Estimation of VIV-parameters based on Response Measurements and Bayesian Machine Learning Algorithms

Abstract

Analysis of structural response levels due to hydro-elastic vortex-induced vibrations (VIV) involves the specification of several parameters both associated with the fluid flow and the structural properties. To the maximum possible extent, the applied values of these parameters should be based on relevant results from experiments and full-scale measurements. This can be achieved by establishing a probabilistic framework which allows continuous learning in relation to the numerical models and associated parameters that are to be applied for the analysis. In this paper, a Bayesian optimization framework for estimating parameters in the VIV time-domain model (VIVANA-TD) is presented. As a case scenario, a simplified VIV model was studied for the purpose of illustration. A simple numerical model of a cylinder with 1 degree of freedom (DOF) was applied in predicting of the time-varying dynamic response. This prediction model is based on the hybrid-analytical concept, which relies on a combination of the time domain model and measured response features. In addition, two methods for estimating the parameter uncertainties are introduced.