According to various statistics concerning digitization, the construction sector is listed at the bottom. Despite current developments in other industry sectors, only little advancements are made, resulting in risks of cost and time overruns. Multiple sources trace these problems back to progress monitoring in the complex and dynamic on-site environments. This thesis provides a novel approach towards automating this critical and time-consuming task. It involves creating a new dataset of annotated site images and training semantic segmentation models to recognize cast-in-place concrete walls, columns, and slabs in panel, rebar, and concrete phases. Continuous site images are segmented and automatically processed, including averaging techniques to detect discrete element-specific progress timestamps. These are coupled with a BIM model or digital twin using the element’s GUID to provide the results for further computations. As verification, in-depth case studies are performed. Hereby, 49 semantic segmentation models are trained, and the derived construction progress timestamps are compared against as-built information for selected elements, resulting in reliable accuracies. Based on a real-world project, monitoring data is coupled with a BIM model, producing promising results and uncovering incentives for further research. Overall, this thesis’ novel approach provides potential improvements for the construction sectors digitization.     «

According to various statistics concerning digitization, the construction sector is listed at the bottom. Despite current developments in other industry sectors, only little advancements are made, resulting in risks of cost and time overruns. Multiple sources trace these problems back to progress monitoring in the complex and dynamic on-site environments. This thesis provides a novel approach towards automating this critical and time-consuming task. It involves creating a new dataset of annota...     »