Probabilistic fatigue life of welded plate joints under uncertainty in Arctic areas

This paper develops a new practical framework of high-cycle fatigue (HCF) probability analysis of local welded plate joints considering the uncertainty in Arctic areas. The framework includes 1) a meta-model of the local energy concentration factor relying on the genetic algorithm and neural network (GA-NN) approach, 2) the identification of uncertainties in the fatigue assessment procedure, and 3) the probability analysis relying on the Monte Carlo simulation with Latin hypercube sampling method. The GA-NN determines the interactive relationship among the geometrical properties for a good estimation of the energy-based indicator concentration factor. The probability analysis considers the uncertainty caused by the meat-model, the local welding geometry, the Young modulus under different temperatures, and the combination of the ice and wave loadings. To determine the uncertainties involved in probability analysis, this study conducts a series of experimental measurements and high-cycle fatigue tests of cruciform welded plate joints. Furthermore, this study discusses the effect of the ice loading, maximum wave stress, and main plate thickness of welded plate joints on the expected fatigue life. The additional ice loading on the welded plate joints can almost cut the expected fatigue life in 33% in comparison with that only under wave loadings, implying the necessity of considering ice loading in Arctic areas. This study also proposes a probability S-N curve with the cumulative probability of 50% based on the maximum local energy-based indicator to estimate the expected fatigue life. (C) 2020 Published by Elsevier Ltd.