Estimation of Aerodynamic Partial Derivatives using Neural Network and Differential Variation Technique

Feed Forward Backpropagation Neural Network (FFBPN) and Differential Variation Technique (DVT) is proposed to extract aerodynamic partial derivatives from flight data of test aircraft. The proposed combination draws its inspiration from FFBPN based the Delta Method (DM) for extracting aerodynamic partial derivatives of longitudinal dynamics of an aircraft. Firstly, FFBPN based network is trained by presenting differential variation in angle of attack, pitch rate and elevator deflection as inputs while lift or moment coefficients as corresponding outputs. For estimation of derivatives, predicted values of lift/moment coefficients are obtained for modified input file of motion and control variables. The modified input file and corresponding predicted output file are then used for the estimation of partial derivatives. Further, the aerodynamic model of estimated derivatives is also validated by comparing the estimated response with flight response which is not used for the estimation of derivatives.