Method and system for preventing parts of high-speed rail overhead line system from falling off based on 4C detection

Abstract

The invention relates to the technical field of high-speed rail operation and maintenance, and provides a high-speed rail overhead line system part falling prevention method and system based on 4C detection, and the method comprises the steps: recognizing original image data through a target detection network model, and obtaining key part data of high-speed rail overhead line system parts; performing feature extraction on the key component data through a fine-grained detection network model to obtain key component feature data; based on the key component feature data, anti-loose state judgment is conducted on the key component, and anti-loose detection data of the key component is obtained; constructing a spatial-temporal characteristic matrix based on the anti-loosening detection data and the historical state data, and calculating a falling risk coefficient of the key component through a risk assessment model; and carrying out anti-falling early warning based on the falling risk coefficient, generating an anti-falling early warning report containing the positioning information, the abnormal type and the maintenance suggestion, and sending the anti-falling early warning report to a manager. According to the invention, the high-precision loosening state monitoring of the key anti-loosening component of the high-speed rail overhead line system is realized, and the falling risk can be early warned in advance.

Description

Technical Field

[0001] The present invention relates to the field of high-speed rail operation and maintenance technology, and in particular to a method and system for preventing high-speed rail contact network components from falling off based on 4C detection. Background Art

[0002] High-speed rail catenary systems are a core component of the power supply system. Their safe and stable operation is directly linked to the reliability and safety of railway operations. The catenary system consists of numerous anti-loosening components, including bolt-nut assemblies, locking washers, and cotter pins. These components are subject to significant vibration, temperature fluctuations, and mechanical stress during high-speed operation, making them prone to loosening or even falling off. Component detachment not only disrupts the normal operation of the power supply system but, in severe cases, can lead to catenary disconnection, equipment damage, and even major safety incidents.

[0003] Existing technologies for monitoring anti-loosening components in high-speed rail contact networks primarily utilize traditional image recognition technology combined with a simple threshold method for status detection, achieving basic component presence determination and obvious anomaly identification. However, this method employs a static judgment model based on a single time point and lacks the ability to dynamically monitor the loosening process of anti-loosening components and analyze spatiotemporal characteristics. It is unable to capture the temporal evolution of component loosening states and their spatial correlation characteristics. Consequently, it can only identify significant loosening that has already occurred, but cannot predict potential detachment risks. This results in low early warning and high hazard levels in monitoring anti-loosening components in high-speed rail contact networks, and results in a significant lag in early warning times. Summary of the Invention

[0004] In view of this, the present invention proposes a method and system for preventing high-speed rail contact network components from falling off based on 4C detection, which solves the problem that the existing technology lacks the ability to dynamically monitor the loosening process of anti-loosening components and analyze the spatiotemporal characteristics, and is unable to capture the laws and spatial correlation characteristics of the evolution of the loosening state of components over time, resulting in only being able to identify obvious loosening that has occurred, but unable to predict potential falling risks, resulting in low warning and high hazard of high-speed rail contact network anti-loosening component monitoring, and serious delay in warning time.

[0005] The technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides a method for preventing components from falling off of a high-speed railway contact network based on 4C detection, comprising the following steps: Acquire multi-source detection data of high-speed rail contact network components, wherein the multi-source detection data includes original image data and historical status data collected by 4C detection devices; The target detection network model is used to identify raw image data and obtain key component data of high-speed rail contact network components, including bolt-nut combinations, stop washers, and cotter pins. Extracting features from the key component data using a fine-grained detection network model to obtain key component feature data; Performing anti-loosening status judgment on the key component based on the key component characteristic data to obtain anti-loosening detection data of the key component; A spatiotemporal feature matrix is ​​constructed based on the anti-loosening detection data and historical status data, and the shedding risk coefficient of the key components is calculated through a risk assessment model; Based on the falling risk coefficient, an anti-falling warning is performed, and an anti-falling warning report including positioning information, abnormality type and maintenance suggestions is generated and sent to the management personnel.

[0006] On the basis of the above technical solution, preferably, the multi-source detection data of the high-speed railway contact network components is obtained, and the multi-source detection data includes original image data and historical status data collected by the 4C detection device, including: Collecting original image data of high-speed rail contact network components using a 4C detection device, wherein the original image data includes image data acquired at different angles and under different lighting conditions; The historical status data of the high-speed railway contact network components are read from the historical status database, wherein the historical status data includes historical anti-loosening detection data and related maintenance records.

[0007] Based on the above technical solutions, preferably, the target detection network model is used to identify the original image data to obtain the key component data of the high-speed rail contact network components, including: An object detection network model was built based on the YOLOv5 network. The model generated candidate regions and an adaptive feature pyramid was used to extract multi-scale features to obtain the coarse-grained locations of the bolt-nut combination, the lock washer, and the cotter pin. A dynamic confidence threshold is used to screen the candidate regions, and the overlapping candidate regions are fused and optimized in combination with the target size distribution characteristics to output the precise bounding box of the key components.

[0008] On the basis of the above technical solution, preferably, the feature extraction of the key component data by a fine-grained detection network model to obtain the key component feature data includes: The fine-grained detection network model includes a multi-branch feature extraction network and a multi-scale key point regression network; Performing fine-grained feature extraction on the key components through a multi-branch feature extraction network, wherein the multi-branch feature extraction network includes a morphological feature branch, a geometric relationship feature branch, and a texture feature branch; The characteristic points of the key components are accurately located based on a multi-scale key point regression network. The characteristic points of the key components include the endpoints of the anti-loosening mark line, the intersection of the long and short ends of the stop gasket, and the root and bending points of the cotter pin.

