This paper presents progress in the field of adaptive civil-engineering structures. Selfdiagnosis, multi-objective shape control and reinforcement-learning processes are implemented within a control framework on an active tensegrity structure. Self-diagnosis extends active structural control to situations of partially defined loads. Multi-objective search is useful for computing commands that control shape while minimizing active strut stroke and stress, and maximizing stiffness. Reinforcement learning improves the control by memorizing, retrieving and adapting previous control events. The control framework is validated experimentally on an active tensegrity structure. This provides an example of an adaptive civil-engineering structure.

This study focuses on improving structural control through reinforcement learning. For the purposes of this study, structural control involves controlling the shape of an active tensegrity structure. Although the learning methodology employs case-based reasoning which is often classified as supervised learning, it has evolved into reinforcement learning, since it learns from errors. Simple retrieval and adaptation functions are proposed. The retrieval function compares the response of the structure subjected to current loading event and the attributes of cases. When the response of the structure and the case attributes are similar, this case is retrieved and adapted to the current control task. The adaptation function takes into account the control quality that has been achieved by the retrieved command in order to improve subsequent commands. The algorithm provides two types of learning: reduction of control command computation time and increase of control command quality over retrieved cases. Results from experimental testing on a full-scale active tensegrity structure are presented to validate performance.

This paper addresses the study of tensegrity active control in case of unknown events, such as applied loading or damage. It describes methodologies for self diagnosis and self repair. Response due to unknown events is measured and analyzed in order to support self diagnosis. Since tensegrities are self-stressed and flexible structures, they exhibit geometrical non-linear behavior. Applied loading and damage thus induce changes in structural response to perturbations. This property is also used to support self diagnosis. Self-diagnosis solutions result in sets of good candidate description of the unknown event. Candidate descriptions exhibit responses to unknown events and perturbations that are close to the response measured on the real structure. These solutions are successfully employed within the framework of shape control and self repair. Self-repair abilities are demonstrated through increasing stiffness and decreasing stresses with respect to the damaged state by modifying the self-stress state of the structure. Validity of the results is demonstrated experimentally on a full-scale active tensegrity structure. The proposed methodologies are attractive for tensegrity active control in situations of unknown events.

A multi-objective search method is adapted for supporting structural control of an active tensegrity structure. Structural control is carried out by modifying the self-stress state of the structure in order to satisfy a serviceability objective and additional robustness objectives. Control commands are defined as sequences of contractions and elongations of active struts to modify the self-stress state of the structure. A two step multi-objective optimization method involving Pareto filtering with hierarchical selection is implemented to determine control commands. Experimental testing on a full-scale active tensegrity structure demonstrates validity of the method. In most cases, control commands are more robust when identified by multi-objective optimization method as compared with a single objective and this robustness leads to better control over successive loading events. Evaluation of multiple objectives provides a more global understanding of tensegrity structure behavior than any single objective. Finally, results reveal opportunities for self-adaptive structures that evolve in unknown environments.

No abstract