A method for rapidly predicting rolling contact fatigue damage of a steel rail

Abstract

This invention discloses a rapid prediction method for rail rolling contact fatigue damage in the field of rail damage prediction technology. The method includes establishing a three-dimensional elastoplastic wheel-rail cyclic rolling contact finite element model, improving the cyclic plastic constitutive model, and simulating the stress-strain evolution of the rail. A rail stress-strain sample library is generated through parametric sampling and batch calculation. A Physically Constrained Generative Adversarial Network (PC-GAN) is constructed, embedding physical constraints such as contact force conservation and energy consistency into the loss function to achieve rapid and high-precision mapping from operating parameters to the stress-strain field. Based on the PC-AN prediction results, cumulative damage and fatigue life are calculated using an incremental multiaxial fatigue damage model, and a damage distribution map of the entire rail line is drawn, identifying high-risk sections. This invention deeply integrates physical modeling with intelligent algorithms, achieving an order-of-magnitude improvement in rail fatigue damage assessment efficiency while ensuring prediction accuracy, providing efficient and reliable technical support for rail life assessment and maintenance decisions across the entire rail line.

Description

Technical Field

[0001] This invention relates to the field of rail damage prediction technology, specifically a method for rapid prediction of rail rolling contact fatigue damage. Background Technology

[0002] Rolling contact fatigue of rails is one of the key issues affecting railway operation safety and track service life. Under wheel-rail cyclic loading, fatigue cracks easily initiate on the surface and subsurface of the rail, which can then develop into damage such as spalling and chipping. In severe cases, this can worsen wheel-rail dynamic interaction, cause abnormal vibration and noise, and significantly increase track maintenance costs. Therefore, accurate and efficient prediction of rolling contact fatigue damage of rails is of great engineering significance.

[0003] Currently, the method based on elastoplastic finite element simulation can accurately simulate the stress-strain field of wheel-rail contact and its evolution under cyclic loading, and combine the material cyclic plasticity model with multiaxial fatigue criteria to achieve damage assessment.

[0004] However, this method is computationally expensive, especially when performing full-line, multi-condition, and long-cycle simulations, making it difficult to meet the analysis efficiency requirements in actual engineering practice.

[0005] In recent years, data-driven methods have provided a new approach for the rapid prediction of mechanical responses. However, purely data-based models often lack clear physical constraints and have limited generalization ability under working conditions outside the training samples.

[0006] Based on this, the present invention designs a rapid prediction method for rolling contact fatigue damage of rails to solve the above problems. Summary of the Invention

[0007] The purpose of this invention is to provide a rapid prediction method for rail rolling contact fatigue damage, enabling efficient and high-precision assessment of rail fatigue damage and life under complex working conditions across the entire railway line, thereby addressing the shortcomings of existing methods in the background art in balancing prediction accuracy and computational efficiency.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for rapid prediction of rolling contact fatigue damage in rails includes the following steps: S1. Construct a three-dimensional elastoplastic wheel-rail cyclic rolling contact finite element model, and use this model to simulate and obtain the stress-strain response of the rail under different working conditions. S2. Based on the finite element model in S1, perform automated calculations for multiple working conditions to generate sample data of rail stress / strain field and construct a structured dataset for network training. S3. Build and train a conditional generative adversarial network based on physical constraints, and use multiple working condition parameters as network inputs to quickly predict the local stress / strain field distribution of the rail under the corresponding working condition. S4. Based on the rail stress / strain field predicted in S3, perform multiaxial fatigue damage calculation and life assessment, and analyze the damage distribution of the entire line.

