A method for railway fastener health state prediction and risk diffusion analysis

Abstract

The application discloses a kind of methods for railway fastener health state prediction and risk diffusion analysis, belong to the technical field of railway fastener health monitoring, it includes: collection track fastener image and environmental parameter;Track fastener image is extracted and health index is calculated to visual feature;According to the visual feature of fastener, health index and environmental parameter, construct space-time graph;Based on the constructed space-time graph, predict fastener health state and evaluate failure risk.The application can realize early detection and trend prediction of fastener defects, avoid systemic track structure damage caused by single-point failure, greatly improve the safety of railway operation and the pertinence of maintenance.Combined with risk diffusion analysis results, maintenance department can scientifically formulate maintenance plan, reduce unscheduled parking and large-scale maintenance cost, improve transportation efficiency, and has important significance for intelligent railway maintenance system construction and track life cycle management.

Description

Technical Field

[0001] This invention belongs to the technical field of railway fastener health monitoring, specifically relating to a method for predicting the health status and risk diffusion analysis of railway fasteners. Background Technology

[0002] The railway track structure is the core load-bearing system for train operation. Fasteners, as key components connecting rails and sleepers, are responsible for maintaining track gauge, buffering dynamic loads from trains, and suppressing rail displacement. Loosening, breaking, aging, or missing fasteners directly leads to decreased track stiffness and rail misalignment, resulting in abnormal track gauge, reduced train stability, and even serious accidents such as derailment. This is especially true in high-speed and heavy-haul railways, where the stress on fasteners is even higher due to high train speeds and significant wheel-rail impact forces. However, current monitoring of fastener health primarily relies on post-event inspections and manual checks, lacking effective predictive methods and making it difficult to identify potential faults in advance. This lack of predictive capability makes it difficult to proactively control track safety risks and increases maintenance costs and operational safety pressures.

[0003] Currently, railway fastener condition inspection mainly relies on manual inspection and image analysis from track inspection vehicles. Manual inspection is inefficient, labor-intensive, and carries the risk of missed detections and misjudgments. Track inspection vehicle visual inspection largely depends on traditional image features and fixed thresholds, lacking robustness to interference from lighting changes and shooting angles, leading to decreased recognition accuracy. Furthermore, existing methods are mostly static inspections, lacking the ability to predict failure trends based on historical fastener health data and operational conditions, and failing to establish risk diffusion models for fastener defects, making it difficult to implement precise maintenance measures in advance. Summary of the Invention

[0004] The purpose of this invention is to address the aforementioned shortcomings in the prior art by providing a method for predicting the health status and analyzing the risk diffusion of railway fasteners. This method enables dynamic assessment of fastener health and short-term life prediction, and combines graph theory and propagation models to deduce the spatial diffusion path of defects, thereby improving prediction accuracy and real-time response, supporting maintenance priority ranking, and reducing operation and maintenance costs.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for predicting the health status and analyzing the risk diffusion of railway fasteners includes the following steps:

[0007] S1. Acquire images of track fasteners and environmental parameters;

[0008] S2. Visual feature extraction and health index calculation of track fastener images;

[0009] S3. Construct a spatiotemporal diagram based on the visual characteristics, health index, and environmental parameters of the fasteners;

[0010] S4. Based on the constructed spatiotemporal diagram, predict the health status of fasteners and assess the risk of failure.

[0011] Furthermore, in step S1, GPS timestamps and mileage information are added to the collected track fastener images and environmental parameters, and the track fastener images are denoised and ROI cropped.

[0012] Furthermore, step S2 specifically includes the following sub-steps:

[0013] S21. Lightweight target detection of track fastener images is used to locate the fasteners and output fastener image blocks;

[0014] S22. Use a visual Transformer to extract the visual feature vector of the fastener image block;

[0015] S23. Based on visual feature vectors, classify the status of fasteners and calculate their health index.

