A New Research of Fault Detection Method for Batch Processes

A phase segmentation strategy to handle the batch processes with multiple operation phases is proposed in this paper. It can not only obtain an accurate stage classification and avoid falling into local optimum, but also improve the precision of modeling and the accurate of fault detection. Firstly, the three dimension batch process is handled by integrating several expansion methods such that the data estimation step may be removed. And then, the spectral clustering method together with graph partitioning criterion is utilized to divide the above data into several clusters which is further modeled via PCA. Finally, a suitable model is applied to monitor these data on line. The effectiveness and flexibility of the proposed method is validated through penicillin simulation process.