Adaptive mask physics-driven neural network for complex flow field prediction using self-supervised learning

Zékai Lu

doi:10.1016/j.ast.2025.110966

Adaptive mask physics-driven neural network for complex flow field prediction using self-supervised learning

2025-09-24

0

PRE

AI

DOI：10.1016/j.ast.2025.110966

Original

Original request for help

Save

Abstract

En 中文

• AMPD-NN achieves unified multi-airfoil modeling across thickness ratios 6 %−15 % and camber 0 %−4 %. • Self-supervised physics-driven training eliminates dependence on extensive labeled datasets. • Maintains 4 % prediction error in subsonic range with 78–109 × computational speedup over CFD. • Demonstrates controlled extrapolation capability with 15–16 % error at 40 % extrapolation degree. • Variable wall condition processing enables single model adaptation to diverse airfoil geometries.

Adaptive mask physics-driven neural network for complex flow field prediction using self-supervised learning

Abstract

Journal

Organization