A method and system for detecting abnormalities in the anti-loosening wires of rail trains

Abstract

This invention discloses a method and system for detecting anomalies in the anti-loosening lines of rail trains, relating to the field of rail transit inspection technology. The method includes: image acquisition and processing, capturing and preprocessing target images of a fixed-point area containing the anti-loosening lines; line integrity judgment, extracting bolts, bolt marking lines, and anti-loosening lines to form a target area map through an analysis model, and judging the integrity of the anti-loosening lines based on line segment continuity and shape; line deformation depth analysis, simultaneously acquiring the intersection position and deformation data of the anti-loosening lines by combining the model with the loosening angle of the bolt marking lines when the lines are intact; line deformation safety judgment, setting a safety threshold group based on historical data, and outputting signals by comparing deformation data; and line deformation anomaly judgment, after receiving the comparison signal, comparing the actual data with the theoretical data output by the mechanical model, and outputting an abnormal or normal signal. This invention achieves high-precision detection of anti-loosening lines across all dimensions, improving automation and real-time performance, and ensuring train operation safety.

Description

Technical Field

[0001] This invention relates to the field of rail transit inspection technology, and more specifically to a method and system for detecting abnormalities in the anti-loosening lines of rail trains. Background Technology

[0002] In the design and manufacturing of railcars, ensuring the reliability of the connections of key components is a crucial aspect. These connections are subjected to complex dynamic loads, high-frequency vibrations, and environmental erosion over long periods of time, which places extremely stringent requirements on the anti-loosening performance of the connectors.

[0003] For a long time, to address the problem of bolt loosening under the high-intensity operating environment of rail trains, the industry has developed and widely applied various anti-loosening technologies. Among them, the figure-eight anti-loosening line, as a mature and effective mechanical anti-loosening method, occupies an important position in the rail transit field due to its unique structural principle. This technology typically involves setting two sets of bolts and installing one or more metal wires, namely the figure-eight anti-loosening line, at a specific angle and tension. This creates a mutually restraining "figure-eight" or "cross" shape between the two bolts. When either bolt attempts to loosen (i.e., rotate) due to vibration or other external forces, the tensile or compressive force generated by the figure-eight anti-loosening line can effectively limit further rotation of the bolt, thereby locking it in its initial installation position. This significantly improves the anti-loosening ability of the bolt connection and effectively ensures the safety of train operation. With its simple structure, ease of installation and maintenance, the figure-eight anti-loosening line has provided a solid guarantee for the safety of critical connections in rail trains over the past few decades.

[0004] However, existing technologies for identifying anomalies in figure-eight anti-loosening wires often rely on periodic, manual visual inspections. This traditional method has many insurmountable drawbacks when dealing with the large and complex structure of train chassis. First, manual inspection is inefficient and cannot cover all critical connection points for high-frequency and real-time monitoring. Second, the inspection results are highly dependent on the experience and subjective judgment of the inspectors and are easily affected by environmental factors such as lighting, viewing angle, and dirt, making it difficult to guarantee the accuracy and consistency of the inspection. In addition, the failure of figure-eight anti-loosening wires is not always a sudden and significant breakage. In many cases, the evolution of the abnormal state is a gradual process. For example, a slight rotation of the bolt may cause slight deformation of the anti-loosening wire, such as thinning, local bending, or decreased tension, or the wire may develop initial cracks that are difficult to detect with the naked eye. These non-obvious and subtle signs of abnormality are easily overlooked in traditional manual or simple automated visual inspections. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of this invention is to overcome the core contradiction of the existing technology in terms of insufficient ability to monitor the status of the anti-loosening line of the rail train and diagnose early anomalies. In particular, it addresses the technical problem of being unable to efficiently, accurately and in real time identify non-obvious abnormal precursors such as minor deformation of the anti-loosening line, initial cracks and improper installation caused by minor bolt loosening.

[0006] Therefore, this invention provides a method and system for detecting anomalies in the figure-eight anti-loosening wires of rail trains. By integrating image acquisition and preprocessing, precise region segmentation, anti-loosening wire integrity judgment, and anti-loosening wire deformation analysis, it achieves comprehensive, high-precision, and real-time detection and diagnosis of macroscopic anomalies such as breakage and loosening of the figure-eight anti-loosening wires, as well as subtle anomalies such as micro-deformation and early signs, thereby significantly improving the operational safety and maintenance efficiency of rail trains.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for detecting abnormalities in the anti-loosening wires of a railcar includes the following steps:

[0009] The image acquisition and processing steps involve visually capturing images of a fixed area containing the anti-loosening line and preprocessing them to obtain the target image.

[0010] The line integrity judgment step involves extracting a target area map from the target image using a preset analysis model, which includes bolts, bolt marking lines, and anti-loosening lines. In the target area map, the anti-loosening lines are judged to be in an intact state based on the continuity of line segments and the shape of the anti-loosening lines.

[0011] The linear deformation depth analysis step involves inputting the target area map into the analysis model when the anti-loosening line is in a complete state. The analysis model analyzes the position of the bolt marking line to obtain the loosening angle of the bolt, and analyzes the intersection position of the anti-loosening line and the deformation data of the anti-loosening line.

[0012] The linear deformation safety judgment step involves defining straightness deviation thresholds and diameter thresholds based on historical data and combining them to obtain a safety threshold group. The anti-loosening linear deformation data is compared with the safety threshold group, and an abnormal signal or comparison signal is output based on the comparison result.

[0013] The abnormal deformation determination step involves, upon receiving a comparison signal, comparing the actual intersection position and actual deformation data of the anti-loosening line in the target area map with the theoretical intersection position and theoretical deformation data of the anti-loosening line output by the preset mechanical model based on the loosening angle, and outputting an abnormal or normal signal based on the comparison result.

