A bearing inner and outer ring synchronous flaw detection method and system
By employing adaptive clustering and feature quantization, the accuracy and reliability issues of existing bearing damage detection algorithms are addressed, enabling automated diagnosis of spalling damage on the inner and outer rings of bearings and improving the accuracy and reliability of the assessment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHAANXI LONGYUE RUIXING TECH CO LTD
- Filing Date
- 2025-12-12
- Publication Date
- 2026-06-05
AI Technical Summary
Existing bearing inner and outer ring damage detection algorithms cannot effectively distinguish between spalling and wear, resulting in low accuracy and reliability. Furthermore, fixed threshold methods are difficult to adapt to complex damage characteristics, leading to frequent missed and false detections.
By adaptively clustering edge pixels, microcrack and pit features are quantified, and microcrack and pit evaluation indicators are constructed. Combined with directional feature coefficients and brightness distribution, the threshold is dynamically adjusted to determine damage.
It improves the accuracy and reliability of bearing inner and outer ring damage assessment, reduces misjudgments and missed detections, realizes automated diagnosis of bearing spalling damage, and avoids the subjectivity of manually setting thresholds.
Smart Images
Figure CN121482019B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of image processing technology, specifically to a method and system for simultaneous flaw detection of the inner and outer rings of a bearing. Background Technology
[0002] Bearing damage is the decisive factor in bearing failure. Real-time and efficient flaw detection of the inner and outer rings of the bearing can detect damage in time, prevent bearing failure caused by damage, and prevent mechanical system failure caused by equipment damage.
[0003] In actual operation, the relative sliding between the contact surfaces of bearings generates an adhesive effect, while the rotating load under high-speed operation can easily lead to localized spalling damage on the inner and outer ring surfaces. However, existing spalling detection algorithms have significant limitations: contrast enhancement-based methods cannot effectively distinguish between "spalling" and other types of damage such as "wear." Relying solely on contrast enhancement can misclassify wear areas as spalling, thus reducing the accuracy and reliability of determining spalling damage on the inner and outer rings of bearings. On the other hand, algorithms based on fixed thresholds are difficult to adapt to the complex morphology of spalling damage—the edges of the damaged area are often accompanied by microcracks of varying lengths and directions, as well as small particles with sharp edges. This complex and variable characteristic makes the detection effect of fixed threshold methods extremely unstable, with frequent missed and false detections, reducing the accuracy and reliability of determining spalling damage on the inner and outer rings of bearings. Summary of the Invention
[0004] To address the aforementioned technical problems, the purpose of this application is to provide a method and system for simultaneous flaw detection of the inner and outer rings of bearings. The specific technical solution adopted is as follows:
[0005] In a first aspect, embodiments of this application provide a method for simultaneous flaw detection of the inner and outer rings of a bearing, the method comprising the following steps:
[0006] Acquire a preset number of bearing images of the inner and outer rings of the same bearing;
[0007] Extract all edge pixels from each bearing image, cluster all edge pixels, and denote the region containing the minimum bounding rectangle of each cluster as a local region; determine the directional feature coefficients of each bearing image based on the distribution of gradient direction angles of all pixels in each local region; determine the linear response intensity of each bearing image based on the variation characteristics of pixel values of all pixels in each local region in different gradient directions, and determine the microcrack evaluation index of each bearing image in combination with the directional feature coefficients.
[0008] Based on the degree of disorder of pixel values of all pixels in each local region and the distribution of gradient direction angles of all pixels, the saliency of pitting features in each bearing image is determined; based on the brightness distribution and gradient amplitude dispersion of pixel values of all pixels in each local region, and in combination with the saliency of pitting features, the pitting evaluation index of each bearing image is determined; based on the microcrack evaluation index and the pitting evaluation index, the confidence level of bearing spalling damage in each bearing image is determined.
[0009] Based on the confidence level of the bearing spalling damage, spalling damage is determined for both the inner and outer rings of the bearing.
[0010] Preferably, the method for determining the intra-cluster metric distance during the clustering process is as follows:
[0011] The distance between edge pixels i and j in the k-th bearing image is measured. The expression is: In the formula, This represents the coordinate distance between edge pixel i and edge pixel j in the k-th bearing image; This represents the distance between edge pixel i and edge pixel j in the k-th bearing image, using the unit gradient vector.
