Wheel-rail damage monitoring method based on damage tolerance design and digital twin driving

Abstract

The application provides a wheel-rail damage monitoring method based on damage tolerance design and digital twin driving, and belongs to the technical field of wheel-rail damage monitoring. The method comprises the following steps: 1) constructing a wheel-rail system digital twin body module; 2) establishing a wheel-rail damage tolerance database; 3) deploying a multi-source sensing system; 4) mapping a damage state in real time; 5) intelligently judging damage to the limit; 6) a machine learning optimization module; and 7) generating a hierarchical early warning instruction. The application can solve the problems that the evaluation index of data feedback of the existing twin data-driven damage monitoring model and the existing damage tolerance design cannot be applied to the service process of the wheel-rail system in real time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wheel-rail damage monitoring, in particular to a wheel-rail damage monitoring method based on damage tolerance design and digital twin driving. BACKGROUND

[0002] As one of the most core key components of rail transit, the service behavior of wheel-rail directly relates to the safe operation of trains. Once the wheel-rail fails due to fatigue and the corresponding repair measures are not taken in time, it will cause disastrous accidents of train destruction and human casualties. Therefore, in order to ensure the safe service performance of the wheel-rail system under complex environment, it is necessary and urgent to carry out research on intelligent monitoring and safe service evaluation of train wheel-rail damage.

[0003] At present, the research on the service safety of wheel-rail mainly focuses on the research of single macro factor, and basically does not involve the safety research of wheel-rail service damage, especially the service reliability evaluation research based on the damage tolerance of wheel-rail material. Therefore, it is necessary to carry out innovative research on the service safety evaluation technology of wheel-rail under complex environment considering the damage tolerance of material. At the same time, the digital twin technology is mainly used for the online monitoring of tool wear based on general wear model in the manufacturing process of lathe and other manufacturing industries, and the application of digital twin technology in the field of rail transit is mainly concentrated in the whole life cycle monitoring of rail transit equipment, and basically does not involve the dynamic monitoring and life prediction based on the service damage behavior of material wheel-rail. Therefore, it is necessary to construct a virtual digital twin based on the physical entity of wheel-rail, realize the virtual-real interaction of wheel-rail damage twin data, take the damage tolerance design value of wheel-rail as the damage state monitoring threshold, and clear the wheel-rail damage state early warning mechanism and the dynamic intelligent monitoring technology of wheel-rail service life under virtual-real interaction. This has important theoretical support and technical guidance for ensuring the safe operation and reliable service of high-speed railway wheel-rail under complex environment. SUMMARY

[0004] The present application provides a wheel-rail damage monitoring method based on damage tolerance design and digital twin driving, which can solve the problems of evaluation index of data feedback of existing twin data driven damage monitoring model and real-time application of existing damage tolerance design in wheel-rail system service process.

[0005] To achieve the above purpose, the present application adopts the following technical scheme:

[0006] The wheel-rail damage monitoring method based on damage tolerance design and digital twin driving comprises:

[0007] 1) Constructing a wheel-rail system digital twin body module: integrating a physical sensor network, a data processing unit and a virtual twin model, for realizing the interactive mapping of wheel-rail physical entity and virtual twin model;

[0008] 2) Establish a wheel-rail damage tolerance database: store the initial damage tolerance threshold determined based on the wheel-rail material fatigue damage analysis research, and the residual life prediction model for calculating the residual life based on the initial damage tolerance threshold, the initial damage tolerance threshold includes the critical crack size, the crack propagation rate threshold and the residual life prediction model; based on the real-time environmental parameters and the material degradation model, the initial damage tolerance threshold is adjusted to obtain the corrected dynamic damage tolerance threshold, the dynamic damage tolerance correction is realized, and the dynamic damage tolerance threshold is updated to the wheel-rail damage tolerance database;

[0009] 3) Deploy a multi-source sensing system: deploy strain sensors, accelerometers, acoustic emission devices and damage topography monitoring units at key parts of the wheel-rail to collect multi-source sensing data during the operation of the wheel-rail;

[0010] 4) Real-time mapping of damage state: process the multi-source sensing data collected by the multi-source sensing system in step 3) through the cross-modal attention fusion network, extract the wheel-rail damage features and generate real-time damage parameters that can represent the current damage of the wheel-rail;

[0011] 5) Damage limit intelligent decision: adopt a crack parameter and damage topography dual criterion mechanism, compare the real-time damage parameters obtained in step 4) with the updated dynamic damage tolerance threshold in the wheel-rail damage tolerance database in step 2), obtain a comparison result containing the damage overrun degree, and trigger an alarm according to the comparison result;

[0012] 6) Machine learning optimization module: based on historical damage data and multi-source sensing data collected in step 3), through self-supervised learning and virtual simulation training of the virtual twin model in step 1), optimize the residual life prediction model and the cross-modal attention fusion network;

[0013] 7) Generate a hierarchical warning instruction: generate a first, second and third level warning signal according to the damage overrun degree in the comparison result in step 5), and push the warning signal to the operation and maintenance terminal.

[0014] In this specification, the wheel-rail system digital twin module in step 1) includes the following three-layer structure:

[0015] i) Physical layer: a distributed sensor network and an edge computing node deployed on the wheel-rail, the distributed sensor network is a specific implementation form of the physical sensor network, and the edge computing node is used for preliminary processing of the data collected by the sensor;

[0016] ii) Virtual layer: including multi-scale modeling architecture, macro-scale simulating crack propagation using wheel-rail finite element model, meso-scale simulating grain slip using crystal plasticity finite element model, micro-scale simulating dislocation evolution using molecular dynamics model, and micro-meso-macro damage correlation is established through the above multi-scale modeling;

[0017] iii) Data layer: integrating real-time monitoring data stream and historical damage database into spatio-temporal correlation database, and deploying distributed machine learning inference engine in the spatio-temporal correlation database, the inference engine is used to perform damage state classification task in real time, and works with the data processing unit.

