Learning Forum

[marco.pedata]

Hello ANSYS community,

I try to perform a mesh convergence study on a problem similar to that of a flow around a cylinder. The problem is transient and I use the SST-k-omega model.

I use 12 inflation layers around my cylinder wall and an unstructered tiangular mesh for the rest. I started with a time step of dt=1e-4 and determined the element of the triangular mesh via CFL=1 even though the simulation is implicit. For the inflation layers I calculated the element size via y+ from boundary layer theory. I am interested in the vibration frequency due to lift forces. Therefore, I performed a FFT for each simulation and looked at the frequency. I tested y+ = 1 -> f = 140 Hz, y+ = 2 -> f = 130 Hz, y+ = 0.5 -> f = 140 Hz. I settled for y+ = 1. Than I tested different time steps: dt = 1e-4 -> f = 140 Hz, 1e-3 -> f = 99 Hz, 1e-5 -> f = 120 Hz. The frequency did not converge for smaller time steps. I simulated a timespan of 0.1 s. Maybe I need to run the simulation for longer time period? What are your thoughts on this. Do I approach the mesh study correctly or should I try a different approach? I attached a figure showing the Cl coefficient over time. Please note that in this figure the simulation for the smallest time step is not yet finished as the system corrupted and I needed to restart the simulation. However the result for 0.1 s was 120 Hz.

As this is my first CFD simulation I welcome any tips and advice on how to proceed :)

[Rob]

It's possible you've not run for long enough. Is it the same flow with just the time step decreasing? If you also plot contours of velocity or similar how's that looking? 

[marco.pedata]

Thank you! 

I will try to run the simulation for longer time periods. Qualitively speaking the velocity contour plots look similar for dt = 1e-4 and dt = 1e-5. 

[Rob]

You may be picking up a fast transient somewhere too. 

[marco.pedata]

Okay I will keep that in mind. As this is a transient problem I need to run it long enough to get good results. How can I determine what long enough is? I suppose that this simulation time depends on the initialization. I used Standard Initialization and speceified a constant flow field computed from the inlet. Would hybrid initialization reduce the simulation time needed?
Can I utilize the residuals plot in some way to determine at what time the results may be of good quality? 

[Rob]

Residuals are good for checking convergence on each time step; monitors are for checking whether the solution has reached the equilibrium transient condition.