Intelligent monitoring of subgrade filling workflows via automatic construction elements recognition using multi-source data fusion

Purpose The increasing automation of construction sites demands real-time monitoring of machinery and operations. Most existing studies identify individual machines or activities, but few recognize construction workflows. This study proposes an intelligent monitoring methodology for detecting subgrade filling workflows to enhance progress tracking and productivity management. Design/methodology/approach This study proposes a multi-source data fusion framework for intelligent monitoring of subgrade filling workflows. The framework integrates three critical approaches: (1) a hierarchical construction sequence library with a knowledge graph capturing temporal dependencies of construction elements operations; (2) a multi-source data fusion approach incorporating YOLOv11-CSA based object detection and Random Forest-based sensor analysis for automatic recognition of construction elements and their states; and (3) an expert decision-making algorithm inferring construction stages by combining multi-source fusion results with the construction sequence library. Findings Validation was conducted in a laboratory scenario using roadside camera video and onboard sensor data. Results demonstrate that construction elements and activity identification achieved accuracy above 98%, construction time recognition errors below 1.7 s, and machinery productivity estimation errors within 5%. Originality/value The main contributions are threefold: (1) a domain-specific recognition framework for subgrade filling workflows supported by a construction-sequence library; (2) a multi-source data fusion approach integrating visual and sensor data to address the limitations of single-source methods; and (3) laboratory validation showing robust performance in identifying construction elements and workflows.