Fault prediction of rolling bearings using a multi-scale convolutional neural network with parallel BiLSTM for noise environment

This study has developed a multi-scale convolutional neural network (MSCNN) with a parallel bi-directional long short-term memory (PBiLSTM) model for improving the prediction accuracy under noise environment. Firstly, the signals are pre-processed using an ensemble empirical mode decomposition (EEMD) method to solve the problem of mode mixing caused by environmental noise. Subsequently, a MSCNN frame-work is designed to extract bearing feature information and predict faults. Additionally, to tackle the problem of overfitting and vanishing gradients in backpropagation, residual connections are incorporated between layers to dynamically adjust the layer count in the MSCNN. To further mitigate the impact of noise, temporal sequence feature information from vibration signals is extensively utilized by introducing a parallel bidirectional long short-term memory (BiLSTM) model at the output layer of the MSCNN structure. The results show that, compared to traditional methods, the proposed model achieves higher fault prediction accuracy under different levels of noise environment.