An Efficient Approach to Quality Diagnosis of PCB Connectors Using Image-Derived Geometric Features

In the era of Industry 4.0, quality research using product images in the manufacturing industry has been on the rise. However, large volumes of image data pose significant challenges in terms of data transmission, storage, management, computation, and analysis. To address these issues, this study proposes a method for extracting key quality information from product images, storing it as text files, extracting geometric features based on this information, and generating a quality classification model. Additionally, it also introduces a method for reconstructing images when necessary for visual verification or further Artificial Intelligence (AI) training. This methodology was applied to Printed Circuit Board (PCB) connectors, and the results demonstrated that the quality classification model based on text files containing extracted quality information outperformed image-based models in terms of storage efficiency and learning performance. This approach enables efficient management and analysis of large-scale image data while also enabling flexible utilization and analysis of image files. However, the proposed methodology is optimized for specific geometric features, and further research is required for broader application to diverse defect types and product characteristics.