[0009] Further, step S1 includes: S1-1, Wheel-rail geometry and contact modeling: A three-dimensional geometric model containing the true contours of the wheel tread and the rail top is established to accurately characterize the shape and geometric transition features of the wheel-rail contact interface. The elastoplastic finite element model can accurately simulate the ratcheting effect and plastic strain accumulation of materials under cyclic loads, ensuring the physical authenticity of the sample data. S1-2, Structural model of wheel-rail material: Based on experimental data of rail material mechanics, the classic Adel-Karim-Ohno cyclic plastic constitutive model was improved by introducing isotropic deformation resistance parameters and ratchet parameters, and combined with Tanaka non-proportional parameters to characterize the cyclic characteristics of rail and ratchet behavior under multi-axis cyclic loading. The UMAT subroutine was written and embedded into the finite element analysis software, in which the wheel is made of elastic material. S1-3, Simulation of wheel-rail cyclic rolling: The DISP and URDFIL subroutines control wheel movement and load, enabling multi-step sequential tasks such as rolling, rewinding, reloading, and releasing. The UMESHMOTION subroutine considers wear effects to capture the cyclic evolution of rail stress / strain. S1-4, Model Validation: By comparing the finite element calculation results with theoretical solutions, experimental data, and literature results, the multi-level prediction capability of the wheel-rail contact model was systematically verified, from elastic contact behavior and constitutive model accuracy to cyclic rolling contact response.

[0010] Furthermore, the wheel tread is reduced to a super element using substructure reduction, and motion simulation is achieved through a UEL user subroutine; the rail top is extended to form a shadow element and periodic boundary conditions are applied to ensure the continuity of displacement and stress; the wheel-rail contact relationship is established using a surface-to-surface contact algorithm, a penalty function method, and a Coulomb friction model.

[0011] Further, step S2 includes: S2-1. Typical Operating Condition Parameter Design and Automated Finite Element Calculation: A multidimensional parameter space for typical working conditions containing variables such as vehicle operating parameters and track structure parameters was constructed. Parameters were sampled using the Latin hypercube sampling method, and rail stress / strain field data were generated in batches using automated finite element calculation scripts, achieving efficient sample generation. S2-2, Data Extraction and Processing: The script automatically extracts and cleans the corresponding finite element calculation results such as contact pressure, equivalent stress, and equivalent plastic strain. After coordinate unification and normalization, a structured high-quality data table is formed. S2-3, Sample set construction: Key feature labels were added to each sample, including maximum contact pressure and location, contact patch size, maximum equivalent stress and location, and maximum equivalent plastic strain and location, to construct a structured sample library corresponding to input and output, providing a high-fidelity training dataset for subsequent generative adversarial networks.

[0012] Furthermore, in step S3, the physical constraint-based conditional generative adversarial network has a generator that uses a multi-branch convolutional structure to extract and fuse the geometric and load features from the input working condition parameters; and a discriminator that uses a PatchGAN structure to judge the local authenticity of the generated stress / strain field.

[0013] Furthermore, in step S3, when training the conditional generative adversarial network, multiple physical constraints are introduced into the loss function, including contact force conservation constraints, energy consistency constraints, and boundary condition matching constraints.

[0014] Furthermore, the multiaxial fatigue damage calculation in step S4 includes: S4-1. Transform the predicted stress / strain field data from the global coordinate system to the potential critical planes of the material points using a rotation matrix. S4-2. An incremental multiaxial fatigue damage model is used to calculate the plastic strain energy increment and fatigue damage increment of each potential critical plane in each load cycle, and the cumulative fatigue damage is obtained by integration. S4-3. Based on the critical plane method, find the maximum cumulative damage and its corresponding plane and location, and predict the fatigue life of the rail according to the critical damage value criterion.

