Chemical processes monitoring based on weighted principal component analysis and its application

Conventional principal component analysis (PCA)-based methods employ the first several principal components (PCs) which indicate the most variances information of normal observations for process monitoring. Nevertheless, fault information has no definite mapping relationship to a certain PC and useful information might be submerged under the retained PCs. A new version of weighted PCA (WPCA) for process monitoring is proposed to deal with the situation of useful information being submerged and reduce missed detection rates of T2 statistic. The main idea of WPCA is building conventional PCA model and then using change rate of T2 statistic along every PC to capture the most useful information in process, and setting different weighting values for PCs to highlight useful information when online monitoring. Case studies on Tennessee Eastman process demonstrate the effectiveness of the proposed scheme and monitoring results are compared with conventional PCA method.

► Weighted PCA is proposed to highlight the useful information for process monitoring. ► The situation of useful information being submerged is analyzed. ► The change of T2 statistic along each principal component is examined. ► Fault information is taken into consideration timely while online monitoring. ► Monitoring result of T2 statistic for both fault detection and diagnosis is improved.