Obtaining data for training a vision system to detect gas inclusions in the glass part of an PS-70E insulator

This paper considers obtaining data for training technical vision systems to detect spherical air inclusions in the glass parts of suspended insulators of the PS-70E brand. The data are provided for air bubbles with diameters of 1 and 2 mm in the head of the stackable part. Based on the distribution patterns of the electric field, the location of the bubbles in which the electrical strength is violated during the operation of the insulator under the worst operating conditions, i.e. under a voltage of 40 kV, are identified. To determine the presence of bubbles from one image, a three-dimensional ray tracing model was created in which the illumination of the glass part is carried out orthogonally from below, and the image is taken from the side of the head. The modeling identifies the locations of the air bubbles in which defects are not distinguishable. There are areas of defects in which there is a violation of electrical strength, which cannot be identified under the applicable conditions of removal. The proportion of regions that simultaneously belong to the two categories is 6.7 and 2.5 % for holes with diameters of 1 and 2 mm, respectively. There are bubble locations for which there is a significant distortion of the defect figures under the described conditions (22.4 and 20.0 % for bubbles of 1 and 2 mm, respectively).