Real-Time Bearing Fault Diagnosis of Induction Motors with Accelerated Deep Learning Approach

This study introduces an efficient deep neural network based bearing fault detection of induction motors. An approach to accelerate and compress convolutional neural networks (CNN) is the basis of the proposed method. As the main advantages, the proposed algorithm is 1) directly applicable to raw data, 2) highly accurate, 3) non-time consuming, 4) applicable to different types of electric machines, 5) merges feature extraction and detection into a single machine learning, and 6) reduces computational burden of conventional CNN. To address and verify the proposed method, the experimental dataset of Case Western Reserve University (CWRU) bearing data center is used. The results show the impressive capability of the proposed CNN method high precision fault detection, comparing with conventional CNN as deep-based structure method and support vector machine (SVM), artificial neural network (ANN), and learning vector quantization (LVQ) as shallow-based structure methods.