A rail breakage risk dynamic prediction and grading early warning method and system

Abstract

The application provides a rail breaking risk dynamic prediction and grading early warning method and system, the method comprises the following steps: continuously laying a distributed optical fiber sensor along the longitudinal whole length of a rail, performing multi-stage signal optimization processing on the original signal collected by the optical fiber, establishing a quantitative physical mapping inversion of the optical fiber sensing characteristics and the wheel-rail impact force and the track state, and acquiring the microscopic state parameters of the crack position by ultrasonic detection; based on the acquired microscopic state parameters, a rail breaking evolution prediction model based on a space-time graph neural network is constructed to realize dynamic prediction of the whole life cycle of rail cracks from initiation to expansion; based on rail material fatigue tests and field measurement data, a crack propagation rate curve is established, and combined with the crack propagation rate, the remaining life and the rail breaking risk probability output by the prediction model, a rail breaking risk grade division and grading early warning is established. The application establishes a grading early warning mechanism based on the crack propagation rate, and realizes the whole life cycle prediction of rail cracks from initiation to expansion.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring and risk early warning technology for rail transit infrastructure, and more specifically, to a method and system for dynamic prediction and graded early warning of rail breakage risk. Background Technology

[0002] As the core load-bearing component of the wheel-rail system, rail transit rails are constantly subjected to multiple factors such as train wheel-rail impact, alternating loads, changes in environmental temperature and humidity, and deformation of the track foundation. This makes them prone to fatigue cracks that gradually propagate, ultimately leading to rail breakage accidents and seriously threatening train operation safety. While existing rail breakage monitoring and early warning technologies have incorporated a combination of methods including manual inspection, fixed-point sensor detection, and ultrasonic flaw detection, significant technical bottlenecks remain in areas such as monitoring continuity, data fusion, prediction accuracy, and the scientific nature of early warnings. These shortcomings can be further elaborated as follows: The monitoring system suffers from "disruptions" and "lag": Traditional manual inspections rely on on-site visual inspection and portable equipment detection, which are limited by manpower, weather, and inspection frequency, making it impossible to achieve long-distance, 24-hour continuous real-time monitoring; fixed-point sensors can only cover local key locations of the rails, resulting in monitoring blind spots, and are mostly post-event detections, making it difficult to capture the dynamic changes of wheel-rail impact and early signals of crack initiation.

[0003] The lack of physical mapping in distributed fiber optic sensing technology: Although existing distributed fiber optic sensing technology can collect rail vibration and acoustic signals, it only stays at the signal acquisition level. It has not established a quantitative physical mapping relationship between fiber optic demodulation signals and acoustic spectrum characteristics, wheel-rail impact mechanical parameters, and track smoothness. It is impossible to accurately invert key information such as the actual force on the rail, impact location, and track deformation from the sensing signals. The engineering application value of sensing data has not been fully explored.

[0004] Low degree of multi-source data fusion and insufficient accuracy of prediction models: Rail crack propagation is the result of the coupling effect of multiple factors such as mechanical load, environmental factors, operating parameters, and structural characteristics. However, existing crack prediction only uses crack geometry data from single-point ultrasonic detection, without integrating spatiotemporal data such as historical wheel-rail impact big data, train operating conditions (load, speed, and frequency of passage), environmental temperature and humidity, and line maintenance records. Furthermore, traditional prediction models (such as empirical formulas and single neural networks) cannot take into account both the spatial neighborhood correlation of crack propagation (such as cracks at a certain monitoring point easily propagating to adjacent points) and the temporal gradual evolution (such as the gradual increase in crack rate due to impact accumulation). The prediction bias for the entire life cycle evolution trend of cracks from initiation to propagation is large, making it difficult to achieve early hazard identification.

[0005] The early warning mechanism lacks scientific grading, resulting in low operation and maintenance response efficiency: existing rail breakage early warning systems mostly use a single crack size threshold as the basis for judgment, without combining the fatigue characteristics of rail materials to establish a correlation between crack propagation rate and remaining life, which easily leads to the problem of "over-warning for minor risks and under-warning for high risks"; moreover, the early warning result is only a binary judgment of "warning / non-warning", without providing clear operation and maintenance handling strategies for different stages of crack propagation, resulting in delayed on-site operation and maintenance response, lack of targeted handling measures, and inability to form a closed-loop management of monitoring and handling.

[0006] Weak system compatibility and engineering capabilities: Some new monitoring technologies require large-scale modifications to existing track lines, and cannot achieve data interoperability with existing track monitoring systems (such as track geometry monitoring systems and train operation monitoring systems). The deployment costs are high, the construction is difficult, and it is difficult to apply them on a large scale in engineering projects. Summary of the Invention

[0007] This invention addresses the technical problems existing in the prior art by providing a method and system for dynamic prediction and graded early warning of rail breakage risk. It establishes a precise mapping relationship between distributed fiber optic sensing signals and acoustic spectral characteristics, wheel-rail impact force, and track condition, enabling long-distance continuous real-time monitoring. By integrating multi-source spatiotemporal data, a high-precision rail breakage evolution prediction model is constructed, achieving full life-cycle prediction of rail cracks from initiation to propagation. A graded early warning mechanism is established based on crack propagation rate, improving the scientific rigor, timeliness, and operability of the early warning system.

