Comparative Analysis of the Fault Diagnosis in CHMLI Using k-NN Classifier Based on Different Feature Extractions

Abstract

Recently, the development of multilevel inverters has great progress in many Industrial applications because of its high efficiency and low switching frequency control methods. To improve the fault diagnosis accuracy, A k-Nearest Neighbors (k-NN) algorithm based on the different feature extractions is used. In this paper, the Principle Component Analysis (PCA) and Probabilistic Principle Component Analysis (PPCA) are used for the feature extraction. Firstly, the data from the output voltage signals under different fault conditions of the Cascaded H-Bridge Multilevel Inverter (CHMI) is optimized by using different feature extractions. Then, the k-NN classifier is used to identify the accurate fault location to diagnosis the fault. Finally, the FFT analysis also applied to evaluate the proposed k-NN technique. To validate the proposed technique the experimental setup has built in the laboratory and verify the simulation results. Based on the experimental and simulation results, the proposed k-NN technique has better performance when the PPCA feature extraction is used.