MultiFluidX will be at the IACM – 3rd Digital Twins in Engineering Conference (DTE 2025) and ECCOMAS – 1st Artificial Intelligence and Computational Methods in Applied Science (AICOMAS 2025), DTE & AICOMAS conference in Paris, 17-21 February 2025.
DTE-AICOMAS 2025 is a great opportunity to present the latest version of #MPflow and its new #AI capabilities for #CFD to experts in Digital Twins in Engineering, Artificial Intelligence and in Applied Science.
Our work is entitled: “A Machine learning Computational Fluid Dynamics solver for simulating flashing jets” and will showcase our contribution in recent developments on new ML-CFD technologies.
Flash-boiling of superheated liquid jets is a common phenomenon observed in many industrial applications such as accidental releases of flammable and toxic pressure-liquefied gases and fuel injection in engines. During flashing, the liquid undergoes a rapid pressure drop inside the nozzle and becomes superheated. Such superheated liquid, in the form of either a jet or droplets, will lead to explosive atomisation with fine droplets and a short spray distance. This makes flashing jets very appealing to modern fuel injection systems for hydrogen engines and their key role in the decarbonisation plans of the EU. Computational fluid dynamics (CFD) is the most popular tool for modelling and simulating flashing jets inside nozzles.
CFD simulation of flash-boiling poses many challenges since the complex underpinning physics requires various numerical models for modelling heat and mass transfer. Most notably, CFD simulations require long computational times. Intermediate steps such as volumetric meshing in mesh-based methods can also significantly increase the computational cost.
The work presented, aims at providing academia and industry with a modelling tool to simulate and investigate the complex multi-facet phenomenon of flash-boiling atomisation deploying a novel machine learning method that could save thousand Central Processing Unit (CPU) hours offering instantaneous CFD predictions. Two techniques are presented and implemented in the ML-CFD software MPflow®: in the first technique, the presented machine-learning-CFD method completely replaces the traditional CFD simulations workflow and requires little simulation expertise from the end-user. This is a novel model that couples for the first time the thermodynamic non-equilibrium (HRM) with convolutional neural networks to simulate flashing liquid hydrogen jets thousand times faster that the standalone CFD solver. In the second technique, the user can bypass only specific parts of the pressure-velocity coupling when solving the Navier-Stokes equations. This is done here by replacing the equations for turbulence models, completely substituting them with a neural network. The accuracy of the novel approaches is evaluated, demonstrating adequate accuracy compared to different unseen simulations and experiments. This work offers the groundwork for further accelerating CFD predictions in multiphase flow problems and could significantly improve testing flash-boiling scenarios in various industrial settings.
See you in Paris !!!
