Research on insulator contamination component identification based on neural network

Glass suspension insulators in power transmission lines are vulnerable to surface contamination over time, especially in harsh environments like metallurgical plants. Analysis of such contamination revealed significant metal deposits, primarily iron particles sized between 2 μm and 20 μm. To study the impact of this metallic contamination on flashover behavior, researchers created artificial pollution using NaCl, diatomaceous earth, and iron powder. Leakage current tests demonstrated that metal content fundamentally alters the current waveform, causing it to exhibit AC superimposed impulses. Key findings include: metal lowers the voltage threshold for impulse inception, shortens the impulse rise and fall times, and increases critical impulse parameters (frequency, maximum amplitude, and discharge magnitude) as the metal proportion rises. Furthermore, a ResNet18-SA deep learning model was developed, integrating a self-attention mechanism. This architecture demonstrates exceptional robustness in interpreting pulsed current signals while accurately classifying levels of metallic contamination, providing a reliable and automated solution for insulator condition assessment.