E-Thesis 114 views 81 downloads
A digital twin of fusion energy components with physics-informed neural networks / PRAKHAR SHARMA
Swansea University Author: PRAKHAR SHARMA
-
PDF | E-Thesis – open access
Copyright: The Author, Prakhar Sharma, 2024 Distributed under the terms of a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0).
Download (26.59MB)
DOI (Published version): 10.23889/SUThesis.68811
Abstract
A fusion reactor presents a highly challenging operational environment, with engineering components exposed to extreme conditions. Temperatures range from 100 million °C at the centre of the plasma to -269 °C in the cryopump, all within a few metres. In addition to these temperature extremes, compon...
| Published: |
Swansea University, Wales, UK
2024
|
|---|---|
| Institution: | Swansea University |
| Degree level: | Doctoral |
| Degree name: | Ph.D |
| Supervisor: | Nithiarasu, P., and Baxter, M. |
| URI: | https://cronfa.swan.ac.uk/Record/cronfa68811 |
| Abstract: |
A fusion reactor presents a highly challenging operational environment, with engineering components exposed to extreme conditions. Temperatures range from 100 million °C at the centre of the plasma to -269 °C in the cryopump, all within a few metres. In addition to these temperature extremes, components must withstand intense electromagnetic loads and irradiationdamage. These extreme conditions were achieved for short periods at JET, formerly the world’s largest fusion device located at Culham Centre for Fusion Energy, UK, before it ceased operations. But one of the greatest engineering challenges of the 21st century will be to construct a machine that can operate under these extremes routinely and produce commercially viable energy. To create a fusion reactor, relevant components must undergo rigorous testing before they can be deployed. Initially, testing at the HIVE facility focused on the sample under test (SUT) under thermal loads alone. However, with the development of testing facilities such as CHIMERA, it will soon be possible to test the SUT under both thermal and magnetic loads. Given the challenges associated with testing, any additional information to understand these components would be extremely useful. This project investigates the potential for creating a foundational digital twin capable of replicating reality from sparse data inputs. A digital twin enables real-time monitoring, provides predictive insights, and supports decision-making to optimise experimental campaigns and ensure safety. In this context, the thesis investigates the suitability of physics-informed neural networks (PINNs) for solving inverse problems and extends their application to develop a foundational physics-informed digital twin platform for replicating physical experiments. PINNs are ideal for scenarios with limited data availability, as they do not require extensive pre-training. They provide a straightforward approach to solving inverse problems without complex algorithms, making them ideal for enhancing testing processes in experimental facilities. This study demonstrates that PINNs can accurately reconstruct thermal fields from sparse data and solve inverse problems with challenging boundary conditions (BCs). The developed PINN- based digital twin reconstructed steady-state temperature distributions of the HIVE sample under varying thermal loads, enabling detailed analysis of the temperature field and the corresponding thermal conductivity variation. These findings establish PINNs as a promising tool for inverse modelling with limited data. |
|---|---|
| Item Description: |
A selection of content is redacted or is partially redacted from this thesis to protect sensitive and personal information. |
| Keywords: |
Mechanical engineering |
| College: |
Faculty of Science and Engineering |
| Funders: |
EPSRC, UKAEA |