A method for detecting surface quality of shaft seal components

Abstract

The present invention belongs to the field of image processing technology, and specifically relates to a method for inspecting the surface quality of shaft seal components. The method comprises: performing edge detection on a collected surface image of the shaft seal component, performing region growing, locating and extracting circular edges, and recording the remaining connected domains as feature regions; obtaining thermal crack regions by calculating feature credibility; fitting a thermal crack trend line, using the length and angle mean of the feature region as parameters, and constructing a sliding window along the trend line in a clockwise direction with the mean of the angles between adjacent thermal cracks as a step size; predicting the window size by combining the window's feature parameters with the thermal crack parameters, and then calculating the window's potential thermal crack index based on the window's pixel distribution and parameter changes; and marking cracks and potential crack regions. The present invention improves the efficiency of shaft seal component surface quality inspection and enables the prediction of potential cracks.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and more particularly to a method for detecting the surface quality of a shaft seal assembly. Background Art

[0002] The shaft seal assembly is a sealing device assembly installed between the rotating shaft and the equipment housing. Its core function is to prevent the fluid inside the equipment from leaking along the gap between the rotating shaft and the housing, while preventing external impurities from invading the interior of the equipment. In industrial scenarios, the shrinkage stress generated by the material during high-temperature cooling and the segregation of the low-melting-point eutectic phase will cause thermal cracks on the surface of the shaft seal assembly. Thermal cracks will directly destroy the structural integrity of the packaging assembly, resulting in a decrease in the sealing performance of the assembly. At the same time, if the surface defects of thermal cracks are not discovered in time, it may cause the shaft seal assembly to break, causing equipment shutdown or even mechanical failure. In order to reduce the failure problems caused by the shaft seal assembly, it is necessary to perform quality inspection on the surface of the shaft seal assembly.

[0003] Image processing technology is currently commonly used to improve the efficiency of detecting thermal cracks on the surface of shaft seal components. For example, Chinese patent application CN114972187B discloses an artificial intelligence-based crack defect assessment method. This method analyzes newly emerged crack intersections in adjacent crack defect images and calculates the grayscale entropy of the detection area as a roughness measure. A greater grayscale entropy indicates a more uneven grayscale distribution in the detection area, corresponding to more dramatic dynamic changes in crack growth. This method combines static image analysis with dynamic time series information to track the real-time status of crack growth.

[0004] However, the AI-based crack defect assessment method proposed in CN114972187B suffers from problems such as a limited detection range, a lack of predictive capabilities, and limited universality when inspecting shaft seal components. For example, surface quality inspection of shaft seal components requires the integration of multiple image features, including edges, regional shapes, and pixel distribution. Focusing solely on crack intersections and grayscale entropy to assess crack growth makes it difficult to fully detect thermal crack areas and potential crack areas. Predicting potential crack areas allows for timely maintenance measures before thermal cracks further expand, leading to component seal failure, impacting equipment heat dissipation efficiency, or causing leaks, thereby improving the reliability and safety of shaft seal components and related equipment. Summary of the Invention

[0005] In order to solve the problems of the above-mentioned existing methods in detecting shaft seal components, such as limited detection range, lack of predictive ability and limited universality, the present invention provides a method for detecting the surface quality of shaft seal components, comprising:

[0006] After edge detection is performed on the collected surface image of the shaft seal component, region growing is performed. The circular edge is located and extracted according to the annular distribution characteristics of the connected domain, and the remaining connected domain is recorded as the feature region; the polar center is obtained according to the distribution characteristics of the circular edge, and a polar coordinate system is constructed; an arc is fitted according to the trend of the line connecting the center points of the feature region, and the first reliability of the feature is calculated by comparing the arc similarity between the fitted arc and the circular edge; the length and angle of the feature region are calculated according to the distance between the pixel point of the feature region and the polar center, and the angle with the baseline of the polar coordinate system; the final reliability of the feature is calculated according to the difference between the length and angle of the feature region and the distribution regularity of the pixel points of the feature region in the polar coordinate system. Reliability, a first threshold is preset to screen the final reliability of the feature to obtain the thermal crack area; according to the distribution of all thermal crack areas and the mean value of the characteristic area length, the fitting crack trend line and the thermal crack characteristic parameters are obtained respectively; the thermal crack characteristic parameters are used as the preset window parameters, and a sliding window is established along the fitting thermal crack trend line in a clockwise direction with the mean value of the angles between adjacent thermal cracks as the step length; the sliding window size is predicted in turn based on the characteristic parameters of the sliding window and the thermal crack parameters of the thermal crack area; the window is set according to the window size, and the window potential thermal crack index is calculated based on the pixel distribution characteristics and the degree of change of the thermal crack parameters in all windows, and the cracks and potential crack areas are marked.

[0007] Existing methods for inspecting these components are limited in scope, only identifying obvious cracks but unable to predict potential problems. They are also not well-suited to diverse situations. This method comprehensively analyzes surface features, incorporating information such as shape and distribution. This method covers the entire component surface, accurately identifying existing cracks and predicting areas where cracks are likely to develop. Furthermore, by adjusting the size and position of the inspection window to accommodate cracks of varying sizes and distributions, it overcomes the issues of limited scope, lack of predictive power, and poor adaptability, resulting in more comprehensive and accurate inspections.

