Phase adaptive RVM model for quality prediction of multiphase batch processes with limited modeling batches

For quality prediction of batch processes under limited modeling batches, the relevance vector machine (RVM) has recently been introduced. By unfolding the three-way dataset through the variable direction, significant nonlinearities are remained in the process data, which in turn explored the nonlinear modeling ability of RVM. For multiphase batch processes, however, different phases may have simultaneous impacts on the final product quality, which should be connected together in the modeling stage. In this paper, a new phase adaptive RVM model is proposed for quality prediction in multiphase batch processes. Based on the information transfer of relevance vectors in each RVM model, different phases are connected one after another, providing simultaneous information for prediction of the final product quality. A detailed industrial case study is given to show the efficiency of the new developed method.