Surrogate Modeling of Fluidic Oscillators

This work presents data-driven surrogate models designed to accelerate the numerical simulation of fluidic oscillators (FOs). High-fidelity two-dimensional OpenFOAM simulations are first performed across a range of operating conditions to resolve the intrinsic jet-switching dynamics and multiscale flow structures that drive FO behavior. These simulations are then decomposed into training datasets for two classes of surrogate models that replace the full FO geometry with an equivalent, time-varying Dirichlet boundary condition. As a baseline approach, a Fourier-series reduced-order model is constructed to reproduce the dominant oscillation frequencies and jet-switching characteristics. To achieve higher fidelity and accommodate parametric variations in driving conditions, a Fourier Neural Operator (FNO) architecture is developed and trained to map inlet pressure histories and near-field flow information to boundary-condition profiles. The results demonstrate that the FNO surrogate captures nonlinear FO response with high accuracy while reducing computational costs significantly, enabling efficient integration of FO dynamics into larger flow configurations.