December 4, 2025 by David Bradley, Inderscience
Collected at: https://techxplore.com/news/2025-12-ai-real-fault-rail.html
Railway infrastructure could be made safer and more reliable using AI, artificial intelligence, according to research published in the International Journal of Information and Communication Technology. The research outlines a new automated, real-time fault detection system based on deep learning that can identify problems with track, bridges, tunnels, and signaling equipment. The work could address long-standing challenges in maintaining complex transportation networks.
Challenges in traditional inspection methods
Faults in railway infrastructure arise through wear and tear, aging, and unexpected failures. Conventional inspection methods remain largely manual and periodic and as such are costly, time-consuming, and prone to human error. This limits their ability to detect problems early. The new AI system can process vast amounts of operational data and quickly identify patterns and anomalies with high precision that can then be followed up by maintenance staff.
One of the major obstacles to the development of automated tools for fault detection has been the scarcity and imbalance of the requisite data. Some types of failures are so rare that training machine-learning models is almost impossible, as there is a dearth of sample data on which to train the model.
Innovative AI techniques for fault detection
The new research tackles this by combining an enhanced Synthetic Minority Over-sampling Technique (ESMOTE) with a class-conditional Generative Adversarial Network (CSGAN). ESMOTE improves data diversity by clustering similar samples and interpolating between them, while CSGAN generates synthetic data that reflects the characteristics of different fault categories. This dual approach creates a more balanced dataset, reducing reliance on expert-labeled data and improving model stability.
Once the data is prepared, the system extracts detailed features from operational signals using a multiscale residual network (ResNet), a type of neural network that captures fine-grained patterns while taking into account variations in operating conditions. A subdomain-adaptive transfer learning strategy allows insights gained from one dataset to be applied to others, enabling accurate fault identification across different environments.
Impact and benefits for railway operators
Tests on the new system gave a diagnostic accuracy of almost 94%, which is better than previous models that struggled with manual feature extraction or unbalanced datasets. The improved precision promises practical benefits for railway operators. Earlier fault detection means limited maintenance resources can be prioritized better and the number of disruptions to service minimized.
More information: Qi An, Intelligent fault diagnosis system for railway infrastructure based on deep learning, International Journal of Information and Communication Technology (2025). DOI: 10.1504/ijict.2025.149991
Leave a Reply