Journals / Demiryolu Mühendisliği / 2026 / Sayı: 25

Vibration-Based Anomaly Detection in High-Speed Railway Infrastructure

Yüksek Hızlı Demiryolu Altyapısında Titreşim Tabanlı Anomali Tespiti

Mustafa Tasci(BANDIRMA ONYEDI EYLUL UNIVERSITY, GÖNEN VOCATIONAL SCHOOL)0000-0002-8073-8587
Pages
—
Publication Date
2026-08-01
🏛️Bibliographic metadata and full text access for this article are provided via DergiPark.
DergiPark Record ↗
✨

AI Summary & Key Takeaways

AI Insight
⚡ TL;DR(Too Long; Didn't Read)

Event aggregation condensed 10,681 anomalies into 21 GPS-referenced events, producing a prioritised maintenance map and confirming that reliable, spatially precise detection is achievable without field annotation.

📌 3 Key Takeaways

🎯
Objective & Hypothesis:Geometry deviations on high-speed railway lines amplify dynamic wheel–rail loads and accelerate track wear and component failure.
📊
Key Findings & Data:This study presents a fully unsupervised, nine-stage pipeline that converts raw onboard inertial measurement unit (IMU) and Global Positioning System (GPS) data, collected at about 250 km/h on the Madrid–Barcelona line, into spatially referenced anomaly indicators.
💡
Conclusion & Impact:Event aggregation condensed 10,681 anomalies into 21 GPS-referenced events, producing a prioritised maintenance map and confirming that reliable, spatially precise detection is achievable without field annotation.
🔬Design:Modeling & Simulation
📍Scope:Comprehensive Sample
⚙️Variables:Anomali tespiti · Demiryolu bakım · Isolation Forest

Abstract

Geometry deviations on high-speed railway lines amplify dynamic wheel–rail loads and accelerate track wear and component failure. Continuous, vehicle-based vibration monitoring is a practical route to condition-based maintenance, yet the scarcity of labelled fault data limits supervised methods. This study presents a fully unsupervised, nine-stage pipeline that converts raw onboard inertial measurement unit (IMU) and Global Positioning System (GPS) data, collected at about 250 km/h on the Madrid–Barcelona line, into spatially referenced anomaly indicators. The pipeline combines data-driven calibration, spectral pre-analysis, sliding-window root mean square (RMS) estimation, median absolute deviation (MAD) based thresholding, crest-factor subtyping, and event aggregation. Seven features were analysed in parallel by four detectors: Isolation Forest (IF), Local Outlier Factor (LOF), One-Class Support Vector Machine (OC-SVM), and a one-dimensional convolutional neural network autoencoder (1D-CNN AE). As no ground-truth labels exist, performance was assessed using RMS overlap, a Cohen-d-like separation score, and the area under the receiver operating characteristic curve (ROC-AUC). IF gave the sharpest separation (ROC-AUC = 0.9722), while the 1D-CNN AE reached the highest RMS overlap (59.6%). A rank-normalised, ROC-AUC-weighted soft-voting ensemble of IF and the 1D-CNN AE outperformed all individual models (RMS overlap = 72.4%; ROC-AUC = 0.9861). Event aggregation condensed 10,681 anomalies into 21 GPS-referenced events, producing a prioritised maintenance map and confirming that reliable, spatially precise detection is achievable without field annotation.

Özet

Yüksek hızlı demiryolu hatlarındaki geometri sapmaları, dinamik tekerlek–ray yüklerini artırarak hat aşınmasını ve bileşen arızalarını hızlandırır. Araç üstü sürekli titreşim izleme, duruma dayalı bakım için pratik bir yol sunsa da, etiketli arıza verisinin kıtlığı gözetimli yöntemleri sınırlamaktadır. Bu çalışma, Madrid–Barselona hattında yaklaşık 250 km/s hızla toplanan ham ataletsel ölçüm birimi (IMU) ve Küresel Konumlama Sistemi (GPS) verisini, mekânsal olarak etiketlenmiş anomali göstergelerine dönüştüren tümüyle gözetimsiz dokuz aşamalı bir işlem hattı sunar. Hat; veri odaklı kalibrasyon, spektral ön analiz, kayan pencere kök ortalama kare (RMS) hesabı, medyan mutlak sapma (MAD) tabanlı eşikleme, tepe faktörü alt tiplemesi ve olay birleştirme adımlarını içerir. Yedi öznitelik, dört algoritmayla paralel çözümlenmiştir: Isolation Forest (IF), Local Outlier Factor (LOF), Tek Sınıf Destek Vektör Makinesi (OC-SVM) ve tek boyutlu evrişimli sinir ağı otokodlayıcısı (1D-CNN AE). Gerçek referans etiket bulunmadığından başarım; RMS örtüşmesi, Cohen-d benzeri ayrım skoru ve alıcı işletim karakteristiği eğrisi altındaki alan (ROC-AUC) ile değerlendirilmiştir. IF en keskin ayrımı (ROC-AUC = 0,9722), 1D-CNN AE ise en yüksek RMS örtüşmesini (%59,6) sağlamıştır. IF ile 1D-CNN AE'nin sıralama-normalize, ROC-AUC ağırlıklı yumuşak oylamalı topluluğu, tüm bireysel modelleri geride bırakmıştır (RMS örtüşmesi %72,4; ROC-AUC = 0,9861). Olay birleştirme, 10.681 anomaliyi 21 GPS etiketli olaya indirgeyerek önceliklendirilmiş bir bakım haritası üretmiş ve güvenilir, mekânsal olarak hassas tespitin saha etiketi olmadan mümkün olduğunu doğrulamıştır.

