Surrogate modeling for conceptual design of blended wing body UAVs

Aerodynamics designs involve multiscale engineering, where extensive investigations at different design stages must be carried out with prior risk anticipation. A cost-effective and reliable approach is to integrate and test more requirements and design variables early in the conceptual design stage. This requires a computationally efficient method to analyze large datasets and achieve feasible designs. One way to meet this requirement is, on the one hand, through appropriate selection of computational fluid dynamics fidelity, and on the other hand, by using statistical techniques and regression methods, which have proven effective for metamodeling computer outputs. In this paper, Gaussian process regression is used to emulate an aerodynamic potential solver for detailed and efficient analysis of key design variables affecting the longitudinal static stability of a tailless blended wing body planform. It is shown how, considering the computer outputs as a realization of a Gaussian process, regression from a Latin hypercube sampling leads to model construction. Cross-validation is used to select the best model that predicts accurately unseen data. The surrogate model predictions were compared against unseen XFLR5 configurations, showing close agreement in pitching moment coefficients. The analysis identified root chord and kink offset as the most influential geometric parameters affecting stability. These findings confirm the robustness of surrogate modeling coupled with panel methods for rapid conceptual design of UAVs. • Gaussian Process Regression used to emulate CFD solver results. • Aerodynamic solver XFLR5 selected for fast BWB conceptual analysis. • New stable BWB UAV designs found with minimal CFD computation. • Design variable interactions analyzed through surrogate modeling.