Qualitive Predictive Model of the Insulation Condition of HV Electrical Machines through a Combination of Diagnostic Techniques and NNs

Electrical Machines (EMs) hold significant importance in electricity generation and industrial applications. One of the most significant characteristics that an EM must have, is the reliability and as much as possible long lifetime. Insulation Resistance (IR), Polarization Index (PI) and Partial Discharges (PDs) Measurements are three of the most important diagnostic factors for the insulation system of EMs. This paper studies the aforementioned factors and their respective techniques combined with another two operational characteristics of an EM (years of operation and times of failures) and suggests a qualitive predictive model for the condition of their stator insulation with the use of Neural Networks (NNs). By leveraging NNs and evaluation metrics, it offers insights into the EM's insulation system state of health (Excellent, Good, Acceptable, Bad) and suggests appropriate actions for maintenance and further inspection. The model could be used for any electrical machine regardless of its power and voltage and it is capable to predict the insulation condition of the EMs with Mean Squared Error (MSE) of 0.0812.