Numerical framework for data-driven model-based monitoring of dynamical systems - Application to earthquake engineering

The EMSI laboratory (CEA Saclay) is equipped with shaking tables that allow to assess the seismic performance of civil engineering constructions. The control laws of the hydraulic actuators that drive them do not directly integrate the health condition of the tested structures, which can suddenly deteriorate and lead to unstable tests. If so, the integrity of the experimental setup is threatened.This thesis work aims to develop a data assimilation framework unified around the concept of modified Constitutive Relation Error (mCRE), in which a digital twin of the tested specimen is used to adapt control laws on-the-fly. First, a fully automated mCRE-based finite element model updating strategy is implemented in an offline context in order to perform robust and accurate parameter identification. The methodology has then been extended to online data assimilation by integrating the mCRE within a Kalman filter (MDKF). All numerical developments have been validated with simulated data from earthquake engineering problems, and successfully applied to the SMART2013 test campaign database. The possibility to monitor in real-time the modal signature of a reinforced concrete specimen from actual acceleration measurements illustrated the relevance and robustness of the proposed strategy.Eventually, the adaptive control issue has been addressed via a proof-of-concept in which a state-feedback command has been tuned in real-time with MDKF to stabilize simulated shaking table tests. This dynamical data-to-model interaction thus paves the way of future investigations for hybrid control strategies and new physics-guided structural health monitoring techniques.