Automated Video Processing and Image Analysis Software to Support Visual Inspection of AGR Cores

摘要

Remote visual inspection of fuel channels in advanced gas-cooled reactor (AGR) cores provides nuclear operators with an understanding of the condition of the UK's fleet of nuclear power plants. During planned, periodic outages, specialist inspection tools equipped with video cameras and other sensors are manipulated inside fuel channels selected for inspection and a video of the entire inner surface of the channel is recorded. If cracks are observed in this process, a montage of the entire crack region needs to be produced, analysed and sentenced (classified with respect to crack morphology, location, orientation and size) before the station is returned to service - provided it is safe to do so. At the present time, the video analysis and crack montage production is done manually by an expert team of inspection engineers. In line with this process, bespoke image stitching software named "ASIST" (automated software image stitching tool) has been trialled in the last 18 months and evaluated using data from Dungeness, Hunterston B, Hinkley Point B, Heysham 1, and Torness outages. The software is now approaching the point of acceptance at which time it is likely to replace the manual process to provide higher quality images with 100% channel visualisation properties in a fraction of the time taken by the current approach. This paper provides a summary of the ASIST evaluation work undertaken and describes recent research endeavours aiming to provide ASIST with algorithms to detect and quantify features in the channel such as brick interfaces, keyways and trepanned holes for which the physical dimensions are known. These features will provide a metric for converting pixel based measurements into millimetres and will support crack sizing in future. Recent research into automated crack detection is also described along with methods to compute structure-from-motion (SfM) which will facilitate the extraction of 3D depth information directly from the 2D video footage. The outcomes of this new research activity will initially provide EDF with new decision support software for crack sentencing. It will also allow station operators to make better use of visual inspection data by facilitating the inference of depth and hence channel bore measurements directly from video.