Revealing hidden fluid dynamics with physics-informed neural networks from droplet impingement experiments

The present work introduces a data-driven approach for the spatio-temporal reconstruction of impinging droplet dynamics from simple optical experiments based on physics-informed neural networks (PINNs) and an advanced shadowgraphy technique with color-coded glare points. The novel approach is first validated on synthetic data obtained by direct numerical simulation by which a high spatial accuracy of the three-dimensional gas-liquid interface, as well as the inferred velocity and pressure fields are revealed. The successful reconstruction of experimental images further underscores the practical applicability and potential of this novel method for real-world fluid dynamics analysis.