Mechancial Engineering

Ultrasonic monitoring of corrosion is an accepted practice in several industries where corrosion of components is a frequent issue and where monitoring this corrosion is difficult to do visually. This is often the case in the petrochemical and nuclear industries where the interiors of steel pipes are corroded by flowing gases and fluids which often cannot be stopped while the monitoring takes place, and it is non-trivial to remove the components for inspection. An operator will have an ultrasonic probe, or a probe is permanently affixed to the surface in the case where an operator cannot reach or would be at great risk. An ultrasonic pulse is sent into the component, reflects off the back wall of the component (in keeping with the pipe example) and is then returned to the probe which detects the signal. A series of these scans are created to create a colour map of the surface of the interior wall or to create a thickness measurement. An operator then compares this map to other prior maps, or to a quoted safe thickness value and makes a judgement on whether a part needs to be replaced.
It is hoped by performing the research in this PhD project that by utilising advancements in machine learning (ML) that some form of network can be created and trained to help operators make these judgements on the level of corrosion, or to potentially allow the ML program to perform judgements on its own. In addition to developing ML techniques to understand the ultrasound signals it is hoped that new practical methods and technologies can be developed to improve transducer performance on non-ideal surfaces. Ultrasonic transducers need to be coupled to the surface of the component. This means there is an interface between the transducer face and the face of the material. Ideally, both of these surfaces are flat. Often this is not the case. The exterior of components can be corroded as well, leading to non-ideal surfaces under the probe which can affect the transmitted and received...