Periodic Reporting for period 1 - BLESSED (Bridging Models at Different Scales To Design New Generation Fuel Cells for Electrified Mobility)
Período documentado: 2023-02-01 hasta 2025-01-31
BLESSED (msca-dn-blessed.eu) is a Marie Skłodowska-Curie Doctoral Network tackling the urgent challenge of reducing greenhouse gas emissions and accelerating the transition to carbon-neutral mobility solutions. The project develops an innovative multi-scale, machine-learning assisted, simulation framework to model hydrogen-powered Proton Exchange Membrane Fuel Cells (PEMFCs) from the atomistic processes up to the complete cell.
At its core, BLESSED focuses on the simultaneous consideration of complex physico-chemical phenomena occurring at all length scales, such as catalytically-assisted chemical reactions, contact of rough surfaces, mechanical/chemical degradation of membranes, fluid flows in porous media etc., at affordable computational cost. Through a combination of multi-disciplinary computational methods with Machine Learning (ML), each length scale’s highly accurate model is bridged to adjacent scales. Following the development of a holistic framework, a top-down length scale approach is planned to be followed in order to optimise simultaneously the PEMFC and its sub-components.
Training is central to BLESSED. Fifteen doctoral researchers are working across disciplines — including chemistry, physics, computer science, and engineering — gaining hands-on expertise in electrochemistry, reacting flows, fluid mechanics, materials, optimisation methods and ML. These skills are critical for driving innovation in efficient, durable and affordable carbon-neutral powertrains.
By revolutionising the design process of next generation PEMFCs and accelerating the introduction of alternative powertrains for electrified vehicles, BLESSED directly supports the European Green Deal and the EU’s vision for carbon neutrality by 2050, promoting sustainable energy conversion, resource efficiency, and circular economy pathways toward a greener, more resilient future.
At its core, BLESSED focuses on the simultaneous consideration of complex physico-chemical phenomena occurring at all length scales, such as catalytically-assisted chemical reactions, contact of rough surfaces, mechanical/chemical degradation of membranes, fluid flows in porous media etc., at affordable computational cost. Through a combination of multi-disciplinary computational methods with Machine Learning (ML), each length scale’s highly accurate model is bridged to adjacent scales. Following the development of a holistic framework, a top-down length scale approach is planned to be followed in order to optimise simultaneously the PEMFC and its sub-components.
Training is central to BLESSED. Fifteen doctoral researchers are working across disciplines — including chemistry, physics, computer science, and engineering — gaining hands-on expertise in electrochemistry, reacting flows, fluid mechanics, materials, optimisation methods and ML. These skills are critical for driving innovation in efficient, durable and affordable carbon-neutral powertrains.
By revolutionising the design process of next generation PEMFCs and accelerating the introduction of alternative powertrains for electrified vehicles, BLESSED directly supports the European Green Deal and the EU’s vision for carbon neutrality by 2050, promoting sustainable energy conversion, resource efficiency, and circular economy pathways toward a greener, more resilient future.
In the first two years of the project, BLESSED has made substantial progress in advancing multi-scale modelling of PEMFC devices.
The scientific work has focused on advancing predictive capabilities across different length and time scales as a starting point of the multi-scale modelling. All doctoral candidates have initiated computational approaches tailored to their individual research projects including, but not limited to, membrane degradation modelling, transport modelling in porous media and electrochemical model of PEMFC cathode. A wide range of tools—such as OpenFOAM, LAMMPS, VASP, SIESTA, and ABAQUS—were employed to develop models spanning CFD simulations, MD studies of Nafion membranes, Pt catalyst chemistry, cDFT calibration, and DEM modelling of catalyst layers. Notable achievements include successful validation of the computational models, creation of machine learning interatomic potentials (MLIPs), and coupling of structural and fluid dynamics models, laying the foundation for integrated multi-scale workflows.
Work has been progressing also on the development of Machine-learning inspired models for PEMFC components, with initial efforts focused on data generation by molecular dynamics models and identification of key descriptors for PEMFC components. Several doctoral candidates are working on descriptor identification to enable accurate structure-to-property predictions, particularly for solid-liquid-gas (SLG) interactions relevant to electrocatalytic activity and transport properties. Advanced ML techniques—including convolutional neural networks (CNNs), Fourier neural operators (FNOs), physics-informed neural networks (PINNs), Gaussian process regression (GPR), support vector regression (SVR), random forest (RF) regressors, and deep neural networks (DNNs)—have been applied to learn transport coefficients, capture structure-property relationships, and optimize gas diffusion layer (GDL) microstructure.
These developments lay the groundwork for data-driven models that enhance predictive accuracy and accelerate design optimization across PEMFC components.
