Ai-Driven Fault Diagnosis And Predictive Maintenance In Electrical Power Networks

Abstract

The fast development of artificial intelligence (AI) and machine learning (ML) technologies have remarkably 
changed the way that fault detection, diagnosis, and control are used and designed in modern power systems. 
This review and meta-analysis presents extensive coverage of the published bill of materials from 2010 to 2024, 
including the results of over 200 peer-reviewed studies exploring deep learning, neural networks, fuzzy logic, 
support vector machines, and hybrid AI architectures as it relates to power system fault identification and 
automated remediation. This paper systematically examines the performance metrics such as accuracy, detection 
latency, false positive rates and scalability reported on various grid environments transmission networks, 
distribution systems and microgrids. It is shown, through meta-analysis, that DNN-based methodologies yield an 
average fault classification accuracy of 97.4% as compared with the conventional relay protection and statistical 
signal processing techniques which achieve less than 85% classification accuracy, a margin of 12% or greater. 
[13] These critical challenges include (i) data imbalance in fault datasets, (ii) computational overhead of real
time deployment, (iii) AI model adversarial vulnerability, and (iv) interoperability with existing legacy SCADA 
infrastructure. The review further identifies important research gaps in explainability, cybersecurity resilience 
and standardized benchmarking. Future research directions are suggested that highlight hybrid AI frameworks, 
federated learning for smarter distributed grids, and physics-informed neural networks to fuse domain knowledge 
with data-driven intelligence. The results confirm the transformational potential of AI-enabled automation for 
backing reliable and resilient and self-healing power grids that are aligned with smart grid and Industry 4.0 
objectives.