Training-Free Learning Applied in GIS X-DR Image Analysis

Abstract

Gas Insulated Switchgear (GIS) plays a crucial role in the electrical grid, particularly in transmission, distribution, and substation applications, and the safe operation of GIS is related to the secure functioning of the power grid. Recent years, X-ray digital radiography (X-DR) as a powerful non-destructive tool has been widely used for inspecting internal defects of GIS. However, the majority of collected GIS X-DR images still require specialized manual analysis, which is time-consuming and labour-intensive. In this paper, we present a novel application of advanced artificial intelligence (AI) models in GIS X-DR analysis. Specifically, we propose a pipeline of components segmentation based on a training-free method, utilizing the foundation models Segment Anything Model (SAM) and SegGPT. Meanwhile we annotate and build a small dataset with nearly 100 GIS X-DR images and three categories covering small-size, larger-size, and grain-type components. Then we carefully conduct several experiments with the dataset. The results demonstrate that the application of this training-free pipeline achieves a high precision of component segmentation in GIS X-DR images. Our method could be directly used for GIS X-DR analysis, like small components counting and image quality test. Also, this pipeline could be used to label the GIS X-DR or other images for next step AI methods application. So, we could develop more and better AI models in GIS XD-R analysis. This advancement in automated GIS X-DR analysis not only contributes to the reliability and efficiency of power distribution systems but also opens avenues for automated inspection and maintenance processes in the energy sector.