grid adaptation for better convergence.
03/30/2004
I have a volume made of conical frustum placed on top of a cylinder. I am trying to solve for natural convection currents in this volume using Boussinesq approximation for density variation. From Ra number calculations, I know that the conditions are turbulent in this volume (Ra>10^10) I have a good hex mesh with max skew ness of 0.51 with all turbulent boundary layers resolved well. Still I am having trouble in getting convergence. The last residual values reported are: Continuity = 6.0564e03 Xvelocity = 2.5158e04 Yvelocity = 1.9871e04 Zvelocity = 2.4165e03 energy = 6.9842e07 k = 3.3916e04 epsilon = 4.3959e04 Thus I guess it is not too bad of a convergence, but I want to use grid adaptation in fluent to improve my results. Can someone suggest as to what type of adaptation I should be focusing on? 
Re: grid adaptation for better convergence.
Check wall y+. It should be between 30 and 300 if you are using standard wall treatment. You may need to use EWT (y+<5) to better capture natural convection.
I assume you have a closed domain, so you can not use mass balance to doublecheck convergence. The high residuals are very likely to be caused by the transient behaviour of the solution. Natural convection is usually an unsteady phenomena that needs to be simulated with a transient simulation. You will see what I mean if you solve the problem with the unsteady solver and animate a contour plot of temperature/velocities (create a video). Solution will change with time. 
Re: grid adaptation for better convergence.
well, i do not have closed domain. I have a vent (pressure outlet BC) at the top.
thanks a lot for all the explanation. But my question was what type of grid adaptation I should use ? should I keep refining my mesh till y+ gets in the correct range of 30 to 300? I donot want my grid size to go too high to keep solution time down. thanks, CO2 
Re: grid adaptation for better convergence.
To adapt the grid you use velocity or pressure bc.
Alamgir 
Re: grid adaptation for better convergence.
last year i did one natural convection problem, in the start i had some convergence problems, but after refinign the mesh that went away, and second we found it conversing with coupled solver (better than segregated), so you can try coupled solver, might help.

Re: grid adaptation for better convergence.
coupled solver: isnt it true that coupled solver is used only for highly compressible flows? I tried coupled solver in my case, the solution was obtained but vel vectors were looking weird  it was as if gravity was acting in X direction, although i had specified it in Z drn.
I solved the steady case yesterday with all hex elements in my geometry  I had to use low underrelax params  around all of them 0.5  IS THAT OK? kEPSILON model was on, since Re numbers in headspace of the tank geometry are around 15000 (due to low viscosity of air)  after gettting solution to the the problem i found out that Y+ max value was 12.81 and Y+ min was 0  Thus I guess now I need to coursen the grid  Can some one suggest me how and where I should coarsen it ? Any other suggestions will be great ! 
Re: grid adaptation for better convergence.
You can either coarsen the mesh (and use standard wall treatment) or refine the mesh (and use enhanced wall treatment). Second option is the most sensible!!!
I do not think you have to mess around with underrelaxation too much. I am pretty sure that the unsteady phenomena do not let you converge the solution in steadystate. Have you tried a transient simulation. It is very likely that a transient simulation will solve your problem!! If there are transient phenomena you will never fully converge your steadystate simulation. 
Re: grid adaptation for better convergence.
thank you very much to everyone for your good answers! your input is certainly helping me to make improvements to my model.
I am about to finalize my grid  I am getting good convergence and realistic values from my steady state model. In the transient case, I will incorporate changing ambient temperature, changing conv. heat transfer coeff at tank walls using profile files, I will simulate the effect of revolving sun and changing sun radiation heat flux through a udf .. I am thinking of allowing large number of iterations per time step so that my solution coverges sufficiently at each time step .. WHAT WILL BE YOUR RECOMMENDATION ON THE NUMBER OF ITERATIONS THAT I SHOULD BE ALLOWING PER TIME STEP? thanks, CO2 
Re: grid adaptation for better convergence.
You can not use a huge time step and wait for hundreds of iteration to converge it.
What you have to do is to choose a timestep size so that you converge the solution in no more than 40 steps. 
Re: grid adaptation for better convergence.
well, I am taking a time step of 30 minutes since my radiation data, ambient temp data, wind velocity data is for every hour. Even if I break down times steps further, I guess BC's are going to remain constant  THEN WHAT IS THE POINT IN FURTHER REDUCING TIME STEP SIZE  I WAS THINKING I CAN EVEN HAVE A TIME STEP OF 1 HOUR  WHAT DO YOU THINK?
Well, I am giving a max of 200 iterations per time step. I am seeing that fluent does not require the whole 200 iterations, but I guess setting 200 iterations keeps me on the safter side  CORRECT ? WHY DO YOU SAY THAT ONLY 40 ITERATIONS SHOULD BE ALLOWED ? WHAT IS SO WRONG IF THE RESIDUALS GO LOWER AND LOWER EVERY TIME STEP ? 
All times are GMT 4. The time now is 12:06. 