A Novel Method for Gearbox Fault Diagnosis Based on Shift-Invariant and Shared Dictionary Learning

Gearbox fault diagnosis is a very crucial part in the field of mechanical diagnosis, and many methods are proposed to solve this problem. Either way, the core idea of fault diagnosis is to extract the class-specific features from the signals collected by different sensors. Although different types of faults own different class-specific features, they also generally possess common features. Especially for gearboxes, there are many complex components in the vibration signals, where many features may belong to more than one fault class, so analyzing and removing these common features before feature extraction is a quite significant process for a more efficient diagnosis. However, at present, few articles are proposed to solve the problem of common features. Inspired by a shared idea, we proposed a novel fault diagnosis method named shift-invariant and shared dictionary learning (SISDL), which consists of shared dictionary and diagnostic dictionary with shift-invariant. The shared dictionary is responsible for learning the common features from different fault classes. Meanwhile, the diagnostic dictionary is responsible for learning the class-specific features. Our proposed method with other five state-of-the-art dictionary learning methods is applied in a experimental gearbox case, and the results verify the advantage of SISDL over the others.