Neural Operators Predict Soft Swimmer Locomotion Hydrodynamics

Simulating the complex interaction between deforming soft swimmers and their surrounding fluid flow is computationally intensive, limiting its use in design and control applications. This research introduces neural operator surrogates as a solution to this challenge, aiming to drastically reduce the cost of repeated evaluations.The developed models are trained on high-fidelity simulation data, specifically regular-grid fields from adaptive fluid-structure simulations. These surrogates are conditioned on factors like swimmer geometry and Reynolds number, enabling them to predict hydrodynamic fields such as velocity, vorticity, and pressure over time for both planar and volumetric eel swimmers.Results show that the planar model achieves a low global relative L^2 error for velocity, vorticity, and pressure. The volumetric model also performs well for velocity and vorticity, though pressure prediction accuracy is identified as a key area for future development. This work demonstrates the potential of neural surrogates for moving-boundary fluid flows, offering a path to more efficient engineering design.