The development and spread of Machine Learning methodologies have also involved the field of anomaly detection, particularly focused on fault detection. This is one of the main goals of Industry 4.0, as it is necessary to optimize repair time and cost. In this regard, Machine Learning is needed to identify precursor features of possible failures that would be difficult for a human operator to discern. However,compared to Deep Learning methodologies, these cannot be fully automatic because of the need to make choices about the features identified by the system. This paper aims to propose a Machine Learning-based system for detecting electrical anomalies attributable to malfunctions in connected industrial machinery. Specifically, the proposal is a fusion of unsupervised learning and traditional methodology to minimize human intervention while maintaining an explainable, white-box approach, contrary to proposals based on Deep Learning. The results demonstrated better performance to techniques of the state of the art.

19^th IMEKO TC10 Conference "Measurement for Diagnostics, Optimisation and Control to Support Sustainability and Resilience"