Physics-Informed Machine Learning for Light Field-Based Fluid–Structure Interaction Diagnostics

We present a new experimental methodology for fluid-structure interaction (FSI) diagnostics that utilizes a single light field camera to simultaneously capture time-resolved, three-dimensional, three-component (3D3C) velocity fields and structural deformations of a flexible body. Conventional experimental FSI diagnostics often analyze fluid and structural phases separately, which limits the accuracy of measurements at their interface. To overcome this limitation, we introduce a framework that leverages physics-informed neural networks (PINNs) to improve velocity field reconstruction, using structural position data as constraints. A plenoptic imaging system is used to independently capture and reconstruct 3D particle tracks and structural motion simultaneously. The PINN framework integrates flow measurements with governing physical equations and structural position data constraints, enabling enhanced flow-field diagnostics at the fluid–structure interface. We validate the approach using simulated FSI data and demonstrate it experimentally in a heart valve model. Preliminary results from synthetic evaluations indicate that incorporating boundary information with PINNs enhances flow reconstruction accuracy by up to ~18% (for particle per microlens of 0.02) at the fluid–structure interface. This integrated, single-camera diagnostic technique presents a novel solution for FSI analysis in complex and constrained environments.