AUTOMATION OF QUALITY CONTROL FOR USED PARTS BASED ON 2D IMAGES AND NON-VISUAL INFORMATION: AN APPROACH TO EXTENDING THE PRODUCT LIFE CYCLE

This paper presents a methodology for automating the quality control of used parts after disassembly, focusing on determining their reusability. The proposed approach uses 2D images of the parts, supplemented with non-visual information, in particular dimensions and service history, to identify worn or damaged components accurately. The methodology consists of four steps: raw data acquisition, hole segmentation, scale classification and optimization, and diameter tolerance classification and validation. Using the Mask R-CNN model with an attention mechanism (SEBlock) in combination with the Random Forest classifier achieved high accuracy in detecting defects, particularly hole diameter mismatches with established tolerances. The study's results demonstrated an average classification accuracy of 79%, with maximum precisions, completeness, and F1 estimation in some instances. The system showed high efficiency, avoiding false negatives and minimizing false positives. The proposed methodology is cost-effective as it eliminates the need for expensive 3D scanning equipment, making it easy to integrate into production lines and repair stations. Integrating a multi-level 3D concept allows the geometric parameters and microstructural characteristics of parts to be considered at different levels, increasing the analysis's accuracy. The system has limitations: it can only analyze one part type at a time and cannot detect features not visible in 2D images. At the same time, it opens perspectives for further research to improve surface analysis methods and the use of multi-angle stereo vision. Implementing the proposed system contributes to more efficient manufacturing processes, reduces the cost of purchasing new components, and supports environmental sustainability by extending the life cycle of parts. Further research will focus on adapting the methodology for different part types and implementing polygonal stereo-vision approaches to improve the accuracy and reliability of quality control.