Spectral and transform-domain based feature extraction approach for wettability and evaporation characteristics evaluation in oil-impregnated paper used in power transformers

The reliability and longevity of power transformers critically depend on the dielectric performance of oil-impregnated paper (OIP) insulation, where surface wettability and evaporation dynamics play a decisive role in moisture migration and ageing mechanisms. In this study, a novel approach based on Gabor filters and Discrete Cosine Transform (DCT) feature extraction is introduced to characterise and model the wettability behaviour and evaporation rate on OIP surfaces. High-resolution droplet images were systematically acquired under controlled laboratory conditions, and texture-domain and frequency-domain features were extracted to quantify temporal variations in droplet morphology. Regression models, including linear, quadratic, and cubic polynomial fitting, were employed to evaluate predictive performance. The results demonstrate that Gabor-based features effectively capture edge structures and spatial texture variations, achieving a high coefficient of determination (R2 = 0.9650) with cubic regression, while DCT-based features provide a compact frequency-domain representation with reliable modelling accuracy (R2 ≈0.87). These findings highlight the complementary strengths of spatial- and frequency-domain approaches in capturing the multiscale dynamics of evaporation and wettability. The proposed contactless, image-based approach enables accurate prediction of contact-angle evolution, advancing diagnostics for transformer insulation health and aging, and establishing a transferable methodology for surface characterization and moisture-related reliability assessment in dielectric and porous materials.