Sensitivity analysis of unsteady aerodynamic reduced-order models robust to freestream parameter variations

In this paper, the sensitivity and performance of an unsteady aerodynamic reduced-order modeling (ROM) approach robust to freestream parameter variations is analyzed. The considered ROM methodology is founded on system identification principles and neurofuzzy networks in order to obtain an efficient aerodynamic model from CFD-based input-output data. In terms of a sensitivity analysis, the user-defined parameters of the neurofuzzy model are varied systematically, while the ROM output for each set-up is compared with the full-order reference CFD solution. In this way, the influence of the ROM parameters is uncovered, while the focus is on a robustness and accuracy evaluation. Additionally, the number of freestream conditions incorporated within the training dataset is increased incrementally in order to study the dependence between the training information and the overall solution quality. The research is carried out based on the NLR 7301 airfoil in the subsonic and transonic flight regime. It is shown that good agreement is obtained between the ROM results and the respective full-order solution. Moreover, the ROM parameters exhibit a robust characteristic which is beneficial for practical applications.