A comparison of adaptive optimizers for nonlinear aerodynamic modeling using flight test data

In this paper, adaptive optimizer-based deep neural network approaches are used to predict nonlinear aerodynamic model using flight test data of standard aircraft. Adaptive optimizers namely Adam and RMSprop algorithms are chosen to model the force and moment coefficients during steady stall phenomena. The effectiveness of these two methods are being investigated and validated. The estimated results from adaptive optimizer based methods are statistically analysed and compared with the conventionally used maximum likelihood method. Comparison results from the above methods are found to be relatively better than the maximum likelihood estimates in terms of RMSE and correlation. Moreover, the adaptive optimization methods are proven to be advantageous over conventionally used methods which strongly depend on solving equations of motion.