Research on Bearing Fault Diagnosis Methods in High Noise Environments

In the fault diagnosis of rolling bearings, due to the characteristics such as complexity, interference, and irregularity of the noisy signal, the problem of decreased accuracy of fault classification will occur. In response to these problems, a fault diagnosis method combining adaptive threshold denoising and convolutional neural network is proposed. Firstly, an Adaptive Threshold Denoising Method is constructed, which uses a sliding window to set the threshold and can enhance the system's noise immunity. Secondly, a Parallel Convolutional Attention Model is designed. Through the way of vector padding, the different-scale features extracted by the square kernel and the vector convolutional kernel are fused, and the channel attention adaptively allocates the weights of different channels according to the features captured by the convolution, which improves the accuracy of fault diagnosis while reducing the calculation amount by k2/d times. Finally, through simulation, it is shown that in the high-noise environment with SNR=0 dB, this method can achieve a relatively high accuracy.