Recurrent Neural Network for Aircraft Parameter Estimation

This paper deals with the implementation of Delta method using Recurrent Neural Network (RNN) for estimation of stability and control derivatives in lateral-directional mode. The proposed method is implemented on simulated flight data and then on real flight data. The generation of data is done using the parameters of the research aircraft, ATTAS. The results obtained using RNN are further compared to results obtained by using Feed Forward Back propagation algorithm (FFBP) in tabular and graphical formats for both simulated as well as real flight data. It is found that the derivatives obtained using RNN are very close to true values of derivatives with lesser standard deviation as compared to derivatives obtained using FFBP algorithm. The results increase level of confidence and suggest that the RNN can be used advantageously to estimate aerodynamic derivatives of an aircraft from real flight data.