[0016] Furthermore, S23 specifically includes:

[0017] The visual feature vector is input into the state classifier, which outputs a fastener state label. This fastener state label is then mapped to a health index, which is represented as follows:

[0018]

[0019] In the formula, For the health index of fasteners, i Indicates the fastener serial number. t Indicates time; Indicates the status of the fastener;

[0020] The fastener health index is smoothed using exponential sliding, and it is expressed as follows:

[0021]

[0022] In the formula, To smooth out health indicators; For smoothing coefficients; This represents the smoothed health index at time t-1.

[0023] Furthermore, step S3 specifically includes the following sub-steps:

[0024] S31. Define each fastener as a graph node, and define node features based on environmental parameters, visual feature vectors, and smoothness health index, as follows:

[0025]

[0026] In the formula, Representing node characteristics, For environmental parameter vectors, , For visual feature dimensions, For the number of environmental parameters, For feature dimensions;

[0027] Fasteners detected on the track k The node features corresponding to each time point form a node feature matrix. , This indicates the number of fasteners inspected on the track;

[0028] S32. Generate a spatial adjacency matrix based on the track mileage distance between fasteners;

[0029] S33. Calculate risk-related edges based on historical smoothed health index sequences;

[0030] S34. Transform the node feature matrix The corresponding spatial adjacency matrix is ​​stored in the cache.

[0031] Furthermore, in step S32, a spatial adjacency matrix is ​​generated based on the track mileage distance between the fasteners, which is represented as follows:

[0032]

[0033] In the formula, Indicates fastener With fasteners Distance; Represents the spatial adjacency matrix; Indicates the smoothing coefficient; This represents the maximum connection distance.

[0034] Furthermore, in step S33, the smooth health index sequences of the i-th fastener and the j-th fastener over a period of time are calculated based on the historical smooth health index sequence. and Pearson correlation coefficient between them:

[0035]

[0036] In the formula, This represents the function for calculating the correlation coefficient. This represents the Pearson correlation coefficient;

[0037] when At that time, add a risk propagation weight term to the spatial adjacency matrix. It is represented as:

[0038]

[0039] In the formula, This represents the enhanced spatial adjacency matrix; Indicates the enhancement coefficient; This is the correlation threshold; As an indicator function, when the Pearson correlation coefficient Greater than the correlation threshold hour, The value is 1 if it is 1, otherwise it is 0.

[0040] Furthermore, step S4 specifically includes the following sub-steps:

[0041] S41. Input the node feature matrix and spatial adjacency matrix into the graph convolution, and perform normalization processing at each time step to output the node feature sequence;

[0042] S42. Input the node feature sequences at each time step into the long short-term memory network for time modeling and generate time-related hidden state vectors.

[0043] S43. Input the hidden state vector into the fully connected layer and map it using the Sigmoid function to obtain the predicted health index, which is expressed as:

[0044]

[0045] In the formula, A predictive health index representing the next Δ steps; Indicates the activation function; Represents the weight matrix; Represents the hidden state vector; Indicates the bias term;

[0046] S44. Reverse map the predicted health index to the failure probability;

[0047] S45. Using the failure probability as the initial risk value, and performing iterative updates based on the independent propagation model, the predicted risk index is obtained.

[0048] S46. Conduct risk classification and early warning output based on risk indicators.

[0049] Furthermore, in step S44, the predicted health index is inversely mapped to the failure probability, which is expressed as:

[0050]

[0051] In the formula, This represents the failure probability in the next Δ steps.

[0052] Furthermore, in step S45, the failure probability is used as the initial risk value, and iterative updates are performed based on the independent propagation model to obtain the predicted risk index, which is expressed as:

[0053]

[0054] In the formula, A predictive risk indicator representing the next Δ step; For a moment Risk indicators For nodes i To the node j The intensity of risk transmission, The risk diffusion attenuation coefficient is the initial risk value. .