[0009] On the basis of the above technical solution, preferably, the step of judging the anti-loosening state of the key component based on the key component characteristic data to obtain the anti-loosening detection data of the key component includes: Calculate the corresponding anti-loosening state parameters based on the characteristic data of different key components. The anti-loosening state parameters include the linear offset of the anti-loosening mark line, the spatial angle between the stop washer and the wire clamp body, and the angle between the two curved edges of the cotter pin. Based on the preset standard values ​​of the anti-loosening state parameters, the calculated anti-loosening state parameters are judged by threshold value, the anti-loosening state level of each key component is determined, and standardized anti-loosening detection data is generated.

[0010] On the basis of the above technical solution, preferably, the construction of a spatiotemporal feature matrix based on the anti-loosening detection data and the historical status data, and the calculation of the risk coefficient of the key component falling off through the risk assessment model include: Construct a spatiotemporal feature matrix, align the anti-loosening detection data with the historical status data in the time and space dimensions, and obtain a multi-dimensional feature matrix that reflects the time-varying and spatial distribution of the loosening status of key components; Through the deep integration risk assessment model, the spatiotemporal feature matrix is ​​identified to obtain the shedding risk coefficient of each key component; The deep integrated risk assessment model adopts a three-layer architecture, where: The first layer establishes a component looseness foundation risk assessment model based on the gradient boosting decision tree to generate an initial risk coefficient; The second layer combines the long short-term memory network to capture the trend characteristics of the time series and perform time dimension correction on the initial risk coefficient; The third layer uses a graph convolutional network to model the spatial topological relationship between components and performs spatial dimension risk correction on the initial risk coefficient.

[0011] On the basis of the above technical solution, preferably, the anti-falling warning is performed based on the falling risk coefficient, and an anti-falling warning report including positioning information, abnormality type and maintenance suggestions is generated and sent to the management personnel, including: Based on the shedding risk factor, combined with component importance and line operation level, a multi-level early warning mechanism is established to generate an anti-shedding early warning report containing precise positioning information, abnormality type analysis, and maintenance priority. The multi-level early warning mechanism includes information level, prompt level, warning level and emergency level, among which: For information level, the test results are recorded but no warning is triggered; for prompt level, a low-priority warning is generated and attention is paid to during the next routine maintenance; for warning level, a medium-priority warning is generated and special maintenance is carried out within a week; for emergency level, a high-priority warning is generated and emergency treatment is carried out within 24 hours; The early warning report is sent to relevant management personnel, and maintenance recommendations and resource allocation plans are generated based on the risk characteristics of component loosening and historical maintenance experience.

[0012] In a second aspect, the present invention further provides a high-speed railway contact network component loss prevention system based on 4C detection, the system comprising: A data acquisition module is used to acquire multi-source detection data of high-speed rail contact network components, wherein the multi-source detection data includes original image data and historical status data collected by 4C detection devices; The key component recognition module is used to identify raw image data through the target detection network model and obtain key component data of the high-speed rail contact network components, including bolt-nut combinations, stop washers, and cotter pins; A feature extraction module is used to extract features from the key component data using a fine-grained detection network model to obtain key component feature data; The anti-loosening detection module is used to judge the anti-loosening state of the key component based on the key component characteristic data, and obtain the anti-loosening detection data of the key component: A falling risk calculation module is used to construct a spatiotemporal feature matrix based on the anti-loosening detection data and the historical status data, and calculate the falling risk coefficient of the key component through a risk assessment model; The anti-fall warning module is used to perform anti-fall warning based on the fall risk coefficient, generate an anti-fall warning report including positioning information, abnormality type and maintenance suggestions, and send it to the management personnel.

[0013] In a third aspect, the present invention further provides an electronic device comprising: at least one processor, at least one memory, a communication interface, and a bus; Among them, the processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement steps such as a method for preventing high-speed rail contact network components from falling off based on 4C detection.

[0014] In a fourth aspect, the present invention further provides a computer-readable storage medium, which stores computer instructions, and the computer instructions enable a computer to implement steps such as a method for preventing high-speed rail contact network components from falling off based on 4C detection.

[0015] The method and system for preventing high-speed railway contact network components from falling off based on 4C detection of the present invention have the following beneficial effects compared with the prior art: (1) Through multi-source data collection, a two-level detection model is constructed to perform anti-loosening status judgment, risk assessment and multi-level early warning. The multi-scale feature extraction of the integrated target detection and fine-grained detection network is used. The state parameter precise quantification method based on geometric features, three-dimensional posture estimation and angle calculation is combined with the dynamic prediction of the spatiotemporal feature matrix and the deep integrated risk assessment model. The high-precision loosening status monitoring of the key anti-loosening components of the high-speed rail contact network is achieved, which can give early warning of the risk of falling off, build a closed-loop management mechanism, effectively prevent the pantograph network failure caused by loose components, and ensure the power supply safety under high-speed operation conditions. (2) By integrating adaptive feature pyramid, dynamic confidence threshold and multi-task loss function into the target detection network model, the ability to accurately locate and identify key components of the high-speed rail contact network in complex environments is achieved. The adaptive feature pyramid dynamically adjusts the feature weights of different levels through the channel attention mechanism, enhancing the detection sensitivity of small components such as cotter pins. The dynamic confidence threshold adaptively adjusts the detection threshold according to the component size characteristics, significantly reducing the error recognition rate of the bolt-nut combination and background, and improving the detection rate of small targets. (3) By combining a multi-branch feature extraction network and a multi-scale key point regression network in a fine-grained detection network model, accurate feature extraction and sub-pixel key point positioning of key anti-loosening components of the high-speed rail contact network are achieved. The multi-branch feature extraction network uses three dedicated feature channels, namely morphology, geometric relationship and texture, to enhance the features of the physical characteristics of different types of components, and uses an adaptive feature enhancement function to dynamically adjust the weights, so that the linear features of the anti-loosening marking line, the planar geometric features of the stop gasket and the curvature features of the cotter pin can be optimally expressed at the same time. The multi-scale key point regression network uses a collaborative regression mechanism to geometrically constrain the key points as a whole, which improves the positioning accuracy in harsh environments such as uneven lighting, partial occlusion and changing angles, reduces the positioning error of the anti-loosening marking line endpoint, and improves the angle measurement accuracy of the stop gasket. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 This is a flow chart of a method for preventing high-speed railway contact network components from falling off based on 4C detection of the present invention; Figure 2 This is a structural diagram of a high-speed railway contact network component loss prevention system based on 4C detection according to the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] See also Figure 1 The present invention provides a method for preventing high-speed railway contact network components from falling off based on 4C detection, comprising the following steps: Acquire multi-source detection data of high-speed rail contact network components, wherein the multi-source detection data includes original image data and historical status data collected by 4C detection devices; Recognizing raw image data through a target detection network model to obtain key component data of high-speed rail contact network components, including bolt-nut assemblies, stop washers, and cotter pins; Extracting features from the key component data using a fine-grained detection network model, locating the anti-loosening marking line, the long and short ends of the stop washer, and the bending end points of the cotter pin, to obtain key component feature data; The anti-loosening state of the key component is judged based on the key component characteristic data to obtain the anti-loosening detection data of the key component, including: a. Linear offset judgment based on the geometric characteristics of the anti-loosening marking line b. Analyze the spatial relationship between the stop washer and the clamp body through 3D posture estimation c. Calculate the angle between the two curved edges of the cotter pin; A spatiotemporal feature matrix is ​​constructed based on the anti-loosening detection data and historical status data, and the shedding risk coefficient of the key components is calculated through a risk assessment model; Based on the falling risk coefficient, an anti-falling warning is performed, and an anti-falling warning report including positioning information, abnormality type and maintenance suggestions is generated and sent to the management personnel.

