Real time estimation of impaired aircraft flight envelope using feedforward neural networks

Abstract

Extensive research in recent years has focused on developing flight envelope estimation methods to improve current loss of control prevention and recovery systems. Such methods are practically efficient only if they are able to evaluate in real time the new flight envelope of damaged aircraft based on the altered dynamics. Due to nonlinear dynamics of aircraft, common approaches to estimate the entire flight envelope of high-fidelity models are numerically intensive and their real time implementation is computationally impossible. So current methods are based on reduced complexity models or flight envelopes are determined locally. This paper presents a novel method to estimate the global flight envelope of impaired aircraft in real-time for any unknown failure degree. In the proposed method, first, numerous flight envelopes are evaluated using a high fidelity model at various failure degrees and different flight conditions and prepared as training data. Then multiple feedforward neural networks are trained offline by a Bayesian regularization backpropagation algorithm. Finally, the trained networks are used to estimate flight envelopes in real time. The method is applied to rudder and aileron failure cases of the NASA Generic Transport Model. Results show that the estimated flight envelopes are good approximations of the high fidelity global flight envelopes.