Neural Network Based Estimator for a Maneuvering Aircraft

Dynamically unstable aircraft and their supermaneuverability produce enormous complexities and tight constraints that existing control strategies cannot handle. Neural networks have emerged as a powerful means to deal with complex non-linear systems such as aircraft parameter identification during complex maneuvers. This paper addresses preliminary research conducted to train a neural network to predict aircraft aerodynamic coefficients, both singularly and simultaneously, during a high alpha maneuver. Using a backpropagation algorithm, we were able to train and optimize network training to predict the coefficients by changing the network characteristics. The network´s ability to reduce the error between actual aircraft data and the trained network output depends on the number of network layers, the number of neurons in each layer, the types of transfer functions used to represent the neurons, and the method by which the training data is presented to the neural network.