Defect detection using 3D computed tomography images and application on nuclear power plants

Abstract

Tools used in nuclear power plant (NPP) inspection are required to be inspected before and after use on a reactor to check their integrity. To address the long duration required for manual inspection, non-destructive testing (NDT) can be implemented. In this thesis, a novel NDT framework is developed using key image processing functions to localize and identify the missing and misplaced components of NPP inspection tools using 3D CT data. Analyzing the limitations in image registration, a new algorithm is proposed to improve existing image registration techniques to handle significant rotational differences. Additionally, to address the annotated data-related issues in deep learning-based approaches, a semi-automated technique for 3D CT data annotation is introduced using the modified image registration process and Computer-aided design (CAD) design files. The annotated data is then used to train a deep learning-based semantic segmentation model to segment each component of the tool in the CT data.