Multi-fidelity multi-objective optimization of a high-altitude propeller

High-altitude propeller optimization aims to discover blade designs that yield superior performance in the low air density environment present at the lower levels of the stratosphere. In our work, a multi-fidelity, multi-objective (MFMO) optimization framework is developed and tested for the application of high-altitude propeller optimization. The optimization employs three levels of fidelity, including Vortex Theory and 3D RANS with the use of γ Reθ transition model, converged with first-order upwind, and second-order upwind for the momentum equations. The Variable Fidelity Expected Improvement Matrix (VFEIM) is used as an MFMO acquisition function and it is extended to account for failed designs, batch submission of infill designs, and geometric filtering of airfoils. Moreover, a Hierarchical Kriging surrogate model is used to fuse the performance data from the three levels of fidelities. Due to the large number of low-fidelity data, a Kriging approach is proposed to fit large, high-dimensional data. The proposed model is found to perform better on the tested propeller performance dataset compared to other popular techniques, such as Sparse Gaussian Processes. Moreover, through a test function example, the benefit of an MFMO optimization approach is examined showing sensitivity to the cost ratio between the different fidelities. Finally, the propeller optimization results demonstrate that high-performing designs are achievable with the proposed optimization framework.