Defect Detection in Manufacturing: Leveraging YOLOv8 for RealTime Quality Control on Production Lines

Abstract

With the rapid advancements in global manufacturing and the increasing shift towards intelligent systems, production line efficiency and product quality have become critical competitive factors. Traditional manual defect detection methods suffer from inefficiency, subjectivity, and high miss rates, which hamper the stability of product quality and production efficiency. This study addresses these challenges by leveraging advanced machine vision and deep learning technologies, specifically focusing on the latest YOLOv8 algorithm, to design and implement an intelligent defect detection system for production line parts. The YOLOv8 algorithm, with its enhanced network structure, loss function, and training strategies, offers superior detection speed and accuracy, making it ideal for real-time industrial applications. The research involved constructing a comprehensive dataset of part defects, followed by meticulous model training and optimization. The experimental results demonstrate that the improved YOLOv8 model significantly outperforms traditional methods in terms of accuracy and efficiency, effectively mitigating the issues of subjectivity and inconsistency. Furthermore, the study explored the deployment of the model using the lightweight ShuffleNetV2 backbone, enhancing its stability and applicability in complex production environments. The findings provide robust support for the intelligent transformation of manufacturing processes and establish a solid foundation for future industrial applications.