AI-based emulators for fast simulations of lava flows.
Amato E., Zago V., Del Negro C.
Lava flows are complex fluids including very high viscosities, solid-fluid interactions, and free-surface flows. The simulations of these flows constitute a challenge for Computational Fluid Dynamics (CFD). Smoothed Particle Hydrodynamics (SPH), a Lagrangian mesh-free numerical method based on a discrete approximation of the Navier-Stokes equations, is a consolidated approach for addressing this kind of problems. However, SPH simulations require long run times and large computational resources. Any speed-up of the simulations is usually obtained by simplifying the model or boosting the computational resources. Here, we discuss a change of paradigm, where instead of relying on upgrading the hardware to make computations faster, we use Artificial Intelligence (AI) to reduce the amount of computation needed in the first place. AI algorithms can be trained over SPH simulated data to emulate the behavior of the model in a faster way. Models obtained in this way are called emulators. Here, we present an AI-based emulator for a SPH model, and validate it with respect to some benchmark tests for viscous fluids, and compare the results with those obtained from the SPH model.