Unsupervised Anomaly Detection for a Machine Whose Vibration Pattern Changes

Abstract

In recent years, there have been many activities to solve problems on manufacturing utilizing digital technologies such as IoT and AI.We have been working on research and development of vibration abnormality detection technologies for mechanical devices with various motion patterns, such as robotics arm and printing machines. Our previous method can quantitatively express the degree of motion abnormality of the observation target without performing preprocessing work for extracting vibration data. However, since this method is based on supervised learning which requires both normal and abnormal data, there is a problem that a highly accurate discrimination model can not be built when the abnormal data is insufficient. To address this problem, we propose a new method to improve our traditional method using unsupervised learning. In addition, we also report on the results of the evaluation experiment using the actual machine and show the effectiveness of our proposed method.