Fault diagnosis of rolling element bearing weak fault based on sparse decomposition and broad learning network

Abstract

Rolling element bearings are widely used in rotating machinery and, at the same time, they are easily damaged due to harsh operating environments and conditions. As a result, rolling element bearings are critical to the safe operation of the mechanical devices. The incipient fault information extraction of rolling bearings mainly faces the following difficulties: (1) The fault signal is too weak. (2) The fault mechanism and the dynamic model of the rolling bearing system are complex. (3) The oscillations caused by the fault shocks are overlapped due to the smaller impact between two adjacent faults. (4) The impact interval of the fault will change randomly. To overcome the aforementioned difficulties, a connection network constructed by resonance-based sparse signal decomposition (RSSD) and broad learning system (BLS) without the need for deep architecture, namely RSSD-BLS, is proposed for intelligent fault diagnosis. We construct RSSD-BLS by input layer, RSSD decomposition layer, feature layer and output layer. So, when the observed vibration signals are the input layer, the network first uses RSSD to decompose the raw vibration signal into high resonance components and low resonance components. Then, the network obtains energy spectrum features of high resonance components which decomposed by RSSD to extract the unique features in the feature. Finally, the network recognizes different fault conditions in the output layer. Through comparing with commonly used intelligent network diagnosis method, the superiority of the proposed RSSD-BLS is verified.