Fault diagnosis of bearings based on a sensitive feature decoupling technique

Abstract

Bearings are commonly used in machine industry, and their faults may result in unexpected vibration and even cause breakdown of a whole rotating machine. This paper proposes a novel fault diagnosis approach for bearings by using a sensitive feature decoupling technique. This approach does not require a training procedure as in machine learning methods and can classify the occurred faults by a simple algebraic computation. Firstly, the features of vibration signals which show the most significant difference under different bearing health conditions are selected and defined as sensitive features. Then those sensitive features under different health conditions are used to construct a feature matrix, and its left null space is computed to obtain the so-called feature decoupling vectors. The bearing faults are finally classified with the help of the decoupling vectors according to a simple decision logic. Since the obtained decoupling vectors may not be unique, we also propose an algorithm to select the optimal ones in order to improve the performance of fault diagnosis. Experiments are carried out to test the proposed approach and the results show that the approach is feasible and effective for the fault diagnosis of bearings.