A digital twin is a synchronised dynamic model of a physical system/s, that is capable of bridging the gap between the virtual and physical worlds in real time. These complex ensembles are developed with the a goal to understand, model, simulate, predict, optimize and control the functionalities, that its own physical twin would be capable of executing under operating conditions.
In this project, highly resolved measurements of velocity and pressure variables will be integrated with Navier-Stokes equations for assessment of dedicated engineering flow-field applications. Physics-Informed Neural Networks (PINNs), that work on the principles of deep learning, will be used to develop new tools for real time management of streamed data, from an in-line coherent image set with the goal of updating a digital twin. A PINN is a network based data assimilation method, which involves loss functions composed of residual data, as well as partial differential equations that can encapsulate knowledge of physics, placing them in the classification of universal function approximators.
Keywords: Experimental mechanics, Digital Twins, Physics Informed Neural Networks, CFD simulations, laser measurements