Machine Learning: 

APRG’s current machine learning efforts focus on the applications of Physics-Informed Neural Networks (PINNs) for fluid simulations as an alternative to conventional CFD methods. One advantage of PINNs is that they will run without fully-defined boundary conditions, which allows for simulation without the need to know the specific values of all the boundary conditions.

PINNs can also perform simulations over a range of parameters. For example, rather than just using a PINN to simulate a flow through a tube, one PINN model can be used to simulate flows through tubes of different sizes, giving the ability to view how flows change in response to changes in tube diameter.

Another advantage of neural networks is their ability to work with data. Neural networks can be applied to both numerical and experimental results to smooth data and separate unknown flow parameters using assumptions about the problem geometry. PINNs can also use a few points of known data as boundary conditions to learn the rest of a flow using governing equations like the Navier-Stokes equations. This can be used to speed up training or to fill in experimental flow results that are missing data in certain regions