Digital twinning of existing reinforced concrete bridges from labelled point clusters

Abstract

Abstract The automation of digital twinning for existing reinforced concrete bridges from point clouds remains an unresolved problem. Whilst current methods can automatically detect bridge objects in point clouds in the form of labelled point clusters, the fitting of accurate 3D shapes to point clusters remains largely human dependent largely. 95% of the total manual modelling time is spent on customizing shapes and fitting them correctly. The challenges exhibited in the fitting step are due to the irregular geometries of existing bridges. Existing methods can fit geometric primitives such as cuboids and cylinders to point clusters, assuming bridges are comprised of generic shapes. However, the produced geometric digital twins are too ideal to depict the real geometry of bridges. In addition, none of the existing methods have explicitly demonstrated how to evaluate the resulting Industry Foundation Classes bridge data models in terms of spatial accuracy using quantitative measurements. In this article, we tackle these challenges by delivering a slicing-based object fitting method that can generate the geometric digital twin of an existing reinforced concrete bridge from four types of labelled point cluster. The quality of the generated models is gauged using cloud-to-cloud distance-based metrics. Experiments on ten bridge point cloud datasets indicate that the method achieves an average modelling distance of 7.05 cm (while the manual method achieves 7.69 cm), and an average modelling time of 37.8 s. This is a huge leap over the current practice of digital twinning performed manually.