Development of a deep neural network framework for sailplane cross-country performance optimisation

Abstract

Traditional sailplane cross-country performance optimisation is a slow and computationally expensive task as it requires the integration of multiple simulation packages and function evaluations for each span-wise station and flight condition. Traditional methods also prohibit expert input to this optimisation process. Deep neural networks, with their powerful generalisation capability, provide a possible solution for replacing the computationally expensive simulation packages with lower fidelity surrogate models, ultimately reducing the time spent on the preliminary sailplane design and optimisation phase. In this research, a deep learning framework for sailplane cross-country performance optimisation is developed to address these shortcomings of the traditional model. Various modelling techniques are employed to build a deep learning based system that is accurate, computationally efficient, and which enables expert input upfront to the optimisation process. For the expert input, deep learning modules are developed, enabling airfoil generation from parsimonious shape and structural variables. Constraining this parsimonious variable search space, therefore, enables intuitive constraint settings on the wing and individual airfoil designs. It is shown that the deep learning framework not only outperforms the traditional model in terms of computational efficiency but is also able to find wing/airfoil combinations that outperform defined baseline designs whilst adhering to all constraints imposed on the final design. Specifically, the efficacy and validity of the proposed deep learning framework is shown where it is applied to realise a 0.9% average increase in the cross-country performance of the JS4 sailplane - a State-Of-The- Art (SOTA) standard class sailplane and currently one of the worlds best performing models. This deep learning framework therefore facilitates rapid, accurate, and targeted wing/airfoil design explorations and ultimately reduces the time spent on design, especially in the preliminary design phases, and has been shown to possess the capability of cross-country performance optimisation of SOTA standard class sailplane designs.