Automated structural crack detection using a deep learningbased YOLO algorithm and robotic dog

4th International Civil Engineering & Architecture Conference, Trabzon, Türkiye, 17 - 19 Mayıs 2025, cilt.1, ss.2137-2149, (Tam Metin Bildiri)

This study aims to automatically detect cracks in structural elements using the YOLO object detection

algorithm, based on Deep Learning techniques, integrated with robotic dog technology. This approach seeks to

enhance both the accuracy and efficiency of structural health monitoring processes. The study begins with the

selection of appropriate datasets via the Roboflow platform, where a dataset consisting of 2,600 annotated crack

images was utilized for model training. Training was conducted in the Google Colab environment, resulting in the

development of an object detection model. Additionally, a segmentation model was created using the same dataset

to ensure more precise crack boundary delineation. The performance of the trained models was evaluated through

real-time tests using images captured by cameras mounted on the Unitree Go2 Edu Plus robotic dog. These tests

were conducted on a reinforced concrete frame with induced cracks at the Structural Health Monitoring Laboratory

of Karadeniz Technical University. The images were processed in real-time using YOLO models, enabling

immediate crack detection. The findings demonstrate that the robotic system, combined with the YOLO algorithm,

effectively detects cracks in real-time with high accuracy. This study highlights the potential of integrating Deep

Learning and robotic technologies for structural health monitoring, offering a reliable and efficient solution for

damage assessment in civil engineering applications. The results validate the proposed approach as a practical and

dependable method for automated structural crack monitoring.