Cross guide rail wear condition monitoring method based on machine vision

Abstract

This invention relates to the field of wear monitoring, and more particularly to a machine vision-based method for monitoring the wear condition of cross-rail guides. The method includes: acquiring an initial image and a standard image of the cross-rail guide; subtracting the initial image from the standard image to obtain an error map; binarizing the error map using the Otsu thresholding method to obtain a binary image; identifying regions composed of white pixels in the binary image as abnormal regions; calculating the texture complexity factor and shape factor of the abnormal regions; normalizing the product of the texture complexity factor and shape factor to obtain a wear feature value; extracting the edge lines of the abnormal regions; calculating the mean gradient magnitude of the edge line pixels; and weighted summing the wear feature value and the mean gradient magnitude of the edge line pixels to obtain a wear index, which is used to determine whether wear has occurred on the surface of the cross-rail guide. This invention improves the accuracy of cross-rail condition monitoring.

Description

Technical Field

[0001] This invention relates to the field of wear monitoring, and more particularly to a machine vision-based method for monitoring the wear condition of cross rails. Background Technology

[0002] Cross-rails, as high-precision linear motion units, are used in CNC machine tools, semiconductor manufacturing equipment, precision measuring instruments, and other fields. After prolonged use, the surface of cross-rails will wear down, which directly affects the machining accuracy and stability of the equipment. Therefore, it is necessary to monitor the wear condition of the cross-rails to ensure the accuracy and stability of the corresponding machining equipment.

[0003] Chinese patent document CN110458824B discloses a method for detecting drill bit wear. This method utilizes image recognition to obtain key feature values ​​of drill bit wear, analyzes and identifies information such as the drill bit body, connector, matrix, cutting teeth, nozzle, blades, bearings, ferrules, and roller cones, compares it with a preset database, outputs drill bit wear detection and evaluation results, and provides suggestions for drill bit optimization and improvement. If a wear cannot be detected, the method can perform self-learning training to expand the database and improve the accuracy of drill bit wear identification.

[0004] Variational Autoencoder (VAE) anomaly detection algorithms train VAE models on massive amounts of images of healthy (i.e., no wear, no interference) guide rails, enabling the model to learn deeply and accurately reconstruct the paradigm of healthy surfaces. During online monitoring, if the input image contains patterns the model hasn't learned, such as physical wear, the VAE will be unable to reconstruct effectively, resulting in significant reconstruction errors in anomalous areas. By detecting these significant errors, preliminary detection of potential defects can be achieved.

[0005] However, due to the working characteristics of cross rails, their surfaces are prone to oil stains, water marks, reflections, and other phenomena. When using VAE models to monitor cross rails, their surfaces are easily interfered with by oil stains, water marks, and reflections, which may lead to misidentification of oil stains, water marks, and reflections as wear defects, resulting in low accuracy of monitoring results and hindering the monitoring of the wear condition of cross rails. Summary of the Invention

[0006] To address the problem of low accuracy in monitoring results caused by interference from the external environment on cross rails, this invention provides a machine vision-based method for monitoring the wear status of cross rails.

[0007] This invention provides a machine vision-based method for monitoring the wear condition of cross rails, employing the following technical solution:

[0008] Obtain the initial image and standard image of the cross rail. Subtract the initial image from the standard image to obtain the error map. Use the Otsu thresholding method to binarize the error map to obtain a binary map. The area composed of white pixels in the binary map is taken as the abnormal area. Calculate the texture complexity factor and shape factor of the abnormal area. Normalize the product of the texture complexity factor and shape factor to obtain the wear feature value.

[0009] The edge lines of the abnormal region are extracted, the mean gradient magnitude of the edge line pixels is calculated, and the wear index is obtained by weighted summation of the wear feature value and the mean gradient magnitude of the edge line pixels, which is used to determine whether wear has occurred on the surface of the cross rail.

[0010] It can efficiently identify abnormal areas on the guide rail surface and form a wear index by fusing wear characteristic values ​​with edge gradient amplitude, thereby achieving accurate identification of actual wear, reducing interference from the external environment during the detection process, and effectively improving the accuracy of cross guide rail condition monitoring.

