Composite insulators are integral to the power grid, providing electrical insulation and mechanical support. However, their prolonged exposure to outdoor conditions renders them vulnerable to ageing. This can result in flashovers and economic losses. Hyperspectral imaging (HSI) has been identified as a promising non-contact detection method to effectively assess the microscopic ageing status of these insulators’ surfaces. This study involved conducting artificial accelerated ageing experiments to mimic extended outdoor operational conditions. The static contact angle of ageing silicone rubber samples served as the basis for ageing classification. High-resolution HSIs of samples at various ageing stages were obtained, covering wavelengths from 400 nm to 1040 nm. Singular spectrum analysis was applied to denoise spectral and spatial domains, eliminating environmental and equipment-related noise effectively. Additionally, a convolutional neural network (CNN) was utilised to extract deep features from the spectral domain and combine them with spatial domain features, thus reducing dimensionality. The resulting fused spatial–spectral features were classified using a support vector machine. The joint denoising and feature fusion in this study’s spectral–spatial domain resulted in a notable improvement of 34.9 % in overall accuracy, achieving a score of 98.3 %. This approach exhibited superior performance, particularly in evaluating severely aged samples. The proposed approach enhances the accuracy in assessing the ageing status of composite insulators and enables pixel-level visualisation of ageing distribution. This assists in guiding the design of insulators’ shapes and materials for various environmental conditions. Moreover, this method is potentially applicable in detecting ageing in other electrical equipment.
Read full abstract