Modeling fluid forces in 2-DOF forced motion experiments using neural network and analytical nonlinear approaches

Abstract

This study explores the prediction of fluid forces on an oscillating cylinder in two degrees of freedom under free-stream flow. It combines experimental data, neural network models, and analytical force mechanisms to analyze hydrodynamic responses across varying motion amplitudes in both in-line and cross-flow directions. Contour plots of the average power coefficient identify distinct regions of free and forced vibration. Neural network models, trained on experimental data, effectively predict forces in free vibration regimes, although they capture actuator-induced noise. To address this, a novel quadratic drag force model is introduced, offering a more interpretable and noise-resistant approach for predicting nonlinear drag forces. Both models show significant potential in aligning with experimental data. This work highlights the importance of integrating data-driven models with physical understanding, improving predictive accuracy and system stability for fluid forces, while addressing challenges like actuator noise and system instability.