부산대학교 도서관

Electric vehicles (EVs) utilize permanent magnet synchronous motors (PMSM) as traction motors due to their advantages like long lifetime, lower maintenance cost, high speed, and noiseless operation. PMSMs are subjected to operating stresses and high temperatures which lead to faults in them. Insulation breakdown results in faulty stator windings and are prevalent. They begin as short circuit in internal turns (ITSC) and when propagated further can lead to other types of faults. Hence, timely identification of ITSC is essential. Machine learning algorithms can detect relationships among features which can be used for fault detection. In this paper, machine learning algorithms are used for identifying early-stage inter-turn short circuit faults under various operating conditions using simulation dataset. Three-phase current signals are used for deriving features like root mean square and negative sequence current, hence no additional sensors are required to detect the fault. Support Vector Machine, Random forest and KNN are trained and the accuracy obtained is 90%, 95.83% and 100% respectively. The model is then validated using simulation data to detect fault under unknown operating conditions and the accuracy obtained is 97.5%.