Abstract. Future utilization of the International Space Station (ISS) exhibits a demand for frequent payload return by means of small unmanned re-entry capsules. Conventional propulsive deorbit systems could be replaced by tether sys-tems that yield high system mass savings. In order to guarantee sufficient landing accuracies, the tether deploy-ment has to be controlled. Beside conventional methods the use of an adaptive neural controller is proposed. The present paper demonstrates the successful identification of the highly non-linear and time-variant tether dynam-ics, using feed-forward-networks. An operating point is selected along a predefined optimal tether deployment path, where disturbances are imposed. The accumulated deviations from the reference path are calculated by forward integration of the equations of motion. The training patterns obtained are transformed into dimension-less state space by applying the Pi-Theorem of Buckingham. The results obtained provide the basis for a future development of an indirect neural controller.

Keywords: tether system, re-enty capsule, frequent payload return, neural controller, system identification, similarity network, Pi-Theorem