Fault detection and diagnosis using correlation coefficients between variables

Abstract

Chemical plants have become increasingly complex and stringent requirements are needed on the desired final product quality. Accurate process fault detection and diagnosis (PFDD) at an early stage of the process is important to modern chemical plants to achieve the above requirements. This paper focuses on the application of fault detection and diagnosis using correlation coefficients between process variables as a PFDD tool. An industrial distillation column is modeled and chosen as the case study. Principal Component Analysis (PCA) and Partial Correlation Analysis (PCorrA) are used to develop the correlation coefficients between the process variables and selected quality variables of interest. Faults considered in this research are sensor faults, valve faults and controller faults. These faults are comprised of single cause faults and multiple cause faults as well as significant faults and insignificant faults. Shewhart Control Chart and Range Control Chart are used with the developed correlation coefficients to detect and diagnose the pre-designed faults in the process. Results show that both methods based on PCA and PCorrA have good PFDD performance. In this study, the PCorrA method was better than the PCA method in detecting insignificant faults.