De-noising and super-resolution of fluid-flow velocity measurements by optimising a discrete loss (ODIL)

This article presents a comparison between two techniques – optimising a discrete loss (ODIL) and physics informed neural networks (PINN) – for reconstructing the velocity field from low resolution and noisy planar PIV measurements. Both techniques are capable of accurately reconstructing the velocity and pressure fields from low resolution and noisy velocity measurements sampled from 2D numerical simulations. Both techniques provide reasonable reconstruction of the in-plane velocity components when provided with two-component velocity measurements in a single plane sampled from a 3D numerical simulation. However, ODIL generally over-fits to any noise in the measurement data, and therefore PINN achieves higher accuracy. While PINN can achieve a reconstruction more accurate than that of ODIL, PINN converges much slower than ODIL, requiring substantially more training epochs and walltime to produce results of similar accuracy to ODIL. Both methods are superior to statistical noise reduction approaches, such as low-pass filtering. • Both ODIL and PINN can improve accuracy and reduce noise. • ODIL converges faster than PINN for the same accuracy. • ODIL is prone to overfitting, so PINN achieves a higher accuracy. • Both methods are superior to statistical noise reduction approaches.