Enhancement of PIV measurements via physics-informed neural networks

Abstract

Physics-informed neural networks (PINN) are machine-learning methods that have been proved to be very successful and effective for solving governing equations of fluid flow. In this work we develop a robust and efficient model within this framework and apply it to a series of two-dimensional three-component (2D3C) stereo particle-image velocimetry (SPIV) datasets, to reconstruct the mean velocity field and correct measurements errors in the data. Within this framework, the PINNs-based model solves the Reynolds-averaged-Navier-Stokes (RANS) equations for zero-pressure-gradient turbulent boundary layer (ZPGTBL) without a prior assumption and only taking the data at the PIV domain boundaries. The TBL data has different flow conditions upstream of the measurement location due the effect an applied flow control via uniform blowing. The developed PINN model is very robust, adaptable and independent of the upstream flow conditions due to different rates of wall-normal blowing while predicting the mean velocity quantities simultaneously. Hence, this approach enables improving the mean flow quantities by reducing errors in the PIV data. The PINNs-predicted data have less than 1\% error and is in excellent agreement with reference data. This shows that PINNs has potential applicability to shear-driven turbulent flows with different flow histories for predicting high-fidelity data.

Past and current components-based detailing of particle image velocimetry: A comprehensive review

“…Recently, neural networks have been used with liquid phase extraction in two-phase flow [ 200 ]. Furthermore, physics-informed neural networks are applied to stereoscopic PIV to enhance flow visualization [ 201 ].…”

Super-resolution Left Ventricular Flow and Pressure Mapping by Navier-Stokes-Informed Neural Networks

“…Among the growing body of DL models, PINNs are well suited to study flow phenomena since they effectively incorporate the laws of physics as constraints in the learning process [34, 65]. PINNs have been used to enhance flow measurements from laboratory experiments [66] and medical imaging acquisitions [31, 36]. However, its applications to recover and visualize intracardiac flows from clinical data are scarce, and only preliminary approaches to the cardiovascular system have been recently conducted [31,67,68].…”

Temporal and spatial flow field reconstruction from low-resolution PIV data and pressure probes using physics-informed neural networks

“…In flow field reconstruction using PIV, the adoption of PINNs is increasing [17,18]. Hasanuzzaman et al [19] developed a robust PINN model to enhance stereo PIV datasets, enabling error correction and mean velocity field reconstruction. Fan et al [20] demonstrated the use of PINNs to extract pressure data from velocity fields in PIV, highlighting their application in synthetic jet analysis.…”