AC flashover voltage prediction of silicone rubber insulators under mechanical defects and non-uniform pollution using experimental, machine learning, and numerical approaches

This study investigates the combined effects of ring non-uniform (RNU) and ring–longitudinal non-uniform (RLNU) pollution, together with mechanical defects, on the AC flashover voltage of silicone rubber insulators under varying humidity conditions. Moving beyond previous studies focused on isolated pollution or defect factors, this work experimentally evaluates their combined influence under realistic environmental conditions. Non-uniform pollution significantly affects flashover performance. Increasing the longitudinal non-uniformity index (P) increases the flashover voltage, whereas increasing the polluted radius under RNU conditions (R) reduces it, indicating that more uniform contamination weakens insulation performance. Mechanical defects further degrade dielectric characteristics by intensifying local electric fields and promoting earlier arc initiation, reducing flashover voltage by up to 31.9% compared with intact insulators. A novel empirical model is proposed to estimate flashover voltage as a function of salt deposit density (SDD), relative humidity (H), and the pollution non-uniformity index (P). Artificial intelligence-based prediction models, including Artificial Neural Network (ANN) and Multilayer Perceptron (MLP), were developed to improve prediction accuracy. The MLP model outperformed the conventional ANN model, while a comparative evaluation of several machine learning regression algorithms identified the CatBoost regressor as the most accurate model. Furthermore, finite element method (FEM) simulations were performed to corroborate the experimental findings and provide a deeper insight into the influence of pollution distribution on electric-field characteristics. Uniform pollution increased the electric-field intensity by 25.7% compared with non-uniform pollution, whereas RLNU pollution produced 11% lower electric-field intensity than RNU, indicating a more stable field distribution. Mechanical defects also increased the maximum electric-field intensity by approximately 63.81% compared with intact insulators. The integrated framework, combining experimental investigation, artificial intelligence-based prediction, machine learning, and finite element simulations, provides a comprehensive understanding of flashover behavior and significantly improves the predictive capability of silicone rubber insulators operating under harsh environmental conditions.