Neural inference of fluid-structure interactions from sparse off-body measurements

We report a novel physics-informed neural framework for reconstructing unsteady fluid-structure interactions (FSI) from sparse, single-phase observations of the flow. Our approach combines modal surface models with coordinate neural representations of the fluid and solid dynamics, constrained by the fluid's governing equations and an interface condition. Using only off-body Lagrangian particle tracks and a moving-wall boundary condition, the method infers both flow fields and structural motion. It does not require a constitutive model for the solid nor measurements of the surface position, although including these can improve performance. Reconstructions are demonstrated using two canonical FSI benchmarks: vortex-induced oscillations of a 2D flapping plate and pulse-wave propagation in a 3D flexible pipe. In both cases, the framework achieves accurate reconstructions of flow states and structure deformations despite data sparsity near the moving interface. A key result is that the reconstructions remain robust even as additional deformation modes are included beyond those needed to resolve the structure, eliminating the need for truncation-based regularization. This represents a novel application of physics-informed neural networks for learning coupled multiphase dynamics from single-phase observations. The method enables quantitative FSI analysis in experiments where flow measurements are sparse and structure measurements are asynchronous or altogether unavailable.