[0011] Preferably, the method for obtaining the standard image is as follows: A historical image set of intersecting guide rails is obtained, containing multiple historical images of intersecting guide rails; a corresponding label is set for each historical image; the initial image is added to the historical image set; the labels in the historical image set are clustered to obtain multiple clusters, and the cluster to which the initial image belongs is obtained; within the cluster to which the initial image belongs, the regions of the intersecting guide rails in the multiple historical images are aligned; the pixels at the same position in the multiple historical images are sorted according to their grayscale values ​​from smallest to largest; the mean of the grayscale values ​​in the middle part is calculated; the mean is used as the standard grayscale value at that position, further obtaining the standard image.

[0012] By clustering historical images to generate standard images, we can ensure that the standard images have good matching and representativeness under environmental variables such as light intensity and vibration frequency, thereby enhancing their adaptability and robustness under complex working conditions.

[0013] Preferably, the method for calculating the texture complexity factor of the abnormal region is as follows: The abnormal region is mapped onto a grayscale image; the gradient direction of each pixel in the abnormal region is calculated using the Canny operator; multiple directional regions are constructed, each with the same angle; the number and frequency of pixels in the abnormal region falling into each directional region are calculated; and the expression for the texture complexity factor of the abnormal region is: In the formula, H represents the texture complexity factor of the abnormal region. This represents the frequency of the k-th directional region, where k represents the index of the directional region ordinal number, and n represents the total number of directional regions.

[0014] By using a texture complexity factor calculation method based on pixel gradient direction, the texture distribution differences in image regions can be accurately quantified. This method is suitable for distinguishing between real wear and pseudo-wear with a single texture, thereby improving the ability to detect anomalies.

[0015] Preferably, before calculating the shape factor of the abnormal region, the method further includes: calculating the straightness of the abnormal region. The calculation method is as follows: In the binary image, construct a rectangular coordinate system with any point as the origin, with one coordinate axis as the x-axis and the other as the y-axis. Obtain the position coordinates of each pixel in the abnormal region, construct a coordinate set using the position coordinates of each pixel, construct a covariance matrix using the coordinate set, calculate the eigenvalues ​​of the covariance matrix, and further obtain the straightness of the abnormal region, expressed as: In the formula, ZX represents the straightness of the abnormal region. , These represent the maximum and minimum values ​​of the eigenvalues, respectively.

[0016] By constructing a straightness index using the covariance matrix and combining it with image geometric features to quantitatively analyze the morphology of the wear area, typical linear scratch-type wear can be accurately identified, thus improving the identification accuracy.

[0017] Preferably, before calculating the shape factor of the abnormal region, the method further includes: calculating the roundness of the abnormal region, which is as follows: in the binary image, extract the edge line of the abnormal region, calculate the perimeter and area of ​​the abnormal region, take the number of pixels in the abnormal region as the area of ​​the abnormal region, and calculate the roundness of the abnormal region, expressed as: In the formula, YD represents the roundness of the abnormal region, S represents the area of ​​the abnormal region, and L represents the perimeter of the abnormal region.

[0018] Preferably, the expression for the shape factor is:

[0019]

[0020] In the formula, ZX represents the shape factor of the abnormal region, YD represents the straightness of the abnormal region, and max represents the maximum value function.

[0021] By integrating straightness and roundness, a comprehensive morphological evaluation index is established, which provides a more comprehensive assessment of the geometric structure of abnormal areas and helps to further screen out irregularly shaped pseudo-wear areas.

[0022] Preferably, the wear index is calculated as follows: The abnormal region is mapped onto a grayscale image; the edge line is extended outwards by multiple pixels to obtain an auxiliary region; the gradient magnitude is binary-classified using the Otsu thresholding method to obtain two clusters; the mean of the gradient magnitude within each cluster is calculated; and the cluster with the larger mean is selected as the first cluster. The expression for the wear index is:

[0023]

[0024] Where, The wear index indicates the abnormal area. This represents the preset weighting coefficient. This represents the mean of the gradient magnitudes within the first cluster. The wear characteristic value represents the abnormal area.

[0025] Introducing auxiliary region and edge gradient clustering mechanisms can enhance the saliency of real wear edge features in images, and jointly calculate the wear index with texture morphology features. While preserving real wear areas, it can effectively reduce the risk of misidentification of non-wear areas such as transparent droplets.

[0026] Preferably, the weighting coefficients are calculated as follows: The within-class variance and between-class variance of the two clusters are calculated, and the expression for the weighting coefficients is:

[0027]

[0028] Where, This represents the preset weighting coefficient. This represents the between-class variance of two clusters. Let represent the within-cluster variance of two clusters, and tanh represent the hyperbolic tangent function.

