A Novel Application Approach for Anomaly Detection and Fault Determination Process based on Machine Learning

Research on the fault is a systematic project, which mainly includes fault detection, identification, and fault handling. The system fault is often manifested in the form of abnormal data (anomaly). However, the traditional fault analysis, such as fault tree analysis (FTA), cannot explore the potential system anomaly. Therefore, it is necessary to combine the FTA with anomaly detection together. Currently, machine learning has been applied to the anomaly detection field widely. But simultaneously, it is unable to determine whether the anomaly is caused by the system fault. Focusing on fault detection and identification, this paper represents a novel approach that associates the application of machine learning in anomaly detection with the FTA organically to determine the characteristics of the anomaly. The effectiveness of the approach is verified through an illustrative example. Finally, we introduce the next stage of work about fault handling.