Prediction of wing aerodynamic coefficients of an unmanned light electric aeroplane with ANN

Purpose To determine the optimal wing taper ratio of an unmanned light electric aeroplane, in this study an artificial neural network is developed and optimised to predict the wing aerodynamic performance at various angles of attach and taper ratios, with the wing lift and drag coefficients obtained from 3D wing CFD simulations. The effects of neuron and hidden layer numbers, i.e. the network configuration, on prediction accuracy are systematically investigated to address the issue of underfitting and overfitting to obtain the optimised network. Design/methodology/approach The 3D wing CFD simulations are first carried out to obtain aerodynamic performance data of an unmanned light electric aeroplane wing at various angles of attach and taper ratios. The effects of neuron and hidden layer numbers on prediction accuracy are then investigated with a feedforward ANN method, using the simulated 3D wing aerodynamic data and the 2D NACA 0012 airfoil aerodynamic data taken from an open source. The ANN model development and optimisation are performed in MATLAB. The ANN model is then used to determine the optimal wing taper ratio or range of the ratios. Findings It is found that the optimal configuration of the ANN model has one hidden layer and six neurons for the 3D data in application. In general, use of more hidden layers more likely causes overfitting than use of more neurons. For the 3D wing in consideration, the Cl/Cd ratios are very similar within the taper ratio range of 0.2–0.6 at the design cruise speed. The taper ratio of the wing design may finally be decided by other factors, such as the needs of structural optimisation. Research limitations/implications The optimised ANN model can be used to replace CFD simulations in aircraft shape optimisation to predict the wing aerodynamic performance in much shorter time and at a lower cost, with more detailed information in a broader design space. Practical implications The ANN model is optimised based on the data available for model training and validation. The detailed information and modelling process presented in this paper on the application of ANN are useful and valuable for aerospace engineers to perform their optimisation work. Social implications The research is about design optimisation of an unmanned light electric aeroplane that got many applications in modern society today. Originality/value The effects of neuron and hidden layer numbers on prediction accuracy of a feedforward ANN are investigated systematically to address the issue of underfitting and overfitting of the network for obtaining the optimised ANN model. The optimal wing taper ratio or range of the ratios of an unmanned light electric aeroplane are obtained based on the wing lift and drag coefficients predicted with the optimised ANN model.