Issue 78

C. Popa et alii, Fracture and Structural Integrity, 78 (2026) 118-128; DOI: 10.3221/IGF-ESIS.78.07

Shear tests Shear tests were performed on a Zwick universal testing system with a 5 kN load cell. All tests were conducted in displacement control at a crosshead speed of 5 mm/min. The specimens were securely gripped, and a tensile load was applied along the loading axis until fracture occurred. Fig. 8 presents the failed specimens after the shear test. The regions of interest are highlighted to facilitate comparison between the two configurations. The typical force–displacement responses are presented in Fig. 9 to enable a direct comparison between configurations. The curve corresponding to the standard specimens (orange) exhibits an average force of 947 N, whereas the curve associated with the optimized configuration (green) shows a lower average force of 765 N. The higher load-carrying capacity of the standard specimens can be attributed to their larger effective cross-sectional area within the critical geometric region.

Standard New design

Figure 9: The relevant Force-Displacement curves for standard specimen and the new design.

The average shear strength results (for five tested specimens of each geometry) are presented in Fig. 10. The measured values are comparable for both configurations; although the standard specimen exhibited a higher force, the resulting shear strengths differ only slightly. The standard specimen reached an average shear strength of 28.24 MPa (standard deviation 1.20 MPa), whereas the optimized configuration achieved 27.38 MPa (standard deviation 1.21 MPa). This corresponds to an absolute difference of 0.86 MPa and a relative deviation of ~3.0%. However, the S parameter is 3.431 for the new design, comparatively with 2.268 for the standard design indicating a more pronounced shear stress field with approximately 50%.

Standard New design

Figure 10: Shear Strength plots.

C ONCLUSIONS

or an optimal design of the shear specimen, 145 simulations were performed in order to train a machine learning routine based on linear regression. Three parameters were considered distance between notches A, notch radius B and notch angle C. The model was trained with 116 data sets and validated with 29 data sets. The linear regression model was used to predict values for 85 new geometries. Regarding the choice of Multiple Linear Regression, we acknowledge that the stress response of geometrical structures is inherently nonlinear. However, the present study investigates a relatively narrow design space around a reference geometry, where the relationships between the selected geometric descriptors and the resulting stress values exhibit an approximately linear behavior. Under these conditions, Multiple Linear Regression provides an interpretable model that allows the contribution of each geometric parameter to be quantified while maintaining computational efficiency. F

126