A method and system for detecting loosening of a wind cylinder clamp of a rail train

Abstract

The application provides a rail train air cylinder hoop loosening detection method, comprising: setting an image acquisition module on a patrol robot, extracting corresponding images in a video as to-be-analyzed images; calling a preset hoop component model, the hoop component model defining a plurality of sub-features of a hoop and spatial constraint conditions between the sub-features, extracting boundary regions of each sub-feature in the to-be-analyzed images through the hoop component model, and respectively performing reliability analysis on the boundary regions of each sub-feature and sorting according to the reliability analysis; according to a reliability analysis sorting result, preferentially identifying a sub-feature with high reliability through a hierarchical identification strategy, taking the identified sub-feature as a reference, predicting a boundary region position of a remaining sub-feature through a prediction positioning strategy based on the spatial constraint conditions; the sub-feature comprises a marker line and a connecting rod, a continuity state of the marker line on the connecting rod is analyzed, and a hoop loosening detection result is output according to the continuity state.

Description

TECHNICAL FIELD

[0001] The present application relates to a rail train detection method, in particular to a rail train air cylinder hoop loosening detection method and system. BACKGROUND

[0002] The air cylinder of the rail train is a key component for ensuring the normal operation of the train braking and pneumatic system, and its fixation stability is directly related to the operation safety of the train. The air cylinder hoop is the core structure for reliable fixation of the air cylinder. With the development of the rail transportation industry towards high density and high speed, higher precision and efficiency are required for the state detection of the air cylinder hoop. In recent years, although some train inspection technologies based on machine vision have been applied, most of the schemes can only realize rough positioning of the air cylinder and the hoop, and cannot accurately extract and analyze the state of the hoop sub-features such as the marker line and the connecting rod. Moreover, there is a lack of evaluation of the reliability of feature recognition and hierarchical positioning strategy, and in the scene of image blur and occlusion, feature misrecognition or positioning failure may easily occur, which cannot effectively judge the loosening state of the hoop. SUMMARY

[0003] In view of the deficiencies in the prior art, the purpose of the present application is to provide a rail train air cylinder hoop loosening detection method and system to overcome the above-mentioned defects in the prior art.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0005] A rail train air cylinder hoop loosening detection method, comprising:

[0006] An air cylinder identification step, an image acquisition module is arranged on the inspection robot, the video of the bottom of the train is shot by using the image acquisition module, and frame extraction detection is performed, the position of the air cylinder is identified by using a preset target detection model, and the hoop on the air cylinder is calibrated by using a preset hoop detection model, and the corresponding image in the video is extracted as an analysis image;

[0007] A model definition step, a preset hoop component model is called, the hoop component model defines a plurality of sub-features of the hoop and a spatial constraint condition between the sub-features, the boundary region of each sub-feature is extracted in the analysis image by using the hoop component model, and the boundary region of each sub-feature is analyzed in reliability and sorted according to the reliability analysis;

[0008] A hierarchical identification and positioning step, according to the sorting result of the reliability analysis, the sub-features with high reliability are preferentially identified by using a hierarchical identification strategy, the identified sub-features are used as a reference, the boundary region positions of the remaining sub-features are predicted based on the spatial constraint condition by using a prediction positioning strategy;

[0009] The state analysis step includes sub-features such as marking lines and connecting rods. The continuity state of the marking lines on the connecting rods is analyzed, and the hoop loosening detection result is output based on the continuity state.

[0010] Preferably, the constraints include the relative positional relationship, relative angle, collinearity or concentricity relationship between the sub-features, and the sub-features include at least a connecting rod, a fixing piece, a connecting block and a marking line.

[0011] Preferably, the predictive localization strategy includes obtaining the boundary region of a sub-feature with high reliability, generating the expected position of the remaining sub-features according to spatial constraints, obtaining the boundary region position of the remaining sub-features, comparing the boundary region position with the expected position to generate an expected confidence level, and preset a confidence level threshold. When the expected confidence level is lower than the confidence level threshold, the identification benchmark is readjusted or a new sub-feature with high reliability is introduced for calculation until the spatial relationship of the sub-features satisfies the constraints and the expected confidence level is higher than the confidence level threshold.

[0012] Preferably, the predictive positioning strategy further includes preset size constraints, automatically completing the connecting rod boundary region using known sub-feature regions, and deducing the simulated boundary region of the connecting rod based on the positional and size constraints between sub-features. Through a verification strategy, the connecting rod boundary region and the simulated boundary region of the connecting rod are obtained. If there is a conflict between the two, the remaining sub-feature boundary regions are repositioned, and the connecting rod boundary region that meets the constraints is corrected and obtained.

