Diesel engine injection faults' detection and classification utilizing unsupervised fuzzy clustering techniques

Abstract

By using the unsupervised fuzzy clustering, this study attempts to design a new scheme for the unsupervised detection and classification of two injection faults using the time–frequency analysis of vibration signals of an internal combustion, four-stroke, diesel engine with six cylinders in-line. To reach this objective, two new methods called modified S-transform and two-dimensional non-negative matrix factorization are used. Three fuzzy clustering algorithms and nine cluster validity indices, for a variable number of classes, are also used to detect and classify the fault classes. The implementation of these methods resulted in a high detection rate of the injection faults.