Automated analysis system for micro-defects in 3D printed concrete

Abstract

The internal micro-defects of 3D printed concrete (3DPC) play a pivotal role in influencing its mechanical properties. Nonetheless, the acquisition of representative internal micro-defect information is hindered by computational inefficiencies and quantification limitations of the current equipment system. This paper proposes a deep learning based system to assist SEM equipment in automatically quantifying micro-defects of 3DPC for in-depth microstructural analysis that surpasses traditional SEM methods. Through optimal resizing approach and model enhancement tactics, the proposed micro-defect segmentation model leverages advantages of both convolutional neural networks and transformer. This improvement segmentation capability achieves higher accuracy and faster speed than current algorithms, enabling system to achieve accurate quantitative analyses of micro-defects. Using this automated analysis system, the relationship among micro-defect areas in 3DPC, mechanical properties, and printer parameters is investigated. Therefore, the proposed system reduces labour and computational time, demonstrating significant potential for applications in analyzing concrete microstructure.