Axle Sensor Fusion for Online Continual Wheel Fault Detection in Wayside Railway Monitoring
arXiv:2602.16101v1 Announce Type: new 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 dete
arXiv:2602.16101v1 Announce Type: new 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 supervised classifier stabilizes anomaly scoring with minimal labels, while a replay-based continual learning strategy enables adaptation to evolving domains without catastrophic forgetting. Experiments show the model detects minor imperfections due to flats and polygonization, while adapting to evolving operational conditions, such as changes in train type, speed, load, and track profiles, captured using a single accelerometer and strain gauge in wayside monitoring.
Executive Summary
This article presents an innovative, semantic-aware, label-efficient continual learning framework for railway fault diagnostics. By leveraging Variational AutoEncoder (VAE) embeddings and AI-driven peak detection on fiber Bragg grating sensors, the framework enhances anomaly detection under unknown operational conditions. The model adapts to evolving domains without catastrophic forgetting, while a lightweight gradient boosting supervised classifier stabilizes anomaly scoring with minimal labels. Experiments demonstrate the model's effectiveness in detecting minor imperfections due to flats and polygonization, as well as adapting to changes in train type, speed, load, and track profiles. This work has significant implications for railway safety and maintenance, as it enables predictive maintenance frameworks to become more reliable and cost-effective.
Key Points
- ▸ The proposed framework uses VAE embeddings and AI-driven peak detection to enhance anomaly detection under unknown operational conditions.
- ▸ The model adapts to evolving domains without catastrophic forgetting, using a replay-based continual learning strategy.
- ▸ Experiments demonstrate the model's effectiveness in detecting minor imperfections due to flats and polygonization.
Merits
Strength in Anomaly Detection
The framework's use of VAE embeddings and AI-driven peak detection enables enhanced anomaly detection under unknown operational conditions, making it a significant improvement over traditional methods.
Efficient Adaptation to Evolving Domains
The replay-based continual learning strategy enables the model to adapt to evolving domains without catastrophic forgetting, making it a valuable addition to predictive maintenance frameworks.
Demerits
Limited Sensor Deployment
The framework relies on a single accelerometer and strain gauge in wayside monitoring, which may limit its applicability to scenarios requiring more comprehensive sensor deployment.
Potential Interference from Electromagnetic Sources
The use of fiber Bragg grating sensors, while resistant to electromagnetic interference, may still be susceptible to interference from other sources, potentially affecting the accuracy of the model.
Expert Commentary
This article presents a novel and innovative approach to railway fault diagnostics, leveraging the power of Variational AutoEncoder embeddings and AI-driven peak detection. The framework's ability to adapt to evolving domains without catastrophic forgetting is a significant contribution to the field of continual learning in deep learning. However, the framework's reliance on a single accelerometer and strain gauge in wayside monitoring may limit its applicability to scenarios requiring more comprehensive sensor deployment. Overall, this work has significant implications for railway safety and maintenance, and its contributions to the field of predictive maintenance and continual learning are substantial.
Recommendations
- ✓ Future research should focus on exploring the applicability of the proposed framework to scenarios requiring more comprehensive sensor deployment.
- ✓ Investigating the potential interference from electromagnetic sources on the accuracy of the model is crucial to ensure its reliability in real-world applications.
