A Schematic Plan for Train Position Identification Using Digital Twin and Positioning Sensors
- Open Access
- 2026
- OriginalPaper
- Buchkapitel
Erschienen in:
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by (Link öffnet in neuem Fenster)
Abstract
1 Introduction
Train positioning, while crucial, was not a primary focus in the earlier days of railway operations. The static traffic block system was sufficient to ensure the safety of trains, with margins typically being adequate. However, with the introduction of the European Rail Traffic Management System (ERTMS), the emphasis on train positioning has grown [1, 2]. ERTMS has been instrumental in modernizing railway operations, emphasizing interoperability, and enhancing safety across European rail networks. Additionally, there’s a shifting focus towards railway track and infrastructure condition monitoring using in-service vehicles. This shift underscores the importance of accurate train positioning, especially in a regional context, as efforts to revitalize regional rail intensify and the need to monitor track conditions using onboard sensing devices becomes paramount [3].
Despite the emphasis on accurate train positioning, numerous challenges persist. Trains often operate in harsh railway environments, where they encounter interference, insufficient satellite availability, and signal blockages. These blockages are especially prevalent in specific areas like tunnels, valleys, or due to meteorological conditions. Such challenges have been identified and discussed by several researchers [4, 5]. The quest for precise train positioning often presents a trade-off: opting for high accuracy necessitates expensive equipment and infrastructure, such as advanced communication systems. Conversely, a moderate infrastructure and equipment setup might result in less accurate positioning.
Anzeige
However, technological advancements, particularly in the realms of digital twins and digital mapping, combined with the emergence of cost-effective and efficient positioning sensors, are paving the way for more affordable and efficient train positioning solutions. Vision-based methods, for instance, are being developed for line-side switch rail condition monitoring and inspection, ensuring safer railway operations [6]. Building on this foundation, this paper presents a schematic plan that leverages digital twin technology and onboard train positioning sensors to accurately determine train positions, offering a comprehensive solution to the challenges faced in the realm of train positioning.
2 Methodology
Fig. 1.
The framework of the proposed approach for fine positioning of trains
As depicted in Fig. 1., the proposed method consists of two phases: an offline phase, which involves generating a digital twin of the railway track environment using a model driven approach with Terrestrial Laser Scanning (TLS); an online phase, for which data captured from on-board sensors such as GNSS sensors, an IMU (Inertial Measurement Unit), and cameras, will be combined and matched with the offline digital twin model to obtain the train’s real-time highly accurate locations.
2.1 Offline Step
There are two tasks in the offline phase: defining CityGML schema for railway track and generating digital twins with semantic information for railway tracks in the test area. So far, there exists research work about integrating railway infrastructure into CityGML [7]. However, in their concept, railway is modelled as Level of Detail 1 (LoD1) -like object without further decomposition in semantical and geometric domains. In this work, we will conduct a comprehensive inventory of objects that could appear on railway tracks to collect information about their geometrical, semantic, and attributive characteristics. Then knowledge graphs will be used to describe the semantic and topological relations of objects on railway tracks. On this base, code lists will be defined for all possible objects on railway tracks but following the sequence of from leaves to roots on the knowledge graphs. Finally, UML diagrams will be generated to describe feature classes and their hierarchical relations of objects on railway tracks.
For each object on the leaf node of the knowledge graph, a 3D digital model will be created to build up a library of 3D models for railway track modeling. In our project, TLS will be used to acquire 3D point clouds with high density, full coverage and high accuracy in positioning and geometries. In the next step, semi-automated method will be used to extract objects from 3D point clouds. At the same time. Line segments, circle-formed objects and planes will be detected from point clouds of the objects. Then outlines of the components of an object can be extracted, and further regularized in terms of shapes and topology. Finally, these primitives of objects can be modelled in CityGML. It should be noted that we will follow the mechanisms similar to city furniture modelling in CityGML. In other words, they will be modelled in a local coordinate system with the centroid of the objects as the original of local coordinate system.
Anzeige
After the library of the primitives of 3D models are made available, a digital twin for railway track in the test area could be generated by using model-driven approach from TLS point clouds [8]. The reconstructed 3D models of railway tracks will be stored in CityGML format in the online phase.
2.2 Online Step
In the online step, the train will be equipped with GNSS sensors, an IMU, and at least four cameras at the front, back, and two sides of the train. Data captured from all the on-board sensors will be combined and matched with the offline digital twin model to obtain the train’s real-time highly accurate locations.
Typically, standard GPS systems offer an accuracy of about 5 to 10 m under open skies. However, this accuracy is often not sufficient for most railway operations. To improve accuracy, many railway systems employ Differential GPS (DGPS) or other augmentation systems like RTK (Real-Time Kinematic) GPS. DGPS can improve accuracy to less than a meter, and RTK GPS can provide centimeter-level accuracy. Nevertheless, the adoption of these technologies is associated with significant financial implications. The proposed method avoids the adoption of such expensive sensors while maintaining high positioning accuracy.