[0018] In the specification, the wheel-rail damage tolerance database establishment process of step 2) includes:

[0019] i) Establishing a three-dimensional wheel-rail fatigue crack propagation model with an initial crack: pre-preparing initial cracks of different lengths and angles in a full-size wheel-rail finite element model, obtaining the stress state of the wheel-rail fatigue crack tip after applying wheel-rail contact load; determining 15% of the current crack length as the crack propagation step to form a new wheel-rail crack morphology, repeating the crack propagation simulation until the crack length reaches the critical crack size, obtaining the function relationship between the cycle period and the crack length through polynomial fitting, and obtaining the function relationship between the crack length and the stress intensity factor range through polynomial fitting based on the fatigue crack propagation theory; extracting stress parameters under different crack lengths, calculating the stress intensity factor and the reference stress at the crack tip, and then obtaining the function relationship between the crack length and the stress intensity factor, and the function relationship between the crack length and the reference stress through polynomial fitting;

[0020] ii) Based on the stress state of the wheel-rail material fatigue crack tip, the safety evaluation of the wheel-rail material containing crack defects is carried out: the ratio of the stress intensity factor to the material fracture toughness is used to represent the fracture ratio of the wheel-rail material resisting fracture failure, and the ratio of the reference stress to the flow stress is used to represent the load ratio of the wheel-rail material resisting plastic limit failure; based on the load ratio and the fracture ratio, the wheel-rail safety evaluation based on the BS7910 standard is carried out, and the first level evaluation is used for the wheel-rail material; the function relationship between the crack length and the stress intensity factor, the function relationship between the crack length and the reference stress, and the standard failure evaluation curve are intersected to calculate the critical thermal crack size of the wheel, and further polynomial fitting is performed on the crack length and the cycle number to obtain the residual life corresponding to the critical crack size of the wheel;

[0021] iii) Compiling damage tolerance design spectrum: combining a large number of wheel-rail fatigue contact test and simulation analysis results to compile the damage tolerance design spectrum, in addition to crack data, the damage tolerance design spectrum also includes material wear amount and damage morphology as damage tolerance evaluation criteria;

[0022] iv) Establishing an environmental-load-material performance dynamic coupling module: the module involves temperature effect correction parameters and load fluctuation correction parameters, and based on the above correction parameters, dynamic critical crack size calculation is carried out to realize dynamic adjustment of damage tolerance threshold, which matches the "adjustment of damage tolerance threshold based on real-time environmental parameters and material degradation model".

[0023] In the specification, the multi-source sensing system in step 3) specifically includes:

[0024] i) Pressure sensor array: for monitoring wheel-rail contact stress distribution, providing stress data support for subsequent damage analysis;

[0025] ii) Acoustic emission sensor: for capturing characteristic frequencies during wheel-rail crack propagation, identifying crack propagation behavior;

[0026] iii) Embedded piezoelectric accelerometer: for collecting vibration energy characteristics during wheel-rail operation, reflecting wheel-rail contact state;

[0027] iv) Damage topography monitoring unit: including high-speed camera and laser confocal scanner, for obtaining the topographic information of wheel-rail damage, providing topographic data for damage state evaluation.

[0028] In the specification, the cross-modal attention fusion network in step 4) includes the following parts:

[0029] i) Modal feature extraction branch: 1D-CNN, 2D-CNN and ResNet-50 are used to process acoustic emission signals, vibration signals and topographic images in multi-source sensing data, respectively, to extract damage features corresponding to each modal;

[0030] ii) Cross-modal attention fusion module: through self-attention mechanism, the correlation weight between different modalities is learned, for example, the strong correlation between acoustic emission frequency and vibration energy during crack propagation, to realize effective fusion of multi-modal features;

[0031] iii) Contrastive learning optimization: NT-Xent loss function is used for training to maximize the similarity of similar damage features and maximize the difference of different damage features, improving the recognition ability of the model for rare damage patterns (such as abnormal wear);

[0032] iv) Damage mode vector construction and risk coefficient calculation: based on the damage topographic features of wheel-rail materials, a damage mode vector is constructed, and the topographic vector is fused with mechanical parameters and input into a virtual twin model, and a topographic risk coefficient is output, and the critical value of the topographic risk coefficient is determined according to engineering experience and relevant evaluation standards, providing basis for subsequent damage tolerance decision.

[0033] In the specification, the damage tolerance intelligent decision-making double-criterion mechanism in step 5) is specifically:

[0034] i) Dynamic crack parameter criterion: when the crack size obtained by real-time monitoring is greater than the damage tolerance design size in the damage tolerance database, or the crack propagation rate is greater than the dynamic threshold in the database, it is determined that the criterion is met;

[0035] ii) Morphology risk coefficient criterion: when the calculated damage morphology risk coefficient is greater than the preset critical value, it is determined that the criterion is met;

[0036] iii) Early warning trigger rule: any of the above criteria is met, and the early warning is triggered, and if both criteria are met, the early warning level is raised to ensure that damage risks are not missed, and the logic of "triggering early warning" is consistent.

[0037] In this specification, the machine learning optimization module of step 6) includes the following processes:

[0038] i) Self-supervised pre-training process: by rotating the unmarked vibration data for enhancement processing, the TCN network is trained to predict the signal rotation angle, and the damage data is enhanced; freeze the bottom structure of the TCN network, only fine-tune the top network for crack propagation direction prediction, so as to realize model pre-training based on normal wheel-rail operation data, and lay a foundation for subsequent model optimization;

[0039] ii) Twin simulation-reinforcement learning closed loop: simulate a large number of virtual damage scenarios in the virtual twin model, such as extreme loads, material defects, etc., to generate virtual training data; use reinforcement learning to train the damage assessment model, and use "minimize prediction error" as the reward function to quickly iterate and optimize the model parameters in the virtual environment;

[0040] iii) Model deployment: when the model prediction error is lower than the preset threshold, the optimized model is deployed to the corresponding physical system of the physical sensor network to realize the actual application of the model, and the "virtual pre-training-physical fine-tuning" iterative optimization process is completed.

[0041] In this specification, the hierarchical early warning instruction in step 7) is associated with the operation and maintenance strategy library, and the specific association relationship is:

[0042] i) First-level (observation level) early warning: trigger the suggestion of rail grinding or wheel lapping, and start the fine-tuning process of the self-supervised model to optimize the monitoring model in time;

[0043] ii) Second-level (maintenance level) early warning: generate a mandatory instruction for rail grinding or wheel lapping, and activate the virtual simulation training process to optimize the maintenance scheme based on virtual scenarios;

[0044] iii) Level 3 (emergency shutdown level) warning: The train control system realizes speed limiting operation, and triggers the reinforcement learning parameter optimization process to quickly improve the adaptability of the model to extreme damage scenarios;

[0045] The associated action of the operation and maintenance strategy library ensures that the warning signal can be converted into actual operation and maintenance measures, realizing the closed-loop management of "monitoring-warning-operation".

[0046] In the specification, the specific specifications of the initial cracks prepared in step 2) i) are: lengths of 0.5 mm, 1.0 mm, and 1.5 mm, and angles of 30°, 45°, and 60°, respectively; the applied wheel-rail contact load is simulated based on the Hertz contact theory under the condition of 21t axle load, ensuring that the crack propagation simulation is consistent with the actual wheel-rail service conditions; the stress intensity factor is calculated using the M integral method, and the reference stress at the crack tip is calculated using the effective net section method, which provides an accurate parameter basis for subsequent function fitting and safety evaluation.

