Dielectric property prediction of coated high voltage glass insulators based on experimental analysis

• Published in: Progress in Organic Coatings
• Main Author: Salem A.A.A.
• Format: Article
• Language: English
• Published: Elsevier B.V. 2025
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85212551895&doi=10.1016%2fj.porgcoat.2024.109003&partnerID=40&md5=8203ad7ada5c755659910e4c88316b9b

High-voltage insulators face serious challenges, including flashovers caused by environmental pollutants. Coatings are an effective approach that may reduce these dangers, although their efficacy varies depending on the surrounding environment. This paper conducts an experimental study on the properties of coated glass insulators under superhydrophobicity coatings of polyurea resin loaded with varying concentrations of micro-nanoscale titanium dioxide (TiO2) (polyurea/TiO2). The study constructs 633 data points for 6 variables, subsequently training four hybrid machine learning models. A data-driven hybrid model (COA-ANN, COA-SVM, COA-RF, and COA-XGBoost) combined with the Coati optimization algorithm (COA) and ANN, SVM, RF, and XGBoost are proposed to predict the dielectric capacity, dielectric strength, and dielectric loss of coated glass insulators in the presence of humidity. We prepared five datasets to evaluate the effects of multiple model outputs. SHapley Additive Explanations (SHAP) were implemented for feature analysis. The results revealed that using the COA algorithm improved the proposed models' prediction accuracy due to hyperparameter optimization. Meanwhile, the hybrid COA-XGBoost model outperforms other models in predicting dielectric capacity, dielectric strength, and dielectric loss in various performance indicators, resulting in high correlations between the actual and predicted data. According to SHAP results, humidity and water contact angle have a distinct influence on the properties of coated insulators. Due to their high accuracy prediction, the developed models are promising for accurately predicting the coated insulator properties. © 2024 Elsevier B.V.