Support:  LG Electronics

Period of Performance: 11/01/2025 – 10/31/2026

Budget: $122, 000

Summary: The project considers a unifying approach combining the PINN and the DeepONet methods, namely Physics-Informed DeepONets (PI-DeepONets), to solve Reynolds-averaged Navier–Stokes (RANS) equations. Traditional numerical methods face computational challenges in solving RANS equations, more generally, in enabling real-time computational fluid dynamics (CFD) simulations of turbulent flows. These challenges include but are not limited to grid sensitivity, numerical diffusion, and convergence difficulties. The goal of the proposed research is to develop an AI model in a completely data-free manner, by means of physics-informed neural operators (e.g. PI-DeepONets), that simulates RANS in real-time, replacing time-consuming traditional CFD simulations. When it is compared with the standard approach of operator learning, the physics-informed one does not require even a single datum, typically generated from running expensive traditional solvers. To this end, a well-trained AI model (e.g. PI-DeepONets) is expected to be deployed in real industrial applications.