[0047] In the specification, the damage mode vector constructed in step 4) iv) specifically includes topographic feature parameters such as the area, depth, and shape irregularity of wheel-rail damage; the mechanical parameters include contact stress monitored by the pressure sensor array, stress intensity factor calculated by stress, etc.; when the topographic vector and mechanical parameters are fused, feature normalization processing is adopted to ensure the balanced weight of different dimension parameters; the determination of the critical value of the topographic danger coefficient needs to be combined with the fatigue performance test data of wheel-rail materials, historical damage failure cases, and industry wheel-rail damage evaluation standards to ensure the rationality and reliability of the critical value, providing an accurate basis for damage-to-limit decision.

[0048] In summary, the present application has at least the following beneficial effects:

[0049] The present application provides a wheel-rail damage monitoring method based on damage tolerance design and digital twin driving. Compared with the existing traditional fatigue analysis method based on nominal stress or crack-free life, the present application considers the inevitable existence of initial defects or small cracks in the wheel-rail system, introduces the mature "damage tolerance" concept in the aviation, nuclear power and other fields into the field of wheel-rail fatigue analysis, and quantitatively calculates the probability and life of these cracks expanding under service load until reaching the critical size. This is more in line with engineering practice, especially for wheel-rail materials that bear high-frequency and high-stress alternating loads.

[0050] Compared with the existing wheel-rail damage monitoring and control measures, the twin data-driven wheel-rail damage monitoring method can not only monitor the wheel-rail damage state in real time, but also can combine the damage tolerance design threshold to feed back to the wheel-rail service system in time by controlling the rail grinding, wheel turning repair and other repair measures, and realize the "virtual-real interaction, and virtual control real". The wheel-rail damage intelligent monitoring new technology of damage tolerance and digital twin fusion will provide important theory and use value for the safe and reliable service of wheel-rail. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0052] Figure 1 The schematic diagram of the wheel-rail damage monitoring method based on damage tolerance design and digital twin driving involved in the present application.

[0053] Figure 2 The schematic diagram of the wheel-rail rolling test bed digital twin involved in the present application.

[0054] Figure 3a The schematic diagram of the wheel-rail rolling test bed digital twin system involved in the present application.

[0055] Figure 3b The schematic diagram of the twin data-driven running process involved in the present application.

[0056] Figure 4a The schematic diagram of the wheel-rail three-dimensional finite element model involved in the present application.

[0057] Figure 4b1 The schematic diagram of the wheel-rail prefabricated crack (angle 30°) involved in the present application.

[0058] Figure 4b2 The schematic diagram of the wheel-rail prefabricated crack (angle 45°) involved in the present application.

[0059] Figure 4b3 The schematic diagram of the wheel-rail prefabricated crack (angle 60°) involved in the present application.

[0060] Figure 5a The schematic diagram of the relationship between the equivalent stress intensity factor and the crack length (the initial crack length is 0.5mm) in the crack propagation process involved in the present application.

[0061] Figure 5bA graph showing the relationship between the equivalent stress intensity factor and the crack length (initial crack length of 1.0 mm) in the crack propagation process involved in the present application.

[0062] Figure 5c A graph showing the relationship between the equivalent stress intensity factor and the crack length (initial crack length of 1.5 mm) in the crack propagation process involved in the present application.

[0063] Figure 6a A graph showing the change of the reference stress with the crack propagation length (initial crack length of 0.5 mm) under a 21t axle load involved in the present application.

[0064] Figure 6b A graph showing the change of the reference stress with the crack propagation length (initial crack length of 1.0 mm) under a 21t axle load involved in the present application.

[0065] Figure 6c A graph showing the change of the reference stress with the crack propagation length (initial crack length of 1.5 mm) under a 21t axle load involved in the present application.

[0066] Figure 7a A graph showing the wheel-rail failure evaluation result (initial crack length of 0.5 mm) under a 21t axle load involved in the present application.

[0067] Figure 7b A graph showing the wheel-rail failure evaluation result (initial crack length of 1.0 mm) under a 21t axle load involved in the present application.

[0068] Figure 7c A graph showing the wheel-rail failure evaluation result (initial crack length of 1.5 mm) under a 21t axle load involved in the present application.

[0069] Figure 8 A graph showing the cross-modal attention fusion network involved in the present application.

[0070] Figure 9 A graph showing the damage-to-limit intelligent decision and grading early warning instruction flow involved in the present application.

[0071] Figure 10 A graph showing the machine learning optimization module flow involved in the present application. DETAILED DESCRIPTION

[0072] Hereinafter, only certain exemplary embodiments are simply described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the embodiments of the present application. Therefore, the drawings and the description are considered to be exemplary in nature rather than limiting.

[0073] The following disclosure provides many different embodiments or examples for implementing different aspects of the present embodiments. For simplicity of the disclosure, the following description often refers to particular examples of components and arrangements of components. But one skilled in the art will appreciate that the embodiments of the present embodiments can be practiced in a variety of arrangements of components and that the embodiments of the present embodiments can be realized in other embodiments that depart from these specific arrangements.

[0074] The embodiments of the present embodiments will be described in detail below with reference to the attached drawings.

[0075] As Figure 1 shown, the present embodiments provide a wheel-rail damage monitoring method based on damage tolerance design and digital twin driving, including the following steps:

[0076] 1) Construct a wheel-rail system digital twin module: This module integrates a physical sensor network, a data processing unit, and a virtual twin model, which is used to realize the interactive mapping of the wheel-rail physical entity and the virtual twin model; wherein the virtual twin model can provide the virtual scene required for subsequent virtual simulation training, and the data processing unit can assist the preliminary operation of the subsequent damage monitoring related model, and provide the basic framework for the overall monitoring process;

[0077] 2) Establish a wheel-rail damage tolerance database: First, store two types of core content - one is the initial damage tolerance threshold determined based on wheel-rail material fatigue damage analysis and research (the initial damage tolerance threshold is a quantitative index, including the critical crack size and the crack propagation rate threshold), and the other is the residual life prediction model based on the initial damage tolerance threshold to calculate the residual life (which belongs to the damage monitoring related model, used to output the residual life combined with real-time damage parameters); then adjust the initial damage tolerance threshold based on real-time environmental parameters and material degradation model to obtain the corrected dynamic damage tolerance threshold, realize dynamic damage tolerance correction, and update the dynamic damage tolerance threshold and the updated residual life prediction model (based on dynamic threshold optimization calculation logic) to the wheel-rail damage tolerance database;

[0078] 3) Deploy a multi-source sensing system: Deploy strain sensors, accelerometers, acoustic emission devices, and damage topography monitoring units at key parts of the wheel-rail, which are used to collect multi-source sensing data during the operation of the wheel-rail (this data is the "real-time monitoring data" mentioned in the subsequent steps, which contains stress, vibration, acoustic emission, damage topography, and other dimension information);