[0008] According to a first aspect of the present invention, a method for dynamic prediction and graded early warning of rail breakage risk is provided, comprising: Distributed fiber optic sensors are continuously laid along the entire longitudinal length of the rail. The raw signals collected by the fiber optic sensors are processed in multiple stages to optimize the signal. A quantitative physical mapping inversion between fiber optic sensing characteristics and wheel-rail impact force and track condition is established. Ultrasonic detection is used to obtain the microscopic state parameters of the crack location. Based on the acquired microscopic state parameters, a rail break evolution prediction model based on spatiotemporal graph neural network is constructed to realize dynamic prediction of the entire life cycle of rail cracks from initiation to propagation. Based on fatigue test data and field measurement data of rail materials, a crack propagation rate curve is established. Combined with the crack propagation rate, remaining life and rail breakage risk probability output by the prediction model, a rail breakage risk level classification and graded early warning are established.

[0009] Based on the above technical solution, the present invention can also be improved as follows.

[0010] Optionally, the acquired raw signals include: optical signals, vibration signals, acoustic signals, and position data; the multi-stage signal optimization processing of the raw signals acquired by the optical fiber includes: Wavelet denoising and adaptive filtering are used to eliminate signal interference caused by environmental noise and fiber loss. Time-frequency analysis, energy spectrum analysis and spectral decomposition are used to extract characteristic parameters of impact sound spectrum, vibration spectrum and energy spectrum, thus completing the transformation of the original signal into engineering interpretable features.

[0011] Optionally, the impact acoustic spectrum features include: frequency distribution, dominant frequency, and bandwidth; the vibration spectrum features include: vibration amplitude, vibration frequency, and vibration duration; and the energy spectrum features include: impact energy and energy decay rate.

[0012] Optionally, the construction of the orbital failure evolution prediction model based on the acquired microscopic state parameters using a spatiotemporal graph neural network includes: Using the spatial location of the rails as the spatial reference and the timestamp as the time reference, a spatiotemporal correlation matrix of multi-source data is established, mapping data with different collection frequencies and different monitoring dimensions to the same spatiotemporal coordinate system, thereby achieving accurate matching of multi-dimensional data of "mileage station number - time point"; Standardized and spatiotemporally registered multi-source data were integrated into a spatiotemporal dataset of rail crack propagation. Key feature parameters that have a significant impact on rail crack propagation were screened out through information analysis, Pearson correlation coefficient and random forest feature importance algorithm, and redundant features were removed. The key feature parameters include: wheel-rail impact cumulative energy, crack depth, ambient temperature change rate and train passing frequency.

[0013] Optionally, the track break evolution prediction model includes: The graph convolution module is used to construct a rail spatial graph structure with the mileage station number of each monitoring point of the rail as the node and the spatial distance between the nodes as the edge weight. The spatial neighborhood association features of crack propagation are extracted through graph convolution operation. The time-series loop module is used to perform loop operations on the time-series data of each monitoring point to extract the time-dependent features of crack propagation. The feature fusion module is used to perform weighted fusion of spatial and temporal features through an attention mechanism, highlighting the spatiotemporal features that have a significant impact on crack propagation and achieving deep coupling of multi-dimensional features. The prediction output module takes the fused spatiotemporal features as input and outputs the crack size, crack propagation rate, and rail breakage risk probability of each monitoring point on the rail within a preset time period through a fully connected layer.

[0014] Optionally, the realization of dynamic prediction of the entire life cycle of rail cracks from initiation to propagation includes: The constructed spatiotemporal dataset was divided into training, validation, and test sets. The prediction error of crack size and the prediction deviation of propagation rate were used as loss functions, and an adaptive moment estimation optimization algorithm was used to train the rail break evolution prediction model. The generalization ability of the rail break evolution prediction model was improved by hyperparameter optimization and cross-validation. At the same time, rail material fatigue test data were introduced as prior knowledge to correct the prediction results of the rail break evolution prediction model and improve the prediction accuracy. An online self-updating mechanism for the rail break evolution prediction model is established. New data collected in real time is continuously input into the rail break evolution prediction model for incremental training, realizing dynamic iterative optimization of the rail break evolution prediction model. This adapts to the changing patterns of rail crack propagation under different lines and operating conditions, ensuring the prediction accuracy of the rail break evolution prediction model in long-term use.

[0015] Optionally, the establishment of the rail break risk level classification includes: Using crack propagation rate, crack size / critical size ratio, and rail breakage risk probability as evaluation indicators, the weight of each indicator is determined by the analytic hierarchy process (AHP), and a comprehensive evaluation model for rail breakage risk is constructed. The rail breakage risk is divided into multiple levels, and different risk characteristics and operation and maintenance strategies are set for each level.