[0008] Preferably, constructing a polar coordinate system includes:

[0009] The RANSAC arc fitting algorithm is used for all circular edges to obtain the fitting center corresponding to each circular edge, and multiple target center points are obtained; for multiple target center points, the sum of squared residuals from the original edge point to each target center point is calculated, and the target center point corresponding to the smallest residual value is selected as the polar center. The ray starting from the polar center and pointing horizontally to the right along the horizontal axis of the target image is recorded as the baseline, and a polar coordinate system is constructed in the target image.

[0010] This method of determining the center and establishing a measurement scheme uses multiple calculations to find the most optimal center position, reducing the impact of erroneous points in edge detection. Using the center as a reference and the horizontal direction as a reference, it allows for a clear positional description of various features on a circular surface, avoiding misjudgments due to angular deviations and enabling more accurate subsequent analysis of surface features.

[0011] Preferably, the first credibility of the feature satisfies the following expression:

[0012] ;

[0013] Where, Indicates the first credibility of the feature, Indicates the minimum cumulative distance corresponding to the fitted arc where the feature area is located; Represents the minimum cumulative distance mean corresponding to all fitted arcs on the target image; It is a minimum positive value, ensuring that the denominator is not 0; represents the absolute value function; represents the normalization function; Represents an exponential function with a natural constant as its base.

[0014] Preferably, calculating the final credibility of the feature includes:

[0015] Calculate the second credibility of the feature; the final credibility of the feature satisfies the following expression:

[0016] ;

[0017] Where, Indicates the final credibility of the feature; g indicates the total number of feature region clusters; Represents the number of pixels in the i-th cluster; Represents the average number of all pixels on the same arc in the feature area; Represents the average distance between adjacent pixels in the i-th cluster; Indicates the second credibility of the feature; represents the absolute value function; Represents an exponential function with a natural constant as its base.

[0018] This refined approach, based on calculating the deviation of features from the average in length and angle, combined with previous judgment results, can more accurately identify areas that match crack characteristics. This allows features that are close to real cracks in size and distribution to stand out more clearly.

[0019] Preferably, calculating the second credibility of the feature includes:

[0020] According to the distance from the pixel points in the feature area to the center of the circle, the first end point and the end point are located, and the length of the feature area and the mean length of the feature area are calculated according to the distance from the first end point and the end point to the polar center; all the pixel points in the feature area are connected to the polar center respectively, and these connecting lines are recorded as the pixel point line segments of the feature area, and the angle of the feature area and the mean angle of the feature area are calculated according to the relationship between the pixel point line segments and the baseline; the difference between the length and angle of the feature area and the mean length and angle of the feature area are calculated respectively, as well as the ratio of the difference to the length and angle of all feature areas, and the second credibility of the feature is calculated in combination with the first credibility of the feature.

[0021] Preferably, obtaining the thermal crack region comprises:

[0022] The final credibility of features of all feature regions is calculated, a first threshold is preset, feature regions greater than the first threshold are recorded as thermal crack regions, and the regions are extracted to obtain thermal crack regions.

[0023] Preferably, respectively obtaining the fitted crack trend line and the thermal crack characteristic parameters includes:

[0024] The center points of all thermal crack areas were connected, and arc fitting was performed using the least squares method to obtain a fitting curve, which was recorded as the crack trend line; the midpoints of two adjacent cracks were connected with the polar center respectively, and the angle between the two connecting lines was calculated, which was recorded as the adjacent crack angle, and the mean of the adjacent crack angles of all adjacent cracks was calculated; the maximum crack area length and the maximum crack area angle of all thermal crack areas were extracted as the length and angle range of the thermal crack; the mean of the adjacent crack angle and the length and angle range of the thermal crack were recorded as the thermal crack characteristic parameters, and the fitting crack trend line and thermal crack characteristic parameters were obtained respectively.

[0025] Preferably, establishing a sliding window includes:

[0026] A sliding window is established on the fitted crack trend line in a clockwise direction with the average of the adjacent crack angles as the step length until it reaches the boundary of the target image and stops, so that the sliding window satisfies the conditions that the window length is the maximum length of the crack area, the window angle is the maximum angle of the crack area, and the step length of the sliding window is the average of the adjacent crack angles.

[0027] Preferably, the prediction sliding window size satisfies the following expression:

[0028] ;

[0029] Where, Indicates the prediction window length of the i-th sliding window; in indicates the in-th sliding window; Indicates the prediction window length of the in-th sliding window; represents the prediction window angle of the i-th sliding window; Indicates the prediction window angle of the in-th sliding window.

[0030] This method of adjusting the window size based on historical conditions, combined with the previous window size to predict the current window size, allows the window to better fit the actual crack situation. This prevents the inclusion of excessive irrelevant areas due to an overly large window, nor does it prevent partial cracks from being missed due to a too small window, thus improving the adaptability of detection.

[0031] Preferably, the window potential thermal crack index satisfies the following expression:

[0032] ;

[0033] Where, represents the potential thermal crack index of the i-th sliding window; Represents the number of all pixels in the feature area in the i-th sliding window; 、 Represents the distance between the nth white pixel in the feature area of ​​the i-th sliding window and the center pixel and the center of the window; 、 Represents the average distance between all white pixels in the i-th sliding window and the center pixel and the center of the window; represents the potential thermal crack index of the in-th sliding window; It is a minimum positive value, ensuring that the denominator is not 0; represents the absolute value function; Represents the normalization function.

