Nondeterministic High-Cycle Fatigue Macromodel Updating and Failure Probability Analysis of Welded Joints of Long-Span Structures

Abstract

Unlike previous models that do not consider uncertainties, this paper proposes a new nondeterministic method to estimate the high-cycle fatigue (HCF) resistance of welded hollow spherical joints (WHSJs) in long-span structures, including bridges, gymnasiums, and factories. This macro damage mechanics model with failure probability is introduced by accounting for the uncertainties of stress concentration and weld defects. Then artificial neural networks (ANNs) for three kinds of load type are built as a substitute for a finite-element (FE) model to obtain the concentrated stress more efficiently. Subsequently, the probability model of the defect factor is identified as a Gaussian distribution, while the stress concentration factor (SCF) has a Gaussian distribution and three-parameter t distribution. Also, a series of fatigue test on WHSJs are used to validate the proposed model, yielding a reasonable fatigue life prediction. Finally, fatigue failure probability analysis, which includes a joint probability density function (PDF), is conducted using the new nondeterministic method, which could provide a reference for fatigue design and damage quantification of WHSJs in long-span structures. Meanwhile, two Monte Carlo simulations corresponding to both possible distributions of concentrated stress were run to verify the accuracy of the HCF failure probability model of WHSJs. The results guarantee the feasibility of the proposed probability model applied in HCF fatigue design for weld joints of long-span structures.