[0016] Optionally, the risk levels of rail breakage are classified as follows: Level 1 warning, attention; Level 2 warning, reminder; Level 3 warning, alert; Level 4 warning, emergency.

[0017] Optionally, establishing a graded early warning system for rail breakage risks includes: Based on the different warning levels, the warning information is accurately pushed to the corresponding level of operation and maintenance management departments: Level 1 / Level 2 warnings are pushed to the on-site monitoring teams and line maintenance stations, Level 3 warnings are pushed to the railway section dispatch center, and Level 4 warnings are pushed to the railway bureau emergency command center. At the same time, a feedback mechanism for handling results is established to feed back the results of on-site operation and maintenance handling to the monitoring and prediction system, update the dataset, and optimize the prediction model for the evolution of track breaks.

[0018] According to a second aspect of the present invention, a dynamic prediction and graded early warning system for rail breakage risk is provided, comprising: The rail breakage data acquisition module is used to continuously lay distributed optical fiber sensors along the entire longitudinal length of the rail, perform multi-stage signal optimization processing on the raw signals collected by the optical fiber sensors, establish a quantitative physical mapping inversion between optical fiber sensing characteristics and wheel-rail impact force and track state, and use ultrasonic detection to obtain the microscopic state parameters of the crack location. The rail breakage risk dynamic prediction module is used to construct a rail breakage evolution prediction model based on spatiotemporal graph neural network based on the acquired micro-state parameters, so as to realize dynamic prediction of the entire life cycle of rail cracks from initiation to propagation. The rail breakage risk classification and early warning module is used to establish crack propagation rate curves based on rail material fatigue test and field measurement data. Combined with the crack propagation rate, remaining life and rail breakage risk probability output by the prediction model, it establishes rail breakage risk level classification and graded early warning.

[0019] The technical effects and advantages of this invention are as follows: This invention provides a method and system for dynamic prediction and graded early warning of rail breakage risk. By constructing a full-process rail breakage risk management system, which includes "real-time acquisition of multi-source data and accurate state inversion - spatiotemporal neural network prediction of rail breakage evolution - graded early warning based on crack propagation rate - closed-loop disposal", the system achieves an upgrade in rail breakage risk from "passive detection" to "active prediction", from "single-point monitoring" to "full-domain perception", and from "single early warning" to "graded disposal", realizing the prediction of the entire life cycle of rail cracks from initiation to propagation. Attached Figure Description

[0020] Figure 1 A flowchart of a method for dynamic prediction and graded early warning of rail breakage risk provided in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Understandably, given the deficiencies in the background technology, this invention proposes a method for dynamic prediction and graded early warning of rail breakage risk, specifically as follows: Figure 1 As shown, it includes the following steps: S1. Distributed fiber optic sensors are continuously laid along the entire longitudinal length of the rail. The raw signals collected by the fiber optic sensors are processed in multiple stages to optimize the signal. A quantitative physical mapping inversion between fiber optic sensing characteristics and wheel-rail impact force and track state is established. Ultrasonic detection is used to obtain the microscopic state parameters of the crack location. The acquired raw signals include: Optical signal: a time sequence of changes in light intensity, phase, and frequency; Vibration signals: vibration amplitude, frequency, and duration; Acoustic signal: impact sound spectrum, dominant frequency, bandwidth, energy attenuation; Location data: Mileage marker, monitoring point number.

[0023] The multi-stage signal optimization processing of the raw signals collected by the fiber optic sensor specifically involves: capturing the vibration and impact sound signals when the train passes by in real time and converting them into intensity, phase, and frequency modulation signals via optical fiber.

[0024] This step breaks through the limitations of traditional sensor data acquisition, which ends as soon as it is collected. By using distributed fiber optic sensing for full-dimensional signal analysis and ultrasonic detection for quantification of microcracks, combined with signal processing and mechanical inversion algorithms, it achieves the synchronous acquisition of the rail's "macroscopic stress state - microcrack state", and establishes a quantitative physical mapping relationship between the sensor signals and the actual state of the rail.

[0025] In this embodiment, the raw signal acquired by the optical fiber undergoes multi-stage signal optimization processing to establish a quantitative physical mapping inversion between optical fiber sensing characteristics and wheel-rail impact force and track state. The microscopic state parameters of the crack location are obtained using ultrasonic detection, specifically including: 1.1. Deployment and full signal acquisition of distributed fiber optic sensing systems; Distributed fiber optic sensors are continuously laid along the entire longitudinal length of the rail. The sensors are attached to the rail web / rail head and employ dual-mode detection technology of optical time domain reflectance (OTDR) and phase-sensitive optical time domain reflectance (Φ-OTDR) to achieve continuous high-frequency acquisition above 100Hz. This captures the vibration and impact sound signals generated by the wheel-rail contact when the train passes through the rail structure and are transmitted to the optical fiber, causing full-dimensional signal changes in optical phase modulation, optical intensity modulation, and optical frequency modulation, thus covering the rail without monitoring blind spots.

