Tensor Slow Feature Analysis for Monitoring Batch Process

For accurately monitoring complex batch processes, the three-dimensional (3-D) structure and implied dynamics should be fully handled. To this end, a novel tensor slow feature analysis (TSFA) method is proposed to improve the batch process monitoring performance. In the proposed method, the 3-D data can be modeled directly without data unfolding to avoid the deficiency of destroying the raw data structure and increasing the modeling parameters in most existing methods. The slowly varying dynamics within batch processes can be efficiently extracted by solving two sub-optimal problems in the proposed TSFA method. Based on the defined monitoring statistics, within-batch detection can recognize the abnormal situation timely. A penicillin fermentation process is used to illustrate the advantages of the proposed method in comparison with the existing methods.