Neural Network-Based Hypersonic Crossflow Transition Model

Abstract

A model is proposed for predicting crossflow transition in hypersonic flows. The model is based on transport equations for the crossflow amplification factor and a modified intermittency. The instability onset is estimated with a correlation function. The form of the transport equations is identical to that of typical transport equations for Reynolds-averaged Navier–Stokes (RANS) turbulence models, thus facilitating an easy integration into existing RANS codes. The crossflow amplification factor is estimated from local flow quantities using a neural network. The neural network is trained with a comprehensive database of linear stability theory data. The sensitivity of the amplification factor growth rate prediction to different dimensionless input parameters is evaluated. The input parameters that provide the most-accurate growth rate predictions are selected for the final form of the model. Based on the volume of the training data set, the model is expected to perform reliably for Mach numbers between 5 and 11 and within given upper bounds of the stagnation temperature and pressure. The model is validated for the HiFiRE-5 geometry, a circular cone at angle of attack, and a blunt swept flat plate. The amplification factor distributions obtained from the model are in good agreement with experimental data and linear stability analysis results.