Irregular Phase Partition and Synchronization Based on Window PCA Model for Uneven Batch Process Monitoring

摘要

In general, batch process exhibit uneven multiphase characteristics in batch-wise, resulting in no significantly statistical characteristics over batches. In order to handle this issue, this paper first proposes a stepwise sequential phase partition method based on window principal component analysis (PCA) model for accurate irregular phase identification. Then, in order to align similar process events of uneven phases, a dynamic time warping (DTW) based method is proposed for synchronization of uneven phase trajectories, where the optimal target length is studied from the aspect of statistical analysis. Besides, for data supplement in phase synchronization, a window-based PCA models are developed for data estimation. In this way, uneven length phases are synchronized to the target length and similar process events over batches correspond at the same time, which makes statistical characteristics well revealed. In the end, a variable-unfolding based modeling method is designed for each phase. So in online monitoring, samples are projected into the corresponding model to achieve accurate monitoring. In order to verify the effectiveness of the proposed method, the proposed methods are performed on gear transmission process with the uneven length problem.