Issue 78

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

deviation in one geometric parameter is expected and falls within the model’s predictive uncertainty. Furthermore, the associated change in the predicted mechanical response is minor, suggesting that both geometries belong to the same near optimal region of the design space rather than representing fundamentally different design solutions.

Figure 7: 3D representation of the machine learning highlighting the optimal configuration (red big circle)

E XPERIMENTAL V ALIDATION

Materials and specimen manufacturing ne selected configuration that was considered is the one proposed in the standard ASTM B831 -14 with distance A= 8.35 mm, radius B =1 mm, and angle C =45 0 . Based on the outcomes of the numerical simulations and machine-learning analysis the second geometry was considered the optimal provided by numerical simulation/machine learning analysis having distance A = 6 mm, radius B = 0.8 mm and angle C = 60°. The specimens were fabricated via fused deposition modeling (FDM) using a Prusa i3 3D printer. White polyethylene terephthalate glycol-modified (PETG) filament was employed as the feedstock material. Printing was performed using a nozzle temperature of 230 °C and a build-plate temperature of 80 °C. The infill raster was oriented at ±45° in alternating layers. All specimens were printed with their primary geometry oriented in the XOY plane. O

Figure 8: Specimens configurations A-standard, and B-new design.

125