Non-intrusive monitoring for condition-based diagnostic of electrical devices via machine learning algorithms

The objective of this thesis is to investigate the literature and test methods for Condition Monitoring of electrical appliances. In particular, the ongoing research area of Non-Intrusive Load Monitoring (NILM) is one of old but of continued interest nowadays. The need for low-cost hardware, the unreliability in sensor placement, and the ever-increasing amount of available data has made re-emerge the research conducted on NILM, as there is a strong interest in new methods of machine diagnostics. At the experimental level, in this thesis were used Machine Learning models capable of predicting taking as input the electrical signature of an appliance such as a dishwasher, with the aim of distinguishing and classifying the different washing phases which are characterized by spray-arms movements. This experimentation was conducted on two different dishwashers. By creating models able to recognize the washing phases for individual dishwashers, having as input the current. Once the models for the single machines were created, they were validated and used for predictions of the corresponding phases. Finally, some thresholds were identified with the current measurement of both electrical appliances, in order to predict with a single model the washing phases of the two dishwashers, having as input an aggregated current measurement.