EFFICIENT METHODS FOR TIME DOMAIN FATIGUE ANALYSIS OF OFFSHORE STRUCTURES

Abstract

Accurate long-term fatigue assessment is a challenging problem which needs to consider all possible sea states over the service life to obtain the mean fatigue damage. The problem becomes much more daunting when including wave directionality. This thesis develops two efficient methods for the long-term fatigue assessment, to account for the variation of wave direction, in addition to wave height and period. The first method enhances auto control variates (ACV) by incorporating artificial neural network, while the second method first applies subset simulation (SS), a probabilistic method traditionally applied for extreme response problem with small failure probability, to do mean estimation. Both are unbiased methods and the error can be quantified. The enhanced ACV has good performance on variance reduction with the speedup of two orders of magnitude while SS has great potential to be incorporated with other variance reduction methods to improve the efficiency of MCS further.