[0079] 4) Real-time damage state mapping: Through the multi-source sensing data collected in step 3) processing by the cross-modal attention fusion network (belongs to the damage monitoring related model, for multi-source data processing), the wheel-rail damage features are extracted and real-time damage parameters (including real-time crack size, real-time crack propagation rate, damage topography feature parameters, etc.) that can represent the current damage situation are generated;

[0080] 5) Damage-to-limit intelligent decision: Adopting a double-criterion mechanism of crack parameters and damage topography, the real-time damage parameters obtained in step 4) are compared with the updated dynamic damage tolerance threshold in the database in step 2), and a clear comparison result (which directly contains the information of "whether the damage is over limit" and "the extent of damage over limit") is obtained, and the corresponding level of warning is triggered according to the comparison result;

[0081] 6) Machine learning optimization module: Based on the historical damage data and the multi-source sensing data (i.e. real-time monitoring data) collected in step 3), through self-supervised learning and virtual simulation training of the virtual twin model in step 1), the remaining life prediction model in step 2), the cross-modal attention fusion network and other damage monitoring related models in step 4) are optimized, and the adjustment logic of the dynamic damage tolerance threshold is corrected;

[0082] 7) Generate graded warning instructions: According to the damage over limit degree in the comparison result in step 5), generate a first (observation level), second (maintenance level), and third (emergency shutdown level) warning signal, and push the warning signal to the operation and maintenance terminal to realize the direct association of warning and operation and maintenance needs.

[0083] In some embodiments, the digital twin module in step 1) includes: i) physical layer: distributed sensor network and edge computing node deployed on wheel-rail; ii) virtual layer: including multi-scale modeling architecture, macro-scale using wheel-rail finite element model to simulate crack propagation, meso-scale using crystal plastic finite element model to simulate grain slip, micro-scale using molecular dynamics model to simulate dislocation evolution, establishing micro-meso-macro damage correlation; iii) data layer: integrating the spatiotemporal correlation database of real-time monitoring data stream and historical damage database, and deploying distributed machine learning inference engine to perform real-time damage state classification tasks.

[0084] In some embodiments, the damage tolerance database establishment process of step 2) includes:

[0085] i) there is an initial crack a 0a three-dimensional wheel-rail fatigue crack propagation model is established. In the wheel-rail full-size finite element model, initial cracks of different lengths and angles are pre-prepared, and the stress state of the wheel-rail fatigue crack tip is obtained after the wheel-rail contact load is applied; 15% of the current crack length is determined as the crack propagation step to form a new wheel-rail crack morphology, and the crack propagation simulation is repeated until the crack length reaches a , and the function relationship between the cycle period N and the crack length a is obtained by polynomial fitting; the function relationship between the crack length a and the stress intensity factor range ΔK is obtained by polynomial fitting according to the Paris formula , wherein C and n are material-related parameters; stress parameters at different crack lengths are extracted, the stress intensity factor is calculated by M integral, the reference stress at the crack tip is calculated by the effective net section method, and the function relationship between the crack length and the stress intensity factor and the function relationship between the crack length and the reference stress are obtained by polynomial fitting.

[0086] ii) Based on the stress state of the wheel-rail material fatigue crack tip, the safety assessment of the wheel-rail material containing crack defects is carried out. The ratio of the stress intensity factor K to the material fracture toughness K IC is used to represent the fracture ratio K r of the wheel-rail material resisting fracture failure; the ratio of the reference stress S ref to the rheological stress S f ( , σ Y is the yield strength, σ u is the tensile strength) is used to represent the load ratio L r of the wheel-rail material resisting plastic limit failure; based on the load ratio L r and the fracture ratio K r , the safety assessment of the wheel-rail based on the BS7910 standard is carried out, and the first-level assessment is used for the wheel-rail material; the function relationship between the crack length and the stress intensity factor and the function relationship between the crack length and the reference stress are intersected with the standard failure assessment curve to calculate the critical thermal crack size of the wheel, and further polynomial fitting of the crack length and the cycle number can obtain the residual life corresponding to the critical crack size of the wheel.

[0087] iii) A damage tolerance design spectrum is compiled by combining a large number of wheel-rail fatigue contact test and simulation analysis results, and in addition to crack data, material wear and damage morphology should also be included as damage tolerance evaluation criteria.

[0088] Primary evaluation formula:

[0089] ;

[0090] : ;

[0091] ;

[0092] iv) Establishing an environmental-load-material performance dynamic coupling module, involving temperature effect correction parameters K IC (T) and load fluctuation correction parameters β , based on which dynamic critical crack size a c is calculated, realizing dynamic adjustment of damage tolerance threshold.

[0093] K IC (T) = K IC0 · exp(- Q / RT); β =1+0.2(Δ P / P 0 ); a c = a c0 · K IC (T) / K IC0 · β ;

[0094] In the formula, Q is the activation energy, R is the gas constant, and Δ P is the load fluctuation amplitude.

[0095] In some embodiments, the multi-element sensing system in step 3) uses a pressure sensor array to monitor the wheel-rail contact stress distribution; an acoustic emission sensor captures crack propagation characteristic frequencies; an embedded piezoelectric accelerometer collects vibration energy characteristics; and a damage topography monitoring unit includes a high-speed camera, a laser confocal scanner, etc.

[0096] In some embodiments, the cross-modal attention fusion network of step 4) comprises: i) a modal feature extraction branch: processing the acoustic emission signal, the vibration signal and the topography image by 1D-CNN, 2D-CNN and ResNet-50 respectively; ii) a cross-modal attention fusion module: learning the correlation weight between modes by a self-attention mechanism, such as the strong correlation between the acoustic emission frequency and the vibration energy during crack propagation; iii) a contrastive learning optimization: using an NT-Xent loss function to maximize the similarity of the same damage features and the difference of the different damage features, and improving the recognition ability of the model to rare damage modes (such as abnormal wear); iv) constructing a damage mode vector based on the wheel-rail material damage topography features, and inputting the topography vector and the mechanical parameters into the twin body model to output a topography risk coefficient ζ The critical value of the topography risk coefficient is determined according to experience and relevant evaluation standards.

[0097] In some embodiments, the damage-to-limit decision in step 5) adopts a double-criteria mechanism: i) a dynamic crack parameter criterion: the real-time crack size is greater than the damage tolerance design size or the crack propagation rate is greater than the dynamic threshold; ii) a topography risk coefficient criterion: the damage topography risk coefficient ζ is greater than the critical value; iii) triggering an early warning when any of the above criteria is met, and upgrading the early warning level when both criteria are met.

