Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.
Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.
Mixture Model, Laminar Flow (mm) interface - anyone using it?
Posted Oct 31, 2012, 11:51 a.m. EDT Computational Fluid Dynamics (CFD) Version 4.3a 13 Replies
Please login with a confirmed email address before reporting spam
Hi,
I wonder if anyone here in the forum is using the Mixture Model, Laminar Flow (mm) interface. I have several issues when using this interface with a fully coupled solver (namely, singular matrices are poppoing out) and I would like to know if anyone else experienced similar problems.
Best,
J
I wonder if anyone here in the forum is using the Mixture Model, Laminar Flow (mm) interface. I have several issues when using this interface with a fully coupled solver (namely, singular matrices are poppoing out) and I would like to know if anyone else experienced similar problems.
Best,
J
13 Replies Last Post Nov 6, 2012, 11:19 a.m. EST
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hello J,
I have some experience with this model.
I also had some convergence issues, wich I resolved by introducing Inconsistent Stabilization for the Dispersed Phase. Be aware if your mesh is fine enough.
Also, I used a segregated solver. Do you wanto to use a Fully coupled Solver?
Best Regards,
Rui Silva
I have some experience with this model.
I also had some convergence issues, wich I resolved by introducing Inconsistent Stabilization for the Dispersed Phase. Be aware if your mesh is fine enough.
Also, I used a segregated solver. Do you wanto to use a Fully coupled Solver?
Best Regards,
Rui Silva
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi Rui,
thanks for replying. This interface is giving me hard time right now. After discussing with COMSOL Support I do not have anymore the issue with the singular matrix (using a time-dependent solver solved the issue). On the other hand you are right regarding the mesh. I'm currently running with P2-P1 elements but my model is pretty big (around 4 millions DoF) and is taking a huge amount of RAM on our cluster. With P1-P1 the oscillations in the gradient of gammadot where extremely big.
The Support suggested to use a segregated solver in any case, so I left behind the idea of running with a coupled solver.
I would like to avoid inconsistent stabilization as far as I can... but it might be a solution if I see that the model doesn't behave as supposed to...
Which slip model do you use? Do you see nicely the migration of particles?
Thanks again for replying,
J
thanks for replying. This interface is giving me hard time right now. After discussing with COMSOL Support I do not have anymore the issue with the singular matrix (using a time-dependent solver solved the issue). On the other hand you are right regarding the mesh. I'm currently running with P2-P1 elements but my model is pretty big (around 4 millions DoF) and is taking a huge amount of RAM on our cluster. With P1-P1 the oscillations in the gradient of gammadot where extremely big.
The Support suggested to use a segregated solver in any case, so I left behind the idea of running with a coupled solver.
I would like to avoid inconsistent stabilization as far as I can... but it might be a solution if I see that the model doesn't behave as supposed to...
Which slip model do you use? Do you see nicely the migration of particles?
Thanks again for replying,
J
Hello J,
I have some experience with this model.
I also had some convergence issues, wich I resolved by introducing Inconsistent Stabilization for the Dispersed Phase. Be aware if your mesh is fine enough.
Also, I used a segregated solver. Do you wanto to use a Fully coupled Solver?
Best Regards,
Rui Silva
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
The Slip Model depends on the size and density of your particles.
As a rule of thumb I estimate the particle Reynolds number with the help of the terminal velocity.
It has served well so far.
With this method I see the particle concentration gradient (I am using stationary study) due to the density ratio.
Best Regards,
Rui Silva
As a rule of thumb I estimate the particle Reynolds number with the help of the terminal velocity.
It has served well so far.
With this method I see the particle concentration gradient (I am using stationary study) due to the density ratio.
Best Regards,
Rui Silva
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi Rui,
yes, I know that the Slip Model depends on the size and density of your particles, that's why I was wondering which kind of model you use - to see how a particular was performing.
Thanks,
J
yes, I know that the Slip Model depends on the size and density of your particles, that's why I was wondering which kind of model you use - to see how a particular was performing.
Thanks,
J
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
I'm sorry, I misunderstood :)
I used the Schiller-Naumann Slip Model.
I have used both the HR and SN models, but for the laminar study I used the Schiller Naumann. Also used it with turbulent studies and it worked very well in both situations.
Best Regards,
Rui
I used the Schiller-Naumann Slip Model.
I have used both the HR and SN models, but for the laminar study I used the Schiller Naumann. Also used it with turbulent studies and it worked very well in both situations.
Best Regards,
Rui
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
One more question: how many DoF has your model? And how much RAM does it take? Do you use a segregated solver with iterative solver or segregated with direct solver?
J
J
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
I have to say that the Iterative Solver looks promising, RAM wise, but have yet to successfully use it without getting some kind of error. I believe the doc should be clearer on this particular subject.
I cannot answer how many dofs right now because I do not have the files with me (workstation), but if I recall correctly, I was using 40G of RAM for 1 meter of pipe section, because I have a very fine mesh. And I used a Direct Solver, usually MUMPS or PARDISO.
