Heat flux estimation of the cylinder in hypersonic rarefied flow based on neural network surrogate model

Abstract

An efficient method to predict thermal loads on hypersonic vehicles in rarefied flows is immediately needed, especially when designing the thermal protection system. To meet the demand, we combine artificial neural networks with the direct simulation Monte Carlo method and build the surrogate model for hypersonic rarefied flows with three inputs (Knudsen number, temperature ration, and Mach number). The heating coefficients at nine points along the surface of a two-dimensional cylinder are output from the model. The results at the stagnation point have errors within 3%, while the biggest error of nine points is 4.8%. The heating coefficients are also compared with the bridge function’s, whose errors reach 14% at the stagnation point and 20% along the surface. The reasons for the errors are discussed in detail. In addition, this framework of building the model with artificial neural networks can be extended to solve problems with more complex mechanisms or configurations.