[0098] In some embodiments, the machine learning optimization module of step 6) comprises: i) a self-supervised pre-training process: enhancing the damage data by training a TCN network to predict the signal rotation angle through rotation enhancement of unlabeled vibration data, freezing the bottom layer network to fine-tune the top layer for crack propagation direction prediction, and realizing model pre-training based on normal wheel-rail operation data; ii) a twin simulation-reinforcement learning closed loop: simulating a large number of virtual damage scenarios in the digital twin, such as extreme loads and material defects, to generate virtual training data; training the damage assessment model using reinforcement learning, taking "minimum prediction error" as the reward function, and quickly iterating and optimizing the model parameters in the virtual environment, and then deploying the optimized model to the physical system. Through this efficient iterative method of "virtual pre-training-physical fine-tuning", the adaptability of the model to extreme working conditions is improved; iii) when the prediction error is lower than the threshold, the optimized model is deployed to the physical system.

[0099] In some embodiments, the hierarchical early warning instruction of step 7) is associated with an operation and maintenance strategy library: i) a first-level early warning triggers a rail grinding / wheel lapping suggestion, and starts self-supervised model fine-tuning; ii) a second-level early warning generates a rail grinding / wheel lapping instruction, and activates virtual simulation training; iii) a third-level early warning links the train control system to limit speed, and triggers reinforcement learning parameter optimization.

[0100] The technical concept of the present application is as follows:

[0101] The wheel-rail damage monitoring method based on damage tolerance design and digital twin driving, as shown in the formula (I), comprises the following steps: Figure 1 The steps are as follows:

[0102] Step 1) Building a wheel-rail system digital twin module: integrating a physical sensor network, a data processing unit and a virtual twin model, specifically involving: i) physical layer: deploying a distributed sensor network and an edge computing node on the wheel-rail; ii) virtual layer: including a multi-scale modeling architecture, a macro-scale using a wheel-rail finite element model to simulate crack propagation, a meso-scale using a crystal plastic finite element model to simulate grain slip, and a micro-scale using a molecular dynamics model to simulate dislocation evolution, establishing a micro-meso-macro damage correlation; iii) data layer: integrating a spatiotemporal correlation database of real-time monitoring data stream and historical damage database, and deploying a distributed machine learning inference engine to perform real-time damage state classification tasks.

[0103] Since the real wheel-rail system is in an open environment, there are many influencing factors such as environment, terrain and operating conditions during its service, and it is difficult and time-consuming to track and measure data on site. Therefore, in this embodiment, a commonly used laboratory double-disc rolling wheel-rail rolling contact simulation test machine is used to establish a wheel-rail rolling test bed digital twin as shown in the formula (II), which involves the multi-source sensor arrangement in step 3) and the real-time mapping of damage state in step 4). Figure 2

[0104] First, the test data is collected and stored. A multi-type sensor system is deployed on the MJP-30A wheel-rail rolling test platform, including pressure, temperature and torque sensors to realize real-time monitoring of test parameters. The system adopts a layered data management architecture to realize full-process data processing in the following ways: i) dynamic data acquisition system: sensors monitor test parameters such as test force, friction coefficient and slip rate in real time, and data streams are stored in the test bed local special directory in txt format. Corresponding data tables are created in the MySQL database in advance, covering static parameters such as sample material properties and classified storage structures such as digital twin system calculation results; ii) data transmission and storage: dynamic data is transmitted across devices through the SMB network protocol. In the same local area network, by configuring the test equipment IP address and shared folder permissions, a PC-based Unity engine is developed to develop a special interface module, which scans the target folder at a set period, intelligently identifies txt file updates and parses structured data, and finally stores them into the corresponding table in the database; iii) static parameters are directly input and stored into the specified data table through the Unity interactive interface; digital twin data is generated by the wear calculation program and written into the database calculation layer through the API interface, forming a complete test data ecosystem.

[0105] ​Secondly, we will conduct wear prediction of wheel-rail samples driven by twin data. Based on Tγ / A The wear model first divides the entire experimental process into intervals 1, 2, 3, ... according to the time column stored in the database. i , i +1、……、 n Based on the real-time data recorded in the experiment, time = i Wear calculations and predictions are performed based on the test force, creep rate, friction coefficient, number of cycles of the wheel specimen, and geometric parameters of the wheel-rail specimen, and the specimen radius is iteratively updated according to the wear threshold.

[0106] Finally, a digital twin system for the wear behavior of wheel-rail samples on a wheel-rail rolling test bench was built in Unity, such as... Figure 3a As shown. Based on this, real-time mapping of the wear state (wear amount, wear depth, etc.) of wheel-rail materials was achieved, and its twin data-driven operation process is as follows. Figure 3b As shown.

[0107] Step 2) Establish a wheel-rail damage tolerance database: Store damage tolerance thresholds determined based on wheel-rail material fatigue damage analysis, including critical crack size, crack propagation rate thresholds, and remaining life prediction models; adjust the damage tolerance thresholds based on real-time environmental parameters and material degradation models to achieve dynamic damage tolerance correction. This embodiment designs damage tolerance based on wheel-rail rolling contact fatigue crack propagation behavior, as follows:

[0108] i) Establishment of an initial crack a 0 A three-dimensional wheel-rail fatigue crack propagation model was established. A three-dimensional finite element model of the wheel-rail system was created. Figure 4a The stress state and cycle period at the tip of the wheel-rail fatigue crack were obtained. N Corresponding wheel-rail fatigue crack propagation length a Initial cracks of different lengths (0.5mm, 1.0mm, 1.5mm) and angles (30°, 45°, 60°) were implanted into the wheel-rail material, such as... Figure 4b1 , Figure 4b2 and Figure 4b3 As shown;

[0109] A simulation of wheel-rail rolling contact under a 21t axle load was established based on Hertz contact theory to obtain the stress state at the wheel-rail fatigue crack tip. Based on this stress state, the stress intensity factor was calculated using the M-integral method, the crack propagation direction was determined using the maximum energy release rate criterion, and the crack propagation rate was calculated using the Paris formula. The formula is as follows: ,in, C Take 4.5966×10⁻¹³, n Take 2.8805;

[0110] 15% of the current crack length is determined as the crack propagation step, a new wheel-rail crack profile is formed, implanted into the original crack position, and the stress state of the current length of the wheel-rail fatigue crack tip is obtained under the load, and the crack propagation is repeated until the crack length reaches a c According to the polynomial fitting, the cycle period N is obtained as a function of the crack length a , and the Paris formula is combined to obtain the function relationship between the crack length a and the stress intensity factor range ΔK through polynomial fitting; the stress parameters under different wheel-rail crack lengths are extracted, the stress intensity factor is calculated through M integral, and the reference stress at the crack tip is calculated through the effective net section method. The function relationship between the wheel-rail crack length and the stress intensity factor and the function relationship between the wheel-rail crack length and the reference stress are polynomial fitted, and the relationship between the equivalent stress intensity factor and the crack length during the crack propagation process under the 21t axle load is as follows: Figure 5a (the initial crack length is 0.5mm), Figure 5b (the initial crack length is 1.0mm), and Figure 5c (the initial crack length is 1.5mm), and the reference stress changes with the crack propagation length under the 21t axle load as follows: Figure 6a (the initial crack length is 0.5mm), Figure 6b (the initial crack length is 1.0mm), and Figure 6c (the initial crack length is 1.5mm).