[0013] Preferably, the predictive localization strategy further includes a cross-validation sub-step. The cross-validation sub-step is used to obtain the expected confidence scores of different sub-features and set expected differences. When the expected confidence score deviation between each sub-feature is greater than the expected difference, a low-reliability sub-feature is obtained as the feature to be adjusted, and the high-reliability sub-features around the feature to be adjusted are obtained as the benchmark feature. According to the spatial constraints between the benchmark feature and the feature to be adjusted, the updated position of the feature to be adjusted is generated. The updated position and the predicted position are compared, and the expected confidence score of the sub-feature is regenerated. Cross-validation is continued until the expected confidence score deviation between each sub-feature is less than the expected difference.

[0014] Preferably, the state analysis step further includes obtaining the boundary area of ​​the connecting rod, extracting the marking lines drawn on the surface of the connecting rod within the boundary area of ​​the connecting rod using a color recognition strategy, and determining the loose state of the clamp based on the position of the marking lines. When the marking lines are stretched or there are gaps in some areas, the clamp is considered loose.

[0015] Preferably, the hierarchical identification and positioning step further includes a connecting rod positioning sub-step, which is used to obtain the reliability of the connecting rod boundary region in the model definition step, and a connecting rod reliability threshold is preset. When the reliability of the connecting rod boundary region is greater than the connecting rod reliability threshold, the process directly enters the state analysis step.

[0016] Preferably, the state analysis step also includes a historical data analysis sub-step, which compares the boundary region positions of each identified sub-feature with historical data in the database to generate deviation data, and updates the clamp component model based on the deviation data.

[0017] A system for detecting loose air cylinder clamps in rail vehicles includes:

[0018] The air cylinder identification module is an image acquisition module installed on the inspection robot. The image acquisition module captures video of the bottom of the train and performs frame extraction detection. The air cylinder position is identified through a preset target detection model, and the clamps on the air cylinder are calibrated through a preset clamp detection model. The corresponding images in the video are extracted as images to be analyzed.

[0019] The model definition module calls a preset clamp component model, which defines multiple sub-features of the clamp and the spatial constraints between the sub-features. The clamp component model extracts the boundary regions of each sub-feature in the image to be analyzed, performs reliability analysis on the boundary regions of each sub-feature, and performs reliability calculation and sorting based on the reliability analysis.

[0020] The hierarchical identification and positioning module, based on the reliability analysis ranking results, prioritizes the identification of sub-features with high reliability through a hierarchical identification strategy. Using the identified sub-features as a benchmark, it predicts the boundary region location of the remaining sub-features based on spatial constraints through a predictive positioning strategy.

[0021] The state analysis module, whose sub-features include marking lines and connecting rods, analyzes the continuity of the marking lines on the connecting rods and outputs the clamp loosening detection results based on the continuity.

[0022] The beneficial effects of this invention are as follows: This method, by pre-setting a clamp assembly model, clarifies the sub-feature marking lines, connecting rods, and spatial constraints between sub-features of the clamp, enabling accurate extraction of the boundary regions of each sub-feature and providing precise data support for subsequent state analysis. Simultaneously, by ranking the recognition results of each sub-feature through reliability analysis, prioritizing high-reliability sub-features as a benchmark, and combining spatial constraints to predict the positions of remaining sub-features, it effectively avoids feature misidentification or positioning failures caused by image blurring, occlusion, and other issues, ensuring the completeness and accuracy of feature extraction. Furthermore, by analyzing the continuity of the marking lines on the connecting rods, the clamp loosening situation is determined. The detection results are direct and objective, reducing the false negative and false positive rates, ensuring the stability of the air cylinder from a technical perspective, and significantly reducing the risk of safety accidents caused by clamp loosening. Through the combination of reliability analysis and hierarchical recognition and positioning, a detection logic of high-reliability feature-priority positioning—spatial constraint prediction and completion is constructed, which can effectively address the impact of complex working conditions such as uneven lighting at the bottom of the train, component occlusion, and surface oil contamination on image quality. For example, when the reliability of identification is low due to occlusion of a connecting rod in the image, a clearly visible marker line can be used as a reference. Combining the spatial constraints between the marker line and the connecting rod, such as their parallelism and spacing, the boundary region position of the connecting rod can be accurately predicted, ensuring the completeness of feature extraction. Simultaneously, the frame-sampling detection stage can flexibly adjust the frame-sampling frequency according to image clarity. For instance, the frame-sampling frequency can be reduced in clear areas and increased in blurry areas, balancing data processing volume while ensuring detection accuracy, avoiding wasted computing power, and enabling the detection system to maintain stable operation under complex conditions. Attached Figure Description