[0020] Specifically, this embodiment collects multi-source data to build a two-level detection model to perform anti-loosening status judgment, risk assessment and multi-level warning, integrates target detection and multi-scale feature extraction of fine-grained detection network, and accurately quantifies state parameters based on geometric features, three-dimensional posture estimation and angle calculation. Combined with the dynamic prediction of spatiotemporal feature matrix and deep integrated risk assessment model, it realizes high-precision loosening status monitoring of key anti-loosening components of high-speed rail contact network, can give early warning of falling risks, build a closed-loop management mechanism, effectively prevent bow network failures caused by loose components, and ensure power supply safety under high-speed operation conditions.

[0021] The multi-source detection data of the high-speed railway contact network components is obtained, and the multi-source detection data includes original image data and historical status data collected by the 4C detection device, including: The original image data of the high-speed railway contact network components are collected by a 4C detection device, and the original image data includes image data obtained at different angles and under different lighting conditions.

[0022] In a specific embodiment, the collected raw image data is preprocessed, and the preprocessing steps include image denoising, image enhancement, image cropping and distortion correction to improve image resolution and accuracy of subsequent target detection.

[0023] The historical status data of the high-speed railway contact network components are read from the historical status database, wherein the historical status data includes historical anti-loosening detection data and related maintenance records.

[0024] In a specific embodiment, the read historical status data is normalized and abnormal information is marked. The normalization is used to eliminate the magnitude differences in the data caused by environmental factors, and the abnormal information marking is used to provide reliable data support for constructing a spatiotemporal feature matrix and risk assessment.

[0025] Specifically, this embodiment achieves high-quality data acquisition and processing capabilities in a complex high-speed rail contact network environment by adopting a four-step multi-stage image preprocessing process and dual-channel data normalization processing.

[0026] The image acquisition strategy for different angles and lighting conditions effectively solves the problems of strong light reflection, limited angles and changing environments in high-speed rail contact network inspection, improves image resolution and the edge clarity of small components such as cotter pins; at the same time, through the normalization processing of historical status data and the abnormality marking mechanism, it eliminates data deviations caused by different environments, equipment and time, and establishes a standardized data set with consistent data level and balanced information density, which improves model training efficiency, reduces sensitivity to environmental changes, and ensures the stability of detection capabilities under various severe weather and lighting conditions.

[0027] The method of identifying the original image data through the target detection network model and obtaining the key component data of the high-speed rail contact network components includes: An object detection network model was constructed based on the YOLOv5 network. The candidate regions were generated through the object detection network model. An adaptive feature pyramid was used to extract multi-scale features to obtain the coarse-grained positions of the bolt-nut combination, the lock washer, and the cotter pin.

[0028] In a specific embodiment, the feature map weight calculation formula of the adaptive feature pyramid is: ; in, For the The fusion weight of the layer feature map, is the channel attention module, For the Layer feature map, is the first learnable parameter, is the number of feature layers.

[0029] A dynamic confidence threshold is used to screen the candidate regions, and the overlapping candidate regions are fused and optimized in combination with the target size distribution characteristics to output the precise bounding box of the key components.

[0030] In a specific embodiment, the dynamic confidence threshold is calculated as follows: ; in, For key components The confidence threshold of the class, The key component in the training set The mean target size of the class, The key component in the training set Standard deviation of target size for the class, is the first empirical coefficient, is the second empirical coefficient, is the first anti-zero constant.

[0031] In a specific embodiment, the loss function calculation formula of the target detection network model is: In a specific embodiment, the loss function calculation formula of the target detection network model is: ; ; ; ; in, is the loss function of the target detection network model, is the Focal loss function, is the CIoU loss function, is the angle loss function, 、 、 They are the balance coefficients of Focal loss function, CIoU loss function, and angle loss function, respectively. For the model The probability value of a sample being predicted as a positive class, is the weight ratio adjustment factor of difficult and easy samples, Image position The spatial attention score at is the total number of samples, is the intersection-over-union (IoU) of the predicted bounding box and the true bounding box, To predict the center point of the bounding box and the center point of the ground truth bounding box The Euclidean distance between is the diagonal length of the minimum enclosing rectangle that covers both the predicted bounding box and the true bounding box, is the trade-off coefficient, is the aspect ratio consistency measurement parameter, To predict the target rotation angle, is the real target rotation angle.