🔖

Cite This Article

Tasci, M. (2026). Vibration-Based Anomaly Detection in High-Speed Railway Infrastructure. Demiryolu Mühendisliği, Sayı: 25. https://doi.org/10.47072/demiryolu.1965385

📄 Full Text PDF Viewer

📄

PDF görüntüleyici hazırlanıyor...

References (29)

  1. A. Jamshidi, A. Núñez, R. Dollevoet, ve Z. Li, “Robust and predictive fuzzy key performance indicators for condition-based treatment of squats in railway infrastructure,” Journal of Infrastructure Systems, c.23, sy.3, 2017, doi: 10.1061/(ASCE)IS.1943-555X.0000357.
  2. M. A. Sayeed, M. A. Shahin, “Three-dimensional numerical modelling of ballasted railway track foundations for high-speed trains with special reference to critical speed”, Transportation Geotechnics, c. 6, sy. 2, ss. 55–65, 2016, doi: 10.1016/j.trgeo.2016.01.003.
  3. M. Arucu, “Magnetic levitation and intelligent transportation systems: Superconductivity and electrodynamics in maglev trains”, Jitsa, c. 8, sy. 2, ss. 70–87, 2025, doi: 10.51513/jitsa.1721729.
  4. H. Tsunashima, R. Honda, ve A. Matsumoto, “Track condition monitoring based on in-service train vibration data using smartphones,” New Research on Railway Engineering and Transportation, A. G. Hessami ve R. Muttram, Ed. London, UK: IntechOpen, 2023, doi: 10.5772/intechopen.111703.
  5. J. Liu, F. Liu, W. Xie, ve X. Ju, "Track vibration sequence anomaly detection algorithm based on LSTM," Advances in Structural Engineering, c. 26, sy. 9, ss. 1682–1695, 2023, doi: 10.1177/13694332231161104.
  6. B. Bergquist ve P. Söderholm, “Data analysis for condition-based railway infrastructure maintenance,” Qual. Reliab. Eng. Int., c. 31, sy. 5, ss. 773–781, 2015, doi: 10.1002/qre.1634.
  7. W. Wang, W. Hu, W. Wang, X. Xu, M. Wang, Y. Shi, S. Qiu, ve E. Tutumluer, "Automated crack severity level detection and classification for ballastless track slab using deep convolutional neural network," Autom. Constr., c. 124, ss. 103484, 2021, doi: 10.1016/j.autcon.2020.103484.
  8. J. Sresakoolchai ve S. Kaewunruen, "Railway defect detection based on track geometry using supervised and unsupervised machine learning," Structural Health Monitoring, c. 21, sy. 4, ss. 1757–1767, 2022, doi: 10.1177/14759217211044492.
  9. C. Esveld, Modern Railway Track, 2nd ed. Zaltbommel, The Netherlands: MRT-Productions, 2001.
  10. G. Lederman, S. Chen, J. H. Garrett, J. Kovačević, H. Y. Noh, ve J. Bielak, “Track monitoring from the dynamic response of an operational train,” Mech. Syst. Signal Process., c. 87, ss. 1–16, 2017, doi: 10.1016/j.ymssp.2016.06.041.
  11. M. Molodova, Z. Li, A. Núñez, ve R. Dollevoet, “Automatic detection of squats in railway infrastructure,” IEEE Transactions on Intelligent Transportation Systems, c. 15, sy. 5, ss. 1980–1990, 2014, doi: 10.1109/TITS.2014.2307955.
  12. R. B. Randall, Vibration-based Condition Monitoring: Industrial, Aerospace and Automotive Applications. Chichester, UK: Wiley, 2011. doi: 10.1002/9780470977668.
  13. Y. Sun, G. Xie, Y. Cao, ve T. Wen, “Strategy for fault diagnosis on train plug doors using audio sensors,” Sensors, c. 19, sy. 1, ss. 3, 2018, doi: 10.3390/s19010003.
  14. T. Ince, S. Kiranyaz, L. Eren, M. Askar, ve M. Gabbouj, “Real-time motor fault detection by 1-D convolutional neural networks,” IEEE Transactions on Industrial Electronics, c. 63, sy. 11, ss. 7067–7075, 2016, doi: 10.1109/TIE.2016.2582729.