[0023] Figure 1 This is an overall flowchart of the present invention;

[0024] Figure 2 This is a flowchart of the hierarchical identification and positioning process of the present invention;

[0025] Figure 3 This is a flowchart illustrating the historical revision process of this invention;

[0026] Figure 4 This is a diagram of the air cylinder of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] It should be noted that when a component is said to be fixed to another component, it can be directly on the other component or it may have a component in between. When a component is said to be connected to another component, it can be directly connected to the other component or it may have a component in between. When a component is said to be set to another component, it can be directly set to the other component or it may have a component in between. The terms vertical, horizontal, left, right, and similar expressions used in this document are for illustrative purposes only.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The terminology used herein includes, and / or encompasses, any and all combinations of one or more of the associated listed items.

[0030] The embodiments of the present invention will be further described in detail below with reference to the accompanying drawings:

[0031] like Figures 1-4 As shown, the present invention provides a method for detecting loose air cylinder clamps in rail trains, comprising:

[0032] The air cylinder identification process involves an image acquisition module mounted on an inspection robot. This module captures video of the train's underside and performs frame-by-frame detection. A pre-defined target detection model identifies the air cylinder's location, and a pre-defined clamp detection model calibrates the clamps on the air cylinder. The corresponding images from the video are then extracted for analysis. The video stream acquired by the image acquisition module employs a dynamic frame-by-frame + keyframe retention strategy, rather than fixed-interval frame-by-frame extraction. When the target detection model identifies the target location, frame-by-frame processing is performed, and the surrounding images are evaluated in real time to obtain a complete and clear image of the clamp. The pre-defined target detection model is built on the YOLOv8 algorithm framework and specifically optimized for the characteristics of the train's air cylinders. When the selected frame images are input into the model, the model first performs multi-scale feature extraction. The C2f module of the Backbone network generates feature maps of different resolutions, then the FPN+PAN structure of the Neck network fuses high and low-level features. Finally, the Head network outputs the predicted bounding box and confidence score of the air cylinder. When the confidence level is greater than the threshold, the cylinder position is directly output; when the confidence level is less than the threshold, it indicates that there is slight occlusion or blurring. The cylinder position in adjacent frames is combined to predict the trajectory. If the deviation of the cylinder center coordinates predicted in 3 consecutive frames is less than 10 pixels, the cylinder position is confirmed.

[0033] The model definition steps involve calling a preset clamp component model. This model defines multiple sub-features of the clamp and the spatial constraints between them. The model extracts the boundary regions of each sub-feature from the image to be analyzed, performs reliability analysis on these boundary regions, and sorts them based on the reliability. The clamp component model is based on the actual structural design of a railway train air cylinder clamp, employing a hierarchical feature classification method to decompose the clamp into multiple sub-features. These sub-features include: the clamp body, a ring-shaped metal structure surrounding the air cylinder with a continuously closed arc edge and a grayscale value lower than the air cylinder surface; fixing plates, rectangular metal plates welded to both ends of the clamp body with a rectangular outline, perpendicularly connected to the edge of the clamp body; and connectors, including... L-shaped and U-shaped connectors are metal components that connect the fixing plate to the vehicle body bracket. They have an asymmetrical geometric structure, with one end fitting against the fixing plate and the other end having a connecting hole. Connecting blocks are cubic blocks welded to the surfaces of the two connecting parts. They have a cube / cubic-shaped outline, a smooth surface, and are aligned with the edges of the connecting parts. Connecting rods are cylindrical bolts that pass through the connecting blocks and fixing plates. They have a cylindrical body, and their axis is perpendicular to the cylinder axis. Holes on the connecting parts are circular through holes opened on the two connecting parts, fixing plates, and connecting blocks. They have a circular outline, smooth edges, and their center coincides with the axis of the connecting rod. Marking lines are high-contrast lines sprayed onto the surfaces of the connecting rods and connecting blocks, appearing as continuous straight lines. Spatial constraints are set to define the relative positions, angles, and morphological relationships of each sub-feature. These constraints include:

[0034] Relative position constraints:

[0035] The clamp body and the fixing plate are connected. The center point of the fixing plate must be located on the extension line of both ends of the clamp body, and the distance between the edge of the fixing plate near the clamp body and the end point of the clamp body is ≤2mm.

