Mixture Discriminant Monitoring: A Hybrid Method for Statistical Process Monitoring and Fault Diagnosis/Isolation

Abstract

In order to better utilize historical process data from faulty operations, supervised learning methods, such as Fisher discriminant analysis (FDA), have been adopted in process monitoring. However, such methods can only separate known faults from normal operations, and they have no means to deal with unknown faults. In addition, most of these methods are not designed for handling non-Gaussian distributed data; however, non-Gaussianity is frequently observed in industrial processes. In this paper, a hybrid multivariate approach named mixture discriminant monitoring (MDM) was proposed, in which supervised learning and statistical process control (SPC) charting techniques are integrated. MDM is capable of solving both of the above problems simultaneously during online process monitoring. Then, for known faults, a root-cause diagnosis can be automatically achieved, while for unknown faults, abnormal variables can be isolated through missing variable analysis. MDM was used on the benchmark Tennessee Eastman (TE) process, and the results showed the capability of the proposed approach.

Variable selection method for fault isolation using least absolute shrinkage and selection operator (LASSO)

“…In addition, all these methods may face difficulties in handling unknown faults. Recently, a hybrid method was proposed to combine fault classification with B&B to deal with both known and unknown faults [16,17]. However, the shortcoming of B&B is inherited.…”

Adaptive partitioning PCA model for improving fault detection and isolation

“…Chemical processes are often slow-varying and with normal disturbances, so the application of a fixed model based fault detection and isolation might lead to mistake and fail to report the warning of fault. Huang et al [13] proposed mixture discriminant monitoring, integrating supervised learning and statistical process control charting techniques, which also utilizes both normal adaptive multi-block PCA, adaptive consensus PCA, and adaptive multiscale PCA algorithms for updating the model structure to deal with changing process. Zhao and Sun [15] presented relative PCA and multiple time region based fault reconstruction modeling algorithms for fault subspace extraction and online fault diagnosis.…”