[0026] 1.2. Refined processing and feature extraction of fiber optic sensing signals; Multi-stage signal optimization processing of the raw signal acquired by optical fiber includes: firstly, eliminating environmental noise and signal interference caused by fiber loss through wavelet denoising and adaptive filtering; then, extracting three core feature parameters—impact spectral characteristics (frequency distribution, dominant frequency, bandwidth), vibration spectral characteristics (vibration amplitude, vibration frequency, vibration duration), and energy spectral characteristics (impact energy, energy decay rate)—through time-frequency analysis (short-time Fourier transform, Hilbert-Huang transform), energy spectrum analysis, and spectral decomposition, thereby transforming the raw signal into engineering-interpretable features.

[0027] 1.3. Quantitative physical mapping and inversion of sensing characteristics and rail condition; Based on the theory of elastic wave propagation and the wheel-rail impact mechanics model, a quantitative mapping formula is established between fiber optic sensing characteristics and wheel-rail impact force and track state, achieving accurate inversion from "characteristic parameters" to "physical parameters." The vibration / acoustic spectrum signals acquired by the fiber optic cable are converted into physical quantities that can be directly used in engineering, such as wheel-rail impact force, track deformation, and crack location.

[0028] Acoustic signal frequency components and fiber optic demodulated signal spectrum characteristics: The characteristic frequency ranges corresponding to different wheel-rail impact types (normal contact, pitting impact, rail gap impact) are determined by frequency matching algorithm; Sound signal amplitude intensity and fiber optic signal amplitude: First, perform on-site mechanical calibration: Using a standard hammer impact / loading test, apply a known force F0 to the rail, and simultaneously record the fiber optic amplitude A0. Establish a quantitative conversion formula between fiber optic signal amplitude and wheel-rail impact force, expressed as follows: F = k×A+b, where F is the wheel-rail impact force, A is the fiber optic signal amplitude, and k and b are calibration coefficients, thus realizing the quantitative calculation of the wheel-rail impact force.

[0029] Example of implementation: Calibration results: k = 0.82 kN / mV, b = 0.15 kN; Field acquisition of fiber optic amplitude: A = 12.5 mV; Inversion impact force: F = 0.82 × 12.5 + 0.15 = 10.4 kN.

[0030] Implementation method: Establish a characteristic frequency library and use spectrum matching to determine the impact type: normal wheel-rail contact: main frequency 800-1500 Hz; rail surface pitting / abrasion: main frequency 2000-4000 Hz; rail gap / joint impact: main frequency 300-600 Hz; track deformation / irregularity: main frequency 50-200 Hz.

[0031] Example of implementation: Fiber optic spectrum peak: 2650Hz; Matching feature library → Judgment: Rail surface pitting impact; Simultaneous location mileage: K12+350.

[0032] The time-domain waveform and duration of the acoustic signal and the train's operating status: By identifying the waveform and analyzing the time window, the train's passing position, speed, and axle load distribution are determined, while the track smoothness status (such as the location and degree of track surface irregularities and track deformation) is inverted.

[0033] Implementation method: The train speed is calculated using a time window Δt, expressed by the formula: v = L / Δt, where v represents the train speed, L represents the length of the fixed monitoring section (e.g., 10 m), and Δt represents the time it takes for the vibration signal to pass through the monitoring section; Inverting the axle load using the vibration envelope area: Axle load ∝ vibration energy integral = ∫A(t) dt; Landing example: L=10m, Δt=0.42s; speed v=10 / 0.42≈23.8m / s=85.7km / h, vibration integral area is too large → judged as heavy-load train.

[0034] 1.4. Quantitative acquisition of microcracks by ultrasonic testing; High-precision ultrasonic testing devices are deployed at key sections of the rail where cracks are prone to initiation, such as the rail head, rail web, and rail bottom. Using phased array ultrasonic testing technology, multi-beam ultrasonic waves are emitted into the rail. By analyzing the propagation time, amplitude, and phase of the echo signals, combined with digital imaging algorithms, quantitative analysis of microscopic parameters such as crack location (accurate to the mm level), depth, length, opening degree, and crack direction is achieved, providing microscopic basic data for crack propagation prediction.

[0035] Ultrasound → Crack Microscopic Parameters (Quantitative Inversion) Implementation method: The formula for calculating depth d by ultrasonic echo time t is: d = v s ×t / 2, where v s The speed of the ultrasonic wave in the rail is approximately 5900 m / s; t represents the transmission-to-reception echo time (μs). Example of landing: echo time t=10.2μs; crack depth d=5900×10.2e-6 / 2≈0.0301 m=30.1mm.

