Detection and classification of partial discharge signals using Hermite transform and expectation-maximization approach

This paper proposes a novel approach for detecting and classifying partial discharge transients in power cable signals recorded from substation environments, based on analysis in the Hermite transform domain. Partial discharges are short-duration, fast-changing phenomena commonly observed within power cables used in substations and communication systems. Accurate partial discharge detection plays a critical role in predictive maintenance of power transmission and distribution networks, as early identification of such events can prevent costly insulation failures. While traditional partial discharge detection techniques rely on energy thresholds, dictionary projections, Fourier transform or time–frequency representations such as spectrograms, recent methods increasingly emphasize sparse representations and entropy-based analysis to improve sensitivity and robustness. In line with this direction, the research introduces an approach based on the Hermite transform for transient events detection, due to the inherent similarity between Hermite functions and the shapes of partial discharge-related transients. To enhance sparsity and representation quality, a scaled and optimized Hermite transform is employed. Concentration measures such as the ℓ1-norm and Rényi entropy, calculated in the Hermite domain, are used to classify transient signals. The proposed method is experimentally validated using real-world signal sets containing three distinct classes of transients recorded from substation equipment. Results demonstrate that the combination of the scaled Hermite transform, concentration measures, and statistical parameters enables effective classification when integrated with the Expectation-Maximization algorithm. Furthermore, the procedure can be automated, and techniques such as the Parzen window method can be employed to further improve classification performance.