Fault Prediction Method of Gear Based on DSAE and GRU Network

Aiming at the problems of excessive reliance on manual experience and processing of complex signals in traditional gear fault prediction methods, a gear fault prediction method based on DSAE and GRU network is proposed. Firstly, the time domain features of the gear vibration signal are extracted and divided into the training samples and the test samples. Secondly, a DSAE network is trained to adaptively extract the gear health index using the training samples. Then, the GRU network is trained with the health index of the training samples. The optimal fault prediction model is created by constantly adjusting the number of hidden layer neurons. Finally, the health index extracted from the test samples are used to verify the fault prediction model. In order to verify the effectiveness of the prediction model, the ARMA model and the traditional BP network are compared with GRU network in this paper. The prediction results indicate that GRU network is more effective than other prediction methods and has better engineering application value.