End-to-end differentiable gust and maneuver loads at large scale via concurrent simulations

This paper introduces a new method for large-scale computation of aerodynamic loads for aircraft and airframe sizing. It is built on a geometrically nonlinear aeroelastic simulation framework with concurrent capabilities that can be deployed on modern hardware architectures. It enables the construction of load envelopes for static and dynamic cases in computational times of the order of seconds, as well as uncertainty quantification analysis via massively-parallel Monte Carlo simulations, while offering seamless integration with industrial-scale aircraft models. Furthermore, sensitivity analysis of the dynamic loads on concurrent computations is achieved via the use of state-of-the-art strategies for automatic differentiation and the extensive use of collective communication primitives. We demonstrate its use on a high-aspect-ratio-wing aircraft configuration, showing the ability of the solvers to capture nonlinear effects for external forces inducing large deformations and also the importance of these effects when studying modern aircraft concepts. Discrete, maneuver and gust loads are assessed for multiple flow and parametric conditions running concurrently. Sensitivities of maximum wing-root stresses and tip displacements are efficiently computed across all those conditions and verified against finite differences.