An algorithm for statistical evaluation of weld toe geometries using laser triangulation

Abstract

Commonly, to evaluate the influence of the local weld geometry in fatigue test, small-scale specimens are used, assuming those represent a longer weld adequately. In this study, a comparison between short specimens and a long weld is performed. A method is developed for the statistical evaluation of weld toe radii and angles, stress concentration factors and weld quality classes. The results show a strong sampling rate dependence and lower ISO 5817:2014 weld quality results for higher sampling rates. Comparable results between short specimens and a long weld can be achieved using modal values of the parameters assuming a lognormal distribution.

Comparison of machine learning and stress concentration factors‐based fatigue failure prediction in small‐scale butt‐welded joints

“…With the development of weld geometry measurement systems, the characterization of stress concentrations at welded joints has seen significant progress; compare earlier research 6,10–14 . To determine SCFs at weld transitions, weld geometries are typically idealized as two‐dimensional geometries, which are described by geometrical properties such as notch opening angle, notch radius, and weld height (see Figure 1).…”

“…This median geometry data were also added to the butt joint dataset. The median was taken instead of the mean as it is less sensitive to outliers 9,13 …”

“…The geometrical variations and misalignment in the investigation (see Figure 1) were determined prior to testing for all specimens. For the measurement of axial and angular misalignment, a test setup with dial gauges was used; see Braun, 50 and the local weld geometry was obtained by laser triangulation and postprocessing by means of the curvature method; see earlier research 13,14 . Figure 5 presents the medians of the measured geometrical and SCF parameters that are used as input for the ML models.…”

“…With the development of weld geometry measurement systems, the characterization of stress concentrations at welded joints has seen significant progress; compare earlier research. 6,[10][11][12][13][14] To determine SCFs at weld transitions, weld geometries are typically idealized as twodimensional geometries, which are described by geometrical properties such as notch opening angle, notch radius, and weld height (see Figure 1). Based on such simplified descriptions of the weld geometry, it is possible to determine SCFs, for example, by means of parametric formulas, for example, Anthes et al 15 An alternative is the direct calculation by FE simulations.…”

“…The median was taken instead of the mean as it is less sensitive to outliers. 9,13 Lastly, before training the models, invalid tests were removed from the data; for example, tests that were aborted without failure (run-out tests). Categorical features were replaced with numerical features with ordinal encoding instead of one-hot (also called one-of-K) encoding since this type of encoding was applied successfully in a previous study.…”