Fault Diagnosis for Time Series Signal based on Transfer Learning in Time-Frequency Domain

Abstract

Time series contributed by sensor signal can be used for fault diagnosis, and machine learning is adopted to identify the causes of failure and the relevant factors in the time-frequency domain. However, the lack of labeled data, incredibly faulty data in various conditions, is one of the significant challenges when applying machine learning approaches. To reduce the barrier of applying those approaches, this study investigated the use of transfer learning. A high accuracy of nearly 95% for classification without the labels in training is found. There is potential research direction in unsupervised domain adaptation and domain generalization.