Finite volume advection of sharp interfaces #22794
Answered
by
lindsayad
ABallisat
asked this question in
Q&A Modules: Navier-Stokes
-
|
Hello, I am trying to model the advection of a sharp interface using the Weakly Compressible Finite Volume Navier Stokes kernels to model a two phase immiscible flow. I am having issues with it not converging with sharp interfaces. To simplify the problem, I have a created a minimum worked example for the advection of a circle in 2D using a single fluid with a constant user specified velocity. To test if the issue is a having a sharp step, I am using a smoothed step function, the idea being the smoothness parameter can be dropped to get closer to an ideal step function. The input file is below. The best results I can get are using crank-nicolson with a Van Leer flux limiter and the default petsc options (although ILU performs almost identically) but there is still quite a lot of excessive diffusion, as in the images below. Pushing the smoothing parameter below 0.0005 (so driving it closer to a step) stops it converging. I can't see anything I am obviously doing wrong, but I suspect I am missing something. My only other thoughts are around trying to implement a more complex flux limiter scheme such as CICSAM. Any suggestions greatly appreciated. Thanks! Input file: Time = 0 Time = 1 (100 time steps) |
Beta Was this translation helpful? Give feedback.
Answered by
lindsayad
Nov 28, 2022
Replies: 4 comments 10 replies
-
|
Hello @lindsayad @grmnptr @tanoret you may want to pitch in First I would remove the temperature variable for now since it does not play a role. This will make the problem faster to run and wont affect results at this point. You may just comment out the kernel and turn the T variable into an auxiliary variable. This is similar to several text book / code studies of the diffusivity / quality of discretization schemes that you may have already found out there VanLeer makes sense as a good candidate Do you have the (non)convergence history for a non-converging case? Guillaume |
Beta Was this translation helpful? Give feedback.
1 reply
-
|
The case has really high viscosity (100) btw. This probably does not help transporting a scalar with no diffusion. Could you please reduce it a bit? |
Beta Was this translation helpful? Give feedback.
-
|
I’ll come back to this tomorrow but the limiters we have (vanLeer and Minmod) fall back to almost upwind where they encounter large gradients (not very smooth solutions). That explains the excess diffusivity with vanLeer. I don’t think we have a less diffusive scheme at the moment (besides the regular weighted average which can result in oscillatory behavior), but I’ll take a look at your input tomorrow. Also, I recommend following @GiudGiud ‘s instructions to see what the issue is with the convergence. |
Beta Was this translation helpful? Give feedback.
0 replies
-
|
Hello both, Thanks for the suggestions. Neither moving out the temperature to an auxvariable nor reducing the viscosity (I dropped it to 1E-5, it was unrealistically high) improved convergence. I have turned debug on and get the following print out: so it does look like phi is the one struggling to converge. Thanks for the links, I will take a look at them. I know that FOAM, specifically interFOAM, has some nice examples of this, it is on my to do list to try to back out of the code what they implement as I have not yet found a good summary of it. |
Beta Was this translation helpful? Give feedback.
2 replies
-
|
one thing to check is that the case converges with 1e-5 viscosity without the scalar. That can be tough on the solver if there is no initialization. ok now that we know that it s linear that is not converging, we should try a stronger preconditioner. The strongest rn is LU with a NONZERO pc_factor_shift. this is not scalable but it s good to get started |
Beta Was this translation helpful? Give feedback.
-
|
It runs through without issue for a viscosity of 1e-5 if I turn off the advection kernel. Using advection, I have just tried with that preconditioner and get a very similar result unfortunately: I am going to have a good dig into the methods suggested and see if any look promising. Looking at the implementation of flux limiters in MOOSE, I am guessing that if I can cast them into a similar form then they should not be too difficult to implement. |
Beta Was this translation helpful? Give feedback.
-
|
You don't need to paste all the diverging timesteps. Probably the first one or two is plenty. Do you need to resolve the velocity and scalar fields at the same time? Can you first compute the velocity field, and then model the transport of phi? |
Beta Was this translation helpful? Give feedback.
7 replies
-
|
Thanks for the suggestion, I will take a look. One question about FVAdvection: only average, upwind and skewness-corrected advection interpolation are supported (from the source code in next https://github.com/idaholab/moose/blob/next/framework/src/fvkernels/FVAdvection.C), whereas the INSFVScalarAdvection supports the various flux limiters. Can I just edit the FVAdvection to support the other limiters? |
Beta Was this translation helpful? Give feedback.
-
|
You could. I just had a look and I dont see why not |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
where do we stand on this? have the results improved or is it still pretty diffusive? |
Beta Was this translation helpful? Give feedback.
-
|
We are now at the stage where I have a working compressive scheme implemented in Python for proof of concept of the algorithm, this week I am hoping to get it working in MOOSE. At the moment I am just working out where in MOOSE all the components I need reside, I will report back with an update when I have something working (or if I hit more issues). |
Beta Was this translation helpful? Give feedback.
Answer selected by
GiudGiud
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
You don't need to paste all the diverging timesteps. Probably the first one or two is plenty. Do you need to resolve the velocity and scalar fields at the same time? Can you first compute the velocity field, and then model the transport of phi?