This paper was produced for the 2019 NAFEMS World Congress in Quebec Canada
Today, numerical simulation is utilised in most scientific fields and engineering domains, enabling accurate designs and virtual evaluation of systems responses, and allowing to drastically reduce the number of experimental tests. These numerical methods in engineering practice (finite element method, finite volume method, etc…) are nowadays ubiquitous in the design of complex engineering systems and their components. Unfortunately, these “virtual twins”, are considered as “static” because they are not able to be dynamically to be improved by incoming (sensor) data. Today’s smart systems connected via sensors are referred to as Dynamic Data-Driven Application System (DDDAS) and impose stringent real-time constraints, especially for control purposes and real-time decision-making. The computation time of standard engineering simulation strategies for high-fidelity models cannot accommodate the real-time constraints posed by DDDAS. “Digital twins” using deep-learning techniques allow for reduced computational time, but do not contain an understanding of the underlying physics and provide coarse approximations when compared with rich high-fidelity simulations. To satisfy these constraints, ESI introduces a “Hybrid Twin" framework, that combines physics-based models within a Model-Order Reduction (MOR) methodology for accommodating real-time feedback, quick decision-making and data science, etc. These techniques do not require simplification of the model; therefore, models continue to provide a well-established and validated description of the physics at hand. Instead, they rely on an adequate approximation of the solution that allows for the simplification of the solution procedure without any sacrifice on the model solution accuracy, in view of accommodating real-time constraints.