Railway Track Fault Detection: History
View Latest Version
Please note this is an old version of this entry, which may differ significantly from the current revision.
Contributor:
Furqan Rustam
,
Abid Ishaq
,
Muhammad Shadab Alam Hashmi
,
Hafeez Ur Rehman Siddiqui
,
Luis Alonso Dzul López
,
Juan Castanedo Galán
,
Imran Ashraf

Railway track faults may lead to railway accidents and cause human and financial loss. Spatial, temporal, and weather elements, and wear and tear, lead to ballast, loose nuts, misalignment, and cracks leading to accidents. Manual inspection of such defects is time-consuming and prone to errors. Automatic inspection provides a fast, reliable, and unbiased solution. However, highly accurate fault detection is challenging due to the lack of public datasets, noisy data, inefficient models, etc. 

  • railway track fault detection
  • trains
  • derailment
  • machine learning
  • acoustic analysis
  • image-processing

1. Introduction

The railway industry has been considered the backbone of a country’s economy, transporting goods and people, and thus offering a potential share in the development of a country. In contrast to road vehicles, trains carry a larger number of people which makes them attractive both to governments and the general population. The public has a low tolerance level for train accidents as they involve a high risk of damage to humans, as well as substantially influencing economic activities. Such accidents put a country’s reputation at risk and political and social risk levels can rise [1]; however, avoiding or reducing the frequency of such accidents to a minimum is very challenging. Derailment, injury, economic burden, death, and loss of public confidence are all undesirable consequences of railway track defects and failure. During railway track maintenance and inspection activities, maintenance staff can also receive injuries or lose their lives [2]. Thus, safe railway operations need proper maintenance, which significantly relies on railway track inspection and fault detection [1]. The safety, reliability, and cost-effectiveness of railway operations are all dependent on railway track condition monitoring. Governments also set regulations for frequent railway track inspections, which generally require a lot of manpower and resources. Therefore, railway track condition monitoring and fault detection are critical due to safety, regulatory, and economic factors [3][4].
Every year a large number of people in Pakistan travel by train. From 2018 to 2019 approximately 70 million people used rail to reach their destinations [5]. Pakistan railway freight also transported 7.4 million tonnes in the year 2020 [6]. However, in the past few years, several serious accidents took place that caused huge human and economic losses [7]. Such accidents can happen due to human error, weather conditions, or faulty railway tracks. According to the Pakistan railway’s annual statistics, train derailments due to railway track faults caused 127 accidents between 2013 to 2020 [8]. The year 2014 was the worst year for the Pakistan railway, as 228 freight trains and 16 passenger trains were derailed, the maximum for any state [9]. In 2019, 23 bogies of freight trains were derailed near Sukkur. Importantly most of the train derailments occurred between the Sukkur–Multan sections [10]. The main reason behind this is the poor condition of railway tracks and the lack of modern resources and techniques to monitor track conditions.
Railway systems around the world operate in a variety of environments where the railway track is threatened by temporal, spatial, and weather factors. The presence of cracks and track conditions are the major factors in rail derailment. Manual inspections consume huge resources and time [11]. They are also prone to human bias and judgement errors [12]. An automated method is required to address the issue of derailment and to ensure the proper investigation of tracks. The objective is to create a system that can assess the given inputs and provide a clear indication of whether the track is faulty or not. This research is concerned with creating a reliable system that can analyze sound signals from tracks and detect whether a track is cracked or not. Railways are one of Pakistan’s most important modes of transportation and have recently experienced a series of rail catastrophes. Keeping this in mind, railway tracks are the most important factor in derailment, and an efficient and effective track fault detection system is needed.

