Machine Fault Classification Using Deep Belief Network

Feature selection is critical to the success of machine fault intelligent diagnosis. However, different feature extraction methods may lead to different diagnostic results. It increases the difficulty in feature selection as well as the diagnostic uncertainty. A deep belief network (DBN), composed of multiple layers of Restricted Boltzmann machine (RBM), is one of the typical deep learning methods, which can obtain more abstract concepts through layer-wise feature learning to discover the data structure. Therefore, a DBN-based fault soft-max classifier was constructed. Bearing experiments were conducted, and the proposed scheme was applied for bearing fault classification. By learning from the raw vibration signal directly and layer-wise training, the DBN could categorize the raw data into the corresponding classes effectively and automatically, which enhanced the intelligent fault diagnosis.