Deep Learning in Fault Diagnosis of Induction Motor Drives

Application of machine learning techniques in fault diagnosis of induction motor is gaining popularity over the years. However, major challenge is the selection of the handcrafted statistical features, which limit performance of the classifiers immensely. Deep learning, a feature representation based method opens up a new horizon, where feature descriptors are extracted from the raw signals. This paper has exploited merit of deep learning and reports a preliminary findings in motor fault detection using novel semi 2D Convolution Neural Networks. The experimental results of the proposed approach show 3-10% enhanced performance compared to the conventional feature engineering based methods. The computation is relatively faster than 2D Convolution Neural Networks and the generalization of the results is promising for real life applications.