Dual Deep Learning-Based Vision Inspection System for Automated Detection of Tube Defects in Medical IV Fluid Bags

다국어 초록 (Multilingual Abstract)

Intravenous (IV) fluid bags are widely used by medical devices, and defects such as foreign particles or improper tube bonding during manufacturing can cause leakage or contamination, threatening patient safety. Manual visual inspection of IV bag tubes is time-consuming, costly, and prone to human error. This paper proposes an AI-based automated inspection system that integrates dual-view imaging, geometric correction, and deep learning–based defect detection to identify multiple tube defects. The proposed algorithm classifies four major defect types: black particles inside the tube, curled tubes, bad welding, and missing tamper-evident seals. To handle differences in object scale and morphological variation, two complementary models are employed: a YOLOv8-based model for large structural components with low shape diversity and a Faster R-CNN–based model for small and highly variable defects. The system uses downscaled, geometrically corrected images for large-object detection and high-resolution region-of-interest inputs for small-object detection. Experimental validation with 2,803 training images and 300 test images achieved 100% accuracy for the defects with bad welding and missing tamper-evident seal and 98–99% accuracy for black particles and curled tubes. The results demonstrate that the proposed method enables reliable, real-time visual inspection of IV bag tubes, offering strong potential for practical deployment in automated medical manufacturing environments.