[0036] S2. Based on the acquired microscopic state parameters, a rail break evolution prediction model based on spatiotemporal graph neural network is constructed to realize dynamic prediction of the entire life cycle of rail cracks from initiation to propagation. This step addresses the issues of existing data being "heterogeneous, isolated, and unrelated." Through data standardization, spatiotemporal registration, and feature fusion, multi-source heterogeneous data is integrated into a unified spatiotemporal dataset, providing high-quality data support for the orbital derailment evolution prediction model.

[0037] In this embodiment, four types of multi-source spatiotemporal data related to rail crack propagation are collected and integrated, and standardized processing is performed according to the different data types, dimensions, and acquisition frequencies. The four types of multi-source spatiotemporal data include: Real-time monitoring data: wheel-rail impact force sequence, impact acoustic spectrum characteristics, track smoothness parameters retrieved by distributed optical fiber, and crack geometric characteristics detected by ultrasonic waves (acquisition frequency: minute-level / hour-level). Operational data: train load, operating speed, passing frequency, axle type distribution, etc. (collection frequency: train number level); Environmental monitoring data: rail surface temperature, ambient temperature and humidity, rainfall, changes in the temperature field of the track foundation, etc. (collection frequency: hourly / daily); Historical and operational data: historical wheel-rail impact big data, rail material fatigue characteristic parameters, track maintenance records (grinding, repair, replacement), past crack detection data, etc. (collection frequency: daily / monthly).

[0038] By using algorithms such as normalization, dimensionless transformation, and interpolation completion, the dimensional differences between different data are eliminated, and the standardization of heterogeneous data is achieved.

[0039] The method for constructing a trajectory failure evolution prediction model based on a spatiotemporal graph neural network based on the acquired microscopic state parameters includes: Spatiotemporal registration of multi-source data: Using the spatial location of the rail (mileage station) as the spatial reference and the timestamp as the time reference, a spatiotemporal correlation matrix of multi-source data is established, mapping data with different collection frequencies and different monitoring dimensions to the same spatiotemporal coordinate system, realizing accurate matching of multi-dimensional data of "a certain mileage station - a certain time point", and solving the problems of "data asynchrony and location mismatch".

[0040] The construction and feature selection of the integrated dataset: Standardized and spatiotemporally registered multi-source data are integrated into a spatiotemporal dataset of rail crack propagation. Through algorithms such as mutual information analysis, Pearson correlation coefficient, and random forest feature importance, key feature parameters that have a significant impact on rail crack propagation (such as wheel-rail impact cumulative energy, crack depth, ambient temperature change rate, train passing frequency, etc.) are selected, redundant features are eliminated, and the training efficiency and prediction accuracy of subsequent prediction models are improved.

[0041] This step overcomes the limitations of traditional prediction models that cannot simultaneously consider both "spatial correlation" and "temporal evolution." Based on the spatial neighborhood correlation and gradual temporal evolution of rail crack propagation, an improved spatiotemporal graph neural network-based rail break evolution prediction model is constructed, enabling dynamic prediction of the entire lifecycle of rail cracks from initiation to propagation. The Improved Spatio-Temporal Graph Neural Network (STGNN model) is specifically designed for dynamic prediction of rail cracks throughout their entire lifecycle, from initiation to propagation and eventual breakage. Its core capability is the simultaneous capture of both the spatial correlation and temporal asymptotic nature of crack propagation. The improved STGNN model consists of four parts: a Graph Convolutional Network (GCN) module, a Temporal Recurrent Model (LSTM / GRU) module, a feature fusion module, and a prediction output module. These modules work together to achieve in-depth mining of spatiotemporal features. Graph Convolution Module (GCN): Using the mileage markers of each monitoring point on the rail as nodes and the spatial distance between nodes as edge weights, a spatial graph structure of the rail is constructed. The spatial neighborhood association features of crack propagation are extracted through graph convolution operations (such as the impact of crack propagation at a certain monitoring point on monitoring points within 50m). The time-series loop module (LSTM / GRU) performs loop calculations on the time-series data of each monitoring point (such as wheel-rail impact accumulation and crack size changes) to extract the time-dependent features of crack propagation (such as the influence of impact accumulation over time on crack propagation rate). Feature fusion module: It uses an attention mechanism to weighted fuse spatial and temporal features, highlighting spatiotemporal features that have a significant impact on crack propagation, and achieving deep coupling of multi-dimensional features; Prediction output module: Taking the fused spatiotemporal features as input, it outputs three core prediction results through a fully connected layer: crack size (depth / length), crack propagation rate (da / dN), and rail breakage risk probability at each monitoring point of the rail within a preset time period (1 day / 7 days / 30 days).

[0042] The implementation of dynamic prediction of the entire life cycle of rail cracks, from initiation to propagation, includes: Model training and optimization: The constructed spatiotemporal dataset is divided into training, validation, and test sets. The adaptive moment estimation (Adam) optimization algorithm is used to train the rail break evolution prediction model, with crack size prediction error and propagation rate prediction deviation as loss functions. The generalization ability of the rail break evolution prediction model is improved through hyperparameter optimization and cross-validation. At the same time, rail material fatigue test data is introduced as prior knowledge to correct the prediction results of the rail break evolution prediction model and further improve the prediction accuracy.