2. Railway Track Fault Detection Methods

Machine learning models can be effective for tackling a variety of problems in such areas as computer vision applications [13][14], text mining [15], image processing [16], and the IoT [17][18], etc. This research also used a machine learning approach for railway track fault detection. In manual railway tracking, fault detection is very difficult, time-consuming, and labor-intensive. AI advancements have led to more precise and accurate railway track fault detection systems while dealing with sensitive data. Railway cracks are the leading cause of derailment all around the world. Machine learning and deep learning models have been proposed to identify these.
Shafique et al. [17] used an acoustic analysis approach to design an automatic railway track fault detection system. They collected data using the traditional railway cart system. Due to their common occurrence, they considered three types of tracks including normal tracks, wheel burnt tracks, and superelevated tracks. They used several machine learning models and showed that RF and DT were able to achieve 97% accuracy. Similarly, by using acoustic analysis, Bhushan et al. [19] proposed a system for the early detection and diagnosis of faults in railway points. An NS-AM-type railway point machine with audio sensors was used for collection of the dataset. This research mainly analyzed faults such as slackened nuts, ballast blast obstruction, and ice obstruction. Two experiments were conducted, one for fault classification on the whole dataset and the other for fault classification. The model evaluation showed an accuracy of 94.1%. Hashmi et al. [20] proposed a conventional acoustic-based system for automatic railway fault detection. They used deep learning models including CONV1D, CONV2D, recurrent neural networks (RNN), and LSTM to address the problem. They considered three types of faults including normal tracks, wheel-burnt tracks, and superelevated tracks. Audio samples of different duration were used to analyze the performance of each model. Each 17 s audio sample was divided into three segments of 1.7 s, 3.4 s, and 8.5 s; the deep learning models were trained and tested against each segment. The performance of the models was investigated using various combinations of audio data augmentation. For the 8.5 s segment, LSTM achieved an accuracy of 99.7%.
Predominantly, image-processing-based methods are utilized for railway track fault detection. For example, Ritika et al. [21] proposed a computer-vision-based system for real-time railway track fault detection. They used a camera mounted on a locomotive to capture images at 30 frames per second. For binary classification, the Inception V3 model was used on the ImageNet dataset. For vegetation overgrowth, the model generalized well on actual vegetation images with a 97.5% precision value. The Sun Kink classifier can professionally classify simulated Sun Kink videos. Similarly, study [22] used different variants of the deep convolutional neural network (DCNN) for railway track fault detection using image data. They used the DCNN-small, DCNN-medium, and DCNN-large networks in their work. The different network architectures were characterized by different sizes and activation functions. The experimental results showed an accuracy of 92% for the large DCNNs.
Manikandan et al. [23] proposed a feed-forward neural network to detect and segment faults from railway track images. They used an adaptive histogram equalization technique to track image enhancement and then features were extracted from the enhanced images. The proposed feed-forward back propagation neural network achieved a 94.9%, 89.99%, and 98.96% accuracy score, sensitivity score, and specificity score, respectively, on the enhanced images. Santur et al. [24] proposed a computer-vision-based system for the inspection of faults in railway tracks. They only inspected faults such as scouring, breaking, and deficient fasteners. The researchers extracted the features from video images containing the healthy railway track, while, for the faulty tracks, virtual faults were generated on the original images. Using a modified RF, the highest accuracy of 98% was obtained with HM features.
Tastimur et al. [25] performed fault detection and classification using railway track images with the AdaBoost classifier. Various image processing techniques were also included in their work and they achieved an accuracy of 94.73% for defect detection and 87% for defect classification. Defect detection refers to confirming if there is a defect present while defect classification refers to deciding the type of defect. Chen et al. [26] proposed a deep-learning-based system using B-scan image recognition of rail defects with an improved YOLOV3 algorithm. The proposed system automatically positions a box in B-scan images and recognizes normal bolt holes, EFBWs (electric flash butt wheels), SSCs (shell spallings or corrugation), and BHBs (bolt hold breaks). The experiments involved used 453 B-scan images as a test dataset. The results demonstrated that the improved YOLOV3 achieved a precision of 87.41%. Similarly, Li et al. [27] proposed an ensemble learning model that uses multiple learning algorithms for better predictive performance. They used multiple backbone neural networks individually to obtain the features and mixed them in a binary format to obtain diverse and improved sub-networks. Different image augmentation and feature augmentation techniques were randomly used to achieve diversity. On an 8-defect class dataset, the proposed MBDA (multi backbone double augmentation) system achieved a 2.8% higher mAP.5 compared with faster R-CNN and a 74% higher mAP.5 compared with YOLOV5.
Nandhini et al. [28] used an unsupervised multi-scale CNN for robust automatic railway tracking for detection. They used vibration data for crack detection. They used an open-source dataset in their study. Different machine learning models with different feature extraction techniques were used; the proposed CNN system achieved an accuracy of 89%. A comprehensive overview of the literature shows that current techniques perform well in the detection of faults. Computer-vision-based techniques are extensively used in this regard; acoustic-based techniques still need development for the efficient detection of railway faults. The results obtained indicate that both image-processing- and acoustic-based approaches perform well with respect to railway track fault detection; however, research into the use of acoustic approaches is lacking. Dedicated research efforts are needed in this context.

