Hello,
I have a strange problem. I have done some simulations of temperature and velocity field for air and water. For air, there is no problem at all. But Elmer can't give me any solution when I increase a density of material for an object in which I run flow simulations. I have tried to create a more dense mesh, but with no luck. What am I doing wrong? I tried to define density = 20, and that gives some results. But when I increased it to 200, it did not give any results.
regards
dk
density of material problem NS
Re: density of material problem NS
hi,
I don't think you are doing anything wrong - and neither is the ElmerSolver. This is a known feature of flows,
see any reference on computational fluid dynamics. If you increase your density keeping other factors
constant,you in effect increase the Reynolds number of the flow. At some point your steady(?) simulation
probably won't give an answer as the flow is not steady anymore, and so on, until finally the flow hits the
fully turbulent regime for high enough Reynolds numbers. In principle such a flow is solvable using the
laminar model with dense enough mesh and small enough timesteps, in practice almost never though,
given your computer resources. Unless you are studying turbulence with a supercomputer. So in practice
you'd probably need some form of modeling turbulence (...or transition to...). Some choices would be the
Reynolds avaraged models: Spalart-Allmaras, K-Epsilon, K-Omega, SST K-Omega, to give a few
(it's a zoo
. These might be good for at least fully turbulent flows, might not be so good in the transition
regime. Or some LES- (Large Eddy Simulation) or VMS- (Variational Multiscale Simulation) model.
That said, in the parameter region where the flow is not quite turbulent, but becoming unsteady, you could
try playing convergence tricks:
I don't think you are doing anything wrong - and neither is the ElmerSolver. This is a known feature of flows,
see any reference on computational fluid dynamics. If you increase your density keeping other factors
constant,you in effect increase the Reynolds number of the flow. At some point your steady(?) simulation
probably won't give an answer as the flow is not steady anymore, and so on, until finally the flow hits the
fully turbulent regime for high enough Reynolds numbers. In principle such a flow is solvable using the
laminar model with dense enough mesh and small enough timesteps, in practice almost never though,
given your computer resources. Unless you are studying turbulence with a supercomputer. So in practice
you'd probably need some form of modeling turbulence (...or transition to...). Some choices would be the
Reynolds avaraged models: Spalart-Allmaras, K-Epsilon, K-Omega, SST K-Omega, to give a few
(it's a zoo
. These might be good for at least fully turbulent flows, might not be so good in the transitionregime. Or some LES- (Large Eddy Simulation) or VMS- (Variational Multiscale Simulation) model.
That said, in the parameter region where the flow is not quite turbulent, but becoming unsteady, you could
try playing convergence tricks:
- o make the simulation time dependent (it could easily be that, even with all loads constant)
o increase mesh density (might make things worse convergencewise, if turbulence really
is setting up)
o increase time resolution ("")
o play with relaxation factors
o play with gradually increased loads and/or material parameter values
o etc...
Re: density of material problem NS
Thanks for your answer.
Now, I understand. For almost a year, I have thought that the only way to do a flow simulation in elmer is in a transient state. Then, I've finally got to one problem in which I could run a simulation in a steady and in a transient state and got the same results! It took me a while to see how by manipulating a Nonlinear System Relaxation Factor I could get correct results. So, I was really happy, since some of previous transient flow simulations took up to almost three days. But when I changed fluid from air to water it again wouldn’t work. Now, I understand why I couldn’t get any correct results for bigger models, like for example building's rooms - it has a big Re number.
So to sum up, I will now always try to simulate some flow problems in steady state first, and when it wouldn’t work, I will simulate same problem in transient state? Am I correct?
I’ve attached files with that problem to my post. You can see that when for body 1 material = 1 (air) everything works. But when you change material to 6 (water), the simulation won't work in correct way.
regards
daniel
Now, I understand. For almost a year, I have thought that the only way to do a flow simulation in elmer is in a transient state. Then, I've finally got to one problem in which I could run a simulation in a steady and in a transient state and got the same results! It took me a while to see how by manipulating a Nonlinear System Relaxation Factor I could get correct results. So, I was really happy, since some of previous transient flow simulations took up to almost three days. But when I changed fluid from air to water it again wouldn’t work. Now, I understand why I couldn’t get any correct results for bigger models, like for example building's rooms - it has a big Re number.
So to sum up, I will now always try to simulate some flow problems in steady state first, and when it wouldn’t work, I will simulate same problem in transient state? Am I correct?
I’ve attached files with that problem to my post. You can see that when for body 1 material = 1 (air) everything works. But when you change material to 6 (water), the simulation won't work in correct way.
regards
daniel
- Attachments
-
- forum_example1.zip
- (73.69 KiB) Downloaded 288 times