High-Lift Optimization Design Using Neural Networks on a Multi-Element Airfoil

Roxana M. GreenmanAerospace EngineerNASA Ames Research CenterMoffett Field, California 94035-1000, U. S. A.Tel: 650-604-3997, Fax: 650-604-2238E-mail: rgreenman@mail.arc.nasa.govKarlin R. RothAerospace EngineerNASA Ames Research CenterMoffett Field, California 94035-1000, U. S. A.Tel: 650-604-6678, Fax: 650-604-2238E-mail: kroth@mail.arc.nasa.govABSTRACTThe high-lift performance of a multi-element airfoil wasoptimized by using neural-net predictions that were trainedusing a computational data set. The numerical data was gener-ated using a two-dimensional, incompressible, Navier-Stokesalgorithm with the Spaiart-Allmaras turbulence model. Becauseit is difficult to predict maximum lift for high-lift systems, anempirically-based maximum lift criteria was used in this studyto detemaine both the maximum lift and the angle at which itoccurs. Multiple input, single output networks were trainedusing the NASA Ames variation of the Levenherg-Marquardtalgorithm for each of the aerodynamic coefficients (lift, drag,and moment). The artificial neural networks were integratedwith a gradient-based optimizer. Using independent numericalsimulations and experimental data for this high-lift configura-tion, it was shown that this design process successfully opti-mized flap deflection, gap, overlap, and angle of attack tomaximize lift. Once the neural networks were trained and inte-grated with the optimizer, minimal additional computerresources were required to perform optimization runs with dif-ferent initial conditions and parameters. Applying the neuralnetworks within the high-lift rigging optimization processreduced Me amount of computational time and resources by83% compared with traditional gradient-based optimization pro-cedures for multiple optimization runs.NOMENCLATURECa drag coefficient, C a - D/(q_c)Ct lift coefficient, C l - L/(q=c)C m moment coefficient, C,. --M/(q=c 2)C e.,is