Efficient prediction method for aerodynamic heating in hypersonic cone boundary-layer transition

The objective is to develop a reliable prediction method to reduce the redundancy in thermal protection system design, thereby decreasing the cost and weight of hypersonic vehicles while enhancing their design efficiency and performance. existing prediction techniques such as engineering algorithms, wind - tunnel tests, and numerical calculations have limitations. hence, this research aims to establish a more effective prediction model.

The methods employed were as follows: First, a VAE(variational autoencoder) - based generative prediction framework was established. Principal component analysis was used to determine the optimal dimension of the latent variable space of the VAE model, which was set to 16. The VAE model, with a structure based on a convolutional neural network and including a residual convolutional encoder and decoder, was trained to extract low - dimensional latent space representations of complex heat flux fields. A fully - connected neural network was then constructed to establish the nonlinear mapping between free - stream parameters and latent variables. The two models were cascaded to form a hypersonic cone transition heat flux prediction model. Additionally, computational fluid dynamics using the Fluent software were carried out on a conical model with specific geometric parameters and flow conditions to generate the training and validation datasets.

The results show that the VAE model can effectively extract heat flux field latent variables and accurately reconstruct the heat flux field structure of the leeward - side streamwise vortex transition. The prediction model can efficiently learn the heat flux distribution characteristics under complex transition mechanisms. The average reconstruction error of the VAE model on the training set is less than 0.03, and on the validation set is less than 0.028. The prediction error of the integrated model for heat flux under different free - stream conditions is not higher than 0.024, indicating high - precision prediction capabilities.

In conclusion, this study successfully combines generative deep - learning methods with a transition heat flux database from numerical simulations. The established artificial intelligence model can accurately and efficiently reconstruct and predict the three - dimensional boundary - layer transition heat flux field of hypersonic cones under different free - stream conditions. The VAE model demonstrates excellent feature extraction and reconstruction abilities, and the fully - connected neural network effectively maps free - stream parameters to latent variables. This research provides a new approach for hypersonic aerodynamic heating prediction, which has important implications for the design and development of hypersonic vehicles.