مهسا برنون

Abstract The railway transportation system is one of the key components of a country’s transport infrastructure an‎d plays a vital role in providing safe, fast, an‎d cost-effective passenger an‎d freight movement. The stability an‎d health of railway tracks are essential requirements for the reliable operation of this system. However, various factors such as mechanical wear, repetitive dynamic loads, an‎d environmental conditions can lead to the development of different surface defects, including cracks, spalling, local fractures, an‎d plastic deformations in rails. Delayed detection of these defects not only increases the risk of sudden failures an‎d accidents but also significantly raises maintenance costs. Traditional inspection methods mainly rely on visual observation an‎d expert judgment, which suffer from limitations such as low accuracy, time-consuming procedures, an‎d lack of repeatability due to human dependency. Therefore, developing intelligent systems based on machine vision an‎d deep learning for automatic defect detection is essential. In this study, an automatic rail surface defect detection model based on a convolutional neural network (CNN) was designed an‎d implemented. The input data were extracted from videos captured by cameras mounted on the inspection system. After preprocessing an‎d labeling, the data were used to train the model. The eva‎luation results showed that the proposed model achieved an accuracy of 96.39%, demonstrating remarkable performance in identifying various surface defects. Thus, it can serve as an effective tool for intelligent monitoring of railway infrastructure.

كليدواژه‌ها: بينايي ماشين، تشخيص عيوب ريل، يادگيري عميق، EfficientNet، شبكه عصبي كانولوشني

Keywords: Machine Vision, Rail Defect Detection, Deep Learning, EfficientNet, Convolutional Neural Network (CNN).

Mahsa Barnoon

Dr, Poorseyed aghaei