Fault Prediction in Energy Systems: Telemetric Data and Machine Learning Approaches

The uninterrupted and reliable provision of energy is of paramount importance for the sustainability of economic activities and the welfare of society. The intricate structures formed by numerous components within the energy infrastructure create a basis for failures to yield severe consequences. This study aims to develop predictive maintenance models using machine learning algorithms by examining telemetry data, such as voltage, rotational speed, pressure, and vibration obtained from machinery, in conjunction with historical failure and maintenance records. The dataset utilized in this study comprises hourly telemetry information, including voltage, revolutions per minute, pressure, and vibration, collected from 100 distinct machines throughout the year 2015. Furthermore, supplementary information, such as the machines past failure occurrences, maintenance logs, and technical specifications, has been considered in the development of the models. In this study, timedependent patterns derived from sensor data, along with historical maintenance and failure information, have been integrated and analyzed using machine learning algorithms, namely Stochastic Gradient Descent Classifier (SGDClassifier), eXtreme Gradient Boosting (XGBoost), and Histogram-based Gradient Boosting (HGBClassifier). The results have been analyzed based on performance metrics such as Precision, Recall and F1-Score. Notably, the XGBoost and HGBClassifier algorithms demonstrated superior performance in early failure detection, particularly within 24-and 48 -hour prediction windows, achieving high F1-Score values. Consequently, this approach aims to transcend the traditional reactive maintenance paradigm, thereby enhancing operational efficiency and preventing unforeseen downtimes.