I am simulating incompressible flow over an articulated lorry using the steady-state simulator. I have meshed the geometry to what I believe is a good quality mesh and have applied appropriate boundary conditions.
I have run the simulation using the K-epsilon model since I am aiming for a Y+ of between 30 and 300. From my understanding, there is no need to use the K-omega SST turbulence model since I am not aiming to fully model the boundary layer.
In run 1, I kept the automatic numerics settings and ran 1000 iterations. This showed promising stability, however, the convergence is really slow and the drag coefficient is just slowly decreasing. From this, I gathered just more simulations are needed.
For the second run, I used the k-omega SST model and changed the relaxation factors to 0.3 0.7 0.3 and 0.3 respectively, as suggested by @Ricardopg in this post, and this led to a lot of instability.
For the third run, I went back to the automatic settings which showed better stability and ran the simulation with 1500 iterations. This still didn’t show convergence so I continued the simulation with a further 500 iterations which caused a spike in the residuals and data at first (not sure why).
I am still getting the same problem. The drag coefficient just seems to slowly keep decreasing and not converge.
From the experimental results I am using to validate my simulation, the drag coefficient I am expecting is around 0.72 and at the current iteration, the drag coefficient is 0.58 and seems that it will just keep decreasing with more iterations.
Not sure what else I can do to increase convergence. I am also unsure of why the drag coefficient is just slowly decreasing and doesn’t converge.
Will someone is able to have a look through the settings and mesh to see if I have made any mistakes or spot something that I can improve on please
Hi, this is Fillia from SimScale support, thank you for all the details!
My first question to you is about the meshing algorithm, how come you chose the Hex dominant instead of the Standard mesher? Did you give the latter a try?
Yes I have and I wasn’t getting a good quality mesh. I moved over to the hex dominant parametric as this is what I saw the majority of users use for similar simulations to mine and it gives you more refined controls over the meshing. I’ve made changes to the geometry since to improve meshing but I haven’t given the standard mesher another go since I switched over.
Will the type of meshing algorithm used have a significant impact on convergence? I assumed as long as the mesh is of relatively high quality then this shouldn’t matter.
We typically suggest to our users to try the Standard mesher, it is reliable and easy to use. Please have a look at this project for example: Aerodynamic Flow Behavior Around a Vehicle Tutorial | SimScale You can use the mesh settings from this (according to your own case), and create a new mesh with this algorithm just to check, what do you think?
Also, we use this algorithm for our validation cases, check here:
Okay, will give the standard mesher another go. `Looking at my project, were you able to find any obvious reasons why my drag coefficient is abnormally low? What are your thoughts on the mesh quality, boundary conditions and initial conditions?
You are using only moving wall, not rotating wheels?
How did you calculate the reference length and reference area?
Can you send me the experimental data please? It is important you are simulating exactly teh same conditions and model in order to compare the results.
I used rotating wheels for my all my previous simulations except the latest two. I was advised to attempt removing the rotating wheels to simplify the simulation and accelerate convergence as they would not play a big role in affecting the wake region of the truck.
The reference length and area was calculated exactly the same way as in the Aerodynamic Flow Behavior Around a Vehicle Tutorial. The reference length being the length of the vehicle in the flow direction and the reference area being the area projected by half the model when looking head-on.
In regards to the experimental data, the model used is pretty much identical except that it does not have a cab deflector. The drag coefficient measured for that model was 0.71. I’ve ran the same simulation but with a much coarser mesh and got a result of 0.42. Currently, the simulation with the finer mesh is showing a drag coefficient of 0.55 but does not seem to have converged yet. Although the models do not exactly match, the use of a cab deflector shouldn’t result in such a massive drop in the drag coefficient.
