About the project
In this project, you will study the influence that different advanced optimization and machine learning algorithms have on the quality of 3D images generated from X-ray computed tomography measurements of industrial components. The focus is on optimising the ability to detect internal object defects that would compromise component safety.
Modern manufacturing processes, such as additive manufacturing (3D printing), offer unrivalled flexibility in the development of component shapes, leading to light-wight and efficient designs not achievable with more traditional manufacturing techniques. For safety critical applications such as aviation or energy production, these components need to be inspected before and during service. Traditional non-destructive inspection methods, such as ultrasound inspection, can however not be used for many objects of complex shapes.
X-ray computed tomography is a promising alternative non-destructive testing methodology. Whilst it can be applied to the inspection of complex geometries, providing detailed images of internal structures, the inspection process in often slow. Furthermore, complex geometries with high aspect ratios or for components where the X-ray path length varies greatly during inspection, significant artefacts and noise can contaminate the images, making defect detection challenging.
To overcome these challenges, the use of advanced iterative or machine learning based image reconstruction is often proposed. Whilst these methods have shown promising performance in simplified and well controlled settings, for real applications, a range of questions however remain. For iterative optimization, there are a range of algorithm dependent parameters, such as the number of iterations or the optimal regularization parameter, that need to be chosen. Unfortunately, it remains unclear how to do this choice or how to efficiently evaluate the resultant images in terms of the visibility of expected defects.
In this project, you will thus use real and simulated X-ray tomography data to evaluate the performance of advanced reconstruction algorithms. Defect visibility is typically a function of both image resolution and image noise. By measuring anisotropic resolution and noise for a wide range of object geometries, parametric studies of key advanced reconstruction approaches can then be explored.
You will be working alongside the team at the University of Southampton’s µ-VIS facility, a dedicated X-ray Computed Tomography (XCT) centre. The centre is part of the UK’s National Facility for lab-based XCT and houses some of the UK’s largest micro-focus CT scanning systems, capable of unveiling sub-surface information from materials, components and structures. With strong links between both research and industry, the centre itself is used for aircraft crash investigations, Formula 1, space technology and much more.