[0036] The connecting rod and the fixing plate must have their axis passing through the center of the hole in the fixing plate, and the distance between the lower surface of the connecting rod head and the upper surface of the fixing plate must be ≤1mm.

[0037] Connector and connecting block: The connecting block must be located in the middle area of ​​the two connectors, the offset of the center point of the connecting block from the axis of symmetry of the two connectors is ≤2mm, and the fit between the edge of the connecting block near the connector and the surface of the connector is ≥90%, with no obvious gap between them.

[0038] Relative angle constraints:

[0039] The clamp body and the connecting rod, the axis of the connecting rod must be perpendicular to the tangent direction of the clamp body, the angle deviation ≤5° is normal, if the deviation >10° is judged as the connecting rod is tilted;

[0040] The included angle between the fixing plate and the connecting parts on both sides must be 90°±3° (L-shaped connecting parts) or 180°±3° (U-shaped connecting parts). An angle deviation greater than 8° is considered as deformation of the connecting parts.

[0041] The marking line and the clamp body: if the marking line is straight, it must be parallel to the axis of the clamp body, with an angle deviation of ≤2°; if it is curved, it must be concentric with the arc of the clamp body, with a radius of curvature deviation of ≤5mm.

[0042] Collinearity constraints:

[0043] The holes on the connectors, the center hole on the fixing plate, the through hole on the connecting block, and the mounting holes on both sides of the connectors must be collinear;

[0044] The marking line and the connecting block: if the marking line crosses the connecting block, it must be collinear with the axis of symmetry of the connecting block;

[0045] Concentric relationship constraints:

[0046] The center of the clamp body and the air cylinder must be concentric.

[0047] The hole on the connector and the connecting rod, the cylindrical surface of the connecting rod must be concentric with the inner wall of the hole on the connector;

[0048] Morphological consistency constraints:

[0049] The connectors on both sides must be symmetrically distributed, with a geometric dimension difference of ≤3% and a corresponding hole position deviation of ≤1mm.

[0050] Based on the sub-feature definition and constraints of the clamp component model, a multi-algorithm fusion extraction strategy is adopted to accurately locate the boundary regions of each sub-feature in the image to be analyzed. For sub-features with regular geometric shapes such as clamp body, fixing piece, and connecting block, the Canny edge detection algorithm is first used to extract the image edges, and then the edge gaps are filled by morphological closing operation to form a complete contour. Holes on circular connectors are identified by Hough circle transform, straight marking lines and connecting rods are identified by Hough line transform, and rectangular fixing pieces and connecting blocks are identified by contour moment calculation. Based on edge integrity, texture matching degree, image quality adaptability, neighborhood correlation, and spatial correlation, the reliability of each sub-feature is obtained, and the sub-features are sorted from high to low according to their reliability.

[0051] Edge integrity focuses on the geometric integrity of sub-features. By determining whether the extracted edges cover the theoretical contour of the sub-features, it reflects whether there are problems such as occlusion or breakage of the sub-features. The core calculation steps are as follows: Based on the preset size and shape of the sub-features in the clamp component model, the theoretical edge contour is generated. For example, the theoretical edge of the connecting rod is a hexagonal head and a cylindrical body, which corresponds to the set of theoretical edge pixels in the image; edge detection is performed on the sub-feature regions in the image to be analyzed, the actual edge pixel set is extracted, noise points are removed, and the edge coverage is calculated, which is the difference between the number of overlapping pixels between the actual edge and the theoretical edge and the total number of theoretical edge pixels;

[0052] Texture matching is used for sub-features with unique textures, such as the hexagonal texture of the connecting rod head and the color texture of the marking lines. By comparing these with preset templates, it determines whether the texture is clear and whether there is rust or oil interference. The calculation steps are as follows: Standard texture samples of sub-features under different working conditions (normal, slight rust, light oil) are collected to construct a texture template library. For example, the connecting rod head template library contains multiple types of templates, such as clear hexagonal textures and light rust textures, with key texture features labeled for each type. Local image cropping is performed on the texture regions of the sub-features, and the LBP algorithm is used to extract texture feature histograms. Parameters such as grayscale distribution and edge gradient are statistically analyzed. The Bach distance between the actual texture histogram and the template histogram is calculated; the smaller the distance, the higher the similarity, which is then converted into texture matching score.