[0055] The method for predicting the health status and analyzing the risk diffusion of railway fasteners provided by this invention has the following beneficial effects:

[0056] In the detection phase, this invention utilizes a deep learning target detection algorithm to accurately identify the fastener status and extract health index sequences from images from track inspection vehicles or drones. In the prediction phase, a spatiotemporal graph neural network (ST-GNN) is introduced to integrate multi-source information such as historical health data, train speed, and environmental temperature and humidity to predict the health change trend of fasteners in future operating cycles. In the diffusion analysis phase, a track segment network model is constructed to simulate the risk propagation process of defects in a spatial range, providing a spatiotemporal basis for maintenance decisions.

[0057] This invention enables early detection and trend prediction of fastener defects, avoiding systemic track structure damage caused by single-point failures, and significantly improving the safety of railway operations and the targeted nature of maintenance. Combined with risk diffusion analysis results, maintenance departments can scientifically formulate maintenance plans, reduce unplanned shutdowns and large-scale maintenance costs, and improve transportation efficiency. This is of great significance for the construction of intelligent railway maintenance systems and the management of the entire track lifecycle.

[0058] This invention realizes dynamic health assessment and trend prediction. By extracting visual features from fastener images and combining them with a smoothed health index, this invention can achieve quantitative assessment of the health status of fasteners (H∈[0,1]). It also uses a spatiotemporal graph neural network and LSTM to capture spatial dependencies and temporal degradation trends, thereby predicting the health index for future Δ-step inspection cycles. This solves the problem that existing technologies can only perform static defect detection and cannot provide early warnings.

[0059] This invention has risk diffusion analysis capabilities. By utilizing risk-related edges and independent propagation models, it can simulate the spatial diffusion process of fastener defects in track sections, generate risk indicators, and classify risks into 1-5 levels, providing a quantitative basis for maintenance decisions. This solves the problem that existing technologies do not consider the spatial correlation of defects and are difficult to make full-line operation and maintenance decisions.

[0060] The health index and risk level output by this invention can directly guide maintenance prioritization, reducing unplanned shutdowns and large-scale repair costs. By identifying potentially faulty fasteners in advance, the risk of accidents can be reduced during the inspection cycle, improving the safety and efficiency of railway operations. Attached Figure Description

[0061] Figure 1 This is a flowchart of the method for predicting the health status and analyzing the risk diffusion of railway fasteners in Example 1.

[0062] Figure 2 This is a flowchart of the system used for predicting the health status and analyzing the risk diffusion of railway fasteners in Example 2. Detailed Implementation

[0063] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0064] Example 1

[0065] The method for predicting the health status and analyzing the risk diffusion of railway fasteners in this embodiment is based on... Figure 1 Specifically, it includes the following:

[0066] S1. Acquire images of track fasteners and environmental parameters;

[0067] Specifically, GPS timestamps and mileage information are added to the collected track fastener images and environmental parameters (temperature, humidity, rainfall, etc.), and the track fastener images are denoised and ROI cropped to reduce subsequent transmission and computation.

[0068] S2. Visual feature extraction and health index calculation of the track fastener image, which includes the following steps:

[0069] S21. Use lightweight target detection (such as MobileNet-YOLO) to detect track fastener images, then locate the fasteners and output the first... Each fastener at time Image blocks ;

[0070] S22. The visual feature vector of the fastener image block is extracted using a visual Transformer, specifically represented as follows:

[0071]

[0072] In the formula, This represents the visual feature vector of the fastener; For feature dimensions; This represents a visual Transformer model used to extract high-level semantic features from images;

[0073] S23. Based on visual feature vectors, classify the status of fasteners and calculate their health index.

