Process Fault Detection Method Based on Time Structure Independent Component Analysis and One-Class Support Vector Machine

摘要

Existing fault detection method based on fast independent component analysis (FastICA) can only extract non-Gaussian independent components (ICs) and cannot consider the serial correlations of monitoring statistics. In this paper, a new fault detection method based on time structure ICA (TSICA) and one-class support vector machine (OCSVM) is proposed. TSICA is used to extract the ICs not restricted to non-Gaussian distributions. Then, the difference statistics are calculated by the different time delays of each traditional monitoring statistic to capture the serial correlations. Furthermore, all the statistics are combined together to construct an OCSVM statistical model and a unified OCSVM-based monitoring statistic is built. The effectiveness of the proposed approach is evaluated using the Tennessee Eastman benchmark industrial process. Simulation results demonstrate that the proposed method achieves superior fault detection performance in comparison to the conventional FastICA-based method.