[0043] Dynamic updating and iteration of the model; establish an online self-updating mechanism for the rail break evolution prediction model, continuously input new data collected in real time into the rail break evolution prediction model for incremental training, realize dynamic iterative optimization of the rail break evolution prediction model, adapt to the law changes of rail crack propagation under different lines and different operating conditions, and ensure the prediction accuracy of the rail break evolution prediction model in long-term use.

[0044] S3. Based on fatigue test data and field measurement data of rail materials, establish crack propagation rate curves. Combine the crack propagation rate, remaining life and rail breakage risk probability output by the prediction model to establish rail breakage risk level classification and graded early warning.

[0045] This step breaks through the limitations of traditional single-threshold early warning. Based on fatigue test data of rail materials and field measurement data, a crack propagation rate curve is established. Combined with the crack propagation rate, remaining life, and rail breakage risk probability output by the prediction model, a multi-dimensional, refined, and implementable rail breakage risk classification early warning mechanism is constructed, forming a closed-loop safety assurance system of "monitoring-prediction-early warning-handling-feedback".

[0046] Based on fatigue test and field measurement data of rail materials, a crack propagation rate curve is established. Combined with the crack propagation rate, remaining service life, and rail breakage risk probability output by the prediction model, a rail breakage risk level classification and graded early warning system is established, specifically including: 3.1. Establishment of rail crack propagation rate curve; By conducting fatigue tests on rail base materials and on-site dimensional tests, and combining fracture mechanics theory, a relationship curve (da / dN~ΔK curve) between rail crack propagation rate and stress intensity factor range was established. This determined the quantitative correlation between crack depth and propagation rate, remaining life, and critical crack size under different rail materials (such as U71Mn and U75V) and different operating conditions, providing a materials science and mechanical basis for graded early warning.

[0047] 3.2. Multi-dimensional classification of track breakage risk levels; Using crack propagation rate, crack size / critical size ratio, and rail breakage risk probability as the three core evaluation indicators, the Analytic Hierarchy Process (AHP) is employed to determine the weights of each indicator, constructing a comprehensive rail breakage risk evaluation model. This model classifies rail breakage risk into four warning levels: Level 1 Warning (Attention); Level 2 Warning (Reminder); Level 3 Warning (Alert); and Level 4 Warning (Emergency). The judgment criteria, risk characteristics, and handling strategies for each level are further refined as follows:

[0048] 3.3. Accurate delivery and closed-loop processing of early warning information; A tiered early warning information push system has been established. Based on different early warning levels, early warning information (including crack location, size, propagation rate, risk level, and recommended handling strategies) is precisely pushed to the corresponding level of operation and maintenance management departments: Level 1 / 2 warnings are pushed to on-site monitoring teams and line maintenance stations; Level 3 warnings are pushed to the railway section dispatch center; and Level 4 warnings are pushed to the railway bureau emergency command center. Simultaneously, a handling result feedback mechanism has been constructed, feeding back the results of on-site operation and maintenance handling (such as crack detection data after grinding, repair, or replacement, and line operating status) to the monitoring and prediction system. This updates the dataset and optimizes the prediction model, forming a closed-loop safety assurance system of "monitoring-prediction-early warning-handling-feedback," achieving full-process control of rail breakage risks.

[0049] According to a second aspect of the present invention, a dynamic prediction and graded early warning system for rail breakage risk is provided, comprising: The rail breakage data acquisition module is used to continuously lay distributed optical fiber sensors along the entire longitudinal length of the rail, perform multi-stage signal optimization processing on the raw signals collected by the optical fiber, establish a quantitative physical mapping inversion between optical fiber sensing characteristics and wheel-rail impact force and track state, and use ultrasonic detection to obtain the microscopic state parameters of the crack location. The rail breakage risk dynamic prediction module is used to construct a rail breakage evolution prediction model based on spatiotemporal graph neural network based on the acquired micro-state parameters, so as to realize dynamic prediction of the entire life cycle of rail cracks from initiation to propagation. The rail breakage risk classification and early warning module is used to establish crack propagation rate curves based on rail material fatigue test and field measurement data. Combined with the crack propagation rate, remaining life and rail breakage risk probability output by the prediction model, it establishes rail breakage risk level classification and graded early warning.

[0050] It is understood that the rail breakage risk dynamic prediction and graded early warning system provided in this embodiment corresponds to the rail breakage risk dynamic prediction and graded early warning method provided in the foregoing embodiment. The relevant technical features of the rail breakage risk dynamic prediction and graded early warning system can be referred to the relevant technical features of the rail breakage risk dynamic prediction and graded early warning method, and will not be repeated here.

