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 S-N 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.

中文翻译：

---

基于人工智能的大跨钢结构焊接接头高周疲劳评估宏观模型学习

焊接球形接头是大跨桥梁和空间钢结构中容易遭受高周疲劳（HCF）载荷的关键部件。在物理实验的小数据中评估该接头的疲劳寿命存在挑战，并且设计规范中没有详细的疲劳分类。本文通过提出基于人工智能的宏观模型学习精细有限元模型和疲劳试验，研究了焊接球形接头的抗疲劳性和非平稳 HCF 载荷下的实时预损伤演化。首先，对焊接球形接头的应力集中和焊缝缺陷的结构特征进行分析和量化，建立修正的SN曲线。然后，基于人工智能（AI-based）学习，通过粒子群优化的人工神经网络获得应力集中因子（SCF），并通过人工蜂群算法导出焊接球形接头的修正疲劳强度。最后，提出宏联动单元作为用户单元子程序（UEL），实现基于时域增量分析方法的非平稳疲劳预损伤演化。结果不仅表明损伤值接近常用的雨流计数方法得出的损伤值，而且还提供了损伤前的演化曲线以供实时监测。