Developing a Computer Vision Application for Crack Detection

The paper addresses a critical issue in 3D concrete printing (3DCP), which is the detection of crack formation during and after the printing process. This problem is significant because cracks in concrete structures can compromise their integrity and needs to be addressed as part of a 3DCP approach implemented at TU Delft for the development of extraterrestrial habitats. Despite the importance of detecting cracks in 3DCP, there is a lack of specific datasets for training computer vision models to perform this task accurately.
Previous efforts in this area have been limited, with most existing solutions focusing on traditional casting methods rather than the challenges presented by 3DCP. This paper presents a solution that leverages Computer Vision (CV) techniques, by retraining the YOLOv7 Object Detection (OD) model using a dataset of cracks in regular cast concrete. This approach acknowledges the absence of a substantial 3DCP-specific dataset and adapts existing models to address the problem. By utilizing CV, the proposed method demonstrates similar performance in precision, recall, and mean average precision between the two models YOLOv7 and YOLOv7 tiny. However, the challenges encountered during crack detection in 3DCP samples highlight the need for an expanded and annotated dataset tailored to this specific application.
The paper aims to provide a comprehensive overview of the problem, the challenges faced, and the proposed solution. By retraining the YOLOv7 model and discussing the advantages and challenges of CV, the paper contributes to advancing the field of crack prevention in 3DCP. Ultimately, the goal is to improve the quality and safety of 3D-printed concrete structures by enabling timely detection and mitigation of crack formation.