Abstract

Fault diagnosis plays an important role to improve maintenance efficiency. The industry faces the challenges to collect history data that include all type of failures. To overcome the limitation of conventional diagnosis approaches, which misclassify new types of faults into existing categories from training, the unsupervised Gaussian mixture model (GMM) and the semi-supervised GMM diagnosis frameworks are presented in this paper for effective detection on new fault categories. For the unsupervised GMM framework, the component number is known and the hard assignment is applied to classify the new types of faults. For the semi-supervised GMM framework, the component number can be auto selected, and the soft assignment is able to first detect whether new types of faults occur and further categorize them in detail via the GMM update. The effectiveness of the two fault diagnosis frameworks is testified on an industrial fault simulator of rotary machine. Compared with existing hard clustering approaches, the semi-supervised GMM framework is able to achieve an average diagnosis accuracy of 99.3% without new fault categories and it can also classify new fault categories with diagnosis accuracy of 94.0%.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call