Abstract This paper presents a Gaussian process regression inspired method to measure the agreement between experiment and computational fluid dynamics (CFD) simulation. Because of misalignments between experimental and numerical outputs in spatial or parameter space, experimental data are not always suitable for quantitative assessing the numerical models. In this proposed method, the cross-validated Gaussian process regression (GPR) model, trained based on experimental measurements, is used to mimic the measurements at positions where there are no experimental data. The agreement between an experiment and the simulation is mimicked by the agreement between the simulation and GPR models. The statistically weighted square error is used to provide tangible information for the local agreement. The standardised Euclidean distance is used for assessing the overall agreement. The method is then used to assess the performance of four scale-resolving CFD methods, such as URANS k-ω-SST, SAS-SST, SAS-KE, and IDDES-SST, in simulating a prism bluff-body flow. The local statistically weighted square error together with standardised Euclidean distance provide additional insight, over and above the qualitative graphical comparisons. In this example scenario, the SAS-SST model marginally outperformed the IDDES-SST and better than the other two other, according to the distance to the validated GPR models.

中文翻译：

---

使用高斯过程回归启发的方法来测量实验和 CFD 模拟之间的一致性

摘要 本文提出了一种受高斯过程回归启发的方法来衡量实验与计算流体动力学 (CFD) 模拟之间的一致性。由于空间或参数空间中实验和数值输出之间的错位，实验数据并不总是适合定量评估数值模型。在该方法中，基于实验测量训练的交叉验证高斯过程回归 (GPR) 模型用于模拟没有实验数据的位置处的测量。实验和模拟之间的一致性通过模拟和 GPR 模型之间的一致性来模仿。统计加权平方误差用于为本地协议提供有形信息。标准化欧几里得距离用于评估整体一致性。然后，该方法用于评估四种尺度解析 CFD 方法（例如 URANS k-ω-SST、SAS-SST、SAS-KE 和 IDDES-SST）在模拟棱柱钝体流动时的性能。除了定性图形比较之外，局部统计加权平方误差和标准化欧几里得距离提供了额外的洞察力。在这个示例场景中，根据与经过验证的 GPR 模型的距离，SAS-SST 模型略微优于 IDDES-SST，并且优于其他两个模型。除了定性图形比较之外，局部统计加权平方误差和标准化欧几里得距离提供了额外的洞察力。在这个示例场景中，根据与经过验证的 GPR 模型的距离，SAS-SST 模型略微优于 IDDES-SST，并且优于其他两个模型。除了定性图形比较之外，局部统计加权平方误差和标准化欧几里得距离提供了额外的洞察力。在这个示例场景中，根据与经过验证的 GPR 模型的距离，SAS-SST 模型略微优于 IDDES-SST，并且优于其他两个模型。