Adaptation and validation of PINNs improve depthwise motion estimations in 3D velocimetry, suggesting enhanced accuracy.
Key Points
This research aims to adapt and validate physics-informed neural networks for improving velocity measurements in rotating three-dimensional velocimetry.
Adaptation of physics-informed neural networks (PINNs) for R3DV applications.
Experimental validation against high resolution stereoscopic particle image velocimetry (stereo-PIV).
Integration with plenoptic particle tracking velocimetry (PTV).
Exploration of design considerations and hyperparameters affecting performance.
Out-of-plane velocity error reduced by an average of 72.6% compared to plenoptic-PTV.
Demonstrated improved accuracy in depthwise motion estimations under various Rossby numbers.
Identified critical adjustments for conventional PINN configurations to minimize errors.