[Andrew Norfolk]

Hi,

I've created a very dense mesh for a heatsink and surrounding fluid.


For unknown reasons I've been struggling to get the model not crash (fatal overflow in the linear solver). This has led me to believe that the mesh is refined enough to be picking up vortex events and the steady state assumption is causing the model to fail. The ANSYS support team suggest upping the timescales (probably to smooth over these flow features) and to use a turbulence model to add stability to the solution.



My calculations show the flow should be laminar, but the ANSYS service request team have suggested running the model with the SST turbulence model as this provides necessary damping. They say that as the velocities are very low it shouldn't effect the heat transfer between the heatsink and fluid much.

My understanding is that the SST model enforces a wall function based around the turbulent velocity profile and the law of the wall. Doesn't this mean that there will be much higher velocity gradients in the near wall region as a result of this wall function being applied and therefore higher rates of heat transfer or am I missing something?

I'm really just looking for an explanation or a direction to some relevant reading. I'm not sure why this is an acceptable solution.

Thanks.

[ghorrocks]

For unknown reasons I've been struggling to get the model not crash (fatal overflow in the linear solver).

Have a look at the FAQ: http://www.cfd-online.com/Wiki/Ansys...do_about_it.3F

And also this FAQ: http://www.cfd-online.com/Wiki/Ansys...gence_criteria

This has led me to believe that the mesh is refined enough to be picking up vortex events and the steady state assumption is causing the model to fail. The ANSYS support team suggest upping the timescales (probably to smooth over these flow features)

Yes, this is discussed in the FAQ.

and to use a turbulence model to add stability to the solution.

I am not sure I agree with this. You use a turbulence model when the model is turbulent. If the flow is laminar, you do not use one. Using a turbulence model to get a laminar flow to converge is ignoring the real problem.

suggested running the model with the SST turbulence model as this provides necessary damping. They say that as the velocities are very low it shouldn't effect the heat transfer between the heatsink and fluid much.

True, when you use SST on a laminar flow the extra dissipation added by the turbulence model is small. But then if the effect is small how is it going to help? It does not add up to me.

My understanding is that the SST model enforces a wall function based around the turbulent velocity profile and the law of the wall. Doesn't this mean that there will be much higher velocity gradients in the near wall region as a result of this wall function being applied and therefore higher rates of heat transfer or am I missing something?

I recommend you use automatic wall treatment which will use wall functions if y+>11 and integrates to the wall for lower y+. Your understanding is incorrect, if you use wall functions it does not cause higher wall gradients, it means it applies the amount of wall shear on the wall as if it where a fully developed boundary layer with the first node at y+>11, which means it will apply too much wall shear. This will cause excessive flow dampening.

I recommend following the FAQ advice on "my simulation converges for a while..." instead of putting a turbulence model in a laminar flow.