A sparse auto-encoder-based deep neural network approach for induction motor faults classification

This paper presents a deep neural network (DNN) approach for induction motor fault diagnosis. The approach utilizes sparse auto-encoder (SAE) to learn features, which belongs to unsupervised feature learning that only requires unlabeled measurement data. With the help of the denoising coding, partial corruption is added into the input of the SAE to improve robustness of feature representation. Features learned from the SAE are then used to train a neural network classifier for identifying induction motor faults. In addition, to prevent overfitting during the training process, a recently developed regularization method called "dropout" which has been proved to be very effective in neural network was employed. An experiment performed on a machine fault simulator indicates that compared with traditional neural network, the SAE-based DNN can achieve superior performance for feature learning and classification in the field of induction motor fault diagnosis.