A bearing clearance dynamic measurement method and system

Abstract

The present application relates to the field of measurement, more particularly, the present application relates to a kind of bearing play dynamic measurement method and system, method includes: processing bearing gray diagram, obtains ball pixel radius, bearing center coordinates and edge graph;With edge graph as input, in the accumulator space vote candidate circle and its parameters;According to characteristic factor, screen real ball and its parameters;To the continuous frame real ball, carry out annular constraint and angle equidistance verification, construct qualified frame ordered ball distance list;The average of the sum of the maximum value and the minimum value of each real ball distance is calculated to obtain the bearing radial play, and then converted to physical units, i.e. Actual radial clearance size.The present application realizes the accurate measurement of bearing radial clearance by processing bearing gray image, screening candidate circle by multi-modal feature fusion, verifying continuous frame and comprehensively calculating radial clearance, which provides strong technical support for quality evaluation and maintenance of bearing.

Description

Technical Field

[0001] This invention relates to the field of measurement. More specifically, this invention relates to a method and system for dynamically measuring bearing clearance. Background Technology

[0002] Bearing clearance is the gap between the rolling elements of a bearing and the inner and outer rings of the bearing housing. Specifically, bearing clearance refers to the amount of radial or axial movement of the unfixed ring when one of the inner or outer rings of the bearing is fixed, without being installed on a shaft or in a bearing housing. Measuring bearing clearance is crucial for ensuring bearing life and stable equipment operation.

[0003] Existing technology, such as patent application CN112747657A, discloses a bearing clearance detection system and its detection method. In this patent application, when detecting bearing clearance, the bearing body is first placed on a moving block. The second knob is operated to move the moving block between four housings. Then, the first knob is turned to adjust the housings, allowing the pressure roller to press against the outer ring and fix the bearing. At this point, the straight rod is on the bearing's central axis. The bearing is placed vertically, and the inner ring moves downwards due to the clearance. The sliding rod, under the action of the counterweight, presses down on the inner ring. Rotating the outer ring causes it to rotate relative to the pressure roller. The inner ring remains stationary due to the anti-slip texture of the rubber pad. The change in clearance of the outer ring compresses the inner ring, causing the arc-shaped pressure plate to move the moving plate. The first elastic element extends and retracts, and the dial indicator probe accurately measures the displacement of the moving plate to obtain clearance data. Simultaneously, the outer ring, through gear transmission, further restricts the inner ring with a clamping component, ensuring accurate dial indicator measurement.

[0004] During the manual rotation of the bearing outer ring, friction inevitably exists between the outer ring and the pressure roller. This friction causes uneven and uneven rotation of the outer ring, resulting in dynamic deviations in the transmission of changes in the clearance between the outer and inner rings. This deviation further prevents the force on the arc-shaped pressure plate from accurately reflecting the actual clearance changes, ultimately leading to errors in the clearance value measured by the dial indicator. This affects the operator's accurate judgment of the bearing's radial clearance, potentially impacting the accuracy of bearing quality assessments, fault diagnosis, and subsequent maintenance decisions based on these measurement results. Summary of the Invention

[0005] To address the aforementioned technical problem of inaccurate bearing clearance measurement, the present invention provides solutions in the following aspects.

[0006] In the first aspect, a method for dynamically measuring bearing clearance includes:

[0007] The acquired bearing grayscale image is processed to obtain the ball pixel radius, bearing center coordinates, and bearing edge image;

[0008] Using the edge map as input, for each edge point, vote in the accumulator space along the gradient direction within a preset radius range to obtain candidate circles, their centers, and corresponding radii; calculate the feature factors of the candidate circles, and filter the real balls, their centers, and corresponding radii in the candidate circles based on the feature factors;

[0009] The real balls in the continuous frame images are verified by ring constraint and angular equidistance. Qualified frame images are selected and an ordered ball distance list of qualified frames is constructed. Each element in the list represents the distance from the center of a real ball to the center of the bearing.

