Process monitoring and predictive analytics in multi-stage manufacturing processes using partial least square methods

The quality of products manufactured by a multistage process is often determined by complex interactions among various quality attributes of the multiple stages of the process. As a result, the quality characteristics of a stage are not only influenced by local variation at that stage but also by the propagated variations from the upstream. Therefore, accurate prediction of variables at different stages of operation of a multistage manufacturing process (MMP) is critical for diagnosis and prognostic purposes and ensuring the high quality of the product. At present, there is no generalized model or variation reduction technique available to effectively monitor and control the processes in MMP setup capable of handling large number of variables, as it is very common in MMP. This paper proposes a methodology which can be used for monitor processes at each stage using multivariate EWMA control chart and design a regression model for accurate prediction of the downstream variables using the partial least square (PLS) method. In addition, the paper presents an optimization technique to minimize overall system variance and a goodness of fit test for finding root causes behind an out of control signal. The proposed methods are validated by real data from an auto manufacturing company in Michigan and a natural gas distribution company in Texas.