Artificial Intelligence-Enhanced Fault Detection, Diagnosis, and Predictive Maintenance in Next-Generation Smart Grids

Abstract

The electrification of the electric power systems into the next generation of smart grids has propelled fault detection, diagnostic and predictive maintenance to new heights in terms of necessity in terms of enhancing reliability and resilience. Traditional methods, which are based on threshold limit values and regular surveys do not usually pick up non-linear trends, undetectable anomalies, and dynamic grid behaviour. The proposed study examines how new artificial intelligence (AI) techniques such as deep learning, graph neural networks (GNNs) and hybrid ensembles, and transformer-based architectures can enter fault detection and predictive maintenance systems. AI-based methods were able to increase accuracy in detecting faults and locating faults as well as estimating transformer life by 83 percent compared to conventional practices using synchrophasor measurements, SCADA data, and dissolved gas analysis (DGA) records. In addition, the application of edge computing enabled more than 50% reduction of detection latency, and federated learning allowed privacy-preserving multiple-institution model training at several substations. The predictions were validated through digital twin simulations, in a manner that was both explainable and trustworthy as a result of correlating AI results with real-world grid models. The results confirm that analytics based on artificial intelligence lead not only to a more efficient operational process and asset life but make the shift toward a more resilient, decentralized, and sustainable ecosystem of the smart grid.