[0111] ii) Based on the stress state at the wheel-rail fatigue crack tip, the safety evaluation of the wheel-rail with crack defects is carried out.

[0112] First, the stress intensity factor is calculated to represent the fracture ratio K r of the wheel-rail resistance to fracture failure, and the formula is:

[0113] ; (1)

[0114] In the formula: K is the stress intensity factor at the crack tip; K ⅠC is the fracture toughness of the material.

[0115] The reference stress S ref is calculated to represent the load ratio L r of the wheel-rail resistance to plastic limit failure, and the formula is:

[0116] ; (2)

[0117] (3)

[0118] where: F is the load-bearing structure, F Y is the plastic limit load of the structure with cracks, S ref is the reference stress, S f is the rheological stress, σ Y is the yield strength, σ u is the tensile strength.

[0119] Then, based on the load ratio L r and the fracture ratio K r The wheel-rail safety assessment based on BS7910 is carried out, and the first-level evaluation is used for wheel-rail materials (see formulas (4)-(7)), and the formulas are as follows:

[0120] (4)

[0121] (5)

[0122] (6)

[0123] (7)

[0124] where: is the yield strength of the material, is the tensile strength, E is the elastic modulus of the material. The wheel-rail failure evaluation results of the 21t axle load are as follows: Figure 7a (initial crack length is 0.5mm), Figure 7b (initial crack length is 1.0mm), and Figure 7c (initial crack length is 1.5mm).

[0125] The cross-modal attention fusion network of step 4) is as follows: Figure 8The shown, including: i) modal feature extraction branch: respectively using 1D-CNN, 2D-CNN and ResNet-50 process acoustic emission signal, vibration signal and topography image; ii) cross-modal attention fusion module: through self-attention mechanism to learn the correlation weight between modal, such as the strong correlation between acoustic emission frequency and vibration energy when crack propagation; iii) contrast learning optimization: using NT-Xent loss function, make the similarity of the same damage feature maximization, the difference of the different damage feature maximization, improve the model's ability to identify rare damage patterns (such as abnormal wear); iv) based on the wheel-rail material damage topography feature to construct damage mode vector, the topography vector and mechanical parameters are fused into the twin body model, and the topography risk coefficient is output ζ , according to experience and related evaluation standard to determine the critical value of topography risk coefficient.

[0126] Step 5) damage to limit intelligent decision and step 7) grading early warning instruction flow chart as Figure 9 shown: double criteria decision: using crack parameters and damage topography double criteria mechanism, compare real-time damage parameters with damage tolerance threshold and trigger early warning, realize the process as Figure 8 , specifically: i) crack size is greater than damage tolerance design size; ii) damage topography risk coefficient ζ is greater than the critical value; iii) meet any of the above criteria to trigger early warning, double criteria meet at the same time to upgrade the warning level.

[0127] Grading early warning instruction: according to the damage overrun degree to generate one level (observation level), two level (maintenance level), three level (emergency shutdown level) early warning signal, and push to operation and maintenance terminal. Specific association operation and maintenance strategy library: i) one level early warning trigger rail grinding / wheel lapping suggestion, start self-supervised model fine-tuning; ii) two level early warning generates rail grinding / wheel lapping instruction, activates virtual simulation training; iii) three level early warning linkage train control system speed limit, trigger reinforcement learning parameter optimization.

[0128] The flow chart of machine learning optimization module of step 6) is as Figure 10As shown, it includes: i) a self-supervised pre-training process: through rotation enhancement on unlabelled vibration data, a TCN network is trained to predict the signal rotation angle to realize the enhancement processing of damage data, the bottom network is frozen to fine-tune the top layer to predict the crack propagation direction, and the model pre-training based on normal wheel-rail operation data is realized; ii) a twin simulation-reinforcement learning closed loop: a large number of virtual damage scenarios such as extreme load, material defects, etc. are simulated in the digital twin, to generate virtual training data; the damage assessment model is trained by using reinforcement learning, and the "minimum prediction error" is taken as the reward function, the model parameters are quickly iterated and optimized in the virtual environment, and then the optimized model is deployed to the physical system. Through this efficient iterative way of "virtual pre-training-physical fine-tuning", the adaptability of the model to extreme working conditions is improved; iii) when the prediction error is lower than the threshold, the optimized model is deployed to the physical system.

[0129] The above-described embodiments are used to illustrate the present application, and are not intended to limit the present application, so the change of example values or the replacement of equivalent elements should still belong to the scope of the present application.

[0130] From the above detailed description, it can be clear to those skilled in the art that the present application can achieve the above-mentioned purposes, and has met the requirements of the Patent Law.

[0131] Although the preferred embodiments of the present application have been described, those skilled in the art can make further changes and modifications to these embodiments once they know the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application. The above description is only the preferred embodiments of the present application and is not intended to limit the present application. It should be noted that any modification, equivalent replacement and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

[0132] It should be noted that the above description of the process is only for example and illustration, and does not limit the scope of the present application. Those skilled in the art can make various modifications and changes to the process under the guidance of the present application. However, these modifications and changes are still within the scope of the present application.

[0133] The above has described the basic concept, and it is obvious that the above-mentioned invention disclosure is only as an example and does not constitute a limitation on the present application for those skilled in the art after reading this application. Although it is not explicitly stated here, those skilled in the art can make various modifications, improvements and modifications to the present application. Such modifications, improvements and modifications are suggested in the present application, so such modifications, improvements and modifications still belong to the spirit and scope of the exemplary embodiments of the present application.

[0134] Also, certain terminology can also be used in the description for the purpose of reference only, and thus is not necessarily limiting. For example, the terms "one embodiment" or "an embodiment" (and / or "one alternative" or "an alternative") are not necessarily mutually exclusive, unless otherwise specified. Additionally, the terms "first", "second", and / or "third" can be used merely as labels, and are not necessarily intended to signify importance or a particular order of occurrence. Moreover, terms such as "front", "back", "top", "bottom", "over", "under", "right", "left", and the like can be used for directionally purposes only, and are not necessarily intended to denote relative position or orientation unless otherwise specified. Furthermore, the terms "coupled" and "connected" and / or similar terms, as used herein, can have different meanings depending upon the context in which they are used. Therefore, these terms are used herein for clarity only, and their use should not be limited by the particular meanings taken from the context in which they are used. For example, the term "connected" can be used herein to indicate that two or more elements are in either physical or logical contact with one another, while the term "coupled" can be used herein to indicate that two or more elements are in either physical or logical contact with one another or that the two or more elements are not in contact with one another, but nonetheless still co-operate. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed terms. As used herein, the term "includes" means includes but is not limited to, or is inclusive of but not limited to.

