Closed guillaume-florent closed 3 months ago
This is expected behavior - changing the leading edge weight should indeed result in an asymmetric airfoil, since the leading-edge mode shape of a CST airfoil parameterization is asymmetric.
For visual of this, see this chart from the NeuralFoil README (blue mode shape of the bottom subplot)
Kulfan's paper on the LEM gives this mathematically in Eqn. 8: https://www.researchgate.net/publication/343615711_Modification_of_CST_Airfoil_Representation_Methodology
Thanks for the clarification. So, the LE radius only depends on the 8 upper and 8 lower weights, with the 'front' weights having a much bigger impact? Is that correct?
Bug Description / Observed Behavior
On a symmetrical Kulfan airfoil (upper weights = - lower weights), using a leading edge weight other than zero creates an assymetrical airfoil. On line 599 of kulfan_airfoil.py, the leading edge modification is added to y_lower. If it is substracted instead, the symmetrical airfoils remain symmetrical. Yet, loading a Kulfan airfoil from a symmetrical library airfoil (e.g. naca00xx) with the above mentioned modification creates an assymetrical airfoil. Which leads to the question: does neuralfoil have to be re-trained if kulfan_airfoil.py is modified to substract the leading edge modification on y_lower?
Steps to Reproduce
Create a symmetrical Kulfan airfoil, tinker with the leading_edge_weight and visualize.
Expected Behavior
Even though changing the leading edge weight should keep symmetrical airfoils symmetrical, it does not.
System Information
Other Information