Soft Fault Diagnosis Using URV-LDA Transformed Feature Dictionary

Dictionary-based fault diagnosis methods, focusing on storing feature patterns of known faults, have been widely used for electromechanical systems. The state of component degradation caused soft faults, however, are continuously changeable. Thus, conventional dictionaries cannot be applied for diagnosis of soft faults with multi-degradation levels. To address this issue, this article develops a new type of dictionary by combining the unit residual signal vector (URV) and the linear discriminant analysis (LDA) for feature transformation, which is referred to as URV-LDA dictionary. The unit residual signal vector keeps the fault feature growth trends but eliminates the degradation severity influence. The linear discriminant analysis is then implemented to find the best projection directions for classification. Specifically, two dictionaries named as the URV-MLDA binary-value dictionary and the URV-SLDA unique-value dictionary are proposed. To validate the efficiency of two developed dictionaries, an electromagnetic relay is carried out and two conventional methods are compared. The comparison results show the developed dictionaries can better solve the soft faults issues with significant increases on diagnostic accuracy.