PSI - Issue 75

Carl-Fredrik Lind et al. / Procedia Structural Integrity 75 (2025) 519–529 Carl-Fredrik Lind et al./ Structural Integrity Procedia (2025)

521

3

forces and moments, making it highly suitable for automation. The method is theoretically sound, empirically validated, and offers a conservative fatigue life estimate, ideal for early design screening. The Master S-N curve method is based on the concept of structural stress, which combines membrane and bending stresses to yield a mesh insensitive metric for fatigue evaluation. A mesh-insensitive structural stress definition aligned with basic structural mechanics, capturing both membrane and bending stress components relevant to fatigue in welded joints. The method can be applied as a post-processing step to conventional solid or shell finite element models. Using this approach, a wide range of existing weld S-N data was reanalysed, showing that different weld classifications can be unified into a single master S-N curve. The curve’s slope is influenced by the balance between membrane and bending stresses, which can also be expressed through an equivalent stress intensity factor (Dong, 2001). The master S-N curve is shown in Fig. 2 illustrating the equivalent structural stress range versus the number of cycles to failure for 1,000 fatigue test data from varying welded joints (Dong et al., 2007).

Figure 2.Master S-N curve based on about 1000 fatigue tests (Dong et al., 2007).

2.3. DBSCAN clustering method DBSCAN is an unsupervised clustering technique that identifies clusters as areas of high data point density. To implement DBSCAN, it is required to specify two main parameters: epsilon (  ), which denotes the maximum distance at which two points are considered neighbours, and the minimum number of points (minPts) needed within this radius to form a dense region or cluster. Based on these two parameters, the data is classified into three categories: core points, which have at least minPts neighbours within  ; border points that fall within  of a core point but do not have enough neighbours on their own, and noise points or outliers, which are assigned to no cluster, meet neither criterion and are typically labelled as -1. A visual demonstration of the DBSCAN method is illustrated in Fig. 3.

Figure 3.Graphical visualisation of the DBSCAN algorithm.