Geometrical quality inspection in 3D concrete printing using AI-assisted computer vision

Abstract

3D concrete printing is an innovative technology poised to transform the construction industry by enabling the automated, layer-by-layer creation of structures directly from digital models. This approach offers numerous advantages over traditional construction methods, including reduced labor costs, faster build times, and the ability to produce complex geometries with high precision. However, unlike conventional mold-cast concrete, 3D printable concrete must support itself without external formwork, posing significant challenges related to material deformation during the printing process. Uncontrolled deformation can lead to structural instability, design deviations, and cumulative errors. Traditional methods for monitoring the geometrical quality of 3D-printed concrete are often insufficient in accuracy and efficiency. Recent advancements in artificial intelligence (AI) present new opportunities for addressing these challenges. AI-assisted methods leverage machine learning to analyze large datasets, enabling more accurate predictions and real-time monitoring and control of deformation during the 3D printing process. In this paper, we explored the application of AI-assisted methods for real-time deformation analysis in 3D concrete printing. Specifically, the Yolo-v5 algorithm, an AI-assisted object detection technique, was employed for the computer vision of extruded concrete filaments. Several quantitative metrics were proposed, including the layer height, layer angle, and curvature. In addition, the rheological properties of 3D-printed concrete were measured to refine the computer vision analysis results. Through experimental validation, we demonstrated the effectiveness of the developed AI-assisted computer vision system in monitoring the 3D concrete printing process.