
[Sponsors] 
December 6, 2005, 15:32 
Time scales and Fluent's unsteady solver issue

#1 
Guest
Posts: n/a

Hi all.
After making dozens of steady state simulations with my 2D problem (bluff body, side view of a car), I am starting to think that my case is unsteady; when I switch to 2nd order schemes, solution starts to oscillate (I mean, residuals, Cd, Cl... oscillate) I've tried to run unsteady simulations with the Automatic Time Stepping (and the NonIterative Time Advancement option enhabled) and after each step, Fluent ALLWAYS reduces time step: it gave me a step of 1e40 and continued reducing it!! (I started from a step of 0.001s) Now I want to try with fixed stepping, but I'm not sure about my problem's time scale. How can I calculate it? In bibliography I allways see Kolmogorov time scale, or viscous diffusion time scale, but in their formula it appears the characteristical Lenght, and this lenght in my opinion can be arbitrary ANY! (from tunnel's dyameter to a screw of the car xD), so time scale is not really defined. Could anyone enlighten me? Or suggest me a time step, or how to fix my ATSNITA problem? Thank you very much! 

December 6, 2005, 20:34 
Re: Time scales and Fluent's unsteady solver issue

#2 
Guest
Posts: n/a

Domp use NITA for this problem and see if you can get some good results, use ITA with very small time step, say 5.0E05 or so. Once you get the results you know it is coming because of the NITA not because of the model set up. If it does not give you good results with ITA also then see whats wrong with the model set up.


December 7, 2005, 08:40 
Re: Time scales and Fluent's unsteady solver issue

#3 
Guest
Posts: n/a

I've tried an ATS without NITA and 20 iterations/step, starting from a step of 1e5, but fluent continued reducing time step to a value as low as 1e22 (I tried it increasing the number of iterations to 35, but it did the same)
Then I've tried an unsteady simulation with fixed time step of 1e4, after some thousands of iterations it could be seen that residuals behaved periodically, as the Cd, Cl and masswaighted average of vel. magnitude in the outlet, but in the SAME manner as it did the steady state simulation, so in this problem I don't understand the difference between steady and unsteady! Two contour plots between iterations of the steady are very similar to those made while running an unsteady simulation... In both cases I start with a ke standard, 1st order schemes. Then I switch to Realizable ke 1st order schemes and noneq wall functions; finally I activate 2nd order schemes and solution starts to oscillate. My grid is yet grid independent, because a 150000tri elements grid gets very similar solutions compared to my actual 50000 tri elem. grid. I've done a quad grid and without switching to 2nd order schemes, solution oscillates too. I quite desperate. Could anybody help me? Thanks a lot! 

December 7, 2005, 08:54 
Re: Time scales and Fluent's unsteady solver issue

#4 
Guest
Posts: n/a

try using standard wall functions. sometimes nonequilibrium wall functions leads to difficulties in convergence.


December 7, 2005, 15:52 
Re: Time scales and Fluent's unsteady solver issue

#5 
Guest
Posts: n/a

Hi Erdem!
I'd tried what you suggested but I get the same results; perhaps I'm not able to see that results in my unsteady simulations are just correct! What should I see in an unsteady simulation of a 2D bluff body, specially in residuals plot? Now, I get residuals that have very short period of oscillation (I suppose this is because of the subiterations in each time step) but with big amplitude, say between 1e3 and 1e7 for the continuity, and they keep in this range as simulation runs. Cd, Cl and Massweighted average of vel. magnitude also oscillates and I think this is obvious because of the unsteady behaviour of the model. But I don't know if this is convergence, because this same behaviour in the solutions when running steady simulation is not considered as convergence. While in unsteady, residuals may go down or it is just correct that they keep in a range and oscilates a bit? Many thanks to all! 

December 12, 2005, 02:08 
Re: Time scales and Fluent's unsteady solver issue

#6 
Guest
Posts: n/a

I think you should solve it in steady mode once and let it iterate as many times as it needs. Then, if convergence happenned you should run it in unsteady mode with as many iterations in first time step as it did in steady mode to reach convergence. In brief what i want to say is that you should increase the number of iterations in each time step. As a question, what are the causes of unsteadyness in your problem? Why dont you choose the steady solution?


December 13, 2005, 15:30 
Re: Time scales and Fluent's unsteady solver issue

#7 
Guest
Posts: n/a

OK, thanks all: finally I get results in unsteady simulations. The problem was time step was quite hight and Fluent started to get diverged solutions.
Ali: I'm simulating a flow past a 2D car shape. My mesh is made with tri elements, so I must switch to 2nd order schemes to get better results. In 1st order schemes simulation converges, but in 2nd order solutions (residuals, Cd, Cl...) starts to oscillate periodically and I convergence is not reached. Can I make you 1 question Ali?: can I consider a 1st order scheme solution got with a mesh with quadrilateral elements as a good solution? The fact is that in Fluent's User manual it is said that with try elements I "must" switch to 2nd order schemes to get the best results because mesh is not aligned to flow, but with Quad elements it says that a 1st order solution "may" be correct. When I switch to 2nd order schemes with a Quad mesh of my problem, I got same results as 2nd order schemes with the Tri mesh, so I'm not sure a 1st order Quad mesh (that I can make it to converge) could be right... What do you think? Thanks! 

Thread Tools  
Display Modes  

