ENACT - Modélisation moléculaire assistée par l'IA du transfert de chaleur dans des nanofluides durables

Keywords

Molecular Simulations, Machine Learning, Graph Theory, Chemical Engineering, Heat Transport Technology, Green Solvents

Subject details

Heat transfer fluids (HTFs) are used to transport thermal energy in industrial and technological applications, from refrigeration systems to solar power plants.1 However, the chemical and physical limitations of conventional fluids (e.g., water, ethylene glycol, and their mixtures) constrain performance, reliability, and durability. Since the 1990s, nanofluids, liquids containing dispersed nanoscale particles, have emerged as a promising alternative, showing improved thermal properties.3 Among them, deep eutectic solvents (DESs) are attracting growing interest,3 due to their low cost and reduced environmental impact,4 particularly natural DESs,5 composed of hydrogen bond acceptors (HBA) and donors (HBD) derived from primary metabolites. Nanoparticle incorporation can further enhance thermal conductivity and heat transfer, while DESs offer wide tunability of viscosity, stability, and interfacial interactions. In these systems, molecular-level understanding of how HBA:HBD combinations affect solvent–nanoparticle interactions is essential. Preliminary simulations indicate that nanoparticle solvation and stability depend strongly on intermolecular forces and structuring at the DES-particle interface.6 In recent years, the DES community has emphasized the need for accurate force fields (FFs) capable of reproducing their strongly non-ideal behavior.7 In this project, we will combine expertise from the LEMTA and LPCT teams to develop AI-enhanced, system-specific interaction models for DES-based nanofluids. Moving beyond fixed empirical FFs, interaction parameters will be learned using supervised machine-learning (ML) approaches trained on ab initio energies and forces.8,9 Although recent work based on the state-of-the-art MACE architecture10 has shown promising results for molten salts,11 a methodology explicitly designed to treat environmentally dependent hydrogen bonds, key to thermal transport in DESs, is still lacking. This project will address this gap by developing a consistent strategy for combining ML-trained interaction parameters with experimental data. Chemically and structurally meaningful descriptors, complemented by hydrogen-bond network representations, will ensure robustness and interpretability, building on recent advances at LEMTA.12 This approach will enable molecular dynamics simulations that retain ab initio accuracy while reaching the time and length scales required to analyze thermal transport. Preliminary results from the IRP project between LPCT and ICCOM (Italy), extending first-principles force-field development to DESs,13 will guide the selection of target descriptors. In parallel, we will develop structural descriptors to capture the complex organization of DESs, which often exhibit nanoscale microdomains that reflect spatial heterogeneity in composition and hydrogen-bond structure.14 These features, commonly studied using geometric approaches (e.g., Voronoi analysis),15 will be complemented by graph-based representations of hydrogen-bond networks, where molecules are nodes and hydrogen bonds form dynamic weighted edges.16 This framework will capture connectivity, clustering, and collective organization in DESs and near the nanoparticle surfaces. We will focus on choline chloride–based DESs combined with glycerol, urea, or ethylene glycol. Target properties include thermal conductivity and heat capacity, with density and viscosity also accurately reproduced. Experimental data for these systems are widely available.3 The most common nanoparticles (TiO2, Al2O3, and graphene oxide) will be considered due to their chemical inertness, thermal stability, and good dispersion without excessive viscosity increases.2 By integrating parameter training, descriptor development, and experimental validation, this project matches the goals of the ENACT program in AI-based scientific discovery and will provide practical guidelines for designing high-performance and sustainable DES-based heat transfer fluids.