Bearing Fault diagnosis based on convolution Neural Network with Multi-Attention Mechanism

In the case of load changes and noise interference environment, the single convolutional neural network(CNN) has insufficient ability to extract key features, which makes the fault diagnosis accuracy of the network model low. In order to improve the ability of convolution neural network to obtain characteristic information in interference environment, and then improve the diagnostic accuracy of bearings. In this paper, a fault diagnosis model combining convolution neural network and bidirectional gated cyclic network (BiGRU) is proposed. In this model, the channel attention mechanism is used to optimize the learning mechanism of convolutional neural network, and BiGRU network is used to extract the timing feature information of vibration signals. Then self attention mechanism is used to filter the important feature information, and finally the fault diagnosis results are output at the full connection layer. The experimental results show that, under the condition of variable load, the average accuracy of convolution neural network with multi-attention mechanism is as high as 97.17%. In the case of noise interference, the accuracy rate is as high as 99.43%. The convolutional neural network with multiple attention mechanisms can effectively resist the interference of load changes and noise, and greatly improve the ability of the convolutional neural network to extract key information.