Artificial intelligence, such as promising ML tools, has shown potential to revolutionise fuel design formulation by building accurate predictive models for key physicochemical properties of fuel utilisation while also helping to screen fuel mixtures in inverse design formulations. Integrated advanced AI tools can be used to leverage fuel screening, significantly dwindling development timelines and costs compared to the conventional fuel design approach. By leveraging large datasets, AI can accurately predict critical physicochemical properties of fuels, including viscosity, density, energy content, along with combustion characteristics and emissions. ML and big data analytics enable mapping fuel composition to key utilisation properties. Additionally, AI-assisted inverse fuel design frames fuel mixture development as a constrained optimisation problem, identifying compound combinations to achieve the desired properties. These advancements accelerate fuel formulation screening, significantly reducing development timelines and cost. For further details, check our paper published at Advances in Applied Energy: Pathways to sustainable fuel design from a probabilistic deep learning perspective

Understanding the thermal decomposition mechanisms of small molecules such as ethylene (C₂H₄) and ammonia (NH₃) is essential for applications in combustion chemistry, astrochemistry, and industrial catalysis. In this study, we employ high-dimensional neural network potentials (NNPs) trained on ab initio data to investigate the atomic-scale dynamics and reaction pathways involved in the thermal decomposition of ethylene and ammonia. The NNPs facilitate large-scale molecular dynamics (MD) simulations with near-quantum mechanical accuracy at a fraction of the computational cost, enabling us to explore complex reaction landscapes and identify key intermediates and transition states. Our simulations reveal detailed decomposition pathways, including the formation of radicals such as H·, NH₂·, and CH₂=CH· under varying temperature conditions. Reaction rates and activation energies derived from trajectory analysis show excellent agreement with available experimental and high-level theoretical data. This work demonstrates the effectiveness of machine learning potentials in modelling high-temperature reaction dynamics and provides new insights into the decomposition mechanisms of fundamental chemical species. For further details, check our paper published at Energy & AI: Neural network potential-based molecular investigation of thermal decomposition mechanisms of ethylene and ammonia

Accurate modelling of turbulent combustion remains a significant challenge due to the complex interplay between turbulence and chemical kinetics. Reactor-based closures, such as those employing perfectly stirred reactors (PSRs) or plug flow reactors (PFRs), provide a reduced-order framework to approximate the effects of unresolved turbulence-chemistry interactions in large-eddy simulations (LES). However, identifying appropriate functional forms for these closures remains largely empirical and limited in generalizability. Novel data-driven approaches for model discovery can be used to identify interpretable, compact, and physics-consistent closures for combustion source terms and scalar quantities. Leveraging high-fidelity simulation data from canonical reactor configurations, we employ symbolic regression to extract algebraic representations of combustion closures as functions of key control variables such as mixture fraction, scalar dissipation rate, and temperature. The resulting models achieve high accuracy with minimal complexity and retain physical interpretability, enabling seamless integration into existing CFD frameworks. These approaches open new avenues for automated, physics-informed model discovery in turbulent reacting flows, bridging the gap between first-principles modelling and data-driven techniques. For further details, check our papers published at Combustion & Flame: Model identification in reactor-based combustion closures using sparse symbolic regression ; and Flow, Turbulence and Combustion: A Data-Driven Approach to Refine the Partially Stirred Reactor Closure for Turbulent Premixed Flames .

The physicochemical properties of liquid alternative fuels are important yet challenging to measure or predict, especially when dealing with complex surrogate fuels. Machine learning can be utilised to develop quantitative structure-property relationship (QSPR) models, and an unprecedented mapping between the fuel composition and key properties of fuel utilisation can be established, where the chemical structure of complex fuels can be described using molecular representations. Such QSPR models can reveal the dependence of physicochemical properties of liquid synthetic fuels on fuel mixture chemical composition/chemical structure, which can potentially lead to fully sustainable fuel utilisation, i.e. the engines are 100% powered by sustainable/renewable synthetic fuels. For further details, check our paper published at Energy & AI: Descriptors-based machine-learning prediction of cetane number using quantitative structure–property relationship