Method and device for predicting damage state of ballastless track

Abstract

The invention provides a ballastless track damage state prediction method and prediction equipment, and relates to the technical field of track prediction. The method comprises the following steps: acquiring multi-dimensional data to be monitored; feature extraction is conducted on the monitoring data of the temperature, the displacement, the strain and the vibration of the track structure, the extracted features are input into a pre-built damage position recognition model, and a first damage result is obtained; inputting the meteorological monitoring data and the first damage result into a pre-constructed cross-scale finite element model, and outputting internal and external damage evolution mechanisms and an updated finite element model under different working conditions; inputting the track structure monitoring data, the first damage result and the internal and external damage evolution mechanisms under different working conditions into a pre-constructed damage state prediction model to output a damage state prediction result; and predicting the service life of the track based on the meteorological monitoring data, the first damage result, the damage state prediction result and the updated finite element model. The track damage state prediction accuracy can be improved.

Description

Technical Field

[0001] This invention relates to the field of track prediction technology, and in particular to a method and device for predicting the damage status of ballastless tracks. Background Technology

[0002] As a core infrastructure of high-speed railways, ballastless track is subjected to the dynamic loads generated by high-speed train operation, as well as the long-term effects of environmental and meteorological factors such as temperature changes, solar radiation, and wind speed. This can lead to the deterioration of materials such as microscopic self-compacting concrete, damage to the mortar layer interface, and even macroscopic destruction of the track slab structure, seriously threatening train safety.

[0003] Currently, there are methods for detecting internal damage to ballastless tracks. However, traditional detection methods use sensors such as strain gauges or accelerometers, which can only acquire data from limited areas. These methods cannot comprehensively reflect the interlayer and surface damage status of the entire track structure, and it is difficult to quantify the evolution of these damages. Current data-driven detection methods often rely on single types of data and lack the ability to comprehensively analyze multi-source heterogeneous data, leading to inaccurate prediction results. Furthermore, while finite element analysis (FEM) methods are used for prediction, current FEM methods only focus on single damage types and lack the ability to perform multi-scale microscopic and macroscopic integrated analysis of material interface damage and overall structural response.

[0004] Improving the accuracy of predicting the internal and external damage conditions of ballastless tracks has become an urgent technical problem to be solved. Summary of the Invention

[0005] This invention provides a method and device for predicting the damage status of ballastless tracks, in order to solve the problem of inaccurate prediction by current methods.

[0006] In a first aspect, embodiments of the present invention provide a method for predicting the damage state of ballastless tracks, including:

[0007] Acquire multidimensional data to be monitored; the multidimensional data includes track structure monitoring data and meteorological monitoring data of the location of the track structure. The track structure monitoring data includes displacement monitoring data, strain monitoring data, vibration monitoring data, and track structure temperature monitoring data.

[0008] Feature extraction was performed on the temperature monitoring data, displacement monitoring data, strain monitoring data and vibration monitoring data of the track structure, and the extracted features were input into the pre-built damage location identification model to obtain the first damage result of the track to be monitored.

[0009] Meteorological monitoring data and the first damage results are input into a pre-constructed multi-scale finite element model to obtain the internal and external damage evolution mechanism of the track under different working conditions and the updated multi-scale finite element model.

[0010] The multidimensional data of the track structure, the first damage result, and the internal and external damage evolution mechanism under different working conditions are input into the pre-constructed damage state prediction model to output the damage state prediction result of the track to be monitored.

[0011] Based on meteorological monitoring data, initial damage results, damage status prediction results, and an updated cross-scale finite element model, the lifetime of the track to be monitored is predicted.

[0012] In one possible implementation, the damage location identification model is constructed based on a clustering algorithm, and the first damage result includes damage type, damage location, damage degree, and damage range;

[0013] Feature extraction was performed on the temperature monitoring data, displacement monitoring data, strain monitoring data, and vibration monitoring data of the track structure, including:

[0014] The time-domain features of temperature monitoring data, displacement monitoring data, strain monitoring data and vibration monitoring data of the track structure are extracted to obtain the extreme values ​​of temperature gradient, peak displacement, vibration dominant frequency shift and area of ​​high stress region of the track structure;

[0015] The extracted features are input into a pre-built damage location identification model to obtain the first damage result of the track to be monitored, including:

[0016] The extreme values ​​of the temperature gradient of the track structure, the peak displacement, the vibration dominant frequency shift, and the area of ​​the high-stress region are input into the trained damage location identification model. Based on the clustering results, the damage type, damage location, damage degree, and damage range of the track to be monitored are obtained.

[0017] In one possible implementation, the multi-scale finite element model includes the train, rails, fasteners, track slabs, mortar layer, base plate, and lower foundation layer;

[0018] The track slab adopts the standard size of the track to be monitored, the base plate is sized according to the design specifications, the thickness of the mortar layer is set according to the construction standards, the rail cross section adopts the CHN60 standard section, the fastener is the WJ-7 type fastener, the lower foundation layer includes two working conditions: roadbed and bridge, the base plate adopts a hexahedral structured grid, the track slab and base plate adopt the plastic damage constitutive model of concrete, the mortar layer adopts the viscoelastic constitutive model, the track slab and mortar layer adopt the nonlinear cohesive force model, and the rail adopts the elastoplastic constitutive model.

[0019] In one possible implementation, a zero-thickness cohesive element is provided at the contact interface between the track slab and the mortar layer. This zero-thickness cohesive element is used to simulate the damage evolution process of the contact interface between the track slab and the mortar layer.

