Improved multivariate statistical process control for chemical process fault detection and diagnosis

Abstract

This thesis demonstrates the application of Multivariate Statistical Process Control (MSPC) monitoring method that is capable of detecting and diagnosing process faults. Conventionally, r Control Chart and Contribution Chart, which have been widely used for these purposes, are not accurate and sensitive enough to detect and diagnose abnormal changes in operating conditions. In order to overcome these problems, the objeGtive of this research is to develop new approaches, which can improve the performance of the present conventional MSPC methods. Three new approaches have been developed i.e., the Outline Analysis Approach for examining the distribution of Principal Component Analysis (PCA) scores, the Correlation Coefficient (Cik) Approach for detecting changes in the correlation structure within the variables, and the Signal Cumulating Approach for gathering more information regarding the fault. In order to implement the three new approaches, this research proposed PCA Outline Analysis Control Chart and Correlation Coefficient (Cik) Control Chart for fault detection; and the r Score Contribution Chart, the Cik Score Contribution Chart, r Score Contribution Chart with Signal Cumulating Approach and the Cik Score Contribution Chart with Signal Cumulating Approach for fault diagnosis. The results from the conventional method and new approaches were compared based on their accuracy and sensitivity. Based on the results of the study, the new approaches generally performed better compared to the conventional approaches, particularly the PCA Outline Analysis Control Chart and C;k Score Contribution Chart with Signal Cumulating Approach.