[0012] Preferably, the method for determining the directional feature coefficients of each bearing image is as follows:
[0013] In each local region of each bearing image, the cosine and sine values of the gradient direction angle of each pixel are calculated. The mean of the cosine value and the mean of the sine value of all pixels are calculated and denoted as the cosine mean and the sine mean, respectively. The square root of the sum of the squares of the cosine mean and the sine mean is denoted as the direction value of each local region.
[0014] The result of subtracting the normalized value of the direction value of each local region from 1 is recorded as the direction index of each local region. The mean of the direction indices of all local regions in each bearing image is used as the direction feature coefficient of each bearing image.
[0015] Preferably, the method for determining the linear response intensity of each bearing image is as follows:
[0016] The maximum response intensity of each pixel in each local region is calculated using convolution kernels across all gradient directions. The average of the maximum response intensities of all pixels in each local region is denoted as the response index of each local region.
[0017] The mean of the response indices of all local regions in each bearing image is taken as the linear response intensity of each bearing image.
[0018] Preferably, the microcrack evaluation index for each bearing image is the result of a linear positive fusion of the directional characteristic coefficient and the linear response intensity of each bearing image.
[0019] Preferably, the expression for the salience of the pit-like features in each bearing image is: ; The significance of the pitting features in the k-th bearing image is represented. The information entropy represents the histogram of pixel values of all pixels in local region a in the k-th bearing image; The response index of local region a in the k-th bearing image is represented. This represents the number of all local regions in the k-th bearing image; represents the preset attenuation coefficient; exp[ ] represents the exponential function with the natural constant as the base.
[0020] Preferably, the method for determining the pitting evaluation index for each bearing image is as follows:
[0021] In each bearing image, the gray-level co-occurrence matrix contrast of all pixels in each local region is calculated and denoted as the contrast feature value of each local region. The mean of the contrast feature values of all local regions is calculated and denoted as the contrast index of each bearing image.
[0022] The degree of dispersion of the gradient magnitude of all pixels in each local region is denoted as gradient dispersion. The mean of the gradient dispersion of all local regions in each bearing image is calculated and denoted as the gradient magnitude exponent of each bearing image.
[0023] The contrast index, gradient magnitude index, and saliency of pit features of each bearing image are positively fused to form the pit evaluation index for each bearing image.
[0024] Preferably, the expression for the confidence level of bearing spalling damage in each bearing image is: In the formula, This represents the confidence level of bearing spalling damage in the k-th bearing image; , These represent the microcrack evaluation index and the pitting evaluation index for the k-th bearing image, respectively.
[0025] Preferably, the determination of spalling damage on the inner and outer rings of the bearing includes:
[0026] A threshold segmentation algorithm is used to select all bearing images other than those with only foreground or only background from a preset number of bearing images as bearing images to be detected.
[0027] Clustering is performed on all bearing images to be inspected. Based on the mean of the bearing spalling damage confidence scores of all bearing images to be inspected in each cluster, the range of spalling damage and the range of non-spalling damage are determined. Based on the range to which the bearing spalling damage confidence score of the bearing image to be inspected belongs, it is determined whether spalling damage exists in the bearing image to be inspected. Secondly, embodiments of this application also provide a bearing inner and outer ring synchronous flaw detection system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described bearing inner and outer ring synchronous flaw detection methods.
