Wavelet-based Noise Extraction for Anomaly Detection Applied to Safety-critical Electronics at CERN

Due to the possible damage caused by unforeseen failures of safety-critical systems, it is crucial to maintain these systems appropriately to ensure high reliability and availability. If numerous units of a system are installed in various areas and permanent access is not guaranteed, remote, data-driven condition monitoring methods can be used to schedule maintenance actions and to prevent unexpected failures. Thereby, failure precursors identified by unsupervised anomaly detection algorithms can be used to detect system malfunctions or to assess the systems condition. The anomaly detection process presented in this paper proposes a novel integrative combination of noise extraction using wavelet transforms and unsupervised algorithms to improve the detectability of a broad variety of anomalies for safety-critical electronics. Here, the performance of this modular process is demonstrated by identifying outlying data samples in datasets generated by the CERN Radiation Monitoring Electronics (CROME) system.