A modeling approach for trailing edge flap forces and moments in the preliminary aircraft design stage

Abstract

Accurately predicting the forces and moments acting on trailing edge devices under different flight conditions is a critical aspect in the design of the kinematics and actuation for high-lift or variable camber applications. However, accurate modeling without elaborated CFD analyses in the subsonic and transonic regimes needs a sophisticated model. Thus, the main objective of this paper is to create such a model that accurately predicts the forces and moments acting on flaps during different flight conditions while remaining applicable in the preliminary aircraft design. In particular, this means remaining sufficiently universal to be applicable to different aircraft types and be computationally efficient and not require excessive computation time. The target values in this model are the 3D forces and moments on the flap, which were obtained through 3D CFD simulations. The chosen input values required for the model include 2D airfoil data, and wing geometry data for three different aircraft types: short-, medium-, and long-range, including a high aspect ratio configuration. Among several potential approaches, a neural network was deemed to be the most promising for predicting the target values. The neural network was used as a regression tool to accelerate the model development process in the preliminary aircraft design. Given the large number of parameters, manual regression would not have been feasible. Consequently, multiple studies were conducted on how the setup of the neural network, including the number of nodes, activation functions, and initialization, impacts the results. The results reveal that the developed neural network accurately predicts the flap forces and moments with a mean deviation of under 2% for the vertical force Fz and the lateral force Fx and under 4% for the moment My. The findings suggest that the trade-off between accuracy and generalization was coped during the training procedure. In addition, the network demonstrated robustness in capturing variations in flight conditions and flap settings in test scenarios outside the training set. The application of the neural network allowed for precise predictions, which in turn allowed for the design of the actuation system based on reliable data. The study highlights the potential of neural networks for other applications within the preliminary aircraft design, which often relies on semi-empirical methods.