Benchmarking of Neural Operator for rigid-body fluid-structure interaction

Simulating rigid-body fluid–structure interaction (FSI) with both accuracy and efficiency remains a persistent challenge in computational science and engineering. While neural operators have recently shown great promise as alternatives to traditional numerical solvers in computational fluid dynamics, their potential in FSI problems—especially those involving freely moving rigid bodies—has not been thoroughly investigated. To address this gap, we present two benchmark datasets tailored for FSI scenarios where rigid-body motion acts as a control signal shaping the surrounding fluid flow. We further conduct a systematic evaluation of state-of-the-art neural operator architectures and explore three strategies for coupling structural motion with fluid initial conditions.