A Physics-Aware Artificial Intelligence Framework for Secure Energy Management in Mission-Critical Defence Power Systems

This paper addresses the reliable fault detection in power distribution networks, which has become increasingly challenging due to the growing disparity between densely interconnected urban feeders and long, high-impedance rural networks. Conventional protection schemes and purely data-driven artificial intelligence methods often exhibit limited sensitivity to weak fault signatures and poor robustness under transient-rich operating conditions. This paper proposes a unified physics-aware artificial intelligence framework for fault detection that is applicable to both urban and rural distribution systems. The proposed approach embeds electrical behaviour constraints directly into the feature representation, enabling physically consistent decision-making while retaining the adaptability of learning-based inference. A comprehensive analytical formulation links feeder dynamics, normalized voltage and current measurements, constrained feature embedding, and probabilistic fault inference within a single framework. The detection logic further incorporates persistence-based decision rules to suppress non-fault disturbances and measurement noise. Extensive MATLAB/Simulink simulations are conducted on representative urban and rural feeders under low-impedance faults, high-impedance faults, marginal fault conditions, and noise-stressed scenarios. The results demonstrate accurate and timely fault detection with improved selectivity and reduced false-alarm rates compared to conventional threshold-based and purely data-driven approaches. Notably, high-impedance rural faults that remain undetected by classical protection schemes are consistently identified through gradual probability accumulation governed by physical feasibility. The proposed framework provides a scalable and interpretable solution for intelligent fault detection in modern power distribution networks.