Nonlinear and Non-Gaussian Dynamic Batch Process Monitoring Using a New Multiway Kernel Independent Component Analysis and Multidimensional Mutual Information Based Dissimilarity Approach

Batch or semibatch process monitoring is a challenging task because of various factors such as strong nonlinearity, inherent time-varying dynamics, batch-to-batch variations, and multiple operating phases. In this article, a novel nonlinear and non-Gaussian dissimilarity method based on multiway kernel independent component analysis (MKICA) and multidimensional mutual information (MMI) is developed and applied to batch process monitoring and abnormal event detection. MKICA models are first built on the normal benchmark and monitored batches to characterize the nonlinear and non-Gaussian variable relationship of batch processes. Then, the kernel independent component (IC) subspaces are extracted from the benchmark and monitored batches. Further, a multidimensional mutual information based dissimilarity index is defined to quantitatively evaluate the statistical dependence between the benchmark and monitored subspaces through the moving-window strategy. With the corresponding control limit estimated from the kernel density function, the integrated MKICA MMI index can be used to detect the abnormal events in dynamic batch processes. The effectiveness of the proposed batch process monitoring approach is demonstrated using the fed-batch penicillin fermentation process, and its performance is compared to that of the MKICA method. The computational results show that the presented dissimilarity approach is faster and more accurate in detecting different types of process faults.