(136h) Concurrent Canonical Variate Analysis for Process Operating Condition Deviations and Dynamic Anomalies Monitoring

Data-driven process monitoring methods are widely applied in industry for fault detection to improve process safety and product quality (e.g., [1]). Conventional process monitoring methods rely on the assumption that the characteristics of the data variations are relatively unchanged at normal operating conditions. However, variations in operating conditions are likely and common, which can result in high false alarm rates. Recursive process monitoring methods update the model with respect to the changing normal operating conditions, but cannot effectively distinguish between a slow drifting fault and a changing normal operating condition (e.g., [2]). The quality monitoring method resolves part of the problem, in the sense that the system can tell whether a deviation affects the quality variables (e.g., [3]). However, the quality monitoring method fails to trigger the alarm when dynamic process anomalies are present. To address these issues, we extend the concurrent canonical correlation analysis of Zhu et al. [4] for static quality monitoring to accommodate system dynamics. Concurrent canonical variate analysis (CCVA) is developed with two groups of monitoring indices to monitor quality-relevant fault and system dynamic anomaly. A real fault is considered as either the deviations of the quality-relevant variables or the abnormal temporal dynamic behaviors. Due to the capability of CVA on dealing with the dynamic data, the proposed approach has higher accuracy compared to other conventional quality monitoring methods, and the corresponding two proposed groups of monitoring indices can distinguish whether the disturbance in operating conditions is normal or a fault, which yields a complete description of the system. The Tennessee Eastman process is used to illustrate the effectiveness of CCVA-based monitoring method.

[1] L.H. Chiang, E.L. Russell, and R.D. Braatz (2000) Fault Detection and Diagnosis in Industrial Systems. Springer Verlag, London, United Kingdom.

[2] W. Li, H.H. Yue, S. Valle-Cervantes, and S.J. Qin (2000) Recursive PCA for adaptive process monitoring. Journal of Process Control, 10, 471-486.

[3] J.F. MacGregor, C. Jaeckle, C. Kiparissides, and M. Koutoudi (1994) Process monitoring and diagnosis by multiblock PLS methods. AIChE Journal, 40(5), 826-838.

[4] Q. Zhu, Q. Liu, and S.J. Qin (2016) Concurrent canonical correlation analysis modeling for quality-relevant monitoring. IFAC-PapersOnLine, 49(7), 1044-1049.