, which is the optimal allocation of work to the different stations of a production line, can be regarded as essentially solved today. The second problem became known as the buffer allocation problem and addresses the issue of how a limited amount of storage capacity for work-in-process should be allocated between the stations of a serial production line in order to minimize blocking and starving phenomena. Until today, very little is known about the buffer allocation problem and its solution for production lines with an unbalanced allocation of work. This thesis determines properties and characteristics of the optimal buffer allocation for this line type with the aim of enhancing the existing knowledge about the problem and making a contribution toward its solution. We examine the validity of several hypotheses regarding optimal buffer allocation which have been conjectured by recent studies and discuss the impact of certain factors on system performance and optimal buffer allocation. Some of the most important among these factors are multiple bottlenecks, line symmetry and line length as well as the different types of imbalances. For the generation of the optimal buffer allocations, a simulation model is combined with an exhaustive search procedure, which is substituted in the second part of the study by an adaptation of the Genetic Algorithm. Besides the characterization of the optimal buffer allocation, a secondary objective of this study is to evaluate the performance of the utilized search heuristic with regard to its practical applicability. For this purpose several designed experiments are performed, which indicate that the search heuristic usually finds a good near-optimal solution under all considered conditions and requires relatively limited computational efforts