Instance Segmentation Models for Real-Time Progress Monitoring and Quality Control in Industrial Plant Construction

Abstract

[Abstract]: This paper proposes a progress monitoring solution through instance segmentation models for real-time production monitoring and quality control in industrial environments. The obtained results address digitalization challenges for automated monitoring of early-stage construction elements including foundation excavations, cast-in-place foundations, steel I-beam and lattice beam placement. Specifically, state-of-the-art instance segmentation architectures are evaluated, consisting of recent YOLO models (YOLOv8-seg, YOLOv11-seg) as well as established approaches (Mask R-CNN, Mask2Former) for industrial monitoring systems integration. An evaluation framework is established that integrates accuracy metrics (mask mean Average Precision (mAP)) with production efficiency measures including real-time processing (FPS) and deployment constraints. The performed analyses reveal accuracy-speed trade-offs essential for real-time applications in Industry 5.0 scenarios, where YOLO variants provide real-time monitoring capabilities and two-stage approaches deliver enhanced precision. Experimental results demonstrate that YOLOv8-seg variants achieve optimal performance with mAP@0.5:0.95 up to 54.09 % at 65 FPS, while YOLOv11-seg models provide superior parameter efficiency for resource-constrained environments.