|
|
diffU.png (178,230 bytes) |
|
|
U2.png (72,693 bytes) |
|
|
my3Dcase.jpg (32,286 bytes) |
|
|
I am running the mixerVessel2D laminar case in OpenFOAM-2.2.x and 2.3.x and find that the behaviour of 2.3.x is more stable than 2.2.x which is rather chaotic and in both cases the results are VERY sensitive to the modelling details, in particular the virtual mass. You mention that you have compared the results to those in a paper; are these results for this particular case or something similar? Can you provide the paper and the equivalent cases setup for OpenFOAM-2.2.x and OpenFOAM-2.3.x so that I can analyse the difference? |
|
|
Hi henry, Thanks a lot for your attention. About the virtual mass force, I didnot pay lots of attention to it. cuz usually I set it to be zero. but u give me a hint, I will try adjust the vm force to see the result. About the paper, these results are particular ones. I compared the epsilon field of my sumilation in 23x with this paper(simulated by 22x): http://www.degruyter.com/view/j/revce.2014.30.issue-1/revce-2013-0015/revce-2013-0015.xml form this graph(I upload it), we can see this epsilon is fairly good. I can get this good epsilon field from openfoam22x, but I can not get it from 23x. besides this, from my simulation. we can see the velocity field is not the same. I think this is the reason why epsilon fields are not the same. if we calculate the volume average epsilon field in my stirred tank case. 22x's result(around 0.05) is half of the result ot 23x(around 0.1). Actually I can use openfoam22x to do simulations. but there is also something weird with the dispersed phase. cuz in my simulation, all the walls are closed, the dispersed phase's volume fraction should be constant.anyway this is another issue. I will upload a 3d case with a coarse mesh, thank u very much. |
|
|
epsilon.PNG (213,367 bytes) |
|
|
https://www.dropbox.com/s/6llgnxn3588yzwj/Test.tar.gz?dl=0 I upload this case,just run it directly u know... Also u can check the log file about the alpha field. I run all of them in parallel mode. besides the U and epsilon, we can see alpha field is not the same. this can be tell from the log: Min(alpha1) = 0.00156777 Max(alpha1) = 0.00271831 in 23x, some cell's alpha is quite large. But I did not put a lot of time to figure it out. To be notice that: in my case, dispersed phase's density is 1022 which is very close to continual phase(998). I wish it should be very homogeneous, in 22x, its quite homogeneous, the max alpha1 is about 0.00203 or something. |
|
|
Could you provide a simplified 2D setup for this case to study the differences between 2.2.x and 2.3.x? Note that I have been running laminar to limit the number of differences we are studying simultaneously. The differences in the turbulence modelling between 2.2.x and 2.3.x are very significant with the latter providing a choice of up-to-date two-phase models. The rather over-simplistic modelling in 2.2.x is not provided in 2.3.x as it is not generally applicable or appropriate. I suggest that in the first instance we concentrate on running laminar and study the effect of the differences in the other models, in particular drag-blending and virtual-mass. |
|
|
|
|
|
|
|
|
half.png (137,121 bytes) |
|
|
I uploaded a laminar case, I think all the configuration is the same. about the drag force I want to set all of these to be noblended, but 23x blows up. So I set all of them to be blended to make a truly comparison. Also I found VM force have an effect on the velocity near to the wall. but it has the effect on both 22x and 23x cases. |
|
|
You MUST blend the interface models for cases with phase-inversion otherwise the system is ill-posed and unphysical. Note that blending in 2.2.x is built-in and linear over the complete range of phase-fraction. You can easily set the blending in 2.3.x to be equivalent. |
|
|
Yes, at first, I want to set it to be non-blended, but this blows up in 23x, so in my uploaded case, all of them are blended. except the governing momentum equation is conservative, I found this code in 23x: phase1.turbulence().divDevRhoReff(U1) does not include this term: (((2.0/3.0)*I)*nuEff1)*tr(gradU1T) even this term should be zero in convergence, but in openfoam22x it was included, in openfoam 23x, it is not included. So is this related to the problem? or there is some tricky way to handle this term in 23x? BTW, I tried to add this term into 23x, looks like it does not improve the result. Thanks very much! |
|
|
For single-phase flow this term is usually handled as part of the pressure but in two-phase flow with a single pressure it cannot be handled as part of the pressure. However, unless the flow is high-speed compressible this term is small and can generally be neglected. For these mixer cases it will be VERY small. |
|
|
oh, very quick response.lol,thanks a lot! Um. so far, for this "small velocity in the stationary zones" problem...Im have tried to figure it out for about ten days. I tried to compared with the codes. also I found dgdt in the alphaEqn is not the same in 23x and 22x: 23x: Sp[celli] -= dgdt_[celli]/max(1.0 - alpha1[celli], 1e-4); Su[celli] += dgdt_[celli]/max(1.0 - alpha1[celli], 1e-4); 22x: Sp[celli] -= dgdt[celli]*alpha1[celli]; Su[celli] += dgdt[celli]*alpha1[celli]; I can understand the code of 22x, cuz the whole equations are the same with your paper long time ago. but 23x this is a little tricky. I transported the whole compressibleTwoPhaseEulerFoam22x codes into openfoam23x. I can get the good result. Good means velocity in the stationary zones is larger. for now, from the laminar case, maybe this problem is caused by the whole algorithm of 23x? I dont have much CFD experience anyway. So I wish U guys can help me out. lol. Actually I can use 22x to run, but I just only run it with single core.. it way too slow.. |
|
|
dgdt is defined differently between 2.2.x and 2.3.x explaining the difference in the Su and Sp terms. Look at the way dgdt is updated in pEqn in the two codes. My understanding of the difference is that it is likely to be due to differences in the modelling and blending between the two codes. Also the fact that 2.3.x is conservative and 2.2.x isn't will affect the results. |
|
|
okay, looks like its very complicated to handle this problem. And if u guys have a solution pls make this update. for now I think I will use 22x to simulate. Thanks very much for your time to discuss this with me. I really appreciate it.! |
|
|
I traced the difference in behavior between OpenFOAM-2.2.x and 2.3.x to the drag-blending as expected but the error is on OpenFOAM-2.2.x not 2.3.x. I have pushed the correction to OpenFOAM-2.2.x: commit 1f35a0ff2a58105bbfce2259a74df47334a0fa53 Now if you choose linear blending in OpenFOAM-2.3.x over the phase-fraction range 0-1 and equivalent models the results between 2.2.x and 2.3.x are the same within reason. |
- Anonymous