[0051] Finally, the embodiments described in this invention, through the deep integration of multiple technologies, data, and models, solve the core pain points of existing rail breakage monitoring and early warning technologies. Compared with existing technologies, it has significant advantages in terms of comprehensive monitoring, precise inversion, intelligent prediction, scientific early warning, and practical engineering application. Specific implementation effects are as follows: 1. Achieve full-area, continuous, real-time monitoring of rails: Distributed fiber optic sensors are continuously laid along the entire length of the rails. Combined with OTDR / Φ-OTDR dual-mode detection technology, it achieves blind-spot-free, 24-hour, high-frequency continuous monitoring, replacing traditional fixed-point sensors and manual inspections. The monitoring coverage is increased to 100%, and the real-time performance is improved to the minute / second level. It can accurately capture the early weak signals of crack initiation.

[0052] 2. Establishing a quantitative physical mapping between sensor signals and rail conditions: Starting from the principles of elastic wave propagation and wheel-rail impact mechanics, the intrinsic relationship between distributed fiber optic sensor signals and acoustic spectrum characteristics, wheel-rail impact force, and rail conditions was clarified. A quantitative conversion formula was established, realizing accurate inversion from "sensor signals" to "physical parameters". The accuracy of state inversion was improved to over 90%, solving the problem of "engineering inability to interpret" existing fiber optic sensor data.

[0053] 3. Significantly improve the prediction accuracy of crack propagation and rail breakage risk: Based on the spatiotemporal graph neural network that integrates multi-source spatiotemporal data, it takes into account both the spatial neighborhood correlation and the gradual evolution of crack propagation. Combined with the fatigue characteristics of rail materials, the model is corrected, the prediction error of crack size is controlled within 5%, and the prediction deviation of propagation rate is controlled within 10%. It can predict the risk of rail breakage 7-30 days in advance, and achieve accurate identification of early hidden dangers.

[0054] 4. Construct a scientific and feasible hierarchical early warning mechanism: Breaking through the limitations of single threshold early warning, a four-level early warning system is constructed based on crack propagation rate curves and multi-dimensional risk assessment indicators. The judgment criteria and operation and maintenance strategies for each early warning level are clarified, which completely solves the problems of "false alarms, missed alarms, and delayed response". The scientific nature and operability of the early warning are greatly improved, providing clear decision-making basis for on-site operation and maintenance.

[0055] 5. Achieve compatibility with existing systems and large-scale engineering application: The monitoring system of this invention can achieve data interoperability and compatibility with existing track geometry monitoring systems and railway operation and maintenance management systems. It does not require large-scale modification of existing track lines. The sensor deployment is simple and cost-effective, and the system has high stability. It can be applied on a large scale in various lines such as urban rail transit.

[0056] 6. Establish a closed-loop safety assurance system for the entire process: A closed-loop management and control system for the entire process has been constructed, which includes "multi-source data collection - accurate status inversion - dynamic prediction - hierarchical early warning - precise handling - result feedback". This system has enabled the transformation of rail breakage risk from "passive response" to "proactive prevention", effectively reducing the incidence of rail breakage accidents and improving the safety operation level and intelligent operation and maintenance capabilities of rail transit infrastructure.

[0057] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0058] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0059] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0060] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic prediction and graded early warning of rail breakage risk, characterized in that, Includes the following steps: Distributed fiber optic sensors are continuously laid along the entire longitudinal length of the rail. The raw signals collected by the fiber optic sensors are processed in multiple stages to optimize the signal. A quantitative physical mapping inversion between fiber optic sensing characteristics and wheel-rail impact force and track condition is established. Ultrasonic detection is used to obtain the microscopic state parameters of the crack location. Based on the acquired microscopic state parameters, a rail break evolution prediction model based on spatiotemporal graph neural network is constructed to realize dynamic prediction of the entire life cycle of rail cracks from initiation to propagation. Based on fatigue test data and field measurement data of rail materials, a crack propagation rate curve is established. Combined with the crack propagation rate, remaining life and rail breakage risk probability output by the prediction model, a rail breakage risk level classification and graded early warning are established.

2. The method for dynamic prediction and graded early warning of rail breakage risk according to claim 1, characterized in that, The acquired raw signals include: optical signals, vibration signals, acoustic signals, and position data; the multi-stage signal optimization processing of the raw signals acquired by the optical fiber includes: Wavelet denoising and adaptive filtering are used to eliminate signal interference caused by environmental noise and fiber loss. Time-frequency analysis, energy spectrum analysis and spectral decomposition are used to extract characteristic parameters of impact sound spectrum, vibration spectrum and energy spectrum, thus completing the transformation of the original signal into engineering interpretable features.

3. The method for dynamic prediction and graded early warning of rail breakage risk according to claim 2, characterized in that, The impact acoustic spectrum features include: frequency distribution, dominant frequency, and bandwidth; the vibration spectrum features include: vibration amplitude, vibration frequency, and vibration duration; the energy spectrum features include: impact energy and energy decay rate.