[0034] This method of determining the likelihood of a potential crack combines the distribution of pixels within a window with the judgment results of the previous window to comprehensively assess whether each area has a hidden crack. The more consistent the pixel distribution is with crack characteristics and the more similar it is to previous suspicious areas, the higher the likelihood, helping to detect potential cracks in advance.

[0035] The present invention effectively detects cracks and potential problems on component surfaces. Undetected thermal cracks on packaged components can lead to leakage, downtime, or even equipment failure. By promptly detecting cracks and potential problems, maintenance measures can be implemented proactively, reducing the risk of equipment damage and lowering repair costs. This also ensures proper component operation and enhances the safety and stability of equipment operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart schematically illustrating a method for detecting the surface quality of a shaft seal assembly in the present invention;

[0037] Figure 2 is a grayscale image schematically showing a shaft seal assembly. DETAILED DESCRIPTION

[0038] The embodiment of the present invention discloses a method for detecting the surface quality of a shaft seal assembly, referring to Figure 1 , including steps S1 to S4:

[0039] S1: After edge detection is performed on the collected surface image of the shaft seal component, region growing is performed to locate and extract the circular edge based on the annular distribution characteristics of the connected domain, and the remaining connected domain is recorded as the feature region.

[0040] It's important to note that shaft seals, as core sealing components of rotating equipment, can cause surface cracks that can directly lead to leakage or equipment failure, necessitating high-precision surface defect detection. Shaft seal cracks can be distributed along a circumference, and the projected position of the cracks can easily vary when photographed from multiple angles, leading to misjudgments or missed detections with traditional rectangular coordinate-based detection methods. This step extracts white pixels and generates connected domains, first using a circular Hough transform to separate the circular edges of the shaft seal from the characteristic regions of the target image.

[0041] Specifically, the shaft seal component image is acquired under sufficient lighting conditions, and the acquired shaft seal image is preprocessed, such as Figure 2 : A grayscale image of the shaft seal assembly; edge detection is performed on the grayscale image of the shaft seal assembly, and the resulting image is recorded as the target image; region growing is performed on the pixels in the target image, and the circular Hough transform algorithm is applied to all connected domains obtained and extracted, which are recorded as circular edges; the remaining connected domains are recorded as feature regions. It should be noted that the circular Hough transform maps edge points in image space to parameter space.

[0042] At this point, the feature area of ​​the target image is obtained.

[0043] S2: The polar center is obtained through the circular edge to construct a polar coordinate system. The arc is fitted according to the trend of the line connecting the center points of the feature area, and its arc similarity with the circular edge is calculated to calculate the first credibility of the feature. The length and angle are calculated by combining the distance between the pixel point of the feature area and the polar center and the angle with the baseline of the polar coordinate system. The final credibility of the feature is calculated based on the difference in length and angle and the regularity of the pixel distribution. The thermal crack area is obtained after screening by the first threshold.

[0044] It should be noted that in the edge detection diagram of the package component, a large number of annular package edge areas will be detected, which appear as white annular feature areas on the target image. These areas can obtain multiple circular edges through circular fitting. Each circular edge corresponds to a fitting circle center. By screening multiple circle centers and determining the final polar center, the constructed polar coordinate system can be adapted to all circular edges and feature areas.

[0045] Specifically, the RANSAC arc fitting algorithm is used for all circular edges to obtain the fitting center corresponding to each circular edge, and multiple target centers are obtained; for multiple target centers, the sum of squared residuals from the original edge point to each target center is calculated, and the target center corresponding to the smallest residual value is selected as the polar center. The ray starting from the polar center and pointing horizontally to the right along the horizontal axis of the target image is recorded as the baseline, and a polar coordinate system is constructed in the target image; it should be noted that the RANSAC arc fitting algorithm can effectively eliminate false detection points in edge detection, select the optimal center through residual analysis, improve the accuracy of polar coordinate system construction, and adapt to the full circumferential crack detection needs of shaft seals. And this process of determining the center and establishing a reference method can make each position on the circular surface have a clear correspondence. Through multiple calculations and comparisons, the center that fits the edge best is selected, which reduces the influence of cluttered points on the circular edge and makes the center position more accurate.

[0046] It should be noted that the surface of the shaft seal assembly is a circular structure, and the growth and distribution of thermal cracks on its surface often have a tendency to expand in the circumferential direction, that is, the overall direction of the crack has a certain curvature similarity with the circular edge of the shaft seal. This step can effectively discover the regular characteristics of cracks by identifying key feature areas and studying their distribution characteristics. That is, the DTW algorithm is used to calculate the curvature similarity between the fitted arc of the feature area and the circular edge, and combined with the average similarity of all fitted arcs, it is quantified whether the shape of the feature area conforms to the circumferential distribution trend of thermal cracks. It can preliminarily exclude non-crack areas with large curvature differences from the circular edge, reducing interference factors in subsequent calculations.

