2022 Volume 71 Issue 3 Pages 296-302
In industrial, agricultural, and chemical plants, rotating machinery is the most regularly used and important equipment, and its troubles and failures have a significant impact on production and quality. Therefore, the development of technology for detecting abnormalities and diagnosing the soundness of rotating machinery has been an important topic of study for many years. Recently, anomaly detection using unsupervised learning methods of machine learning has been studied in various fields. In this study, we attempted to develop an unsupervised anomaly detection method that does not require damage data in advance. Using a Variational Autoencoder (VAE), which is one of the machine learning techniques, we conducted an anomaly detection experiment through vibration experiments simulating some typical damages of rotating machines and investigated whether the system can recognize situations different from normal appropriately.
