Aiming at the problems that traditional deep learning methods do not make full use of the timing characteristics of bearing signals, and are difficult to process dynamic data, an improved CNNBiGRU intelligent diagnosis method for bearing faults is proposed. The convolutional neural network is used to extract representative features from the input signal, and the bidirectional gated recurrent neural network is introduced to mine the semantic information in the time dimension of the fault data, and the attention mechanism is used to adaptively assign different weights to the feature map to achieve high precision diagnosis of bearing faults. Experiments on public bearing data sets show that the method can correctly classify bearing faults with a classification accuracy of 996%。