Aiming at the problem of poor robustness of rolling bearing fault diagnosis model under noise interference,a fault diagnosis method based on the combination of wavelet threshold denoising (WTD),AR spectrum and mind evolutionary algorithm (MEA) optimization back-propagation neural network (BPNN) was proposed.Taking the original vibration signal as the input,the wavelet method was used to decompose and reconstruct the original signal to filter out high-frequency noise,then the Burg algorithm was used to estimate the parameters of the AR model to extract the power spectrum features of the noise-reduced signal,finally the feature vector and the corresponding label were used as the input and output of MEA-BPNN neural network to train,then the diagnosis was realized.The proposed method was compared with some advanced artificial neural network diagnostic methods to test the performance of the diagnostic model.The research results show that the WTD-AR spectrum-MEA-BPNN diagnostic model can effectively reduce the noise interference of bearing vibration signals,achieve feature enhancement and has higher resolution;compared with traditional neural networks,the training speed is faster,and the fault type can be identified in a shorter time with high recognition rate.