[0032] In a specific embodiment, the target detection network model adopts an improved YOLOv5 network architecture and is optimized in three aspects to meet the component detection requirements in the complex environment of the high-speed rail contact network.

[0033] First, an adaptive feature pyramid structure is introduced to assign optimal weights to feature maps at different levels. In experimental verification, when detecting small cotter pins, the weight of shallow feature maps automatically increases to 0.65, while the weight of deep feature maps decreases to 0.20, effectively enhancing the detection capability of objects with side lengths less than 30 pixels. Second, a dynamic confidence threshold mechanism is designed to adaptively adjust the threshold based on the target size distribution. For bolt-nut combinations (average size 120 pixels) in a high-speed rail catenary environment, the confidence threshold is set to 0.42, while for cotter pins (average size 25 pixels), the confidence threshold is lowered to 0.32, reducing the missed detection rate of small objects. Third, a composite loss function is constructed that includes focal loss, CIoU loss, and angle loss. In particular, the introduction of angle loss enables the model to accurately capture the pose of components in a rotated state, reducing the angle prediction error from the original ±12° to ±3.5° during model training.

[0034] Specifically, this embodiment achieves high-precision identification of key anti-loosening components of high-speed rail contact networks in complex environments through the improved target detection network model, effectively overcoming the three major difficulties of traditional target detection in high-speed rail scenarios: first, it solves the detection problem of small-sized components such as cotter pins (with a diameter of only 2-3 mm), and the detection rate is increased from 68.3% of the baseline model to 94.7%; second, it significantly improves the detection balance of components of different sizes, and the difference in F1-score for large, medium and small targets is reduced from the original 0.18 to 0.05; finally, through the angle perception mechanism, the model can maintain stable detection performance under a wide range of viewing angles of 30°-150°, providing accurate target positioning and segmentation for fine-grained anti-loosening status analysis, reducing the computational redundancy in the feature extraction stage, and improving the overall inference speed, while improving the adaptability and stability in actual high-speed rail line inspections.

[0035] The feature extraction of the key component data by the fine-grained detection network model to obtain the key component feature data includes: The fine-grained detection network model includes a multi-branch feature extraction network and a multi-scale key point regression network; Fine-grained feature extraction is performed on the key components through a multi-branch feature extraction network, where the multi-branch feature extraction network includes a morphological feature branch, a geometric relationship feature branch, and a texture feature branch.

[0036] In a specific embodiment, the multi-branch feature extraction network adopts an adaptive feature enhancement function to dynamically adjust the feature weight coefficient of each branch according to the characteristics of different key components, specifically: For the anti-loosening marking line, the weight of the texture feature branch is enhanced; For stop washers, increase the weight of the geometric relationship feature branch; For cotter pins, the weight of the morphological feature branch is enhanced; At the same time, multi-scale response fusion is performed on each feature map to eliminate the feature imbalance problem caused by size differences.

[0037] In a specific embodiment, the calculation formula of the adaptive feature enhancement function is: In a specific embodiment, the target detection network model adopts an improved YOLOv5 network architecture and is optimized in three aspects to meet the component detection requirements in the complex environment of the high-speed rail contact network.

[0038] First, an adaptive feature pyramid structure is introduced to assign optimal weights to feature maps at different levels. In experimental verification, when detecting small cotter pins, the weight of shallow feature maps automatically increases to 0.65, while the weight of deep feature maps decreases to 0.20, effectively enhancing the detection capability of objects with side lengths less than 30 pixels. Second, a dynamic confidence threshold mechanism is designed to adaptively adjust the threshold based on the target size distribution. For bolt-nut combinations (average size 120 pixels) in a high-speed rail catenary environment, the confidence threshold is set to 0.42, while for cotter pins (average size 25 pixels), the confidence threshold is lowered to 0.32, reducing the missed detection rate of small objects. Third, a composite loss function is constructed that includes focal loss, CIoU loss, and angle loss. In particular, the introduction of angle loss enables the model to accurately capture the pose of components in a rotated state, reducing the angle prediction error from the original ±12° to ±3.5° during model training.

[0039] Specifically, this embodiment achieves high-precision identification of key anti-loosening components of high-speed rail contact networks in complex environments through the improved target detection network model, effectively overcoming the three major difficulties of traditional target detection in high-speed rail scenarios: first, it solves the detection problem of small-sized components such as cotter pins (with a diameter of only 2-3 mm), and the detection rate is increased from 68.3% of the baseline model to 94.7%; second, it significantly improves the detection balance of components of different sizes, and the difference in F1-score for large, medium and small targets is reduced from the original 0.18 to 0.05; finally, through the angle perception mechanism, the model can maintain stable detection performance under a wide range of viewing angles of 30°-150°, providing accurate target positioning and segmentation for fine-grained anti-loosening status analysis, reducing the computational redundancy in the feature extraction stage, and improving the overall inference speed, while improving the adaptability and stability in actual high-speed rail line inspections.

[0040] The feature extraction of the key component data by the fine-grained detection network model to obtain the key component feature data includes: The fine-grained detection network model includes a multi-branch feature extraction network and a multi-scale key point regression network; Fine-grained feature extraction is performed on the key components through a multi-branch feature extraction network, where the multi-branch feature extraction network includes a morphological feature branch, a geometric relationship feature branch, and a texture feature branch.

[0041] In a specific embodiment, the multi-branch feature extraction network adopts an adaptive feature enhancement function to dynamically adjust the feature weight coefficient of each branch according to the characteristics of different key components, specifically: For the anti-loosening marking line, the weight of the texture feature branch is enhanced; For stop washers, increase the weight of the geometric relationship feature branch; For cotter pins, the weight of the morphological feature branch is enhanced; At the same time, multi-scale response fusion is performed on each feature map to eliminate the feature imbalance problem caused by size differences.

