Buffer Capacity Allocation in Job-Shop Production Systems with Fixed Service Orders and Routes

Abstract

We consider optimal resource allocations, especially buffer space allocation, in job shop production systems that can be modeled by a class of queueing network systems with variable service times, fixed order of services at each server and fixed route of each part. Under a given topology of the production network, processing schedule, service time distributions and total amount of buffer spaces available, the objective is to find an allocation of buffers for the servers that maximizes system throughput. To resolve this type of problem, one needs to perform two major tasks. One is to develop an approximation scheme or model for calculating the performance index (throughput values) since it is generally hard to evaluate them exactly. The other is how to solve the optimal buffer space allocation problem which is NP hard. Our approach is based on the so-called "sample-path optimization" process combined with the "common random number" technique. Under randomly generated service times, the sample-path of departure times of parts from each server, which can be determined by a set of simple recursion formulae (simulation run), is used for estimating the system throughput, and we search for the optimal buffer allocation that maximizes throughput. As a solution method for the approximate buffer space allocation problem, we use a simple genetic algorithm. Numerical experiments shown that our approach is applicable for larger sized problems.