Empirical-Cumulative-distribution-based Outlier Detection
Assembly Line
Evaluation of Anomaly Detection Algorithms on Circuit Breaker Lifetime Test Data
Industrial condition monitoring based on AI is often hampered by the lack of properly labeled data. One possible solution to this problem is to use anomaly detection algorithms. In this paper we test four machine learning anomaly detection algorithms (DeepSVDD, GMM, KNN, ECOD) on endurance test condition monitoring data recorded from two medium voltage circuit breakers in a long-term laboratory experiment. The paper shows the suitability of these algorithms for the early detection of faults in a deteriorating mechanical system. We benchmark the results against the interpretation of traditionally used metrics. The comparison leads to the preferable alternative to those metrics.