[0028] This application has at least the following beneficial effects:
[0029] This application defines local regions by adaptively clustering edge pixels. Furthermore, it quantifies microcrack features from two dimensions: directional disorder and linear significance, constructing a microcrack evaluation index reflecting the severity of spalling damage. This effectively overcomes the problem of traditional algorithms being insensitive to the randomness of microcrack direction and length, helping to improve the accuracy and reliability of simultaneous flaw detection of bearing inner and outer rings. Further, this application quantifies pit features by analyzing the pixel disorder and directional distribution of local regions, and constructs a pit-like evaluation index by combining brightness and gradient dispersion. This index is then fused with the microcrack evaluation index to obtain the bearing spalling damage confidence level, which helps to effectively identify bearing spalling damage, reduce false positives and false negatives, and improve the accuracy and reliability of bearing spalling damage determination. Finally, this application automatically learns and determines a dynamic threshold for damage determination through adaptive clustering analysis combined with the bearing spalling damage confidence level. This objectively classifies the bearing image to be detected into two categories: those with spalling damage and those without, thereby achieving the final automated diagnosis of the bearing spalling state. This effectively avoids the subjectivity of manually setting thresholds and improves the accuracy and reliability of determining the spalling damage of the bearing inner and outer rings. Attached Figure Description
[0030] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0031] Figure 1 A flowchart illustrating the steps of a method for simultaneous flaw detection of the inner and outer rings of a bearing, as provided in one embodiment of this application;
[0032] Figure 2 This is a schematic diagram of the bearing spalling damage confidence extraction process provided in one embodiment of this application. Detailed Implementation
[0033] To further illustrate the technical means and effects adopted by this application to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a bearing inner and outer ring simultaneous flaw detection method and system proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.
[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.
[0035] The following description, in conjunction with the accompanying drawings, details the specific scheme of the bearing inner and outer ring synchronous flaw detection method and system provided in this application.
[0036] Please see Figure 1 The diagram illustrates a flowchart of a method for simultaneous flaw detection of the inner and outer rings of a bearing according to an embodiment of this application. The method includes the following steps:
[0037] Step S1: Obtain a preset number of bearing images of the inner and outer rings of the same bearing.
[0038] Using an industrial high-magnification CCD camera, bearing images are acquired from the inner and outer rings of the bearing. During the acquisition process, areas outside the inner and outer rings are avoided, and damage at each location is comprehensively captured. For example, the image may only show damage to the inner and outer rings and their surfaces. Therefore, damage in different areas of the same bearing's inner and outer rings is captured as completely as possible, or multiple images are used to reduce the influence of accidental factors during the actual acquisition process. In this embodiment, a preset number of bearing images of the same bearing's inner and outer rings are acquired using an industrial high-magnification CCD camera and are denoted as bearing images. Furthermore, for ease of analysis, the bearing images are converted to grayscale. Therefore, in the following content, all pixel values mentioned refer to grayscale values.
[0039] It should be noted that the preset quantity is set manually. In this embodiment, the preset quantity is 1000. In actual application, as other implementation methods, implementers can also set it according to specific circumstances. This embodiment does not impose any special restrictions.
[0040] Furthermore, since the bearing images of the inner and outer rings may be affected by equipment and lighting during the acquisition process, resulting in noise and uneven lighting in the bearing images, this embodiment uses a Gaussian filtering algorithm to suppress noise in the bearing images and reduce the interference of noise on feature extraction. In practical applications, as other implementation methods, implementers may also use other filtering methods such as median filtering algorithms according to specific circumstances. This embodiment does not impose any special restrictions on the selection of filtering algorithms. Furthermore, this embodiment uses the Limit Contrast Adaptive Histogram Equalization (CLAHE) algorithm to enhance the contrast of the bearing images. In this embodiment, the contrast limit value is set to 2.0, and the local processing area size is 16×16 to eliminate the influence of uneven lighting.
[0041] Gaussian filtering and contrast-limited adaptive histogram equalization (CLAHE) are both well-known techniques. The specific processes of using Gaussian filtering to reduce noise in images and using contrast-limited adaptive histogram equalization (CLAHE) to enhance contrast in images will not be described in detail here.
[0042] Step S2: Quantitatively analyze the microcrack and pit features in the local area at the edge of the bearing image to comprehensively assess the confidence level of bearing spalling damage.
[0043] When the inner and outer rings of a bearing experience spalling damage, irregular connected regions will appear in the images of the bearing's inner and outer rings. These irregular connected regions are the spalling damage areas. Through high-magnification camera acquisition, microcracks and small pits can be observed at the edges of the spalling damage areas. During the acquisition of bearing images of the inner and outer rings, two types of pixels can be identified as representing spalling damage: microcrack pixels and small pit pixels. These two types of pixels exhibit different regional characteristics in the bearing image. For example, when microcracks appear at the edge of the spalling damage area, the regions formed by microcrack pixels are characterized by inconsistent orientation and length; when small pits appear at the edge of the spalling damage area, the regions formed by small pit pixels are randomly distributed and have sharp-edged small particles around their edges. Both microcracks and small pits are located in the neighborhood of the edge of the spalling damage area, and interference between the two must be avoided.
