Normal Data-Based Motor Fault Diagnosis Using Stacked Time-Series Imaging Method

In most engineering systems, the acquisition of faulty data is difficult or sometimes not feasible, while normal data are secured. To solve these problems, this paper proposes an fault diagnosis method for electric motor using only normal data with self-labeling based on stacked time-series imaging method. Since only normal data are used for fault diagnosis, a selflabeling method is used to generate a new labeled dataset based on pretext task. To emphasize faulty features from nonstationary faulty data, stacked time-series imaging method is developed. The overall procedure includes the following steps: (1) transformation of a one-dimensional current signal to a two-dimensional image in time-domain, (2) adding sparse features with sparse dictionary learning, ( 3) stacked images through every window size, and (4) fault classification based on convolutional neural network (CNN) and Mahalanobis distance. Transformation of the time-series signal is based on recurrence plots (RP). The proposed RP method develops from sparse dictionary learning that provides the dominant fault feature representations in a robust way. To verify the proposed method, data from real-field manufacturing line is used.