Session: 02-02-03 Probabilistic Models of Forces and Motions

Paper Number: 122594

122594 - Bayesian Updating and Gaussian Process Regression for Estimation of Viv Load Parameters 

The purpose of the present paper is to demonstrate the application of Bayesian updating and Gaussian Process Regression to VIV cross-flow response. The Bayesian updating approach is based on first formulating of a prior statistical model for the quantities of interest (i.e. the parameters entering the VIV load model in the time domain response analysis). Subsequently this statistical model is modified based on a set of available observations. The statistical properties of the observations conditional on a given set of values for the load parameterst are represented by the corresponding likelihood function. By combining the prior model with the likelihood function, the resulting posterior statistical model is obtained.

The following parameter studies are performed: (i)Different types of fitting of continuous “response functions” to represent the discrete-valued data base (ii) The number measurement samples (iii)The noise level associated with measurement of the response amplitude/diameter-ratio and the response oscillation frequency. (iv)Application of different types of likelihood functions.Connections is also made between the search for the maximum value of the posterior density function and the so-called Bayesian optimization scheme which is based on a sequence of Gaussian Process Regression models. Generalization of the algorithm to cases with both cross-flow and in-line VIV response is also adressed.

 

Keywords: VIV-response, Bayesian updating, Load Parameter.s

Presenting Author: Bernt Leira NTNU

Presenting Author Biography: Master Civil Engineering, 1978
PhD Marine Technology, 1987
Professor Marine Structures, 1979

Authors:

Bernt Leira NTNU
Martin Andersen IMT, NTNU
Jie Wu SINTEF Ocean
Svein Savik NTNU, Dept. Marine Technology