TECHNOLOGIES
Automatic quality control of blanked parts based on segmentation of shear and fracture zones using a convolutional neural network
A method for automatic quality control of blanked parts based on the ratio of the plastic deformation (shear) zone to the brittle fracture (rupture) zone on the cut surface is presented. Unlike known approaches that require strictly controlled shooting conditions, the proposed convolutional AI model (R-CNN) demonstrates robustness to a wide range of illumination: variations in brightness and light incidence angle reduce segmentation accuracy by no more than 3%. This is achieved through the use of local texture features that are invariant to brightness scale. The system takes into account the material grade — reference ratios of the zones are defined for each steel grade. Experiments were carried out on samples of three grades (AISI 1010, S235JR, AISI 1066). The classification accuracy of acceptable and defective parts was 98% relative to expert assessment.