• A rapid multi-fidelity framework for preliminary wing sizing and aeroelastic analysis • Integration of structural, aeroelastic, and load-driven constraints within an MDO workflow • Quantified sensitivity of wing mass and flutter to discretisation, constraints, and load cases • Composite wings achieve ∼ 30% mass reduction while maintaining aeroelastic margins compared to metallic wings. • Interpretable surrogate models enable uncertainty-aware, data-driven wing design To accelerate the overall design of next-generation aircraft towards Net Zero aviation, a flexible toolset of structural and aeroelastic analysis integrating multi-fidelity models within a Multidisciplinary Design Optimisation (MDO) framework has been developed. First, a discrete beam-based low-fidelity method is used to capture the dominant global behaviour of the wing, such as global stiffness and mass distribution. Second, two medium-fidelity tools, the Structural Layout Tool (SLoT) and the Structural Analysis Tool (SAIT), are used to generate parametric structural layouts and to evaluate stresses, buckling, and stiffness characteristics of semi-monocoque structures. SLoT and SAIT bridge low- and medium-fidelity methods to enable rapid and accurate sizing and optimisation, achieving computation times on the order of minutes rather than the days typically required by high-fidelity workflows at the preliminary design stage. Parametric wing box models of the NASA Common Research Model, for both composite and metallic configurations, were sized under Certification Standards CS-25 manoeuvre/gust loads, aeroelastic divergence, prescribed flutter margins, and structural safety requirements. Sensitivity analysis quantified how the design inputs (material choice, load case selection and design constraints) govern the outputs (structural mass, stiffness and aeroelastic performance). Aspect ratio and taper ratio were varied to assess transferability across wing planforms while targeting minimum structural weight subject to flutter constraints. The workflow also generated a curated dataset to support data-driven design: random-forest models with SHapley Additive exPlanations (SHAP) for feature attribution, Gaussian process regression surrogates for uncertainty-aware prediction, and an inverse-design loop for rapid design-space exploration. Findings identify the highest-leverage design inputs for reliable preliminary sizing and demonstrate a practical route to data-driven MDO with controlled accuracy and reduced computational cost.
Pan et al. (Sun,) studied this question.