PSI - Issue 75

Kris Hectors et al. / Procedia Structural Integrity 75 (2025) 102–110 Hectors et al. / Structural Integrity Procedia (2025) and the notch depth (Radaj et al., 2006). Fig. 1 illustrates the geometric features for both a V-notch and a U-notch. The severity of a notch can be quantified by the elastic stress concentration factor which quantifies the discrepancy between the elastic notch stress and the far-field or nominal stress (Sonsino et al., 2012). Many parametric analytical equations have been developed to determine -values of notches with a given notch opening angle, radius and depth (Noda & Takase, 2006). However, the real, as-machined, geometry can vary (significantly) from the idealized one. As this can impact the severity of the stress concentration, fatigue assessment based on the real notch geometry becomes of interest. Theoretically, models based on the as-machined notch geometry derived from 3D scanning should yield more accurate stress concentration factor predictions than those based on idealized geometries. However, leveraging this detailed data with Finite Element Analysis (FEA) presents significant challenges. Generating FEA models from complex 3D scan data requires considerate expertise. Furthermore, accurately capturing the manufacturing features that influence values necessitates high mesh densities, which leads to prohibitive computational costs (Niederwanger et al., 2020). These challenges can be addressed with machine learning (ML) approaches. For a comprehensive overview reference is made to (Nath et al., 2024). This work investigates the potential of ML models to overcome these limitations. The objective is to rapidly predict values for V-notched specimens directly from their 3D scanned notch profiles. This approach bypasses the demanding model development and expensive simulations required by traditional FEA. 103 2 Fig. 1: Illustration of the notch opening angle , notch radius and notch depth for a sharp and blunt V-notch. 2. Dataset of as-machined specimen geometries To develop a machine learning model capable of predicting stress concentration factors for as-machined specimens, a comprehensive database detailing notch geometry was required. We manufactured and scanned 177 cylindrical notched bar steel specimens to create this database. The specimens studied in this work are rotating bending fatigue specimens that conform to the ISO 12107:2012 standard. The nominal dimensions of the specimens are shown in Fig. 2. Using a 3D measurement profilometer (Keyence VR-5200), we conducted wide-area scans at 12x magnification, achieving a spatial resolution of 23.50 µm. Fig. 3 displays an example of a scanned notch. From these scans, we extracted the relevant notch parameters (depth, opening angle, and radius) as illustrated in Fig. 3c. Aside from this, the detailed notch profile was also captured and stored as shown in Fig. 4. All the data that was captured in this work has been made available in an open-access database (Hectors & De Waele, 2023). Results of the rotating bending fatigue experiments (not considered in this study) are published in the same database (Hectors et al., 2023).

Fig. 2: Technical drawing of the notched specimens considered in this study. Geometry based on the ISO 12107:2012 standard.