EFFICIENT MULTI-FIDELITY REDUCED-ORDER MODELING FOR FLUTTER PREDICTIONS ACROSS MULTIPLE MACH NUMBERS

Flutter prediction is an important phenomenon that need to be considered in aircraft design. However, highfidelity predictions for transonic flutter are difficult to make due to the associated computational costs. This paper proposed a multi-fidelity reduced-order modelling framework for flutter predictions to achieve high-fidelity predictions with less computational costs. Here, the high-fidelity data is obtained from a Navier-Stokesequation- based solver, while the low-fidelity solution is taken from an Euler-equation-based flow solver. Using the multi-fidelity neural network trained based on the multi-fidelity data, this approach can achieve online predictions of high-fidelity results. To demonstrate the multi-fidelity process, a widely used pitching and plunging airfoil case is considered. Verification of the approach is done by comparing with results from the time-domain aeroelastic solvers. The results show that the proposed multi-fidelity neural network modelling framework can realize the online predictions of unsteady aerodynamic forces and flutter results across multiple Mach numbers. Compared with the typical Co-kriging method, the proposed method has higher accuracy and stronger generalization capability. Finally, the method’s potential for reducing the computational effort of high-fidelity aeroelastic analyses is demonstrated.