This entry is adapted from the peer-reviewed paper 10.3390/s23167018

References

  1. Cannon, D.; Edel, K.O.; Grassie, S.; Sawley, K. Rail defects: An overview. Fatigue Fract. Eng. Mater. Struct. 2003, 26, 865–886.
  2. Beeck, F. Track Circuit Monitoring Tool: Standardization and Deployment at CTA; Technical Report; The National Academies of Sciences: Washington, DC, USA, 2017.
  3. Rail Defects Handbook. Available online: https://extranet.artc.com.au/docs/eng/track-civil/guidelines/rail/RC2400.pdf (accessed on 5 May 2022).
  4. Ji, A.; Woo, W.L.; Wong, E.W.L.; Quek, Y.T. Rail track condition monitoring: A review on deep learning approaches. Intell. Robot 2021, 1, 151–175.
  5. British Broadcasting Corporation. Pakistan Train Fire: Are Accidents at a Record High? Available online: https://www.bbc.com/news/world-asia-50252409 (accessed on 5 May 2022).
  6. Statista. Number of Rail Accidents and Incidents in the United States from 2013 to 2020. Available online: https://www.statista.com/statistics/204569/rail-accidents-in-the-us/ (accessed on 5 May 2022).
  7. Mustafa, A.; Rasheed, O.; Rehman, S.; Ullah, F.; Ahmed, S. Sensor Based Smart Railway Accident Detection and Prevention System for Smart Cities Using Real Time Mobile Communication. Wirel. Pers. Commun. 2023, 128, 1133–1152.
  8. Pakistan Railways Achieves Record Income in 2018–2019. Available online: https://www.railjournal.com/news/pakistan-railways-achieves-record-income-in-2018-19/ (accessed on 5 May 2022).
  9. Auditor General of Pakistan. Audit Report on the Accounts of Pakistan Railways Audit Year 2019–2020; Ministry of Railways Govt of Pakistan: Lahore, Pakistan, 2020. Available online: https://agp.gov.pk/SiteImage/Policy/Audit%20Report%202019-20%20Railways..pdf (accessed on 5 May 2022).
  10. Outrage over train crash that killed 20 in Pakistan. Available online: https://gulfnews.com/world/asia/pakistan/outrage-over-train-crash-that-killed-20-in-pakistan-1.70052037 (accessed on 5 May 2022).
  11. Asber, J. A Machine Learning-Based Approach for Fault Detection of Railway Track and Its Components. Master’s Thesis, Luleå University of Technology, Operation, Maintenance and Acoustics, Lulea, Sweden, 2020.
  12. Zhuang, L.; Wang, L.; Zhang, Z.; Tsui, K.L. Automated vision inspection of rail surface cracks: A double-layer data-driven framework. Transp. Res. Part C Emerg. Technol. 2018, 92, 258–277.
  13. Mendieta, M.; Neff, C.; Lingerfelt, D.; Beam, C.; George, A.; Rogers, S.; Ravindran, A.; Tabkhi, H. A Novel Application/Infrastructure Co-design Approach for Real-time Edge Video Analytics. In Proceedings of the 2019 SoutheastCon, Huntsville, AL, USA, 11–14 April 2019; IEEE: Piscataway, NJ, USA, 2019; pp. 1–7.
  14. George, A.; Ravindran, A. Distributed middleware for edge vision systems. In Proceedings of the 2019 IEEE 16th International Conference on Smart Cities: Improving Quality of Life Using ICT & IoT and AI (HONET-ICT), Charlotte, NC, USA, 6–9 October 2019; IEEE: Piscataway, NJ, USA, 2019; pp. 193–194.
  15. Rustam, F.; Ashraf, I.; Mehmood, A.; Ullah, S.; Choi, G.S. Tweets classification on the base of sentiments for US airline companies. Entropy 2019, 21, 1078.
