Sim Specialists recently traveled to Mexico City to deliver onsite FEA and CFD simulation services....
Validating CFD Simulation, Part 2 -- Mesh Refinement
Note: This is a continuing series on CFD validation. Please subscribe to the blog or check back on a regular basis for previous and subsequent installments.
In Part 1 of this series, we explained why validation is so important when using Computational Fluid Dynamics (CFD) as an effective engineering tool. One approach to CFD software validation is to benchmark it directly against reliable physical test results.
For Part 2, the impact of mesh refinement on the fidelity of results will be determined by comparing the CFD results to the classic testing performed by Restivo ("Turbulent Flow in Ventilated Rooms, 1979"). The setup of the test model, which is a common reference in CFD whitepapers, is depicted in Figure 1. Air is introduced into the space at a velocity of 0.455 m/s at the inlet in the upper left corner. Note that this example is also a great validation benchmark for the many AEC simulations that we perform.
Figure 1: Dimensions of rectangular cavity
The software used is Autodesk CFD (formerly known as CFdesign), which uses a segregated Navier-Stokes solver and has a variety of turbulence models, where k-epsilon (used here) is the default. A global mesh refinement strategy was employed to simulate three different automatic mesh sizes: default autosize, 50% refinement and 25% refinement.
Figure 2: Autosize, 50% and 25% mesh refinement
When comparing the results of these three different simulation runs to the physical test, the impact of mesh refinement can clearly by seen. In Figures 3 and 4, correlation with testing improves with increasing mesh density, a process known as mesh convergence.
Figure 3: Vertical velocity profile @ X = 3 meters
Figure 4: Vertical velocity profile @ X = 6 meters
The key takeaway is that a proper mesh is a fundamental requirement for achieving the desired level of accuracy. The accuracy required will vary based on whether a quick qualitative assessment is needed or if the objective is a more stringent quantitative comparison.
There are other aspects to the mesh (e.g., boundary layers) and alternative refinement approaches (e.g., adaptive meshing) that will be examined in upcoming installments of this blog.