LAUSR

Abstract In process monitoring, some specific performance indexes need to be paid attention to. Therefore, the performance-triggered process monitoring scheme is proposed. Different from the traditional process monitoring method, the process is considered normal if there is no apparent anomaly happens on the performance index. In order to predict the values of performance indexes that cannot be measured in real time, ridge regression is used. And, the regression coefficients are used to pick the most relevant process variables for subsequent modelling. In this scheme, after the performance index exceeds the control limit, the monitoring of the relevant process variables is triggered to determine whether the prediction is abnormal due to the occurrence of a fault. Then, dictionary learning method and Low rank representation (LRR) are used for feature extraction and construction of the statistic. Finally, the effectiveness of the proposed method is verified by a numerical example and the Tennessee Eastman (TE) process.