Fault Signal Separation Method for Rolling Bearing of Wind Turbine Based on Multi-Channel DCNN

Maoju Ran

Abstract: The operating conditions of rolling bearings of wind turbines are variable and complex, so it is of great significance to accurately analyze the type of bearing failure, damage degree and fault location for improving the safety of wind turbines. In this study, a method of fault signal separation for rolling bearing of wind turbine based on multi-channel DCNN is proposed, and a kurtosis and envelope spectrum comparison method is proposed to evaluate the blind source signal separation effect. Firstly, the bearing signal was analyzed by Short Time Fourier Transform (STFT) to generate a Binary Time-Frequency Mask (BTFM) that can characterize the bearing fault. Secondly, the multi-channel DCNN signal separation model is established, and the time spectrum and binary time-frequency mask are used as training samples to train the model. Then, the mixed bearing time spectrum is used as the input of the model to obtain the binary mask of each fault. Finally, multiplying the mixed time-frequency spectrum by the corresponding fault time-frequency mask can obtain the time-domain signal that contains various faults in the mixed signal. Experimental results on vibration data set of Case Western Reserve University show that the proposed method can effectively separate bearing fault signals and obtain accurate information such as fault type, damage degree and fault location. This method can realize automatic bearing fault feature learning and fault signals and the separation of mixed signal, to improve the safety of the wind turbine.

Keywords: Wind turbine, rolling bearing, fault characteristic analysis, multi-channel DCNN, blind source signal separation, binary time-frequency mask