Digital twin framework using agent-based metaheuristic optimization

Despite advances in instrumentation and measurement techniques, it is still necessary to update numerical models to simulate or predict some structural responses, for example. Thus, this work proposes a metaheuristic framework based on hybrid agents, an approach within the Artificial Intelligence (AI) topics for updating Finite Element (FE) numerical models. This framework aims to provide flexible non-deterministic strategies to guide the updating process, ranging from simple local search procedures to complex learning processes. Two case studies are presented: (i) a free-free aluminium beam tested under laboratory conditions and; (ii) a catamaran tested during a sea trial under real operating conditions. The updating process aimed to optimize the stiffness matrix while maintaining the mass matrix unchanged. The objective function seeks to minimize the differences between numerical and experimental modal parameters, namely, natural frequencies and vibration modes. Results from the digital twin framework showed that the difference in natural frequencies significantly decreased, for example, 9% to 1% for the free-free aluminium beam and 15% to 4% for the catamaran's main deck, when comparing the experimental with the updated FE model. As for the updated FE vibration modes, the Modal Assurance Criteria (MAC) values decreased slightly in both cases but within the acceptable MAC values (above 0.9), thus showing good consistency with the experimental vibration modes. In the end, the proposed framework was able to update the FE model directly using its respective reduced model, circumventing the"black box"of commercial packages.