Investigating Artificial Intelligence Applications in Electrical System Monitoring and Fault Diagnosis

Abstract

This review paper provides a comprehensive evaluation of the paradigm shift in electrical system monitoring and fault diagnosis, tracing its evolution from manual inspections and rule-based expert systems to dynamic, self-learning artificial intelligence (AI) models. As modern power grids face heightened complexity due to the bidirectional power flows of Distributed Energy Resources (DERs), traditional threshold-based and signal-processing techniques increasingly struggle. To address these challenges, advanced deep learning architectures—including Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks—are deployed alongside advanced computer vision frameworks to extract multi-resolution temporal and spatial features from operational data. Furthermore, non-contact sensing paradigms such as Infrared Thermography and airborne acoustic emission allow for non-invasive anomaly capture, while Graph Neural Networks (GNNs) optimize fault localization across dynamically reconfiguring spatial topologies. To overcome data scarcity and ensure consistency, Physics-Informed Neural Networks (PINNs) embed differential mathematical equations directly into loss functions, delivering significant computational speedups over traditional simulation tools. This review also explores intelligent domain knowledge integration within Dissolved Gas Analysis (DGA) for transformer health monitoring, alongside privacy-preserving Federated Learning strategies that enable decentralized edge processing. Finally, the critical need for operational transparency is addressed via Explainable AI (XAI) frameworks like LIME and SHAP, which demystify complex "black-box" models. By synthesizing algorithmic innovations, practical bottlenecks, and diagnostic performances across sub-domains, this paper outlines a clear roadmap for the deployment of resilient, interpretable, and autonomous AI architectures within modern, next-generation smart grids.