System identification of an airship using trust region reflective least squares algorithm

Abstract

Lighter than air vehicles present feasible solutions to several problems in aviation industry. Dynamic modeling of airships, however, poses enhanced complexities due to the effects of buoyancy-based static lift and virtual mass and inertia. System identification is an established technique for modeling aerial vehicles, but it generally requires huge amount of flight data, acquired through costly sensors operating at high sampling rates. Earlier airship identification works have used output/filter error methods, evolution strategies and subspace identification methods; all using large sets of estimation data. In this research, the longitudinal dynamics of a 30 ft long unmanned airship have been modeled using very less estimation data. During airship’s flight experiment, flight data was recorded at a minimal sampling frequency of 8 Hz, using low-cost sensors. The less estimation data was compensated by iterative estimation technique, instead of one-step estimation. The flight data was subjected to trust region reflective least squares algorithm that is based on a relatively new and efficient optimization method. The model estimation quality was quantified by residual analysis and Akaike’s criterion of final prediction error. The promising cross-validation results show that the adopted identification approach is suitable and cost-effective for modeling of complex airship dynamics.