Buildability Analysis in 3D Concrete Printing Using Computer Vision and Automated Annotation (2025-10)¶

Advanced 3D concrete printing (3DCP) requires real-time monitoring and feedback control to ensure build quality. This study developed a vision-based deep learning workflow to delineate interlayers in 3DCP structures, enabling both layer-wise analysis and overall structural assessment. To improve reliability and efficiency, a grayscale pattern recognition technique was introduced to generate training annotations and reduce manual labeling effort. The automated annotation reduced manual labeling time from nearly an hour to under one minute per batch of images. The resulting dataset was used for transfer learning with DeepLabV3+, achieving an F1-score of 94.28% at a 2-pixel tolerance and real-time inference (98 ms per frame). Based on the segmented interlayers, layer thickness and inclination distributions were computed. By integrating the trained model with image processing techniques, a buildability analysis system was developed to detect structural deformations and visualize results. This workflow enables real-time, robust analysis of diverse 3DCP structures, thereby facilitating efficient monitoring.