[vivekcfd]

I am solving a transient FSI problem where we have been facing an accuracy issue due to hard-coded "Total Force" transfer to the structure side. Our FSI problem requires only the transient component of the Total force across the FSI interface. Therefore, I would like to subtract the steady- or mean-part of the fluid-forces from the Total Force on each node at the FSI interface and consequently supply only the time-dependent part of the fluid force to the structure side.

For this operation, we require:

1) steady or mean value of total force at each node which we may get from a steady-state solution

2) Defining a new force (by subtracting Step 1 Force from the Total Force) term in transient simulation and transferring this term across the FSI surface.

Could anyone give some suggections on carrying out such a step in CFX?

thanks in advance
Vivek

[stumpy]

Could you explain where the accuracy issue comes in? What you describe is possible, I'm just not sure if it's really needed.

[vivekcfd]

we are only interested in transient or fluctuating part of the force. As in our problem, the transient force is the measured variable. In the FSI problem we have, the transient or fluctuating part of the force is 1e-6 to 1e-9 smaller than the mean force or mean signal. Moreover we want to resolve the variations in transient part of the order 0.1% that means 1e-9 to 1e-12 times smaller than the mean force or mean signal.

Resolving such small numbers are not possible in full FSI simulations. By filtering the mean values of fluid force from the total force, we can improve the accuracy (or signal to noise ratio) in our simulations. Hope you understand my point. I am very much convinced that this is needed in our case.

Vivek

[ghorrocks]

Can you explain why only the fluctuating component is important?

Writing your own FSI routine sounds like a really bad idea to me. You should try to use what is there as much as is possible. If we have some context we may be able to come up with a way to remove the DC component of the force.

[vivekcfd]

I do not know in which context you mean bad idea. Our problem demands such type of approach as we measure only AC (fluctuating) part of the force in our experiments and we want to simulate this situation. My dear friend, I have spent many months on CFX-FSI coupling before reaching to this conclusion. We have carried out various simulations supporting the need of this idea. As I said earlier, including or letting the DC part in the signal (or force) lead to too much noise in the numerical signal being transferred to the structure and consequently it does not allow us to resolve flow effects which we measure in the lab.

I hope, I have answered your question. Will be grateful, if someone may suggest an idea to filter out DC or mean force from the total force.
thanks

[stumpy]

Could you describe where the mean and fluctuating parts are physically coming from? For example, if the mean force is due to a high system pressure, then you could just set an appropriate reference pressure to remove this. My other concern is that if the fluctuating component is so much smaller than the mean component, then won't you have the same accuracy issues on just the fluid side (forget the FSI) in resolving this force? If your fluid simulations can resolve this component accurately, then that would be reflected in the total force. Is the real problem the fact that the structural side or interface loads are not converged sufficiently to resolve the changes in Total Force? So perhaps you just need tighter convergence targets? In any case, simply set "Send To ANSYS" to None, then in the "Additional Coupling Sent Data" section send your own force variable (an Additional Variable of your choice I would assume).

[vivekcfd]

it not just the magnitude of fluctuating part. The small numbers I mentioned earlier were the ratio between fluctuating to mean force. And this ratio always keeps on changing if e.g. one varies mean flow velocity. Here mean is not due the static pressure rather mean flow velocity (mean kinetic energy, shear stresses etc.). The ratio is simply the kinteic energy of the fluctuating flow to the kinetic energy of the mean flow. Remember, here fluctuations are not due to turbulence but due to the induced motion of the structure.

If we filter out mean force from the total force, we will get what we want and the accuracy issue will be left only to convergence criteria. By the way we are converging 5 orders of magnitude (forces and displacement) at the interface. I do not think that the are not sufficiently converged.

Stumpy, regarding your suggestion: in CFX12.0, I do not see any way to send AV across interface. There are only some pre-defined variables (force, shear force etc) which are available in the GUI and one is not allow (atleast in the gui) to type a new variable as there is only drop-down menu available (try yourself to see what i mean). That is why force transfer I say is hard-coded.

[stumpy]

Hmm, I see. Did you try creating a vector AV? I believe only vector quantities can be sent.