[0029] Adaptive adjustment based on intra-class and inter-class variance enables the wear index to dynamically balance the contributions of texture and edge features according to the current image characteristics, thereby further improving the model's generalization ability and computational accuracy.

[0030] Preferably, the label includes light intensity and the vibration frequency of the cross rails.

[0031] Preferably, the DBSCAN clustering algorithm is used to cluster the labels in the historical image set to obtain multiple clusters.

[0032] Using the DBSCAN clustering method to perform cluster analysis on image labels avoids dependence on the number of clusters, has greater adaptability and practicality, and ensures the effectiveness of the standard image construction process.

[0033] The present invention has the following technical effects:

[0034] This invention can efficiently identify abnormal areas on the surface of the guide rail and form a wear index by fusing wear characteristic values ​​with edge gradient amplitude, thereby achieving accurate identification of real wear. It reduces the interference of the external environment on the detection process, can distinguish between real wear and pseudo wear, and effectively improves the accuracy of cross guide rail condition monitoring. Attached Figure Description

[0035] Figure 1 This is a flowchart of the cross-rail wear condition monitoring method based on machine vision according to the present invention. Detailed Implementation

[0036] 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, not all, of the embodiments of the present invention. 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.

[0037] This invention discloses a machine vision-based method for monitoring the wear condition of cross rails, referring to... Figure 1 The process includes the following steps, as detailed below:

[0038] S1: Obtain the initial image of the cross rails, process the initial image to obtain a binary image, and further obtain the abnormal region.

[0039] A historical image set of the cross rails is obtained, containing multiple historical images of the cross rails. The historical images are grayscale images. Each historical image is assigned a corresponding label, including the light intensity and the vibration frequency of the cross rail. An initial image of the cross rail and the label are acquired using an industrial camera. The initial image is converted into a grayscale image and added to the historical image set. The DBSCAN clustering algorithm is used to cluster the labels in the historical image set to obtain multiple clusters. The cluster to which the initial image belongs is further determined. It can be understood that each cluster corresponds to multiple historical images. The acquisition background (light intensity and vibration frequency) corresponding to the images within the same cluster is similar. Therefore, the grayscale values ​​corresponding to the images within the same cluster are also basically similar.

[0040] In the cluster to which the initial image belongs, there are multiple historical images. The regions where the intersecting guide rails belong in the multiple historical images are aligned. The pixels at the same position in the multiple historical images are sorted in ascending order of gray value. The mean of the gray value in the middle part is calculated and used as the standard gray value at that position to obtain a standard image. The difference between the standard image and the initial image is used to obtain an error map. The error map is binarized using the Otsu thresholding method to obtain a binary map. The connected component processing of the binary map is performed to obtain the abnormal region (the region composed of white pixels).

[0041] In one embodiment, the intermediate grayscale value specifically refers to: the sorted grayscale values ​​forming a grayscale value sequence, obtaining the median of the grayscale value sequence, and selecting multiple grayscale values ​​before and after the median as the intermediate grayscale value.

[0042] In one embodiment, the method for obtaining the intermediate grayscale values ​​is as follows: The sorted grayscale values ​​form a grayscale value sequence; the median of the grayscale value sequence is obtained; and several grayscale values ​​before and after the median are selected as the intermediate grayscale values. For example, the grayscale values ​​of the 11th pixel in multiple historical images are sorted, and the 4th, 5th, 6th, 7th, and 8th grayscale values ​​are selected as the intermediate grayscale values. In historical images, noise data values ​​typically exhibit abnormal increases and decreases. Therefore, by taking the average of the intermediate grayscale values ​​as the grayscale value in the standard image, the error of the standard image is reduced.

[0043] S2: Calculate the texture complexity factor of the abnormal region.

[0044] In the working environment of cross rails, the common real wear is usually scratches and pits, while the pseudo wear is usually oil stains and water stains. The internal structure of real wear is usually rich in texture details, that is, it has high complexity, while pseudo wear is usually liquid spread, and its texture details are relatively simple. Therefore, it is necessary to calculate the texture complexity factor of abnormal areas.

