[frossi]

I am conducting a study of the effects caused by changing the winglet cant angle. In particular, I am observing how drag, lift, and L/D of the whole wing change as the winglet cant angle is changed.

Wing
My wing has a dihedral of 6 deg, and thus I consider the winglet to be "flat" (0 cant angle) when it has an angle of 6 deg from the horizontal.

Winglet
The winglet cant angle is changed from 0 cant angle (same deflection as wing dihedral) to 60 deg, at increments of 3 deg.

Conditions
My wing (as seen in the picture) has an angle of attack of 7.5 deg, a velocity of Mach 0.392. The ambient conditions are those at 3,000 m altitude.

As you can see from my graphs:
1) The values before 42 deg winglet angle seem to oscillate

2) There is a huge jump in both lift and drag from 42 deg winglet deflection to 45 deg winglet deflection. How can a 3 deg change give this much of a difference?

3) The pattern after 45 deg winglet seem more stable and reasonable. Why is this?

It looks like when the winglet is at 45 deg (39 deg from the horizonatal) something changes, either in the physics or the math, so that my results change drastically. I thought it may be an error in the way in which the software calculates lift, or maybe the force components, due to the fact that the winglet is inclined compared to the wing, but I cant figure it out.
I also thought it may be the way in which I define the force components, but they look right.

I define lift as:
x-component: -sin(7.5)
y-component: cos(7.5)

I define drag as:
x-component: cos(7.5)
y-component: sin(7.5)

What could be the cause of my results? should I define lift and drag differently for the wing? Could it be that Fluent is not resolving the components on the winglet because it is angled? Is it a geometry problem or a physics problem?

Thank you all for your help.

[ghorrocks]

This is the CFX forum. Try the fluent forum for help with fluent.

But I hope you have done sensitivity analysis of mesh, time step and convergence criteria. If you have not shown that your mesh, time step and convergence criteria is adequate then everything you have done so far is meaningless and will need to be redone on the correct mesh, time step and/or convergence criteria.

[frossi]

Hi ghorrocks,

I posted here since I thought this question could be general enough. Yes, I have a fine mesh and my results do not change significantly if I further refine it. Also, my simulation is steady so there is no time step. And yes, my convergence is good too.

Since I expect induced drag to decrease as the winglet angle is increased, I was thinking that maybe the software is not able to solve the physics of induced drag?
I am using a spalart allmaras turbulent model for a Mach 0.4. Should i switch to realizable k-eps?
I really cant figure out why I cant show a good pattern of decreasing drag (due to decrease in lift induced drag). Any idea of what could be wrong?

Best regards

[ghorrocks]

Fluent and CFX are both Navier Stokes solvers. If the flow you describe is a Navier Stokes flow then they should be able to handle it. There is nothing you have mentioned so far which makes me think a NS solver should not be able to handle it.

Don't forget to check convergence tolerance. You need to be adequately converged.

There are LOTS of possible reasons why your simulation is not accurate. Here are a few I can think of:
* Inappropriate turbulence model
* Laminar to turbulent transition, meaning the front of the wing is laminar and the rear is turbulent. This has major effects on separations and drag.
* Geometry faceting and fidelity
* Separations leading to vortex shedding. These are often not accurately captured by RANS and need LES/DES/SAS approaches.
* Discretisation accuracy (1st order, second order etc)
* Boundary proximity
* Interaction with fuselage

So I recommend you have a close look at the results in the stable and unstable regions and try to work out the difference. There should be some flow feature which changes. Then decide whether it is going to be real or not.