Knowledge-augmented multi-modal data fusion and reasoning for automated crane lift monitoring

Crane lift operations are critical to construction productivity, yet their monitoring remains largely manual, fragmented, and inefficient. This paper introduces a knowledge-augmented, multi-modal data fusion and reasoning framework for the automated tracking and analysis of crane lifts. The proposed approach integrates a domain ontology to fuse computer vision, sensor signals, and schedule data, enabling a hierarchical hybrid reasoning pipeline that infers transient behaviours, segments complete operations, and maps them to scheduled tasks via similarity metrics. In a two-day field experiment, the system accurately recognised eight distinct behaviours with an overall accuracy of 0.911 and an average F1 score of 0.907, segmented lift cycles with a median duration error under 10 s, and mapped operations to scheduled orders with an F1 score of 0.944 and accuracy of 0.905. These results demonstrate the technical feasibility and robustness of the framework, which transforms low-level data into high-level, context-rich knowledge to support productivity assessment and workflow optimisation in construction environments.