Fatigue evaluation based on the effective notch stress approach requires stress concentration factors for idealized notch geometries. In this paper, stress concentration factors are calculated numerically for different idealizations of the weld geometry. The joints covered here are one-sided Y-butt joints and two-sided DY-butt joints, each with partial and full penetration. The variable parameters of the finite element models are the weld flank angle, the notch radius, the sheet thickness, the ratio of weld seam width to sheet thickness, and the ratio of height of non-fused root face to sheet thickness. Existing estimation formulae for stress concentration factors will be compared with new methods for stress concentration factor estimation: (a) polynomial regression using mixed terms and (b) artificial neural networks. These two methods show similar or superior quality compared with the existing estimations which is expressed through lower estimation errors with respect to the numerically calculated stress concentration factors. The database for the regression analysis using methods (a) and (b) consists of a total of 11,871 design alternatives obtained by finite element analysis. In addition to the improved quality of prognosis, the range of allowable geometrical parameters is significantly increased compared with the existing formulae. The methods (a) and (b) provide a possibility to a fast estimation of stress concentration factors of sufficient quality for a large range of geometrical weld seam parameters. The presented formulae could also be part of a programmed solution.

基于有效缺口应力方法的疲劳评估需要应力集中系数来实现理想的缺口几何形状。在本文中，针对焊接几何形状的不同理想化，通过数值计算了应力集中系数。此处涵盖的接缝为一侧的Y型对接接头和两侧的DY型对接接头，每个接头均具有部分和完全穿透的能力。有限元模型的可变参数是焊缝侧面角，切口半径，薄板厚度，焊缝宽度与薄板厚度的比率以及非熔融根面高度与薄板厚度的比率。现有的应力集中系数估算公式将与应力集中系数估算的新方法进行比较：（a）使用混合项的多项式回归和（b）人工神经网络。与现有估计相比，这两种方法显示出相似或更高的质量，而现有估计是通过相对于数值计算得出的应力集中因子的较低估计误差来表示的。使用方法（a）和（b）进行回归分析的数据库由通过有限元分析获得的总共11,871个设计方案组成。除了改善预后质量外，与现有公式相比，允许的几何参数范围也显着增加。方法（a）和（b）为大范围的几何焊缝参数快速评估足够质量的应力集中因子提供了可能性。提出的公式也可以是编程解决方案的一部分。