Deep Filament Extraction for 3D Concrete Printing

TL;DR

This work tackles the challenge of quality control for filament geometry in large-scale 3D concrete printing, addressing both extrusion-based and shotcrete 3D printing. It introduces a sensor-agnostic QC pipeline that projects diverse 3D data (from cameras, SLS, or TLS) into 2D via a virtual camera, performs filament instance segmentation with YOLOv11-seg, and quantifies filament height using a distance transform. The approach is validated across multiple datasets and sensor modalities, achieving fast online inference (≈76 FPS) and enabling back-projection of results into 3D space for integration with digital twins and BIM/FIM. The work demonstrates a scalable framework for automated, data-driven QC in digital construction workflows, with clear paths toward closed-loop control and broader applicability to non-planar surfaces and varied materials.

Abstract

The architecture, engineering and construction (AEC) industry is constantly evolving to meet the demand for sustainable and effective design and construction of the built environment. In the literature, two primary deposition techniques for large-scale 3D concrete printing (3DCP) have been described, namely extrusion-based (Contour Crafting-CC) and shotcrete 3D printing (SC3DP) methods. The deposition methods use a digitally controlled nozzle to print material layer by layer. The continuous flow of concrete material used to create the printed structure is called a filament or layer. As these filaments are the essential structure defining the printed object, the filaments' geometry quality control is crucial. This paper presents an automated procedure for quality control (QC) of filaments in extrusion-based and SC3DP printing methods. The paper also describes a workflow that is independent of the sensor used for data acquisition, such as a camera, a structured light system (SLS) or a terrestrial laser scanner (TLS). This method can be used with materials in either the fresh or cured state. Thus, it can be used for online and post-printing QC.

Figures (10)

• Figure 1: The prediction results of the filament instances are projected back into 3D space. (a) Extrusion-based object from Photogrammetry: The results shown are for the data from Fig. \ref{['Fig: Results_imposed']}b. (b) SC3DP object from TLS: The results shown for the data from Fig. \ref{['Fig: Results_imposed']}c. The left image shows the raw data of the whole object superimposed with the predicted instances in 3D. The middle image shows the same as the left image, but zoomed in. The right-hand image shows the zoomed-in view, but only the predicted results in 3D space.
• Figure 2: Workflow of the proposed method.
• Figure 3: Euclidean transformation between the world and camera coordinate frames, Adopted from Hartley_Zisserman_2004. K: Intrinsic matrix, R: Rotation matrix, t: Transformation vector, and GSD: ground sampling distance.
• Figure 4: Different VC orientation scenarios: (a) predefined four sides, where the camera sensor plane is parallel to one side of the bounding box sides. (b) Frustum Establishment from a VC camera. (c) The points inside the frustum are only considered.
• Figure 5: VC view direction axes—direct (gold) vs. horizontal-projected (blue) through the point-cloud centroid. Camera placements along these axes to achieve the required working distance and GSD.