[0015] Furthermore, the full-line damage distribution analysis in step S4 includes: S4-4. Based on the operating characteristics of different line sections, the predicted damage results are spatially mapped and superimposed. S4-5. By using segmented statistics, Kriging interpolation, and spatial weighting algorithms, a rail fatigue damage distribution map of the entire line is constructed. S4-6. Conduct parameter sensitivity analysis to identify the dominant damage factors and identify potential high-risk fatigue zones based on the damage distribution map.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention achieves a leap from time-consuming high-precision simulation to "second-level" intelligent prediction by deeply integrating a high-fidelity elastoplastic finite element model with a physical constraint generative adversarial network (PC-GAN). This method achieves prediction accuracy comparable to finite element methods while significantly improving computational efficiency by orders of magnitude, greatly reducing analysis costs and resolving the core contradiction of traditional methods in balancing efficiency and accuracy. 2. This invention ensures that the prediction results strictly conform to physical laws by embedding physical constraints such as contact mechanics and energy conservation into the artificial intelligence training. Furthermore, the subsequent fatigue damage model based on the critical plane method makes the entire process from operating condition input to lifespan output logically clear and the results reliable, providing a credible basis for engineering decision-making. 3. This method can not only quickly predict single-point damage, but also has the macroscopic analysis capability to generate a fatigue risk distribution map of the entire rail line, which can directly serve the accurate decision-making of maintenance. Moreover, the framework of this invention is universal and can be extended to other key components such as wheels and turnouts, providing an innovative technical path for intelligent condition inspection of rail transit infrastructure. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a roadmap of the overall research technology of this invention; Figure 2 A technical roadmap for a three-dimensional elastoplastic wheel-rail cyclic rolling contact finite element model; Figure 3 This is a roadmap for rapid prediction of rail stress / strain fields based on PC-GAN. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see the appendix Figure 1 This invention provides a method for rapid prediction of rolling contact fatigue damage in rails, comprising: A high-precision and high-efficiency three-dimensional elastoplastic wheel-rail cyclic rolling contact finite element model was constructed, and a high-fidelity rail stress / strain field sample library covering typical working conditions was calculated and generated. Establish a PC-GAN for wheel-rail rolling contact to achieve efficient generation and rapid prediction of rail stress / strain field across the entire line; Based on the incremental multiaxial fatigue damage model, the cumulative damage characteristics of materials are calculated. Combined with the track characteristics and vehicle operating conditions, the rail damage distribution and life statistics of different sections of the entire track are obtained, providing a quantitative decision-making basis for track maintenance and life management.

[0021] By achieving the above objectives, the project will break through the computational efficiency bottleneck in predicting rail contact fatigue damage, and realize a full-line fatigue risk assessment that balances high accuracy and high efficiency.

[0022] The specific steps of the above method are as follows: Step 1: Construct a three-dimensional elastoplastic wheel-rail cyclic rolling contact finite element model. (See technical route below.) Figure 2 As shown, the specific components include wheel-rail geometry and contact modeling, wheel-rail material constitutive modeling, wheel-rail cyclic rolling simulation, and model verification.

[0023] Specifically, wheel-rail geometry and contact modeling includes: 1. Establish a three-dimensional geometric model containing the true contours of the wheel tread and the rail top to accurately represent the shape and geometric transition characteristics of the wheel-rail contact interface; 2. The wheel tread is generated by substructure reduction to super elements, and finite rotational motion simulation is achieved through the custom element subroutine UEL; 3. The rail top extends to form a shadow unit, and periodic boundary conditions are applied at both ends of the rail to ensure the continuity of displacement and stress.

[0024] 4. Wheel-rail contact is established using a surface-to-surface contact algorithm. The penalty function method and the Coulomb friction model are used to solve the tangential contact behavior of the wheel-rail method, respectively. Friction coefficient and creep rate parameters are introduced to simulate the actual wheel-rail contact mechanical response under different rolling-slip contact conditions. Based on parametric modeling, parameters for multiple working conditions can be flexibly adjusted, providing a repeatable and efficient modeling framework.

[0025] Specifically, the constitutive model of wheel-rail materials includes: 1. Based on experimental data of rail material mechanics, the classic Adel-Karim–Ohno cyclic plastic constitutive model is improved. Taking into account the effects of multiaxial stress and ratcheting effect, isotropic deformation resistance parameters and ratcheting parameters related to cumulative plastic strain are introduced into the isotropic hardening law and the nonlinear kinematic hardening law, respectively. Combined with Tanaka nonproportional parameters, the cyclic characteristics and ratcheting behavior of rail under multiaxial cyclic loading are characterized.

[0026] 2. The improved cyclic plastic constitutive model adopts the implicit stress integration method and constructs a consistent tangential stiffness matrix to achieve finite element numerical solution.

[0027] 3. Write a UMAT subroutine and embed it into the finite element analysis software. The wheel is modeled using an elastic material.

[0028] Specifically, the wheel-rail cyclic rolling simulation includes: 1. The wheel-rail cyclic rolling simulation is mainly achieved by controlling the boundary and load conditions of the wheel through the DISP and URDFIL subroutines, which are completed in sequence by multiple steps such as rolling, rewinding, reloading and releasing.

[0029] 2. Simultaneously, the UMESHMOTION subroutine is used to calculate the superposition of rail wear during wheel-rail cyclic rolling contact, as well as the competitive relationship between wear and rolling contact fatigue.

