Cross correlation-based blocking whitening slow feature analysis for coupled-data industrial process monitoring

Modern industrial systems are characterized by high complexity, strong coupling, and multi-source data interactions. Practical industrial process data are mostly nonlinear, dynamical, and manifested as temporal correlation, which makes it challenging for traditional monitoring methods to accurately capture the changes in internal state variables. To this end, a multi-block whitening slow feature analysis (MBW-SFA) approach is proposed in this paper, which utilizes the cross maximum information coefficient (CMIC) for nonlinear correlation analysis, blocking, as well as selective whitening transformation (SWT) for the nonlinear and strongly coupled process variables, so as to avoid the information loss caused by global whitening. The proposed MBW-SFA approach performs manifold mapping on the data with strong correlations, preserving key information while reducing dimensionality, and the selective whitening transformation is applied to prevent information loss across different variables. In addition, for scenarios involving partially known fault data, this study proposes a control limit optimization (CLO) function that balances fault identification and false alarms to calculate control limit thresholds based on slow feature monitoring statistics, achieving objective monitoring of industrial processes. The proposed approach is experimentally validated on Tennessee Eastman process, where the correlation between process variables is investigated, and the results show that the proposed method achieves excellent performance in process monitoring.