Swept-source optical coherence tomography for ageing assessment of high voltage cellulose insulation

Oil- impregnated cellulose materials (such as Kraft paper) form an essential part of a transformer insulation system. The mechanical strength of the Kraft paper directly impacts the safe operation of the power transformer. Obtaining both spatially localized absorption profiles and subsurface structural images of the transformer paper would offer considerably more information about the samples than those obtained from optical methods such as microscopy. This could allow a continuous estimation of the operational age of the transformer insulation, beyond only classifying it as belonging to a broad age group. In this thesis, optical coherence tomography (OCT), a novel and non-invasive diagnostic technique, is used for the first time to acquire subsurface images of the thermally aged Kraft paper samples. To achieve this goal, we implemented a swept-source quadrature OCT (SS-QOCT) system for both structural and spectroscopic OCT image acquisition. Then frame theory was used to develop a generalized OCT image reconstruction method using redundant and non-uniformly spaced frequency domain samples. We also introduced a new concept of applying compressed sensing techniques to acquire a sample's spectroscopic properties, (i.e., its spatially localized light absorption profile). As the typical number of samples acquired by OCT was not sufficient for this purpose, we developed a gradient-based multidimensional signal recovery from incomplete samples in arbitrary separable dictionaries to reduce the number of required measurements for successful spectroscopic signal recovery. Then accelerated thermal ageing was used to generate samples of oil-impregnated Kraft paper with different levels of deterioration. From these aged paper samples, subsurface structural images at different depths were obtained using swept-source optical coherence tomography (SS-OCT). Statistical texture analysis was applied to these OCT images to convert each image into a spatial grayscale level dependence m