AI-based macro model learning for high cycle fatigue assessment of welded joints in large-span steel structures

Welded spherical joints are critical components in large-span bridges and spatial steel structures prone to high cycle fatigue (HCF) loadings. Challenges exist for assessing fatigue life of this joints in small data from physical experiments and no detailed fatigue classification in design codes. This paper investigates the fatigue resistance of welded spherical joints and the real-time pre-damage evolution under non-stationary HCF loadings by proposing AI-based macro model learning from fine FE models and fatigue tests. Firstly, the structural features of welded spherical joints about stress concentration and weld defects were analyzed and quantified, so modified SN curve was built. Then, based on artificial intelligence (AI-based) learning, stress concentration factors (SCF) were obtained by an artificial neural network with particle swarm optimization and the modified fatigue resistance of welded spherical joints was derived by artificial bee colony algorithms. Finally, the macro linkage element was proposed as a user element subroutine (UEL), by which non-stationary fatigue pre-damage evolution based on the time-domain incremental analytical method was realized. The results not only indicated the damage value close to the ones derived by commonly used rain flow count methods, but also provided the pre-damage evolution curve for real-time monitoring.