First, the GNSS signal will be employed to determine an approximate position of the train. In addition, the train’s IMU data will also be utilized, which provides the train’s orientation and velocity. This IMU data, based on the train’s latest known position, can further refine the current estimated location of the train. Such refinement is crucial for enhancing the accuracy of GNSS data, particularly in scenarios where GNSS signals are compromised, such as in tunnels or valleys, or when signals are completely obstructed due to extreme weather conditions. Utilizing the approximate position derived from both GNSS and IMU inputs, a buffer zone is then created around this position within the offline 3D digital twin.
Next, four cameras, ideally positioned at the train’s front, rear, will capture images of the surrounding environment. Railway track and trackside objects will be extracted from the multi-view images using image detection and segmentation neural networks, and their neighborhood contexts will be established. These objects will be matched with the objects in the buffer zone of the digital twin. The buffer zone can largely reduce the searching area and increase the positioning efficiency.
2.3 Control Environment Testing
Fig. 2.
Test Environment at NTNU
Before implementing this methodology on an actual train, it’s prudent to test it in a controlled environment, primarily due to safety and security concerns. From a safety perspective, the installation of equipment, especially cameras, necessitates strict regulations and approvals from train operators and owners. On the security front, the train’s position, especially if derived from on-board data, may not be readily accessible to external or unauthorized individuals. Thus, it’s advisable to initially test the methodology in a broader setting and subsequently adapt it to the railway domain.
Given the generic nature of the approach, the strategy is to conduct tests in an open environment without compromising safety. A suitable location for this would be the campus of the Norwegian University of Science and Technology (NTNU), as depicted in Fig. 2.
The campus environment offers a variety of features that can simulate railway conditions. Buildings in close proximity can mimic valleys, building underpass can represent tunnels or enclosed areas, and open spaces can serve as typical open terrains. Specific landmarks and objects within this environment will be positioned in the digital twin for accuracy.
To simulate the train, a remote-controlled car will be employed. This car will be equipped with the same sensors as those on the train (i.e., GNSS sensors, cameras, and an IMU). First, an offline 3D digital twin of the test site will be generated using the methods described in Sect. 2.1. Next, the car will traverse the campus, while the data captured by all its sensors will be combined and matched with the digital twin to obtain the car’s real-time locations. The algorithm will be continuously tested and revised until the positional accuracies achieve satisfactory.
3 Potential Challenges
Several challenges can be anticipated at this stage:
Monotonous Objects:
While railway tracks are lined with numerous objects, many of these are repetitive in nature. For instance, the shapes of rails and sleepers remain consistent over long stretches. Relying solely on images for object identification can be problematic when environmental differences are minimal. Hence, it’s crucial to incorporate other data sources, such as velocity, train position, and high-frequency trackside objects.
Weather Conditions:
Conditions like heavy rain, snowfall, fog, or a snow-covered environment can obstruct or degrade camera visibility, compromising image quality for identification. In such scenarios, alternative data sources or methods should be considered to address these challenges.
Difficult Locations:
Locations like tunnels and valleys are notorious for obstructing GPS signals. Additionally, these areas are often dimly lit, further complicating positioning efforts.
Lighting:
Proper lighting is essential for accurate camera image identification. In low-light conditions, the accuracy of train location identification may be compromised. Potential solutions could include using cameras capable of functioning in low light or installing additional lighting on the vehicle.
Odometer Data:
The resolution of odometer data logging can vary, presenting challenges if there’s significant deviation. Different odometers have varying logging frequencies, which can affect the accuracy of positioning.
Communication System:
Transmitting real-time position data to a central system can be challenging, especially when commercial communication infrastructures are unavailable, particularly in regional contexts. It’s essential to process and compress position data to ensure it’s transferable, even though systems like GSM-R.
System Integration:
Once developed, system compatibility might pose challenges. The on-board system could differ from the testing version. If the system is to be implemented on-board, seamless integration with existing systems is vital to avoid substantial investments. Ideally, a plug-and-play solution would be most beneficial in this context.
4 Expected Outcomes
Through this methodology, the ambition is to pinpoint train positions with a precision of ±25 cm. This level of accuracy is a marked improvement over certain maintenance vehicles, which currently have positioning accuracies more than ±15 m. Another anticipated outcome is the potential for this methodology to be seamlessly implemented across all trains in Europe. The ideal scenario would be a plug-and-play solution, subject, of course, to rigorous security checks and full compliance with established standards.
Acknowledgements
This research project is funded by Norwegian Railway Directorate (NRD) and Europe’s Rail Joint Undertaking (grant number 101101962-FP6-FutuRe-HORIZON-ER-JU-2022-01).
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
