面向内流的激波/边界层湍流模型数据同化及应用

Data assimilation and application for turbulence models of shock wave/boundary layer interactions in inward flows

为研究压缩拐角激波/边界层干扰问题和机器学习方法在湍流模型参数辨识中的有效性,提出一种面向内流的激波/边界层湍流模型数据同化方法,以Kriging代理模型传播参数不确定量化过程,基于贝叶斯框架构建似然函数作为评判标准,最后利用粒子群优化算法近似获取参数的最大似然估计并进行参数验证。结果表明,通过校准大角度(24°)压缩拐角获取的湍流模型参数,可以应用到相同条件下相对较小的小角度(20°,16°和8°)压缩拐角,获取的壁面压力、摩阻系数和速度剖面均与试验值基本吻合。在Ma=2.85下校准的壁面压力,均方根误差由60.29%下降到16.56%。将大角度下获取的参数应用到Ma=2.9和不同的入射边界层厚度的条件下,获取的壁面压力和速度剖面仍与试验值基本吻合,验证了小范围马赫数内湍流模型参数的适用性。

To investigate the shock wave/boundary layer interactions in compressed corners and the effec⁃tiveness of machine learning in turbulence model parameter identification,a data assimilation method for shock/boundary layer turbulence models tailored to internal flows is proposed.This approach first propagates the uncer⁃tainty quantification process of propagation parameters through Kriging surrogate models.Then,the evaluation cri⁃teria are constructed by likelihood function constructed based on Bayesian framework.And finally,Particle Swarm Optimization(PSO)is employed to approximate maximum likelihood estimates of parameters for subse⁃quent validation.The results indicate that the parameters of turbulence model calibrated using a large-angle(24°)compression corner can be applied to smaller angles(20°,16°,and 8°)under similar conditions,with the calculated wall pressure,friction coefficients,and velocity profiles correspondingly aligning well with experi⁃mental data.Calibrated wall pressure at Ma=2.85 exhibited a reduction in root mean square error from 60.29%to 16.56%.When parameters obtained from large angles are applied at Ma=2.9 with different incident boundary lay⁃er thicknesses,the resulting wall pressure and velocity profiles remain largely consistent with experimental out⁃comes,that effectively verifies the applicability of turbulence model parameters within a small range of Mach numbers.