Advanced Design of Heat Exchangers using multiscale models and machine learning

Informacje na temat projektu

ADeHEx

Identyfikator umowy o grant: 101106842

DOI

10.3030/101106842

Projekt został zamknięty

Data podpisania przez KE 24 Kwietnia 2023

Data rozpoczęcia 1 Października 2023

Data zakończenia 30 Września 2025

Finansowanie w ramach

• Marie Skłodowska-Curie Actions (MSCA)

Koszt całkowity

Brak danych

Wkład UE

€ 230 774,40

Koordynowany przez

SYDDANSK UNIVERSITET
Denmark

Periodic Reporting for period 1 - ADeHEx (Advanced Design of Heat Exchangers using multiscale models and machine learning)

Okres sprawozdawczy: 2023-10-01 do 2025-09-30

Podsumowanie kontekstu i ogólnych celów projektu

The project addressed the urgent need for more energy-efficient cooling technologies, which are critical for a sustainable and climate-neutral society. Modern industries such as electronics, renewable energy, and transportation face increasing challenges in managing heat while reducing energy consumption and greenhouse gas emissions. Traditional design methods for heat exchangers and cooling devices are limited in their ability to fully exploit new manufacturing techniques such as additive manufacturing.

The project set out to develop advanced design methods based on topology optimisation, which is a computational approach that automatically generates highly efficient structures. By combining mathematical models, multi-scale simulation, and emerging techniques in machine learning, the project aimed to deliver design frameworks capable of producing cooling systems with unprecedented efficiency and manufacturability.

The overall objectives are as follows:

1. To create novel homogenisation- and de-homogenisation-based topology optimisation methods for fluid and heat transfer.
2. To integrate machine learning for surrogate modelling and faster optimisation.
3. To demonstrate pathways towards industrial applications through collaboration with international partners and industry.

These objectives are aligned with European strategic goals for climate neutrality, industrial innovation, and sustainable energy systems. The project pathway to impact lies in bridging fundamental research and industrial practice, enabling new generations of heat exchangers and cooling technologies that are both high-performing and manufacturable.

Prace wykonane od początku projektu do końca okresu sprawozdawczego oraz najważniejsze dotychczasowe rezultaty

During the project, several key scientific and technical activities were carried out. A multi-scale optimisation framework for microchannel cooling was developed, allowing designs that bridge fine microstructures and large-scale heat sink geometry. New de-homogenisation methods were introduced, including a phasor-based approach that drastically reduces computational cost while ensuring realistic and connected flow channels. Neural networks were trained to serve as surrogate models, enabling rapid evaluation of unit-cell properties and accelerating optimisation workflows.

The project also produced a high-resolution 3D topology optimisation workflow for conjugate heat transfer, which integrates both fluid and solid domains. This workflow was coupled with a graphical user interface, making the methods more accessible to engineers, and a prototype heat exchanger was manufactured and scanned for validation.

In parallel, the project explored scientific machine learning approaches such as Physics-Informed Neural Networks and Convolutional Neural Networks during a research secondment at Brown University. These methods were tested for solving fluid and heat equations and for linking low- and high-fidelity models.

The work was disseminated through multiple high-quality journal publications, international conference presentations, and industrial collaborations. Achievements were recognised with a Best Paper Award at the IEEE ITherm 2025 conference.

Innowacyjność oraz oczekiwany potencjalny wpływ (w tym dotychczasowe znaczenie społeczno-gospodarcze i szersze implikacje społeczne projektu)

The project delivered several results that advance the state of the art. The homogenisation and de-homogenisation based approach has been extended to thermo-fluid problems. The phasor-based dehomogenisation method represents a breakthrough by reducing computational effort by orders of magnitude while improving design quality. The neural network surrogate model provides a novel solution to learn the nonlinear operator in the convection-dominated heat transfer. The 3D conjugate heat transfer workflow, with integrated GUI and prototyping, demonstrates a direct pathway from fundamental research to industrial practice.

These results have the potential to transform how cooling devices and heat exchangers are designed. They offer practical benefits such as improved thermal performance, reduced design cost, and compatibility with additive manufacturing. To ensure further uptake, continued follow-up research is required to extend the methods to two-fluid systems and fully three-dimensional Navier–Stokes flow. Industrial validation through large-scale prototyping, demonstration in real-life environments, and access to additional funding will be key steps for future success.

3D-printed metal-based heat exchanger prototype

Multi-scale design of a diffuser

Neural network architecture for learning nonlinear operator in convection-dominated heat transfer

Velocity streamlined in a multi-scale design of microchannel cooling heat sink

Ultra-high resolution microchannel cooling using a reduced-dimensional model