[0010] For each real ball, find the maximum and minimum distances from the center of the real ball to the center of the bearing in all qualified frames. Calculate the average of the sum of the differences between the maximum and minimum distances for all real balls to obtain the radial clearance of the entire bearing. Convert the radial clearance to physical units to obtain the actual radial clearance size of the bearing.

[0011] Preferably, processing the acquired bearing grayscale image further includes:

[0012] Mark the highlight areas on the grayscale image, repair the highlight areas, and mark the repaired highlight areas as 1 in the mask, while marking other areas as 0.

[0013] Preferably, before calculating the feature factors of the candidate circles, each candidate circle is sampled multiple times at equal intervals on its circumference, with the same total number of sampling points each time.

[0014] Preferably, the feature factors of the candidate circle include: edge integrity, gradient direction symmetry, and specular mask feature factors.

[0015] Preferably, the process of screening candidate circles includes:

[0016] The multimodal feature credibility is obtained by weighted summation of the edge integrity, gradient direction symmetry and specular mask feature factors of the candidate circle. If the multimodal feature credibility of the candidate circle is greater than or equal to the preset screening threshold, the candidate circle is retained; otherwise, the candidate circle is discarded.

[0017] Preferably, the process of obtaining the edge integrity includes:

[0018] Edge points are selected from the sampling points obtained in each sampling. The difference between the total number of sampling points and the average number of edge points obtained in all sampling is taken as the number of missing points. Combined with the number of missing points and the preset size parameters, the edge integrity of the candidate circle is calculated using a Gaussian function.

[0019] Preferably, the process of obtaining the gradient direction symmetry includes:

[0020] Calculate the minimum angle between the gradient direction of the edge point and the vector direction pointing from the center of the candidate circle to the edge point. If the minimum angle is less than the preset angle threshold, the edge point is determined to be an edge point that meets the gradient direction symmetry. Calculate the average of the ratios of the number of edge points that meet the gradient direction symmetry in all samplings to the average number of edge points obtained in all samplings as the gradient direction symmetry of the candidate circle.

[0021] Preferably, the process of obtaining the specular mask feature factors includes:

[0022] If the value of a sampling point on the mask of the restored highlight area is 1, then the sampling point is located in the restored area. Calculate the average number of sampling points located in the restored area for all samplings, and use the ratio of the average number of sampling points located in the restored area to the total number of samples as the highlight ratio.

[0023] The highlight ratio is transformed using an exponential function to obtain the highlight mask feature factor of the candidate circle.

[0024] Preferably, the verification condition for the ring constraint is: calculate the mean and standard deviation of the distances from all real balls to the bearing center; if the standard deviation is greater than a preset first threshold, the ring distribution of real balls in the current frame is determined to be abnormal; the verification condition for the angular equidistantness is: calculate the mean of the angle differences between all adjacent real balls, calculate the sum of the deviations of all angle differences from their mean; if the sum of the deviations of all angle differences from their mean is greater than a preset second threshold, the distribution of real balls in the current frame is determined to be uneven.

[0025] If either the ring constraint verification or the angular isometry verification is determined to be abnormal, the result of the current frame is discarded.

[0026] In a second aspect, a dynamic bearing clearance measurement system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the dynamic bearing clearance measurement method described in any one of the claims is implemented.

[0027] The present invention has the following beneficial effects:

[0028] By processing the grayscale image of the bearing, including marking and repairing highlight areas, the interference of factors such as lighting on image quality is reduced, providing a clearer and more accurate base image for subsequent edge detection and ball recognition. Before calculating the feature factors of the candidate circles, multiple samplings are performed on the circumference of each candidate circle at equal intervals, which can obtain more comprehensive information about the candidate circles. Increasing the number of sampling points helps to analyze the characteristics of the candidate circles in more detail.

[0029] The multimodal feature credibility is obtained by weighted summation of the edge integrity, gradient direction symmetry and specular mask feature factors of the candidate circle. By comprehensively considering multiple factors, it is possible to determine whether the candidate circle is a real ball. Compared with the method of screening based on a single feature, this multimodal feature fusion method can more comprehensively evaluate the authenticity of the candidate circle, greatly improve the accuracy and reliability of screening, and effectively reduce the occurrence of misjudgment.

