rel="stylesheet">
Creating and sharing knowledge in communications and information technology

Automating Electrical Grid Asset Inspection: From Current Challenges to Future Directions

Rocha, P. ; Lopes, F. ; Cruz, L. A. S. C.

IEEE Access Vol. 13, Nº , pp. 201392 - 201438, December, 2025.

ISSN (print):
ISSN (online): 2169-3536

Scimago Journal Ranking: 0,88 (in 2025)

Digital Object Identifier: 10.1109/ACCESS.2025.3636718

Download Full text PDF ( 7 MBs)

Downloaded 4 times

Abstract
Automated visual inspection systems are becoming increasingly important for improving the reliability, safety, and efficiency of electrical grid infrastructure. Despite significant advances in computer vision and artificial intelligence, inspection applications have not yet reached the performance of human experts. This review surveys current technologies, application domains, and methodological trends in automated grid asset inspection, while striving to uncover the root causes for the human-machine performance gap, advancing the state of the art through the analysis of common limitations and the formulation of future research directions. In addition, this work expands the scope of previous surveys beyond power lines to multiple electrical grid assets, acknowledging that most technologies and practices are shared across different application scenarios. The paper unique structure reflects the key dimensions for modeling an inspection system, ranging from inspected assets and tasks to model architectures and algorithms, including data acquisition platforms and sensors. This novel concept of defining a vision system through the intersection of independent domains promotes a system engineering perspective, which highlights often overlooked issues that result from improper subsystem decomposition, manifesting as dataset limitations, inconsistent validation practices, and divergence between metrics optimization and operational testing. To address these issues and related problems, this work proposes a novel inspection system design, combining embedded vision for data validation, with multimodal datasets and models, and an active learning strategy for continuous model tuning. Overall, this paper provides a roadmap for developing inspection systems that combine accuracy, scalability, and cost-effectiveness, ultimately supporting maintenance and sustainability goals in electrical infrastructure management.