Research Projects

High-voltage cable failures cause economic losses and threaten grid stability. The current diagnostic methods require multiple distributed sensors, centralized processing, and continuous high-bandwidth communication, making them cost-prohibitive for widespread deployment. The fundamental challenge lies in separating multiple overlapping fault signals (partial discharge, corona, arcing) from a single measurement point, an under-determined blind source separation (BSS) problem that conventional digital signal processing cannot solve efficiently. This research addresses the critical need for low-cost, real-time, edge-deployable diagnostic solutions by leveraging the sparse, event-driven computation of spiking neural networks (SNN). The scientific objectives include developing a single-sensor multi-fault signal separation methodology using neuromorphic computing principles. It involves establishing optimal time-frequency preprocessing pipelines, designing and validating SNN architectures capable of BSS under under-determined conditions, and demonstrating real-time neuromorphic hardware implementation. The proposed work is expected to give better inference capabilities and to quantify performance gains in accuracy, power efficiency, and cost-effectiveness compared to the conventional approaches. The main experimental framework encompasses: 1) controlled multi-fault signal generation using systematic fault induction in high voltage cables to establish ground-truth data; 2) Two source SNN architecture development implementing supervised learning with Bayesian uncertainty quanftification; 3) Development of multi source speration (more than two) using iterative technique on the developed two source SNN based BSS; 4) neuromorphic hardware validation with real-time performance evaluation; and 5) field deployment validation through pilot testing in operational substation environments. The significance to the research field spans both fundamental understanding and practical applications. In terms of fundamental understanding, this work advances signal processing theory by demonstrating SNN effectiveness for under-determined BSS problems, establishes new paradigms for industrial edge intelligence, and proves SNNs' viability for complex real-world signal analysis. For practical applications, this research enables grid modernization through widespread deployment of intelligent monitoring systems with higher cost reduction potential. Further, it provides scalable diagnostic solutions for remote and resource-constrained environments and enhances power system reliability. This research represents a convergence of advanced computational neuroscience, signal processing theory, and critical infrastructure needs, positioning neuromorphic computing as a transformative technology for industrial diagnostics and smart infrastructure applications.