[0014] Furthermore, the line integrity determination step includes:

[0015] The continuity judgment sub-step traverses the pixel sequence of the straight line segment of the anti-loosening line and calculates the number of times the gray value of adjacent pixels changes abruptly and exceeds the color range of the anti-loosening line. If the number is zero and the pixel length of each straight line segment is not less than the preset minimum length threshold, then the continuity and length compliance conditions are triggered.

[0016] The geometric shape judgment sub-step determines whether the two straight line segments of the anti-loosening line intersect and whether the intersection point is located between the two sets of bolts. If they intersect and the intersection point is located between the two sets of bolts, the shape compliance condition is triggered.

[0017] The connection reliability judgment sub-step determines whether the two endpoints of the anti-loosening line straight segment overlap with the boundary frames of the two sets of bolts. If so, the reliability compliance condition is triggered.

[0018] Furthermore, the analysis model includes a convolutional neural network architecture with a YOLOv8 module at the front end. The YOLOv8 module is used to identify and extract bolts, bolt marking lines, and anti-loosening lines. The back end includes a feature extraction module, which includes multiple convolutional layers, pooling layers, and fully connected layers. The feature extraction module calculates the angle between the bolt marking line and the horizontal axis by using image gradient analysis and least squares fitting based on the pixel distribution of the bolt marking line. This yields the reference angle of the bolt during initial installation and the real-time installation angle of the bolt. The loosening angle of the bolt is then calculated based on the reference angle and the real-time installation angle.

[0019] Furthermore, the analysis model selects n uniformly distributed sampling points on the two straight segments of the loosening line and outputs the pixel coordinates of each sampling point. It calculates the root mean square value of the distance from each sampling point to the loosening line segment fitted by the least squares method as the straightness deviation. Then, it selects n measurement points on the two straight segments of the loosening line and outputs the pixel diameter of the loosening line at each measurement point through an image processing algorithm. The straightness deviation and the pixel diameter are combined to obtain the deformation data of the loosening line.

[0020] Furthermore, the safety threshold group definition strategy for the linear deformation safety judgment step includes calculating the first mean and first standard deviation of the sampling points corresponding to the non-abnormal anti-loosening wires based on historical data, setting a straightness deviation threshold based on the first mean and first standard deviation, calculating the diameter of the measurement points corresponding to the non-abnormal anti-loosening wires based on historical data, calculating the second mean and second standard deviation, and setting a diameter threshold based on the second mean and second standard deviation.

[0021] Furthermore, the linear deformation anomaly determination step includes a theoretical data calculation strategy. This strategy includes calculating the effective elongation of the anti-loosening line at the bolt connection point caused by bolt rotation based on the bolt loosening angle output by the analysis model through geometric relationships, then calculating the resulting tensile force change based on the effective elongation, and calculating the theoretical position coordinates of the anti-loosening line intersection point based on the geometric configuration and tensile force change of the anti-loosening line, and calculating the theoretical diameter of the anti-loosening line based on Poisson's ratio.

[0022] Furthermore, the linear deformation anomaly determination step also includes an actual data calculation strategy, which includes extracting the actual intersection coordinates of the two straight line segments of the anti-loosening line in the target area map, and recalculating the actual straightness deviation and actual diameter in the target area map based on the sampling point position and measurement point position selected by the analysis model.

[0023] Furthermore, the process includes a wire tightening connection analysis step. This involves locating the tightening connection, extracting the edge contour of the tightening connection structure using an edge detection algorithm, calculating the contour area, contour roundness, and contour center offset using image morphology analysis, comparing the contour area, contour roundness, and contour center offset with the corresponding parameter ranges in a preset standard morphological parameter database, and outputting the morphological analysis results. Then, with the geometric center of the tightening connection as the origin, rays are emitted along 360° evenly divided into n scanning directions. The total number of intersection events between the rays and the edge of the winding structure is counted. The total number of intersection events is divided by 2 and rounded to obtain the actual number of winding turns. The actual number of winding turns is compared with a preset standard number of turns range, and the number of turns analysis results are output. A comprehensive judgment is made based on the morphological analysis results and the number of turns analysis results. If both are normal, the tightening connection status of the anti-loosening wire intersection is determined to be normal; otherwise, a signal indicating a loose anti-loosening wire is output.

[0024] Furthermore, the analysis step of the wire tightening connection includes a standard morphological parameter database construction strategy. Based on the contour area, contour roundness, and contour center offset data of the tightening connection under various normal conditions in historical data, the standard range of contour area, the standard range of contour roundness, and the standard range of contour center offset are determined by clustering algorithm to form a standard morphological parameter database.

[0025] An abnormality detection system for the figure-eight anti-loosening line of a rail train includes:

[0026] The image acquisition and processing module captures images of a fixed area containing the anti-loosening line through vision and performs preprocessing to obtain the target image;

[0027] The line integrity judgment module extracts a target area map from the target image by using a preset analysis model, which includes bolts, bolt marking lines and anti-loosening lines. In the target area map, the anti-loosening lines are judged to be in an intact state by the continuity of line segments and the shape of the anti-loosening lines.

[0028] The linear deformation depth analysis module, when the anti-loosening line is in a complete state, inputs the target area map into the analysis model. The analysis model analyzes the position of the bolt marking line to obtain the loosening angle of the bolt, and analyzes the intersection position of the anti-loosening line and the deformation data of the anti-loosening line.

[0029] The linear deformation safety judgment module defines straightness deviation threshold and diameter threshold based on historical data and combines them to obtain a safety threshold group. It compares the anti-loosening linear deformation data with the safety threshold group and outputs an abnormal signal or comparison signal based on the comparison result.

[0030] The linear deformation anomaly determination module, when receiving a comparison signal, compares the actual intersection position and actual deformation data of the anti-loosening line in the target area map with the theoretical intersection position and theoretical deformation data of the anti-loosening line output by the preset mechanical model based on the loosening angle, and outputs an abnormal signal or a normal signal based on the comparison result.