[0030] 3. By capturing the stress / strain evolution of the rail through the loading and unloading of the wheel's cyclic rolling motion, the formation process of residual stress and cumulative plastic strain is revealed. The subroutines are nested in an orderly manner and interact with the finite element main program in real time, achieving efficient cyclic rolling contact simulation.

[0031] Specifically, model validation includes: 1. By comparing the finite element calculation results of elastic materials with the solutions of classical contact theory, the differences in contact patch morphology and contact stress distribution are verified.

[0032] 2. By comparing the stress / strain response with that of uniaxial and multiaxial cyclic loading experiments, the accuracy of the improved constitutive model in describing the cyclic deformation and ratchet effect of the rail is verified.

[0033] 3. The stress distribution, plastic zone morphology, and ratchet strain evolution results predicted by finite element analysis under elastoplastic materials were compared with the data from the wheel-rail rolling contact test bench and the literature results to verify the predictive ability of the model under cyclic rolling contact.

[0034] Step 2: Generation of rail stress / strain field samples and construction of dataset; see the technical route below. Figure 3 As shown, this includes the design of typical working condition parameters and automated finite element calculation, data extraction and processing, and sample set construction.

[0035] The design of typical operating condition parameters and automated finite element calculation specifically include: 1. To obtain comprehensive rail rolling contact response data, a multi-dimensional typical working condition parameter space is first constructed, which includes vehicle operating parameters (such as profile, axle load, running speed, creep rate, and friction coefficient) and track structure parameters (such as profile, curve radius, superelevation, and irregularity).

[0036] 2. The Latin hypercube sampling method is used for parameter sampling to ensure the uniform distribution and representativeness of the samples in the high-dimensional space.

[0037] 3. Based on the established finite element model, write automated calculation control scripts to achieve full automation of the process, including typical working condition parameter input, model calling, calculation submission, and result post-processing.

[0038] 4. Batch calculation and generation of rail stress / strain field data under typical working conditions to achieve efficient sample generation.

[0039] The data extraction and processing specifically includes: 1. After the finite element calculation is completed, the script automatically extracts the contact pressure, equivalent stress, and equivalent plastic strain data of the nodes in the rail contact area.

[0040] 2. Perform anomaly detection on the extracted data and remove samples with numerical oscillations or non-convergence.

[0041] 3. To ensure comparability between samples, the results are processed by unifying the coordinate system and normalizing the scale, ultimately forming a structured, high-quality data table.

[0042] The construction of the sample set specifically includes: 1. Use scripts to add key feature labels to each sample, including maximum contact pressure and location, contact patch size, maximum equivalent stress and location, maximum equivalent plastic strain and location, etc., to form a structured sample library with corresponding input and output.

[0043] 2. Finally, an extensible typical working condition rail stress / strain dataset is constructed to provide a high-fidelity training data foundation for subsequent generative adversarial networks.

[0044] Step 3: Rapid prediction of rail stress / strain field based on PC-GAN, which includes conditional generation network design, physical constraint embedding and loss function optimization, as well as model training strategy and performance evaluation.

[0045] The specific design of conditional generation networks includes: 1. Construct a conditional generative adversarial network based on physical constraints to achieve efficient mapping from multiple operating parameters to local stress / strain fields. The model input includes key vehicle operation parameters and track structure parameters (N=9) for the entire line, and the output is the multidimensional stress / strain field distribution tensor of the rail.

[0046] 2. To improve the model's ability to represent features under different working conditions, the generator adopts a multi-branch convolutional structure to extract geometric features and load features separately, and achieves multi-source feature coupling in the fusion layer.

[0047] 3. The discriminator adopts the PatchGAN structure to determine the authenticity of local features of the output field, thereby preserving stress gradient and local peak information.

[0048] Specifically, physical constraint embedding and loss function optimization include: To avoid the network relying solely on data distribution while ignoring physical laws, multiple physical constraint mechanisms are introduced during training: ① Contact force conservation constraints are added to the loss function to ensure that the predicted stress field satisfies the balance between normal and tangential contact forces; ② Energy consistency constraints are introduced to ensure that the evolution of the predicted plastic work or strain energy density with the number of cycles matches the finite element results; ③ Boundary condition matching constraints are used to ensure that the predicted rail stress / strain distribution under extreme working conditions conforms to physical laws and avoids non-physical anomalies.

