Feature representation based on a rearrangement of principal components method for fault diagnosis of multilevel inverter

Multilevel inverters are widely applied to medium and high voltage industrial fields because of their low harmonics and low requirements of withstand voltages. However, it will be more difficult for fault diagnosis of the power switching devices when the levels of the inverters increase. One of the key factors is the decreased sensitivity between the extracted features and the fault categories. Therefore, a Rearrangement of Principal Components (RePCs) method is proposed for the faulty data of the switching devices in multilevel inverters, which makes the features more representative and enhances the diagnostic efficiency of the inverter system. Besides, a diagnostic strategy is built based on the proposed method. Finally, the proposed feature representation method is verified by the experimental platform.