[0031] The beneficial effects of this invention are as follows: 1. By image acquisition and preprocessing, line integrity judgment, line deformation depth analysis, line deformation safety judgment, and line deformation anomaly judgment, combined with the analysis model, pixel-level features of bolts, bolt marking lines, and anti-loosening lines can be accurately extracted. This not only enables the rapid identification of obvious anomalies such as broken anti-loosening lines and failure to be effectively connected to bolts, but also captures latent precursors such as minute deformations, initial cracks, and abnormal prestresses of anti-loosening lines caused by minor bolt loosening through bolt loosening angle calculation and anti-loosening line straightness deviation and diameter change analysis. This achieves full-dimensional anomaly detection from macro to micro, significantly improving the comprehensiveness and accuracy of monitoring the status of key connection parts of rail trains.

[0032] 2. Based on historical normal working condition data, safety threshold groups such as straightness deviation threshold and diameter threshold are set by statistical analysis of mean and standard deviation. Combined with clustering algorithm, a database of standard morphological parameters of tightened connections is constructed to avoid the subjectivity of threshold setting. In addition, through the dual verification logic of comparing actual data with theoretical data, a mechanical model is constructed based on the bolt loosening angle to calculate the theoretical intersection position and theoretical deformation data of the anti-loosening line. This data is then compared with the actual collected deformation data to eliminate the risk of misjudgment from a single data dimension. At the same time, the morphological analysis of the tightened connection and the verification of the number of winding turns complement each other, further improving the reliability of anomaly judgment and effectively reducing missed and false judgments. Attached Figure Description

[0033] Figure 1 This is the overall flowchart of the present invention;

[0034] Figure 2 This is the visual fixed-point acquisition image in this invention;

[0035] Figure 3 This is a flowchart of the line integrity judgment process in this invention;

[0036] Figure 4 This is a flowchart of the linear deformation anomaly determination process in this invention. Detailed Implementation

[0037] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are denoted by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0038] Current technologies for identifying anomalies in figure-eight anti-loosening braces often rely on periodic, manual visual inspections. This traditional method has several insurmountable drawbacks when dealing with the large and complex structures of train chassis. First, manual inspections are inefficient, failing to cover all critical connection points for high-frequency and real-time monitoring. Second, inspection results are highly dependent on the experience and subjective judgment of the inspectors, easily affected by environmental factors such as lighting, viewing angle, and dirt, making it difficult to guarantee accuracy and consistency. Furthermore, the failure of figure-eight anti-loosening braces is not always a sudden and significant break; in many cases, the evolution of the abnormal state is a gradual process. During the process of installation, for example, a slight rotation of the bolt may cause minor deformation of the anti-loosening wire, such as thinning, local bending, or decreased tension, or the wire may develop initial cracks that are difficult to detect with the naked eye. These non-obvious and subtle abnormal precursors are easily overlooked in traditional manual or simple automated visual inspection. Therefore, this invention designs an abnormal detection method and system for the figure-eight anti-loosening wire of a rail train, which can efficiently identify subtle deformation, initial cracks, and non-obvious abnormal precursors of the anti-loosening wire caused by slight bolt loosening, which are difficult to detect by traditional inspection methods. This significantly improves the operational safety and maintenance efficiency of rail trains.

[0039] like Figure 1 As shown, the process includes image acquisition and processing steps, line integrity judgment steps, line deformation depth analysis steps, line deformation safety judgment steps, and line deformation anomaly judgment steps. Specifically:

[0040] The image acquisition and processing steps involve acquiring images of a fixed area using existing train under-body vision cameras or vision cameras mounted on inspection robots, such as... Figure 2As shown, the area includes bolts, anti-loosening lines, and the background. To ensure uniform and stable illumination, the vision camera integrates an LED array supplementary light, which can automatically adjust the light intensity in low-light environments. Secondly, the acquired raw image undergoes multi-level preprocessing to eliminate image noise, enhance features, and compensate for illumination. These multi-level preprocessing methods are existing image processing techniques. The initial noise reduction uses a two-dimensional Gaussian filtering algorithm to effectively remove Gaussian and salt-and-pepper noise while preserving image edge information to the greatest extent. Further, image enhancement processing is performed. An adaptive histogram equalization algorithm divides the image into 8×8 sub-regions, and histograms are calculated for each sub-region. The image is equalized, and noise is prevented from being excessively amplified by limiting contrast gain, thereby adjusting the grayscale distribution of the image and significantly improving the image contrast. This makes the grayscale differences between the bolt surface, the red marking line of the bolt, the figure-eight anti-loosening line and the background more obvious. In addition, if the image has low contrast or blurred details due to environmental factors, the multi-scale Retinex algorithm is used for illumination compensation and detail restoration. This algorithm effectively removes the haze effect in the image and compensates for local uneven illumination by performing logarithmic transformation, Gaussian filtering and color restoration on the image at multiple scales. It restores the inherent reflective characteristics of the image and ensures the continuity of the anti-loosening line, the intersection features and the clear visibility of the red marking line of the bolt.

[0041] like Figure 3 As shown, the line integrity judgment steps include precise segmentation and extraction of the bolt and figure-eight anti-loosening line areas. The purpose is to accurately segment the bolt area, the red marker line area, and the figure-eight anti-loosening line area from the preprocessed image, eliminating complex background interference and providing accurate target area masks and pixel-level features for subsequent anti-loosening line integrity judgment and fine anomaly analysis. First, bolt area segmentation is performed based on the shape and color characteristics of the bolts. The RGB image is converted to the HSV color space. According to the surface metal material commonly used for railway bolts, such as high-strength alloy steel, its surface is silver-gray or black. For special colors, such as those resulting from anti-corrosion treatment, threshold ranges for H (hue), S (saturation), and V (brightness) are set in the HSV color space. Binarization is then performed to initially screen out possible foreground bolt areas. The H, S, and V threshold ranges are obtained through statistical analysis of numerous normal bolt images under different lighting and environmental conditions. For example, for silver-gray bolts, the H range can be set to [0,180], the S range to [0,30], and the V range to [100,255]; for black bolts, the H range can be set to [0,180], the S range to [0,255], and the V range to [0,50].