[0053] Image quality adaptability reflects the image quality of the region where the sub-feature is located, including whether the sharpness, contrast, and noise meet the requirements for feature extraction, avoiding misidentification due to poor image quality. The calculation steps are as follows: Crop the local region where the sub-feature is located, such as the marked line region, and evaluate the quality through three indicators: Sharpness: Calculate the edge gray-level gradient using the Laplacian operator. A gradient value >80 is sharp, and <50 is blurry; Contrast: Calculate the dynamic range of the gray-level histogram. A dynamic range >180 is high contrast, and <120 is low contrast; Noise: Calculate the noise variance using Gaussian filtering. A variance <10 is low noise, and >20 is high noise. The three indicators are scored separately, and the image fit is calculated according to the corresponding weights. The neighborhood correlation is based on the spatial constraint compliance between the sub-feature and the surrounding high-reliability features, such as the hoop body and the marker line, to determine whether the sub-feature position is reasonable and avoid misidentification of isolated features. The calculation steps are as follows: Select high-reliability features within a fixed pixel range around the sub-feature and determine the preset spatial constraint between the two; Measure the actual spatial relationship between the sub-feature and the surrounding high-reliability features, compare it with the preset constraint, and calculate the constraint compliance.

[0054] Constraints include the relative positional relationship, relative angle, collinearity or concentricity between sub-features. Sub-features include at least connecting rods, fixing plates, connecting blocks and marking lines.

[0055] The hierarchical identification and positioning step, based on the reliability analysis ranking results, prioritizes the identification of sub-features with high reliability through a hierarchical identification strategy. Using the identified sub-features as a benchmark, a predictive positioning strategy is used to predict the boundary region locations of the remaining sub-features based on spatial constraints. According to the reliability of the sub-features, priority is assigned, including high priority, medium priority, and low priority. Based on the boundary region of the high-reliability benchmark feature, the spatial constraints in the model definition step are invoked. Through the process of expected location calculation - actual location comparison - deviation correction, the boundary regions of the remaining sub-features are accurately derived.

[0056] The hierarchical identification and localization step also includes a connecting rod localization sub-step, used to obtain the reliability of the connecting rod boundary region in the model definition step. A connecting rod reliability threshold is preset. When the reliability of the connecting rod boundary region is greater than the threshold, the process directly proceeds to the state analysis step. Based on the reliability score of the connecting rod boundary region in the model definition step, a connecting rod reliability threshold is preset. When the calculated connecting rod reliability > the threshold, it indicates accurate connecting rod localization, requiring no further verification from other sub-features, and the process directly jumps to the state analysis step. If the reliability is ≤ the threshold, it needs to be incorporated into the predictive localization strategy, combining other high-reliability sub-features to further optimize the localization.

[0057] The predictive localization strategy includes: acquiring the boundary regions of highly reliable sub-features; generating the expected locations of the remaining sub-features based on spatial constraints; acquiring the boundary region locations of the remaining sub-features; comparing the boundary region locations with the expected locations to generate expected confidence scores, with a preset confidence threshold. When the expected confidence score is lower than the confidence threshold, the identification benchmark is readjusted or new highly reliable sub-features are introduced for calculation until the spatial relationships of the sub-features satisfy the constraints and the expected confidence score is higher than the confidence threshold. From the reliability ranking results, 2-3 highly reliable sub-features are selected, and their precise boundary region coordinates are obtained. Based on the spatial constraints in the model definition steps, the expected locations of the remaining sub-features are calculated. The actual extracted boundary regions of the remaining sub-features are compared with the expected locations, and the expected confidence scores are calculated using the location deviation rate. A confidence threshold is preset. If the expected confidence is greater than the threshold, the location result of the sub-feature is retained. If it is less than the threshold, the identification benchmark is readjusted, such as replacing another high-reliability sub-feature as the benchmark or introducing a new high-reliability sub-feature, such as selecting the third high-reliability feature from the ranking results, and the expected position is recalculated until the confidence reaches the standard.

