| In this dissertation, we study the effects of nonstationary demand processes on a Just-in-time manufacturing system that uses the Kanban system as a vehicle. We first show that the Kanban system can be measured by a series of M/M/l/K queues bounded together by some approximation algorithm using which we can derive performance measures such as service level, throughput, inventory, and backlog queue sizes.;We develop a transient analysis of the M/M/l/K queue that enables us to calculate the time-dependent parameters of the queue when the process is undergoing a nonstationary 'warming up' phase which gives us some idea on what to expect in terms of performance when the queue in consideration has not reached steady state yet, as well as what the behavior of the queue is.;We also present a numerical-experimental maximum likelihood approximation (Meier-Hellstern, 1987) to fit a given demand data to get the parameters lambda 1, lambda2, sigma1, and sigma2, that denote two classes of arrivals and the duration of Markov-modulated poisson arrivals. We study the behavior of the system under such nonstationary arrival stream after allocating varying levels of kanban-cushioning throughout the upstream, downstream, and middle cells. The simulation results show that more kanbans in the downstream cells improves the performance of a kanban system under nonstationary demand by a substantial amount.;An empirical study of the Semiconductor Equipment Manufacturing Center (SEMC) of Air Products and Chemicals Inc. (APCI) is done to demonstrate the results developed in this dissertation. First, a time series analysis of the demand data of SEMC is conducted. The analysis shows that demand for certain SKU's is indeed nonstationary. Finally, a simulation of the system with high cushioning of kanbans toward the demand front demonstrates considerable mitigation of the effects of nonstationarity. |
