In view of the phenomenon of the degraded diagnostic performance of many rolling bearing fault diagnosis methods based on deep learning under the small sample data set

a bearing fault diagnosis model based on convolution gated recurrent neural network was proposed. This model used a two-layer convolution network to extract features from the input signal

using the tanh function as the activation function

and the pooling layer used the large pooled kernel for overlapping downsampling. The extracted high-level features were connected to the bidirectional gated recurrent network. The last states of the forward and reverse directions of the recurrent network were combined

and a layer of fully connected layers was connected for output. The bearing fault data set of Case Western Reserve University was used to verify the diagnostic performance of the model under the small sample data set. The experimental results showed that the model still maintained 97% accuracy with only 20 training samples compared with other types of models.