[0058] The predictive localization strategy also includes preset dimensional constraints. Using known sub-feature regions, it automatically completes the connecting rod boundary region. Based on the positional and dimensional constraints between sub-features, it deduces the simulated boundary region of the connecting rod. A verification strategy is used to obtain both the connecting rod boundary region and the simulated boundary region. If a conflict exists, the remaining sub-feature boundary regions are repositioned, and a new connecting rod boundary region conforming to the constraints is obtained. Preset dimensional correlation rules exist between sub-features, such as connecting rod diameter = hole diameter on the connector - 0.5mm; connecting block side length = connecting rod diameter × 1.8. A dimensional parameter library for different vehicle models is also stored. When the initial extraction of the connecting rod is incomplete, such as only recognizing the head and not the main body, it uses known sub-features, such as the dimensions of holes on the connector, to reverse-engineer the boundary region. Obtain the initially extracted boundary region of the connecting rod and the completed simulated boundary region. Determine if there is a conflict by calculating the overlap. If the overlap is greater than 80%, the boundary region of the connecting rod is confirmed to be valid. If the overlap is less than 50%, reposition the remaining sub-features such as holes and connecting blocks on the connector, correct the center coordinates and diameter measurements of the holes, and then re-determine the simulated boundary region of the connecting rod based on the corrected dimensional constraints until the overlap is greater than 80%, ensuring that the positioning of the connecting rod conforms to the dimensional rules.

[0059] The predictive localization strategy also includes a cross-validation sub-step. This sub-step is used to obtain the expected confidence scores of different sub-features and sets expected differences. When the expected confidence score deviation between sub-features exceeds the expected difference, a low-reliability sub-feature is selected as the feature to be adjusted, and high-reliability sub-features surrounding the feature to be adjusted are selected as the baseline features. Based on the spatial constraints between the baseline features and the feature to be adjusted, an updated position of the feature to be adjusted is generated. The updated position is compared with the predicted position, and the expected confidence score of the sub-feature is regenerated. Cross-validation continues until the expected confidence score deviation between all sub-features is less than the expected difference. For all located sub-features, high, medium, and low reliability are calculated based on different baseline features. For example, a connecting rod simultaneously calculates two confidence scores based on the marking line and the clamp body, forming a confidence score matrix. A preset expected difference is used to compare the confidence score deviation of the same sub-feature under different baselines, as well as the confidence score deviation between different sub-features. If all deviations are less than the expected difference, it indicates that the localization logic of each sub-feature is consistent; if any deviation exceeds the expected difference, a correction process is triggered. Sub-features with excessive confidence deviations are marked as features to be adjusted. From the high-reliability sub-features, select 2-3 features most closely related to the features to be adjusted as baseline features. Based on the spatial constraints between the baseline features and the features to be adjusted, such as the perpendicular relationship between the marker line and the connecting rod, and the concentric relationship between the hole and the connecting rod, generate updated positions for the features to be adjusted. Compare the updated positions with the original predicted positions and recalculate the expected confidence. Repeat the above steps, continuously adjusting and recalculating the confidence of sub-features with deviations, until the expected confidence deviations between all sub-features are less than the expected difference, ensuring that the sub-feature positioning of the entire clamp assembly forms a logically self-consistent system.

[0060] The state analysis step includes sub-features such as the marking line and the connecting rod. It analyzes the continuity of the marking line on the connecting rod and outputs the clamp loosening detection result based on this continuity. The marking line and connecting rod are direct characteristics reflecting clamp loosening—under normal conditions, the marking line is continuous and regular on the surface of the connecting rod; when the clamp is loose, displacement or angular shift of the connecting rod will cause the marking line to stretch, break, or become gapped.

[0061] The state analysis step also includes obtaining the boundary area of ​​the connecting rod, extracting the marking lines drawn on the surface of the connecting rod through a color recognition strategy within the boundary area of ​​the connecting rod, and judging the looseness of the clamp based on the position of the marking lines. When the marking lines are stretched or there are gaps in some areas, it is considered that the clamp is loose. The coordinates of the connecting rod boundary area output from the hierarchical recognition and positioning steps are called, and the HSV color space is used for specific extraction of the marking lines. For the red and yellow marking lines commonly used in railcar clamps, a specific color threshold is set. By filtering through this threshold, the marking line pixel area can be accurately separated from the metallic color of the connecting rod surface. Based on the extracted marking line pixel coordinates, the continuity is judged from the dimension of morphological integrity. The specific judgment criteria are as follows: the continuous length ratio of the marking line is calculated. The continuous length ratio is the ratio of the actual continuous length of the marking line to the theoretical total length. If the continuous length ratio is ≥95% and there is no obvious break, it is in a normal state; if the continuous length ratio is 80%-94%, or there are 1-2 short breaks, and there is no tensile deformation, it is slightly loose; if the continuous length ratio is <80%, or there are long breaks, or there is obvious tensile deformation, it is seriously loose.

