[Brock17]

Hi all,

I'm a 3rd year Engineering student undergoing a CFD-based project based around investigating flow over a cricket ball, and eventually hoping to establish a simulation that can successfully replicate the phenomena of 'swing' (theoretically, by angling the seam relative to the direction of the flow, the protruding seam 'trips' the boundary layer into turbulence on the seam side and therefore results in a pressure differential and a side-force causing lateral movement, known as 'swing').

I've attempted to run the simulation using a steady, RANS solver (SST k-omega) under the impression that the flow i'm expecting shouldn't be time dependent and therefore a steady solver can be acceptable(?). My initial sim produced fairly decent results in-line with what I should be expecting; a side force was generated in the expected direction of a magnitude somewhere in the right ball park of what my research suggested. I was a little concerned by the appearance of the residuals however (image attached) as they appear to oscillate beyond approx 300 iterations. Despite this, the force coefficient appears to have resolved (image attached also) and doesn't oscillate which I assumed to suggest the flow had successfully resolved.

Despite the concerns, I was satisfied with the simulation as it gave me a reasonable output, however when I refined the mesh further and ran again I encountered bigger problems. My initial grid was 300,000 cells (polyhedral) and my next attempt was 430,000 cells. This attempt gave an even more concerning residual plot (attached) and this time, the force coefficient didn't resolve and appeared to change periodically by the latter iterations? This suggested to me it was due to a transient feature and an unsteady solver might be required?

All in all, i'm unsure how to proceed, because the initial steady solver appeared to resolve but further attempts don't(?). From my research, every other investigation similar to mine (involving simulating flow over a cricket ball) used LES, but I would be keen to use RANS if it's plausible. Any advice would be greatly appreciated.

Thanks.

EDIT: Using Star CCM+ by the way.

InitialResiduals.jpg

IntialForceCoeff.jpg

430kResiduals.jpg

430kForceCoeff.jpg

[flotus1]

Originally Posted by Brock17

This suggested to me it was due to a transient feature and an unsteady solver might be required?

This.
There is an inherent instability in the flow that requires a time-dependent approach.
The fact that the simulation converged better on a coarse mesh (more numerical diffusion) further adds to this conclusion.

[Brock17]

Originally Posted by flotus1

This.
There is an inherent instability in the flow that requires a time-dependent approach.
The fact that the simulation converged better on a coarse mesh (more numerical diffusion) further adds to this conclusion.

Thanks for the quick response.

Do you have any advice regarding specific models? Do you think this simulation can be achieved using uRANS modelling?

Also, on a side note, do you have any advice regarding a grid sensitivity exercise for an unsteady flow? I would usually conduct a sensitivity study on a steady flow grid by taking velocity profiles, refining the mesh and re-simming until the velocity is no longer sensitive to refinement but this will prove difficult I envisage for an unsteady sim, which will take significantly longer to conduct?

Thanks again.

[arjun]

I see two plots for force coefficient one is positive and other one is negative.

Which one is correct one?

[flotus1]

Originally Posted by Brock17

Thanks for the quick response.

Do you have any advice regarding specific models? Do you think this simulation can be achieved using uRANS modelling?

Also, on a side note, do you have any advice regarding a grid sensitivity exercise for an unsteady flow? I would usually conduct a sensitivity study on a steady flow grid by taking velocity profiles, refining the mesh and re-simming until the velocity is no longer sensitive to refinement but this will prove difficult I envisage for an unsteady sim, which will take significantly longer to conduct?

Thanks again.

My somewhat personal opinion on this subject is that LES is the only right way to perform such an analysis.
You might be able to get some results with an URANS approach, but I would not put my money on this method for doing actual research. There is a reason why most of the research on similar subjects uses LES: it is more accurate than RANS/URANS.
Unfortunately, there is no way around the increased computational cost for unsteady simulations.