The “digital twin” of the track is no longer a conference concept. A review published in 2026 in the journal Sensors lines up the state of the art of digital twin applied to railway infrastructure maintenance: which sensors are needed, which algorithms, what benefits are already measurable, and—most importantly—what's still missing to move from pilot projects to full scale.
What is a digital twin, in practice?
The authors define the digital twin as an integrated cyber-physical system: a physical asset, its virtual representation, and a continuous, bidirectional exchange of data between the two, enabling real-time monitoring, forecasting, and decision support. Translated onto the track: a digital copy of the line that updates with data collected in the field and allows for the transition from scheduled maintenance to planned maintenance. based on conditions effective.
The data that feeds the model
The system's strength lies in the variety of sources. The review lists accelerometers and vibration sensors for crack detection, temperature sensors, strain gauges for track stresses, ultrasonic sensors for internal rail defects, low-voltage differential pressure transducers (LVDTs), and, for ballast and substructure assessment, ground penetrating radar (GPR) and moisture sensors. Monitoring can be fixed on-line, high precision but localized, or on board the train, with wider spatial coverage but lower localization accuracy.
Machine learning and deep learning algorithms work on this data alongside physical simulations of finite elements and BIM platforms, with convolutional neural networks (CNN) for defect classification and estimates of the residual useful life (Remaining Useful Life) of the components.
Numbers: how much does it yield today?
The picture of the benefits quantified by the collected literature is concrete:
- reduction of unplanned failures 20–40%;
- maintenance delays down 15–30%;
- inspection accuracy better than 25–50% compared to manual methods;
- reduced life cycle costs 10–25% and useful life of the asset extended up to 20%;
- on-board accelerometers with accuracy of 96–100% in detecting corrugations, cracks and geometric irregularities.
The most interesting finding isn't a single number, but the consistency of the direction: every lever—failures, delays, inspections, costs, useful life—moves in the same direction when monitoring becomes continuous and structured.
The limit that remains
However, the revision is also clear on the brakes. Most of the implementations remain conceptual or pilot scale, with limited validation in real-world operating conditions: beyond the 70%, railway IoT applications remain limited to small projects or individual subsystems. The key challenges are the lack of standardization and interoperability, real-time scalability, data governance and security, and the still insufficient integration between multiple sensors and advanced analytics.
The P-WAY reading
There is one point that, in our method, comes before any algorithm: a digital twin is worth as much as the relief from which it arises. Track geometry, dimensions, connections, and constraints are the reference state against which the model measures any deviation over time. If the initial snapshot is inaccurate, even the best predictive model inherits the error.
This is why we insist on a simple sequence—survey, analysis, design, verification—and on traceable and documented data. Predictive maintenance does not replace armament engineering: it provides an additional tool, provided we start from the right place.
Sources: Kazemi MJ, Rashidi M., Kang W.-H., Siahkouhi M., «Toward Smart Railway Infrastructure: Predictive and Optimized Maintenance Through Digital Twin (DT) System», Sensors 2026, 26(8):2333, Western Sydney University; University of Birmingham, «Digitalised Predictive Maintenance in Railways: A Systematic Review of AI, BIM and Digital Twins», Infrastructures 2026.