Stochastic analysis of multi-item flow lines

This research provides important results for modelers and managers of multi-item production systems. An inherent characteristic of multi-item production systems is the service time dependencies between adjacent stations. The service times for different job types at the same station are generally different. As distinct job types move through the system, they create processing time dependencies between stations. Even systems that produce similar jobs may have dependencies between station service times because the outcome of tasks at an upstream station may affect the service times at a downstream station. However, previous studies typically model this type of production system by a single "aggregate" job type with jobs having independent service times. We develop a model to describe these dependencies and evaluate their impact on system performance. Our results indicate an invalid assumption of independent service times can lead to poor estimates of system performance and poor design decisions. The overall objective of this research is to provide guidelines for system design and control procedures that will improve the performance of multi-item production lines.;The contributions of this research include useful analytical and numerical results that suggest the effect of service time parameters such as, variability, distribution shape, and dependencies on line capacity, WIP, and system time distribution. This research also provides results on how to improve delivery performance in production line systems with finite and infinite capacity buffers. We show the practical demonstration of this research to several important design problems which include order of stations in flow lines, allocation of product to multiple lines, job release process, and buffer allocation. We provide a new methodology for estimating job flow allowances. Our results suggest that models based on this new approach provide improved delivery performance when compared to established methods. In addition, they require less data collection, less data maintenance, and less computational effort to estimate model parameters. Our results also indicate the impact of several shop factors: number of stations, number of buffers, service time dependencies, service time variability, inter-arrival time variability, and line congestion on delivery performance.