POD-based surrogate modeling of transitional flows using an adaptive sampling in Gaussian process

Abstract

A surrogate model, based on proper orthogonal decomposition (POD) with the adaptive sampling method, was proposed to predict the transitional flow past rough flat plates simulated by a four-equation k-ω-γ-Ar transition model. Gaussian process regression was used to map the input parameters to the POD expansion coefficients. The variance and gradient of Gaussian process were taken as the criteria for the adaptive sampling. The proposed methodology was applied to a one-dimensional heat conduction problem and two-dimensional transitional flow past rough flat plates. At the same time, the results were compared with those of Halton sequences. With the same sample size, the adaptive method achieved a higher accuracy on the test set, and the proposed adaptive criterion could serve as an indicator for the model discrepancies.