[0049] The model training strategy and performance evaluation specifically include: 1. First, pre-training is performed on low-dimensional simplified samples to quickly learn the global distribution characteristics of the stress / strain field. Then, complex working condition data is gradually introduced for refined training.

[0050] 2. The Adam optimizer and gradient penalty mechanism are used to suppress mode collapse, and cross-validation and early stopping strategies are combined to prevent overfitting.

[0051] 3. After training, the model's prediction accuracy in key indicators such as stress / strain distribution, magnitude, and evolution is evaluated by comparing it with the finite element results.

[0052] Furthermore, error space mapping analysis is used to evaluate the robustness and generalization performance of the model under different working conditions.

[0053] Step 4: Calculation of rail fatigue damage and life assessment based on rapid prediction results, specifically including multi-axis fatigue damage prediction and damage distribution and statistical analysis of the entire line.

[0054] Specifically, multiaxial fatigue damage prediction includes: 1. After obtaining the rail stress / strain field predicted by PC-GAN under multiple working conditions, the stress / strain is transformed from the global coordinate system to the coordinate system on the rotation plane based on the rotation matrix. The transformed stress / strain data is then input into the incremental multiaxial fatigue damage model to calculate the plastic strain energy increment and fatigue damage increment of each node on any rotation plane step by step.

[0055] 2. By integrating the fatigue damage increment over the entire cycle, the cumulative fatigue damage value of each node is obtained, thus revealing the distribution law of cumulative fatigue damage of the rail.

[0056] 3. Based on the critical plane method, the plane and location corresponding to the maximum cumulative fatigue damage are identified. Finally, fatigue life is calculated according to the critical damage value criterion, realizing direct prediction from local mechanical response to fatigue life.

[0057] The damage distribution and statistical analysis of the entire line specifically includes: 1. Based on the operating characteristics of different line sections, spatial mapping and damage overlay analysis are performed on the PC-GAN prediction results. The fatigue damage energy density distribution of different sections is calculated using a segmented statistical method, and a method for constructing the fatigue damage distribution of the entire line is built using Kriging interpolation and spatial weighting algorithms.

[0058] 2. Further combine vehicle operating conditions and track structure parameters to conduct sensitivity analysis, identify dominant damage factors, and reveal the regional differences and evolution patterns of rail fatigue damage.

[0059] 3. Based on the fatigue damage calculation results, draw a fatigue risk distribution map of the rails along the entire line and identify potentially high-risk sections.

[0060] As can be seen from the above, this invention deeply integrates elastoplastic finite element method with generative adversarial network for rapid prediction of rolling contact fatigue damage of rails, realizing cross-domain coupling from "high-precision numerical calculation" to "intelligent prediction generation".

[0061] This invention establishes a rapid prediction method for rail rolling contact fatigue damage based on elastoplastic finite element analysis and physical constraint generative adversarial networks. This method can be widely applied to key aspects such as railway rail service performance evaluation, life prediction, and condition-based maintenance, providing a fast, low-cost, and high-precision prediction tool for intelligent track maintenance and structural health monitoring. Furthermore, this model framework can be extended to fatigue damage analysis of wheels, turnouts, and welded joints, providing a new technical approach for the intelligent operation and maintenance of rail transit infrastructure.

[0062] The ultimate goal of this invention is to create a fatigue damage and life distribution map of rails along the entire railway line, identify potentially high-risk sections, promote the integration of wheel-rail contact mechanics calculations with artificial intelligence, and enrich the theoretical system in the field of rolling contact fatigue. Simultaneously, this achievement helps extend the service life of rails, reduce maintenance costs, and improve the safety and economy of railway transportation, playing a significant role in promoting the high-quality development of my country's rail transit equipment manufacturing industry through science, technology, and socio-economic means.