[0047] Preferably, feature regions are extracted, minimum circumscribed rectangles are generated for all feature regions, and the central pixel points of the feature regions are located and recorded as feature center points; based on the distribution positions of all feature center points, arc fitting is performed using the least squares method to obtain multiple fitted arcs; the similarity between the multiple fitted arcs and the circular edge is calculated using the DTW algorithm to obtain the minimum cumulative distance corresponding to each fitted arc, and the average of the minimum cumulative distances corresponding to all fitted arcs on the target image is calculated; the minimum cumulative distance corresponding to each fitted arc is used; the credibility of each fitted arc formed by fitting the central pixel points of multiple feature regions and connected by the thermal crack region is calculated, which is recorded as the first credibility of the feature; the first credibility of the feature satisfies the following expression:

[0048] ;

[0049] Where, Indicates the first credibility of the feature; Indicates the minimum cumulative distance corresponding to the fitted arc where the feature area is located; Represents the minimum cumulative distance mean corresponding to all fitted arcs on the target image; It is a minimum positive value, ensuring that the denominator is not 0; represents the absolute value function; represents the normalization function; Represents an exponential function with a natural constant as its base.

[0050] Where, Indicates the normalization of the minimum cumulative distance corresponding to the fitting arc where the feature area is located; The smaller it is, the higher the first credibility of the feature; It represents the normalization of the deviation between the minimum cumulative distance corresponding to the fitted arc where the feature area is located and the mean of the minimum cumulative distances corresponding to all fitted arcs on the target image. It represents the difference between the minimum cumulative distance corresponding to the fitted arc where the feature area is located and the mean of the minimum cumulative distances corresponding to all fitted arcs on the target image. The smaller it is, the smaller the deviation between the minimum cumulative distance corresponding to the fitting arc where the feature area is located and the mean of the minimum cumulative distances corresponding to all fitting arcs on the target image is, and the greater the first credibility of the feature is; The product of the distance to the fitting arc where the feature area is located and the degree of distance deviation quantifies the first credibility of the feature. The larger the value, the greater the first credibility of the feature.

[0051] At this point, the first credibility of the feature is obtained.

[0052] It should be noted that thermal crack growth on the shaft seal surface is directional, typically expanding simultaneously in the radial and circumferential directions. The target area length and angle correspond to the extent of expansion in these two directions, respectively. By calculating the relative deviation between the characteristic area length and angle and the corresponding average of the characteristic area length and angle, we can effectively screen out areas that meet the thermal crack size distribution pattern and reduce interference from non-crack defects such as surface depressions.

[0053] Preferably, extract any feature area, calculate the distance from all pixels in this feature area to the polar center, record the point with the largest distance from the pixel in the feature area to the polar center as the end point, and record the point with the smallest distance from the pixel in the feature area to the polar center as the starting point; the distance from the starting point to the polar center is recorded as the starting distance, and the distance from the end point to the polar center is recorded as the ending distance; calculate the difference between the starting distance and the ending distance, and record it as the feature area length; calculate the length of all feature areas, and find the average, and record it as the average feature area length. It should be noted that the feature area length reflects the radial extension range of the crack, and the difference in distance between the starting and ending points can quantify the radial size characteristics of the crack. This method is more advantageous for the shaft seal annular structure, and avoids the distortion of the length calculation caused by angle deviation in the rectangular coordinate system.

[0054] It should be noted that the formation and expansion of thermal cracks on the shaft seal assembly, an annular sealing component, exhibit significant directional characteristics, typically developing simultaneously along the radial and circumferential polar centers. This requires the crack size characteristics to be described by both radial length and circumferential angle. This step specifically analyzes the bidirectional expansion characteristics of the shaft seal thermal crack by calculating the second credibility. Specifically, this analysis uses the deviation between the length of the characteristic region and the mean length, i.e., radial expansion, and the angular deviation analysis in the polar coordinate system, i.e., circumferential expansion. Combined with the first credibility as a basic multiplier, this allows for a multi-dimensional assessment of arc similarity and the length and angle of the characteristic region, more accurately capturing the typical characteristics of thermal cracks on the shaft seal surface.

[0055] Preferably, all pixel points in the feature area are connected to the pole center respectively, and these lines are recorded as feature area pixel line segments. The angles between all pixel line segments and the baseline are calculated counterclockwise along the baseline of the polar coordinate system, and are recorded as pixel line angles; the difference between the maximum pixel line angle and the minimum pixel line angle in the feature area is calculated, and recorded as the feature area angle; all feature area angles are calculated and averaged, and recorded as the feature area angle mean; the feature area length and angle mean are used, and combined with the feature first credibility to calculate the second credibility that the feature area is a thermal crack, and recorded as the feature second credibility.

[0056] The second reliability of the feature satisfies the following expression:

[0057] ;

[0058] Where, Indicates the second credibility of the feature; The length of the characteristic region representing the characteristic region, Indicates the mean length of the feature area in the target image; The feature area angle representing the feature area, Indicates the mean angle of the feature area in the target image; Indicates the confidence that the feature region is a feature region; represents the absolute value function; represents the normalization function; Represents an exponential function with a natural constant as its base.