[0030] By performing ring constraint and angular equidistant verification on the real balls in continuous frame images, qualified frame images can be further screened out, ensuring that the selected ball position information for radial clearance calculation conforms to the actual working state and geometric characteristics of the bearing. This avoids calculation errors caused by factors such as abnormal ball positions in individual frame images, thereby improving the stability of the entire detection process and the reliability of the results, and making the final radial clearance calculation results more accurately reflect the real condition of the bearing.

[0031] The radial clearance of the bearing is obtained by finding the maximum and minimum distances from the center of each real ball to the bearing center in all qualified frames, and calculating the average of the sum of the differences. This is then converted into physical units to obtain the actual radial clearance size. This comprehensive analysis method can fully consider the situation of the ball in different positions and orientations, avoiding the deviation caused by relying on measurements of only a few specific positions. This makes the radial clearance calculation results more accurate and objective, providing strong technical support and data basis for bearing quality assessment, fault diagnosis, and subsequent maintenance. Attached Figure Description

[0032] Figure 1 This is a flowchart of steps S1-S4 in a dynamic measurement method for bearing clearance according to an embodiment of the present invention. Detailed Implementation

[0033] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0034] In this invention, the correlation between radial clearance and axial clearance is considered in practical applications. For full complement ball bearings, axial clearance is typically indirectly influenced and controlled by controlling radial clearance. Under certain operating conditions, the magnitude of axial clearance can be affected by radial clearance; therefore, prioritizing the detection of radial clearance helps to effectively assess the overall clearance condition of the bearing.

[0035] Reference Figure 1 A method for dynamic measurement of bearing clearance includes steps S1-S4, as detailed below:

[0036] S1: Process the acquired bearing grayscale image to obtain the ball pixel radius, bearing center coordinates, and bearing edge image.

[0037] In one embodiment, the diameter of each ball is first measured using a micrometer, and the median of all diameters is taken as the representative value of the ball diameter. The preset calibration scale is k = x pixels / mm (which needs to be calibrated using standard gauge blocks or a precision grid). For example, if 1 mm corresponds to 2.5 pixels, then k = 2.5 pixels / mm. The representative value of the ball diameter is then multiplied by k to obtain the pixel diameter of the ball, and the pixel radius of the ball is calculated. A coordinate system is established by observing the imaging of the markers on the fixture in the image, and the position of the bearing center in this coordinate system is accurately calibrated with an accuracy of ±0.1 pixels.

[0038] Through the above preliminary preparations and parameter calibration, the ball pixel radius and bearing center coordinates are obtained to ensure that all measurements are based on a unified spatial reference, thus avoiding deformation errors caused by viewing angle deviations.

[0039] Furthermore, an industrial camera with a ring light source was used to vertically photograph the surface of the rotating bearing. The acquired RGB images were converted into grayscale images and then Gaussian blurred to reduce noise interference.

[0040] In another embodiment, considering that the metal ball bearings are prone to reflection during imaging, resulting in highlight areas that can affect subsequent processing, areas with grayscale values ​​exceeding, for example, 240 in the grayscale image are marked as highlight areas. Then, image inpainting techniques are used to fill in the highlight areas. During the inpainting process, interpolation or diffusion is performed using information from the normal areas surrounding the highlight areas to restore the continuity of the edges, making the inpainted areas visually more natural and harmonious with the surrounding normal areas. A binary mask of the same size as the original image is created, and the inpainted highlight areas are marked as 1, while the remaining areas are marked as 0.

[0041] In one embodiment, edge detection is performed on the grayscale image to obtain the corresponding edge map.

[0042] After the above series of operations, the ball pixel radius, bearing center coordinates, and bearing edge map are finally obtained.