The scientific work has focused on advancing predictive capabilities across different length and time scales as a starting point of the multi-scale modelling. All doctoral candidates have initiated computational approaches tailored to their individual research projects including, but not limited to, membrane degradation modelling, transport modelling in porous media and electrochemical model of PEMFC cathode. A wide range of tools—such as OpenFOAM, LAMMPS, VASP, SIESTA, and ABAQUS—were employed to develop models spanning CFD simulations, MD studies of Nafion membranes, Pt catalyst chemistry, cDFT calibration, and DEM modelling of catalyst layers. Notable achievements include successful validation of the computational models, creation of machine learning interatomic potentials (MLIPs), and coupling of structural and fluid dynamics models, laying the foundation for integrated multi-scale workflows.
Work has been progressing also on the development of Machine-learning inspired models for PEMFC components, with initial efforts focused on data generation by molecular dynamics models and identification of key descriptors for PEMFC components. Several doctoral candidates are working on descriptor identification to enable accurate structure-to-property predictions, particularly for solid-liquid-gas (SLG) interactions relevant to electrocatalytic activity and transport properties. Advanced ML techniques—including convolutional neural networks (CNNs), Fourier neural operators (FNOs), physics-informed neural networks (PINNs), Gaussian process regression (GPR), support vector regression (SVR), random forest (RF) regressors, and deep neural networks (DNNs)—have been applied to learn transport coefficients, capture structure-property relationships, and optimize gas diffusion layer (GDL) microstructure.
These developments lay the groundwork for data-driven models that enhance predictive accuracy and accelerate design optimization across PEMFC components.
Our work has generated several breakthroughs in the physical and chemical modeling of PEMFC components. Large-scale molecular dynamics simulations of oxygen and water in hydrated Nafion membranes revealed fundamental transport mechanisms and enabled the derivation of diffusion laws for macro-scale models. We introduced a novel nanostructure-informed multiscale framework to predict proton conductivity under varying hydration, temperature, and chemical degradation conditions—integrating Coarse-Grained MD with Finite Difference and Random Walk models to quantify both vehicle and Grotthuss proton transport mechanisms for the first time. This approach uncovered the critical impact of chemical degradation on water retention and nanostructural heterogeneity, advancing understanding of durability-performance trade-offs. Additionally, we developed an end-to-end mesoscale DEM workflow to model Pt/C and ionomer aggregation dynamics during catalyst layer manufacturing, providing unprecedented insights into microstructural evolution as a function of solvent composition.
In the field of Machine Learning, we introduced physics-informed machine learning frameworks that fundamentally advance predictive modeling for PEMFC components. A key innovation is the augmentation of classical density functional theory (cDFT) with neural corrections trained on molecular dynamics data, embedding compact neural networks within Helmholtz free-energy functionals to preserve thermodynamic consistency while capturing missing correlations. This approach accurately predicts interfacial properties such as density profiles, surface tension, and droplet morphology far beyond the training regime, bridging particle and continuum scales. Additionally, we developed a high-accuracy ML-driven optimization strategy for gas diffusion layer (GDL) microstructure, achieving R² scores above 90% for multiple properties and reducing computational time from hours to seconds. Our models identified critical descriptors—such as fiber length, bond density, and fiber-to-bond ratio—that outperform traditional metrics like porosity in predicting electrical contact resistance under compression.
Finally, in the field of macro-scale optimisation, a new continuous adjoint model of PEMFC was developed, by formulating and implementing the adjoint transport equations.
Collectively, these innovations establish new modeling paradigms that bridge scales and inform the design of high-performance, durable PEMFC systems.
In the field of Machine Learning, we introduced physics-informed machine learning frameworks that fundamentally advance predictive modeling for PEMFC components. A key innovation is the augmentation of classical density functional theory (cDFT) with neural corrections trained on molecular dynamics data, embedding compact neural networks within Helmholtz free-energy functionals to preserve thermodynamic consistency while capturing missing correlations. This approach accurately predicts interfacial properties such as density profiles, surface tension, and droplet morphology far beyond the training regime, bridging particle and continuum scales. Additionally, we developed a high-accuracy ML-driven optimization strategy for gas diffusion layer (GDL) microstructure, achieving R² scores above 90% for multiple properties and reducing computational time from hours to seconds. Our models identified critical descriptors—such as fiber length, bond density, and fiber-to-bond ratio—that outperform traditional metrics like porosity in predicting electrical contact resistance under compression.
Finally, in the field of macro-scale optimisation, a new continuous adjoint model of PEMFC was developed, by formulating and implementing the adjoint transport equations.
Collectively, these innovations establish new modeling paradigms that bridge scales and inform the design of high-performance, durable PEMFC systems.