Reducing uncertainty in time domain fatigue analysis of offshore structures using control variates

This study is concerned with time domain fatigue analysis of offshore structures subjected to random waves. The fatigue damage calculated from a single realization of the stress time history is random, thus the mean damage is typically estimated via Monte Carlo simulation (MCS), by averaging over multiple realizations. This approach is time-consuming, because each realization involves a time domain dynamic analysis, and MCS has a slow convergence rate. Variance reduction techniques can improve the efficiency of MCS, but successful implementation necessitate prior information on the system behavior, which is difficult to acquire for this high-dimensional problem. Herein, a method is developed for reducing the variance of the MCS estimator of the damage, based on a new technique known as auto control variates. The control function is constructed via artificial neural network trained from existing MCS data, thus avoiding the need for prior information or additional simula-tions. The proposed method has several advantages; it is unbiased, and an error estimate is available. Besides, variance reduction is implemented at the post-processing stage; allowing multiple stress locations to be evaluated from the same dynamic simulation results. The case studies include a nonlinear single-degree-of-freedom system under different scenarios, and a full nonlinear model of a floating system. The proposed method enhances the MCS efficiency for all cases, with speedups ranging from one to two orders of magnitude. (c) 2020 Elsevier Ltd. All rights reserved.

Automated summary

In this study, a method for reducing the variance of Monte Carlo simulation estimator of fatigue damage is proposed, utilizing auto control variates technique, which successfully enhances efficiency. The method is unbiased, provides an error estimate, and the variance reduction is implemented at the post-processing stage.