[0042] Secondly, an edge detection algorithm is applied to the initially screened foreground region. The high threshold is automatically determined using the Otsu method, and the low threshold is set to half of the high threshold to extract the precise edges of the region. Considering that railway bolts usually have circular or hexagonal contours, Hough circle transform is used for positioning of circular bolts. The minimum detection radius is set to the pixel value corresponding to the actual bolt size and camera calibration parameters, and the maximum detection radius is set to n times the actual bolt size to ensure that only complete and obvious circular bolts are detected. For hexagonal bolts, a contour approximation algorithm based on random sampling consistency is used. By finding the smallest bounding polygon and evaluating its vertex count and included angle, the hexagonal features are accurately matched. Through the above methods, the center position, radius or side length of the two sets of bolts are accurately located, their precise pixel regions are determined, and bolt region mask images are generated to ensure that the bolt regions are complete and free of background redundancy.

[0043] Next, the figure-eight anti-loosening wire region is segmented. This segmentation is based on the color and thin, elongated characteristics of the anti-loosening wire. First, for the black metal wire typically used in figure-eight anti-loosening wires, precise H, S, and V thresholds are set in the HSV color space to filter out black areas in the image, initially obtaining candidate regions for the anti-loosening wires. Second, considering that the anti-loosening wires are thin and elongated and may contain a few breaks or noise, a series of morphological operations are used for optimization: first, an erosion operation is performed to remove small noise blocks and thin connections in the candidate regions; then, a dilation operation is performed to restore the continuity of the anti-loosening wire lines and compensate for the thinning of the lines caused by erosion. Finally, straight line segments in the optimized candidate regions are extracted through Hough line transform. Based on the "figure-eight" shape of the figure-eight anti-loosening wire, it consists of two intersecting straight line segments, and the two straight line segments connect two sets of bolts respectively. Straight line segments that meet the conditions are selected to determine the pixel region of the figure-eight anti-loosening wire and generate a mask image of the anti-loosening wire region.

[0044] The integrity of the segmented anti-loosening line area is verified. This verification mechanism checks whether the segmented anti-loosening line area contains two continuous straight line segments with the required length, and whether these two straight line segments intersect and are connected to the areas of the two pre-positioned bolts. The specific verification method includes: calculating the pixel length of each straight line segment; if its length is less than the preset minimum length threshold, it is marked as insufficient length; determining whether there is a valid intersection by calculating the equations of the two straight line segments and solving for the intersection; and determining whether the two endpoints of each straight line segment fall within the bounding boxes of the two bolts, and whether the length of the line segment falling within the bolt area is not less than the preset connection length threshold to ensure the physical connection between the anti-loosening line and the bolts. If a straight line segment is broken in any of the above verification steps, i.e., the length is insufficient, there is no intersection, or it is not properly connected to the bolt area, the anti-loosening line area is determined to be incompletely segmented. In this case, the image preprocessing parameters or segmentation parameters will be automatically adjusted, and the image acquisition and processing steps and precise region segmentation and extraction will be re-executed until the anti-loosening line area is accurately and completely segmented.

[0045] The integrity verification of the figure-eight anti-loosening line sets three key indicators, including: the continuity and length standard of the anti-loosening line segment, the geometric structure standard of the "figure-eight" shape, and the reliability standard of the connection between the anti-loosening line and the bolt.

[0046] The first indicator is the continuity and length standard of the anti-loosening line segment. Both straight line segments in the anti-loosening line area must be continuous without breakage. By traversing the pixel sequence of each straight line segment, the number of times the gray value of adjacent pixels changes abruptly and exceeds the color range of the anti-loosening line is counted. If the number of breaks is 0 and the pixel length of each straight line segment is not less than the preset minimum length threshold, then the continuity and length requirements are met.

[0047] The second indicator is the geometric structure standard of the "figure-eight" shape. The extensions of the two straight line segments must intersect and form an angle, which is between 30° and 150°. This is set according to the standard installation specifications of the anti-loosening line for railway train bolts. For example, the typical design angle is 60°±30°, and the intersection point is located near the middle of the line connecting the two sets of bolts.

[0048] The third indicator is the reliability standard for the connection between the anti-loosening line and the bolt. Either endpoint of the two straight line segments must fall completely within the area of ​​the two sets of bolts. Specifically, this is determined by calculating the number of overlapping pixels between the endpoints of the line segments and the boundary boxes of the bolts. This ensures that the length of the line segment falling within the bolt area is not less than the preset connection length threshold, thus ensuring that the connection between the anti-loosening line and the bolt is normal and secure.

[0049] The segmented anti-loosening line areas are checked one by one. The specific check process is as follows: By performing pixel-level traversal and connected component analysis on the two segmented line segments, the pixel continuity is checked, and the actual pixel length of each line segment is calculated. If a break point is found in the line segment, that is, the pixels are discontinuous or the actual length is less than the minimum length threshold, the continuity requirement is not met. The equations of the two line segments are calculated and the coordinates of the intersection point are solved. Then, the angle between the direction vectors of the two line segments is calculated using the vector dot product formula. If the calculated angle is not within the preset range of 30°-150°, or the position of the intersection point does not meet the geometric requirements near the perpendicular bisector, the "figure-eight" shape requirement is not met. By comparing the pixel coordinates of the endpoints of each line segment with the pixel coordinates of the two sets of bolt areas, it is determined whether it falls completely within the corresponding bolt area, and the connection length is verified to reach the preset threshold. If the endpoint of any line segment fails to connect correctly with the bolt area, the connection reliability requirement is not met.

