Heat Flux Prediction of Radiation Balance Wall by Deep Convolutional Neural Networks

Abstract

Aerodynamic thermal prediction plays a crucial role in the design of a hypersonic vehicle, particularly with regard to the thermal protection system. Traditional methods of aerodynamic thermal prediction encounter several primary challenges, including slow convergence rates, rigorous computational grid requirements, and the need to simplify by assuming isothermal wall conditions. In this research, we propose using the Convolutional Neural Network (CNN) Hybrid Feature (HF) model to facilitate rapid aerothermal predictions for both isothermal wall conditions with varying wall temperatures and radiation balance wall conditions. The CNN HF model is trained separately for isothermal wall conditions under identical inflow conditions as well as for diverse inflow conditions and radiation balance wall temperature scenarios. The model’s predictions are then compared to numerical simulation results. Our findings demonstrate that the CNN HF model efficiently provides rapid aerothermal predictions by leveraging macroscopic converged flowfield data. In the majority of cases, the model achieves a threefold enhancement in computational efficiency while maintaining predictive accuracy within a 5% range when compared to numerical simulation results. The application of the CNN HF approach in aerothermal prediction for different wall temperatures and radiation balance scenarios has significantly reduced the time required to obtain aerodynamic heating results.