Surrogate Model Based Aerodynamic Shape Optimization of a Hydrogen Powered Aircraft

This paper describes both the low and high fidelity methods used in implementation of machine learning based optimization techniques for aerodynamic shape optimization of hydrogen powered commercial aircraft. Using vortex lattice method based aerodynamic performance estimations in conjunction with empirical solvers used to estimate an arbitrarily generated aircraft’s overall structural weight and drag, a Bayesian optimization algorithm is applied to optimize the wing to maximize the aircraft’s overall range. This method was applied to a test case of the Boeing 737-800 where it was seen to predict the plan form shape of the real world wing with a high degree of accuracy. This methodology is then applied to the design of transonic supercritical airfoils in which the airfoil is parameterized and optimized to reduce drag at a specified lift coefficient. The resultant airfoils are seen to closely resemble modern supercritical airfoils and reduce the drag significantly. Lastly a method of airfoil surrogate modelling using convolutional neural networks to predict aerodynamic performance of airfoils at a fraction of the computational cost is examined. This method is seen to provide highly accurate estimates for the drag coefficient of airfoils, however is seen to fall short of being able to fully optimize the airfoils. Ultimately the previously described methodologies are all used in conjunction to create a computer program which is able to fully optimize the geometry and airfoil distribution of a wing with high degree of accuracy at the maximum possible computational efficiency.