[0050] Based on the above judgment results, corresponding instructions are output. If the anti-loosening line simultaneously meets all the above integrity indicators, namely continuity, length, "figure-eight" shape, and connection reliability, the figure-eight anti-loosening line is determined to be intact and proceeds to subsequent in-depth analysis. If any indicator is not met, such as line segment breakage, lack of intersection, excessive deviation of intersection position, abnormal angle, or failure to be effectively connected to the bolt, the figure-eight anti-loosening line is directly determined to be broken or seriously abnormal, and a replacement instruction is generated and output. The instruction includes the precise location information of the abnormal bolt, the type of anti-loosening line abnormality, the suggested replacement time, and the corresponding operation procedure guidance. The precise location information includes the train number, carriage number, component code, and specific coordinates of the bolt. The type of anti-loosening line abnormality includes complete breakage, partial breakage, and loosening. This instruction is transmitted to the train maintenance management system or central control platform in real time, and an alarm prompt is displayed in red on the display screen of the inspection equipment terminal, prompting maintenance personnel to intervene and handle the situation in a timely manner.

[0051] like Figure 4As shown, the linear deformation depth analysis step aims to input the collected and finely preprocessed image into a preset image analysis model when the anti-loosening line is initially judged to be intact. This model accurately analyzes the bolt loosening angle, the intersection position of the anti-loosening line, and subtle deformation data, providing a high-precision, quantitative basis for subsequent anomaly judgment. First, a preset image analysis model is constructed and trained. This model adopts a convolutional neural network architecture, with a YOLOv8 module integrated at its front end. This YOLOv8 module is used for high-precision target detection and localization, and can quickly and robustly identify two sets of bolts and figure-eight anti-loosening lines in the image. The model identifies the intersection of the loosening lines and outputs their precise center point coordinates and bounding box information. The backend of the model integrates a custom feature extraction module, consisting of multiple convolutional layers, pooling layers, and fully connected layers. This module extracts fine pixel-level features from local image regions. Specifically, it can extract the angle features of the red marker lines in the bolt area and the deformation features such as the curvature and diameter changes of the lines in the anti-loosening line area. The model is trained through supervised learning on a large dataset of labeled images, which includes images of normal working conditions, bolt loosening conditions of varying degrees, and anti-loosening line deformation conditions of different types.

[0052] The pre-processed image is input into a trained analysis model. The model first uses a YOLOv8 module to output the precise center point coordinates of the two bolts, their minimum bounding rectangle information, and the pixel coordinates of the intersection of the figure-eight anti-loosening lines. Then, a custom feature extraction module extracts the pixel features of the red marking lines on each bolt region. These red marking lines are pre-painted baselines on the bolt heads, indicating the initial installation angle. Based on the pixel distribution of the red marking lines, the model calculates the angle between the marking lines and the horizontal or vertical axis of the image using image gradient analysis combined with least squares fitting, determined during camera calibration according to the actual installation direction of the bolts. This calculated angle serves as the basis for... The real-time installation angle of the bolt is calculated by combining it with the reference angle of each bolt during initial installation. For example, if the real-time angle change of the marking line of one bolt (A) is positive, it means that the bolt is rotating clockwise, while the real-time angle change of the marking line of another bolt (B) is negative, it means that the bolt is rotating counterclockwise. The absolute values ​​of the angle changes of the two bolts satisfy a specific logical relationship. For example, if the sum of the absolute values ​​is greater than a preset threshold, it is preliminarily determined that bolt (A) is loose. Conversely, if the angle change of bolt (A) is negative and that of bolt (B) is positive, it is preliminarily determined that bolt (B) is loose. At the same time, the absolute value of the angle change is quantified as the rotation angle θ of the loose bolt.

[0053] In addition, the analysis model can also output deformation data of the figure-eight anti-loosening line, specifically, data related to the tension of the anti-loosening line. This data is obtained by selecting n evenly distributed sampling points on each of the two straight segments of the anti-loosening line, where the value of n is determined based on the pixel length of the anti-loosening line in the image. The model outputs the precise pixel coordinates of each sampling point. ;

[0054] The output is the diameter data of the anti-loosening line. This data is obtained by selecting m measurement points on each straight line segment. The model outputs the pixel diameter of the anti-loosening line at each measurement point through an image processing algorithm. The diameter data will be used to evaluate whether the anti-loosening line has been overstretched, causing the diameter to become thinner or compressed, causing the diameter to become thicker.

[0055] The purpose of the linear deformation safety judgment step is to preliminarily verify the deformation data output by the analysis model by setting a safety threshold for the deformation of the anti-loosening line, and to determine whether it is within the safe range, thereby quickly eliminating abnormal situations caused by severe deformation. First, the safety threshold for the deformation of the anti-loosening line is determined. This threshold is obtained by statistical analysis of a large amount of anti-loosening line deformation data under normal operating conditions. Specifically, at least 1,000 sets of images of the figure-eight anti-loosening line of the track train under normal operating conditions are collected. These images are input into the image analysis model, and the deformation data of the anti-loosening line in each set of images is extracted, including the straightness deviation of n sampling points and the diameter data of each measurement point.

[0056] For straightness deviation data, calculate its mean. and standard deviation Straightness deviation is defined as the root mean square value of the distance from n sampling points to the anti-loosening line segment fitted by the least squares method, based on the normal distribution. In principle, a safe threshold for the degree of tension is set as follows: When the actual detected straightness deviation of the anti-loosening line is less than the threshold, the tension of the anti-loosening line is determined to be within a safe range.

[0057] For the diameter data, the diameter data at each measurement point were statistically analyzed, and the mean value was calculated. and standard deviation The diameter data is the pixel width of the anti-loosening line at each measurement point, and the safe threshold range for diameter deformation is set as follows. When the actual measured diameter of the anti-loosening wire is within this range, the diameter deformation is determined to be within the safe range.

