Turbulence model parameter calibration method based on the combination of deep neural network surrogate model and genetic algorithm in supersonic flow over cavity-ramp

Abstract

The traditional turbulence models have the problem of low accuracy and poor applicability of normal value when predicting complex separation flows (such as shock wave/turbulent boundary-layer interaction). Therefore, cavity-ramp is chosen as the research object in this paper, and a turbulence model parameter calibration method based on a combination of deep neural network surrogate model and genetic algorithm is proposed. The Latin Hypercube Sampling method is used to obtain the sample space of nine uncertain parameters of the SST turbulence model, and then the hypersonic inside-outflow coupled numerical simulation software (AHL3D) is used to carry out the calculation. The cavity-ramp wall pressure samples corresponding to different turbulence model parameters are obtained, which are used to construct a deep neural network turbulence surrogate model. Finally, through the deep neural network turbulence surrogate model and experimental wall pressure data, genetic algorithm is used to optimize and calibrate the turbulence model parameters. Experimental results show that the deep neural network turbulence surrogate model is highly accurate, with a coefficient of determination above 0.99 for the predicted wall pressure curve. At the same time, the computational time of the deep neural network turbulence surrogate model is on the order of milliseconds, which can considerably improve the acquisition efficiency of the wall pressure; In addition, the calibrated turbulence model is closer to the experimental data in calculating the wall pressure, which validates the feasibility of the method and is expected to improve the computational accuracy of the current turbulence models.