[0042] In a specific embodiment, the calculation formula of the adaptive feature enhancement function is: ; in, For the Enhanced features of key components of the class, For the Basic features of key components of the class, For the Heat map response of key components of the class, For the The heat map mean of key components of the class, For the Standard deviation of the heat map of key components of the class, For the The enhancement factor of key components of the class, is the hyperbolic tangent activation function.

[0043] The characteristic points of the key components are accurately located based on a multi-scale key point regression network. The characteristic points of the key components include the endpoints of the anti-loosening mark line, the intersection of the long and short ends of the stop gasket, and the root and bending points of the cotter pin.

[0044] In a specific embodiment, the multi-scale key point regression network adopts a collaborative regression mechanism to constrain the key feature points as a geometric structure as a whole, and improves the geometric consistency of feature point positioning by introducing a geometric relationship loss function between points. The specific methods include: for anti-loosening marking lines, the endpoint positioning accuracy is enhanced through linearity constraints; for stop washers, the accuracy of the long and short end intersection points is improved through plane orthogonal constraints; for cotter pins, the geometric rationality of the bending points is ensured through angle constraints.

[0045] In one specific embodiment, the fine-grained detection network model adopts a dual-core architecture consisting of a multi-branch feature extraction network based on an improved ResNeXt-50 and a multi-scale keypoint regression network. First, the multi-branch feature extraction network achieves accurate feature extraction of different types of anti-loosening components through three dedicated feature channels: the morphological feature branch uses a deformable convolutional network, which is particularly suitable for extracting the curvature features of cotter pins. In experiments, the offset field of the deformable convolutional network can adaptively adjust to the curvature variation of the cotter pin edge, effectively improving edge accuracy; the geometric relationship feature branch is constructed based on a graph convolutional network. By establishing the geometric relationship constraints of the stop gasket plane, the gasket edge feature points are processed as a graph structure, accurately capturing the relative position relationship of the long and short ends of the gasket, and reducing the spatial pose estimation error to 2.8°; the texture feature branch uses a depthwise separable convolution to enhance the subtle texture changes of the anti-loosening marking line, extracting a marking line offset of only 0.15mm under low contrast.

[0046] Secondly, in this embodiment, an adaptive feature enhancement function dynamically assigns weights based on the characteristics of different components. Through a heatmap response mechanism, the weight of the texture branch is automatically increased to 0.58 when detecting anti-loosening marking lines, the weight of the geometry branch is increased to 0.62 when detecting stop washers, and the weight of the morphology branch is increased to 0.65 when detecting cotter pins. This embodiment's dynamic weight adjustment strategy improves feature expression by 18.7% compared to fixed weights, maintaining high feature extraction quality, especially under partial occlusion conditions.

[0047] By incorporating a collaborative regression mechanism and geometric constraint loss, the multi-scale keypoint regression network applies a linearity constraint of 0.35 to the anti-loosening marking line, ensuring endpoint positioning accuracy remains at 0.18mm even under vibration conditions at 350 km / h. A plane orthogonality constraint is introduced for the stop washer to ensure the angular measurement error between the washer and the clamp is within ±2.5°. A curvature-based angle constraint is applied to the cotter pin, establishing a loosening warning criterion by calculating the angle between the pin's base and the bending points at both ends. In actual line testing, the model maintained a 98.3% keypoint detection rate and submillimeter positioning accuracy across a temperature range of -10°C to 40°C, as well as in adverse weather conditions such as rain, fog, and snow.

[0048] Specifically, the fine-grained detection network model of this embodiment achieves high-precision feature extraction and key point positioning of key anti-loosening components of high-speed rail contact networks. The multi-branch feature extraction strategy accurately captures the key physical features of different anti-loosening devices, enabling the model to simultaneously process components of varying shapes, increasing the accuracy from 87.2% of the standard convolutional network to 96.8%. The adaptive feature enhancement function reduces feature instability under different lighting and angle conditions, improving feature quality in shadowed areas by 42% and feature detection capabilities in overexposed edge areas by 37%. The geometric constraint mechanism of the multi-scale key point regression network ensures the physical rationality of key point positions, solving the key point drift problem in traditional methods and increasing the average accuracy of overall anti-loosening device identification from 76.5% of the baseline model to 94.2%.

[0049] The step of judging the anti-loosening state of the key component based on the key component characteristic data to obtain the anti-loosening detection data of the key component includes: Based on the characteristic data of different key components, the corresponding anti-loosening state parameters are calculated. The anti-loosening state parameters include the linear offset of the anti-loosening marking line, the spatial angle between the stop gasket and the wire clamp body, and the angle value of the two curved edges of the cotter pin.

[0050] In a specific embodiment, for the anti-loosening marking line, the straight line segments are extracted by Hough transform, and the angle deviation and endpoint offset between adjacent marking lines are calculated to characterize the looseness of the bolt-nut combination; for the stop washer, the spatial position of the stop washer is reconstructed using a three-dimensional posture estimation algorithm, and the fitting angle between its long end and the side end face of the wire clamp body and the fitting angle between the short end and the nut in six aspects are calculated; for the cotter pin, the skeleton lines of the two curved edges are extracted, the angle between the two edges is calculated, and compared with the standard installation angle to analyze its deformation degree.

[0051] Based on the preset standard values ​​of the anti-loosening state parameters, the calculated anti-loosening state parameters are judged by threshold value, the anti-loosening state level of each key component is determined, and standardized anti-loosening detection data is generated.

