Efficient uncertainty quantification of stochastic problems in CFD by combination of compressed sensing and POD-Kriging

This paper proposes an uncertainty quantification method that combines compressed sensing and POD-Kriging that inherits the benefits of each key element. The compressed sensing is used to calculate the covariance matrix from a coarse grid. POD is used to significantly reduce the number of response surface constructions. The Gaussian process is used to compensate for the lack of predictive ability of the trend function in the highly non-linear region, thereby improving the ability to predict the local details of the flow field. Two CFD test cases are used to test the effectiveness and accuracy of the method proposed in this paper. The test cases examine the compressible flow around NLR7301 wing and heat removal by forced convection in a cooling system using fins. The results show that compared to full PCE, the compressed sensing combined with POD-Kriging method can significantly reduce the number of low and high-fidelity CFDs required to quantify uncertainty with the same computation accuracy, and computational cost is significantly reduced.