Treffer: Integrated vision-based in-line surface defect detection system for realizing zero-defect manufacturing of tapered rollers.

Achieving Zero Defect Manufacturing (ZDM) during the mass production of critical components, such as tapered rollers, requires 100% in-line inspection to ensure the supply of defect-free products to consumers. Integrating vision systems with the production line can enable 100% in-line surface inspection of mass-manufactured components, facilitating the realization of ZDM. This work presents a vision-based system designed for integration with a tapered roller production line to detect surface defects. The system features indigenously designed hardware elements for in-line image acquisition, a hybrid algorithm combining image processing and deep learning to detect defective rollers, and seamless synchronization with the production line through an interactive user interface. The image acquisition hardware comprises a single-camera system and a horizontal belt conveyor for controlled translation and rotation of tapered rollers. The image processing framework encompasses algorithms for Region of Interest (ROI) detection, noise removal, and Convolutional Neural Network (CNN)-based surface defect detection, enabling robust prediction capabilities. The developed system is comprehensively evaluated through a performance analysis and Strengths, Weaknesses, Opportunities, and Threats (SWOT) assessment. The outcomes of the present study demonstrate that the vision-based system can be effectively integrated with the tapered roller manufacturing line to achieve reliability and efficacy during in-line inspections. The study also demonstrated that vision-based inspection systems can be implemented to achieve ZDM, enabling manufacturing industries to improve product quality and enhance global competitiveness. • Vision-based system ensuring 100% in-line inspection and facilitating ZDM. • Indigenous hardware arrangement capturing high-quality images of rollers in motion. • Hybrid algorithm to detect defects combining image processing and EfficientNet-b0. [ABSTRACT FROM AUTHOR]