Simple continuous optimal regions of the space of data

The definition of efficient programs for both maintenance and optimization is a struggling task in many industrial sectors. In this context, data analysis can significantly improve the state-of-the-art techniques, employed, for instance, to determine if a particular component or product is showing an anomalous behavior with respect to a defined nominal state. In fact, through the analysis of data collected on field, it is possible to detect optimal operating regions and to detect anomalies in advance. In this context, we propose a multi-purpose algorithm for unsupervised or semi-supervised learning in order to determine a simple continuous region of points. This region can be adopted in order to describe a component or a product nominal behavior and can be used in order to detect anomalies which are outside it. Such a region can be defined adopting a finite ensemble of thresholds, whose value can be physically interpreted. In order to show the effectiveness of our approach, the proposed method has been tested in an Anomaly Detection problem concerning Predictive Maintenance, exploiting data coming from a naval vessel, characterized by a combined diesel-electric and gas propulsion plant.