Url https://www.cimne.com/sgp/rtd/Project.aspx?id=908
LogoFeder
Acronym SMiLE
Project title Machine Learning for data-driven modeling
Reference PID2020-113463RB-C33
Principal investigator Pedro DÍEZ MEJÍA - pdiez@cimne.upc.edu
Matteo GIACOMINI - mgiacomini@cimne.upc.edu
Start date 01/09/2021 End date 31/08/2024
Coordinator UNIZAR
Consortium members
  • UPC
Program P.E. de I+D+i Orientada a los Retos de la Sociedad Call Proyectos de I+D+i 2020
Subprogram Retos Investigación Category Nacional
Funding body(ies) MCIU Grant 96.800,00 €
Abstract SMiLE builds upon Scientific Machine Learning as an emerging research field focusing on the exploitation of successful machine learning techniques from computer science for the solution of complex problems in physical sciences and engineering. Specifically, SMiLE will contribute in three industrial challenges jointly determined with SEAT, World Sensing and ESI-Group. It is designed as a unique project where everybody works together and contributions from the different teams are interweaved to exploit synergies and complementarities. The project will develop novel computational strategies to simulate industrial problems combining data-intensive technology and frontier physics-based models and solvers (including reduced order models). Each group, will bring their own individual expertise in every angle of the research. But the project has been designed as a unique project where every team works together. Thus, the three teams will contribute to every task, although only one leader will be in charge of the execution of each task. Each scientific and technological objective in SMiLE has lead researcher from any the three institutions. They are chosen depending on their expertise and assigned tasks independently of their location. CIMNE will contribute in every task. But it will lead two specific ones setting the foundations for cognitive digital twins. These tasks are: T1.1 Robust full-order solvers in robust and efficient computational engineering solvers and T2.2 Uncertainty Quantification in model updating, data assimilation and Uncertainty Quantification. The first one entails the development of novel extremely robust solvers to train machine learning to avoid failure induced by output scatter. By contrast, the second one focusses on credibility (accounting for uncertainty of data) and oriented to rare events, resulting from exceeding critical values of the quantities of interest.
Proyecto PID2020-113463RB-C33 financiado por MCIN/ AEI /10.13039/501100011033