Advancing CAD-Based Wing Geometry Optimization Using High-Accuracy Sensitivity Analysis

Abstract

A key enabler in designing highly efficient aircraft is multi-disciplinary optimization (MDO), which considers multiple disciplines simultaneously. The disciplines share information through interdisciplinary gradients. In aircraft wing design, structure and aerodynamics are the most tightly coupled disciplines. These two disciplines share information through the wing geometry. This makes the parameters related to geometry interdisciplinary. A well-established method of deforming wing geometries is the free-form-deformation. However, accurately translating changes of design variables to the control points can be challenging. This can create inconsistencies across multiple disciplines, resulting in poorly conditioned optimizations. In contrast, this work employs a CAD-based approach to alter geometries. Parameters of the CAD model can be directly used as design variables for the MDO, mitigating the need to translate design variables onto control points. Additionally, by using a centralized CAD model, it ensures that multiple disciplines share and utilize the same information effectively. Previous studies that utilized the CAD-based method typically relied on finite differences to compute sensitivities. While this technique offers simplicity and ease of implementation, it has been observed to negatively impact the convergence rate of the optimization process. This circumstance is addressed using a specialized CAD kernel providing high precision algorithmic differentiation techniques and a process to propagate the surface gradients through the structural optimization. The benefit of numerical exact geometrical gradients is demonstrated on a wing box use case, where a combined structural and geometrical optimization reduces the mass by 1.1% and converges within roughly half of the iterations needed compared to finite difference gradients.