[0044] In the peeling images of the inner and outer rings of the bearing, the microcracks in the neighborhood of the peeling damage area vary in direction and length, are sensitive to linear structures, and the pits have a granular texture with randomness. Therefore, the microcracks have a clear dominant direction and a strong linear response, that is, the neighborhood of the peeling damage area contains pixel connected regions with obvious linear structures; the pits have no dominant direction and obvious texture details, that is, the neighborhood of the peeling damage area has randomly distributed high-frequency pixel regions.
[0045] Based on the above analysis, this embodiment extracts all edge pixels in each bearing image, clusters all edge pixels, and denotes the region containing the minimum bounding rectangle of each cluster as a local region. Based on the distribution of gradient direction angles of all pixels in each local region, the directional feature coefficients of each bearing image are determined. Based on the variation characteristics of pixel values of all pixels in each local region under different gradient directions, the linear response intensity of each bearing image is determined. Combined with the directional feature coefficients, the microcrack evaluation index of each bearing image is determined. The process is as follows:
[0046] S2.1: Extract all edge pixels in each bearing image, cluster all edge pixels, and denote the region where the minimum bounding rectangle of each cluster is located as a local region; determine the directional feature coefficient of each bearing image based on the distribution of gradient direction angles of all pixels in each local region; determine the linear response intensity of each bearing image based on the variation characteristics of pixel values of all pixels in each local region in different gradient directions, and determine the microcrack evaluation index of each bearing image in combination with the directional feature coefficient.
[0047] (1) Extract all edge pixels in each bearing image, cluster all edge pixels, and denote the region where the minimum bounding rectangle of each cluster is located as the local region.
[0048] In this embodiment, an edge detection algorithm is used to identify all edge pixels in each bearing image, and a clustering algorithm is used to cluster all edge pixels. The region where the minimum bounding rectangle of each cluster is located is recorded as a local region.
[0049] It should be noted that there are many commonly used edge detection algorithms. This embodiment uses the Canny edge detection algorithm to extract edge pixels from the bearing image. In practical applications, as other implementation methods, implementers may also use the Sobel algorithm to extract edge pixels according to specific circumstances. This embodiment does not impose any special restrictions on the selection of edge detection algorithms.
[0050] It should be noted that there are many commonly used clustering algorithms. This embodiment uses the DBSCAN density clustering algorithm to cluster edge pixels. In practical applications, as other implementation methods, implementers may also use other clustering methods such as the k-means clustering algorithm according to specific circumstances. This embodiment does not impose any special restrictions on the selection of clustering algorithms.
[0051] The Canny edge detection algorithm and the DBSCAN density clustering algorithm are both well-known technologies. The specific process of extracting edge pixels using the Canny edge detection algorithm and the specific process of clustering edge pixels using the DBSCAN density clustering algorithm will not be described in detail.
[0052] It should be noted that the number of clusters is determined using the elbow rule during the clustering process, and the specific process for determining the distance metric is as follows:
[0053] As one implementation method, in this embodiment, the distance between edge pixel i and edge pixel j in the k-th bearing image is measured. The expression is: In the formula, This represents the coordinate distance between edge pixel i and edge pixel j in the k-th bearing image; This represents the distance between edge pixel i and edge pixel j in the k-th bearing image, using the unit gradient vector.
[0054] It should be noted that the distance between unit gradient vectors in this embodiment is Euclidean distance. In practical applications, as other implementation methods, implementers may also use other distance calculation methods such as DTW distance according to specific circumstances. This embodiment does not impose any special restrictions.
[0055] Among them, the elbow rule and the calculation method of the unit gradient vector are well-known techniques. The specific process of using the elbow rule to determine the number of clusters, the specific calculation process of the unit gradient vector, and the calculation process of Euclidean distance are all well-known techniques, and will not be elaborated here. (2) Based on the distribution of the gradient direction angle of all pixels in each local region, the directional feature coefficients of each bearing image are determined.
[0056] In this embodiment, in each local region of each bearing image, the cosine and sine values of the gradient direction angle of each pixel are calculated, and the mean of the cosine values and the mean of the sine values of all pixels are calculated and recorded as the cosine mean and the sine mean, respectively. The square root of the sum of the squares of the cosine mean and the sine mean is recorded as the direction value of each local region.