[0059] Where, The difference between the length of the characteristic region and the mean length of the characteristic region; It represents the difference between the feature region length and the mean feature region length, and its proportion in the total length of all feature regions. The smaller it is, the greater the second reliability of the feature; The difference between the feature area angle and the mean feature area angle of the feature area; Indicates the difference between the feature area angle of the feature area and the mean feature area angle, and its proportion in the total angle of all feature areas. The smaller it is, the greater the second reliability of the feature; The second reliability of the feature is expressed by combining the relative deviations of the length and angle of the feature area. It should be noted that the angle calculation method in the polar coordinate system can avoid the angle quantification deviation caused by the circumferential distribution of the crack in the rectangular coordinate system. The maximum and minimum difference of the angle of the feature area can accurately reflect the extension range of the crack along the circumference. Combined with the deviation analysis of the mean length and mean angle, the statistical characteristics of the thermal crack in size and distribution can be highlighted. Then, the reliability of the feature area is integrated to improve the specificity of thermal crack identification.

[0060] At this point, the second reliability of the feature is obtained.

[0061] It should be noted that thermal cracks on the surface of the shaft seal assembly often show multi-arc features in radial distribution, while non-crack defects such as scratches and oxidation spots that may exist on the surface often have a chaotic pixel distribution. Although the second credibility of the feature in the previous step has been screened by length, angle and arc similarity, it does not involve the distribution uniformity and density of pixels in different radial arcs, and may misjudge discrete noise as cracks. This step uses the DBSCAN clustering algorithm to group pixels by segment length and calculate the final credibility of the feature. It can accurately distinguish the pixel distribution of thermal cracks along different radial arcs. Compared with single-dimensional judgment, it can effectively eliminate interference from non-crack defects, provide more comprehensive feature support for the final credibility, improve the specificity of thermal crack identification, and provide a reference basis for subsequent potential crack prediction.

[0062] Preferably, the length of the pixel line segment in the feature area is extracted and recorded as the pixel line segment length; the length of all pixel line segments in a feature area is recorded as a pixel length set; the pixels in this pixel length set are clustered according to the length data size, and each pixel whose pixel length difference is within 2 pixels is clustered using the DBSCAN clustering algorithm, and it is stipulated that the number of pixels in each cluster is greater than or equal to 3, and multiple pixel clusters are obtained; the number of pixels in each cluster is extracted and recorded as the number of pixels in the same arc, and the mean number of all pixels in the same arc in the feature area is calculated; the average distance between adjacent pixels in these clusters is calculated; and the final credibility of the feature is calculated.

[0063] The final credibility of the feature satisfies the following expression:

[0064] ;

[0065] Where, Indicates the final credibility of the feature; g indicates the total number of feature region clusters; Represents the number of pixels in the i-th cluster; Represents the average number of all pixels on the same arc in the feature area; Represents the average distance between adjacent pixels in the i-th cluster; Indicates the second credibility of the feature; represents the absolute value function; Represents an exponential function with a natural constant as its base.

[0066] Where, Represents the difference between the number of pixels in the i-th cluster in the feature area and the mean number of pixels in the same arc in the feature area; It represents the sum of the difference between the number of pixels in all clusters in g feature regions and the mean number of pixels in the same arc in the feature regions; Represents the total number of pixels in all clusters in g feature regions; Indicates the difference in the number of pixels in each cluster in the feature area. The smaller it is, the more the number of pixels in each cluster in the feature area is approximately equal, indicating that the arc length of this feature area on the same arc is more regular, and the greater the final credibility of the feature; Represents the sum of the average distances between adjacent pixels in all clusters in the feature area, recorded as the sum of the average distances between adjacent pixels; It represents the ratio of the average distance between adjacent pixels in the i-th cluster in the feature area to the sum of the average distances of adjacent pixels, which is recorded as the ratio of the average distance between adjacent pixels to the sum of the average distances; It represents the sum of the ratio of the average distance between adjacent pixels to the sum of the average distances, and is used to describe the distance between adjacent pixels in the entire feature area. The smaller it is, the denser the pixels in the feature area are, and the greater the final credibility of the feature. The final credibility of the feature region is expressed by the product of three indicators: the uniformity of the number of cluster pixels, the distance between adjacent pixels, and the credibility of the feature region itself.

[0067] It should be noted that the DBSCAN clustering algorithm groups pixels by line segment length, without specifying the number of clusters. This algorithm can distinguish cracks with different radial directions and pixel distributions within the same arc segment. Pixel uniformity reflects the crack's continuity and regularity along the radial direction, while the distance between adjacent pixels reflects the density of crack pixels. Combining pixel uniformity with the distance between adjacent pixels effectively distinguishes true thermal cracks from discrete noise points, improving the reliability of the feature's ultimate credibility.

[0068] At this point, the final credibility of the feature is obtained.

[0069] Preferably, the final credibility of the features of all feature regions is calculated, a first threshold is preset, and feature regions greater than the first threshold are recorded as thermal crack regions and extracted. It should be noted that the first threshold is set by the implementer based on the actual implementation situation. For example, in the Chinese patent document with authorization announcement number CN115423816B, entitled "Method for Detecting Metal Surface Grinding Quality", the preset threshold is 0.55. In actual industrial production, in addition to real thermal cracks, the shaft seal surface often also has various interference factors such as processing marks and stains. By calculating the final credibility of the features, a comprehensive multi-dimensional evaluation is performed on each feature region.