[0045] The calculation method is as follows: The abnormal region is mapped onto a grayscale image. In the grayscale image, the gradient direction of each pixel in the abnormal region is calculated using the Canny operator. Multiple directional regions are constructed, each with the same angle. For example, since the spatial angle is 360°, the spatial angle is divided into 8 directional regions, each with a 45° angle. The first directional region is (0–45°], the second directional region is (45–90°], ..., and the eighth directional region is (270–360°]. The number and frequency of pixels in the abnormal region whose gradient directions fall into each directional region are calculated. For example, if there are 100 pixels in the abnormal region, and 20 pixels have gradient directions within the first directional region (0–45°], then the frequency of the first directional region is 20 / 100 = 0.2. The expression for the texture complexity factor of the abnormal region is: In the formula, H represents the texture complexity factor of the abnormal region. This represents the frequency of the k-th directional region, where k represents the index of the directional region, and n represents the total number of directional regions. A larger texture complexity factor indicates more texture details within the abnormal region, making it more likely to be genuine wear; conversely, a smaller texture complexity factor indicates fewer texture details within the abnormal region, making it more likely to be pseudo-wear.

[0046] S3: Calculate the shape factor of the abnormal region.

[0047] Since true wear is usually manifested as scratches and pits, scratches typically exhibit an approximately straight-line structure, while pits usually exhibit an approximately circular structure. Pseudo-wear areas formed by oil stains and water stains usually exhibit irregular shapes. For example, oil stains and water stains can form a structure with a large top and a thin bottom under the influence of gravity, or form a composite of straight lines and circles under the influence of external forces. Therefore, it is necessary to calculate the shape factor of abnormal areas to further distinguish between true wear and pseudo-wear.

[0048] S31: Calculate the straightness of the abnormal region.

[0049] The calculation method is as follows: In the binary image, construct a Cartesian coordinate system with any point as the origin, with one coordinate axis as the x-axis and the other as the y-axis. Obtain the position coordinates of each pixel in the abnormal region, construct a coordinate set using the position coordinates of each pixel, and construct the covariance matrix using the coordinate set. The expression is:

[0050]

[0051] Where, Indicates abnormal area The covariance matrix, This represents the variance of the coordinate set in the x-direction. This represents the covariance of the coordinate set in the x and y directions. Let represent the variance of the coordinate set in the y-direction. The covariance matrix is ​​symmetric, therefore... equal .

[0052] Calculate the eigenvalues ​​of the covariance matrix to obtain the straightness of the outlier region, expressed as:

[0053]

[0054] In the formula, ZX represents the straightness of the abnormal region. , These represent the maximum and minimum values ​​of the eigenvalues, respectively. The eigenvalues ​​of the covariance matrix reflect the dispersion of the coordinate set in various directions. For a long, thin linear region, the variance of its pixel coordinates along the principal axis is greater than the variance along the secondary axis. Therefore, when the shape of an abnormal region tends to be linear, its pixel distribution is highly concentrated in one direction, leading to… Much larger , When the value of approaches 0, the straightness value approaches 1; conversely, when the region is an irregularly shaped cluster, and The values ​​are similar in size. When the value of approaches 1, the value of straightness approaches 0.

[0055] S32: Calculate the roundness of the abnormal region.

[0056] The calculation method is as follows: In the binary image, extract the edge lines of the abnormal region and calculate the perimeter and area of ​​the abnormal region. Here, the number of pixels in the edge line is taken as the perimeter of the abnormal region, and the number of pixels in the abnormal region is taken as the area of ​​the abnormal region. Then, calculate the roundness of the abnormal region, expressed as:

[0057]

[0058] In the formula, YD represents the roundness of the abnormal region, S represents the area of ​​the abnormal region, and L represents the perimeter of the abnormal region. When the shape of the abnormal region is close to a circle, the value of the roundness of the abnormal region approaches 1. When the shape of the abnormal region is irregular, its value approaches 0. Therefore, the roundness of the abnormal region can be used to determine how close the shape of the abnormal region is to a circle.

[0059] S33: Calculate the shape factor of the abnormal region using the straightness and abnormality of the abnormal region.

[0060] The expression for the shape factor is:

[0061]

[0062] Where, ZX represents the shape factor of the abnormal region, YD represents the straightness of the abnormal region, and max represents the maximum value function.

[0063] On cross guide rails, the shape of the wear area is usually manifested as linear scratch wear and pit wear formed by material detachment. The straightness of linear scratch wear is close to 1, and its roundness is close to 0. The roundness of pit wear is close to 1, and the straightness is close to 0. For the pseudo-wear area formed by droplets and oil stains, due to its irregular shape, its shape factor value is small. Therefore, for an abnormal area, the maximum value of straightness and roundness is taken to characterize the shape factor of the abnormal area. It can be understood that the closer the shape factor value is to 1, the greater the possibility that it is real wear.

