E-Thesis 104 views 5 downloads
Simulation driven machine learning methods to optimise design of physical experiments and enhance data analysis for testing of fusion energy heat exchanger components / RHYDIAN LEWIS
Swansea University Author: RHYDIAN LEWIS
DOI (Published version): 10.23889/SUthesis.66041
Abstract
Plasma facing components (PFCs) must be designed to routinely withstand the harsh environment of a fusion device, where temperatures at the core of the plasma exceed 150,000,000 °C. The heat by induction to verify extremes (HIVE) experimental facility was established to replicate the thermal loads a...
| Published: |
Swansea University, Wales, UK
2024
|
|---|---|
| Institution: | Swansea University |
| Degree level: | Doctoral |
| Degree name: | Ph.D |
| Supervisor: | Evans, Ll. |
| URI: | https://cronfa.swan.ac.uk/Record/cronfa66041 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Abstract: |
Plasma facing components (PFCs) must be designed to routinely withstand the harsh environment of a fusion device, where temperatures at the core of the plasma exceed 150,000,000 °C. The heat by induction to verify extremes (HIVE) experimental facility was established to replicate the thermal loads a PFC is subjected to during normal operation of a fusion device.To maximise its impact on the design of PFCs, HIVE must deliver smarter testing and improved component insight. Currently, the experimental parameters required to deliver a certain response to the component are decided at the point of testing through a combination of previous experience, intuition, and trial & error, which is both time-consuming and unreliable. To assess a PFC’s suitability, knowledge of its mechanical performance while operating at high temperatures is desirable, however HIVE only records pointwise temperature measurements on the component’s surface using thermocouples. Currently, HIVE has no method of inferring a component’s mechanical response using the temperature measure-ments.Both the challenges of smarter testing and improved component insight can be achieved through the identification of inverse solutions. A popular approach to solving engineering inverse problems is surrogate assisted optimisation, where a machine learning model is trained using finite element (FE) simulation data. Much of the work in literature use single value surrogate models on quite simplistic problems, however HIVE is a real-world, multi-physics problem which requires full field (FF) surrogate models to solve its multitude of inverse problems.The development of a method which can easily construct FE data driven FF surrogates would be invaluable for a variety of tasks in engineering, as well as solving inverse problems. In this work, it demonstrates that it can provide a much more robust and comprehensive method of characterising a PFC’s strengths and limitations, enabling more informed decisions to be made during its design cycle. |
|---|---|
| Item Description: |
Part of this thesis has been redacted to protect personal information |
| Keywords: |
Fusion Energy, Machine Learning, Surrogate Model, Simulation |
| College: |
Faculty of Science and Engineering |