[0074] The visual feature vector is input into the state classifier, which outputs a fastener state label. This fastener state label is then mapped to a health index, which is represented as follows:

[0075]

[0076] In the formula, For the health index of fasteners, i Indicates the fastener serial number. t Indicates time; (0: Normal, 1: Mild, 2: Moderate, 3: Severe) indicates the status of the fastener;

[0077] The fastener health index is smoothed using exponential sliding, and it is expressed as follows:

[0078]

[0079] In the formula, To smooth out health indicators; The smoothing coefficient, 0 ≤ <1, used to reduce sporadic detection noise; This represents the smoothed health index at time t-1.

[0080] S3. Based on the visual characteristics, health index, and environmental parameters of the fasteners, construct a spatiotemporal diagram, which includes the following steps:

[0081] S31. Define each fastener as a graph node, and define node features based on environmental parameters, visual feature vectors, and smoothness health index, as follows:

[0082]

[0083] In the formula, Representing node characteristics, This is a vector of environmental parameters (temperature, humidity, etc.). , For visual feature dimensions, For the number of environmental parameters, For feature dimensions;

[0084] Fasteners detected on the track k The node features corresponding to each time point form a node feature matrix. , This indicates the number of fasteners inspected on the track;

[0085] S32. Generate a spatial adjacency matrix based on the track mileage distance between fasteners, which is represented as follows:

[0086]

[0087] In the formula, Indicates fastener With fasteners Distance; Represents the spatial adjacency matrix; Indicates the smoothing coefficient; Maximum connection distance;

[0088] S33. Calculate risk-related edges based on historical smoothed health index sequences;

[0089] Specifically, the smoothed health index sequences of the i-th and j-th fasteners over a period of time are calculated based on the historical smoothed health index sequences. and Pearson correlation coefficient between them:

[0090]

[0091] In the formula, This represents the function for calculating the correlation coefficient. denoted as Pearson correlation coefficient, which is used to capture the same / opposite trends of health index changes, and the result is limited to [-1, 1].

[0092] when At that time, add a risk propagation weight term to the spatial adjacency matrix. It is represented as:

[0093]

[0094] In the formula, This represents the enhanced spatial adjacency matrix; This represents the enhancement factor, which is used to control the overall magnitude of risk weights, and its value depends on the specific environment. The correlation threshold is a parameter set based on historical data and engineering experience. As an indicator function, when the Pearson correlation coefficient Greater than the correlation threshold hour, The value is 1 if it is 1, otherwise it is 0.

[0095] Risk propagation weighting Its function is to increase the weight of the risk propagation edge only when the correlation between fasteners is high enough.

[0096] S34. Calculate the node feature matrices for the k most recent time steps. The corresponding spatial adjacency matrix is ​​stored in the cache.

[0097] S4. Based on the constructed spatiotemporal diagram, predict the health status of fasteners and assess the failure risk, which specifically includes the following steps:

[0098] S41. Input the node feature matrix and spatial adjacency matrix into the graph convolution, and perform normalization processing at each time step to output the node feature sequence, which is specifically represented as follows:

[0099]

[0100] In the formula, For node feature sequences; ; Indicates the first The feature node matrix of all nodes at time 1. For trainable weight matrix, ReLU is a non-linear activation function. Given trainable bias vectors, graph convolutional layers are used to extract spatial dependencies;

[0101] S42. Analyze the node feature sequences at each time step. The input is a Long Short-Term Memory (LSTM) network for time modeling, generating time-dependent hidden state vectors to capture the fastener degradation trend, which is represented as follows:

[0102]

[0103] In the formula, Represents the hidden state vector; Represents the Long Short-Term Memory network; Represents the node feature sequence;

[0104] S43. Input the hidden state vector into the fully connected layer and map it to the interval using the Sigmoid function. The predicted health index is obtained, and the prediction result represents the health level of the fastener in the future inspection cycle. The predicted health index is expressed as:

[0105]

[0106] In the formula, A predictive health index representing the next Δ steps; Indicates the activation function; Represents the weight matrix; Represents the hidden state vector; Indicates the bias term;

[0107] S44. The predicted health index is inversely mapped to the failure probability. The higher the failure probability value, the greater the risk of structural or functional failure of the fastener in the future. The failure probability is expressed as:

[0108]

[0109] In the formula, This represents the failure probability in the next Δ steps.

