minimize time step, adapt region
Good morning everyone,
i did a geometry on Gambit for a Catheter into a vessel to simulate an injection of an oil in blood.
for the catheter of an exterior diameter of 0.5mm i made 10 meshes.
the problem that in Fluent i need 10^-5 as time step to obtain a stable flow.
so i have to wait to attend 10s more than a week
do you suggest anything for me?
can the option ADAPT REGION reduce the time step??
thank you in advance and it's serious i need a response
if ure mesh is very fine (small cells), u will need smaller time steps to not violate the shannon-nyquist theorem.
Rememebr that VOF also needs very small time steps.
1-Maybe ure mesh is very fine....Try coarser mesh. Maybe u need less cells than u think (verify grid indepenant solution)
2-The adapt function cant help u, because it can only refine the mesh (so u will need smaller time steps). It can only coarsen a mesh that u have alerady refined (it is a big limitation in Fluent)
If u want more help, tell more about ure flow and about ure turbulence model....
THANK YOU KARINE
things became more complicated;
first i am not using laminar flow.
i have a structured meshwith 1/0.025
diameter for cappilary tube 0.5mm and for the tube 7mm
first i took a time step 10^-5
i can never see what's happening in experiments cases.i can see a formation of a ¨¨finger¨ not a drop
when i took for example a time step 5*10^-5 i can see the phenomon that i expected.
this is bizarre cz i felt that increasing the time step would reduce the accuracy.
thank you for ur response in advance
why u dont run a DNS??? ure geometry is very small and ure flow is must be laminar.
Use the laminar model of Fluent (DNS).
So my adcices:
use VOF with DNS. Geo-reconstruct. Verify that solution does not depend on grid size.
It will take some time to do this, but at least, u are sure to have the best solution.
When u got ure bizarre solution, what turbulence model have u been using??
i am already in laminar model and using VOF geo-reconstruct.
the form bizarre obtained is a propagation in lenghth but no drop was formed, i can't observe the detachement.i have to validate an experimental study but i couldn t
do you think it's a mesh problem?i am really so confused i have to present my project as soon as i can :S
maybe the problem is because of VOF.
Vof is the most used interface tracking method but have some liitations: artificial coalescence. I will explain to you:
imagine that u have 2 droplets. The VOF will show u two droplets unless the distance between the two is less than the cell size: VOF allows ONE interface by cell. So when the 2 droplets approach, they will coalesce more rapidly than in experiments.
So when u are using a time step of 5*10-5, the droplets detaches and get more far than a cell size: u see droplets.
When u are using a time step of 10^-5, the droplets detach but dont get more far than a cell distance: VOF make them coalsece artificially.
So, u must use a smaller mesh. I told u before that u must verify that the solution does not depend on the grid size. The problem is that u will need smaller time steps :(
Tell me more about ure droplet size so i can be sure what i told u is true.
Verify also that ure problem is well posed.
Hope that u understood me with my bad english
i think ur right concerning the mesh size.it seems logical.
do you mean you want the size of the droplet formed in experiments?
is there anyway to show you some of the results?
thank you for everything
Yes, i mean is the experimental droplet size less then the mesh size????
for ure results, there is a way to show them on the forum (i dont know how) or send it to me via mail
DNS so ure mesh size must be at least two time less the smallest flow strucure (edy, droplet...) and ure time step smaller than the smallest lifetime of any of ure flow eddies, else u violate shannon-nyquist theorem
VOF: ure mesh must be at least two time less than ure interface lengthsclae for the same reason.
This is why VOF cant be used for largely mixed flows: we must use eulerian-eulerian empirical models, else the simulation will take years to converge....)
excuse me KArine DNS is to choose the laminar flow?i am using VOF with laminar flow but don't know what did you mean by DNS.
do you suggest for me to change the method and go to Eulerian?
i am really so confused
DNS is Direct Numerical simulation. It means that u try to solve directly the Navier-Stokes equations without any modeling (no turbulence model). So DNS will give u the solution of the NS equations, but the problem is that for turbulent flow u will need very very lot of cells, and very very small time steps, so it is not feasible in practice with our computers. LES (large eddy simulation) resolves the big scales in the flow, and models the small ones. RANS models will model the whole flow....
the Laminar model of FLUENT is a DNS.
In ure case the best way is VOF of course, with DNS. U have separated flow (so VOF) . Eulerian models will not be good, are very empirical, and cannot give a sharp interface.
Tell me more about ure flow. is it axismetric????
Ure computer is very slow?? Can u access the university cluster?? Can u tell me about ure scientific level (student, PHD....)?
If u have skype i can give u the adress of a friend (a teacher) who maybe can have better solutions)
Karine plz can you help me i have an urgent question.
i am now trying to impose 2 different inlets for the flow..(injection from a capillary tube into another tube containing a fluid in motion).
my problem is that i need to identify 2 different UDF functions to each one.
and when i interpret the 2 UDF, fluent take just one.
so what should i do.
plz reply i need ur help
|All times are GMT -4. The time now is 18:52.|