PSI - Issue 75

Kris Hectors et al. / Procedia Structural Integrity 75 (2025) 102–110 Hectors et al. / Structural Integrity Procedia (2025)

104

3

a)

b)

c)

Fig. 3: a) A 40x zoom photograph of the V-notch, with measurement points defining the notch opening width (1.322 mm). The light blue line across the center defines the cross-section that is shown in in Fig 2c. b) Three-dimensional surface topography obtained via profilometry, with height variations color-mapped. c) The extracted cross-sectional profile, indicating a notch depth of 1.512 mm, an opening displacement of 0.368 mm, and a notch opening angle of 135.036°, as determined from the intersection of the notch faces.

Fig. 4: Illustration of the V-notch profile and local geometry extraction from the 3D scan of the specimen.

3. Training database 3.1. Database structure

As a prerequisite to the model training, train and test datasets have to be developed. These must pair geometric data with corresponding accurate values. To achieve this, a script was developed to convert the profilometer measurement into finite element models. The script was developed in the Python-based Abaqus v2023 scripting interface. Given the cylindrical shape of the specimens, an assumption of axial symmetry was adopted. Assuming axisymmetric specimens significantly simplifies the conversion of point cloud data to a finite element model. This is because it enables the model to be constructed using only the two-dimensional cross-sectional profile data of a specimen, thereby eliminating the need to incorporate the entire point cloud dataset. To validate the assumption of axial symmetry, 36 high-resolution profile scans were conducted around the circumference of a single specimen. These scans were performed at 40 × magnification with a 3.70 µm spatial resolution. The results showed close alignment of the profile scans near the notch root, which is also the main region of interest. Areas with greater variance were also apparent, however further away from the notch root.