[0052] In a specific embodiment, based on the mechanical properties of the anti-loosening component and historical fault data, a three-level anti-loosening state judgment standard is established, wherein: normal state: the anti-loosening mark line offset is less than 0.3mm, the stop gasket position deviation angle is less than 5°, and the cotter pin bending angle is between 120° and 130°; slightly loose state: the anti-loosening mark line offset is between 0.3mm and 0.5mm, the stop gasket position deviation angle is between 5° and 10°, and the cotter pin bending angle deviates from the standard range by no more than ±3°; seriously loose state: the anti-loosening mark line offset is ≥0.5mm, the stop gasket position deviation angle is ≥10°, and the cotter pin bending angle deviates from the standard range by more than ±5°; at the same time, the stress state, installation position and environmental factors of the component are comprehensively considered to correct the judgment result, so as to improve the adaptability and reliability of the anti-loosening state judgment.

[0053] Specifically, this embodiment establishes an accurate anti-loosening state calculation and graded judgment mechanism based on physical characteristics for key components of the high-speed rail contact network, thereby achieving millimeter-level precision in loosening state quantification and early risk identification capabilities.

[0054] For the anti-loosening marking line, the Hough transform is used to extract the straight line segment and calculate the offset, which improves the accuracy of bolt-nut loosening detection; the three-dimensional posture estimation algorithm is used to reconstruct the spatial position of the stop gasket and calculate the fitting angle, which solves the problem that the traditional method cannot accurately judge the spatial posture change of the gasket in the complex environment of high-speed rail, and the measurement accuracy of the position deviation angle is improved; the skeleton line extraction and angle calculation method are used for the cotter pin to improve the sensitivity of deformation detection.

[0055] At the same time, the three-level anti-loosening status judgment standard (normal state, slightly loose state and severely loose state) established based on the mechanical properties of anti-loosening components and historical failure data, combined with the comprehensive correction mechanism of component stress state, installation position and environmental factors, has greatly improved the reliability of anti-loosening status judgment, allowing it to maintain an accuracy rate of 96.3% even in the complex high-speed rail operating environment.

[0056] The method of constructing a spatiotemporal feature matrix based on the anti-loosening detection data and the historical status data and calculating the shedding risk coefficient of the key components through the risk assessment model includes: A spatiotemporal feature matrix is ​​constructed, and the anti-loosening detection data and historical status data are aligned in the time dimension and space dimension to obtain a multidimensional feature matrix that reflects the temporal changes and spatial distribution of the loosening status of key components.

[0057] In a specific embodiment, the spatiotemporal feature matrix is ​​composed of the following static features, time series features, and spatial correlation features, wherein: Static features include component type, installation location, load characteristics, and line grade. Time series features include the historical trend, rate of change, periodic fluctuations, and abnormal mutation points of component anti-loosening parameters. Spatially correlated features include the correlation of loosening states between adjacent components, the loosening associations between different components within the same assembly, and regional environmental influencing factors. By aligning and standardizing features in both time and space, inconsistent inspection frequencies and environmental interference are eliminated, enhancing the consistency and representativeness of the feature matrix.

[0058] Through the deep integration of risk assessment model, the spatiotemporal feature matrix is ​​identified, and the component type, operating environment, stress condition and historical change trend are comprehensively considered to obtain the shedding risk coefficient of each key component.

[0059] The deep integrated risk assessment model adopts a three-layer architecture, where: The first layer establishes a component looseness foundation risk assessment model based on the gradient boosting decision tree to generate an initial risk coefficient; The second layer combines the long short-term memory network to capture the trend characteristics of the time series and perform time dimension correction on the initial risk coefficient; The third layer uses a graph convolutional network to model the spatial topological relationship between components, comprehensively analyzes the impact of the loose state of adjacent components on the target component, and performs spatial dimension risk correction on the initial risk coefficient; In a specific embodiment, the initial risk coefficient is mapped to the interval [0, 1] through normalization processing, and is divided into three levels according to the risk level: low risk (0-0.3), medium risk (0.3-0.7) and high risk (0.7-1.0).

[0060] Specifically, this embodiment achieves high-precision prediction capability for the risk of falling off of high-speed rail contact network anti-loosening components by constructing a three-dimensional spatiotemporal feature matrix containing static features, time series features and spatial correlation features, combined with the three-layer architecture of the deep integrated risk assessment model.

[0061] The spatiotemporal feature matrix of this embodiment organically integrates the gradual characteristics of the loose state of components in the time dimension and the correlation in the spatial dimension, solving the problem that traditional single-point detection methods cannot capture the evolution law of looseness. In the deep integrated risk assessment model, the first-layer gradient boosting decision tree can identify the key factors of the basic risk of component loosening with an accuracy rate of 92.5%; the second-layer long-short-term memory network can capture the historical trend characteristics of up to 30 days, effectively predict the acceleration period of loosening, and extend the time warning lead time from the original 1-2 days to 3-7 days; the third-layer graph convolutional network models the spatial topological relationship between components and mines the loose transmission chain in the same connection group, improving the accuracy of spatial correlation identification by 41%.

[0062] The anti-fall warning is performed based on the fall risk coefficient, and an anti-fall warning report including positioning information, abnormality type and maintenance suggestions is generated and sent to the management personnel, including: Based on the shedding risk factor, combined with component importance and line operation level, a multi-level early warning mechanism is established to generate an anti-shedding early warning report containing precise positioning information, abnormality type analysis, and maintenance priority. The multi-level early warning mechanism includes information level, prompt level, warning level and emergency level, among which: If the information level dropout risk coefficient is below 0.3, the detection result is recorded and no warning is triggered; The warning-level shedding risk coefficient is between 0.3 and 0.5, generating a low-priority warning to be paid attention to during the next routine maintenance; The warning-level shedding risk coefficient is between 0.5 and 0.7, generating a medium-priority warning and requiring special maintenance within one week. The emergency-level drop risk coefficient is above 0.7, generating a high-priority warning that requires emergency processing within 24 hours.

[0063] In a specific embodiment, the risk of clustering of abnormalities occurring simultaneously in multiple components within the same section is assessed. When the density of abnormal components in a section exceeds a preset threshold, the overall warning level of the section is increased to prevent systemic risks.