[0020] A zero-thickness cohesive unit is also provided between the mortar layer and the base plate. The parameters of the zero-thickness cohesive unit are determined based on the roughness and bonding strength of the contact interface between the mortar layer and the base plate.

[0021] In one possible implementation, meteorological monitoring data includes ambient temperature monitoring data, solar radiation monitoring data, and wind speed monitoring data; in the multi-scale finite element model simulation, the train dynamic load is simulated by applying a moving wheel-rail force to the top surface of the rail; the ambient temperature load includes two parts: overall temperature rise and fall, and extreme gradient temperature. The range of overall temperature rise and fall is determined based on the extreme temperature conditions of different climate regions, while the extreme gradient temperature is determined based on solar radiation monitoring data, annual temperature cycle changes, and daily temperature cycle; the thermal boundary conditions include solar radiation thermal boundary conditions and wind speed thermal boundary conditions. The solar radiation thermal boundary condition is based on the radiative heat flux applied to the exposed surfaces of the track slab and fastening system to simulate the heating process of the track structure by shortwave solar radiation; the wind speed thermal boundary condition is used to simulate the convective heat transfer capacity of wind speed on the track structure surface.

[0022] The thermal response calculated by the multi-scale finite element model is compared with the temperature, stress and displacement responses obtained from the track structure monitoring data. The material parameters and boundary conditions of the multi-scale finite element model are then corrected and updated. For example, the elastic modulus and heat transfer coefficient of the track slab are corrected and updated to obtain an updated multi-scale finite element model.

[0023] In one possible implementation, the damage state prediction model is constructed based on an improved particle swarm optimization algorithm and a long short-term memory network. The improved particle swarm optimization algorithm is used to optimize the hyperparameters in the long short-term memory network, and the long short-term memory network is used to extract the evolution law of the internal and external damage evolution mechanism in the time dimension under different working conditions.

[0024] In one possible implementation, the training sample set of the damage state prediction model includes multiple training samples. Each training sample includes acquired multidimensional data, the first damage result corresponding to the multidimensional data, the train dynamic load, ambient temperature load, and thermal boundary conditions in the updated cross-scale finite element model obtained through the inversion algorithm, as well as the damage distribution map of each component corresponding to the multidimensional data, the damage index change curve over time, the damage evolution trend prediction result of the preset location, and the overall health level.

[0025] In one possible implementation, the evolution mechanism of internal and external damage under different working conditions includes material damage variables, interface crack evolution process, and interlayer stress distribution of components.

[0026] In one possible implementation, the damage status prediction results of the track to be monitored include damage distribution maps of multiple components, damage index change curves over time, damage evolution trend prediction results of preset locations, and overall health level.

[0027] In a second aspect, embodiments of the present invention provide a device for predicting the damage state of ballastless tracks, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation of the first aspect.

[0028] In this embodiment of the invention, the extracted features are clustered using a damage location identification model to obtain the first damage result of the track to be monitored, thus preliminarily determining the basic information of external damage. To further reveal the microscopic damage mechanism at the material level, a cross-scale finite element model is constructed to determine the internal and external damage evolution mechanism of the track under different working conditions, laying a theoretical foundation for subsequent damage state prediction models. The constructed damage state prediction model can then predict the damage state of the track to be monitored. This invention, by fusing real-time acquired data with the internal and external damage evolution mechanism provided by the finite element model, avoids the overfitting and extrapolation failure problems of purely data-driven damage state prediction models. By establishing a closed-loop mechanism of "monitoring → identification → modeling → prediction," an intelligent health monitoring framework covering the entire process from theoretical model to engineering practice is constructed, enabling more accurate prediction of the damage state of ballastless tracks. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating the implementation of the method for predicting the damage state of ballastless track provided in this embodiment of the invention.

[0030] Figure 2 This is a schematic diagram of the integrated air-ground-space-orbit monitoring system provided in an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of real-time multidimensional data acquisition provided in an embodiment of the present invention;

[0032] Figure 4 This is a schematic diagram of the structure of the multi-scale finite element model provided in the embodiment of the present invention;

[0033] Figure 5 This is a flowchart illustrating the implementation of lifetime prediction provided in an embodiment of the present invention;

[0034] Figure 6This is a schematic diagram of fatigue life prediction results provided in an embodiment of the present invention;

[0035] Figure 7 This is a schematic diagram of the structure of the device for predicting the damage status of ballastless track provided in an embodiment of the present invention. Detailed Implementation

[0036] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0037] As described in the background section, current predictions of internal and external damage to ballastless track structures suffer from numerous problems, including narrow coverage, insufficient accuracy, and poor real-time performance, making it difficult to meet the requirements for smoothness and safety in high-speed railway train operation.

[0038] The inventors discovered that although the finite element method is currently used to assess damage, the current assessment method lacks the ability to dynamically analyze the evolution mechanism of damage inside and outside the ballastless track. It cannot accurately simulate and predict the entire process from the deterioration of key materials, interface damage, structural failure to the safety and comfort of the train, resulting in the prediction results not matching the actual situation.

[0039] To address the aforementioned technical problems, this invention provides a method and apparatus for predicting the damage status of ballastless tracks.

[0040] See Figure 1 The document illustrates a flowchart of the method for predicting the damage state of ballastless track provided in an embodiment of the present invention, which is described in detail below:

[0041] S110. Obtain the multidimensional data to be monitored.