[0058] Next, threshold verification and result processing are performed on the anti-loosening line deformation data output by the analysis model. First, the actual straightness deviation of n sampling points is calculated. If the actual straightness deviation is greater than the preset tension safety threshold, the result is processed. If the anti-loosening wire is excessively loose, severely bent, or locally deformed, it indicates that the deformation has exceeded the safe range. Secondly, check the actual diameter of the anti-loosening wire at m measurement points. If the diameter at any measurement point exceeds the preset safe diameter threshold (i.e., less than...), it indicates that the anti-loosening wire is not properly secured. or greater than If the deformation is too large, it indicates that the anti-loosening line is either overstretched at that point, causing the diameter to become thinner, or compressed, causing the diameter to become thicker, and the deformation is determined to be beyond the safe range.

[0059] Based on the above judgment results, the system outputs corresponding instructions. If the tension of the anti-loosening line and the diameter data of all measuring points are within their respective safety threshold ranges, the system determines that the deformation of the current anti-loosening line is in a normal state and enters the abnormal deformation judgment step. If any deformation data exceeds the safety threshold, the system directly determines that the figure-eight anti-loosening line is abnormal and generates a detailed abnormality report. The abnormality report includes the abnormality type, such as excessive slack, local bending, excessive stretching, local compression, the specific value of the deformation data exceeding the threshold, the precise location information of the bolt, and the suggested maintenance measures.

[0060] The purpose of the linear deformation anomaly determination step is to calculate the theoretical deformation data of the anti-loosening line caused by bolt loosening when the initial deformation of the anti-loosening line is judged to be normal. This data is then compared with the actual deformation data of the anti-loosening line extracted from the actual acquired images with high precision to assess whether the deviation is within a reasonable range. This ultimately determines whether there are deeper, non-obvious anomalies in the figure-eight anti-loosening line, such as material fatigue, internal damage, or changes in prestress. First, theoretical data calculations are performed based on a mechanical model of the figure-eight anti-loosening line established by the bolt loosening angle. This mechanical model comprehensively considers the physical dimensions of the anti-loosening line, material properties, and bolt specifications. Physical dimensions include wire diameter and initial free length; material properties include elastic modulus and Poisson's ratio; and bolt specifications include pitch, thread diameter, and bolt head diameter. When the analysis model determines that a bolt has loosened and accurately calculates its rotation angle θ, the mechanical model calculates the effective elongation of the anti-loosening line at the bolt connection point caused by the bolt rotation through geometric relationships. Specifically, the rotation of a bolt changes the position of its anti-loosening line anchor point relative to the anchor point of another bolt, thus causing a change in the length of the anti-loosening line segment connecting the two bolts.

[0061] According to Hooke's Law, the elongation of the anti-loosening thread... The resulting change in tension ,in To prevent the original length of the line from loosening, Given the cross-sectional area of ​​the anti-loosening wire, and further, combining the geometric configuration and force equilibrium conditions of the anti-loosening wire, the theoretical coordinates of the intersection points of the anti-loosening wires can be accurately calculated. The location of this intersection point is determined by solving for the intersection of two circles with the two bolt anchor points as centers and the two new lengths of the anti-loosening line as radii.

[0062] At the same time, the mechanical model calculates the theoretical deformation data of the anti-loosening line, based on the elongation of the anti-loosening line. Through Poisson's ratio Definition: ,in For axial strain, This represents the change in diameter. Given the original diameter, calculate the theoretical diameter change of the anti-loosening wire. Thus, the theoretical diameter is obtained. Furthermore, under ideal conditions of uniform stress and no damage, the anti-loosening line under tension should be a perfectly straight line. Therefore, theoretically, the n sampling points should be completely located on the same straight line, thus setting the theoretical straightness deviation to 0.

[0063] Further, actual data extraction is performed. From the images actually collected and preprocessed in step one, high-precision image coordinate positioning tools, such as linear equations fitted by the least squares method and precise intersection point calculation algorithms, are used to accurately extract the actual intersection point of the two straight line segments of the figure-eight anti-loosening line, and read the actual position coordinates of the intersection point. Simultaneously, the actual deformation data of the anti-loosening line is extracted. Within the anti-loosening line area of ​​the actual image, sampling point positions consistent with the analysis model output are determined, i.e., n sampling points and measurement point positions, i.e., m measurement points. The actual straightness deviation and actual diameter are then recalculated. The actual diameter is obtained by calculating the root mean square value of the distance from n sampling points to the actual anti-loosening line fitted by the least squares method. By using an edge detection algorithm to extract the two edges of the anti-loosening line at m measurement points consistent with the model output, calculating the pixel distance between the edges, and taking the average value as the final actual average diameter. .

[0064] Next, calculate the deviation between theoretical and actual data, and set a reasonable range for the deviation. Calculate the deviation of the intersection point of the anti-loosening line, and use the Euclidean distance formula to calculate the deviation value between the theoretical and actual positions. Calculate the deviation of the anti-loosening wire diameter, that is, the absolute deviation between the theoretical diameter and the actual average diameter. Calculate the straightness deviation, which is the difference between the theoretical straightness deviation of 0 and the actual straightness deviation. .

[0065] Finally, a reasonable range for deviation is set, and anomaly judgment is performed. The reasonable range for deviation is determined by referring to the statistical results of theoretical and actual deviation data under a large number of normal bolt loosening conditions. Specifically, at least 500 sets of working condition data that have been confirmed as normal but with slight bolt loosening are collected, and the above theoretical calculation and actual extraction process is repeated to statistically analyze each deviation index. mean and standard deviation According to the normal distribution The final reasonable deviation range is set as follows: When the calculated If any value exceeds this reasonable range, it is determined that there is a deep anomaly in the figure-eight anti-loosening line. This unexpected deviation indicates that although the bolt loosening has produced a certain theoretical deformation, the actual deformation is significantly different from the theoretical prediction. This may indicate that the anti-loosening line material has undergone fatigue, there is minor internal damage, abnormal initial installation prestress, or other structural problems not directly caused by bolt loosening.

