PhD Studentship - Development of a system for automatic defect recognition from ultrasonic data (NSIRC/BUL)
Since their introduction, advanced ultrasonic technologies (AUT), such as phased array or full matrix capture have become rapidly accepted by the welding industry. They offer versatile inspection solutions, broadly applicable to a wide range of applications, and they can achieve performance with a high probability of detection (POD) if applied appropriately.
From the collected data, engineers cannot only reach a joint decision over a single component's quality but also make educated decisions over the overall quality of the manufacturing environment. Therefore, the acquired data, if appropriately deciphered, may provide valuable information and room for improvements to all the involved departments and the opportunity for the stakeholders to reduce and eliminate unnecessary sources of waste. However, as the inspection data set sizes now being acquired are very large, in many cases, the information they contain is not gathered or shared across the entire production line. Therefore, a more effective process and evaluation of these data are required.
Simultaneously, when AUT are implemented, a human inspector evaluates the achieved weld quality by completing the data acquisition process. The data post-processing can be a time-consuming course in itself. A human operator must go through all available scans and evaluate every indication. Besides the fact that this approach is prone to human error, it can also lead to substantial delays during fabrication. To complicate things even more, if, for instance, an error in the application of the inspection technique is discovered at this stage, then the inspector may have to go back on-site to undertake the data acquisition anew. In cases with limited access or when the inspected structure is already under operation, the need for such corrective engagements may increase the maintenance costs substantially. Consequently, faster decision-making and evaluation are required with less impact from human errors.
Industry 4.0 and smart fabrication methods are already fuelling demand for more agile inspection systems with an improved capability of detecting flaws during fabrication with approaches that meet the increasingly demanding fabrication targets. The lack of rapid scan-data processing presents a barrier to developing in-line and in-situ inspection systems operating during fabrication. Increasingly automated inspection methods are required, and the aspiration within the industry is to establish inspection solutions capable of simultaneously performing data acquisition and non-destructive evaluation (NDE) of the final product quality.
We need to address the issues of inspection time and quality assurance to allow the industry to tackle challenges for in-process detection and NDE of remote welding (i.e. electron beam) and/or additive manufacturing imperfections and to commission new structures promptly, with zero waste and unnecessarily spend of resources.
This PhD will aim to develop an algorithm that replicates the steps a human inspector follows during the ultrasonic data analysis but implements an artificial intelligence (AI)-based assisted defect recognition (ADR) solution instead.
The NDT industry, in general, has been slow to adopt AI, but advancements for AI in other fields and their proven use in the last few years mean that best practices can be adopted. By using a training set of, i.e. front, side and sectorial ultrasonic image views that contain both defective and non-defective images, an AI algorithm will learn (to some extent) to recognise potential defective areas. AI algorithms typically require very large datasets to learn before being effective. In most cases, datasets with such a quantity are seldom available.
Therefore, the PhD candidate shall investigate efficient methods to expand data sets by potentially merging simulation tools with AI architectures. Effectively the executed work will explore relevant AI technologies for use with AUT, determine what is currently commercially available, advance and demonstrate the capabilities of a developed AI solution and examine its applicability to the broader NDT community on other applications. In addition, the aim is to provide recommendations regarding using AI standards for UT data (relevant for any standards research committee) and to offer guidance for using AI in industrial NDT inspection in general.
Candidates should have a relevant degree at 2.1 minimum, or an equivalent overseas degree in (academic and industry requirements) Overseas applicants should also submit IELTS results (minimum 6.5) if applicable.
This project is funded by Brunel University London and TWI.
How to apply
Please contact Hannah Stedman @ email@example.com for details on how to apply and Emma Smith @ firstname.lastname@example.org for information on how to apply to Brunel University London