Texture analysis of microscopy images from power transformer cellulose insulation for aging condition assessment
MetadataShow full item record
Electrical paper insulation used in power transformers thermally deteriorates during the normal transformer operation. When this deterioration becomes significant, tensile strength is reduced, and the risk for insulation failure is increased. The most accurate method for paper condition assessment is the degree of polymerization (DP) measurement, where a sample of paper is removed from the transformer for chemical analysis. However, the quantity of paper required for DP measurement is often considered too invasive. In this research, methods for texture analysis on microscopic images of thermally aged insulation paper, is presented as an alternative approach for condition assessment. An experimental setup was developed to artificially aged oil-impregnated Kraft paper samples. The samples were thermally stressed in an oven at temperatures above those normally present in power transformers, to produce a sample set with varying levels of insulation deterioration. Microscopy images of the paper samples were analyzed using two texture analysis methods. The first method is a statistical-based texture analysis method called the spatial grey level dependence method (SGLDM). SGLDM converts images into matrices containing information about the statistical variation of pixel grey-level intensities in an image. Mathematical operators applied to the SGLDM are used to extract 22 statistical texture features for each sample image. The second method uses a two-dimensional Wavelet transform to extract detail information from Wavelet decomposition coefficient matrices. The Wavelet method is applied recursively, with four decompositions, producing a total of 12 Wavelet texture features per sample image. Analysis of the microscopy images obtained from thermally aged samples show that thermal deterioration of the insulation paper produces changes in the surface morphology and physical structure. These changes are detectable by the texture features extracted from the SGLDM and Wavelet texture analysis. Correlations between texture features and DP measurements performed on the paper samples are analyzed, and statistical classification is performed on the feature set to demonstrate that differentiation between oil-impregnated paper samples with different levels of thermal degradation is reliable with low error rates. Therefore, development of a practical method to assess condition of oil-impregnated paper insulation using optical microscopy and texture analysis is promising.