摘要
This study evaluates the recently developed general framework for solution verification methods for large eddy simulation（LES） using implicitly filtered LES of periodic channel flows at friction Reynolds number of 395 on eight systematically refined grids.The seven-equation method shows that the coupling error based on Hypothesis I is much smaller as compared with the numerical and modeling errors and therefore can be neglected. The authors recommend five-equation method based on Hypothesis II, which shows a monotonic convergence behavior of the predicted numerical benchmark（S_C）, and provides realistic error estimates without the need of fixing the orders of accuracy for either numerical or modeling errors. Based on the results from seven-equation and five-equation methods, less expensive three and four-equation methods for practical LES applications were derived. It was found that the new three-equation method is robust as it can be applied to any convergence types and reasonably predict the error trends. It was also observed that the numerical and modeling errors usually have opposite signs, which suggests error cancellation play an essential role in LES. When Reynolds averaged Navier-Stokes（RANS） based error estimation method is applied, it shows significant error in the prediction of S_C on coarse meshes. However, it predicts reasonable S_C when the grids resolve at least 80% of the total turbulent kinetic energy.
This study evaluates the recently developed general framework for solution verification methods for large eddy simulation（LES） using implicitly filtered LES of periodic channel flows at friction Reynolds number of 395 on eight systematically refined grids.The seven-equation method shows that the coupling error based on Hypothesis I is much smaller as compared with the numerical and modeling errors and therefore can be neglected. The authors recommend five-equation method based on Hypothesis II, which shows a monotonic convergence behavior of the predicted numerical benchmark（S_C）, and provides realistic error estimates without the need of fixing the orders of accuracy for either numerical or modeling errors. Based on the results from seven-equation and five-equation methods, less expensive three and four-equation methods for practical LES applications were derived. It was found that the new three-equation method is robust as it can be applied to any convergence types and reasonably predict the error trends. It was also observed that the numerical and modeling errors usually have opposite signs, which suggests error cancellation play an essential role in LES. When Reynolds averaged Navier-Stokes（RANS） based error estimation method is applied, it shows significant error in the prediction of S_C on coarse meshes. However, it predicts reasonable S_C when the grids resolve at least 80% of the total turbulent kinetic energy.
