AeroSandbox: A Differentiable Framework for Aircraft Design Optimization

Abstract

This work presents a new computational framework for conceptual aircraft design called AeroSandbox. This framework leverages modern techniques for automatic differentiation developed in the optimal control and machine learning communities. By combining these efficient gradient calculations with robust optimizers such as IPOPT, multidisciplinary aircraft design problems of practical interest can be solved in seconds. We demonstrate this speed with several canonical aircraft design problems in this work, showing that performance and flexibility equals or exceeds that of state-of-the-art tools in many cases.

This framework's modular approach to engineering analysis allows sophisticated aerospace problems to be constructed by connecting plug-and-play building blocks in code. This decreases the time required to go from a qualitative vehicle and mission concept to a quantitative, optimized performance estimate. The framework's emphasis on rapid development time and run time enables an engineer to interactively pose design questions, enabling human insight to be more readily applied to the computational design process.