[0042] The multidimensional data includes track structure monitoring data and meteorological monitoring data of the track structure's location. The track structure monitoring data includes displacement monitoring data, strain monitoring data, vibration monitoring data, and track structure temperature monitoring data. It should be noted that the track structure temperature monitoring data here refers to the detected temperature of the track structure.

[0043] Meteorological monitoring data includes ambient temperature monitoring data, solar radiation monitoring data, and wind speed monitoring data at the location of the orbital structure.

[0044] To enable multi-dimensional monitoring of the orbital structure, an integrated air-ground-space-orbit monitoring system can be adopted, such as... Figure 2 As shown, this includes BeiDou satellites and automatic weather stations in the air, high-definition cameras on the ground, and multi-parameter fiber optic sensors for temperature, humidity, stress, and vibration in orbit, which collect data in real time, such as... Figure 3 The data shown includes multidimensional data such as displacement, strain, vibration, and temperature.

[0045] BeiDou satellites, through monitoring stations and base stations, measure uneven settlement and regional settlement of ballastless track and its underlying foundation with an accuracy of millimeters, which is used to identify the location of track slab deformation and other defects caused by uneven settlement.

[0046] Automatic weather stations are installed near the track location to monitor ambient temperature, humidity, solar radiation, and wind speed. Multidimensional Copula functions are used to analyze the nonlinear relationship between environmental factors and temperature, stress, and displacement responses within the track.

[0047] High-definition cameras are installed on the railings to capture images of the multi-layered surface of the track. Through 3D reconstruction and YOLO optimization algorithms, hidden defects such as track slab cracks are identified.

[0048] The multi-parameter fiber optic sensor, based on the Brillouin scattering principle, is installed within the multi-layer structure of the track to monitor temperature, humidity, stress, and vibration, and to identify interlayer damage characteristics.

[0049] After acquiring multidimensional data, the acquired multidimensional data can be input into a lightweight model for real-time prediction. When the degree of damage or displacement value exceeds a preset threshold, an early warning signal is triggered.

[0050] The integrated air-space-ground-orbit monitoring system serves as the starting point for the entire process, providing multi-dimensional and high-precision structural status monitoring data, which forms the basis for all subsequent feature extraction, modeling, and prediction.

[0051] S120. Extract features from the track structure temperature monitoring data, displacement monitoring data, strain monitoring data and vibration monitoring data respectively, and input the extracted features into the pre-built damage location identification model to obtain the first damage result of the track to be monitored.

[0052] As Figure 2 As shown, after acquiring multidimensional data, the collected data can be preprocessed and features extracted through edge computing terminals, reducing the computing load on the cloud.

[0053] In some embodiments, the time-domain features of temperature monitoring data, displacement monitoring data, strain monitoring data, and vibration monitoring data of the track structure can be extracted to obtain the extreme values ​​of the temperature gradient, the peak displacement, the vibration dominant frequency shift, and the area of ​​the high-stress region. Then, the extreme values ​​of the temperature gradient, the peak displacement, the vibration dominant frequency shift, and the area of ​​the high-stress region are input into a trained damage location identification model. Based on the clustering results, the damage type, damage location, damage degree, and damage range of the track to be monitored are obtained.

[0054] In some embodiments, the pre-built damage location identification model is a model based on a clustering algorithm. K-means clustering and / or DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithms are used to perform unsupervised classification of the extracted features to identify interlayer damage results. The damage location identification model is trained using pre-collected sample data during the training process, which will not be elaborated here.

[0055] K-means is suitable for feature sets with relatively regular structures and uniform sample distribution. DBSCAN is suitable for identifying outliers (such as local damage) and can automatically identify clusters and noise. Methods for identifying interlayer damage results observe the migration of cluster centers and variations in cluster shape to discover areas where data deviates from the normal distribution, focusing on identifying "outliers" and "isolated clusters" within the clusters, which often correspond to damaged parts or abrupt changes in the structure. Specific damage patterns, such as interlayer interface peeling, may exhibit consistent "abnormal clustering" behavior across multiple feature dimensions.

[0056] The damage location identification model achieves comprehensive diagnosis by unifying and fusing clustering results of multi-dimensional features through decision-level fusion methods such as weighted averaging. First, the outputs of different clustering algorithms, such as K-means and DBSCAN, across multiple feature dimensions are spatially mapped and uniformly labeled. Then, a weighted average is used for decision-level integration. The weighting coefficients are set based on historical clustering accuracy (Silhouette Coefficient, ARI index) and feature reliability.

[0057] Damage types, such as surface cracks or interlayer delamination, can be determined by clustering patterns with different feature combinations. Damage locations are given by the coordinates of the measurement points corresponding to the abnormal clusters. Damage severity can be determined by the offset of the abnormal cluster centers and the inter-cluster spacing. Damage extent can be judged by the spatial distribution density and area of ​​the sample points in the abnormal clusters.

[0058] By transforming raw multi-source monitoring data into actionable structural diagnostic criteria, and through the complementary fusion of multiple clustering results, the accuracy of damage identification can be improved.

[0059] However, this step primarily focuses on identifying interface damage or track slab damage, such as interlayer interface gaps and surface cracks, emphasizing anomalous clustering behavior in localized areas. It does not yet address material layer damage or the overall impact of these localized damages on train safety. This step mainly involves preliminary damage identification and does not predict broader material damage or systemic safety risks. It is a data-driven diagnostic step, capable of identifying abnormal patterns and locating potential defects. However, this process relies heavily on monitoring data and lacks the ability to explain the physical mechanisms of damage evolution and predict future trends.