[0135] Furthermore, to the extent that the terms "includes", "containing", "having", "with", "wherein", or the like can be interpreted to imply a numerical limitation or the necessity of concomitant recitation of a "comprising" or "consisting of" limitation, those terms shall not be interpreted so as to exclude other embodiments of the application. It will be apparent to one of ordinary skill in the art that aspects of the application can be practiced by other than the methods, systems and materials as described. Such variations are not to be regarded as a departure from the spirit and scope of the application, and all such modifications as would be recognized by one of ordinary skill in the art are intended to be included within the scope of the application. Accordingly, the application is not to be limited by the specific examples described herein, but is intended to include all such embodiments as fall within the scope of the present application.

[0136] Computer program code for carrying out operations of portions of the application can be written in any combination of one or more programming languages, including an object-oriented programming language such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, conventional procedural programming languages, such as the C programming language, Visual Basic, Fortran 2103, Perl, COBOL 2102, PHP, ABAP, dynamic programming languages, such as Python, Ruby and Groovy, or another programming language. The program code can execute entirely on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider), or in a cloud computing environment or as a service such as Software as a Service (SaaS).

[0137] Furthermore, the order of processing elements or sequences, or the use or appearance of certain terminology, throughout the above description should not be construed as limiting the application. Other steps, components, or configurations can be determined and implemented in a manner most beneficial to a particular application. For example, although the implementation of the various components described above can be embodied in hardware devices, it can also be implemented as a pure software solution, for example, as an installation on an existing server or mobile device.

[0138] Similarly, it is to be noticed that the term "comprising", used in the description, should not be interpreted as being restricted to the means listed thereafter; it does not exclude other elements or steps. It is thus to be interpreted as specifying the presence of the stated features, integers, steps or components as referred to, but does not preclude the presence or addition of one or more other features, integers, steps or components, or groups thereof. Furthermore, the description of the application is not intended to limit the application to the form disclosed herein. Various modifications and changes can be made without departing from the spirit and scope of the application as set forth in the following claims.

Claims

1. A wheel-rail damage monitoring method based on damage tolerance design and digital twin drive, characterized in that, include: 1) Construct a digital twin module for the wheel-rail system: integrate physical sensor networks, data processing units, and virtual twin models to realize interactive mapping between wheel-rail physical entities and virtual twin models; 2) Establish a wheel-rail damage tolerance database: Store the initial damage tolerance thresholds determined based on fatigue damage analysis of wheel-rail materials, as well as the remaining life prediction model calculated based on the initial damage tolerance thresholds. The initial damage tolerance threshold includes the critical crack size, crack propagation rate threshold, and remaining life prediction model; the initial damage tolerance threshold is adjusted based on real-time environmental parameters and material degradation model to obtain the corrected dynamic damage tolerance threshold, thereby realizing dynamic damage tolerance correction, and the dynamic damage tolerance threshold is updated to the wheel-rail damage tolerance database. The process for establishing the wheel-rail damage tolerance database includes: A three-dimensional wheel-rail fatigue crack propagation model with initial cracks was established: Initial cracks of different lengths and angles were prefabricated in the full-size wheel-rail finite element model. After applying wheel-rail contact load, the stress state at the crack tip was obtained. 15% of the current crack length was determined as the crack propagation step size to form a new wheel-rail crack morphology. The crack propagation simulation was repeated until the crack length reached the critical crack size. The functional relationship between the cycle period and crack length was obtained by polynomial fitting. Combined with fatigue crack propagation related theories, the functional relationship between crack length and stress intensity factor range was obtained by polynomial fitting. Stress parameters under different crack lengths were extracted, and the stress intensity factor and reference stress at the crack tip were calculated. Then, the functional relationship between crack length and stress intensity factor, and the functional relationship between crack length and reference stress were obtained by polynomial fitting. 3) Deploy a multi-source sensing system: Install strain sensors, accelerometers, acoustic emission devices and damage morphology monitoring units at key wheel-rail locations to collect multi-source sensing data during wheel-rail operation; 4) Real-time damage state mapping: The multi-source sensing data collected by the multi-source sensing system in step 3) is processed by a cross-modal attention fusion network to extract wheel-rail damage features and generate real-time damage parameters that can characterize the current damage status of the wheel and rail. 5) Intelligent damage limit judgment: The dual-criteria mechanism of crack parameters and damage morphology is adopted. The real-time damage parameters obtained in step 4) are compared with the updated dynamic damage tolerance threshold in the wheel-rail damage tolerance database in step 2) to obtain a comparison result including the degree of damage exceeding the limit. The warning is triggered according to the comparison result. 6) Machine learning optimization module: Based on historical damage data and multi-source sensor data collected in step 3), the remaining life prediction model and cross-modal attention fusion network are optimized through self-supervised learning and virtual simulation training of the virtual twin model in step 1); 7) Generate graded early warning instructions: Based on the degree of damage exceeding the limit in the comparison results of step 5), generate level 1, level 2, and level 3 early warning signals, and push the early warning signals to the operation and maintenance terminal.

2. The wheel-rail damage monitoring method based on damage tolerance design and digital twin drive according to claim 1, characterized in that, The construction of the wheel-rail system digital twin module in step 1) includes the following three-layer structure: i) Physical layer: Distributed sensor network and edge computing nodes deployed on wheel tracks. The distributed sensor network is a specific implementation of the physical sensor network, and the edge computing nodes are used to perform preliminary processing on the data collected by the sensors. ii) Virtual layer: includes a multi-scale modeling architecture. At the macro scale, the wheel-rail finite element model is used to simulate crack propagation. At the meso scale, the crystal plasticity finite element model is used to simulate grain slip. At the micro scale, the molecular dynamics model is used to simulate dislocation evolution. The micro-meso-macro damage correlation is established through the above multi-scale modeling. iii) Data layer: A spatiotemporal correlation database integrating real-time monitoring data streams and historical damage databases, and a distributed machine learning inference engine deployed in the spatiotemporal correlation database. The inference engine is used to perform damage state classification tasks in real time and works in collaboration with the data processing unit.

