Cycle time modeling

In today's manufacturing environments, accurately estimating cycle time is of great importance in achieving on-time delivery. In order to predict the expected system cycle time with confidence, and therefore have a higher chance of meeting on-time delivery to customers, it is essential to understand the behavior of the cycle time variance as well as the mean. This research presents one example of this effort.; Closed queueing networks are often best suited to capture the behavior of manufacturing systems where the number of jobs in the system is a controlled parameter. The first part of this research presents an analytical expression of cycle time mean and variance as a function of the system's work-in-process (WIP) level. The system we consider is a single-class closed Jackson queueing network. Insights gained from the obtained analytical expressions reveal the strong impact of the bottleneck on the cycle time variance as well as the mean. We also learn that, at high levels of work-in-process (WIP), both cycle time mean and variance increase at a constant rate.; Although the obtained exact results on cycle time mean and variance provide useful insights about their behavior as a function of the number of jobs in the system, one can very well argue that the considered system is too simple to represent many production manufacturing systems. First, it is very unlikely that factories today produce one single type of product. In fact, offering a diversified product portfolio is one of the elements of customer satisfaction. Secondly, in a complex manufacturing environment such as the semiconductor wafer fabrication line, processing time at each step is difficult to be characterized.; Consequently, we present a statistical approach to cycle time modeling. Three models of cycle time prediction for data collected at a complex semiconductor manufacturing process are proposed. The models allow us to capture changes in cycle time mean and variance as the WIP level increases. We model cycle time mean by a two-segment piecewise linear function of the current WIP level. We consider three variance models: stepwise constant, piecewise linear and piecewise quadratic. We use the software BUGS to perform a Bayesian analysis to estimate the model parameters. With three competing cycle time models, Bayesian model selection is used to identify the most plausible model. Model averaging is then performed on the selected model.