[0060] S130. Input the meteorological monitoring data and the first damage result into the pre-constructed cross-scale finite element model to obtain the internal and external damage evolution mechanism of the track under different working conditions and the updated cross-scale finite element model.

[0061] It should be noted that the evolution mechanism of internal and external damage under different working conditions includes material damage variables, the evolution process of interface cracks, and the stress distribution between components.

[0062] Since step S120 only performs preliminary damage identification, it achieves preliminary identification of surface damage such as track slab cracks and interface damage such as delamination between the track slab and mortar layers. However, step S120 mainly focuses on the local damage morphology at the mesoscale, without revealing the microscopic damage mechanism at the material level, nor establishing a complete mechanical chain from damage occurrence to its expansion and eventual impact on the overall performance of the track structure and train operation safety.

[0063] In some embodiments, the multi-scale finite element model is a three-dimensional geometric model built on the COMSOL Multiphysics platform, such as... Figure 4 As shown, it includes a train, rails 41, fasteners 42, track slabs 43, mortar layer 44, base plate 45, and lower foundation layer 46.

[0064] The track slab 43 adopts the standard size of the track to be monitored, the base plate 45 is sized according to the design specifications, the thickness of the mortar layer 44 is set according to the construction standards, the cross section of the rail 41 adopts the CHN60 standard section, the fastener is the WJ-7 type fastener, the lower foundation layer 46 includes two working conditions: roadbed and bridge, the base plate 45 adopts a hexahedral structured grid, the track slab 43 and the base plate 45 adopt the plastic damage constitutive model of concrete, the mortar layer 44 adopts the viscoelastic constitutive model, the nonlinear cohesive force model is adopted between the track slab and the mortar layer, and the rail 41 adopts the elastoplastic constitutive model.

[0065] The model employs a combined structured and unstructured meshing strategy. For regular geometries such as track slabs and base plates, hexahedral structured meshes are used, with element sizes controlled within a reasonable range to balance computational accuracy and efficiency. For complex geometries such as fastener systems and contact interface areas, tetrahedral unstructured meshes are used, with localized refinement in stress concentration areas. The minimum element size is determined based on stress gradient requirements. At least multiple layers of elements are used along the mortar layer thickness direction to accurately capture the stress gradient. Zero-thickness cohesive elements are used at the contact interface to simulate interface damage behavior.

[0066] The track slab and base plate employ a concrete damaged plasticity constitutive model, which can simultaneously describe the damage evolution process of concrete under tensile and compressive conditions. Material parameters include key mechanical indices such as elastic modulus, Poisson's ratio, compressive strength, tensile strength, fracture energy, and damage evolution parameters, with values ​​determined based on material test results. The mortar layer uses a viscoelastic constitutive model, considering its creep characteristics under long-term loads; the elastic modulus, Poisson's ratio, and viscous damping coefficient are calibrated based on experimental data. The rail uses an elastoplastic constitutive model, with yield strength determined according to steel standards. The fastening system is simplified to spring units, with vertical and lateral stiffness set according to product technical parameters. Cohesive contact units are set between the track slab and mortar layer to consider the interfacial bonding effect of the nonlinear cohesive model, thereby simulating the interfacial damage evolution process. An exponential nonlinear cohesive constitutive model is used, and parameters such as peak stress, tangential peak stress, Type I fracture energy, and Type II fracture energy are obtained through interfacial tests. The contact algorithm employs a penalty function method, setting appropriate contact stiffness to avoid excessive penetration. Contact elements are also set between the mortar layer and the base plate, with parameters determined based on interface roughness and bonding strength.

[0067] In this embodiment, the load and boundary conditions of the multi-scale finite element model are set in the following ways: simulation of train dynamic load, application of ambient temperature load, definition of thermal boundary conditions, and setting of structural boundary conditions.

[0068] The simulation of train dynamic load involves applying a moving wheel-rail force to the top surface of the rails to simulate the dynamic load effect of the train. The load amplitude is determined based on the vehicle type, and the dynamic load coefficient is obtained through wheel-rail dynamics analysis, taking into account the effects of track irregularities, wheel flats, and other factors. The load moves in the time domain according to the train's operating speed, covering a range of common operating speed levels. Hertz contact theory is used to determine the wheel-rail contact area, converting concentrated forces into surface loads.

[0069] The application of ambient temperature load takes into account the impact of ambient temperature changes on the track structure, applying a non-uniform temperature field. The temperature load includes two parts: overall temperature rise and fall, and extreme gradient temperature. The range of overall temperature rise and fall is determined based on the extreme temperature conditions of different climatic regions in my country. The gradient temperature considers the effects of solar radiation, annual temperature variations, and daily temperature differences, resulting in a significant temperature difference between the top and bottom surfaces of the track slab. The ambient temperature load primarily reflects heat conduction.

[0070] In some embodiments, to refine the impact of solar radiation and wind speed on the orbital structure, the meteorological monitoring data also includes solar radiation monitoring data and wind speed monitoring data.

[0071] The temperature field simulation is further refined by introducing solar radiation and wind speed as thermal boundary conditions.

[0072] The thermal boundary condition of solar radiation is based on the radiative heat flux applied to the exposed surfaces of the track slab and fastening system, used to simulate the heating process of the track structure by shortwave solar radiation. The thermal boundary condition of solar radiation reflects the effect of thermal radiation.

[0073] The thermal boundary condition for wind speed is used to simulate the convective heat transfer capacity of wind on the surface of the track structure. The thermal boundary condition for wind speed reflects the heat exchange effect.