[0062] The status analysis step also includes a historical data analysis sub-step. This sub-step compares the boundary region locations of each identified sub-feature with historical data in the database to generate deviation data, and updates the clamp component model based on this deviation data. Historical inspection data for the same train and the same air cylinder clamp is stored. Each data entry includes: inspection time, coordinates of the boundary regions of each sub-feature, marking line, connecting rod, and clamp body, continuity parameters of the marking line, percentage of continuous length, axis deviation, and the inspection result at that time. The database establishes a three-level index based on train number, air cylinder number, and clamp number to ensure accurate data retrieval. After the current inspection is completed, the system automatically retrieves the last three historical data. If this is the first inspection of a clamp, it retrieves the standard reference data for clamps of the same model and location. The system compares the key parameters of the current data with those of the historical data to generate a deviation data matrix. It focuses on analyzing the following two types of deviations: sub-feature position deviation: calculating the difference between the center coordinates of the boundary areas of the connecting rod and the marking line and the historical data; marking line continuity deviation: calculating the difference between the current continuous length ratio and the axis deviation degree and the historical data. Based on the analysis results of the deviation data, the spatial constraint reliability evaluation parameters of the clamp component model are dynamically updated to optimize the subsequent inspection accuracy: Spatial constraint update: If the historical data of a clamp for a certain model shows that the concentricity deviation between the connecting rod and the hole is generally within 1 pixel, the concentricity constraint threshold for that model is lowered from 2 pixels to 1.5 pixels to improve positioning accuracy. If it is found that the vertical deviation between the marking line and the connecting rod of a certain batch of clamps is generally large, the vertical constraint threshold for that batch is raised to 8° to avoid misjudgment.

[0063] Reliability assessment parameter update: If historical data shows that the texture matching degree of the connecting rod generally decreases by 5-8 points in winter, then during winter detection, the image quality adaptability weight will be increased from 15% to 20% to reduce the impact of texture matching degree fluctuations on reliability scores and ensure accurate selection of the benchmark for layered recognition and positioning.

[0064] A system for detecting loose air cylinder clamps in rail vehicles includes:

[0065] The air cylinder identification module is an image acquisition module installed on the inspection robot. The image acquisition module captures video of the bottom of the train and performs frame extraction detection. The air cylinder position is identified through a preset target detection model, and the clamps on the air cylinder are calibrated through a preset clamp detection model. The corresponding images in the video are extracted as images to be analyzed.

[0066] The model definition module calls the preset clamp component model. The clamp component model defines multiple sub-features of the clamp and the spatial constraints between the sub-features. The clamp component model extracts the boundary regions of each sub-feature in the image to be analyzed, performs reliability analysis on the boundary regions of each sub-feature, and performs reliability calculation and sorting based on the reliability analysis.

[0067] The hierarchical identification and positioning module, based on the reliability analysis ranking results, prioritizes the identification of sub-features with high reliability through a hierarchical identification strategy. Using the identified sub-features as a benchmark, it predicts the boundary region location of the remaining sub-features based on spatial constraints through a predictive positioning strategy.

[0068] The state analysis module includes sub-features such as marking lines and connecting rods. It analyzes the continuity of the marking lines on the connecting rods and outputs the clamp loosening detection results based on the continuity.

[0069] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting loose air cylinder clamps in railcars, characterized in that, include: The air cylinder identification process involves setting up an image acquisition module on the inspection robot, using the image acquisition module to capture video of the bottom of the train, performing frame extraction detection, identifying the position of the air cylinder through a preset target detection model, calibrating the clamps on the air cylinder through a preset clamp detection model, and extracting the corresponding images from the video as images to be analyzed. The model definition step involves calling a preset clamp component model, which defines multiple sub-features of the clamp and the spatial constraints between the sub-features. The clamp component model is used to extract the boundary regions of each sub-feature in the image to be analyzed, and the reliability analysis of the boundary regions of each sub-feature is performed and sorted according to the reliability analysis. The hierarchical identification and positioning steps involve prioritizing the identification of highly reliable sub-features based on the reliability analysis ranking results, using a hierarchical identification strategy, and then using the identified sub-features as a benchmark to predict the boundary region locations of the remaining sub-features through a predictive positioning strategy. The predictive localization strategy includes obtaining the boundary regions of sub-features with high reliability, generating the expected positions of the remaining sub-features based on spatial constraints, obtaining the boundary region positions of the remaining sub-features, comparing the boundary region positions with the expected positions, generating the expected confidence level, and setting a confidence level threshold. When the expected confidence level is lower than the confidence level threshold, the identification benchmark is readjusted or a new high-reliability sub-feature is introduced for calculation until the spatial relationship of the sub-features satisfies the constraints and the expected confidence level is higher than the confidence level threshold. The state analysis step includes sub-features such as marking lines and connecting rods. The continuity state of the marking lines on the connecting rods is analyzed, and the hoop loosening detection result is output based on the continuity state.

