![[2602.16101] Axle Sensor Fusion for Online Continual Wheel Fault Detection in Wayside Railway Monitoring](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16101] Axle Sensor Fusion for Online Continual Wheel Fault Detection in Wayside Railway Monitoring
Summary
This paper presents a novel framework for online continual wheel fault detection in railway systems using axle sensor fusion and machine learning techniques.
Why It Matters
As railways face increasing demands for safety and efficiency, this research addresses the critical need for reliable predictive maintenance. The proposed method enhances fault detection capabilities while adapting to changing operational conditions, which is vital for ensuring safety and minimizing downtime in railway operations.
Key Takeaways
- Introduces a semantic-aware continual learning framework for railway fault diagnostics.
- Utilizes Variational AutoEncoders for unsupervised encoding of accelerometer signals.
- Fuses semantic metadata with VAE embeddings to improve anomaly detection.
- Employs a lightweight gradient boosting classifier for stable anomaly scoring.
- Demonstrates adaptability to evolving operational conditions with minimal labeled data.
Computer Science > Machine Learning arXiv:2602.16101 (cs) [Submitted on 18 Feb 2026] Title:Axle Sensor Fusion for Online Continual Wheel Fault Detection in Wayside Railway Monitoring Authors:Afonso Lourenço, Francisca Osório, Diogo Risca, Goreti Marreiros View a PDF of the paper titled Axle Sensor Fusion for Online Continual Wheel Fault Detection in Wayside Railway Monitoring, by Afonso Louren\c{c}o and 3 other authors View PDF HTML (experimental) Abstract:Reliable and cost-effective maintenance is essential for railway safety, particularly at the wheel-rail interface, which is prone to wear and failure. Predictive maintenance frameworks increasingly leverage sensor-generated time-series data, yet traditional methods require manual feature engineering, and deep learning models often degrade in online settings with evolving operational patterns. This work presents a semantic-aware, label-efficient continual learning framework for railway fault diagnostics. Accelerometer signals are encoded via a Variational AutoEncoder into latent representations capturing the normal operational structure in a fully unsupervised manner. Importantly, semantic metadata, including axle counts, wheel indexes, and strain-based deformations, is extracted via AI-driven peak detection on fiber Bragg grating sensors (resistant to electromagnetic interference) and fused with the VAE embeddings, enhancing anomaly detection under unknown operational conditions. A lightweight gradient boosting supervise...