[0074] Solar radiation is applied to the exposed surfaces of the track slab and fastening system through the boundary condition of "radiative heat flux." A solar radiation intensity curve varying over time is input, and the "external radiation source" parameter is set in conjunction with the material absorptivity to simulate the heating process of shortwave solar radiation. The solar radiation intensity curve varying over time is based on solar radiation monitoring data.

[0075] Wind speed affects the convective heat transfer capacity of the track structure surface. The convective heat transfer coefficient corresponding to the wind speed is calculated by wind speed monitoring data and empirical formulas, and the coefficient and the ambient temperature function are input into the "convective heat flux" boundary to dynamically reflect the wind cooling effect.

[0076] The aforementioned thermal boundary conditions are applied to the "heat conduction module" and coupled with the structural mechanics module. Taking into account heat conduction, convective heat transfer and radiative heat transfer, the non-uniform temperature distribution of the orbital structure under natural weather conditions is solved.

[0077] Structural boundary conditions are set by applying fixed constraints at the bottom of the model and using elastic constraints at the side boundaries to simulate the effects of a semi-infinite foundation. For bridge applications, hinged and sliding support boundary conditions are set at the beam ends. Non-reflective boundary conditions are used to avoid the effects of stress wave reflection at the boundaries.

[0078] The multi-field coupled solution includes solver configuration, damage evolution calculation, and post-processing and result extraction. Solver configuration utilizes the built-in multiphysics coupled solver in COMSOL to achieve thermo-mechanical coupled analysis. Adaptive control is used for the time step, with an initial step size of 0.001 s, automatically adjusted based on convergence. Spatial discretization uses Lagrangian elements, and time integration employs the backward Euler method to ensure numerical stability. The nonlinear solution uses the Newton-Raphson iterative method, with a convergence criterion set at a residual of less than 1 × 10⁻⁶. -6 .

[0079] Damage evolution calculation is performed by updating damage variables through user-defined functions, and the damage evolution is based on the strain energy release rate criterion. Within each time step, the stress-strain state is first calculated, then the material damage parameters are updated, and finally the material stiffness matrix is ​​corrected. Interface damage is characterized by the damage variables of cohesive elements; when the damage variable reaches 1, it indicates that the interface has completely failed.

[0080] Post-processing and result extraction involve extracting key response parameters after calculation, including temperature distribution, stress distribution, displacement field, and vibration response. A damage cloud map is created to visualize the damage evolution process, and time-history response curves of feature points are extracted to analyze dynamic characteristics. The parameter scanning function analyzes the structural response under different working conditions, providing training data for the damage state prediction model.

[0081] The operational safety of the train is determined based on its derailment coefficient, wheel load reduction rate, lateral force on the wheel and axle, vertical force on the wheel and rail, and vertical and longitudinal acceleration response.

[0082] In this step, multi-field coupling analysis is used to input the temperature, stress, displacement, and vibration response of the track structure, meteorological monitoring data, and damage types, locations, degrees, and extents obtained in previous steps into a multi-scale finite element model. This establishes a multi-scale damage evolution law from "microscopic material damage - mesoscopic interface failure - macroscopic component failure - system service safety." The model can capture multi-level damage modes such as the initiation and propagation of microcracks inside the mortar layer, the separation of the track slab and mortar layer interface, and cracking on the track slab surface, revealing the evolution mechanism of damage in spatial and temporal dimensions.

[0083] The thermal response calculated by the multi-scale finite element model is compared with the temperature, stress and displacement responses obtained from the track structure monitoring data. The material parameters and boundary conditions of the multi-scale finite element model are corrected and updated, such as the elastic modulus and heat transfer coefficient of the track slab, to obtain an updated multi-scale finite element model.

[0084] The multi-scale finite element model provides a solid theoretical foundation and numerical tool for the evolution of internal and external damage to track structures. It can not only accurately predict the damage development trend under different service conditions, but also provide a scientific basis for structural optimization design, maintenance strategy formulation and safety assessment, and support the subsequent development of intelligent prediction algorithms based on machine learning.

[0085] To verify the accuracy of the cross-scale finite element model provided by this invention, continuous monitoring was conducted on existing typical railway lines such as the Beijing-Shanghai High-Speed ​​Railway and the Wuhan-Guangzhou High-Speed ​​Railway using the cross-scale finite element model provided by this invention. The monitoring data was compared with the simulation results of the finite element model under the same working conditions, such as different seasons, train loads, and operating speeds. The model's predictive ability was evaluated from dimensions such as response morphology (strain curve, temperature response), amplitude deviation (maximum displacement, peak stress), and evolution trend (fatigue damage accumulation). Internal structural damage was confirmed by combining actual on-site detection results (such as infrared thermal imaging, ultrasonic testing, and ground-penetrating radar), including mortar layer voids, track slab cracks, and interface separation. The measured damage information was matched and verified with the finite element prediction results in terms of spatial location, damage level, and damage range to verify the spatial diagnostic accuracy and timeliness of the model.

[0086] S140. Input the track structure monitoring, the first damage result, and the internal and external damage evolution mechanism under different working conditions into the pre-constructed damage state prediction model to output the damage state prediction result of the track to be monitored.

[0087] In some embodiments, the damage state prediction model is constructed by an improved particle swarm optimization algorithm and a long short-term memory network. The improved particle swarm optimization algorithm automatically adjusts the number of neuron layers and the learning rate of the long short-term memory network, and the long short-term memory network is used to extract the evolution law of the internal and external damage evolution mechanism in the time dimension under different working conditions.