[0070] S3: Fit the thermal crack trend line, using the characteristic area length and angle mean as parameters, and establish a sliding window along the trend line in the clockwise direction with the mean of the adjacent thermal crack angles as the step size. Combine the characteristic parameters of the window with the thermal crack parameters to predict the window size, and then calculate the potential thermal crack index of the window based on the window pixel distribution and parameter changes.

[0071] It's important to note that thermal cracks on the surface of shaft seal components often don't exist in isolation, but rather are distributed along a specific circumferential trend, and potentially contain hidden cracks. While the previous step identified the thermal crack area, it didn't analyze the overall crack distribution, making it difficult to predict potential cracks. This step, by fitting a thermal crack trend line, captures the overall crack propagation direction. By combining thermal crack characteristic parameters to establish a sliding window, this approach better reflects the crack distribution characteristics of the shaft seal annular structure, providing a trend basis for potential crack prediction, compared to the previous step, which only identified a single crack area.

[0072] Specifically, all the center points of the thermal crack areas are connected, and arc fitting is performed using the least squares method to obtain a fitting curve, which is recorded as the crack trend line; all the thermal crack areas in the target image are extracted, and the center point of each thermal crack area is located, which is recorded as the crack midpoint; the midpoints of two adjacent cracks are connected to the pole center, and the angle between the two lines is calculated, which is recorded as the adjacent crack angle; all the adjacent crack angles in the target image are calculated and averaged, which is recorded as the mean of adjacent crack angles; the characteristic area length and characteristic area angle are extracted, and the characteristic area length and characteristic area angle are used as the crack area length and crack area angle respectively, and the maximum crack area length and the maximum crack area angle of all thermal crack areas are extracted as the length range and angle range of the thermal crack in the target image; the mean of the adjacent crack angles and the length range and angle range of the thermal crack are recorded as the thermal crack characteristic parameters.

[0073] It's important to note that thermal cracks on the surface of shaft seal components follow a circular distribution pattern, and potential cracks often correlate with identified cracks in terms of circumferential spacing and size. While the previous step fitted the crack trend line and extracted characteristic parameters, it didn't retrieve the characteristic regions along the trend line, making it difficult to cover areas where potential cracks might exist. This step creates a sliding window with a specific step size and size, extending it to the image boundary to cover the possible areas of potential cracks.

[0074] Preferably, a sliding window is established on the fitted crack trend line in a clockwise direction with the average of the angles between adjacent cracks as the step length until it reaches the boundary of the target image and stops, so that the sliding window satisfies the window length being the maximum length of the crack area, the sliding window satisfies the window angle being the maximum angle of the crack area, and the sliding window step length being the average of the angles between adjacent cracks.

[0075] It should be noted that in the existing method of the present invention, the length and angle of the target window are determined based on the maximum length and angle of the thermal crack region, which may be insufficiently adaptable to cracks of different sizes. For example, small cracks may be covered by an overly large window, resulting in increased noise interference; large cracks may be truncated by the window, resulting in partial areas being missed. This step arranges the thermal crack length and angle data along the trend line and uses a weighted average formula to predict the length and angle of the sliding window, achieving the purpose of dynamically adjusting the window parameters.

[0076] Preferably, the window length of the sliding window and the length of all thermal cracks are extracted, and these length data are arranged in a clockwise direction along the fitted crack trend line with the position of the prediction window to obtain a length set; the window angle of the sliding window and the angle of all thermal cracks are extracted, and these angle data are arranged in a clockwise direction along the fitted crack trend line to obtain an angle set; based on the length set and the angle set, the sliding window size is predicted in sequence along the clockwise direction of the fitted crack trend line, and the sliding window size includes the prediction window length of the sliding window and the prediction window angle of the sliding window.

[0077] The prediction sliding window size satisfies the following expression:

[0078] ;

[0079] Where, Indicates the prediction window length of the i-th sliding window; in indicates the in-th sliding window; Indicates the prediction window length of the in-th sliding window; represents the prediction window angle of the i-th sliding window; Indicates the prediction window angle of the in-th sliding window.

[0080] Where, Indicates the length of the sliding window from the 1st to the i-th Perform weighted summing and then take the average; is the weight, indicating that as n increases, the weight of the window before the i-th sliding window that is farther away from the i-th sliding window is smaller, and the window closer to the i-th window has a greater impact on the length and angle prediction of the i-th window; It also represents the angle from the 1st to the i-th sliding window After weighted summation, the average is taken. The other parts of the angle prediction formula have the same function as the length prediction principle.

[0081] It should be noted that the potential thermal cracks on the surface of the shaft seal assembly are often manifested as weak pixel features and scattered distribution, which are difficult to accurately identify by adapting the window size alone. Although the previous step achieved dynamic adjustment of the window size, it did not combine the analysis of the impact of the pixel distribution details and the previous window credibility on the current window, which may have the potential risk of missing the detection of potential cracks. This step calculates the window potential thermal crack index by quantifying the number and dispersion of white pixels in the window and combining the credibility accumulation of the window before the i-th sliding window. That is, this step first extracts the window prediction value and the previous window credibility and arranges them into a final credibility set, and then combines the current window pixel distribution characteristics with the weighted evaluation value of the final credibility set through a formula to calculate the window potential thermal crack index.

