Parametric Framework for Automated Structural Layout Modelling and Optimisation of Composite Airframe

Although considerable progress has been made in the rapid automated generation of computational fluid meshes, literature on analogous tools for structural models remains scarce. This work presents an automated framework for the parametric modeling of composite airframe structures, enabling rapid design changes and evaluation of alternative structural layouts. Python modules, using HyperMesh TCL to generate GFEMs of wings and fuselage segments with control over geometry and layout parameters, and automatically assign materials, properties, loads and boundary conditions are developed. Generated wing models enable composite sizing optimization in OptiStruct, allowing systematic layout exploration under strain, stress, and buckling constraints for selected design load cases. The approach is demonstrated on the Simple Transonic Wing (STW) wingbox benchmark by varying three layout variables: number of ribs, number of stringers per skin panel and spar-cap width. An optimized Latin Hypercube Sampling set of 18 layouts is evaluated, where each configuration is sized via composite optimization. A staged multi-run strategy is also applied to mitigate local-minimum sensitivity. A Gaussian Process surrogate is trained to map the optimized wingbox mass to the layout parameters, achieving a leave-one-out Root Mean Square Error of around 0.79\% relative to the mean observed mass. The surrogate model is then embedded in a Bayesian optimization loop using the Monte-Carlo batch Log Noisy Expected Improvement (qLogNEI) acquisition function in BoTorch. Starting from the initial 18 designs, only four additional sizing evaluations are required to identify a configuration, 2.2\% lighter than the best design of the initial sample set. The result is compared against the direct global search over the full set of sizing and layout variables, performed by MIDACO . MIDACO recovers a design with the same three layout parameters and a mass 1.46\% lower than the surrogate-based result, at the expense of requiring nearly ten times more FE analyses. Overall, the study proved the proposed surrogate-assisted approach capable of identifying near-global wingbox layouts at substantially reduced computational cost, while providing a flexible parametric modeling tool extensible to full-airframe optimization.