3. The wheel-rail damage monitoring method based on damage tolerance design and digital twin drive according to claim 1, characterized in that, Step 2) of the wheel-rail damage tolerance database establishment process also includes: Safety assessment of wheel-rail materials with crack defects is carried out based on the stress state at the fatigue crack tip. The ratio of stress intensity factor to fracture toughness is used to characterize the fracture ratio of wheel-rail materials resisting fracture failure, and the ratio of reference stress to flow stress is used to characterize the load ratio of wheel-rail materials resisting plastic limit failure. The wheel-rail safety assessment is carried out based on the load ratio and fracture ratio, and a first-level assessment is adopted for wheel-rail materials. The critical thermal crack size of the wheel is calculated by intersecting the above functional relationships between crack length and stress intensity factor, and between crack length and reference stress with the standard failure assessment curve. Furthermore, the remaining life corresponding to the critical crack size of the wheel is obtained by polynomial fitting of crack length and cycle number. Compile the damage tolerance design spectrum: The damage tolerance design spectrum is compiled based on the results of wheel-rail fatigue contact test and simulation analysis. In addition to crack data, the damage tolerance design spectrum also includes material wear amount and damage morphology as the criteria for judging damage limit. A dynamic coupling module for environment-load-material properties is established: the module involves temperature effect correction parameters and load fluctuation correction parameters. Based on the above correction parameters, dynamic critical crack size calculation is carried out to realize dynamic adjustment of damage tolerance threshold, which matches the adjustment of damage tolerance threshold based on real-time environmental parameters and material degradation model.

4. The wheel-rail damage monitoring method based on damage tolerance design and digital twin drive according to claim 1, characterized in that, The multi-source sensing system in step 3) specifically includes: i) Pressure sensor array: used to monitor the stress distribution in wheel-rail contact, providing stress data support for subsequent damage analysis; ii) Acoustic emission sensor: used to capture characteristic frequencies during the wheel-rail crack propagation process and identify crack propagation behavior; iii) Embedded piezoelectric accelerometer: used to collect vibration energy characteristics during wheel-rail operation and reflect the wheel-rail contact state; iv) Damage morphology monitoring unit: including a high-speed camera and a laser confocal scanner, used to acquire morphological information of wheel-rail damage and provide morphological data for damage status assessment.

5. The wheel-rail damage monitoring method based on damage tolerance design and digital twin drive according to claim 1, characterized in that, The cross-modal attention fusion network in step 4) includes the following components: i) Modal feature extraction branch: 1D-CNN, 2D-CNN and ResNet-50 are used to process acoustic emission signals, vibration signals and morphological images in multi-source sensor data to extract damage features corresponding to each mode; ii) Cross-modal attention fusion module: Learns the correlation weights between different modalities through a self-attention mechanism to achieve effective fusion of multimodal features; iii) Contrastive learning optimization: The NT-Xent loss function is used for training to maximize the similarity of damage features of the same type and the difference of damage features of different types, thereby improving the model's ability to identify rare damage patterns. iv) Damage mode vector construction and hazard coefficient calculation: Based on the damage morphology characteristics of wheel-rail materials, a damage mode vector is constructed. The damage mode vector is fused with mechanical parameters and then input into a virtual twin model to output the damage morphology hazard coefficient. The critical value of the damage morphology hazard coefficient is determined to provide a basis for subsequent damage limit judgment.

6. The wheel-rail damage monitoring method based on damage tolerance design and digital twin drive according to claim 5, characterized in that, The damage-to-limit intelligent decision dual-criteria mechanism in step 5) is as follows: i) Dynamic crack parameter criterion: When the crack size obtained by real-time monitoring is greater than the damage tolerance design size in the damage tolerance database, or the crack propagation rate is greater than the dynamic threshold in the database, the criterion is deemed to be satisfied. ii) Morphological risk coefficient criterion: When the calculated damage morphological risk coefficient is greater than the preset critical value, the criterion is deemed to be satisfied; iii) Warning triggering rules: A warning is triggered if any of the above criteria are met. If both criteria are met at the same time, the warning level is increased to ensure that no damage risk is overlooked.

7. The wheel-rail damage monitoring method based on damage tolerance design and digital twin drive according to claim 1, characterized in that, Step 6) of the machine learning optimization module includes the following process: i) Self-supervised pre-training process: By performing rotation enhancement processing on unlabeled vibration data, the TCN network is trained to predict the rotation angle of the signal, thereby enhancing the damage data; the bottom layer structure of the TCN network is frozen, and only the top layer network is fine-tuned for crack propagation direction prediction, thus realizing model pre-training based on normal wheel-rail operation data; ii) Twin simulation-reinforcement learning closed loop: Simulate virtual damage scenarios in a virtual twin model to generate virtual training data; use reinforcement learning to train the damage assessment model, use the minimization of prediction error as the reward function, and rapidly iterate and optimize the model parameters in the virtual environment; iii) Model deployment: When the model prediction error is lower than the preset threshold, the optimized model is deployed to the physical system corresponding to the physical sensor network to realize the actual application of the model and complete the iterative optimization process of virtual pre-training-physical fine-tuning.

8. The wheel-rail damage monitoring method based on damage tolerance design and digital twin drive according to claim 7, characterized in that, The hierarchical early warning instructions in step 7) are associated with the operation and maintenance strategy library. The specific association relationship is as follows: i) Level 1 warning: Triggers a recommendation for rail grinding or wheel turning and initiates the fine-tuning process of the self-supervised model to optimize the monitoring model in a timely manner; ii) Level 2 warning: Generates a mandatory command for rail grinding or wheel turning, and activates the virtual simulation training process to optimize the maintenance plan based on the virtual scenario; iii) Level 3 early warning: The train control system is linked to implement speed limit operation, and at the same time, the reinforcement learning parameter optimization process is triggered to quickly improve the model's adaptability to extreme damage scenarios; The associated actions in the operation and maintenance strategy library ensure that early warning signals can be transformed into actual operation and maintenance measures, realizing closed-loop management of monitoring, early warning, and operation and maintenance.

9. The wheel-rail damage monitoring method based on damage tolerance design and digital twin drive according to claim 3, characterized in that, The initial crack specifications are as follows: lengths of 0.5mm, 1.0mm, and 1.5mm, and angles of 30°, 45°, and 60°, respectively; the applied wheel-rail contact load is based on Hertz contact theory to simulate a 21t axle load condition to ensure that the crack propagation simulation is consistent with the actual wheel-rail service conditions; The stress intensity factor was calculated using the M-integral method, and the effective net section method was used to calculate the reference stress at the crack tip.

10. The wheel-rail damage monitoring method based on damage tolerance design and digital twin drive according to claim 5, characterized in that, The damage mode vector specifically includes the area, depth, and shape irregularity of wheel-rail damage; the mechanical parameters include the contact stress monitored by the pressure sensor array and the stress intensity factor obtained through stress calculation; when fusing the morphology vector with the mechanical parameters, feature normalization is used to ensure the weight balance of parameters in different dimensions; the determination of the critical value of the damage morphology risk coefficient needs to be combined with the fatigue performance test data of wheel-rail materials, historical damage failure cases, and industry wheel-rail damage assessment standards.

Images