here is an overview of the problem:
DPM results change with CFD mesh resolution
I have cylindrical walls (representing screen wires) that are subtracted from the bulk air duct. The wire diameter is 100 microns and the screen opening is 200 microns. My domain is purely made of air with the walls subtracted. The walls are solid and not porous and a "trap" DPM boundary condition is applied to them. I am trying to count the number of the particles that will hit the walls and then will be eliminated from the domain and will be counted as captured particles in the report.
Model (Eulerian and Lagrangian)
Ansys Fluent 2020 R2 is used. A steady K-Omega SST turbulent model is used to solve the flow. There is a one-way interaction between the particles and the flow. Therefore, the flow is solved first and after the convergence (errors down to 1e-10 or lower), the particles are injected at the inlet using surface injection (about 700 parcels) with the same velocity as the flow at the inlet. the particle size is 1 micron. Random walk model in DPM settings is off.
velocity inlet (1 – 15 m/s) pressure outlet (0 gauge) Symmetry walls on 4 side walls of the duct, and no-slip walls for the screen walls.
Using the "body of influence" meshing feature, I give Fluent Meshing software two mesh element sizes: one in the vicinity of the walls (called vicinity region) and one away from that (which I call up/downstream). The cell size in the vicinity zone is 4e-6m. the cell size in the up/downstream zone varies between 16e-6, 8e-6 or 4e-6 for 3 different cases. 7 layers of inflation are added.
The problem is that the number of trapped particles on the walls dramatically change when mesh size changes. I have checked the Z-velocity (the velocity along the duct) as a grid study parameter and the graphs align very well for different mesh size cases (please refer to the following photo). So, the grid resolution seems to be sufficient for the Eulerian phase. However, DPM results are different and we know that for Lagrangian solutions, as long as the flow is converged and mesh independent (for calculation of the drag forces), DPM results should not vary with the grid size since it calculates the particles' trajectories by solving F=ma for each particle along its path and is a mesh-less method.
Here are some of the particle trap results for different mesh resolutions:
What I have tested so far:
1- Grid-size independency
In order to make sure that the flow solution is mesh independent, for different mesh results, I drew a line parallel to the airflow direction and plotted Z_velocity-position on that line. Moreover, to study the velocity field close to the walls, I drew a ring with a diameter of 500 microns around one of the third layer wires and again, plotted Z_velosity on that ring for different mesh size cases. Both the plots indicate a very well alignment of the data for all the mesh cases. Therefore, it seems that the Eulerian phase is mesh-independent now.
2- Number of injected parcels
I know that a large number of parcels is required to get statistically satisfying results. In general, I am injecting 716 particles using the surface injection which will inject a parcel from the center of each mesh element on the inlet plane. To test whether it is enough, using the group injection feature, I once injected 1900 particles and then injected half of that number, 950 particles. The percentage of trapped particles was exactly the same for both of the injections. Please refer to the screenshot. therefore, I believe the number of injected particles is sufficient.
1- The impact of random walk model
I have recently realized that the DPM results are similar for different mesh sizes when Random Walk model is activated. The results of random walk model on-off are very different. The incoming turbulence intensity is about 4% (depending on inlet velocity) and drops to the order of 10^-2 at the outlet since the mesh layers damp turbulence. Although there seems to be no problem with the random walk model activated, one of my study goals is to investigate the impact of this model. So, I need to get the model giving reasonable results when Random Walk model is off, too.