  16. Rustam, F.; Reshi, A.A.; Aljedaani, W.; Alhossan, A.; Ishaq, A.; Shafi, S.; Lee, E.; Alrabiah, Z.; Alsuwailem, H.; Ahmad, A.; et al. Vector mosquito image classification using novel RIFS feature selection and machine learning models for disease epidemiology. Saudi J. Biol. Sci. 2022, 29, 583–594.
  17. Shafique, R.; Siddiqui, H.U.R.; Rustam, F.; Ullah, S.; Siddique, M.A.; Lee, E.; Ashraf, I.; Dudley, S. A novel approach to railway track faults detection using acoustic analysis. Sensors 2021, 21, 6221.
  18. George, A.; Ravindran, A.; Mendieta, M.; Tabkhi, H. Mez: An adaptive messaging system for latency-sensitive multi-camera machine vision at the iot edge. IEEE Access 2021, 9, 21457–21473.
  19. Bhushan, M.; Sujay, S.; Tushar, B.; Chitra, P. Automated vehicle for railway track fault detection. In IOP Conference Series: Materials Science and Engineering; IOP Publishing: Bristol, UK, 2017; Volume 263, p. 052045.
  20. Hashmi, M.S.A.; Ibrahim, M.; Bajwa, I.S.; Siddiqui, H.U.R.; Rustam, F.; Lee, E.; Ashraf, I. Railway track inspection using deep learning based on audio to spectrogram conversion: An on-the-fly approach. Sensors 2022, 22, 1983.
  21. Ritika, S.; Rao, D. Data Augmentation of Railway Images for Track Inspection. arXiv 2018, arXiv:1802.01286.
  22. Faghih-Roohi, S.; Hajizadeh, S.; Núñez, A.; Babuska, R.; De Schutter, B. Deep convolutional neural networks for detection of rail surface defects. In Proceedings of the 2016 International Joint Conference on Neural Networks (IJCNN), Vancouver, BC, Canada, 24–29 July 2016; IEEE: Piscataway, NJ, USA, 2016; pp. 2584–2589.
  23. Manikandan, R.; Balasubramanian, M.; Palanivel, S. An efficient framework to detect cracks in rail tracks using neural network classifier. Am.-Eurasian J. Sci. Res. 2017, 12, 218–222.
  24. Santur, Y.; Karaköse, M.; Akin, E. Random forest-based diagnosis approach for rail fault inspection in railways. In Proceedings of the 2016 National Conference on Electrical, Electronics and Biomedical Engineering (ELECO), Bursa, Turkey, 1–3 December 2016; IEEE: Piscataway, NJ, USA, 2016; pp. 745–750.
  25. Tastimur, C.; Yetis, H.; Karaköse, M.; Akin, E. Rail defect detection and classification with real time image processing technique. Int. J. Comput. Sci. Softw. Eng. 2016, 5, 283.
  26. Chen, Z.; Wang, Q.; Yang, K.; Yu, T.; Yao, J.; Liu, Y.; Wang, P.; He, Q. Deep learning for the detection and recognition of rail defects in ultrasound B-scan images. Transp. Res. Rec. 2021, 2675, 888–901.
  27. Li, H.; Wang, F.; Liu, J.; Song, H.; Hou, Z.; Dai, P. Ensemble model for rail surface defects detection. PLoS ONE 2022, 17, e0268518.
  28. Nandhini, S.; Mohammed Saif, K.V.P.S.E. Robust Automatic Railway Track Crack Detection Using Unsupervised Multi-Scale CNN. Available online: https://ijarsct.co.in/Paper2423.pdf (accessed on 5 May 2022).
More
© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license; additional terms may apply. By using this site, you agree to the Terms and Conditions and Privacy Policy.
This entry is offline, you can click here to edit this entry!
Video Production Service
Feedback