Fault Diagnosis of Asymmetric Cascaded Multilevel Inverter using Ensemble Machine Learning

Cascaded Multi Level Inverters (CMLI) are widely used in a wide range of high-power industrial drives and for integrating solar PV system. Asymmetric Cascaded Multilevel Inverter (ACMLI) minimizes the components and produces an output voltage with the highest number of levels with reduced Total Harmonic Distortion (THD) when compared to Symmetric Cascaded Multilevel Inverter (SCMLI). ACMLI comprises of more semiconductor devices and thus reliability is of major concern. Efficient, high speed and precise fault detection is required for ACMLI to reduce failure rates and avoid unplanned shutdown. Thus, Fast and accurate fault diagnosis scheme is necessary for increasing the reliability of the system. Various voltage signals from the CMLI various fault conditions are used as fault features. Fault diagnosis method for ACMLI based on probabilistic principal component analysis (PPCA) and Ensemble Machine Learning (EML) is presented. PPCA is used to optimize data and reduce the size of fault features. Finally, an EML classifier combining Support Vector Machine (SVM), K-Nearest Neighborhood (KNN) and Decision Tree (DT) is employed to diagnose the various fault open circuit faults. The proposed fault diagnosis method is validated using an ACMLI experimental setup. The simulation and experimental results show that using EML improves the accuracy of 99.32% in diagnosing the fault location.