RO3 : Hybrid twins: simulation, learning

Activités

The aim of this research operation is to integrate the data observed in the experimental framework and synthesised by numerical simulation, in order to establish a new "grey box" paradigm for the design and maintenance of infrastructures, critical structures and buildings. In order to face the necessary adaptation of structures to climate change and natural disasters, the activities of the RO aim, in coordination with the other ROs of the OMEIR team, to define and build hybrid and multi-physical twins of future and existing structures, to make the design phase robust with respect to uncertainties (intrinsic and linked to external environmental agents), to plan and adapt maintenance, to assess the health status of the structure, and to predict its ultimate life time. The framework is that of machine learning, which is necessary to design fast meta-models that integrate or replace numerical simulations or experimental campaigns. The hybrid twins provide for direct and inverse simulation and metamodelling to adjust the prediction and investigate the uncertainty and correlation of the input parameters with respect to model sensitivity.

In order to achieve these objectives, the RO aims at exploiting BIM, high performance computing, learning and AI algorithms and data acquisition, uncertainty processing and sensitivity analyses to propose a local and global estimation of the risk (simple or due to cascading effects) associated to each infrastructure in its environment.