Machine learning to locate defects in ultrasonic inspection images

It is important that manufactured components are inspected to identify defects which may cause early failure, particularly in safety critical systems. Non-destructive techniques, such as ultra-sound, are used regularly to be able to see below the surface to identify hidden defects. This project aims to develop automatic techniques to help identify the defects. The student will combine image analysis and machine learning methods to build a system that can reliably distinguish between normal parts and regions with abnormalities. The student will be part of the Research Centre for Non-Destructive Evaluate (RCNDE) (www.rcnde.ac.uk) - a collaboration between six universities and many industrial partners. This project will investigate the application of novel analysis techniques to ultrasonic NDE inspection, aiming to support the analysis of phased-array or Time-of-Flight-Diffraction images.

The project will be actively supported by BAE Systems Maritime who deploy these techniques in a large scale manufacturing environment. The benefits of a robust automated analysis process would be very significant and could potentially reduce the inspection cost and duration for large scale welded structures across many industrial sectors.