4. The method for dynamic prediction and graded early warning of rail breakage risk according to claim 1, characterized in that, The method for constructing a trajectory failure evolution prediction model based on a spatiotemporal graph neural network based on the acquired microscopic state parameters includes: Using the spatial location of the rails as the spatial reference and the timestamp as the time reference, a spatiotemporal correlation matrix of multi-source data is established, mapping data with different collection frequencies and different monitoring dimensions to the same spatiotemporal coordinate system, thereby achieving accurate matching of multi-dimensional data of "mileage station number - time point"; Standardized and spatiotemporally registered multi-source data were integrated into a spatiotemporal dataset of rail crack propagation. Key feature parameters that have a significant impact on rail crack propagation were screened out through information analysis, Pearson correlation coefficient, and random forest feature importance algorithm, and redundant features were eliminated. The key feature parameters include: wheel-rail impact cumulative energy, crack depth, ambient temperature change rate, and train passing frequency.

5. The method for dynamic prediction and graded early warning of rail breakage risk according to claim 1, characterized in that, The track break evolution prediction model includes: The graph convolution module is used to construct a rail spatial graph structure with the mileage station number of each monitoring point of the rail as the node and the spatial distance between the nodes as the edge weight. The spatial neighborhood association features of crack propagation are extracted through graph convolution operation. The time-series loop module is used to perform loop operations on the time-series data of each monitoring point to extract the time-dependent features of crack propagation. The feature fusion module is used to perform weighted fusion of spatial and temporal features through an attention mechanism, highlighting the spatiotemporal features that have a significant impact on crack propagation and achieving deep coupling of multi-dimensional features. The prediction output module takes the fused spatiotemporal features as input and outputs the crack size, crack propagation rate, and rail breakage risk probability of each monitoring point on the rail within a preset time period through a fully connected layer.

6. The method for dynamic prediction and graded early warning of rail breakage risk according to claim 1, characterized in that, The implementation of dynamic prediction of the entire life cycle of rail cracks, from initiation to propagation, includes: The constructed spatiotemporal dataset was divided into training, validation, and test sets. The prediction error of crack size and the prediction deviation of propagation rate were used as loss functions, and the adaptive moment estimation optimization algorithm was used to train the rail break evolution prediction model. The generalization ability of the rail break evolution prediction model was improved by hyperparameter optimization and cross-validation. The rail material fatigue test data was introduced as prior knowledge to correct the prediction results of the rail break evolution prediction model. An online self-updating mechanism for the rail break evolution prediction model is established, which continuously inputs new data collected in real time into the rail break evolution prediction model for incremental training, thereby realizing the dynamic iterative optimization of the rail break evolution prediction model.

7. The method for dynamic prediction and graded early warning of rail breakage risk according to claim 1, characterized in that, The establishment of a risk level classification for rail breakage includes: Using crack propagation rate, crack size / critical size ratio, and rail breakage risk probability as evaluation indicators, the weight of each indicator is determined by the analytic hierarchy process (AHP), and a comprehensive evaluation model for rail breakage risk is constructed. The rail breakage risk is divided into multiple levels, and different risk characteristics and operation and maintenance strategies are set for each level.

8. The method for dynamic prediction and graded early warning of rail breakage risk according to claim 7, characterized in that, The risk levels of rail breakage are classified as follows: Level 1 Warning, Attention; Level 2 Warning, Reminder; Level 3 warning: alert; Level 4 warning: emergency.

9. The method for dynamic prediction and graded early warning of rail breakage risk according to claim 1, characterized in that, Establishing a graded early warning system for rail breakage risks includes: Based on the different warning levels, the warning information is accurately pushed to the corresponding level of operation and maintenance management departments: Level 1 / Level 2 warnings are pushed to the on-site monitoring teams and line maintenance stations, Level 3 warnings are pushed to the railway section dispatch center, and Level 4 warnings are pushed to the railway bureau emergency command center. At the same time, a feedback mechanism for handling results is established to feed back the results of on-site operation and maintenance handling to the monitoring and prediction system, update the dataset, and optimize the prediction model for the evolution of track breaks.

10. A dynamic prediction and graded early warning system for rail breakage risk, characterized in that, include: The rail breakage data acquisition module is used to continuously lay distributed optical fiber sensors along the entire longitudinal length of the rail, perform multi-stage signal optimization processing on the raw signals collected by the optical fiber sensors, establish a quantitative physical mapping inversion between optical fiber sensing characteristics and wheel-rail impact force and track state, and use ultrasonic detection to obtain the microscopic state parameters of the crack location. The rail breakage risk dynamic prediction module is used to construct a rail breakage evolution prediction model based on spatiotemporal graph neural network based on the acquired micro-state parameters, so as to realize dynamic prediction of the entire life cycle of rail cracks from initiation to propagation. The rail breakage risk classification and early warning module is used to establish crack propagation rate curves based on rail material fatigue test and field measurement data. Combined with the crack propagation rate, remaining life and rail breakage risk probability output by the prediction model, it establishes rail breakage risk level classification and graded early warning.

Images