I have been getting started by trying to modify the 2d-inc-cylinder tutorial to be a flow around a rhombus instead of around a round cylinder. To do so I produced a new mesh using Gmsh but otherwise ran the same configuration (.ini file). However, I observed numerical artifacts, specifically negative pressures at the four corners (wall and outlet).
I believe the physics should not be too different, so I imagine the issue is the mesh. Because I am using a rhombus obstacle, I made a first-order mesh rather than the second-order one given in the demo. At a glance the triangle sizes look roughly similar.
Is there any guidance about mesh generation for this basic setting? A text or programmatic description about how the original mesh was generated might also suffice. Here is a link to the rhombus mesh I have been using.
In general you should not expect configuration rules tuned for one case to directly port other to another. My first suggestion would be to look at the convergence history of the pseudo time steps.
Thank you for the quick response. I’m not entirely sure what you mean by convergence history, but I tried decreasing both dt and pseudo-dt by 10x: the pressure residual is noticeably (1.5-2x) worse throughout the simulation, whereas the velocity residuals go down by 3x by the end of the simulation. The negative pressure artifacts remain.
Is there a possibly slower but more robust scheme that would work for similar-enough meshes? I’m investigating the possibility of using of PyFR for a learning application in which I would need to run thousands of related simulations, so tuning them separately is not really feasible.
I had a look at the mesh and it looks very coarse (there are only two elements along each side of the diamond shape). There will likely be very strong spatial gradients as flow separates from the lower and upper points of the diamond shape, which I don’t think this mesh will be able to resolve. What Reynolds number are you running at?
Thank you for taking a look. Since my initial post I realized the top and bottom boundary conditions should be inlets (not walls) and have tried significantly refining the mesh around the obstacle (here). Running this with a 20x smaller dt and pseudo-dt (dt=0.0025 and pseudo-dt=0.00025) still leads to substantial negative pressures in the wake region and top, bottom, and right corners of the obstacle.
I’m using the same flow velocity, kinematic viscosity, and object diameter as the original simulation, so Re=200.
Sure, below is a zoom in on the pressure around the obstacle at the last timestep (t=75). My concern stems from the magnitude of the negative pressure being comparable to that of the surrounding positive pressure, and so I’m not sure it’s something that can be safely clipped to zero (as maybe it could be if it was an order-of-magnitude or two smaller). In my case I will likely be interested in the pressure quantity (in addition to the velocity) as a simulation output.
