Remote Expert AI System for Industrial Maintenance: A Technical Approach

Industrial maintenance has become increasingly important in industrial settings for maintaining operational efficiency and reducing downtime. This study evaluates YOLO object detection algorithms versions 5, 8, and 9 in combination with Microsoft's HoloLens 2 for detecting anomalies in PLCnext equipment. The research compares model performance using accuracy metrics and explores on-device and API-based implementation approaches. Results show YOLOv9 and YOLOv8 outperforming YOLOv5 in accuracy, with YOLOv9 demonstrating slightly higher performance. However, both YOLOv8 and YOLOv9 exhibited high CPU usage (51.75% and 61.32% respectively) during testing. Given the limited computational resources of HoloLens 2, an API-based approach is recommended to offload processing to server-side infrastructure. This research contributes to the advancement of augmented reality-assisted predictive maintenance in industrial settings by identifying effective methods for deploying object detection models. The findings suggest potential improvements in operational efficiency and reduced downtime in manufacturing environments through optimized implementation strategies.