Mardi 20 février à 11h00 / Amphithéâtre Canac, LMA
Abstract : The real-time monitoring of real-life engineering systems, by coupling simulation tools and observed data (Dynamic Data Driven Application Systems – DDDAS), is made very difficult in practice due to several issues. In particular, the complex nonlinear multiscale phenomena which are involved may be associated with computationally intensive simulations (hardly compatible with real-time) which requires reduced order modeling and/or model coarsening. In addition, the problem is plagued with model bias, uncertain environment, and measurement noise, which need to be taken into account for robust data assimilation, accurate diagnosis and prognosis, and safe decision-making. An appealing trend is to refer to hybrid techniques, in which an a priori physics-guided model is updated and enriched with data-based information, thus making benefit of all knowledge available.
In this context, we present some effective methods and recent developments for the construction of such hybrid twins, with practical applications on the online monitoring of engineering systems (e.g., control of additive manufacturing processes, or structural health monitoring) with sequential assimilation of real data. We also give some insights on the beneficial use of deep learning in such hybrid strategies.