Multi-product Manufacturing Systems With Setup Times: Performance Analysis And Design

Multi-product production systems play an important role in the modernmanufacturing industry. In the manufacturing of products such as semiconductor chips,flexible lines can produce multiple product types with limited resources. In such lines, asetup time is often needed when switching from one product type to another, and thesetup time is usually sequence-dependent. In addition, multi-product systems typicallyhave other features such as finite buffers, randomness (such as machine breakdown) andscheduling policies. For researchers and practitioners, it is critical to study theperformance analysis of such systems under given input, as well as the design of systemparameters and scheduling policies.This thesis studies a single-machine, multi-product production system withrandomness. When the machine switches between different products, there aresequence-dependent setup times. In addition, each product type has its own finite buffer,and certain scheduling policy is used to coordinate different products. A continuoustime Markov chain is used for system modeling, and its stationary distributions aresolved for system throughput calculation. Based on this, we further study threeproblems with practical significance: Firstly, how will the system throughput changewith the system parameters? Secondly, when there are various scheduling policies touse, which policy can lead to higher throughput? Thirdly, we want to study the systemrobustness problem by finding a robust scheduling policy, which delivers relativelygood performance under a wide range of system parameters.When analyzing system performance, we obtained an analytical solution for thesimplified two-product case, and used numerical methods for more general cases. Westudied the trend of system throughput when changing parameters (arrival rate, setuptimes, and so on), discovered non-monotonic phenomenon when setup times are longand buffers are small, and proposed a hypothesis of the condition for such phenomenonto occur. When studying different scheduling policies, we have compared sevendifferent policies. Our research indicates that: the best policy is generally among longestqueue policy (LQ), longest processing time policy (LPT) and shortest overall timepolicy (SOT), and bigger buffer sizes are in favor of the SOT policy. To facilitate the practical use of the policy comparison results, and to deal with incomplete or inaccuratedata in real practice, we studied the robustness of scheduling policies. We proposed twoindicators of system robustness, and employed a series of numerical results whichshows: The LQ and Cyclic scheduling policies have demonstrated robust performanceswith the change of system parameters.The results of this study has not only contributed to the theoretical research ofmulti-product systems, but also offered insights for the management of realmulti-product systems.