The goal is to control a serviceability criterion related to the structure in order to improve its usefulness. Our serviceability criterion consists of maintaining a constant slope of the upper surface of the structure. When a load alters the slope of the structure, we apply a control command. This modifies the self stress state and makes it possible to recover the initial slope.

However, finding an efficient control command is not an easy task due to high coupling between the elements and geometric non-linearity. Consequently we use a generate-analyze-verify process with stochastic search and case-based reasoning (Domer and Smith, 2005). The PGSL (Probabilistic Global Search Lausanne) stochastic search algorithm used in our case is a direct search algorithm developed at EPFL (Raphael and Smith, 2003). In order to take advantage of previous experience, altered configurations and corresponding control commands are stored in a case-base. When the structure is subjected to a load, the nearby configuration is retrieved from the case base and its corresponding control command is adapted to the new task. As cases are added in the case-base the average time necessary to identify a control command decreases (learning). Since the structure is able to improve performance progressively using past experience we considerer this to be a characterization of an intelligent structure.

In the case of local damage, localization is the “inverse problem” of determining a cause given an effect. It is possible to perturb the structure through micro-movements (±1mm) and measure its response through six indicators: RMS variation of the vertical displacements at the three measured nodes (37, 43, 48, see Figure 2) due to the micro-movement and slope variation in three directions due
to the micro-movement. For this simple case, we used the algorithm described below. Measured indicator variations on the damaged physical structure are compared with numerically simulated indicator variations due to the same micro-movement and assuming a candidate local damage position. If measured and simulated variations vary in the same way, the assumption is kept. Otherwise the candidate is rejected and another local damage assumption is evaluated. The space of possible damage positions reduces iteratively with micro-movements until the damage is localized.

Once the damage is localized, a self-compensating control command can be applied in order to satisfy the serviceability criterion. The structure can then be controlled despite a loss of carrying capacity. The principal constraint of this task is serviceability and not structural safety. Contrary to most traditional civil structures, the serviceability and structural safety of tensegrities are often conflicting objectives since measures that increase serviceability (for example, prestress) may lower structural safety through lower load-carrying capacity.