Adaptive multiwavelets via two-scale similarity transforms for rotating machinery fault diagnosis

Abstract

Fault diagnosis of rotating machinery is very important and critical to avoid serious accidents. However, the complex and non-stationary vibration signals with a large amount of noise make the fault detection to be challenging, especially at the early stage. Based on the inner product principle, fault detection using wavelet transforms is to match fault features most correlative to basis functions, and its effectiveness is determined by the construction and choice of wavelet basis function. In this paper, a new method based on adaptive multiwavelets via two-scale similarity transforms (TSTs) is proposed. Multiwavelets can offer multiple wavelet basis functions and so have the possibility of matching various fault features preferably. TSTs are simple and straightforward methods to design a series of new biorthogonal multiwavelets with some desirable properties. Using TSTs, a changeable and adaptive multiwavelet library is established so as to provide various ascendant multiple basis functions for inner product operation. By the rule of kurtosis maximization principle, optimal multiwavelets most similar to the fault features of a given signal are searched for. The applications to a rolling bearing of outer-race fault and a flue gas turbine unit of rub-impact fault show that the proposed method is an effective approach to detecting the impulse feature components hidden in vibration signals and performs well for rotating machinery fault diagnosis.