期刊
MARINE STRUCTURES
卷 78, 期 -, 页码 -出版社
ELSEVIER SCI LTD
DOI: 10.1016/j.marstruc.2021.103002
关键词
Fatigue analysis; Control variates; Variance reduction; Artificial neural network; Monte Carlo simulation
资金
- NUS Research Scholarship
Accurate fatigue assessment is crucial in riser design and must consider various sea states. An efficient method based on time domain simulation, considering wave directionality, has been proposed to reduce computational cost.
Accurate fatigue assessment is a challenging and crucial aspect of riser design. The prediction of the long-term fatigue damage must account for numerous sea states of different wave heights, periods, and directions. Each sea state entails a dynamic analysis, often performed in the time domain owing to the significant nonlinearities. Because of the short-term uncertainties from irregular waves, the simulation duration must be sufficiently long for results to converge. To alleviate the hefty computational cost of long-term fatigue analysis, researchers have proposed efficient methods, but these are not without drawbacks; in particular, wave directionality is commonly neglected. This paper presents an efficient method for long-term fatigue analysis based on time domain simulation, considering wave directionality among other things. The proposed method is based on an enhanced version of control variates to reduce the variance in Monte Carlo simulations (MCS). The control function is constructed by training artificial neural network (ANN) models using existing MCS data. Here, a customized scheme is developed to allow for the situation that the training data and ANN prediction cases have different wave directions. The proposed method is unbiased and provides an error estimate. Simulations are performed on a floating system, and the proposed method is found to improve the efficiency of MCS significantly. Different scenarios such as fixed and random wave directions are compared, confirming that wave directionality is critical and should be included in a long-term fatigue assessment.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据