Nuclear power plays a important role to the United Kingdom electricity generation infrastructure, providing a reliable baseload of low carbon electricity. The Advanced Gas-cooled Reactor (AGR) design makes up approximately 50% of the existing fleet, however, many of the operating reactors have exceeding their original design lifetimes.To ensure safe reactor operation, engineers perform periodic in-core visual inspections of reactor components to monitor the structural health of the core as it ages. However, current inspection mechanisms deployed provide limited structural information about the fuel channel or defects.This thesis investigates the suitability of image-based 3-D reconstruction techniques to acquire 3-D structural geometry to enable improved diagnostic and prognostic abilities for inspection engineers. The application of image-based 3-D reconstruction to in-core inspection footage highlights significant challenges, most predominantly that the image saliency proves insuffcient for general reconstruction frameworks. The contribution of the thesis is threefold. Firstly, a novel semi-dense matching scheme which exploits sparse and dense image correspondence in combination with a novel intra-image region strength approach to improve the stability of the correspondence between images.This results in a percentage increase of 138.53% of correct feature matches over similar state-of-the-art image matching paradigms. Secondly, a bespoke incremental Structure-from-Motion (SfM) framework called the Constrained Homogeneous SfM (CH-SfM) which is able to derive structure from deficient feature spaces and constrained environments. Thirdly, the application of the CH-SfM framework to remote visual inspection footage gathered within AGR fuel channels, outperforming other state-of-the-art reconstruction approaches and extracting representative 3-D structural geometry of orientational scans and fully circumferential reconstructions.This is demonstrated on in-core and laboratory footage, achieving an approximate 3-D point density of 2.785 - 23.8025NX/cm² for real in-core inspection footage and high quality laboratory footage respectively. The demonstrated novelties have applicability to other constrained or feature-poor environments, with future work looking to producing fully dense, photo-realistic 3-D reconstructions.
|Date of Award||28 Jul 2020|
- University Of Strathclyde
|Sponsors||University of Strathclyde|
|Supervisor||Graeme West (Supervisor) & Paul Murray (Supervisor)|