Online batch process monitoring based on multi-model ICA-PCA method

Multiway principal component analysis (MPCA) has been widely used to monitor batch processes. However, due to the nature and complicated changes of batch processes, the data in fact contains inherent non-Gaussian information. Besides, when used for on-line monitoring, MPCA needs future value estimation. These shortcomings may lead to poor monitoring performance. In this paper, a new statistical batch process monitoring approach based on Multi-model independent component analysis (ICA) and PCA is proposed, using ICA to monitor non-Gaussian information of the process, and then PCA is applied for the rest Gaussian part. Further more, the proposed method does not require prediction of the future values, since we build sub-models for every sample time of the batch, and it can also be used for batch processes in which the batch length varies. The simulation results of penicillin batch process show the power and advantages of the proposed method, in comparison to MPCA.