[AS_Aero]

Hi All

I want to do an LES computation for incompressible internal duct flow and hence I am looking for some best practice tutorial which could explain me and make sure I am not going wrong anywhere. Few main thiings I need to know is

1. How to initialize the case , I have inlet total pressure and outlet static pressure.

2. I have RANS results .res file for a particular coarse mesh, can I use the same result to intialize my LES which has more finer mesh ? Or should I run an RANS with the fine mesh first to intialize the LES ?

3. How to decide the deltaT value and CFL condition in CFX

Kindly help me with these information
Thanks in advance.

Regards
Anand

[ghorrocks]

This is not a simple topic. Initial conditions and boundary conditions for LES require careful thought as there are fundamental differences between LES and RANS. Have you done background reading into LES so you understand what issues you face in defining initial and boundary conditions?

dt and CFL - First of all, forget about CFL. CFX is an implicit solver and does not have a strong dependence on CFL. You can do a time step sensitivity check and work out what time step you require, but a much better way is to use adaptive time stepping, homing in on 3-5 coeff loops per iteration. Much simpler and it gets it right just about every time.

[Sergeyevichsky]

Originally Posted by ghorrocks

This is not a simple topic. Initial conditions and boundary conditions for LES require careful thought as there are fundamental differences between LES and RANS. Have you done background reading into LES so you understand what issues you face in defining initial and boundary conditions?

dt and CFL - First of all, forget about CFL. CFX is an implicit solver and does not have a strong dependence on CFL. You can do a time step sensitivity check and work out what time step you require, but a much better way is to use adaptive time stepping, homing in on 3-5 coeff loops per iteration. Much simpler and it gets it right just about every time.

Hi Glenn
5 coeff loops per iteration has been set in my simulation and residuals value was set to 10e-7 ,using LES in CFX, but after 3000 time-steps, results on flow fields of inlet was very bad, in contrast, flow filed where below 1R inner region at inlet was similar with some paper. Some paper shows that the residuals value should drop to 10e-9 and coeff loops was set to 20. I found that residuals value drops slowly after 15 coeff loops in my CFX simulation. So should I set 15 coeff loops per iteration in my simulation for preventing too time being cost? And, if less residuals value was set, more precise results was capture? Can you recommend some books or literature to me about some fundamental concepts as mention above?
Sincerely
Lintao

[ghorrocks]

The literature you are quoting does not sound like they are using CFX. CFX does not need convergence to 1e-7 and definitely not to 1e-9, and 20 coeff loops per iteration is not very efficient for CFX. But other software may work with this. So you cannot simply take settings from other software and apply it to CFX as they have different meanings.

Turbulence Modelling for CFD by Wilcox is a key textbook for turbulence modelling. Also all the introductory CFD textbooks like Anderson discuss it to some extent.

[Sergeyevichsky]

Originally Posted by ghorrocks

The literature you are quoting does not sound like they are using CFX. CFX does not need convergence to 1e-7 and definitely not to 1e-9, and 20 coeff loops per iteration is not very efficient for CFX. But other software may work with this. So you cannot simply take settings from other software and apply it to CFX as they have different meanings.

Turbulence Modelling for CFD by Wilcox is a key textbook for turbulence modelling. Also all the introductory CFD textbooks like Anderson discuss it to some extent.

Thanks for your recommendation, Glenn. Your suggestions is always timely and useful.
Today I look up some threads which has been submitted in this forums before. some people think that adaptive time-step should be choose in LES, but other people think that fixed time-step should be choose. So which time-step type should I choose?
And how should I add synthetic turbulence at inlet? I noticed that some people used harmonic function to generate synthetic turbulence, but I don't know how to realize that using CFX. Should I use CEL or other methods?
Sincerely
Lintao

[ghorrocks]

So which time-step type should I choose?

If you are experienced in finding suitable time step sizes then use either. It you are using second order time stepping (which you should if you are using LES) then you should probably use fixed time stepping. But if you are not familiar with how to determine a suitable time step then use the 3-5 coeff loop adaptive time step option.

And how should I add synthetic turbulence at inlet?

Good question

Whether you can introduce the harmonic function in CEL will depend on what the function is. Please post an example of the function you wish to apply.

[Sergeyevichsky]

Originally Posted by ghorrocks

If you are experienced in finding suitable time step sizes then use either. It you are using second order time stepping (which you should if you are using LES) then you should probably use fixed time stepping. But if you are not familiar with how to determine a suitable time step then use the 3-5 coeff loop adaptive time step option.



Good question

Whether you can introduce the harmonic function in CEL will depend on what the function is. Please post an example of the function you wish to apply.

In fact, I just have known harmonic function today. And then I looked up some literature about harmonic function. I'm not sure if the picture is display.

[ghorrocks]

That appears to be a function describing the flow over a wavy surface, not flow through an inlet plane with turbulence. Are you sure that function is relevant?

[Sergeyevichsky]

Originally Posted by ghorrocks

That appears to be a function describing the flow over a wavy surface, not flow through an inlet plane with turbulence. Are you sure that function is relevant?

I'm sorry. I looked up some literature again. Some function which was used to generate synthesis turbulence at inlet, are similar with the harmonic function (for example , eddy interaction models, random walk models). But it is so complexly. It made me dispirited. I have defined specific velocity profiles, come from experimental data, at inlet. So need I still add synthesis turbulence?

[ghorrocks]

This issue is one of the big unsolved problems in LES modelling - how to generate a realistic inlet turbulence profile. So there is no easy answer to this. For the latest thinking do a literature study on the topic.

Options include:
* Move the inlet boundary upstream and use a simple boundary on it. Design it so the turbulence develops to the correct level in the extra flow length
* Use a translational periodic boundary condition with a source term to drive the flow to "feed" turbulent structures which go out the exit back in the inlet.
* Do a separate model of the upstream conditions and "record" the flow as it travels out the outlet. Then "play back" this flow as an inlet condition. You will probably need to write a fortran routine to do this.

There are other ways of doing it, but hopefully that sparks some ideas.