Performance-driven ensemble learning ICA model for improved non-Gaussian process monitoring

Abstract

Although successful application studies of independent component analysis (ICA) have been reported for non-Gaussian process monitoring, there are several drawbacks of this method, which make it cumbersome for practical utilization. First, due to the random initialization of the ICA algorithm, the monitoring performance of the ICA-based method is unstable, which may confuse the result. Second, the number selection of retained independent components (ICs) is still an open problem. Third, how to measure the importance of each IC for process monitoring purpose is also a difficult task so far. To address these issues, this paper intends to improve the ICA statistical monitoring method by incorporating the ensemble learning approach and the Bayesian inference strategy. Besides, a new performance-driven approach for IC number selection is also proposed. As a result, the stability of the non-Gaussian process monitoring result is greatly improved. Meanwhile, the monitoring performance is also boosted up, which is illustrated through the Tennessee Eastman (TE) benchmark case study.