[0064] S4: Calculate the wear characteristic value of the abnormal area.

[0065] The expression is:

[0066]

[0067] In the formula, T represents the wear characteristic value of the abnormal area. H represents the shape factor of the abnormal region, H represents the texture complexity factor of the abnormal region, and tanh represents the hyperbolic tangent function, which is used to map the value to (0, 1).

[0068] If the abnormal area is a real wear area, its shape factor and texture complexity factor are larger, so the wear feature value obtained is larger. The wear feature value obtained by calculation can further determine the possibility that the abnormal area is a real wear area.

[0069] S5: Calculate the wear index of abnormal areas using wear characteristic values.

[0070] The working environment of cross-guide rails is relatively complex. When detecting wear on the surface of cross-guide rails, even small droplets containing impurities, such as water or oil droplets, will usually result in a large calculated wear characteristic value. Therefore, using the wear characteristic value to determine the presence of wear can lead to misjudgments. Real wear areas have distinct edge lines, with larger gradient amplitudes at the pixels along the edge. Droplets, on the other hand, are usually transparent, lacking clear edge lines, and the gradient amplitude changes at the boundary pixels are not significant. Therefore, calculating the wear index of abnormal areas based on these characteristics reduces the likelihood of misjudgments.

[0071] The abnormal region is mapped onto a grayscale image, and its edge line is extracted. This edge line is then extended outwards by a certain distance to obtain an auxiliary region (for example, 5 pixels outwards). The Canny operator is used to calculate the gradient magnitude of each pixel within the auxiliary region. This gradient magnitude is then normalized using a linear normalization method. Finally, the Otsu thresholding method is used to perform binary classification of the gradient magnitudes, resulting in two clusters. The mean gradient magnitude within each cluster is calculated, and the cluster with the larger mean is designated as the first cluster, and the cluster with the smaller mean is designated as the second cluster. The expression for the wear index is:

[0072]

[0073] Where, The wear index indicates the abnormal area. This represents the preset weighting coefficient. This represents the mean of the gradient magnitudes within the first cluster. The wear characteristic value represents the abnormal area.

[0074] The weighting coefficients are calculated as follows: the inter-cluster variances of the first and second clusters are calculated. The weighted sum of the variances of data points within the first cluster and the variances of data points within the second cluster is taken as the intra-cluster variance. ;

[0075]

[0076]

[0077]

[0078] Where, This represents the preset weighting coefficient. This represents the between-class variance of two clusters. This represents the within-class variance of two clusters. This represents the mean of the data points within the i-th cluster. This represents the mean gradient magnitude of pixels within the auxiliary region. This represents the number of data points within the i-th cluster, where i is the cluster index, tanh represents the hyperbolic tangent function used for normalization, and N represents the total number of data points within the two clusters. This represents the variance of the data points within the i-th cluster.

[0079] Within the actual wear area, the gradient magnitude of edge pixels is larger. Therefore, pixels corresponding to data points in the first cluster are considered edge pixels, while pixels corresponding to data points in the second cluster are considered non-edge pixels. When calculating the wear index, if... A larger value indicates a more pronounced distinction between edge and non-edge pixels, in which case the gradient magnitude weight should be increased; conversely, if... A small value indicates that the distinguishing features between edge pixels and non-edge pixels are not obvious enough. In this case, the weight of the gradient magnitude should be reduced and the weight of the wear feature value should be increased to improve the accuracy of the wear index calculation result.

[0080] S6: Issues a warning about cross rail wear when the wear index exceeds the preset warning threshold.

[0081] When the wear index exceeds the preset warning threshold, it indicates that there is a real wear area on the surface of the corresponding cross rail, and a warning should be issued to facilitate repair or replacement of the cross rail. When the wear index is less than the preset warning threshold, it indicates that the abnormal area on the surface of the cross rail is a pseudo-wear area, and no additional treatment is required. The warning threshold is set manually according to the actual situation; for example, the warning threshold is 0.8.