[0088] In this embodiment, the training sample set of the damage state prediction model includes multiple training samples. Each training sample includes acquired multidimensional data, the first damage result corresponding to the multidimensional data, the train dynamic load, ambient temperature load, and thermal boundary conditions in the updated cross-scale finite element model obtained through the inversion algorithm, as well as the damage distribution map of each component corresponding to the multidimensional data, the damage index change curve over time, the damage evolution trend prediction result of the preset part, and the overall health level.

[0089] In terms of scenario simulation, extreme temperature gradients, heavy load cycles, and interface voids are applied as boundary conditions in a calibrated multi-scale finite element model to calculate the structural response and potential damage risk. Subsequently, simulated response data and measured data are jointly analyzed to construct a complete response set containing actual characteristics and theoretical deductions, which is used to train the damage state prediction model.

[0090] Although preliminary identification has been performed in the preceding steps, focusing solely on static diagnosis based on existing data is insufficient for predicting structural damage evolution under future or extreme conditions. Therefore, this invention, after obtaining the initial damage results of the track under test and the internal and external damage evolution mechanisms under different operating conditions, introduces a deep learning-based damage state prediction model to construct a structural state prediction framework that is interpretable, forward-looking, and adaptive. This framework predicts the damage response and evolution path of the structure under extreme environments such as sudden temperature changes and high-frequency loads.

[0091] IPSO is used to optimize the hyperparameters of the Long Short-Term Memory network, while LSTM is used to learn the evolution of damage over time. Together, they achieve accurate prediction of internal and external damage states of the structure. With continuous acquisition of real-time monitoring data, incremental learning or transfer learning methods are used to dynamically update the neural network parameters and physical model boundary conditions, enabling the damage state prediction model to continuously optimize as the working conditions evolve. Based on the feature intensity, location, and expansion trend of the predicted output, three state intervals—healthy, warning, and failure—are defined to determine the damage level and provide time-series warnings.

[0092] By integrating the internal and external damage evolution mechanisms under different working conditions provided by cross-scale finite element models—that is, physical boundaries and mechanical interpretations—overfitting and extrapolation failure problems of purely data-driven damage state prediction models can be avoided. Furthermore, in situations lacking historical data or facing sudden working conditions, information can be supplemented through model deduction, achieving "predictable" structural states. Relying on real-time data to continuously update the damage state prediction model, it possesses strong robustness and real-world adaptability. It provides a basis for judging the potential evolution trend of the structure, offering a time window and reliable criteria for subsequent fatigue life assessment and maintenance decisions. This enables accurate prediction of the internal and external damage states of the structure.

[0093] S150, based on meteorological monitoring data, first damage results, damage status prediction results, and an updated cross-scale finite element model, predicts the lifetime of the track to be monitored.

[0094] After obtaining meteorological monitoring data, initial damage results, and damage state prediction results, the Palmgren-Miner linear damage accumulation criterion and SN curves of different concrete structures can be combined to calculate the failure cycle number of the mortar layer interface and important components such as track slabs and mortar layer and base plate. This allows for the quantification of the fatigue life distribution of the interface and components, providing decision support for formulating scientific maintenance strategies, optimizing resource allocation, and ensuring train operation safety.

[0095] Based on the updated multi-scale finite element model from the previous steps, the Palmgren-Miner linear damage accumulation criterion is introduced using the COMSOL platform to simulate the stress / strain variation of the component under multi-field coupled loading, including train dynamic load and temperature field, with the number of cycles. Figure 5 and 6 As shown, the stress history of key locations, such as the upper surface of the track slab and the interface of the mortar layer, is extracted. Combined with material SN curves, such as stress-cycle number curves, the fatigue life of each element is calculated. Starting from the initial stress, the cyclic stress is extracted, and the damage variables are iteratively updated through the "fatigue calculation" module until the damage accumulation threshold "D≥D" is reached. n "To determine fatigue failure and predict the fatigue life of interfaces and components."

[0096] The prediction method provided by this invention uses a damage location identification model to perform cluster analysis on the extracted features to obtain the first damage result of the track to be monitored, thus initially determining the basic information of external damage. To further reveal the microscopic damage mechanism at the material level, a cross-scale finite element model is constructed to determine the internal and external damage evolution mechanism of the track under different working conditions. An updated cross-scale finite element model can then be developed, laying a theoretical foundation for subsequent damage state prediction models. The constructed damage state prediction model can then predict the damage state of the track to be monitored. By fusing real-time acquired data with the internal and external damage evolution mechanism provided by the finite element model, this invention avoids the overfitting and extrapolation failure problems of purely data-driven damage state prediction models. By establishing a closed-loop mechanism of "monitoring → identification → modeling → prediction," an intelligent health monitoring framework covering the entire process from theoretical model to engineering practice is constructed, enabling more accurate prediction of the damage state of ballastless tracks.

[0097] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0098] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0099] Figure 7 A schematic diagram of the structure of the device for predicting the damage state of ballastless track provided in an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below:

[0100] like Figure 7 As shown, the ballastless track damage prediction device 700 includes:

[0101] The data acquisition module 710 is used to acquire multidimensional data to be monitored; the multidimensional data includes track structure monitoring data and meteorological monitoring data of the location of the track structure, and the track structure monitoring data includes displacement monitoring data, strain monitoring data, vibration monitoring data and track structure temperature monitoring data.

