Improving Ultrasonic Imaging using Machine Learning

Ultrasonic non-destructive evaluation (NDE) describes the practice of transmitting sound waves through solid objects and analysing the resulting scattered waves to construct images of the object's interior, facilitating defect detection and materials characterisation.
The ultimate aim of this project is to investigate enhanced ultrasonic NDE of metal components directly at the point of manufacture, to deliver high-quality components right, first time. Importantly, this in-process approach requires the development of real-time data processing algorithms. Machine learning algorithms used in conjunction with models and simulations of wave propagation will be explored to facilitate this.
Specifically, this project will examine how to compensate for the effects that thermal gradients, complex build geometries and heterogeneous microstructures have on the probing ultrasonic waves. Three research questions will be addressed:
i) Can complex and dynamic build geometries be accurately mapped out in near real-time using in-process ultrasonic inspection data, where extreme thermal gradients (generated by the manufacturing process) cause distortion of the expected wave paths?
ii) Can this knowledge of the component geometry, coupled with models of the thermal gradient be used to better constrain and drive the microstructure characterisation problem?
iii) Can the knowledge from steps i) an ii) be used to reliably image components as they are built?
The final deliverable of the project will be a fully automated in loop capability, to be demonstrated on test structures manufactured for aerospace and energy applications (with other industrial members applications supported as appropriate).