Multivariate Statistical Process Monitoring Using Modified Factor Analysis and Its Application

A new concept-sensitive factor and a modified factor analysis (FA) modeling approach called process monitoring based on sensitive factor analysis (SFA) are proposed to improve process monitoring performance. The aim of the present study is to solve the factors selection problem, which is a key step in process monitoring based on FA and can directly affect the monitoring result. Process monitoring based on FA usually employs the first several factors that represent the maximum information of normal operating sample, and the determination of factors is rather subjective. Generally, because the GT(2) statistic measures the variation along each of the loading vectors directly, it is possible to find some factors that reflect the dominant variation of abnormal observations, which are termed as sensitive factors in this paper. Additionally, the change rate of the GT(2) of each factor is defined to determine the sensitive factors. Then, a new fault diagnosis approach based on sensitive factors is proposed. Furthermore, a case study on the TE process demonstrates the performance of the SFA model on online monitoring. Results show that the performance based on SFA is improved signally, compared with the classical FA model.