[maba]

Hi every one.
I am simulating turbulent oscillating flow over flat plate. I started with laminar oscillating flow(2nd stokes problem) that has an analytical solution, and to check if the setup works. The domain is a rectangle, with the translational periodic boundary condition in stream-wise direction, wall at the bottom, and symmetry at the top. The domain is 40 stokes length in stream wise and 200 stokes length normal to flow. There is a harmonic pressure gradient in flow direction to make flow oscillate with prescribed frequency. The number of time steps per period is 720. The residuals are less than 10^-5 during simulation, but The result differ from analytical solution. What is the problem? What did I forget?

[ghorrocks]

Looks like an FAQ to me: https://www.cfd-online.com/Wiki/Ansy..._inaccurate.3F

The big one most people forget to do is: Have you done sensitivity studies on your time step size, mesh size, domain size and convergence criteria?

If you want us to be more specific, you will need to show details of what results you are getting and what results you expect, and how you modelled it (post an image of your domain and mesh, and the output file).

[maba]

Originally Posted by ghorrocks

Looks like an FAQ to me: https://www.cfd-online.com/Wiki/Ansy..._inaccurate.3F

The big one most people forget to do is: Have you done sensitivity studies on your time step size, mesh size, domain size and convergence criteria?

If you want us to be more specific, you will need to show details of what results you are getting and what results you expect, and how you modelled it (post an image of your domain and mesh, and the output file).

Thank you for your answer Glenn.
Yes, I have done sensitivity analysis for time step and mesh size. I reduced the residual target(rms) to 10^-6, but the result didn't change. I continued the solution for 70 period of oscillation, it didn't help.
When I tried to get the analytical results with turbulence model, it became even worse. The wall shear stress became zero!(The model was k-epsilon)
The mesh and velocity profile is attached.

[ghorrocks]

Can you upload your output file?

[Opaque]

It would be great to compare the analytical solution setup, i.e. exact boundary conditions, and the Ansys CFX setup.

Differences around the moving wall could be explained by mesh sensitivity, but the far-field velocity is way off. That is a sign of different boundary conditions at the top boundary.

As Glenn suggested, we need more information.

[maba]

Originally Posted by Opaque

It would be great to compare the analytical solution setup, i.e. exact boundary conditions, and the Ansys CFX setup.

Differences around the moving wall could be explained by mesh sensitivity, but the far-field velocity is way off. That is a sign of different boundary conditions at the top boundary.

As Glenn suggested, we need more information.

Thank you for your answer. In analytical solution, we have free stream flow top of the boundary layer. So I moved the top boundary condition up to BC not to affect the boundary layer. The boundary layer thickness is about 4 mm, and the domain height is 200 mm. What else can be done?

That's the output file link. It would be great if you could look at it.

https://drive.google.com/file/d/1ida...ew?usp=sharing

Originally Posted by ghorrocks

Can you upload your output file?

[ghorrocks]

Your linked file is the results file. I don't have time to go through that. Just upload the text output file please.

Also, I thought Stokes second problem was oscillating flow between two plates, meaning that you can assume symmetry in the middle. You appear to have it transitioning to free stream. Is this what you intended?

[Opaque]

Originally Posted by maba

Thank you for your answer. In analytical solution, we have free stream flow top of the boundary layer. So I moved the top boundary condition up to BC not to affect the boundary layer. The boundary layer thickness is about 4 mm, and the domain height is 200 mm. What else can be done?

Effectively you are not solving the same problem, but an approximation of it; therefore, the current differences are inconclusive of the quality of the solution.

As I always suggest in the forum, write the mathematical model using equations (not plain English) used for the analytical solution, post it and let us see/compare it to an equivalent Ansys CFX model. By waving approximations we will not be able to help you narrow down the source of the differences.