Learning, self-diagnosis and multi-objective control of an active tensegrity structure

Abstract

This paper presents a full-scale active tensegrity structure at EPFL and demonstrates how it can learn as well as carry out self-diagnosis and self-compensation. Tensegrities are generally flexible structures: small loads may lead to large displacements. We thus control slope by actively modifying the self-stress state between cables and struts. The structure benefits from past experience through case-based reasoning. It memorizes past control commands and adapts them in order to react to new applied loads up to forty times more rapidly than without this previous control information. Redundancy of this structure provides opportunities for "fault tolerant" behavior. The active control system can also be used to perform self-diagnosis and then to self-compensate local damage. For many cases of local damage, the structure remains capable of satisfying control goals. This paper also summarizes a multi-objective optimization method for control according to four criteria. In contrast with other applications involving multiple objectives, such as design where users prefer choices, this is a control task, thereby requiring identification of a single solution only. Also, the single dominant objective usually generates hundreds of possible solutions. Four objectives are evaluated firstly using Pareto optimality and then a unique solution is chosen through successive filtering of candidate solutions using a hierarchy of objectives. The combination of advanced computing techniques with structural control of serviceability criteria is providing many new possibilities for structural engineers. These results are expected lead toward more autonomous and self-adaptive structures that are able to evolve as their environment changes. © 2006 Springer.