Prediction of Breakdown Time of Insulation Material for Use in Medium-Voltage Inverter-Fed Marine Propulsion Motor Applications

Large voltage transients and high frequencies from power electronic inverters are the major cause and concern for insulation failure within inverter-fed machines. This is especially true as more wide bandgap devices and electric motor drives penetrate various applications, including renewable integration. To understand the impact of high-voltage and high-frequency of emerging power electronic inverters on insulation lifetime, a medium-voltage H-bridge inverter testbed was built to stress Kapton HN (polyimide) films. These films of different thicknesses were subjected to voltage endurance tests at different inverter voltages and frequencies to evaluate their breakdown time. Three different machine learning (ML) regression algorithms—linear, random forest, and support vector machine (SVM)—were employed on time to breakdown data to predict the insulation lifetime. Results show that random forest regression performs best and gives a reasonably accurate prediction of time to failure of Kapton HN film given its thickness at any inverter voltage and switching frequency tested. In addition, a closed-form equation of insulation breakdown time is derived from the linear regression to predict breakdown time for various insulation thicknesses, under varying inverter voltage, and inverter switching frequency.