Digital twins for rail damage evolution in railway curves - Field data, model calibration and reduced-order models
Doctoral thesis, 2025
Rails in railway tracks are subjected to complex rolling-sliding contact conditions that can cause surface and subsurface damage in the form of wear, plastic deformation, and Rolling Contact Fatigue (RCF). The prevailing operational conditions and maintenance practices of the rails strongly influence the evolution of this damage. If not managed properly, these mechanisms can degrade vehicle steering performance, increase noise and vibration levels, and in severe cases lead to rail failure or derailment. However, maintenance is costly and is difficult to plan correctly. Therefore, the ability to predict damage evolution is essential for effective maintenance planning and extending the rails' service life.
This thesis presents a digital twin framework for the long-term prediction of rail damage evolution in railway curves under operational traffic conditions. The framework integrates field measurements of rail profiles with numerical simulations using models for the mechanical behavior of the rails. This enables continuous calibration and updating of model parameters based on observed rail behavior. The simulations are performed using models of dynamic vehicle-track interactions and accumulative rail damage computations considering plastic deformation, wear, and surface RCF. The calibration process adjusts parameters related to vehicle, track, contact model, and material properties, such as wear coefficients or yield limits, to ensure that the simulated degradation mimics the field-measured rail profile evolution. After calibration, the digital twins can predict railhead damage more accurately.
To speed up numerical efficiency, a reduced order modeling framework has been developed for 3D rail geometry and plastic deformations. The framework adopts a convective coordinate system that follows a moving contact load and assumes a steady-state response. This transforms the transient moving contact problem into a stationary formulation, where the material history is tracked along streamlines in the spatial mesh. An iterative scheme couples the displacement and plastic strain fields. In addition, the Proper Generalized Decomposition (PGD) has been adopted to provide an efficient representation of the 3D displacement field through domain decomposition and parametrization of the contact loads. The framework achieves accuracy comparable to that of a full 3D finite-element analysis at a fraction of the cost.
The resulting digital twins efficiently and accurately forecast profile geometry changes of railheads in curved tracks. Thus, this work demonstrates the feasibility of using digital twins to predict rail damage evolution and provides a step towards building a foundation for data-informed maintenance planning.
This thesis presents a digital twin framework for the long-term prediction of rail damage evolution in railway curves under operational traffic conditions. The framework integrates field measurements of rail profiles with numerical simulations using models for the mechanical behavior of the rails. This enables continuous calibration and updating of model parameters based on observed rail behavior. The simulations are performed using models of dynamic vehicle-track interactions and accumulative rail damage computations considering plastic deformation, wear, and surface RCF. The calibration process adjusts parameters related to vehicle, track, contact model, and material properties, such as wear coefficients or yield limits, to ensure that the simulated degradation mimics the field-measured rail profile evolution. After calibration, the digital twins can predict railhead damage more accurately.
To speed up numerical efficiency, a reduced order modeling framework has been developed for 3D rail geometry and plastic deformations. The framework adopts a convective coordinate system that follows a moving contact load and assumes a steady-state response. This transforms the transient moving contact problem into a stationary formulation, where the material history is tracked along streamlines in the spatial mesh. An iterative scheme couples the displacement and plastic strain fields. In addition, the Proper Generalized Decomposition (PGD) has been adopted to provide an efficient representation of the 3D displacement field through domain decomposition and parametrization of the contact loads. The framework achieves accuracy comparable to that of a full 3D finite-element analysis at a fraction of the cost.
