Vibration Analysis with AI: Physics-Informed Neural Network Approach for Vortex-Induced Vibration

Abstract

Vortex-induced vibration (VIV) of structures exposed to fluid flow is a complex phenomenon that can lead to fatigue damage and failure. Physics-informed neural networks (PINNs) are a promising approach to model VIV by incorporating both data and physical laws. This study develops a PINN framework to analyze VIV of a cylinder in cross-flow. The model integrates the fluid dynamics equations, cylinder equations of motion, and vibration data into a neural network. Nonlinearities and fluid forces are learned by the network through minimizing loss functions representing physics and data. The trained PINN model accurately predicts displacement and stress for varying flow speeds. A parametric study explores the effects of mass, damping, and flow parameters on VIV amplitude and frequency. The PINN model provides insights into energy transfer mechanisms and key parameters governing VIV. The integration of data and physics-based losses in PINNs is demonstrated as an effective approach for analysis and knowledge discovery in fluid-structure interaction problems.