The Study on Oil-paper Insulation Classification with Scattering Features of Raman spectrum

Abstract

Raman spectroscopy detecting technique rapidly develops in the field of material composition detection, including electrical insulation materials. In this paper, three kinds of mineral insulating oil are selected to form three experimental control groups with insulating paper. Take samples of aging oil according to specified aging time to collect Raman spectrums, establishing a data set for each group, respectively. In addition, a new diagnostic model is introduced, which consists of wavelet scattering network and kernel SVM fulfilling recognition of different aging stages. It shows that all classification accuracy results are above 90%, where validates that scattering features transformed from Raman spectrums are effectively classified by proposed model.