AI-Based Generative Algorithms Applied to the Design of Blended Wing Body Aircraft

This paper presents a generative modeling framework for the synthesis of Blended Wing Body (BWB) aircraft geometries that meet specified aerodynamic targets. The approach integrates a Denoising Diffusion Probabilistic Model (DDPM) with a tailored 1D U-Net architecture, conditioned on lift, drag, and moment coefficients. Geometries are parameterized through a compact 24-dimensional vector capturing both planform and sectional characteristics and evaluated using a mid-fidelity simulation pipeline coupling DUST and NeuralFoil via GEMSEO. A large dataset of valid configurations is generated under realistic aerodynamic constraints and used to train the model. Quantitative evaluations demonstrate that the model can generate geometrically diverse and aerodynamically accurate wing configurations, achieving medium-high fidelity with respect to target aerodynamic features, despite the absence of structural or manufacturability constraints. The results confirm the DDPM's capacity to generate novel, non-redundant designs, supporting its application as a creative tool in conceptual aircraft design.