Intelligent system based on Yolo v11 for detecting defects in bottle filling at a bottling plant

Abstract

This study addressed the challenges of efficiency and accuracy in quality inspection at bottling plants, where manual methods lead to errors, high costs, and risks to brand reputation. An automated system based on artificial vision was implemented and evaluated to identify filling defects in water bottles at the Cielo brand plant in Iquitos during 2025. The objective was to determine the relationship between the implementation of the system and the improvement in defect detection. A quantitative and experimental approach was used, collecting and preprocessing a set of image data to train and validate a model from the YOLO family in Python. The model's performance was evaluated using metrics such as Mean Average Precision (mAP), accuracy, and recall. The results demonstrated high effectiveness in identifying bottles in good condition (87% recall) and moderate ability to detect defects (52% recall), reaching an inference speed of 30 FPS, suitable for real-time operation. It is concluded that the system is technically viable for improving the speed and accuracy of quality control, although further training is required to reduce the false negative rate in defective bottles.