[0066] Based on the final anomaly judgment result, the system outputs corresponding instructions. If all deviation indicators are within a reasonable range, the system determines that the figure-eight anti-loosening line is functioning normally. If any deviation indicator exceeds a reasonable range, the system determines that the figure-eight anti-loosening line has a potential deep anomaly and generates a detailed anomaly diagnosis report. The report includes the specific value of the deviation exceeding the threshold, the possible anomaly type, the precise location of the abnormal bolt, and the suggested in-depth inspection or replacement instructions.

[0067] The process also includes a wire tightening connection analysis step. As a verification step, the tightening connection is located based on the intersection point. The tightening connection is extracted using target detection. If the tightening connection is an intersection point, the intersection point can be directly located as the tightening connection. Then, the edge contour of the tightening connection structure is extracted using an edge detection algorithm. Image morphological analysis is used to calculate the contour area, contour roundness, and contour center offset. The contour area is the number of pixels enclosed by the edge contour of the tightening connection. The contour roundness reflects the regularity of the tightening connection structure. The contour center offset is obtained by comparing the calculated geometric center coordinates of the tightening connection contour with the preset standard center coordinates to obtain the offset distance in the X and Y axes. ;

[0068] A large number of images of tightened connections under "normal conditions" were collected in advance. Morphological parameters, including area, roundness, and center offset, were extracted using the aforementioned techniques. The K-means clustering algorithm was used to analyze the parameters and determine the standard range of each parameter to construct a standard morphological parameter database. The real-time morphological parameters extracted during the detection process were compared one by one with the standard range in the database. If all parameters were within the standard range, the tightened connection was determined to be normal in morphology, without deformation, misalignment, or other problems. If any parameter exceeded the standard range, it was marked as "morphologically abnormal," and it was initially determined that the anti-loosening line might be at risk of loosening.

[0069] Next, the number of turns at the tightening connection is calculated. Taking the geometric center of the tightening connection as the origin, rays are emitted radially (360° evenly divided into 16 scanning directions). When each ray passes through the edge of the winding structure, the edge crossing event is recorded sequentially. Since each turn of the winding structure will cause the ray to cross twice, the actual number of turns of the tightening connection can be obtained by counting the total number of crossing events of all rays, dividing by 2 and taking the integer. The actual number of turns is then compared with the initially recorded number of turns or the preset standard number of turns, and the number of turns analysis results are output. Finally, a multi-dimensional fusion judgment is performed, mainly by comprehensively judging the morphological analysis results and the number of turns analysis results. If both are normal, the tightening connection status of the anti-loosening wire intersection is judged to be normal; otherwise, the anti-loosening wire loosening signal is output.

[0070] A system was designed based on the abnormal detection method for the figure-eight anti-loosening line of railcars, including:

[0071] The image acquisition and processing module captures images of a fixed area containing the anti-loosening line through vision and performs preprocessing to obtain the target image;

[0072] The line integrity judgment module extracts a target area map from the target image by using a preset analysis model, which includes bolts, bolt marking lines and anti-loosening lines. In the target area map, the anti-loosening lines are judged to be in an intact state by the continuity of line segments and the shape of the anti-loosening lines.

[0073] The linear deformation depth analysis module, when the anti-loosening line is in a complete state, inputs the target area map into the analysis model. The analysis model analyzes the position of the bolt marking line to obtain the loosening angle of the bolt, and analyzes the intersection position of the anti-loosening line and the deformation data of the anti-loosening line.

[0074] The linear deformation safety judgment module defines straightness deviation threshold and diameter threshold based on historical data and combines them to obtain a safety threshold group. It compares the anti-loosening linear deformation data with the safety threshold group and outputs an abnormal signal or comparison signal based on the comparison result.

[0075] The linear deformation anomaly determination module, when receiving a comparison signal, compares the actual intersection position and actual deformation data of the anti-loosening line in the target area map with the theoretical intersection position and theoretical deformation data of the anti-loosening line output by the preset mechanical model based on the loosening angle, and outputs an abnormal signal or a normal signal based on the comparison result.

[0076] 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 abnormalities in the anti-loosening wires of a railcar, characterized in that: Includes the following steps: The image acquisition and processing steps involve visually capturing images of a fixed area containing the anti-loosening line and preprocessing them to obtain the target image. The line integrity judgment step involves extracting a target area map from the target image using a preset analysis model, which includes bolts, bolt marking lines, and anti-loosening lines. In the target area map, the anti-loosening lines are judged to be in an intact state based on the continuity of line segments and the shape of the anti-loosening lines. The linear deformation depth analysis step involves inputting the target area map into the analysis model when the anti-loosening line is in a complete state. The analysis model analyzes the position of the bolt marking line to obtain the loosening angle of the bolt, and analyzes the intersection position of the anti-loosening line and the deformation data of the anti-loosening line. The linear deformation safety judgment step involves defining straightness deviation thresholds and diameter thresholds based on historical data and combining them to obtain a safety threshold group. The anti-loosening linear deformation data is compared with the safety threshold group, and an abnormal signal or comparison signal is output based on the comparison result. The abnormal deformation determination step involves, upon receiving a comparison signal, comparing the actual intersection position and actual deformation data of the anti-loosening line in the target area map with the theoretical intersection position and theoretical deformation data of the anti-loosening line output by the preset mechanical model based on the loosening angle, and outputting an abnormal or normal signal based on the comparison result.

2. The method for detecting abnormalities in the anti-loosening wire of a railcar according to claim 1, characterized in that: The line integrity determination step includes: The continuity judgment sub-step traverses the pixel sequence of the straight line segment of the anti-loosening line and calculates the number of times the gray value of adjacent pixels changes abruptly and exceeds the color range of the anti-loosening line. If the number is zero and the pixel length of each straight line segment is not less than the preset minimum length threshold, then the continuity and length compliance conditions are triggered. The geometric shape judgment sub-step determines whether the two straight line segments of the anti-loosening line intersect and whether the intersection point is located between the two sets of bolts. If they intersect and the intersection point is located between the two sets of bolts, the shape compliance condition is triggered. The connection reliability judgment sub-step determines whether the two endpoints of the anti-loosening line straight segment overlap with the boundary frames of the two sets of bolts. If so, the reliability compliance condition is triggered.

