Rolling bearing fault diagnosis method based on multi-scale pooling residual convolutional neural network under noisy environment

To address the issues of unstable performance and poor generalization ability of bearing fault diagnosis model caused by strong noise and variable operating conditions, a novel method based on multi-scale pooling residual convolutional neural network (MSPRCNN) is proposed in this paper. Firstly, by converting vibration signals to frequency domain with Fourier Transform (FT) and utilizing wide convolution kernels for feature extraction, the approach enhances fault detection. Then, a multi-scale pooling feature extraction (MSPFE) module is presented, which captures information at different scales to simplify complexity, while an up-sampling position attention (UPA) module is designed to establish correlations between frequency domain positions. Finally, the MSPRCNN model is built, which employs gated convolution (GC) instead of standard convolution to reduce the impact of noise and solve the problem of vanishing gradient, and the IReLU activation function is put forward to improve model feature representation. Experimental results on two datasets show that the fault recognition accuracy is 98.71% under variable loads and 98.2% under variable speeds. The MSPRCNN model outperforms other methods in fault recognition accuracy and generalization in noisy environments.