In collaboration with the UK Atomic Energy Authority (UKAEA), this project aims to leverage the transformative potential of machine learning to unravel the operational behaviour of Silicon Carbide/Silicon Carbide (SiC/SiC) materials. SiC/SiC, a prominent class of ceramic matrix composites, plays a vital role in the advancement of fusion nuclear reactors. With increasing global demand for clean energy sources, the development of controllable fusion power generation becomes increasingly critical. As we approach the breakthroughs necessary for practical fusion power, understanding the behaviour of vital materials like SiC/SiC under operational conditions forms a crucial aspect of transforming fusion energy into a feasible and reliable energy source.
The inherent complexity of characterising and modelling SiC/SiC composites under in-service conditions is attributed to their multiscale and multiphysics nature. The materials encounter a spectrum of degradation mechanisms, including micro-cracking and the effects of neutron irradiation, high temperature and potentially lithium corrosion. These factors contribute to uncertainties pervading all length scales of the material system. The intricacy of these interacting phenomena, some of which remain not fully understood, poses significant challenges. As such, existing approaches are insufficient, unable to adequately account for the full spectrum of degradation mechanisms at play in SiC/SiC materials.
Advanced techniques, such as X-ray tomography and diffraction, electron microscopy, acoustic emission, along with high-fidelity numerical simulations, have enabled the generation of large and information-rich datasets with intricate details on the material microstructures and their responses to harsh environments. However, a significant challenge lies in efficiently extracting the wealth of information embedded in these multi-modal data and effectively merging them to derive valuable insights. To tackle this challenge, this project will develop a hybrid ML framework, which synergistically combines early fusion (feature-level fusion) and late fusion (decision-level fusion) algorithms. This integrated approach will not only enhance the interpretability of the ML output, but also provide a more comprehensive understanding of the SiC/SiC material system. An active learning approach will be developed to alleviate the requirement of extensive data labelling, which may be difficult for large amount of multi-modal data. Furthermore, to accommodate the stochastic nature of the SiC/SiC materials, Bayesian methods will be adopted, providing a quantifiable measure of confidence in the fused data. This is of utmost importance for the safe operation of nuclear reactors, as it offers a robust foundation for decision-making under uncertainty.
To further enhance the transparency and interpretability of our machine learning framework, we will incorporate physics-informed neural networks (PINNs) into our approach. This unique approach seamlessly integrates the partly known physics and multi-modal data, offering a holistic perspective of the material system. With its proficiency at solving both forward and inverse problems, PINNs facilitate the uncovering and refinement of the fundamental physics of SiC/SiC materials, thereby shedding light on the primary factors influencing their performance in extreme conditions. This knowledge will pave the way for optimizing material design and manufacturing processes, while also supporting informed decision-making on the most suitable applications and optimal timing for deploying these advanced materials.
In this interdisciplinary project, the PhD candidate will work closely with computer science experts to develop cutting-edge AI techniques. By efficiently merging multi-modal data and quantifying uncertainties, the project will enhance material design and safe operation of nuclear reactors.
The project will be carried out as part of an interdisciplinary integrated PhD in the UKRI Centre for Doctoral Training in Accountable, Responsible and Transparent AI (ART-AI). The ART-AI CDT aims at producing interdisciplinary graduates who can act as leaders and innovators with the knowledge to make the right decisions on what is possible, what is desirable, and how AI can be ethically, safely and effectively deployed. We value people from different life experiences with a passion for research. The CDT's mission is to graduate diverse specialists with perspectives who can go out in the world and make a difference.
Successful applicants will have, or expect to receive, a master's degree or first or upper-second bachelor's degree in a relevant subject.
Formal applications should include a research proposal and be made via the University of Bath’s online application form. Enquiries about the application process should be sent to [Email Address Removed]. Enquiries about the research should be directed to Dr Chen.
Start date: 2 October 2023.
Continue with Facebook