The resulting digital twins efficiently and accurately forecast profile geometry changes of railheads in curved tracks. Thus, this work demonstrates the feasibility of using digital twins to predict rail damage evolution and provides a step towards building a foundation for data-informed maintenance planning.
dynamic vehicle-track interaction
Rolling Contact Fatigue (RCF)
Proper Generalized Decomposition (PGD)
railway curves
plasticity
digital twin
Reduced-Order Modeling (ROM)
Steady state in rolling contact
wear
Virtual Development Labrotory, Chalmers Tvärgata 4C, Gothenburg
Opponent: Pedro Díez, Universitat Politènica de Catalunya, Spain
Author
Caroline Ansin
Chalmers, Industrial and Materials Science, Material and Computational Mechanics
Simulation and Field Measurements of the Long-term Rail Surface Damage due to Plasticity, Wear and Surface Rolling Contact Fatigue Cracks in a Curve
CM 2022 - 12th International Conference on Contact Mechanics and Wear of Rail/Wheel Systems, Conference Proceedings,;(2022)p. 591-601
Paper in proceeding
Influence of model parameters on the predicted rail profile wear distribution in a curve
Simulations and field measurements of long-term rail profile wear in curves
Other conference contribution
Fast simulation of 3D elastic response for wheel–rail contact loading using Proper Generalized Decomposition
Computer Methods in Applied Mechanics and Engineering,;Vol. 417(2023)
Journal article
Prediction of evolving plasticity in rails under steady state rolling contact based on Reduced-Order Modeling
Computer Methods in Applied Mechanics and Engineering,;Vol. 438(2025)
Journal article
When a train rolls over a track, the contact patch between each wheel and rail is no larger than the size of a thumbnail, yet that tiny area carries the weight of five to ten tonnes. Under such extreme loads, the rail steel gradually wears down, permanently deforms, and can develop small fatigue cracks. If this damage is not properly maintained, it can lead to costly repairs or safety risks. Therefore, maintenance must be carefully planned and performed at the right time to keep the railway infrastructure reliable and cost-efficient.
This thesis investigates how digital twins can be used to forecast rail damage in curved tracks. A digital twin is a virtual copy of a physical object, in this case, a section of a curved railway track, which combines measurement data from the field with advanced computer simulations. By linking data on rail profiles, grinding interventions, and traffic information with computational models, the virtual copy can use that information to forecast how the rail will evolve under various operational conditions. The results can then support maintenance decisions that balance safety, cost, and impact on traffic.
To truly understand how the rail geometry changes over time requires both reliable field data and accurate computational models. The measurements must cover traffic conditions, maintenance actions, and rail profiles. Simulations should capture the interaction between the vehicle and the rail and the resulting damage mechanisms such as wear, permanent deformation, and material fatigue. Because many train passages must be represented, the calculations must be highly efficient, which is achieved by simplifying complex models.
The digital twins developed in this thesis demonstrate that combining measurement data, simulations, and improving computational models through calibration can give reliable predictions of rail damage for many train passages. This marks a step toward integrating digital twins into future railway maintenance planning.
This thesis investigates how digital twins can be used to forecast rail damage in curved tracks. A digital twin is a virtual copy of a physical object, in this case, a section of a curved railway track, which combines measurement data from the field with advanced computer simulations. By linking data on rail profiles, grinding interventions, and traffic information with computational models, the virtual copy can use that information to forecast how the rail will evolve under various operational conditions. The results can then support maintenance decisions that balance safety, cost, and impact on traffic.
To truly understand how the rail geometry changes over time requires both reliable field data and accurate computational models. The measurements must cover traffic conditions, maintenance actions, and rail profiles. Simulations should capture the interaction between the vehicle and the rail and the resulting damage mechanisms such as wear, permanent deformation, and material fatigue. Because many train passages must be represented, the calculations must be highly efficient, which is achieved by simplifying complex models.
The digital twins developed in this thesis demonstrate that combining measurement data, simulations, and improving computational models through calibration can give reliable predictions of rail damage for many train passages. This marks a step toward integrating digital twins into future railway maintenance planning.
Subject Categories (SSIF 2025)
Applied Mechanics
DOI
10.63959/chalmers.dt/5777
ISBN
978-91-8103-320-5
Doktorsavhandlingar vid Chalmers tekniska högskola. Ny serie: 5777
Publisher
Chalmers
Virtual Development Labrotory, Chalmers Tvärgata 4C, Gothenburg
Opponent: Pedro Díez, Universitat Politènica de Catalunya, Spain