[0110] S45. Using the failure probability as the initial risk value, and performing iterative updates based on the independent propagation model, the predicted risk index is obtained, which is expressed as:

[0111]

[0112] In the formula, A predictive risk indicator representing the next Δ step; For a moment Risk indicators For nodes i To the node j The intensity of risk transmission, The risk diffusion attenuation coefficient is the initial risk value. ;

[0113] The number of iteration steps is determined by the range of values ​​required for the observation area.

[0114] S46. Classify risks and issue early warnings based on risk indicators;

[0115] The risk classification includes:

[0116] Final risk indicators Mapped to levels 1–5:

[0117] →Level 1 (Safe);

[0118] 0.2–0.4 → Level 2 (Attention);

[0119] 0.4–0.6 → Level 3 (Medium Risk);

[0120] 0.6–0.8 → Level 4 (High Risk); ≥0.8 → Level 5 (Immediate Action Required);

[0121] When a high-level trigger occurs, immediate manual review and emergency repair suggestions will be provided.

[0122] Early warning output includes: receiving prediction results, determining whether to trigger an early warning based on health indices, and outputting concise alarm information to facilitate timely handling by maintenance personnel.

[0123] Example 2

[0124] This embodiment provides a system for predicting the health status and analyzing the risk diffusion of railway fasteners, referencing... Figure 2 It includes:

[0125] Data acquisition and processing unit; used to acquire images of track fasteners and environmental parameters.

[0126] Feature extraction and health index calculation unit; used for visual feature extraction and health index calculation of track fastener images;

[0127] Specifically, this unit is used to detect the position of fasteners, extract visual features, identify the status level, and calculate a smoothed health index.

[0128] Spatiotemporal diagram construction unit; used to construct a spatiotemporal diagram based on the visual characteristics, health index, and environmental parameters of the fastener;

[0129] Specifically, this unit is used to construct a graph-structured data that can be used by spatiotemporal prediction models by combining fastener data after feature extraction and health index calculation with track spatial location and historical health changes. Its functions include node generation, feature fusion, adjacency matrix construction, risk-related edge calculation, and time-series data caching.

[0130] Prediction and risk analysis unit; used to predict the health status of fasteners based on the constructed spatiotemporal diagram;

[0131] The visualization and early warning output unit is used to receive prediction results, determine whether to trigger an early warning based on health indices, and output concise alarm information for timely handling by maintenance personnel.

[0132] Although specific embodiments of the invention have been described in detail with reference to the accompanying drawings, this should not be construed as limiting the scope of protection of this patent. Various modifications and variations that can be made by a person skilled in the art without inventive effort within the scope described in the claims still fall within the scope of protection of this patent.

