Abstract:

The healthy state of bearings is very important for the normal operation of rotating machinery such as radar driving structure and helicopter transmission mechanism. Aiming at the characteristics of complex working conditions, noise, and insufficient and unbalanced samples of the fault labels of the vibration signals of rolling bearings, this paper proposes an improved one-dimensional convolution neural network fault diagnosis method for rolling bearings based on deformable convolution of the disturbance training samples and depth residual block structure. The deformable convolution is set to improve the ability of extracting local fault features, and the improved depth residual block is introduced to improve the generalization ability and sensitivity of the model to the training data. When the training data are fed, the training disturbance layer is set to add disturbance samples to improve the robustness of the model. The Case Western Reserve University bearing data set is used as the experimental data set to divide the training set and the test set. The experimental results prove the effectiveness of the proposed method. TD-DCCNN algorithm can still achieve an average accuracy of 90.35% when the signal-to-noise ratio is 0, which has certain advantages compared with other diagnostic algorithms.