Optimal AI Models for Process Data Analysis in Industry 4.0 Regarding Digital Twin Implementation

This study compares YOLOv8 and Mask R-CNN object detection algorithms for defect detection in digital twin applications for smart manufacturing. With the ongoing transformation of manufacturing processes in Industry 4.0 and 5.0, integrating digital twin technology and advanced visual inspection techniques has become crucial for improving quality control and process optimization. Using the expanded MVTec Metal Nut Dataset, we evaluated performance based on mean average precision (mAP), precision, recall, F1-score, and inference time. Results show YOLOv8 consistently outperformed Mask R-CNN in accuracy and computational efficiency, achieving higher scores across all metrics and quicker inference times. This suggests YOLOv8 is more suitable for real-time industrial applications, particularly in digital twin implementations. However, the study emphasizes the importance of considering specific manufacturing contexts when selecting an object detection model. The integration of AI-powered visual inspection with digital twin technology has significant potential to enhance product quality, reduce downtime, and optimize production processes, contributing valuable empirical evidence to guide the implementation of advanced defect detection systems in industry.