Two-Phase Anomaly Detection and Confidence Analysis in Electromechanical Systems with Implicit Operating Condition

Abstract

Many studies related to failure analysis have been carried out utilizing condition-based maintenance (CBM) to ensure quality monitoring in electromechanical systems. Different operating conditions of the system's state affect the sensor signal, the signal may be different according to operating conditions. There may be insufficient data set about operating conditions. If the signal is implicit in information about operating conditions, it is challenging to accurately determine whether the current system’s status is normal or abnormal and to which operating conditions. To tackle these issues, this study proposes a two-phase anomaly detection procedure consisting of (i) if the operating condition information is not explicit, the data is preprocessed for the operating conditions and generate boundary for normal data for each operating condition. (ii) We propose an alternative method like a confidence interval, which uses a different term, confidence score, which considers the probability distribution of the data under operating conditions and determines under which operating conditions a test instance occurred. The confidence score indirectly measures confidence in the decision about abnormality. The proposed method is demonstrated with case studies of (i) the elevator system and (ii) the rolling-element-bearing vibration system.