A novel use of machine learning for the modeling of wing acoustics: A study on the effects of winglet cant angle

Since the 1970s, winglet devices have been widely utilized to ameliorate the aerodynamic performance of aircraft by increasing L/D of the wing configuration. Modern designs can change cant angle during the flight in order to maximize their benefits to all flight states, which is limited to on-design points in passive/fixed designs. This paper discusses the effects of atmospheric and operating parameters on the aeroacoustics of a double-swept wing configuration facilitated with a rotary winglet device. To do so, the turbulent flow is simulated via 3D RANS formulation and k-ω SST turbulence model. Then, machine learning tools, consisting of Multi-layer Perceptron (MLP) neural networks and supervised classification methods are used to generate scaler regression models based upon numerical aeroacoustic datasets. Moreover, deep Convolutional Neural Network (CNN) is used to estimate the aeroacoustic field. The results depicted that changing the cant angle significantly affects the aerodynamic noise of the wing configuration regardless of operating conditions. Secondly, artificial intelligence is a practical modeling tool for acoustic parameters with reasonable accuracy and cost compared with RANS simulations.