Implementation of System Identification on Unmanned Aerial Vehicle via Subspace and Prediction Error Method

摘要

This paper describes an investigative modeling effort through system identification technique on an unmanned aerial vehicle (UAV). Linear state-space models corresponding to the longitudinal and lateral dynamics of a fixed-wing air vehicle are obtained through the implementation of two well-established identification methods, namely subspace method and prediction error method (PEM). Although both methods differ in terms of their fundamental principles, they complement each other very well and may be applied to the same dynamic system together thus improving the fidelity of the identified model. The combined subspace-PEM identification routine proposed in this paper is applied on the Spoonbill UAV system which is the latest long endurance unmanned aerial vehicle (UAV) system currently under development by the Remotely Piloted Vehicle and Microsatellite Laboratory (RJVIRL), National Cheng Kung University, Taiwan. Actual flight tests are being carried out to gather necessary input-output data for system identification. Specifically designed input flight maneuvers are executed during flight tests in order to yield better quality data which is crucial in producing good identified models. Identification results corresponding to longitudinal and lateral dynamics are presented and discussed in this paper. They show encouraging accuracy and the simplicity of the identification algorithm accentuate the benefits of the proposed identification scheme. Furthermore, the algorithm is applicable to any UAV system as long as the associated air vehicle is of conventional design.