Development and Application of Improved Tandem Neural Networks for Inverse Design of Rotorcraft Airfoil

Abstract

Designing an airfoil shape with specific performance characteristics is a fundamental problem in the field of rotorcraft blade design. Traditional aerodynamic design methodologies often involve iterative optimization of the shape using low-fidelity modeling techniques during the early design phase, owing to the demanding computational costs of high-fidelity adjoint-CFD based optimization. In this study, an efficient accurate approach using Deep Neural Networks for the inverse design of rotorcraft airfoils is investigated. We leverage the Tandem Neural Network (T-NN) architectures to design the airfoil for a required performance curves (lift, lift-to-drag ratio, and pitching moment) in a novel and efficient way. The T-NN architecture allows for a modified and flexible cost function making them highly efficient and accurate for the inverse design of airfoils. Three different improvements to the T-NN architecture are proposed in this work to allow for improved accuracy and better constraint handling capabilities, enabling a paradigm shift in the rotorcraft component design methodologies.