Automatic creation method of extraction condition of sensor data for anomaly detection based on machine learning

Abstract

Anomaly detection methods are becoming important for condition-based maintenance. This paper proposes an automatic creation method of extraction condition of sensor data for an anomaly detection system. To achieve high performance for anomaly detection, it is important to extract a machine steady state in the learning phase and the anomaly detection phase. This creation method can generate extraction condition candidates automatically using the observed frequency and expected frequency of machine sensor data. Experimental results demonstrate that the proposed method can create extraction condition candidates more accurately than the conventional method does.