ABSTRACT
Convolutional neural networks (CNNs) have shown powerful superiority in feature representation learning for their intrinsic characteristics of locality and translation equivariance. In this chapter, we introduce CNNs for intelligent fault diagnosis and present a convolutional discriminative feature learning (CDFL) approach for induction motor fault diagnosis. We introduce the basic concepts of CNN and provide a brief review of the methodology of one-dimensional (1D) CNN used for fault diagnosis. Then we present a 1D-CNN-based CDFL method for motor fault diagnosis, which demonstrates how to design the unsupervised convolutional pooling architecture for fault diagnosis. The superiority of CNN structure has been proved through the experiments performed on a machine fault simulator. Compared with the state-of-the-art methods for fault diagnosis, CDFL method shows significant performance gains, and it is effective and efficient for induction motor fault diagnosis.
