Direct numerical Simulation is the most accurate method for simulating turbulent flows which require very dense discrete meshes and is therefore computationally very expensive. Large Eddy Simulation method (LES) is performed on meshes that are up to ten times coarser in each direction than the ones used in the DNS, which significantly decreases computation time. LES is based on semi-empirical models, which approximately describe turbulent diffusion in the smallest eddies that are not captured by the discrete grid. In the master’s thesis, we are analyzing the accuracy of the LES method through a comparison with existing results that are computed with the DNS method in turbulent flow geometry in a backward facing step domain. Our LES was running on approximately 15 times coarser mesh and consumed two orders of magnitude less computation time than the DNS. Simulations are comparable since they have the same Reynolds number, which is 7100. We are comparing the time-averaged velocity fields in different planes, which show considerable similarities between the methods in the flow throughout the step. Small but noticeable differences in dimensions and shapes of the vortices are attributed to the inaccuracy of the large eddy method. Additionally, we evaluated the accuracy of the LES through the comparison of time-averaged velocities and their fluctuations on selected line segments in the domain after the step. From the latter, we can conclude that the flow in the LES is faster in the center of the domain and slower near the walls in comparison to the flow in the DNS. Although the LES method uses empirical models only for the smallest eddies, this approximation affects the geometry of the largest vortices in the system.
|
