Hybrid modeling for intelligent dynamical systems in mechanical engineering

Dimitri Goutaudier 

CR CNRS 

Laboratoire PIMM 

Amphi Esquillan

ABSTRACT

In this talk, I will first introduce myself as a new member of the PIMM Laboratory and briefly present my past works, before presenting my research project. The latter deals with the development of hybrid twins for intelligent dynamical systems in mechanical engineering. In the scope of my project, I think of intelligent dynamical systems as engineering structures or industrial processes evolving in time, equipped with sensors, and capable of using the sensor measurements with algorithms to predict future states, to detect anomalies and damages, or to adapt control laws to reach better performances. In this context, my objective is to develop an innovative framework for modeling such complex systems evolving under varying operational conditions. I will try to combine advanced model order reduction techniques, physics-based and/or data-based, with original inverse methods using as less sensors as possible. One of the novel ingredients of my approach is to relax the usual accuracy constraint on the reduced order model with respect to its underlying high-fidelity model, and instead to build it right from the start with the objective of identifying parameters in an efficient manner. Such methodologies could hopefully welcome various applications of interest for the laboratory.