3. The method for detecting abnormalities in the anti-loosening wire of a railcar according to claim 2, characterized in that: The analysis model includes a convolutional neural network architecture with a YOLOv8 module at the front end. The YOLOv8 module is used to identify and extract bolts, bolt marking lines, and anti-loosening lines. The back end includes a feature extraction module, which includes multiple convolutional layers, pooling layers, and fully connected layers. The feature extraction module calculates the angle between the bolt marking line and the horizontal axis by using image gradient analysis and least squares fitting based on the pixel distribution of the bolt marking line. This yields the reference angle of the bolt during initial installation and the real-time installation angle of the bolt. The loosening angle of the bolt is then calculated based on the reference angle and the real-time installation angle.

4. The method for detecting abnormalities in the anti-loosening wire of a railcar according to claim 3, characterized in that: The analysis model selects n uniformly distributed sampling points on the two straight segments of the loosening line and outputs the pixel coordinates of each sampling point. It calculates the root mean square value of the distance from each sampling point to the loosening line segment fitted by the least squares method as the straightness deviation. Then, it selects n measurement points on the two straight segments of the loosening line and outputs the pixel diameter of the loosening line at each measurement point through an image processing algorithm. The straightness deviation and the pixel diameter are combined to obtain the deformation data of the loosening line.

5. The method for detecting abnormalities in the anti-loosening wire of a railcar according to claim 4, characterized in that: The linear deformation safety judgment step safety threshold group definition strategy includes calculating the first mean and first standard deviation of the sampling points corresponding to the non-abnormal anti-loosening line based on historical data, setting the straightness deviation threshold based on the first mean and first standard deviation, calculating the diameter of the measurement points corresponding to the non-abnormal anti-loosening line based on historical data, calculating the second mean and second standard deviation, and setting the diameter threshold based on the second mean and second standard deviation.

6. A method for detecting abnormalities in the anti-loosening wire of a railcar according to claim 1 or 5, characterized in that: The linear deformation anomaly determination step includes a theoretical data calculation strategy. This strategy involves calculating the effective elongation of the anti-loosening line at the bolt connection point caused by bolt rotation based on the bolt loosening angle output by the analysis model through geometric relationships, then calculating the resulting tensile force change based on the effective elongation, and calculating the theoretical position coordinates of the anti-loosening line intersection point based on the geometric configuration and tensile force change of the anti-loosening line, and finally calculating the theoretical diameter of the anti-loosening line based on Poisson's ratio.

7. The method for detecting abnormalities in the figure-eight anti-loosening wire of a railcar according to claim 6, characterized in that: The linear deformation anomaly determination step also includes an actual data calculation strategy, which includes extracting the actual intersection coordinates of the two straight line segments of the anti-loosening line in the target area map, and recalculating the actual straightness deviation and actual diameter in the target area map based on the sampling point position and measurement point position selected by the analysis model.

8. The method for detecting abnormalities in the anti-loosening wire of a railcar according to claim 1, characterized in that: The process also includes a wire tightening connection analysis step. This involves locating the tightening connection, extracting the edge contour of the tightening connection structure using an edge detection algorithm, calculating the contour area, contour roundness, and contour center offset using image morphology analysis, comparing the contour area, contour roundness, and contour center offset with the corresponding parameter range in a preset standard morphological parameter database, and outputting the morphological analysis result. Then, with the geometric center of the tightening connection as the origin, rays are emitted along 360° evenly divided into n scanning directions. The total number of intersection events between the rays and the edge of the winding structure is counted. The total number of intersection events is divided by 2 and rounded to obtain the actual number of winding turns. The actual number of winding turns is compared with a preset standard number of turns range, and the number of turns analysis result is output. A comprehensive judgment is made based on the morphological analysis result and the number of turns analysis result. If both are normal, the tightening connection status of the anti-loosening wire intersection is determined to be normal; otherwise, a loosening signal for the anti-loosening wire is output.

9. The method for detecting abnormalities in the anti-loosening wire of a railcar according to claim 8, characterized in that: The analysis step of the wire tightening connection includes a standard morphological parameter database construction strategy. Based on the contour area, contour roundness, and contour center offset data of the tightening connection under various normal conditions in historical data, the standard range of contour area, the standard range of contour roundness, and the standard range of contour center offset are determined by clustering algorithm to form a standard morphological parameter database.

10. An abnormal detection system for the figure-eight anti-loosening line of a rail train, characterized in that: include: The image acquisition and processing module captures images of a fixed area containing the anti-loosening line through vision and performs preprocessing to obtain the target image; The line integrity judgment module extracts a target area map from the target image by using a preset analysis model, which includes bolts, bolt marking lines and anti-loosening lines. In the target area map, the anti-loosening lines are judged to be in an intact state by the continuity of line segments and the shape of the anti-loosening lines. The linear deformation depth analysis module, when the anti-loosening line is in a complete state, inputs the target area map into the analysis model. The analysis model analyzes the position of the bolt marking line to obtain the loosening angle of the bolt, and analyzes the intersection position of the anti-loosening line and the deformation data of the anti-loosening line. The linear deformation safety judgment module defines straightness deviation threshold and diameter threshold based on historical data and combines them to obtain a safety threshold group. It compares the anti-loosening linear deformation data with the safety threshold group and outputs an abnormal signal or comparison signal based on the comparison result. The linear deformation anomaly determination module, when receiving a comparison signal, compares the actual intersection position and actual deformation data of the anti-loosening line in the target area map with the theoretical intersection position and theoretical deformation data of the anti-loosening line output by the preset mechanical model based on the loosening angle, and outputs an abnormal signal or a normal signal based on the comparison result.

Images