[0063] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0064] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for rapid prediction of rolling contact fatigue damage of a steel rail, characterized by, Includes the following steps: S1. Construct a three-dimensional elastoplastic wheel-rail cyclic rolling contact finite element model, and use this model to simulate and obtain the stress-strain response of the rail under different working conditions. S2. Based on the finite element model in S1, perform automated calculations for multiple working conditions to generate sample data of rail stress / strain field and construct a structured dataset for network training. S3. Build and train a conditional generative adversarial network based on physical constraints, and use multiple working condition parameters as network inputs to quickly predict the local stress / strain field distribution of the rail under the corresponding working condition. S4. Based on the rail stress / strain field predicted in S3, perform multiaxial fatigue damage calculation and life assessment, and analyze the damage distribution of the entire line.

2. The method for rapid prediction of rolling contact fatigue damage of a rail according to claim 1, characterized in that, Step S1 includes: S1-1, Wheel-rail geometry and contact modeling: Establish a three-dimensional geometric model that includes the real contours of the wheel tread and the rail top to accurately characterize the shape and geometric transition features of the wheel-rail contact interface; S1-2, Structural model of wheel-rail material: Based on experimental data of rail material mechanics, the classic Adel-Karim-Ohno cyclic plastic constitutive model was improved by introducing isotropic deformation resistance parameters and ratchet parameters, and combined with Tanaka non-proportional parameters to characterize the cyclic characteristics of rail and ratchet behavior under multi-axis cyclic loading. The UMAT subroutine was written and embedded into the finite element analysis software. S1-3, Simulation of wheel-rail cyclic rolling: The DISP and URDFIL subroutines control wheel movement and load, enabling multi-step sequential tasks such as rolling, rewinding, reloading, and releasing. The UMESHMOTION subroutine considers wear effects to capture the cyclic evolution of rail stress / strain. S1-4, Model Validation: By comparing the finite element calculation results with theoretical solutions, experimental data, and literature results, the multi-level prediction capability of the wheel-rail contact model was systematically verified, from elastic contact behavior and constitutive model accuracy to cyclic rolling contact response.

3. The method of claim 1, wherein the method is characterized by: Step S2 includes: S2-1. Typical Operating Condition Parameter Design and Automated Finite Element Calculation: A multidimensional parameter space for typical working conditions, including vehicle operating parameters and track structure parameters, is constructed. Parameters are sampled using the Latin hypercube sampling method, and rail stress / strain field data are generated in batches using automated finite element calculation scripts. S2-2, Data Extraction and Processing: The script automatically extracts and cleans the corresponding finite element calculation results such as contact pressure, equivalent stress, and equivalent plastic strain. After coordinate unification and normalization, a structured high-quality data table is formed. S2-3, Sample set construction: Key feature labels were added to each sample, including maximum contact pressure and location, contact patch size, maximum equivalent stress and location, and maximum equivalent plastic strain and location, thus constructing a structured sample library corresponding to input and output.

4. The method of claim 1, wherein, The physical constraint-based conditional generative adversarial network in step S3 uses a multi-branch convolutional structure to extract and fuse the geometric and load features from the input working condition parameters. Its discriminator uses a PatchGAN structure to determine the local authenticity of the generated stress / strain field.

5. The method of claim 1, wherein: In step S3, when training the conditional generative adversarial network, multiple physical constraints are introduced into the loss function, including contact force conservation constraints, energy consistency constraints, and boundary condition matching constraints.

6. The method of claim 1, wherein, The multiaxial fatigue damage calculation in step S4 includes: S4-1. Transform the predicted stress / strain field data from the global coordinate system to the potential critical planes of the material points using a rotation matrix. S4-2. An incremental multiaxial fatigue damage model is used to calculate the plastic strain energy increment and fatigue damage increment of each potential critical plane in each load cycle, and the cumulative fatigue damage is obtained by integration. S4-3. Based on the critical plane method, find the maximum cumulative damage and its corresponding plane and location, and predict the fatigue life of the rail according to the critical damage value criterion.

7. The method of claim 1, wherein, The full-line damage distribution analysis in step S4 includes: S4-4. Based on the operating characteristics of different line sections, the predicted damage results are spatially mapped and superimposed. S4-5. By using segmented statistics, Kriging interpolation, and spatial weighting algorithms, a rail fatigue damage distribution map of the entire line is constructed. S4-6. Conduct parameter sensitivity analysis to identify the dominant damage factors and identify potential high-risk fatigue zones based on the damage distribution map.

Images