A lagrangian AI-based CFD emulator with reconstruction of Hydrodynamic forces components

• AI-based CFD Lagrangian emulator that reconstructs the unknown individual terms of the viscous Navier-Stokes equation. • The emulator learns from total accelerations of the fluid and reproduces pressure and viscous forces independently. • The proposed approach isolates the physical terms of the model and retrieves unknown quantities. • Applications of changing the simulated fluid viscosity without dedicated training are shown. The growing combination of Artificial Intelligence (AI) and Computational Fluid Dynamics (CFD) represents an important paradigm shift from traditional numerical methods, where fluid dynamics are explicitly modeled, to data-driven approaches, where physics is inferred from data. AI-based CFD emulators learn from existing CFD simulations to replicate the behavior of numerical models, providing enhanced capabilities. Emulators can be faster and take advantage of the data-based approach to manage physical properties that are not explicitly modeled. Here, we present an AI-based CFD emulator for Lagrangian simulations of viscous fluids. This emulator learns fluid dynamics from Weakly Compressible Smoothed Particle Hydrodynamics (WCSPH) simulations and reconstructs physical quantities not explicitly available in the reference data. Building on our previous Artificial Neural Network (ANN) SPH emulator for particle interaction forces, we now extend the approach to enhance modeling functionalities. Specifically, the emulator reproduces fluid behavior and separately reconstructs unknown pressure and viscous hydrodynamic forces for individual particle pair interactions. We verify the reconstruction accuracy against reference CFD data and demonstrate the emulator’s capability to reproduce fluids with various viscosities without retraining.