This presentation by Francisco Chinesta was made at the NAFEMS eSeminar "Simulation & Digital Twins - Behind the Buzzwords" on the 2nd of May 2018.

In the third industrial revolution, “virtual twins"" emulating a physical were major protagonists. Nowadays, numerical simulation is present in most scientific fields and engineering domains. However, virtual twins are usually static. The reason is that the characteristic time of standard simulation strategies is not compatible with the real-time constraints, which is compulsory for control purposes.
Even if HPC could alleviate that issue, it renders the accessibility to the appropriate simulation resources challenging for small and medium-sized companies. Thus, in order to democratize simulation, new solutions are required. Model Order Reduction (MOR) techniques open new possibilities for more efficient simulations, among them Proper Orthogonal Decomposition – POD-, Reduced Bases – RB- and Proper Generalized Decomposition - PGD.

The next generation of twins, the “digital twins"", allowed assimilating data collected from physical sensors, to be the main tool of identifying parameters to represent the model. It also allowed for predictive capabilities based on time evolution of these parameters, in real-time, to anticipate actions. Thus, simulation-based control was possible, and implemented even in deployed computing devices (e.g. Programmable Logic Controller, PLC).
Despite an initial euphoric period, unexpected difficulties appeared. In practice, significant deviations between the predicted and observed responses were noticed. The origin of the just referred deviations is due to inaccuracy in the employed models. In order to address this inevitable “ignorance"", one approach lies in constructing “on-the-fly"" a data-driven model able to fill the gap between prediction and measurement. Indeed, Hybrid TwinTM consists of three main ingredients: (i) a simulation core able to solve complex mathematical problems representing physical models under real-time constraints; (ii) advanced strategies able to proceed with data-assimilation, data-curation and data-driven modeling; and (iii) a mechanism to adapt the model online to evolving environments (control).

Indeed, Hybrid TwinTM  consists of three main ingredients: (i) a simulation core able to solve complex mathematical problems representing physical models under real-time constraints; (ii) advanced strategies able to proceed with data-assimilation, data-curation and data-driven modeling; and (iii) a mechanism to adapt the model online to evolving environments (control).