Fault diagnosis of gearbox based on stacked sparse autoencoder

Aiming at the complex feature extraction and low recognition accuracy of gear box fault diagnosis, a fault diagnosis method based on stack sparse self-encoder (SSAE) is proposed. The fault signal is pre-processed by time domain analysis, and then it was input into sparse self-encoder network for feature optimization and dimensionality reduction to extract tables. The feature of intrinsic information of signature signal is input into Softmax classifier to classify surface defects of strip. The experimental results show that the proposed method can achieve high recognition accuracy under the same and mixed conditions. For the mixed conditions, the recognition accuracy reaches 99.5%, which is higher than other models proposed in this paper. Therefore, the proposed method can be effectively applied to gear box fault diagnosis.