Author(s): Linlin Yin; Zhenju Chuang; Ankang Hu
Linked Author(s):
Keywords: 2D hydrofoil; Hydrodynamics; Physics-informed neural networks; Parameter identification
Abstract: This study presents a physics-informed neural network (PINN) framework designed for the efficient prediction of the hydrodynamic performance of two-dimensional hydrofoils. Using the NACA0020 hydrofoil as a case study, initial hydrodynamic performance data were generated across a range of attack angles (α) from 0 to 12 degrees utilizing the computational fluid dynamics (CFD) software STAR-CCM+. This process produced high-fidelity flow field data, encompassing velocity components (u, v) and pressure (p). These datasets were subsequently employed to train the PINN model, with spatial coordinates (x, y, α) serving as inputs and the corresponding u, v velocity components and pressure p of the flow field as outputs. In comparison to conventional CFD methods, the proposed PINN framework substantially reduces computational expenses while preserving prediction accuracy. By integrating the Navier-Stokes equations into the loss function, PINNs diminish their dependence on extensive labeled datasets and are capable of deriving solutions that adhere to physical principles. The principal aim of this research is to demonstrate the efficacy of the PINN framework in addressing complex hydrodynamic challenges and to investigate its potential for optimizing hydrofoil design. The results indicate that the proposed PINN framework accurately captures the flow field characteristics of the NACA0020 hydrofoil across various attack angles, thereby providing a novel tool for the efficient performance assessment and rapid iterative design of hydrofoils.
Year: 2026