Overall, I think that Direct Solvers are more robust, but the downside is the RAM requirements.
Like I wrote, the problem with Iterative Solvers is the "fine-tunning", which, I think should be better depicted in the documentation.
Hope this helps.
Best Regards,
Rui
I cannot answer how many dofs right now because I do not have the files with me (workstation), but if I recall correctly, I was using 40G of RAM for 1 meter of pipe section, because I have a very fine mesh. And I used a Direct Solver, usually MUMPS or PARDISO.
Overall, I think that Direct Solvers are more robust, but the downside is the RAM requirements.
Like I wrote, the problem with Iterative Solvers is the "fine-tunning", which, I think should be better depicted in the documentation.
Hope this helps.
Best Regards,
Rui
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi Rui,
thanks for the info. My model it actually takes a lot more (segregated solver and direct solver), it takes around 160 GB or real RAM (the virtual is much more) for a problem with "Number of degrees of freedom solved for: 3880077 (plus 96454 internal DOFs)".
I'm curious about many DoF you have. Would you be so kind to check whenever you have the chance? I'm trying to understand why my model is taking so much RAM.
Best,
J
thanks for the info. My model it actually takes a lot more (segregated solver and direct solver), it takes around 160 GB or real RAM (the virtual is much more) for a problem with "Number of degrees of freedom solved for: 3880077 (plus 96454 internal DOFs)".
I'm curious about many DoF you have. Would you be so kind to check whenever you have the chance? I'm trying to understand why my model is taking so much RAM.
Best,
J
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
As soon as I get to my office I will post the file.
If you would like me to take a look or reconstruct the mesh in my station, it's no problem.
I have found that using some meshes (swept and tringular) increases signifcantly the DOFs.
Best Regards,
Rui
If you would like me to take a look or reconstruct the mesh in my station, it's no problem.
I have found that using some meshes (swept and tringular) increases signifcantly the DOFs.
Best Regards,
Rui
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
That's perfect, thanks. I'm currently using a free tetrahedral mesh, without the boundary layer mesh (it gave me some convergence issue)...
I'll wait to hear from you... thanks again,
J
I'll wait to hear from you... thanks again,
J
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hello J,
as promised, here is the link to download my case study.
Due to size limitations i have uploaded it to Sendspace.
www.sendspace.com/file/8tmg6v
Best Regards,
Rui
as promised, here is the link to download my case study.
Due to size limitations i have uploaded it to Sendspace.
www.sendspace.com/file/8tmg6v
Best Regards,
Rui
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi Rui,
many thanks for uploading your model!
It make sense for your model to take around 40GB, I'm around that number of DoF and amount of RAM occupied.
And I see now why you can model such a long tube (I was actually quite surprised), you use P1-P1 elements and do not discretize the scalar shear rate with an additional PDE. I'm using a model in which the particle fluxes are function of the gradient of the shear rate - see the tutorial dense_suspension.mph. I'm forced to use P2-P1 elements and discretized the shear rate with an additional PDE... this is adding a lot more DoF.
But the idea of using a swept mesh is probably a good one... I'm gonna try it and see what comes out. Mainly what I need is to control the nasty oscillations in the derivatives of the velocity - check ux, vy, wz and you'll notice them, mine at the moment are much worse.
Thanks for your help, I'll might come back to you if something more comes into my mind... and feel free to contact me if anything comes in yours.
Thanks again,
Best,
J
many thanks for uploading your model!
It make sense for your model to take around 40GB, I'm around that number of DoF and amount of RAM occupied.
And I see now why you can model such a long tube (I was actually quite surprised), you use P1-P1 elements and do not discretize the scalar shear rate with an additional PDE. I'm using a model in which the particle fluxes are function of the gradient of the shear rate - see the tutorial dense_suspension.mph. I'm forced to use P2-P1 elements and discretized the shear rate with an additional PDE... this is adding a lot more DoF.
But the idea of using a swept mesh is probably a good one... I'm gonna try it and see what comes out. Mainly what I need is to control the nasty oscillations in the derivatives of the velocity - check ux, vy, wz and you'll notice them, mine at the moment are much worse.
Thanks for your help, I'll might come back to you if something more comes into my mind... and feel free to contact me if anything comes in yours.
Thanks again,
Best,
J
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
I'm glad I could help.
From your explanation, yes, my model is simpler.
I used this simulation to serve as a comparison with previous turbulent simulations but with some turbulence atenuation by the particles.
The Swept mesh will probably be more coherent (no inversed mesh elements), but I believe will be more RAM demanding. But I agree that you should try it.
Feel free to ask further questions.
Best Regards,
Rui
From your explanation, yes, my model is simpler.
I used this simulation to serve as a comparison with previous turbulent simulations but with some turbulence atenuation by the particles.
The Swept mesh will probably be more coherent (no inversed mesh elements), but I believe will be more RAM demanding. But I agree that you should try it.
Feel free to ask further questions.
Best Regards,
Rui