[0102] The preliminary damage identification module 720 is used to extract features from the track structure temperature monitoring data, displacement monitoring data, strain monitoring data and vibration monitoring data respectively, and input the extracted features into the pre-built damage location identification model to obtain the first damage result of the track to be monitored;

[0103] The damage mechanism identification module 730 is used to input meteorological monitoring data and the first damage result into a pre-constructed cross-scale finite element model to obtain the internal and external damage evolution mechanism of the track under different working conditions and the updated cross-scale finite element model.

[0104] The state prediction module 740 is used to input track structure monitoring data, first damage results and internal and external damage evolution mechanisms under different working conditions into a pre-built damage state prediction model, so as to output the damage state prediction results of the track to be monitored.

[0105] The lifetime prediction module 750 is used to predict the lifetime of the track to be monitored based on meteorological monitoring data, the first damage result, the damage status prediction result, and the updated cross-scale finite element model.

[0106] In one possible implementation, the damage location identification model is a model built based on a clustering algorithm, and the first damage result includes damage type, damage location, damage degree, and damage range;

[0107] The preliminary damage identification module 720 is used to extract the time-domain features of track structure temperature monitoring data, displacement monitoring data, strain monitoring data and vibration monitoring data to obtain the extreme value of track structure temperature gradient, displacement peak value, vibration dominant frequency shift and high stress area.

[0108] The extreme values ​​of the temperature gradient of the track structure, the peak displacement, the vibration dominant frequency shift, and the area of ​​the high-stress region are input into the trained damage location identification model. Based on the clustering results, the damage type, damage location, damage degree, and damage range of the track to be monitored are obtained.

[0109] In one possible implementation, the multi-scale finite element model includes the train, rails, fasteners, track slabs, mortar layer, base plate, and lower foundation layer;

[0110] The track slab adopts the standard size of the track to be monitored, the base plate is sized according to the design specifications, the thickness of the mortar layer is set according to the construction standards, the rail cross section adopts the CHN60 standard section, the fastener is the WJ-7 type fastener, the lower foundation layer includes two working conditions: roadbed and bridge, the base plate adopts a hexahedral structured grid, the track slab and base plate adopt the plastic damage constitutive model of concrete, the mortar layer adopts the viscoelastic constitutive model, the track slab and mortar layer adopt the nonlinear cohesive force model, and the rail adopts the elastoplastic constitutive model.

[0111] In one possible implementation, a zero-thickness cohesive element is provided at the contact interface between the track slab and the mortar layer. This zero-thickness cohesive element is used to simulate the damage evolution process of the contact interface between the track slab and the mortar layer.

[0112] A zero-thickness cohesive unit is also provided between the mortar layer and the base plate. The parameters of the zero-thickness cohesive unit are determined based on the roughness and bonding strength of the contact interface between the mortar layer and the base plate.

[0113] In one possible implementation, during the simulation of the multi-scale finite element model, the train dynamic load is obtained by applying a moving wheel-rail force to the top surface of the rail; the temperature load includes two parts: overall temperature rise and fall, and extreme gradient temperature. The range of overall temperature rise and fall is determined based on the extreme temperature conditions of different climatic regions, while the extreme gradient temperature is determined based on solar radiation, annual temperature cycle variation, and daily temperature cycle difference; the thermal boundary conditions include solar radiation thermal boundary conditions and wind speed thermal boundary conditions. The solar radiation thermal boundary condition is based on the radiative heat flux applied to the exposed surfaces of the track slab and fastening system to simulate the heating process of the track structure by shortwave solar radiation; the wind speed thermal boundary condition is used to simulate the convective heat transfer capacity of wind speed on the track structure surface. The thermal response calculated by the multi-scale finite element model is compared with the temperature, stress, and displacement responses obtained from track structure monitoring data. The material parameters and boundary conditions of the multi-scale finite element model are then corrected and updated, such as the elastic modulus and heat transfer coefficient of the track slab, to obtain an updated multi-scale finite element model.

[0114] In one possible implementation, the damage state prediction model is constructed based on an improved particle swarm optimization algorithm and a long short-term memory network. The improved particle swarm optimization algorithm is used to optimize the hyperparameters in the long short-term memory network, and the long short-term memory network is used to extract the evolution law of the internal and external damage evolution mechanism in the time dimension under different working conditions.

[0115] In one possible implementation, the training sample set of the damage state prediction model includes multiple training samples. Each training sample includes acquired multidimensional data, the first damage result corresponding to the multidimensional data, the train dynamic load, ambient temperature load, and thermal boundary conditions in the updated cross-scale finite element model obtained through the inversion algorithm, as well as the damage distribution map of each component corresponding to the multidimensional data, the damage index change curve over time, the damage evolution trend prediction result of the preset location, and the overall health level.

[0116] In one possible implementation, the evolution mechanism of internal and external damage under different working conditions includes the material damage variables of ballastless track under different temperatures coupled with train loads, the evolution process of interface gaps, and the local stress distribution of components.

[0117] In one possible implementation, the damage status prediction results of the track to be monitored include damage distribution maps of multiple components, damage index change curves over time, damage evolution trend prediction results of preset locations, and overall health level.

[0118] This invention also provides a prediction device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the above method embodiments.

