RBF Neural Network Modeling and Validation for FADS System Applied to the Vehicle with Sharp Wedged Fore-Bodies

In allusion to the difficulty in aerodynamic modeling of the flush air data sensing(FADS) system for sharp wedged fore-bodies, an extended minimum resource allocating network radial basis function(RBF) neural network was developed as the FADS system model for its good generalization capabilities and compact algorithm structure. Single hidden layer artificial neural networks based on RBF were designed to generate regression model for the FADS system. The neural networks were trained and tested with pressure measurements inputs to model the aerodynamic relationship between aircraft surface pressure and air data states. RBF neural network model was then used to calculate and predict the Mach number, angle of attack, angle of sideslip and free stream static pressure displacing, rather than the traditional aerodynamic model. Computational fluid dynamic(CFD) simulations were implemented to identify the ideal pressure port locations, and wind-tunnel tests and CFD results were carried out to train and test the extended minimum resource allocating network RBF neural network. These models were then applied to pressure measurements during flight tests of the aircraft and sensing system to judge the viability of the method. The results show that, the RBF neural network model for the FADS system is accurate enough. Neural network outputs agree well with the real test data, and the testing error distribution (absolute error distribution) for angle of attack, angle of sideslip, Mach number and free stream static pressure is less than 0.25°, 0.5°, 0.05, and 250 Pa, respectively. The results also present that the RBF neural network model for the FADS system will be further developed in the future.