Aircraft system identification using artificial neural networks with flight test data

This paper presents linear system identification results using noisy data from a six-degree-of-freedom aircraft simulation and data obtained from flight test of an Unmanned Aerial System using the recently developed novel Artificial Neural Network System Identification algorithm. The method uses an artificial neural network with a single input layer and single output layer to learn the elements of the state transition and control distribution matrices directly. This results in a discrete time state-space model of the aircraft dynamics which is then converted to continuous time using standard techniques. For simulated data, the true linear model is used for verifying the identified model. Linear models are generated from the flight data and are compared with results obtained using the well-established Observer/Kalman Identification algorithm. Results show that the neural-network method identifies valid models for both longitudinal and lateral/directional axes, and performs comparably to Observer/Kalman Identification. Advantages of the method include ease-of-use and low complexity.