[0043] S2: Using the edge map as input, for each edge point, vote in the accumulator space along the gradient direction within a preset radius range according to its gradient direction to obtain candidate circles, their centers, and corresponding radii; calculate the feature factors of the candidate circles, and filter the real balls, their centers, and corresponding radii in the candidate circles according to the feature factors.

[0044] The Hough transform is a commonly used method for identifying geometric shapes from images. In the specific application of bearing clearance measurement, the balls exhibit obvious circular characteristics. Using the Hough transform, the parameters corresponding to the circular contour, namely the center coordinates and radius, can be efficiently extracted from complex edge images.

[0045] Furthermore, the edge image obtained after image preprocessing and edge detection may contain a large number of edge points, which may originate from various factors such as ball bearings, background noise, and scratches. Directly analyzing all edge points one by one is computationally intensive and inefficient. The Hough transform, by transforming the problem from image space to parameter space, can quickly identify possible candidate circles, thereby narrowing down the scope of further analysis and verification and improving the overall processing efficiency of the system.

[0046] In one embodiment, taking the edge image obtained in S1 as input, for each edge point, according to its gradient direction, within a preset radius range (the lower limit of the radius can be defined as 0.9×R and the upper limit as 1.1×R according to the process error tolerance, where R is the radius of the ball pixel in S1), voting is performed in the accumulator space along the gradient direction to obtain the candidate circle center and its corresponding radius.

[0047] Further verification of the candidate circles is conducted to filter out the center and radius information corresponding to the actual ball bearings.

[0048] The specific screening process described above is as follows:

[0049] First, based on the characteristic that the edge of a real ball bearing should be continuous, each candidate circle is sampled K times at equal intervals on its circumference. During each sampling, a point is selected every 2 degrees, for a total of 180 points.

[0050] In one embodiment, the coordinates of each sampling point are calculated using the parametric equation of a circle. For each sampling point, it is checked whether the sampling point is within the edge image range to avoid invalid detections that exceed the image boundary. Then, it is also checked whether the point is an edge point on the edge image, that is, whether the pixel value of the point is 255 (assuming that edge points are represented by 255 on the binarized edge image).

[0051] The number of edge points that satisfy both of the above two inspection conditions among all sampling points is counted as the sampling result, denoted as N. After repeating the sampling K times, the mean of all sampling results is calculated.

[0052] The difference between the above sampling number 180 and the mean of the sampling results is taken as the number of missing points. This indicates how many edge points are missing on average in the candidate circle compared to the complete edge.

[0053] It is important to consider that low-noise images have clear and continuous edges, and the number of missing points is relatively small. In this case, a smaller scale parameter can be set to strictly require edge integrity (to improve precision and reduce false positives). For high-noise images, the edges may be broken due to noise, and the number of missing points is relatively large. In this case, a larger scale parameter needs to be set to tolerate more missing points (to improve recall and avoid missing true balls).

[0054] Therefore, combining the number of missing points and the scale parameter, a Gaussian function is used to calculate the edge integrity of the candidate circle. That is, it satisfies the following relationship:

[0055]

[0056] In the formula, For the edge integrity of the candidate circle, The number of missing points in the candidate circle. For scale parameters, This represents an exponential function with the natural constant e as the base. The initial value of the scale parameter is set to 15, which can be adjusted according to actual needs.

[0057] In one embodiment, since the imaging of a real ball against a bright background has specific gradient direction characteristics, the angle between the vector directions from the center of the circle to the edge point is calculated based on the aforementioned edge point to determine whether the gradient direction is symmetrical.

[0058] Ideally, for a bright circle against a dark background, the gradient direction should differ from the vector direction pointing from the center of the candidate circle to the edge by 180 degrees. However, since angles are periodic (ranging from 0 to 360 degrees), the minimum angle between these two directions is calculated, satisfying the following relationship:

[0059]

[0060] In the formula, Let be the minimum angle between the gradient direction at the edge point and the vector direction pointing from the center of the candidate circle to the edge point. The gradient direction at the edge point. Let be the angle between the vector directions pointing from the center of the candidate circle to the edge point.

