at the moment I am testing some simulation runs for natural convection. The testing object is a block made of copper hanging in the air. Its a quite simple setup:
- No extra contact definitions, all contacts are bounded
- Standard Initial Conditions
- Heat Source of the block (3.5 Watt)
- Standard Boundary Conditions for Natural Convection (Made by your User Guide)
- At first Standard Numeric settings
- 2000 iterations
I did two runs for now. Both worked without any error.
Now to my question:
During the first Run the Residuals are oscilating. So I changed the relaxation settings a little and started Run 2. The Residuals are more or less constant now, but they are still very high.
How can I decrease them? Especially the pressure residual is really high…
I added two screenshots of the residuals.
Thanks for your help!
here is the project link.
Hey @mloft and sorry for getting back to you quite late!
You assumed that the flow is steady-state, why? Oscillations can occur if the flow is unsteady from nature but your boundary conditions are (at least at first glance) constant in time thus not time dependent. However we could argue that the solver tries to find a steady state solution in an asymmetric mesh and/or that the mesh does not properly resolve gradients in the temperature field.
One way we could approach this is to run the simulation with a transient setting and see if there’s any change . Numerically you can try to lower the relaxation factor (if the factor is < 1 then we have under-relaxation) which might decrease the oscillations but slows down the speed of convergence (as you see there’s a trade-off). If I have missed something here I would be happy if our CFD Squad can jump in to add their two cents.
Thanks for your reply. From my point of view the flow condition is at the frontier between steady state (laminar) and transient flow (concerning the to Rayleigh number). But I hope I can deal with a steady state simulation to save extra computation time.
I will give the reduction of the relaxation factors a try. And I will use a more accurate mesh.
Do you have any other ideas what might cause these high residuals?
Personally I would go with under-relaxation first and then see if numerical schemes can fix this issue. Keep us up-to-date and we will follow up with the numerical schemes if that is even necessary. I mean if you fall below a certain level (which looks quite good in your case) we can assume the solution to be converged however I would say that the nature hides something transient Let’s see what my colleagues the PowerUsers have/want to add here.
thanks again for your reply!
I will give the underrelaxation a try. But I got two question, first, what exactly do you mean with numerical schemes? And Second (other context now), is it possible to use the boussinesq approximation in CHT Analysis? The material properties are quite weird, e.g. I have no clue what “amount of substance” is.
(Im sorry, if these are stupid questions, I am just a “newbie” in CFD)
And I an also curious what the PowerUsers say
Legit questions and not stupid at all, that’s the purpose of the forum - to ask and be curious!
Indeed the Boussinesq approximation is possible in CHT, you can find all the details here: Convective Heat Transfer | Analysis Types | SimScale - please let us know if that helped and if you have any other questions.
And what do you mean by numerical Schemes?
And oh, I think I have made myself misleading. I mean Conjugated Heat Transfer with CHT. Not the Convective Heat Transfer. The material properties are confusing here, (e.g. Amount of Substance), And I dont find the option of the boussinesq approximation here. May you help me?
Under Numerics you can find some settings that can help to increase accuracy and stability but a prerequisite for this is a good quality mesh If the mesh is bad you will either not get the simulation started or deal with numerical oscillations in any case - this is some sort of “last resort” if everything else does not want to go your way.
For CHT (I should have known that you mean Conj. HT here) no Boussinesq option is available but I will ask my colleagues if there’s anything planned.
Is there anything more I can do in Numerics what might improve the residuals (except the relaxation factors and the different tollerances)?
Can it be helpful to vary the type of solver (Smooth Solver e.g) ?
By the way, I found one certain reason which caused my bad residuals. Its the boundary condition “Pressure inlet outlet velocity”. When I change the boundary type to Wall with constant temperature, the residuals look a lot better! Can you explain this?
This should be the right reading material for you: http://www.wolfdynamics.com/wiki/OFtipsandtricks.pdf
Good question. We have had some simulations in the past (as far as I can remember) where the same issue occurred but I would have to dig a bit deeper to find a mathematical and physically meaningful interpretation of this phenomenon. Some oscillations come from 2nd order discretization and additionally truncation does not help this to get even better. You could try to work with a finer grid for the inlet/outlet regions but that is just a quick idea here.