Predictive capability assessment for physics-guided learning of vortex-induced vibrations

Abstract

We seek here a computationally parsimonious and credible means to simulate the complex phenomena of vortex-induced vibrations (ViV), as one tool to assist in mitigating risk associated with ViV-induced instabilities that can cause non-negligible structural acoustic response. To address current limitations in data-driven modeling, for which credibility assessment proves challenging, or physics-based simulation (i.e., constrained by governing partial differential equations (PDEs)), which often includes prohibitive computational expense, we explore recent state-of-the-art approaches to optimally combine these engineering disciplines via a physics-guided machine learning framework. One can expect that intersecting data-driven modeling with physics-guided simulation offers one means to both maximize the credibility of machine learning based approaches and minimize the computational expense of physics-based modeling approaches.