Journal article 47 views 11 downloads
Integrated environment for machine learning-aided heat transfer optimisation in internal flows: hammerhead software
,
A Liptak ,
Antonio Gil ,
Michael G. Edwards,
A Dubas,
A Davis
Machine Learning: Science and Technology, Volume: 7, Issue: 1, Start page: 015011
Swansea University Authors:
Daniela Segura Galeana , Antonio Gil , Michael G. Edwards
-
PDF | Version of Record
© 2026 The Author(s). Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 license.
Download (3.84MB)
DOI (Published version): 10.1088/2632-2153/ae3053
Abstract
The optimisation of the cooling system performance in a tokamak reactor requires the accurate analysis of the thermo-hydraulic interactions under complex flow regimes. Although computational fluid dynamics (CFD) provides high-fidelity insight at a lower cost than experimental testing, optimisation r...
| Published in: | Machine Learning: Science and Technology |
|---|---|
| ISSN: | 2632-2153 |
| Published: |
IOP Publishing Ltd
2026
|
| Online Access: |
Check full text
|
| URI: | https://cronfa.swan.ac.uk/Record/cronfa71237 |
| Abstract: |
The optimisation of the cooling system performance in a tokamak reactor requires the accurate analysis of the thermo-hydraulic interactions under complex flow regimes. Although computational fluid dynamics (CFD) provides high-fidelity insight at a lower cost than experimental testing, optimisation remains computationally prohibitive due to the large number of simulations required. To address this, we introduce Hammerhead, the first open-source, transparent and fully integrated machine learning (ML)-CFD framework with automated high-fidelity database generation and modular multi-surrogate modelling for optimisation in pipe flow heat and mass transfer problems. The software provides seamless Python-OpenFOAM integration for automatic high-fidelity model database building and multi-surrogate model comparison from 3 ML algorithms with modular architecture within Python infrastructure. For the high-fidelity database, the software constructs a parametric space using up to four shape parameters to deform pipe wall geometries and perform the conjugate heat transfer simulations. The surrogate models are constructed on the basis of radial basis function interpolation, feed-forward neural networks, and Gaussian processes, and trained on the high-fidelity database to approximate thermo-hydraulic responses efficiently. The framework provides a comparative environment for systematically assessing surrogate model accuracy and reliability within the same optimisation workflow. The approach offers a practical tool that increases the feasibility of design optimisation on small datasets for cooling system applications. On a database of up to 400 cases for a 2D domain, a geometry with thermo-hydraulic performance enhancement of over 500% with respect to that of a smooth surface pipe at Re = 1000 was found with the aid of surrogate modelling, validated through high-fidelity simulation. |
|---|---|
| Keywords: |
machine learning, computational fluid dynamics, neural networks, heat transfer, Gaussian processes |
| College: |
Faculty of Science and Engineering |
| Funders: |
The research was developed through the ESPRC DTP ADDED project with the financial support of RCUK Energy Programme, Grant Number EP/T012250/1, and ERDF via the Welsh Government. |
| Issue: |
1 |
| Start Page: |
015011 |