Artificial intelligence based microcracks research in 3D printing concrete

3D printing concrete (3DPC), which employs extruded-filament and no-framework technologies, is more vulnerable to effects of internal microcracks than traditional concrete. However, the existing detection and analysis method of microcracks relies on manual visual inspection and the experience of observers, which fall short in quantifying representative information. This study proposes a fast-speed and automated segmentation method of microcracks in 3DPC through dual deep learning (DL) algorithms and scanning electron microscopy (SEM). This method applies transformer based networks and a range of improvement tactics to respectively achieve high-precision results of mIoU at 0.7586 and mPA at 0.8332. These performances for segmenting microcracks in 3DPC are better than results obtained from current popular networks. To significantly reduce the time and cost of acquiring high-quality SEM data, a super-resolution reconstruction strategy is adopted to obtain higher-resolution data with 34.23 of PSNR and 0.87 of SSIM. Using the proposed method, a comprehensive analysis reveals that 3DPC has a higher risk for microcrack progression into macrocracks than cast concrete. Therefore, this method surpasses the existing techniques related to microscopic imaging in terms of intuitiveness, automation, informational content, and accuracy.