In-process filament repair in 3D concrete printing with instance segmentation

Abstract

3D Concrete Printing (3DCP) stands as an automated method using robotics to construct concrete structures layer by layer, offering time efficiency, cost savings, and improved safety. Despite its practical relevance, on-site operators are still necessary to identify printing issues. While research focuses primarily on material optimisation, lesser attention is given to processing, software, and building integration aspects. This study delves into the gaps and correction behaviour of 3DCP cementitious materials, employing simulation tools and computer vision. Using varied nozzle speeds (60, 30, and 20 mm/s), repairs were attempted above simulated gaps on the base layer. Observations revealed a slump-like behaviour in filament gaps, increased layer width for 33.3% and 50% speeds, and the necessity for smaller repair adaptability. Optimal repair speed for filament tears fell between 33.3% and 50% of the nozzle travel speed. Assessing real-time in-process scanning and repair for a closed-loop system proved promising, though hardware limitations introduced unforeseen noise. Future work suggestions encompass improved simulated models, nozzle travel speed optimisation, enhanced inference models for higher layers, and nozzle adaptations for smaller repairs.

Keywords

• Computer vision
• YOLO
• Concrete 3D Printing
• additive manufacturing
• sustainability