[0064] Early warning reports are sent to relevant managers through the intelligent distribution system, and maintenance recommendations and resource allocation plans are automatically generated based on the risk characteristics of component loosening and historical maintenance experience.

[0065] In a specific embodiment, the intelligent distribution system adopts a hierarchical, multi-channel information delivery method, including: Automatically select notification channels based on warning levels, including system messages, emails, text messages, and emergency voice calls; intelligently identify warning recipients based on organizational structure and job responsibilities to ensure that key information is simultaneously delivered to three types of personnel: on-site maintenance, technical support, and management decision-making; automatically generate standardized maintenance recommendations for different types of loosening anomalies, including anomaly descriptions, possible causes, recommended tools, spare parts requirements, and labor time estimates, in combination with the historical maintenance database; provide closed-loop management of the maintenance process, requiring maintenance personnel to feedback actual processing results and transmit processing information back to the historical status database to improve the data samples for loosening pattern recognition and continuously improve the accuracy of the risk assessment model.

[0066] Specifically, this embodiment realizes the full-process automated management of high-speed railway contact network anti-loosening components from risk assessment to warning processing by establishing a multi-level warning mechanism and intelligent distribution system based on the shedding risk coefficient.

[0067] The multi-level warning mechanism divides the risk of shedding into four levels: information, prompt, warning, and emergency. Dynamic adjustments are made based on component importance and line operation levels, addressing the crudeness and blindness of traditional binary warning models and improving warning accuracy. An intelligent recognition algorithm for clustered risks automatically raises the overall warning level when the density of abnormal components within a 5-kilometer section exceeds a set threshold.

[0068] At the same time, a hierarchical, multi-channel intelligent distribution system automatically selects the optimal notification method based on the warning level, reducing emergency warning response time from 120 minutes to less than 15 minutes. Intelligent warning routing based on job responsibilities ensures simultaneous information transmission to maintenance, technical, and management teams, eliminating information silos. Standardized maintenance recommendations automatically generated for different loosening types include anomaly descriptions, possible causes, and resource requirement estimates, improving maintenance efficiency and shortening average processing time. A closed-loop management mechanism continuously optimizes risk assessment models by collecting maintenance feedback data, increasing warning accuracy from an initial 87.5% to 96.8% within six months and reducing false alarm rates by 71%.

[0069] See also Figure 2 The present invention also provides a high-speed railway contact network component falling-off prevention system based on 4C detection, the system comprising: A data acquisition module is used to acquire multi-source detection data of high-speed rail contact network components, wherein the multi-source detection data includes original image data and historical status data collected by 4C detection devices; The key component recognition module is used to identify the original image data through the target detection network model and obtain the key component data of the high-speed rail contact network components; A feature extraction module is used to extract features from the key component data using a fine-grained detection network model to obtain key component feature data; The anti-loosening detection module is used to judge the anti-loosening state of the key component based on the key component characteristic data, and obtain the anti-loosening detection data of the key component: A falling risk calculation module is used to construct a spatiotemporal feature matrix based on the anti-loosening detection data and the historical status data, and calculate the falling risk coefficient of the key component through a risk assessment model; The anti-fall warning module is used to perform anti-fall warning based on the fall risk coefficient, generate an anti-fall warning report including positioning information, abnormality type and maintenance suggestions, and send it to the management personnel.

[0070] Specifically, this embodiment of a high-speed rail contact network component loss prevention system based on 4C detection implements intelligent closed-loop monitoring and early warning of high-speed rail contact network components through six functional modules: data acquisition, key component identification, feature extraction, anti-loosening detection, loss risk calculation, and loss prevention early warning. This improves the system's integration and synergy. By adopting a modular design and distributed computing architecture, seamless data flow and processing are achieved between functional modules, reducing system response latency and improving data processing consistency.

[0071] Especially in the high-speed rail operating environment, a high-speed rail contact network component prevention system based on 4C detection in this embodiment shows three major advantages: First, through the deep integration of the data acquisition module and the key part identification module, adaptive acquisition and preprocessing of images in a mobile environment are realized, so that the system can maintain extremely high recognition accuracy under high-speed operation conditions of 350km / h; second, the linkage mechanism between the feature extraction module and the anti-loosening detection module enables maintenance decisions to be transformed from traditional experience judgment to data-driven intelligent analysis, and resource allocation efficiency is improved; third, the shedding risk calculation module and the anti-shedding warning module realize dynamic distribution of computing loads based on the cloud-edge collaborative architecture, and the processing capacity is increased by 68% under high load conditions, while ensuring the real-time nature of emergency warnings.

[0072] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory communication interface and a bus: wherein the processor, memory and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement a method for preventing high-speed rail contact network components from falling off based on 4C detection.

[0073] The present invention also discloses a computer-readable storage medium storing computer instructions that cause the computer to implement all or part of the steps of a method for preventing high-speed rail contact network components from falling off based on 4C detection, as described in an embodiment of the present invention. The storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0074] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for preventing components from falling off in a high-speed railway contact network based on 4C detection, characterized in that: The following steps are involved: Acquire multi-source detection data of high-speed rail contact network components, wherein the multi-source detection data includes original image data and historical status data collected by 4C detection devices; The target detection network model is used to identify raw image data and obtain key component data of high-speed rail contact network components, including bolt-nut combinations, stop washers, and cotter pins. Extracting features from the key component data using a fine-grained detection network model to obtain key component feature data; Performing anti-loosening status judgment on the key component based on the key component characteristic data to obtain anti-loosening detection data of the key component; A spatiotemporal feature matrix is ​​constructed based on the anti-loosening detection data and historical status data, and the shedding risk coefficient of the key components is calculated through a risk assessment model; Based on the falling risk coefficient, an anti-falling warning is performed, and an anti-falling warning report including positioning information, abnormality type and maintenance suggestions is generated and sent to the management personnel.

