Statistics Mahalanobis distance for incipient sensor fault detection and diagnosis

Abstract

For modern industrial processes, many sensors equipped operate in harsh environments and the large number of sensors increases the probability of sensor malfunction. In order to guarantee an optimal and efficient operating condition, incipient sensor fault detection and diagnosis become necessary and important. In the present work, a new data-driven process monitoring method called statistics Mahalanobis distance (SMD) is proposed for incipient fault detection of three common sensor fault types. Detectability analysis of SMD is provided and compared theoretically with the conventional approach. Besides, the effects of parameter selection in SMD on its fault detectability is briefly discussed. Then, a hierarchical strategy is proposed for subsequent fault diagnosis, including the fault isolation and fault classification aspects. Simulation studies on a numerical example and a benchmark process are carried out, which demonstrate the effectiveness and merits of the SMD based fault detection and diagnosis methods, in comparison with conventional approaches.