[0119] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0120] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for predicting the damage state of ballastless track, characterized in that, include: Acquire multidimensional data to be monitored; wherein, the multidimensional data includes track structure monitoring data and meteorological monitoring data of the location of the track structure, and the track structure monitoring data includes displacement monitoring data, strain monitoring data, vibration monitoring data, and track structure temperature monitoring data; Feature extraction is performed on the temperature monitoring data, displacement monitoring data, strain monitoring data and vibration monitoring data of the track structure respectively, and the extracted features are input into the pre-constructed damage location identification model to obtain the first damage result of the track to be monitored; The meteorological monitoring data and the first damage result are input into the pre-constructed cross-scale finite element model to obtain the internal and external damage evolution mechanism of the track under different working conditions and the updated cross-scale finite element model. The track structure monitoring data, the first damage result, and the internal and external damage evolution mechanism under different working conditions are input into the pre-constructed damage state prediction model to output the damage state prediction result of the track to be monitored. Based on the meteorological monitoring data, the first damage result, the damage state prediction result, and the updated cross-scale finite element model, the lifetime of the track to be monitored is predicted.

2. The method for predicting the damage state of ballastless track according to claim 1, characterized in that, The damage location identification model is constructed based on a clustering algorithm, and the first damage result includes damage type, damage location, damage degree, and damage range; The feature extraction of the temperature monitoring data, displacement monitoring data, strain monitoring data, and vibration monitoring data of the track structure includes: The time-domain features of the temperature monitoring data, displacement monitoring data, strain monitoring data and vibration monitoring data of the track structure are extracted respectively to obtain the extreme value of the temperature gradient of the track structure, the peak value of the displacement, the vibration dominant frequency shift and the area of ​​the high stress region; The step of inputting the extracted features into a pre-built damage location identification model to obtain the first damage result of the track to be monitored includes: The extreme values ​​of the temperature gradient, peak displacement, vibration frequency shift, and area of ​​the high-stress region of the track structure are input into the trained damage location identification model. Based on the clustering results, the damage type, damage location, damage degree, and damage range of the track to be monitored are obtained.

3. The method for predicting the damage state of ballastless track according to claim 1, characterized in that, The multi-scale finite element model includes the train, rails, fasteners, track slabs, mortar layer, base plate, and lower foundation layer; The track slab adopts the standard dimensions of the track to be monitored, the base plate is sized according to design specifications, the thickness of the mortar layer is set according to construction standards, the cross-section of the rail adopts the CHN60 standard section, the fastener is a WJ-7 type fastener, the lower foundation layer includes two working conditions: roadbed and bridge, the base plate adopts a hexahedral structured grid, the track slab and base plate adopt a concrete plastic damage constitutive model, the mortar layer adopts a viscoelastic constitutive model, the track slab and mortar layer adopt a nonlinear cohesive force model, and the rail adopts an elastoplastic constitutive model.

4. The method for predicting the damage state of ballastless track according to claim 3, characterized in that, A zero-thickness cohesive unit is provided at the contact interface between the track slab and the mortar layer. This zero-thickness cohesive unit is used to simulate the damage evolution process of the contact interface between the track slab and the mortar layer. A zero-thickness cohesive unit is also provided between the mortar layer and the base plate. The parameters of the zero-thickness cohesive unit are determined based on the roughness and bonding strength of the contact interface between the mortar layer and the base plate.

5. The method for predicting the damage state of ballastless track according to claim 3, characterized in that, The meteorological monitoring data includes ambient temperature monitoring data, solar radiation monitoring data, and wind speed monitoring data; In the simulation process of the multi-scale finite element model, the dynamic load of the train is simulated by applying a moving wheel-rail force on the top surface of the rail; the ambient temperature load includes two parts: overall temperature rise and fall and extreme gradient temperature. The range of overall temperature rise and fall is determined according to the extreme temperature conditions of different climate regions, and the extreme gradient temperature is determined based on solar radiation monitoring data, annual temperature cycle changes, and daily temperature cycle differences; the thermal boundary conditions include solar radiation thermal boundary conditions and wind speed thermal boundary conditions. The solar radiation thermal boundary conditions are based on the radiative heat flux applied to the exposed surfaces of the track slab and fastening system to simulate the heating process of the track structure by solar shortwave radiation. The thermal boundary condition of wind speed is used to simulate the convective heat transfer capacity of wind speed on the surface of the track structure.

6. The method for predicting the damage state of ballastless track according to claim 5, characterized in that, The damage state prediction model is constructed based on an improved particle swarm optimization algorithm and a long short-term memory network. The improved particle swarm optimization algorithm is used to optimize the hyperparameters in the long short-term memory network, and the long short-term memory network is used to extract the evolution law of the internal and external damage evolution mechanism under different working conditions in the time dimension.

7. The method for predicting the damage state of ballastless track according to claim 6, characterized in that, The training sample set of the damage state prediction model includes multiple training samples. Each training sample includes acquired multidimensional data, the first damage result corresponding to the multidimensional data, the train dynamic load, ambient temperature load, and thermal boundary conditions in the updated cross-scale finite element model obtained through the inversion algorithm, as well as the damage distribution map of each component corresponding to the multidimensional data, the damage index change curve over time, the damage evolution trend prediction result of the preset part, and the overall health level.

8. The method for predicting the damage state of ballastless track according to any one of claims 1-7, characterized in that, The evolution mechanism of internal and external damage under different working conditions includes material damage variables, interface crack evolution process, and interlayer stress distribution of components.

9. The method for predicting the damage state of ballastless track according to any one of claims 1-7, characterized in that, The damage status prediction results of the track to be monitored include damage distribution maps of multiple components, damage index change curves over time, damage evolution trend prediction results of preset locations, and overall health level.

10. A device for predicting the damage state of ballastless track, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 9.

Images