Data analysis and exploration for a fault detection, diagnosis, and prognosis system

Abstract

The subject of this paper is a statistical fault detection system with the scope of detection, diagnosis and prognosis. It was designed using the fundamental procedures of data analysis and exploration: recognizing atypical elements (outliers), clustering, and classification, based on the nonparametric methodology of kernel estimators. Employing a homogenous mathematical apparatus for all three of the above tasks significantly facilitates practical implementation. The formula for the proposed concept is universal in character, and the investigated system can be applied in a wide range of tasks, particularly in engineering and management. Experimental tests showed its effectiveness in identifying abrupt as well as slowly progressing anomalies. For the latter case in particular, the still rarely-used function for prediction of faults prevailed.