مسعود افروغ

Concrete sleepers (railroad ties), as key components of railway infrastructure, play a vital role in the safety an‎d stability of railway tracks, an‎d the presence of surface cracks can be an indicator of potential failure. The objective of this research is to investigate an‎d eva‎luate image processing an‎d deep learning-based methods for detecting surface cracks an‎d analyzing the severity of damage in concrete sleepers. In this study, images of sleepers were collected under both static an‎d dynamic data acquisition conditions. After preparation an‎d labeling, they were used to train an‎d eva‎luate various models from the YOLO family, including YOLOv8, YOLOv11, YOLOv12, YOLO-NAS, an‎d RF-DETR. The performance of the models was assessed using quantitative metrics such as Precision, Recall, mAP50, an‎d F1-Score, an‎d a qualitative analysis of the outputs was also conducted. Furthermore, using the geometric dimensions of the cracks, the crack damage severity was classified into different levels—"Low," "Medium," an‎d "High"—to enable a more accurate assessment of the sleeperʹs health condition. The results demonstrated that the YOLOv12 model delivered the best performance in terms of detection accuracy an‎d stability, an‎d exhibited suitable generalizability to new data. The quantitative results for this model in terms of Precision, Recall, mAP50, an‎d F1-Score were 86.8%, 77.5%, 82.8%, an‎d 81.89%, respectively. Although the images acquired under dynamic conditions presented challenges such as vibration an‎d varying lighting, the selec‎ted model was able to provide acceptable performance, indicating the applicability of the system for field monitoring. The crack severity analysis also showed that combining automated detection with damage severity assessment can assist in prioritizing maintenance actions an‎d enhancing railway track safety. This research provides a foundation for the development of intelligent monitoring systems for concrete sleepers an‎d supports preventive maintenance management in railway infrastructure. It can serve as a basis for future research an‎d applications in this field.

تراورس هاي بتني , تشخيص ترك سطحي , يادگيري عميق , يولو , تحليل شدت خرابي , پايش خطوط ريلي

Concrete sleepers , surface crack detection , deep learning , YOLO , crack severity analysis , railway condition monitoring

Masoud Afrough

Prof. Jabbar Ali Zakeri, Dr. Hamid Reza Heydari