[0082] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A machine vision-based method for monitoring the wear condition of cross-guide rails, characterized in that, Including the following steps: Obtain the initial image and standard image of the cross rails. Subtract the standard image from the initial image to obtain the error map. Use the Otsu thresholding method to binarize the error map to obtain a binary map. The area composed of white pixels in the binary map is taken as the abnormal area. Calculate the texture complexity factor and shape factor of the abnormal region, and normalize the product of the texture complexity factor and shape factor to obtain the wear characteristic value; The method for calculating the texture complexity factor of anomaly regions is as follows: Map the anomaly region onto a grayscale image, use the Canny operator to calculate the gradient direction of each pixel in the anomaly region, construct multiple directional regions with the same angle, and calculate the number and frequency of pixels in the anomaly region whose gradient directions fall into each directional region. The expression for the texture complexity factor of the anomaly region is: In the formula, H represents the texture complexity factor of the abnormal region. This represents the frequency of the k-th directional region, where k represents the index of the directional region ordinal number, and n represents the total number of directional regions. The expression for the shape factor is: In the formula, ZX represents the shape factor of the abnormal region, YD represents the straightness of the abnormal region, and max represents the maximum value function. The edge lines of the abnormal region are extracted, the mean value of the gradient magnitude of the edge line pixels is calculated, and the wear index is obtained by weighted summation of the wear feature value and the mean value of the gradient magnitude of the edge line pixels, which is used to determine whether wear has occurred on the surface of the cross guide rail.

2. The machine vision-based method for monitoring the wear condition of cross rails according to claim 1, characterized in that, The standard image acquisition method is as follows: acquire the historical image set of the cross rails, which contains multiple historical images of the cross rails; set corresponding labels for each historical image, add the initial image to the historical image set, cluster the labels in the historical image set to obtain multiple clusters, and obtain the cluster to which the initial image belongs; Within the cluster to which the initial image belongs, the regions to which the intersecting guide rails belong in multiple historical images are aligned. Pixels at the same position in multiple historical images are sorted in ascending order of gray value. The mean of the gray value in the middle part is calculated and used as the standard gray value at that position to further obtain the standard image.

3. The machine vision-based method for monitoring the wear condition of cross rails according to claim 1, characterized in that, Before calculating the shape factor of the abnormal region, the following steps are taken: The straightness of the abnormal region is calculated. The method is as follows: In the binary image, construct a Cartesian coordinate system with any point as the origin, with one coordinate axis as the x-axis and the other as the y-axis. Obtain the position coordinates of each pixel in the abnormal region. Construct a coordinate set using the position coordinates of each pixel. Construct a covariance matrix using the coordinate set. Calculate the eigenvalues ​​of the covariance matrix to further obtain the straightness of the abnormal region. The expression is: In the formula, ZX represents the straightness of the abnormal region. , These represent the maximum and minimum values ​​of the eigenvalues, respectively.

4. The machine vision-based method for monitoring the wear condition of cross rails according to claim 3, characterized in that, Before calculating the shape factor of the abnormal region, the following steps are included: calculating the roundness of the abnormal region. The calculation method is as follows: in the binary image, extract the edge line of the abnormal region, calculate the perimeter and area of ​​the abnormal region, use the number of pixels in the abnormal region as the area of ​​the abnormal region, and calculate the roundness of the abnormal region. The expression is: In the formula, YD represents the roundness of the abnormal region, S represents the area of ​​the abnormal region, and L represents the perimeter of the abnormal region.

5. The machine vision-based method for monitoring the wear condition of cross guide rails according to claim 1, characterized in that, The wear index is calculated as follows: Abnormal regions are mapped onto a grayscale image. An auxiliary region is obtained by extending the edge lines outwards by multiple pixels. The gradient magnitude is then binary-classified using the Otsu thresholding method to obtain two clusters. The mean gradient magnitude within each cluster is calculated, and the cluster with the larger mean is selected as the first cluster. The expression for the wear index is: Where, The wear index indicates the abnormal area. This represents the preset weighting coefficient. This represents the mean of the gradient magnitudes within the first cluster. The wear characteristic value represents the abnormal area.

6. The machine vision-based method for monitoring the wear condition of cross guide rails according to claim 5, characterized in that, The weighting coefficients are calculated as follows: The within-class variance and between-class variance of the two clusters are calculated, and the expression for the weighting coefficients is: In the formula, This represents the preset weighting coefficient. This represents the between-class variance of two clusters. Let represent the within-cluster variance of two clusters, and tanh represent the hyperbolic tangent function.

7. The machine vision-based method for monitoring the wear condition of cross rails according to claim 2, characterized in that, The labels include light intensity and the vibration frequency of the cross rails.

8. The machine vision-based method for monitoring the wear condition of cross rails according to claim 2, characterized in that, The DBSCAN clustering algorithm was used to cluster the labels in the historical image set to obtain multiple clusters.

Images