Application of Denoising Autoencoder in Intelligent Fault Diagnosis for Bearings under Varying Working Conditions

Bearing fault diagnosis is an important issue and useful tool to keep the equipment safe and reliable. With sustainable progress of deep learning and the use of large-scale industrial data, intelligent fault diagnosis based on the deep learning has become one of main research directions in this field. To improve the practicability of results, the denoising autoencoder (DAE) is applied to the bearings with complex working conditions. Two experiments are conducted to compare the performances of the DAE and the stack autoencoder (SAE). In the experiments, the data from one rotation speed are used for training, and the data from different conditions, including the changes of the rotation speed, load torque, and radial force, are used for test. The results indicate that only the rotation speed may influence the classification accuracy, and its changes may lead to the misclassification of bearing health states. For the changes of the other two conditions, the DAE and SAE have high diagnosis accuracies, and the former shows better results.