Incorporating scaled conjugate gradient (SCG) in system identification for aircraft stability and control derivatives

The current investigation has been carried out on a unique approach i.e. supervised machine learning algorithm. The proposed scaled conjugate gradient (SCG) training method is used for trained the flight simulated data and to evaluate all the stability and control derivatives parameter estimation has been used for longitudinal short period mode. For the parameters like angle of attack, Pitch rate, elevator deflection as neural network inputs, lift and moment coefficients as output respectively considered in investigation. The aim is to train the delta method using scaled conjugate gradient to learn and predict aerodynamic derivative coefficients. This approach is applied on simulated flight data with various neural network model (SCG) thoroughly evaluated and validate with true flight data. It has been observed that the mean square error (MSE) value is very minimal for scaled conjugate gradient (SCG) as compared to Levenberg-Marquardt Optimization (LMO) method. This method’s accuracy is benchmarked against standard backpropagation algorithms to analyze the problem and predict aerodynamic parameters using neural network-based methods. In this study, all the models are compared based on their training performance and standard deviation of their estimates and actual values. The suggested SCG model will reduce the effort cost as well as experimental time of researchers to find accurate parameter estimated derivatives.