Analytical Esitmation of Throughput Distribution for Serial Manufacturing Systems with Muli-State Machines and its Application.

An accurate estimate of throughput is necessary for design, improvement and management of a manufacturing system. In literature, most studies focus on the steady state behavior of manufacturing systems, estimating only their average performance. Due to uncertainties, such as machine failures, the standard deviation associated with a manufacturing system over a week, might be over 10% of its mean. By estimating the throughput distribution of a manufacturing system, higher order estimates of a system's performance can be achieved to increase system predictability and dependability. Further, in contrast to conventional two state machine models, this study also incorporates multi-state machine models, to include condition monitoring within manufacturing system decision making. The throughput distribution of a serial manufacturing system with multi-state machines was analytically estimated by first, extending a conventional two state machine model to a three state continuous time Markov chain model, to include preventive maintenance (PM). Secondly, the throughput distribution and performance measures for a generalized two machine one buffer system were estimated by uniformizing the machine models and by modeling the non linear machine buffer interactions. Further, considering Markov nature of the system, an aggregation approach was developed to obtain the overall throughput distribution of the system. Finally, a segmentation approach was developed to divide a serial manufacturing system into smaller segments. For each of the segments, modes of operation, mode dynamics and mode probabilities were estimated to obtain the overall throughput distribution. The PM feasibility and optimal rate of PM were derived for a three state machine model. The optimal PM was shown to provide significant improvements in efficiency. The developed analytical approach for estimating throughput distribution was found to be an order of magnitude faster than a discrete event simulation approach. Further, an application of estimating throughput distribution to condition based maintenance decision making and optimal buffer design was also shown. The developed analytical approach is based on the fundamental principles of manufacturing system interactions and can be extended to analyze different forms of complex manufacturing systems, and shows great potential for being used as a condition based production and maintenance decision making tool.PhDMechanical EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate StudiesRemoved from view by the University of Michigan.http://deepblue.lib.umich.edu/bitstream/2027.42/84489/4/SaumilAmbaniDisserationRevised_final.pd