Neural Network Modeling of Transonic Buffet on the NASA Common Research Model

The application of reduced-order modeling (ROM) techniques in the context of aerodynamic nonlinear system identification of realistic aircraft configurations gained increasing attention in recent years. Therefore, in the present study the application of a recurrent neuro-fuzzy model (NFM) that is serial connected with a multilayer perceptron (MLP) neural network is introduced concerning the computation of transonic buffet aerodynamics. In particular, the intention of the ROM is the prediction of coefficient time-series trends in contrast to a precise resolution of detailed flow effects. Further, a reduction of computational time compared to a full-order reference Computational Fluid Dynamics (CFD) solution is pursued. The training of the ROM is accomplished based on a data set computed by means of unsteady Reynolds-averaged Navier-Stokes (URANS) simulations. The performance of the trained ROM is demonstrated by predicting the buffet flow characteristics of the NASA Common Research Model (CRM) investigated at transonic flow conditions. Therefore, the wing of the configuration is excited by an external pitching motion beyond buffet onset. By comparing the ROM result with a reference URANS solution, a precise prediction capability of the aerodynamic characteristics as well as a reduction in computational time is demonstrated.