Time Step Size-Unsteady Flow

ck3 · May 14, 2009, 09:26

Hello,
I using unsteady flow option in fluent to solve a separated turbulent flow over a wavy wall.
The purpose of this test is to compare the predictions of FLUENT's Realizable k-e turbulence model, against the DNS results of Maaß and Schumann.
For more detail, you can consult this link:
http://cfd.me.umist.ac.uk/ercofold/d...77/test77.html
I chose the computational domain to cover only one period of the wavy channel. The length of the periodic domain is 1 m. An 64x96 quadrilateral mesh was generated and the velocity is 0,103 m/s.
I need to define the number of steps and the steps size.
Please anyone could tell me what should be the number of steps and the step size, and also the max. number of iteration. How should I calculate them?
Thank you very much.

erkan · May 15, 2009, 10:05

Hi,

As a first thought, I can recommend you to make a time-step refinement study. To have time-step independent solution you can run the case for different step-sizes and the highest step size, where the change in any reference property is small enough, can be your time-step size.

ak6g08 · May 15, 2009, 14:31

Hi,

If you were solving a DNS problem I suppose the way you would calculate the required time-step is by calculating the kolmogorov timescale, but since you are doing k-epsilon which is a RANS model, you arent going to resolve the kolmogorov scales, if you know (approximately) how much energy is going into the system, equate this to the turbulent dissipation (epsilon) and calculate the kolmogorov timescale (formula is on wikipedia), you should know your integral timescale (based on the largest domain or eddy size divided by some characteristic large scale velocity, use what you are using to calculate your reynolds number), your timestep should be somesort of average of the two...as a ball park figure (closer to the kolmogorov scale to be conservative). As far as how many iterations per timestep...this is hard to say, it depends on what residuals you use, something like 1e-4 for momentum and continuity should be ok...set the iterations really high for ONE timestep (set them to 500) and then look at how many timesteps it takes for the solution converges on the first timestep, will give you a good indication of how many you will need. Hope that helps

akour

ck3 · May 15, 2009, 17:06

Thank you (erkan and ak6g08) for your answer.
Dear akour, Give me your e-mail to send you the artical of Maass and Schumann for more details.

May 14, 2009, 09:26	Time Step Size-Unsteady Flow	#1
ck3 Member Khaled Join Date: Mar 2009 Posts: 53 Rep Power: 17	Hello, I using unsteady flow option in fluent to solve a separated turbulent flow over a wavy wall. The purpose of this test is to compare the predictions of FLUENT's Realizable k-e turbulence model, against the DNS results of Maaß and Schumann. For more detail, you can consult this link: http://cfd.me.umist.ac.uk/ercofold/d...77/test77.html I chose the computational domain to cover only one period of the wavy channel. The length of the periodic domain is 1 m. An 64x96 quadrilateral mesh was generated and the velocity is 0,103 m/s. I need to define the number of steps and the steps size. Please anyone could tell me what should be the number of steps and the step size, and also the max. number of iteration. How should I calculate them? Thank you very much.

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
DPM UDF particle position using the macro P_POS(p)[i]	dm2747	FLUENT	0	April 17, 2009 01:29
Phase locked average in run time	panara	OpenFOAM	2	February 20, 2008 14:37
General grid/element size and time step questions	stu	CFX	1	May 24, 2007 05:37
time step for inviscid supersonic wedge flow.	yaseer	Main CFD Forum	1	March 8, 2007 09:40
unsteady calcs in FLUENT	Sanjay Padhiar	Main CFD Forum	1	March 31, 1999 12:32

May 15, 2009, 10:05		#2
erkan New Member Join Date: Mar 2009 Location: Turkey Posts: 15 Rep Power: 17	Hi, As a first thought, I can recommend you to make a time-step refinement study. To have time-step independent solution you can run the case for different step-sizes and the highest step size, where the change in any reference property is small enough, can be your time-step size.

May 15, 2009, 14:31		#3
ak6g08 Member Akour Join Date: May 2009 Posts: 79 Rep Power: 16	Hi, If you were solving a DNS problem I suppose the way you would calculate the required time-step is by calculating the kolmogorov timescale, but since you are doing k-epsilon which is a RANS model, you arent going to resolve the kolmogorov scales, if you know (approximately) how much energy is going into the system, equate this to the turbulent dissipation (epsilon) and calculate the kolmogorov timescale (formula is on wikipedia), you should know your integral timescale (based on the largest domain or eddy size divided by some characteristic large scale velocity, use what you are using to calculate your reynolds number), your timestep should be somesort of average of the two...as a ball park figure (closer to the kolmogorov scale to be conservative). As far as how many iterations per timestep...this is hard to say, it depends on what residuals you use, something like 1e-4 for momentum and continuity should be ok...set the iterations really high for ONE timestep (set them to 500) and then look at how many timesteps it takes for the solution converges on the first timestep, will give you a good indication of how many you will need. Hope that helps akour

May 15, 2009, 17:06		#4
ck3 Member Khaled Join Date: Mar 2009 Posts: 53 Rep Power: 17	Thank you (erkan and ak6g08) for your answer. Dear akour, Give me your e-mail to send you the artical of Maass and Schumann for more details.