[0082] Preferably, the results of all predicted sliding window sizes are extracted, and all sliding windows are set in combination with the fitted crack trend line and the average of the angles between adjacent cracks as the step size; all thermal crack areas and the final credibility of the features corresponding to the thermal crack areas are extracted, and the potential crack credibility of the window before the i-th sliding window is calculated; along the clockwise direction of the fitted crack trend line, the final credibility of the features corresponding to these thermal crack areas and the potential crack credibility of the window before the i-th sliding window are arranged according to the position of the sliding window to obtain a final credibility set; based on the pixel distribution characteristics in the window and the final credibility set, the possibility of potential cracks in the current window is calculated, which is recorded as the window potential thermal crack index.

[0083] The window potential thermal crack index satisfies the following expression:

[0084] ;

[0085] Where, represents the potential thermal crack index of the i-th sliding window; Represents the number of all pixels in the feature area in the i-th sliding window; 、 Represents the distance between the nth white pixel in the feature area of ​​the i-th sliding window and the center pixel and the center of the window; 、 Represents the average distance between all white pixels in the i-th sliding window and the center pixel and the center of the window; represents the potential thermal crack index of the in-th sliding window; It is a minimum positive value, ensuring that the denominator is not 0; represents the absolute value function; Represents the normalization function.

[0086] Where, Indicates the possibility of potential thermal cracks in the i-th sliding window is evaluated based on the distribution characteristics of the white pixels in the characteristic area of ​​the sliding window. The difference between the distance between the nth white pixel in the feature area of ​​the i-th sliding window and the center pixel of the window and the average distance between all white pixels in the i-th sliding window and the center pixel of the window is recorded as the dispersion degree of a single pixel in the window. Represents the total dispersion of all pixels in the sliding window, It represents the difference between the distance between the nth white pixel in the feature area of ​​the i-th sliding window and the polar center, and the average distance between all white pixels in the feature area of ​​the i-th sliding window and the polar center. It is used to express the dispersion degree of a single pixel in the polar coordinate system. Indicates the degree of dispersion of all pixels in the polar coordinate system; represents the weighted average of the potential thermal crack index of the sliding window before the i-th sliding window, is the weight, The closer the sliding window is to the i-th sliding window, the greater its influence on the i-th sliding window, and the greater the calculated corresponding credibility; It means multiplying the pixel distribution feature evaluation result of the i-th window by the weighted average result of the potential thermal crack index before the i-th sliding window to obtain the potential thermal crack index of the i-th sliding window; The larger the value, the higher the possibility of the existence of potential thermal cracks in the i-th sliding window.

[0087] At this point, the potential thermal crack area is obtained.

[0088] S4: Mark cracks and potential crack areas.

[0089] It's important to note that the shaft seal assembly, a critical sealing component, requires a visual representation of surface cracks for subsequent maintenance decisions. While the previous step calculated the potential thermal crack index, it didn't directly identify defective areas. This step, by presetting a second threshold to filter high-risk windows and mark all problematic areas, transforms abstract data into intuitive visual markers, enhancing the practicality of the technical solution.

[0090] Specifically, the second threshold is preset. A sliding window greater than the second threshold is regarded as a sliding window with potential thermal cracks; all thermal crack areas and potential crack areas in the target image are marked. It should be noted that the first threshold is set by the implementer according to the actual implementation situation. For example, in the Chinese patent document with authorization announcement number CN115423816B, entitled A Metal Surface Grinding Quality Detection Method, the preset threshold is 0.55. Marking thermal cracks and potential crack areas can intuitively present the full circumferential distribution of shaft seal surface defects, provide a complete basis for quality assessment, and facilitate the distinction between clear cracks that require immediate treatment and suspected cracks that require further verification.

[0091] At this point, a method for detecting the surface quality of a shaft seal assembly has been completed.

[0092] Although this specification has shown and described several embodiments of the present invention, it is obvious to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and substitutions without departing from the idea and spirit of the present invention.

Claims

1. A method for detecting the surface quality of a shaft seal assembly, characterized in that: include: After edge detection is performed on the collected surface image of the shaft seal component, region growing is performed to locate and extract the circular edge based on the annular distribution characteristics of the connected domain, and the remaining connected domain is recorded as the feature region; The polar center is obtained based on the distribution characteristics of the circular edge, and a polar coordinate system is constructed. An arc is fitted based on the trend of the line connecting the center points of the feature area, and the arc similarity between the fitted arc and the circular edge is calculated to calculate the first reliability of the feature. Calculate the length and angle of the feature area based on the distance between the pixel point of the feature area and the line connecting the polar center, as well as the angle with the reference line of the polar coordinate system; The final credibility of the feature is calculated based on the difference between the length and angle of the feature area and the distribution regularity of the pixel points of the feature area in the polar coordinate system. The first threshold is preset to screen the final credibility of the feature to obtain the thermal crack area; According to the distribution of all thermal crack areas and the mean length of characteristic areas, the fitted crack trend line and thermal crack characteristic parameters are obtained respectively; Using the thermal crack characteristic parameters as the preset window parameters, a sliding window is established along the fitted thermal crack trend line in the clockwise direction with the average of the angles between adjacent thermal cracks as the step size. The sliding window size is predicted in sequence by combining the characteristic parameters of the sliding window and the thermal crack parameters of the thermal crack area. The window is set according to the window size, and the potential thermal crack index of the window is calculated by combining the pixel distribution characteristics and the degree of change of the thermal crack parameters in all windows, and the cracks and potential crack areas are marked.

