Different solutions to same setup on different versions of vspAERO

[Lin Livrin]

Hey guys,

I've noticed recently that there are quite big differences between results with the same geometry (or .vsp3 file) across different vspAERO versions:

What I tested was V3.36, 3.40 and 3.49. Now, to make the test similar i used only thin bodies in 3.49 and only VLM in 3.36 & 3.40. Solver setup and geometry were completely the same across all versions.

I was looking at CMy(alpha) chart to compare the results. In 3.36 & 3.40 the results looked to be more or less the same -> the curve had a negative trend which was satisfying my requirements. Then, in 3.49 -> the curve was actually trending positively with increasing alpha. (see attached screenshots)

I am really interested if any information could be provided on if something in the solver changed fundamentally over time or which version could be wrong here? With such discrepancies it is hard to trust any result. Any information will be very much appreciated, thank you for your time!

[Rob McDonald]

A ton of stuff changed in the solver during that time.

The adjoint capability in the solver was totally re-written.  It used an algorithmic approach which was replaced with a totally hand-coded approach.

The new adjoint introduced a new implicit wake model and other formulation changes.

On the OpenVSP side, the geometry handling was changed to support mixed thick/thin geometry and ngon geometry.  Now, merging of tris into quads is handled by OpenVSP instead of VSPAERO.

The main thing I would investigate is to make sure that you are converging your wake in both before and after cases.

Rob