THE OPTIMIZATION OF RAILWAY FASTENER DEFECT DETECTION VIA ACTIVATION FUNCTION ADAPTATIONS

Year 2024, Volume: 25 Issue: 4, 557 - 566, 27.12.2024

Rıdvan Özdemir , Mehmet Koç

https://doi.org/10.18038/estubtda.1479970

Abstract

Manual control of rail defect detection is slow and costly. Deep learning methods can detect some of these defects to a certain extent. However, existing systems produce too many false positives due to environmental factors, resulting in labor and cost losses. One of the most important components in railway systems is the fastener, and their failure can lead to severe accidents. In this study, we developed a deep learning-based method that is designed to remain robust against foreign objects and environmental conditions when detecting railway fasteners. By employing various activation functions and expanding the training dataset through data augmentation techniques, our method significantly reduces false alarms. The best-performing activation function in our tests achieved an F1-score of 0.99 and a mean average precision (mAP) of 100%. Testing on a dataset provided by TCDD Railway Research & Technology Centre (DATEM) confirms the efficacy of our approach, demonstrating a notable decrease in unnecessary work and associated costs.

Keywords

YOLOv4, Railway component, Deep learning, Activation function, Fastener defect

Supporting Institution

Eskisehir Technical University Scientific Research Projects Commission

Project Number

21GAP081

Thanks

This research received funding from the Eskisehir Technical University Scientific Research Projects Commission, under grant number 21GAP081. The authors express their gratitude to TCDD Railway Research & Technology Centre (DATEM) for supplying the dataset.

References

• [1] Dinhobl G, Petz M, Kümmritz S, Hutterer H. On Measurement of Railway Noise: Usability of Acoustic Camera. In: Degrande G, Lombaert G, Anderson D, de Vos P, Gautier P-E, Iida M, Nelson JT, Nielsen JCO, Thompson DJ, Tielkes T, Towers DA, editors. Noise and Vibration Mitigation for Rail Transportation Systems. Cham: Springer International Publishing, 2021, pp. 234–241.
• [2] Assali P, Viguier F, Pollet N. Contribution of Terrestrial Laser Scanning for monitoring and inspection of railway infrastructure. In Proceedings of the World Congress on Railway Research; 2013 Nov 25-28; Sydney, Australia. 2013.
• [3] Szandała T. Review and comparison of commonly used activation functions for deep neural networks. In: Bio-inspired Neurocomputing 903, Bhoi AK, Mallick PK, Liu CM, Balas VE, editors. Studies in Computational Intelligence, vol. 903. Singapore: Springer Singapore, 2021, pp. 203–224.
• [4] Lin Y-W, Hsieh C-C, Huang W-H, Hsieh S-L, Hung W-H. Railway track fasteners fault detection using deep learning. In: 2019 IEEE Eurasia Conference on IOT, Communication and Engineering, 2019, pp. 187–190.
• [5] Qi H, Xu T, Wang G, Cheng Y, Chen C. MYOLOv3-Tiny: A new convolutional neural network architecture for real-time detection of track fasteners. Computers in Industry; 2020; 123: 103303.
• [6] Güçlü E, Aydin İ, Şahbaz K., Akin E, Karaköse M. Detection of defects in railway fasteners using YOLOv4 and fuzzy logic (article in Turkish with an abstract in English). Railway Engineering 2021; 14: 249-262.
• [7] Liao X, Lv S, Li D, Luo Y, Zhu Z, Jiang C. YOLOv4-MN3 for PCB surface defect detection. Applied Sciences 2021; 11(24):1-17.
• [8] Şener A, Ergen B, Toğaçar M. Fault detection from images of railroad lines using the deep learning model built with the tensorflow library. Turkish Journal of Science & Technology 2022; 17(1): 47-53.
• [9] Ozdemir R, Koc M. On the enhancement of semi-supervised deep learning-based railway defect detection using pseudo-labels. Expert Systems With Applications 2024; 251: 124105.
• [10] Sevi̇ M, Aydın İ, Karaköse M. Classification of railway fasteners by deep learning methods (article in Turkish with an abstract in English). European Journal of Science and Technology 2022; 35: 268-274.
• [11] He J, Wang W, Yang N. Research on track fastener service status detection based on improved YOLOv4 model. Journal of Transportation Technologies; 14: 212-223.
• [12] Yılmazer M, Karaköse M, Aydın İ, Akın E. Multiple fault detection in railway components with mask R-CNN deep neural network (article in Turkish with an abstract in English) Cukurova University Journal of the Faculty of Engineering 2022; 37(4); 1103-1111.
• [13] Mi Z. Research on steel rail surface defects detection based on improved YOLOv4 network. Frontiers in Neurorobotics 2023; 17: 1-11.
• [14] Alzubaidi L, Zhang J, Humaidi AJ, Al-Dujaili A, Duan Y, Al-Shamma O, Santamaria J, Fadhel MA, Al-Amidie M, Farhan L. Review of deep learning: concepts, CNN architectures, challenges, applications, future directions 2021; 8(1); 1-74.
• [15] Bochkovskiy A, Wang C-Y, Liao H-YM. YOLOv4: Optimal speed and accuracy of object detection. arXiv preprint 2020; 2004.10934.
• [16] Wang C-Y, Liao H-YM, Wu YH, Chen P-Y, Hsieh J-W, Yeh I-H, CSPNet: a new backbone that can enhance learning capability of CNN. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops; 2020; pp. 390-391.
• [17] Liu S, Qi L, Qin H, Shi J, Jia J. Path aggregation network for instance segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2018; pp. 8759-8768.
• [18] He K, Zhang X, Ren S, Sun J. Spatial pyramid pooling in deep convolutional networks for visual recognition IEEE Transactions of Pattern Analysis and Machine Intelligence 2015; 37(9): 1904-1916.
• [19] Getting Started with YOLOv4 - MATLAB & Simulink. Accessed: Feb. 23, 2024. [Online]. Available: https://www.mathworks.com/help/vision/ug/getting-started-with-yolo-v4.html