1. Obtaining 3D3C PIV results from 2D2C data using PINNs

Particle Image Velocimetry (PIV) is an experimental fluid velocity and property estimation method. The procedure seeds minute tracer particles into fluids, which are then illuminated by a light source before being photographed by one or more high-speed cameras to capture two successive images with a very small capture time between them. The distance covered by the particles in the images are used to estimate velocity. More recently, with the advancement in neural networks, different applications of CNNs, GANs, and PINNs have allowed researchers to innovate alternate methods to perform PIV without the need for traditional processing with significantly improved accuracy and processing time.

Typical PIV images with various seeding densities

PIV is traditionally limited to two dimensions with one camera. Setups involving multiple cameras offset by a certain angle allow capturing depth information in a volume instead of a single plane, where volumetric processing then allows resolving for three-dimensional velocities. However, obtaining 3D data to process into 3D velocities is not efficient. The requirement of multiple cameras and illuminating entire volumes with powerful lasers makes the experimental setup more costly and complicated. As such, there is a need to obtain realistic three-dimensional velocities in a more cost-effective manner.

PIV Principle of Operation

This project combines optics information from the camera and its associated lens physics to transform 2D velocities to 3D velocities. Through the use of PINNs, we are developing a new method that takes in time and spatial coordinates and outputs the velocity of the fluid in three dimensions along with its pressure. This will be of great benefit as we will be capable of producing realistic flow estimates without the need for complicated and expensive experimental hardware.