[0061] If this minimum included angle If the angle is less than the set threshold (e.g., 10 degrees), the gradient direction of the edge point is considered to be symmetrical.

[0062] For K repeated samplings, count the number of edge points that satisfy gradient direction symmetry in each sampling. Calculate the average of the ratios of the number of edge points satisfying gradient direction symmetry in the K repeated samplings to the above sampling results. This average is used as the gradient direction symmetry of the candidate circle, satisfying the following relationship:

[0063]

[0064] In the formula, For the gradient direction symmetry of the candidate circle, Let be the number of edge points that satisfy the gradient direction symmetry under any number of samplings. The number of edge points in any number of samplings. This represents the total number of samples taken.

[0065] Gradient direction symmetry, by examining whether the gradient direction of the edge points of the candidate circle matches the optical characteristics of a bright circle against a dark background, helps to further screen out candidate circles that are more likely to be real balls, thus improving the accuracy of the measurement.

[0066] It should be noted that the above It could be 0, and the appropriate processing method needs to be selected according to the specific situation, such as skipping the sampling, setting a default value, using smoothing processing, increasing the number of samplings, etc.

[0067] In one embodiment, since highlight areas are repaired during image preprocessing, these repaired areas may contain some missing information or errors. By checking whether sampling points on the circumference of the candidate circle are located within the highlight repaired areas and calculating the highlight ratio, a highlight mask feature factor is constructed to assess the degree to which the candidate circle is affected by repair errors. This indirectly reflects the reliability of the candidate circle's edges, because excessive highlight repair areas may lead to inaccurate edge information, thus affecting the judgment of the candidate circle's authenticity.

[0068] Based on each of the above sampling points, check the value of that sampling point on the highlight restoration mask. If the mask value is 1, it indicates that the sampling point is located in the restoration area. For K repeated samplings, calculate the mean number of sampling points located in the highlight restoration area, i.e., the average number of sampling points located in the highlight restoration area.

[0069] Further calculate the highlight ratio, which is the ratio of the mean number of sampling points in the highlight restoration area to the total number of sampling points (i.e., 180).

[0070] To assess the extent to which the candidate circle is affected by highlight restoration errors, the aforementioned highlight ratio is mapped to a range of 0 to 1 using an exponential function, satisfying the following relationship:

[0071]

[0072] In the formula, The specular mask feature factor of the candidate circle. For highlight ratio, This represents an exponential function with the natural constant e as its base.

[0073] In summary, the edge integrity, gradient direction symmetry, and specular mask feature factors of the candidate circle are considered in relation to its edge. A comprehensive assessment of whether a candidate circle represents a true ball bearing circle is made from various dimensions, including edge integrity, gradient direction rationality, and the degree of influence from specular restoration. Verification using these edge features effectively improves the accuracy of identifying true balls, reduces false positives, and provides a reliable data foundation for subsequent clearance measurements.

[0074] Therefore, the multimodal feature confidence is obtained by weighted summing of the candidate circle's edge integrity, gradient direction symmetry, and specular mask feature factors, which satisfies the following relationship:

[0075]

[0076] In the formula, To assess the credibility of the multimodal features of candidate circles, , and These are the weight coefficients for the candidate circle's edge integrity, gradient direction symmetry, and specular mask feature factor, respectively, and the sum of these three weight coefficients is 1. For example, weighting is assigned based on the importance of each feature. It is 0.6. It is 0.3. It is 0.1.

[0077] Finally, a screening threshold is set. If the confidence level of the multimodal features of a candidate circle is greater than or equal to the screening threshold, the candidate circle is considered to have high authenticity and is retained; otherwise, the candidate circle is discarded.

[0078] Based on the above operations, the actual balls in all candidate circles, as well as the center and corresponding radius of the actual balls, are finally obtained.

[0079] Since more precise sub-pixel accuracy is required in practical applications, sub-pixel refinement technology is used to refine the center coordinates to obtain the final center and radius of the ball.

