GMM-Aided DNN Bearing Fault Diagnosis Using Sparse Autoencoder Feature Extraction

Abstract

Deep learning techniques are gaining popularity due to their ability of feature extraction, dimensionality reduction, and classification. However, one of the biggest challenges in bearing fault diagnosis is reliable feature extraction. When using the bearing fault vibration spectrum, the deep neural network (DNN) model can learn the relationships in data that are unrelated to the task. In this work, a simple approach to bearing fault diagnosis using the elimination of unrelated data artifacts for DNN is proposed. The proposed fault diagnosis pipeline is explained and the comparison with popular fault diagnosis methods is performed.