Intelligent condition monitoring of power cables using advanced machine learning models
Year: 2025
Type: article
Abstract: Power cables are susceptible to ageing phenomena, which can lead to unexpected failures and catastrophic damage. Consequently, investigating the degradation of power cables and predicting their health condition is crucial for avoiding irreversible destruction, power delivery stoppage, and power quality problems in power systems. This paper offers an intelligent health monitoring framework that uses advanced machine learning (ML) techniques for various power cables, which is trained and validated on a new dataset of 15 kV and 20 kV cross-linked polyethylene (XLPE) power cables, incorporating features such as partial discharge, age, visual inspection, and neutral corrosion. To identify the most effective algorithm, 18 ML techniques were systematically benchmarked, with hyperparameters for each model tuned via Bayesian optimisation to ensure a fair comparison. The results indicate that boosting-based models consistently outperform linear, tree-based, and neural network-based models, achieving accuracies higher than 98 %. Furthermore, the framework's generalisation capabilities are tested on a new dataset of a different cable, i.e., 138 kV ethylene propylene rubber (EPR) power cables, showcasing high F1-scores exceeding 98 % in identifying health indexes. Finally, to evaluate its robustness under realistic imperfections, the framework was subjected to challenging stress test scenarios—including multi-feature missing measurements and severe class imbalance—demonstrating the CatBoost model’s high resilience with accuracies consistently above 95 % for XLPE and above 75 % for EPR power cables. The proposed framework forms the core of intelligent, automated systems for condition-based asset management in modern electric network. By predicting a reliable health index for both new and existing assets, it serves as a robust tool for optimising capital-intensive expansion planning, network reconfiguration, and maintenance scheduling. Ultimately, this validated framework could serve as the foundation for developing a digital twin for power cables, enabling a shift from reactive maintenance to predictive, intelligent asset management.
Source: Results in Engineering
Authors: Mohsen Abdolahi, Wenjuan Song, Mohammad Yazdani-Asrami
Institution: University of Glasgow
Language: English
Subfield: Materials Chemistry
Field: Materials Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $1,940
Award: G377745559
OpenAlex ID: W7106323877