2. A method for detecting the surface quality of a shaft seal assembly according to claim 1, characterized in that: The constructing of the polar coordinate system includes: The RANSAC arc fitting algorithm is used for all circular edges to obtain the fitting center corresponding to each circular edge, and multiple target center points are obtained; for multiple target center points, the sum of squared residuals from the original edge point to each target center point is calculated, and the target center point corresponding to the smallest residual value is selected as the polar center. The ray starting from the polar center and pointing horizontally to the right along the horizontal axis of the target image is recorded as the baseline, and a polar coordinate system is constructed in the target image.

3. A method for detecting the surface quality of a shaft seal assembly according to claim 1, characterized in that: The first credibility of the feature satisfies the following expression: ； Where, Indicates the first credibility of the feature, Indicates the minimum cumulative distance corresponding to the fitted arc where the feature area is located; Represents the minimum cumulative distance mean corresponding to all fitted arcs on the target image; It is a minimum positive value, ensuring that the denominator is not 0; represents the absolute value function; represents the normalization function; Represents an exponential function with a natural constant as its base.

4. A method for detecting the surface quality of a shaft seal assembly according to claim 1, characterized in that: The calculation of the final credibility of the feature includes: Calculate the second credibility of the feature; the final credibility of the feature satisfies the following expression: ； Where, Indicates the final credibility of the feature; g indicates the total number of feature region clusters; Represents the number of pixels in the i-th cluster; Represents the average number of all pixels on the same arc in the feature area; Represents the average distance between adjacent pixels in the i-th cluster; Indicates the second credibility of the feature; represents the absolute value function; Represents an exponential function with a natural constant as its base.

5. A method for detecting the surface quality of a shaft seal assembly according to claim 4, characterized in that: The calculating the second credibility of the feature includes: According to the distance from the pixel points in the feature area to the centroid, the starting point and the end point are located, and the length of the feature area and the mean length of the feature area are calculated according to the distance from the starting point and the end point to the centroid; all the pixel points in the feature area are connected to the centroid respectively, and these connecting lines are recorded as the pixel point line segments of the feature area, and the angle of the feature area and the mean angle of the feature area are calculated according to the relationship between the pixel point line segments and the baseline; the difference between the length and angle of the feature area and the mean length and angle of the feature area are calculated respectively, as well as the ratio of the difference to the length and angle of all feature areas, and the second credibility of the feature is calculated in combination with the first credibility of the feature.

6. A method for detecting the surface quality of a shaft seal assembly according to claim 1, characterized in that: The obtaining of the thermal crack region comprises: The final credibility of features of all feature regions is calculated, a first threshold is preset, feature regions greater than the first threshold are recorded as thermal crack regions, and the regions are extracted to obtain thermal crack regions.

7. A method for detecting the surface quality of a shaft seal assembly according to claim 1, characterized in that: The respectively obtaining of the fitted crack trend line and the thermal crack characteristic parameters includes: The center points of all thermal crack areas were connected, and arc fitting was performed using the least squares method to obtain a fitting curve, which was recorded as the crack trend line; the midpoints of two adjacent cracks were connected with the polar center respectively, and the angle between the two connecting lines was calculated, which was recorded as the adjacent crack angle, and the mean of the adjacent crack angles of all adjacent cracks was calculated; the maximum crack area length and the maximum crack area angle of all thermal crack areas were extracted as the length and angle range of the thermal crack; the mean of the adjacent crack angle and the length and angle range of the thermal crack were recorded as the thermal crack characteristic parameters, and the fitting crack trend line and thermal crack characteristic parameters were obtained respectively.

8. The method for detecting the surface quality of a shaft seal assembly according to claim 1, wherein: The step of establishing a sliding window comprises: A sliding window is established on the fitted crack trend line in a clockwise direction with the average of the adjacent crack angles as the step length until it reaches the boundary of the target image and stops, so that the sliding window satisfies the conditions that the window length is the maximum length of the crack area, the window angle is the maximum angle of the crack area, and the step length of the sliding window is the average of the adjacent crack angles.

9. The method for detecting the surface quality of a shaft seal assembly according to claim 1, wherein: The prediction sliding window size satisfies the following expression: ； Where, Indicates the prediction window length of the i-th sliding window; in indicates the in-th sliding window; Indicates the prediction window length of the in-th sliding window; represents the prediction window angle of the i-th sliding window; Indicates the prediction window angle of the in-th sliding window.

10. The method for detecting the surface quality of a shaft seal assembly according to claim 1, wherein: The window potential thermal crack index satisfies the following expression: ； Where, represents the potential thermal crack index of the i-th sliding window; Represents the number of all pixels in the feature area in the i-th sliding window; 、 Represents the distance between the nth white pixel in the feature area of ​​the i-th sliding window and the center pixel and the center of the window; 、 Represents the average distance between all white pixels in the i-th sliding window and the center pixel and the center of the window; represents the potential thermal crack index of the in-th sliding window; It is a minimum positive value, ensuring that the denominator is not 0; represents the absolute value function; Represents the normalization function.

Images