[0080] S3: Verify the circular constraint and angular equidistance of the real balls in the continuous frame images, filter the qualified frame images and construct an ordered list of ball distances for the qualified frames. Each element in the list represents the distance from the center of a real ball to the center of the bearing.

[0081] In the Hough transform and candidate circle verification stages of S2 mentioned above, the detection and verification are mainly performed on individual balls. However, each stage of detection may have certain errors or random factors that lead to false detections. Although the candidate circle verification stage uses edge integrity, gradient direction symmetry, and specular mask feature factors for screening, there may still be some special cases that allow erroneous candidate circles to pass verification.

[0082] Therefore, considering the overall ball distribution, there is a specific relative positional relationship between the balls, that is, they are evenly distributed on a circumference with the bearing center as the center. By checking whether all the balls meet the ring constraint and angular equidistant distribution, we can avoid misjudging some isolated objects that meet the characteristics of some balls as balls.

[0083] In one embodiment, the above-described operation for performing ring constraint verification is as follows:

[0084] In the current frame image, calculate the mean and standard deviation of the distances from all real balls to the bearing center. Verification condition: if the standard deviation is greater than the preset first threshold (set to 3 in this embodiment), it is determined that the ring distribution of real balls in the current frame is abnormal; otherwise, it indicates that the distribution of real balls is normal.

[0085] In one embodiment, the above-described operation for verifying angular equidistance is as follows:

[0086] First, for a full complement ball bearing, the balls should be evenly distributed at an angle on the circumference. Assuming the total number of balls is Y, the theoretical angle interval is 360 degrees divided by Y.

[0087] For each real ball, calculate its angle relative to the bearing center (obtained in S1), and for a single real ball, calculate the angle difference between it and the next ball, i.e.: In the formula, For the first The angle difference between one real ball and its next ball For the first The angle of a real ball relative to the center of the bearing. For the first The angle of a real ball relative to the center of the bearing.

[0088] The angle difference between the last and first real balls is calculated using the following formula: Since the angle is circular, the formula is as follows: In the formula, For the first The angle difference between each real ball and the first real ball. For the first The angle of a real ball relative to the center of the bearing.

[0089] Calculate the mean of the angle differences of all real balls, and calculate the sum of the deviations of all angle differences from their mean.

[0090] The verification condition for angular equidistance is as follows: if the sum of the deviations of all angle differences from their mean is greater than the second threshold (set to 5 in this embodiment of the invention), it indicates that the distribution of the real balls in the current frame is not uniform; otherwise, it indicates that the distribution of the real balls in the current frame conforms to angular equidistance.

[0091] Finally, if either the ring constraint verification or the angular equidistant verification is deemed abnormal, the result of the current frame is discarded, and the next frame's data is used for re-detection. If both verifications pass, the current frame is considered a qualified frame. The distances from all actual balls to the bearing center in the qualified frame are calculated, and an ordered ball distance sequence is constructed according to the angle.

[0092] S4: For each real ball, find the maximum and minimum distances from the center of the real ball to the center of the bearing in all qualified frames. Calculate the average of the sum of the differences between the maximum and minimum distances corresponding to all real balls to obtain the radial clearance of the entire bearing. Convert the radial clearance to physical units to obtain the actual radial clearance size of the bearing.

[0093] Based on the above S3, qualified frames are selected from multiple frames of images. For each real ball, the maximum and minimum distances from the center of the real ball to the center of the bearing are found in all qualified frames. Then, the average of the sum of the differences between the maximum and minimum distances corresponding to all real balls is calculated and divided by the number of real balls to obtain the radial clearance of the entire bearing (the unit is sub-pixel).

[0094] Based on the known actual physical length represented by each pixel (e.g., determined through calibration), the radial clearance of the sub-pixel units obtained above is converted into actual physical unit values, thereby obtaining the actual radial clearance size of the bearing.

[0095] This completes the dynamic measurement of the bearing radial clearance, and subsequent analysis can be performed based on the measured radial clearance.

