Neural network and regression approximations in high-speed civil aircraft design optimization

Nonlinear mathematical-programming-based design optimization can be an eleg ant method. However, the calculations required to generate the merit functi on, constraints, and their gradients, which are frequently required, make t he process computationally intensive. The computational burden can be subst antially reduced by using approximating analyzers derived from an original analyzer utilizing neural networks and linear regression methods. The exper ience gained from using both of these approximation methods in the design o ptimization of a high-speed civil transport aircraft is the subject of this paper. The NASA Langley Research Center's Flight Optimization System was s elected for tbe aircraft analysis. This software was exercised to generate a set of training data with which a neural network and regression method we re trained, thereby producing the two approximating analyzers. The derived analyzers were coupled to the NASA Lewis Research Center's CometBoards test bed to provide the optimization capability, Both approximation methods wer e examined for use in aircraft design optimization, and both performed sati sfactorily. The CPU time for solution of the problem, which had been measur ed in hours, was reduced to minutes with the neural network approximation a nd to seconds with the regression method. Instability encountered in the ai rcraft analysis software at certain design points was also eliminated. Howe ver, there were costs and difficulties associated with training the approxi mating analyzers. The CPU time required to generate the I/O pairs and to tr ain the approximating analyzers was seven times that required for solution of the problem.