Deep Robust Autoencoder based Framework for Bearing Fault Detection

Bearings are important components in many systems and widely used in rotating machineries. The condition of bearing influences the reliability and useful life of rotating machinery greatly. Thus, it is of significant importance to find early-stage bearing faults timely and effectively. Recently, due to the increasing of data size and computation capacity, more and more attentions have been drawn in machine learning-based methods for fault detection of bearings. A new bearing fault detection method based on deep autoencoder and support vector machine (SVM) is proposed in this paper. Signal processing and feature extraction is first performed to calculate envelope spectra of the measured raw vibration signals. Then, a robust deep autoencoder is applied to reduce the dimension of the spectra points; the output of robust deep autoencoder is then used to train an SVM for the purpose of fault detection. A case study using bearing data obtained from Case Western Reserve University Bearing Data Center is provided to show the effectiveness of our fault detection method.