Bug in particle tracking
Hi experts,
I wrote UDF for brownian motion of particles , and I'm sure that brownian forces are correct. But particle trajectories didn't match with analytic result (ave(r*r)=6*D*t). does anyone have any idea to change scheme of particle tracking ? thanks in advance . |
Could you please provide a copy of your UDF to see what the problem could be?
|
thanks for your response,
as I said, there is not any problem in my UDF. I just want to change particle tracking scheme . it seems that at the beginning of particles paths , their positions are not realistic ! |
tracking
I am not sure wether I understand it well, but as far as I know the only way to influence particle trajectory is through the setable parameters and body forces, but not using another numerical scheme.
Once I implemented a Brownian model by selecting random displacements and wanted to add these dispalacement to the ones due to the deterministic forces, computed by FLUENT, but I failed. Later I was told by FLUENT developers that this would be possible only if I had the source code (which is obviously not the case). Why do you implement Brownian force, when it is included in FLUENT DPM? You have to switch it on and that's all. |
notice that if we want to use brownian force , the time step should be one order lower than particle relaxation time , but FLUENT time step automation is not good here at all.
also many researchers state that FLUENT brownian forces are also wrong . it's preferable to implement UDF for brownian motion. |
I perfectly agree with you! Actually, this was the reason (relaxation times) why I wanted to implement my own Brownian motion UDF. I had long discussions with developers, but they simply did not recognize that there is any kind of time step problem. I have a general equation, which is valid for any time step (Chandrsekhar), and turns into FLUENT's equation in case the time step is much lower than the relaxation time.
I also agree with you concerning the shortcomings of using forces instead of displacements, but as I told you it turned out that only forces can be implemented into FLUENT and not displacements. |
I implement brownian forces to FLUENT by UDF and I want to validate particles paths that FLUENT computed. forces are absolutely correct and I validate them.
I found that the problem were in FLUENT particle tracking at first time step and I'm going to handel. |
And what was the problem with FLUENT particle tracking during the first time step?
|
the particles have a relative big jump at first time step when small time steps use. at higher time steps this problem disappear but brownian forces were wrong !:rolleyes:
|
Do you use the original formula for the brownian force? (eq. 21.2-18 in FLUENT documentation)
|
yes. I don't have any doubt about brownian force . as I said before , I checked these forces with validated FORTRAN code .
|
OK. Have you tried to switch to another tracking scheme. I don't know which version are you using, but in the newer versions there is a possibility to select from different schemes (e.g. trapezoidal, runge kutta).
|
I use FLUENT 6.3 and all schemes like analytic,implicit,runge kutta and trapezoidal were tested . I think this is a bug in particle tracking and may be removed in newer versions.
|
Then you should contact FLUENT (ANSYS) helpdesk and report the bug.
In an earlier version (6.1.18 I think) there was a bug like if the particle diameter was less than 0.01 micron than it evaporated. I reported the bug and they fixed it. |
Maybe try to use an earlier version and see wether the problem persists or not.
|
Quote:
I have Drag, Pressure force and virtual mass in a cyclone particle tracking Fortran code. When i switch time step (Dt) from 0.1 of relaxation time (tav) to 1 tav my results show changes. I think it should be independent to Dt. what do you think? How much should be set for Dt? |
particle_tracking
In case of Brownian force it is important to compare the time step with the relaxation time, because the distribution from which we select the random force is different in the two cases (dt<<tau and dt>>tau). If there is no random force than I do not know about such restriction. However, the smallest the time step the more accurate the simulation is. If your result changes significantly than most probably the larger time step is not that correct. If your computer is performant enough, than I would recommend the usage of the smaller time step.
Regards, Arpad. Quote:
|
Quote:
In each transient case such as lagrangian models, you have to implement time step study. For brownian motion, we have a rule of thumb, i.e. 0.1 tau, but for any other forces you have to obtain its criterion; maybe it's lower or higher according to its nature. Regards, Amir |
particule tracking
I work on the taylor couette problem, i want to track a defined particules from the fluid to calculate their parametres like pressure or velocity, using fluent.
haw to do that (tracking particules)? ps; i use an udf for the wall of the moving cylinder, but not for the particules. |
tracking particules
I work on the taylor couette problem, i want to track a defined particules from the fluid to calculate their parametres like pressure trajectory or velocity, using fluent.
haw to do that (tracking particules)? ps; i use an udf for the wall of the moving cylinder, but not for the particules.[/QUOTE] |
Hi Amir/Flanker,
I have question regarding usage of Brownian force in Turbulent flow. It was mentioned in the Fluent 6.3.26 manual (section 22.2.1) that "Brownian force is intended only for nonturbulent models". is it?? I have already tested Lagrangian model (Fluent in built) for very fine particles such as 1nm and 5 nm. No. of particles deposited is almost same in both cases. Could please give me some suggestion how i can improve my results. Thanks John |
Quote:
About brownian motion in turbulent flow, I think manual statement makes sense because it can be ignored in comparison with turbulent dispersion mechanism but I didn't check that before. And about deposition issues; I saw number of papers in which this issue was reported and 2 UDFs were written for both brownian motion and deposition; it's stated that there is a problem in particle velocity interpolation near boundaries which affects deposition; but as a general recommendation, it's better to use eulerian model for such range of diameters to get rid of such problems. Bests, |
Hi Amir,
Thanks for your reply. Could you please name the papers who used UDF for Brownian motion in DPM model. As you suggested i am planning to go for Eulerian method. Thanks John |
Quote:
Effectiveness of Direct Lagrangian Tracking Models for Simulating Nanoparticle Deposition in the Upper Airways; P. Worth Longest and Jinxiang Xi, Aerosol Science and Technology, 41:380–397, 2007 This is the most famous paper of this group, you can also find other papers of these authors in this field. Bests, |
hi.
as you know, in DPM of Fluent, 0.3 *(K/epsilon) is used as eddy life time. is it possible to write a udf to use another formula for eddy life time instead of above equation? and if yes how? |
UDS(nanofluid) problems
Hi experts,
I would like to add the transport equation of source term(Temperature) into the UDS for fluent and how can I do ? My geometry is a 3D cubical and it's enclosure, the fluid is water+nanofluid(Al2O3). I want to observe the natural convection heat transfer phenomena of nanofluids likes Rayleigh and number Nusselt number. Do I need to supply more details ? please help me :) |
All times are GMT -4. The time now is 07:11. |