Claims

1. A method for predicting the health status and analyzing the risk diffusion of railway fasteners, characterized in that, Includes the following steps: S1. Acquire images of track fasteners and environmental parameters; S2. Visual feature extraction and health index calculation of track fastener images; S3. Construct a spatiotemporal diagram based on the visual characteristics, health index, and environmental parameters of the fasteners; S4. Based on the constructed spatiotemporal diagram, predict the health status of fasteners and assess the risk of failure; S2 specifically includes the following steps: S21. Lightweight target detection of track fastener images is used to locate the fasteners and output fastener image blocks; S22. Use a visual Transformer to extract the visual feature vector of the fastener image block; S23. Based on visual feature vectors, classify the fasteners according to their status and calculate their health index; S23 specifically includes: The visual feature vector is input into the state classifier, which outputs a fastener state label. This fastener state label is then mapped to a health index, which is represented as follows: In the formula, For the health index of fasteners, i Indicates the fastener serial number. t Indicates time; Indicates the status of the fastener; The fastener health index is smoothed using exponential sliding, and it is expressed as follows: In the formula, To smooth out health indicators; For smoothing coefficients; The smoothed health index represents the value at time t-1. S4 specifically includes the following steps: S41. Input the node feature matrix and spatial adjacency matrix into the graph convolution, and perform normalization processing at each time step to output the node feature sequence; S42. Input the node feature sequences at each time step into the long short-term memory network for time modeling and generate time-related hidden state vectors. S43. Input the hidden state vector into the fully connected layer and map it using the Sigmoid function to obtain the predicted health index, which is expressed as: In the formula, A predictive health index representing the next Δ steps; Indicates the activation function; Represents the weight matrix; Represents the hidden state vector; Indicates the bias term; S44. Reverse map the predicted health index to the failure probability; S45. Using the failure probability as the initial risk value, and performing iterative updates based on the independent propagation model, the predicted risk index is obtained. S46. Conduct risk classification and early warning output based on risk indicators.

2. The method for predicting the health status and analyzing the risk diffusion of railway fasteners according to claim 1, characterized in that, In step S1, GPS timestamps and mileage information are added to the collected track fastener images and environmental parameters, and the track fastener images are denoised and ROI cropped.

3. The method for predicting the health status and analyzing the risk diffusion of railway fasteners according to claim 1, characterized in that, S3 specifically includes the following steps: S31. Define each fastener as a graph node, and define node features based on environmental parameters, visual feature vectors, and smoothness health index, as follows: In the formula, Representing node characteristics, For environmental parameter vectors, , For visual feature dimensions, For the number of environmental parameters, For feature dimensions; Fasteners detected on the track k The node features corresponding to each time point form a node feature matrix. , This indicates the number of fasteners inspected on the track; S32. Generate a spatial adjacency matrix based on the track mileage distance between fasteners; S33. Calculate risk-related edges based on historical smoothed health index sequences; S34. Transform the node feature matrix The corresponding spatial adjacency matrix is ​​stored in the cache.

4. The method for predicting the health status and analyzing the risk diffusion of railway fasteners according to claim 3, characterized in that, In step S32, a spatial adjacency matrix is ​​generated based on the track mileage distance between fasteners, which is represented as follows: In the formula, Indicates fastener With fasteners Distance; Represents the spatial adjacency matrix; Indicates the smoothing coefficient; This represents the maximum connection distance.

5. The method for predicting the health status and analyzing the risk diffusion of railway fasteners according to claim 4, characterized in that, In step S33, the smooth health index sequence of the i-th fastener and the j-th fastener over a period of time is calculated based on the historical smooth health index sequence. and Pearson correlation coefficient between them: In the formula, This represents the function for calculating the correlation coefficient. This represents the Pearson correlation coefficient; when At that time, add a risk propagation weight term to the spatial adjacency matrix. It is represented as: In the formula, This represents the enhanced spatial adjacency matrix; Indicates the enhancement coefficient; This is the correlation threshold; As an indicator function, when the Pearson correlation coefficient Greater than the correlation threshold hour, The value is 1 if it is 1, otherwise it is 0.

6. The method for predicting the health status and analyzing the risk diffusion of railway fasteners according to claim 1, characterized in that, In step S44, the predicted health index is inversely mapped to the failure probability, which is expressed as: In the formula, This represents the failure probability in the next Δ steps.

7. The method for predicting the health status and analyzing the risk diffusion of railway fasteners according to claim 6, characterized in that, In step S45, the failure probability is used as the initial risk value, and iterative updates are performed based on the independent propagation model to obtain the predicted risk index, which is expressed as: In the formula, A predictive risk indicator representing the next Δ step; For a moment Risk indicators For nodes i To the node j The intensity of risk transmission, The risk diffusion attenuation coefficient is the initial risk value. .

Images