A novel fault diagnosis method of wind turbine bearings based on compressed sensing and AlexNet

Abstract

Abstract The bearing is the core component which ensures the normal operation of the wind turbine. The vibration signal based on fault diagnosis is non-linear, non-stationary and causes serious noise pollution. Conventional methods are difficult to demodulate, and the operation is complex. With the increasing capacity of wind turbine assemblers, the signal samples based on Shannon sampling theorem are also increasing, which brings great pressure to data transmission and storage. Deep learning based on big data-driven for wind turbine running condition monitoring plays an effective role in the field of fault diagnosis. However, data training depends on a large amount of data and takes a long time. Therefore, a novel fault diagnosis method based on compressed sensing (CS) and AlexNet is proposed. This method used small sample data. Firstly, the signal is sparsely processed by stagewise orthogonal matching pursuit algorithm, so that the sparse signal is convenient for signal transmission and can alleviate the pressure of signal storage. Secondly, the CS theory is used to restore the signal and reduce the noise of the signal. Thirdly, a fault-free signal is selected and compared with the fault signal with the same phase to obtain the characteristic residual signal. Next, they are subjected to continuous wavelet transform to obtain the wavelet spectrum of the signal. Finally, it is constructed into a pseudo-trichromatic graph and put into the improved AlexNet network to obtain the effect of fault diagnosis. Compared with other methods, experiment shows that the proposed method has higher accuracy in wind turbine fault diagnosis