2. A method for preventing high-speed railway contact network components from falling off based on 4C detection according to claim 1, characterized in that: The multi-source detection data of the high-speed railway contact network components is obtained, and the multi-source detection data includes original image data and historical status data collected by the 4C detection device, including: Collecting original image data of high-speed rail contact network components using a 4C detection device, wherein the original image data includes image data acquired at different angles and under different lighting conditions; The historical status data of the high-speed railway contact network components are read from the historical status database, wherein the historical status data includes historical anti-loosening detection data and related maintenance records.

3. The method for preventing components from falling off of a high-speed railway contact network based on 4C detection according to claim 1, characterized in that: The target detection network model is used to identify raw image data and obtain key component data of high-speed rail contact network components, including: An object detection network model was built based on the YOLOv5 network. The model generated candidate regions and an adaptive feature pyramid was used to extract multi-scale features to obtain the coarse-grained locations of the bolt-nut combination, the lock washer, and the cotter pin. A dynamic confidence threshold is used to screen the candidate regions, and the overlapping candidate regions are fused and optimized in combination with the target size distribution characteristics to output the precise bounding box of the key components.

4. The method for preventing high-speed railway contact network components from falling off based on 4C detection according to claim 1, characterized in that: The feature extraction of the key component data by the fine-grained detection network model to obtain the key component feature data includes: The fine-grained detection network model includes a multi-branch feature extraction network and a multi-scale key point regression network; Performing fine-grained feature extraction on the key components through a multi-branch feature extraction network, wherein the multi-branch feature extraction network includes a morphological feature branch, a geometric relationship feature branch, and a texture feature branch; The characteristic points of the key components are accurately located based on a multi-scale key point regression network. The characteristic points of the key components include the endpoints of the anti-loosening mark line, the intersection of the long and short ends of the stop gasket, and the root and bending points of the cotter pin.

5. The method for preventing high-speed railway contact network components from falling off based on 4C detection according to claim 1, characterized in that: The step of judging the anti-loosening state of the key component based on the key component characteristic data to obtain the anti-loosening detection data of the key component includes: Calculate the corresponding anti-loosening state parameters based on the characteristic data of different key components. The anti-loosening state parameters include the linear offset of the anti-loosening mark line, the spatial angle between the stop washer and the wire clamp body, and the angle between the two curved edges of the cotter pin. Based on the preset standard values ​​of the anti-loosening state parameters, the calculated anti-loosening state parameters are judged by threshold value, the anti-loosening state level of each key component is determined, and standardized anti-loosening detection data is generated.

6. The method for preventing components from falling off of a high-speed railway contact network based on 4C detection according to claim 1, characterized in that: The method of constructing a spatiotemporal feature matrix based on the anti-loosening detection data and the historical status data and calculating the shedding risk coefficient of the key components through the risk assessment model includes: Construct a spatiotemporal feature matrix, align the anti-loosening detection data with the historical status data in the time and space dimensions, and obtain a multi-dimensional feature matrix that reflects the time-varying and spatial distribution of the loosening status of key components; Through the deep integration risk assessment model, the spatiotemporal feature matrix is ​​identified to obtain the shedding risk coefficient of each key component; The deep integrated risk assessment model adopts a three-layer architecture, where: The first layer establishes a component looseness foundation risk assessment model based on the gradient boosting decision tree to generate an initial risk coefficient; The second layer combines the long short-term memory network to capture the trend characteristics of the time series and perform time dimension correction on the initial risk coefficient; The third layer uses a graph convolutional network to model the spatial topological relationship between components and performs spatial dimension risk correction on the initial risk coefficient.

7. A method for preventing high-speed railway contact network components from falling off based on 4C detection as claimed in claim 6, characterized in that: The anti-fall warning is performed based on the fall risk coefficient, and an anti-fall warning report including positioning information, abnormality type and maintenance suggestions is generated and sent to the management personnel, including: Based on the shedding risk factor, combined with component importance and line operation level, a multi-level early warning mechanism is established to generate an anti-shedding early warning report containing precise positioning information, abnormality type analysis, and maintenance priority. The multi-level early warning mechanism includes information level, prompt level, warning level and emergency level, among which: For information level, the test results are recorded but no warning is triggered; for prompt level, a low-priority warning is generated and attention is paid to during the next routine maintenance; for warning level, a medium-priority warning is generated and special maintenance is carried out within a week; for emergency level, a high-priority warning is generated and emergency treatment is carried out within 24 hours; The early warning report is sent to relevant management personnel, and maintenance recommendations and resource allocation plans are generated based on the risk characteristics of component loosening and historical maintenance experience.

8. A high-speed railway contact network component prevention system based on 4C detection, used to execute a high-speed railway contact network component prevention method based on 4C detection according to any one of claims 1 to 7, characterized in that: The system comprises: A data acquisition module is used to acquire multi-source detection data of high-speed rail contact network components, wherein the multi-source detection data includes original image data and historical status data collected by 4C detection devices; The key component recognition module is used to identify raw image data through the target detection network model and obtain key component data of the high-speed rail contact network components, including bolt-nut combinations, stop washers, and cotter pins; A feature extraction module is used to extract features from the key component data using a fine-grained detection network model to obtain key component feature data; The anti-loosening detection module is used to judge the anti-loosening state of the key component based on the key component characteristic data, and obtain the anti-loosening detection data of the key component: A falling risk calculation module is used to construct a spatiotemporal feature matrix based on the anti-loosening detection data and the historical status data, and calculate the falling risk coefficient of the key component through a risk assessment model; The anti-fall warning module is used to perform anti-fall warning based on the fall risk coefficient, generate an anti-fall warning report including positioning information, abnormality type and maintenance suggestions, and send it to the management personnel.

9. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

Images