(113a) Fault Isolation Using a Multiscale PCA Reconstruction-Based Approach

Process monitoring and control have become more automated in the past decade through the utilization of data-driven techniques. This is motivated by the need to improve the efficiency of operation and to avoid undesirable outcomes that may result from failing sensors. Thus, effective fault detection and isolation methods are essential for efficient operation of industrial processes. This work proposes a new fault isolation technique for process monitoring applications, which extends the abilities of PCA reconstruction-based isolation using wavelet multiresolution (MR) analysis.

Multiresolution wavelet analysis is a well-established and widely-implemented tool for fault detection. The wavelet coefficients inherently possess characteristics that enable efficient separation of deterministic and stochastic features in measured data, and help deal with autocorrelated noise [1]. PCA reconstruction-based isolation has also shown advantages over other PCA isolation techniques, such as structured residuals and contribution plots [2]. This is due to its ability to deal with the residual smearing effects, which hinder effective fault isolation.

The main contribution of this work is the development of an algorithm for isolation that utilizes the benefits of both MR analysis and PCA reconstruction-based fault isolation. The effectiveness of the developed algorithm will be demonstrated through a simulated example of a sensor fault that occurs in a chemical process unit, e.g. a CSTR reactor. The isolation results obtained by implementing PCA reconstruction-based contribution with MR analysis is compared to those obtained by implementing only the PCA reconstruction-based method (without MR analysis) in order to examine the advantages achieved by the developed algorithm. Further validation of the results will be carried out, by comparing the developed algorithm to previous work that performed MR analysis with contribution plot isolation [3].

References

[1] B. Bakshi, Multiscale PCA with application to multivariate statistical process monitoring, AIChE J. 44 (1998) 1596–1610.

[2] R. Dunia, F. Systems, S.J. Qin, T.F. Edgar, T.J. Mcavoy, Identification of faulty sensors using principal component analysis, AIChE J. 42 (1996).

[3] M. Misra, H.H. Yue, S.J. Qin, C. Ling, Multivariate process monitoring and fault diagnosis by multi-scale PCA, Comput. Chem. Eng. 26 (2002) 1281–1293.