LAUSR

Partial discharge (PD) defects are initiated in the electrical machines over different life stages. The intensity of PD defects increases continuously with time, which may lead to insulation failure. In some cases, multiple PD defects may occur concurrently leading to a faster equipment failure.This article presents a methodology for the separation of multiple PD defects in low-voltage motors. The characteristics of the recorded PD signals have been investigated to estimate the severity of these defects. The cumulative energy (CE) function in the time domain has been calculated from the PD signals and its significant features, which include amplitude, range, skewness, kurtosis, autocorrelation functions, cross correlation functions, and width parameter of CE signals, have been compared for the separation of various defects. Therefore, 6-D feature space consisting of peak value, dispersion, symmetry, sharpness, similarity, and shape features of CE functions has been produced for the separation of multiple defects in the motors. Finally, the K-mean clustering classification algorithm has been adopted using significant features of CE functions to discover their clusters in the feature space. The proposed algorithm has been validated based on the adjusted Rand index (ARI) function. Thus, it has been observed that the proposed procedure is effective for the separation of mixed PD signals from multiple defects.