Machine Learning Based Prognostics Techniques for Power Equipment: Comparative Study

Power distribution systems operate under thermal and mechanical stresses after their installation. Improper handling of such equipment, abrupt surges, overloading, loose connections and negligence in maintenance not only result in electricity short falls and power losses but also deteriorates the remaining useful life of the equipment. The scenario emphasizes the importance of asset management for power devices in order to utilize equipment up till their maximum service life. Indication and identification of fault and its location and intensity, prior to the occurrence of fault, is a proactive approach for fault avoidance and monitoring of remaining useful life. Over the years, statistical approaches have been used for the prediction and forecasting. The aim of this study is to investigate computationally intelligent techniques suitable for the prognosis of remaining useful life of power devices. This paper provides state-of-the-art review of some selected ML techniques in connection with prognostics relevance, and discusses the significant features connected with modeling development for prognostics techniques. The paper also discusses the strength and limitations of the selected ML techniques compared with other available dynamic modeling approaches.