A Bayesian Model of Cycle Time Prediction

In this paper, we propose a Bayesian statistical approach to cycle time modeling. Three models of cycle time in complex manufacturing environments are proposed. These models capture changes in cycle time mean and variance at different levels of work-in-process. We model cycle time mean during a period as a two-segment piecewise linear function of the periodʹs work-in-process and consider three variance models. The challenge is to estimate the breakpoint between the two segments, and the parameters of each model. To accomplish this, we use the Gibbs sampler and Metropolis–Hastings algorithm to perform a Bayesian analysis. With three competing models, Bayesian model selection is used to identify the most plausible and model averaging is performed on the selected model. We compare the resulting model to an analytical non-linear model on an example and provide some insights.