A study on anomaly prediction method of machine tools - Feature extraction for anomaly prediction

This research aims to develop an anomaly prediction method only based on sensor data acquired from machine tools. In order to realize this aim, first of all, it is necessary to extract the feature to classify whether the machine tool is in an anomalous state or normal state. This paper shows the result of examining the feature extraction method of machine tool applying distribution and chaos theory to acquired sensor data. In the distribution method, an anomalous state and a normal state are compared by creating a frequency distribution diagram using the heat map. In chaos theory, the anomalous state and the normal state are compared by reconstructing attractors using Takens' embedding theorem. In addition, quantitative evaluation using topological geometry to the anomaly prediction of machine tools is considered and its result is shown. Both methods, the distribution and chaos theory, were confirmed their features for classifying the normal state and the anomalous state. And, we also confirmed the effectiveness of the quantitative analysis of the reconstructed attractors by using the method of topological geometry.