Characterization of the impact of coating defects on ceramic insulator efficiency: Machine learning approaches

• Published in: ELECTRIC POWER SYSTEMS RESEARCH
• Main Authors: Salem, Ali Ahmed; Alawady, Ahmed Allawy; AL-Gailani, Samir A.; Amer, Abdulrahman Ahmed Ghaleb; Al-Ameri, Salem Mgammal; Abd-Rahman, Rahisham; AL-Gailani, Nasir Ahmed
• Format: Article
• Language: English
• Published: ELSEVIER SCIENCE SA 2025
• Subjects: Engineering
• Online Access: https://www-webofscience-com.uitm.idm.oclc.org/wos/woscc/full-record/WOS:001349283600001
• ISSN: 0378-7796; 1873-2046
• DOI: 10.1016/j.epsr.2024.111158

Porcelain insulators are essential for the safety and efficiency of electrical systems, yet their hydrophobicity needs improvement through anti-pollution flashover coatings. However, these coatings are prone to damage from factors such as pollution. This study aims to investigate the electrical characteristics, specifically flashover voltage and leakage current, of insulator strings under seven different RTV-coating damage scenarios, as well as fully coated and bare scenarios, in the presence of pollution and humidity. Machine learning models were utilized to estimate the condition of insulators under these coating damage scenarios. Statistical features were extracted from the leakage current (LC) and flashover voltage (Uf) of the insulators using signal analyzer techniques. These features were then used to train machine-learning algorithms, including Support Vector Machines (SVM), Multi-layer Perceptron (MLP), and the Random Under Sampling with Adaptive Boosting algorithm (RUSBoost). A Self-Organizing Map (SOM) was employed to cluster the flashover voltage and leakage current datasets before training the machine learning models. The results indicated that risky insulator conditions were more prevalent in bare insulators and those with severe RTV coating deterioration, such as scenarios S5 and S8. The proposed models effectively identified and classified the conditions of insulator strings under various coating damage scenarios, demonstrating strong performance metrics. The use of statistical features improved classifier performance by up to 6.37 %. This approach offers a robust, automated method for the maintenance and monitoring of ceramic insulators, potentially reducing inspection time and enhancing the reliability of electrical infrastructure.