Model updating of structures with uncertain multiple parameters: from traditional trial-and-error approaches to autonomous intelligent agents

The assessment of the health state of existing structures over time represents a main challenge for structural engineers, which need to control the occurrence of damage. To this end, continuous structural health monitoring (SHM) is suggested, which allows to elaborate measures from environmental actions to define the dynamic behavior and its evolution. Additionally, to enhance the performance level, calibrated numerical models can be used through a model updating (MU) process, in which numerical outputs are matched with the experimental ones. In this context, different approaches can be exploited for efficiently performing MU. The most traditional method consists of a trial-and-error approach, in which the uncertain parameters of numerical model are varied until matching numerical and experimental data. Nevertheless, the reliability of the results is affected by the base assumptions and by the analyst’s experience. When increasing the number of uncertain parameters, the above approach is not efficient and a viable solution for enhancing MU is represented by artificial intelligence. The paper presents, through some case studies, a possible paradigm shift from traditional to innovative MU, in which an intelligent agent can be trained to tune uncertain parameters and then improve the prediction of the behavior of existing structures.