An Active Structure that Learns

Abstract

Tensegrity structures are composed of cables and struts that become stable through self stress. They are good candidates for implementation of active structural control because their flexibility may mean that they cannot meet serviceability criteria. Changes to the self stress influence the form of the structure. A reliable closed-form solution for obtaining control commands for telescopic compression elements in order to obtain a required shape does not exist for such a closely coupled and geometrically nonlinear structure. Simulating the structural behavior after all possible control commands and testing against constraints and the objective function requires computational times that grow exponentially with the number of actuators. This paper demonstrates that search time can be reduced through use of stochastic search methods and that incrementally storing successfully applied control commands in a case-based reasoning system increases performance during service lives (learning). Such results demonstrate that enhancing control with advanced computing methods provides opportunities for innovative structures.