Minimizing Computational Data Requirements for Multi-Element Airfoils Using Neural Networks

Artie cial neural networks were used to e ll in a design space of computational data to optimize the e ap position for maximum lift for a multi-element airfoil. Multiple-input, single-output networks were trained using NASA Ames Research Center’ s variation of the Levenberg ‐Marquardt algorithm. The computational data set was generated using an incompressible Navier ‐Stokes algorithm with the Spalart ‐Allmaras turbulence model. An empirically-based criteria, designated the “ pressure difference rule,” was applied to the training set because numerical inaccuracies for the computational method were identie ed near maximum lift. The neural networks were trained with only three values of the inputs: e ap dee ection, gap, and overlap at various angles of attack. The entire computational data set was thus sparse, and yet by using only 52 ‐70% of the computed data, the trained neural networks predicted the aerodynamic coefe cients within 1.7% of the maximum lift coefe cient. In addition, a genetic algorithm and a gradient-based optimizer were integrated with the neural networks to optimize the high-lift rigging. This new optimization process had a higher e delity and a reduction in CPU time when compared with an optimization procedure that excluded the genetic algorithm.