Constant component separation for nonlinear time‐varying process monitoring based on stationary subspace analysis and one‐class <scp>SVM</scp>

Abstract Chemical processes often exhibit non‐stationary characteristics, making traditional multivariate statistical methods unsuitable. Stationary subspace analysis (SSA) is an effective approach for handling non‐stationary variables by decomposing multivariate non‐stationary time series to identify stationary components. Additionally, the complexity of modern industrial processes introduces nonlinear relationships among variables. One‐class support vector machine (OCSVM) uses kernel tricks to map low‐dimensional nonlinear data into high‐dimensional space for linear separability and employs hyperplanes in high‐dimensional space for anomaly detection. Industrial data typically possess both non‐stationary and nonlinear characteristics; however, most existing methods, including SSA and OCSVM, address only one aspect, failing to comprehensively extract features and resulting in suboptimal monitoring performance. This study proposes an SSA‐OCSVM collaborative monitoring strategy for non‐stationary processes. By extracting stationary components with SSA and capturing nonlinear features with OCSVM, a monitoring statistic (MS) is constructed to enhance process monitoring. The proposed strategy effectively integrates the synergistic effects of both models on monitoring performance. Two case studies were conducted to validate the strategy and compare it with other methods. The results show that this method can comprehensively extract non‐stationary and nonlinear features, making it suitable for monitoring non‐stationary processes. Moreover, the monitoring results outperform those of single‐objective methods, enhancing the robustness and accuracy of the monitoring performance.