Construction of a Deep-learning-based Reusability Assessment System for Large Vehicle Tires

AI advancements in anomaly detection enable inspections that are more efficient and precise than human operators, enhancing recycling efficiency and supporting Sustainable Development Goals.Tire manufacturers currently rely on manual visual inspection to assess used tires for damage and manufacturing dates; however, this process suffers from skilled labor shortages and excessive time requirements.In this research, we aim to automate tire damage detection and manufacturing date recognition using deep learning on videos from standard cameras, thereby improving operational efficiency.Four experiments were conducted to validate the system.In Experiment 1, we evaluated four object detection models for damage recognition efficacy.In Experiment 2, we proposed a system for automatic reusability determination based on confidence thresholds.In Experiment 3, we utilized optical character recognition (OCR) and generative pre-trained transformer (GPT) for manufacturing date recognition, achieving 88.2% accuracy with GPT after applying image rotation and cropping.In Experiment 4, we tested an automated image cropping method, resulting in a 5.06% relative error in bounding box areas compared with manual annotation.Future work will be on combining damage and manufacturing date recognition systems and incorporating slip sign detection to further improve the classification accuracy of reusable tires.