Neural Network based Robust Integrated Autopilot for an Aircraft

This work focuses on the usage of Neural Networks (NN) for uncertainty and disturbance estimation in nonlinear flight control systems. The Uncertainty and Disturbance Estimator (UDE) control strategy is considered for mitigating the effects of uncertainties and disturbances present as system nonlinearities, parametric variances and variable external disturbances in the aircraft. The need to overcome difficulties of complexity, nonlinearity and uncertainty has led to utilization of Neural Networks. Hence, Neural Network models that are capable of mimicking the uncertainty and disturbance estimation in the controller employing the UDE strategy are explored in this paper. Multilayer Perceptron Neural Networks have been trained on-line with the Back-propagation algorithm and used in these controllers to obtain the desired tracking performance. In the Neural Network-based controller, two types of activation functions have been applied, and their performances are compared. Finally, the performance of the Neural Network based controller in autopilot systems has been compared with that of the conventional controller employing the UDE strategy.