Automated monitoring of insulation by ultraviolet imaging employing deep learning

The corona effect on the surface of electrical system equipment and components generally indicates undesirable phenomena that can lead to physical degradation of materials or even equipment failure. One of the most promising techniques for monitoring corona discharges is the use of specialized cameras for the detection of ultraviolet radiation. This paper introduces an innovative algorithm for classifying the criticality of insulation based on attributes extracted from videos recorded using an ultraviolet detection camera. The attributes extracted from each facula origin include maximum persistence, area, and the minimum distance between the facula origin and the insulation. To obtain this distance, a technique combining a deep convolutional neural network model with an adaptive segmentation thresholding method is proposed. To validate the proposed methodology, inspections were conducted at a 500 kV substation. A total of 96 videos were recorded, within which 99 facula origins were identified. The object detection model applied demonstrated an accuracy of 85.5% in detecting insulation in images, based on a validation set comprising 1,985 images and 8,730 instances. The results of the classification showed that 72.7 % of the facula origins recorded originated from regions far from the insulation (mainly cables and corona rings). These results demonstrate that the distance between the insulation and the facula origin is an essential attribute for video analysis, providing context for recorded discharges and allowing differentiation between cases where ultraviolet radiation originates from insulation and those where discharge location is less critical.