2. The method for detecting loose air cylinder clamps in a railcar according to claim 1, characterized in that, The constraints include the relative positional relationship, relative angle, collinearity or concentricity relationship between sub-features, and the sub-features include at least connecting rods, fixing plates, connecting blocks and marking lines.

3. The method for detecting loose air cylinder clamps in a railcar according to claim 1, characterized in that, The predictive positioning strategy also includes preset size constraints. It automatically completes the boundary region of the connecting rod using known sub-feature regions, and deduces the simulated boundary region of the connecting rod based on the positional and size constraints between sub-features. It obtains the boundary region of the connecting rod and the simulated boundary region of the connecting rod through a verification strategy. If there is a conflict between the two, the remaining sub-feature boundary regions are repositioned, and the obtained connecting rod boundary region that meets the constraints is corrected.

4. The method for detecting loose air cylinder clamps in a railcar according to claim 1, characterized in that, The predictive localization strategy also includes a cross-validation sub-step. This sub-step is used to obtain the expected confidence scores of different sub-features and set expected differences. When the expected confidence score deviation between each sub-feature is greater than the expected difference, a low-reliability sub-feature is obtained as the feature to be adjusted, and the high-reliability sub-features around the feature to be adjusted are obtained as the benchmark feature. Based on the spatial constraints between the benchmark feature and the feature to be adjusted, the updated position of the feature to be adjusted is generated. The updated position and the predicted position are compared, and the expected confidence score of the sub-feature is regenerated. Cross-validation continues until the expected confidence score deviation between each sub-feature is less than the expected difference.

5. The method for detecting loose air cylinder clamps in a railcar according to claim 1, characterized in that, The state analysis step also includes obtaining the boundary area of ​​the connecting rod, extracting the marking lines drawn on the surface of the connecting rod through a color recognition strategy within the boundary area of ​​the connecting rod, and judging the loose state of the clamp based on the position of the marking lines. When the marking lines are stretched or there are gaps in some areas, it is determined that the clamp is loose.

6. The method for detecting loose air cylinder clamps in a railcar according to claim 1, characterized in that, The hierarchical identification and positioning step also includes a connecting rod positioning sub-step, which is used to obtain the reliability of the connecting rod boundary region in the model definition step, and a connecting rod reliability threshold is preset. When the reliability of the connecting rod boundary region is greater than the connecting rod reliability threshold, the process directly enters the state analysis step.

7. The method for detecting loose air cylinder clamps in a railcar according to claim 1, characterized in that, The state analysis step also includes a historical data analysis sub-step, which compares the boundary region locations of each identified sub-feature with historical data in the database to generate deviation data, and updates the clamp component model based on the deviation data.

8. A system for detecting loose air cylinder clamps in railcars, characterized in that, include: The air cylinder identification module is an image acquisition module installed on the inspection robot. The image acquisition module captures video of the bottom of the train and performs frame extraction detection. The air cylinder position is identified through a preset target detection model, and the clamps on the air cylinder are calibrated through a preset clamp detection model. The corresponding images in the video are extracted as images to be analyzed. The model definition module calls a preset clamp component model, which defines multiple sub-features of the clamp and the spatial constraints between the sub-features. The clamp component model extracts the boundary regions of each sub-feature in the image to be analyzed, performs reliability analysis on the boundary regions of each sub-feature, and performs reliability calculation and sorting based on the reliability analysis. The hierarchical identification and positioning module, based on the reliability analysis ranking results, prioritizes the identification of sub-features with high reliability through a hierarchical identification strategy. Using the identified sub-features as a benchmark, it predicts the boundary region location of the remaining sub-features through a predictive positioning strategy. The predictive localization strategy includes obtaining the boundary regions of sub-features with high reliability, generating the expected positions of the remaining sub-features based on spatial constraints, obtaining the boundary region positions of the remaining sub-features, comparing the boundary region positions with the expected positions, generating the expected confidence level, and setting a confidence level threshold. When the expected confidence level is lower than the confidence level threshold, the identification benchmark is readjusted or a new high-reliability sub-feature is introduced for calculation until the spatial relationship of the sub-features satisfies the constraints and the expected confidence level is higher than the confidence level threshold. The state analysis module, whose sub-features include marking lines and connecting rods, analyzes the continuity of the marking lines on the connecting rods and outputs the clamp loosening detection results based on the continuity.

Images