Effects of varying initial conditions of ship encountering wave groups in computing extreme motion statistics

TL;DR

This paper evaluates methods for incorporating initial ship conditions into wave group-based motion statistics calculations, demonstrating the superiority of the natural initial condition approach and accelerating it with Gaussian process regression.

Contribution

It introduces and benchmarks three methods for initial condition incorporation, highlighting the effectiveness of the natural initial condition method and its acceleration via Gaussian process regression.

Findings

Natural initial condition method outperforms others in accuracy.

Sequential sampling with Gaussian process regression accelerates computations.

Method robustness across different wave spectra and ship dynamics.

Abstract

In computing ship motion statistics (e.g., exceeding probability) in an irregular wave field, it is a common practice to represent the irregular waves by a large number of wave groups and compute the motion statistics from the distribution of these wave groups. While this procedure significantly reduces the computational cost, the uncertainties introduced in this reduced-order computation have not been quantified. In general, the representation of a continuous wave field by separated groups loses information about wave phases, frequency modulation and initial conditions of ship when encountering the wave groups, among which the last one is arguably the most influential factor for the ship motion statistics. In this paper, we test three methods to incorporate the ship initial conditions into the computation of roll motion exceeding probability, namely the methods of natural initial condition, pre-computed initial condition and constant initial condition. For different input wave spectra and ship motion dynamics (in terms of the parametric excitation level modeled by a nonlinear roll equation), the performances of the three methods are carefully benchmarked and the superiority of the natural initial condition method is demonstrated. We finally show that the computation using the natural initial condition method can be greatly accelerated through a sequential sampling algorithm making use of the variational heteroscedastic Gaussian process regression.