[0096] The system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the bearing clearance dynamic measurement method according to the first aspect of the present invention.

[0097] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0098] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for dynamic measurement of bearing clearance, characterized in that, include: The acquired bearing grayscale image is processed to obtain the ball pixel radius, bearing center coordinates, and bearing edge image; Using the edge map as input, for each edge point, voting is performed in the accumulator space along the gradient direction within a preset radius range based on its gradient direction to obtain candidate circles, their centers, and corresponding radii. Feature factors of the candidate circles are calculated, including edge integrity, gradient direction symmetry, and specular mask feature factors. The edge integrity, gradient direction symmetry, and specular mask feature factors are weighted and summed to obtain the multimodal feature confidence level. If the multimodal feature confidence level of the candidate circle is greater than or equal to a preset screening threshold, the candidate circle is retained; otherwise, it is discarded, resulting in the actual ball within the candidate circle, its center, and its corresponding radius. The process of obtaining the edge integrity includes: Edge points are selected from the sampling points obtained in each sampling. The difference between the total number of sampling points and the average number of edge points obtained in all sampling is taken as the number of missing points. Combining the number of missing points and the preset size parameters, the edge integrity of the candidate circle is calculated using a Gaussian function. The process of obtaining the gradient direction symmetry includes: Calculate the minimum angle between the gradient direction of the edge point and the vector direction pointing from the center of the candidate circle to the edge point. If the minimum angle is less than the preset angle threshold, the edge point is determined to be an edge point that meets the gradient direction symmetry. Calculate the average of the ratios of the number of edge points that meet the gradient direction symmetry in all samplings to the average number of edge points obtained in all samplings as the gradient direction symmetry of the candidate circle. The process of obtaining the feature factors of the specular mask includes: If the value of a sampling point on the mask of the restored highlight area is 1, then the sampling point is located in the restored area. Calculate the average number of sampling points located in the restored area for all samplings, and use the ratio of the average number of sampling points located in the restored area to the total number of samples as the highlight ratio. The highlight ratio is transformed using an exponential function to obtain the highlight mask feature factor of the candidate circle; The real balls in the continuous frame images are verified by ring constraint and angular equidistance. Qualified frame images are selected and an ordered ball distance list of qualified frames is constructed. Each element in the list represents the distance from the center of a real ball to the center of the bearing. For each real ball, find the maximum and minimum distances from the center of the real ball to the center of the bearing in all qualified frames. Calculate the average of the sum of the differences between the maximum and minimum distances for all real balls to obtain the radial clearance of the entire bearing. Convert the radial clearance to physical units to obtain the actual radial clearance size of the bearing.

2. The method for dynamic measurement of bearing clearance according to claim 1, characterized in that, Processing the acquired bearing grayscale image also includes: Mark the highlight areas on the grayscale image, repair the highlight areas, and mark the repaired highlight areas as 1 in the mask, while marking other areas as 0.

3. The method for dynamic measurement of bearing clearance according to claim 2, characterized in that, Before calculating the eigenfactors of the candidate circles, each candidate circle is sampled multiple times at equal intervals on its circumference, with the same total number of sampling points each time.

4. The method for dynamic measurement of bearing clearance according to claim 1, characterized in that, The verification condition for the ring constraint is: calculate the mean and standard deviation of the distances from all real balls to the bearing center. If the standard deviation is greater than the preset first threshold, the ring distribution of the real balls in the current frame is determined to be abnormal. The verification condition for the angular equidistance is as follows: calculate the mean of the angle difference between all adjacent real balls, calculate the sum of the deviations of all angle differences from their mean, and if the sum of the deviations of all angle differences from their mean is greater than a preset second threshold, it is determined that the distribution of real balls in the current frame is uneven. If either the ring constraint verification or the angular isometry verification is determined to be abnormal, the result of the current frame is discarded.

5. A dynamic measurement system for bearing clearance, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the dynamic measurement method for bearing clearance according to any one of claims 1-4.

Images