Multifidelity Surrogate Aerodynamic Model with a New Adaptive Sampling for Rotorcraft Design

This paper presents a multifidelity surrogate modeling architecture to estimate the aerodynamics of complete rotor systems. A novel adaptive sampling method, rCorSpa, is designed to achieve global aerodynamics estimation and avoid failure design space. It introduces the concepts of a safe ball domain and a compound shell region, which guide sampling near complex failure boundaries. These regions are sampled independently to balance global exploration and failure avoidance. The core design space for rotor aerodynamics is uncovered, which enables the complex domain to be reconstructed and simplified. As this core design space inherently represents a lowerdimensional subspace, the cost of failure estimation is significantly reduced via space projection. Building on this foundation, a sphere contraction technique and the concept of a convergent radius are introduced to further minimize the number of redundant sampling iterations. Overall, the new sampling method achieves substantial computational savings by both simplifying core formulations and reducing the total number of required sampling steps. The proposed surrogate aerodynamic model is applied to a multicopter rotor system modeling scenario. Compared with a conventional hierarchical kriging model, the rCorSpa-enhanced surrogate model achieves a 95% improvement in global prediction accuracy, reducing the error to 3.5% after convergence.