  15. A. Akkaya, “Aşırı sınıf dengesizliği altında ince doğrusal yapıların segmentasyonu: Demiryolu veri setlerinde doğruluk paradoksu ve mimari verimlilik”, Demiryolu Mühendisliği, sy. 24, 2026, doi: 10.47072/demiryolu.1910383.
  16. A. Malekjafarian, C.-A. Sarrabezolles, M. A. Khan, ve F. Golpayegani, "A machine-learning-based approach for railway track monitoring using acceleration measured on an in-service train," Sensors, c. 23, sy. 17, ss. 7568, 2023, doi: 10.3390/s23177568.
  17. F. T. Liu, K. M. Ting, ve Z.-H. Zhou, “Isolation forest,” in Proceedings of the 8th IEEE International Conference on Data Mining (ICDM), 2008, ss. 413–422. doi: 10.1109/ICDM.2008.17.
  18. M. M. Breunig, H.-P. Kriegel, R. T. Ng, ve J. Sander, “LOF: Identifying density-based local outliers,” in Proceedings of the 2000 ACM SIGMOD International Conference on Management of Data, 2000, ss. 93–104. doi: 10.1145/342009.335388.
  19. H. J. Shin, D.-H. Eom, ve S.-S. Kim, "One-class support vector machines—an application in machine fault detection and classification," Comput. Ind. Eng., c. 48, sy. 2, ss. 395–408, 2005, doi: 10.1016/j.cie.2005.01.009.
  20. M. Sakurada ve T. Yairi, “Anomaly detection using autoencoders with nonlinear dimensionality reduction,” in Proceedings of the MLSDA 2014 2nd Workshop on Machine Learning for Sensory Data Analysis, 2014, ss. 4–11. doi: 10.1145/2689746.2689747.
  21. Y. Zhang, Y. Chen, J. Wang, ve Z. Pan, "Unsupervised deep anomaly detection for multi-sensor time-series signals," IEEE Trans. Knowl. Data Eng., c. 35, sy. 2, ss. 2118–2132, 2023, doi: 10.1109/TKDE.2021.3102110.
  22. M. J. Gomez, C. Castejon, E. Corral, ve M. Cocconcelli, "Railway axle early fatigue crack detection through condition monitoring techniques," Sensors, c. 23, sy. 13, ss. 6143, 2023, doi: 10.3390/s23136143.
  23. İ. Aydın, T. K. Şener, ve M. Sevi, “Demiryolu ray ve çevresinin anlamlandırılması için derin öğrenme tabanlı yöntemlerin karşılaştırmalı analizi”, Demiryolu Mühendisliği, sy. 19, ss. 1–16, Oca. 2024, doi: 10.47072/demiryolu.1336812.
  24. B. Li, L. Tan, F. Wang, ve L. Liu, "A railway intrusion detection method based on decomposition and semi-supervised learning for accident protection," Accident Analysis & Prevention, c. 189, ss. 107124, 2023, doi: 10.1016/j.aap.2023.107124.
  25. M. Memarzadeh, B. Matthews, T. Templin, A. S. Rohani, ve D. Weckler, "Semi-supervised active learning for anomaly detection in aviation," Journal of Aerospace Information Systems, c. 20, sy. 4, ss. 181–194, 2023, doi: 10.2514/1.I011083.
  26. A. Zimek, R. J. G. B. Campello, ve J. Sander, “Ensembles for unsupervised outlier detection: Challenges and research questions,” ACM SIGKDD Explorations Newsletter, c. 15, sy. 1, ss. 11–22, 2014, doi: 10.1145/2594473.2594476.
  27. S. Wang, F. Liu, ve B. Liu, "Semi-supervised deep learning in high-speed railway track detection based on distributed fiber acoustic sensing," Sensors, c. 22, sy. 2, ss. 413, 2022, doi: 10.3390/s22020413.
  28. F. J. Bris-Peñalver, R. Verdecia-Peña, ve J. I. Alonso, "A survey of AI-enabled predictive maintenance for railway infrastructure: Models, data sources, and research challenges," Sensors, c. 26, sy. 3, ss. 906, 2026, doi: 10.3390/s26030906.
  29. A. L. Bustamante ve J. T. Castán, “Train track environmental data for predictive maintenance”. Zenodo, Nov. 14, 2025. doi: 10.5281/zenodo.17607068, [Erişim: 08 Haz. 2026].