Acoustic Analysis Framework for Detecting Internal Electric Discharges in Glass Insulators under Fluctuating Voltage

Abstract

Partial discharges (PDs) within high-voltage glass insulators are critical indicators of insulation degradation that can compromise the operational reliability and safety of electrical power systems. Conventional PD detection methods, primarily based on electrical measurements, often face limitations under fluctuating voltage conditions and in environments with significant electromagnetic or mechanical noise. Acoustic-based diagnostics offer a non-intrusive and complementary approach, leveraging the mechanical vibrations generated by PD events to detect and localize insulation faults. This research presents a comprehensive acoustic analysis framework for detecting internal electric discharges in glass insulators subjected to variable applied voltages. The framework integrates high-fidelity acoustic signal acquisition with advanced signal processing techniques, including discrete wavelet transform (DWT) for multiresolution analysis, coupled with radial basis function neural networks (RBFNNs) optimized via particle swarm optimization (PSO) for robust classification. Laboratory-scale experiments were conducted on glass insulators under controlled fluctuating voltage profiles, capturing transient acoustic emissions from multiple PD events while simulating industrial noise conditions. The results demonstrate that the proposed framework reliably identifies PD events with high sensitivity, achieving classification accuracy above 94%, even under significant ambient interference. Time-frequency analysis revealed consistent patterns in wavelet coefficients corresponding to PD signatures, enabling precise temporal localization and differentiation from non-PD acoustic events. Comparative evaluation against traditional electrical PD detection methods highlights the framework's superior performance under variable voltage and noisy environments, emphasizing its practical relevance for predictive maintenance and risk mitigation. The study further identifies limitations associated with acoustic signal propagation, such as sensitivity to insulator geometry and mounting conditions, and proposes strategies to mitigate these effects, including adaptive thresholding and multi-sensor configurations. By combining rigorous signal processing, machine learning classification, and empirical validation, the research provides a scalable and non-intrusive methodology for early detection of insulation degradation, contributing to enhanced reliability, reduced maintenance costs, and extended service life of high-voltage insulation systems. The proposed framework establishes a foundation for future development of real-time monitoring systems and intelligent diagnostics for glass insulators and other critical high-voltage components, offering actionable insights for utilities and industrial stakeholders aiming to optimize operational safety and predictive maintenance strategies (Nattrass, 1988).