[0057] The result of subtracting the normalized value of the direction value of each local region from 1 is recorded as the direction index of each local region. The mean of the direction indices of all local regions in each bearing image is used as the direction feature coefficient of each bearing image.
[0058] It should be noted that there are many commonly used normalization methods. In this embodiment, the direction values of all local regions are used as the input of the maximum-minimum normalization method, and the output is the result after the direction values are normalized. In practical applications, as other implementation methods, implementers can also use other normalization methods such as z-score normalization method according to specific circumstances. This embodiment does not impose any special restrictions on the selection of normalization methods.
[0059] It should be noted that, unless otherwise specified, all normalization processes in this embodiment employ the maximum-minimum value normalization method.
[0060] The methods for obtaining the gradient direction angle of each pixel and the process of normalizing the data using the minimum-maximum normalization method are well-known techniques, and their specific acquisition processes will not be elaborated further. Based on the directional feature coefficient of each bearing image, it can be understood that the directional feature coefficient reflects the degree of disorder in the gradient direction within a local region. It is used to determine whether there is a dominant, consistent gradient direction within the local region. The directional feature coefficient is mainly affected by the distribution of the gradient direction angles of all pixels within the local region. The gradient direction angle is the normal direction of the linear texture, that is, the direction perpendicular to the texture direction. If the directional feature coefficient of the current bearing image is larger, it indicates that the gradient directions of the pixels in the local region of the current bearing image are more dispersed, and there is no obvious dominant direction. This suggests that the local region may contain random textures with multiple directions, indicating a greater possibility of peeling damage to the inner and outer rings of the bearing.
[0061] Conversely, if the directional feature coefficient of the current bearing image is smaller, it means that the gradient direction of the pixels in the local area of the current bearing image is more concentrated and there is a clear dominant direction. This indicates that the local area may be a smooth area with uniform gray level. Therefore, it is less likely to be judged as complex peeling damage.
[0062] (3) Based on the variation characteristics of pixel values of all pixels in each local area in different gradient directions, the linear response intensity of each bearing image is determined.
[0063] As one implementation method, in this embodiment, the maximum response intensity of each pixel in each local region in all gradient directions is calculated using a convolution kernel, and the average of the maximum response intensity of all pixels in each local region is recorded as the response index of each local region.
[0064] The mean of the response indices of all local regions in each bearing image is taken as the linear response intensity of each bearing image.
[0065] It should be noted that all gradient directions in this embodiment include: The specific calculation process for the response intensity of a pixel in each gradient direction is as follows:
[0066] This embodiment uses gradient direction For example, each pixel is divided into 8 neighborhoods centered on it. The pixel values of each pixel and all pixels within its 8 neighborhoods are arranged in order from left to right and from top to bottom according to the pixel's position coordinates, forming an 8-neighborhood matrix. The 8-neighborhood matrix and the gradient direction are then calculated. The convolution kernels are multiplied, and the sum of all elements in the resulting matrix is used as the gradient of the pixel along the gradient direction. The response intensity of a pixel in the gradient direction is calculated using the same method as described above for the gradient direction. Methods for calculating the time response intensity.
[0067] Where, when the gradient direction is When the convolution kernel is When the gradient direction is When the convolution kernel is When the gradient direction is When the convolution kernel is When the gradient direction is When the convolution kernel is .
[0068] Based on the linear response intensity of each bearing image, it can be understood that the linear response intensity reflects the maximum average response intensity of pixels in a local area to a multi-directional linear filter. It is used to characterize the density and clarity of linear features in the bearing image. If the linear response intensity of the current bearing image is greater, it indicates that there are a large number of clear and obvious linear structures in the current bearing image. These structures have generated a strong response with the convolution kernel in at least one direction. This reflects the presence of microcrack-type spalling damage in the inner and outer rings of the bearing, indicating that the bearing is highly likely to have spalling damage.
[0069] Conversely, if the linear response intensity of the current bearing image is smaller, it indicates that the linear structure in the current bearing image is sparse, blurry, or indistinct, and the pixel grayscale changes relatively smoothly in all directions, failing to strongly match any linear detection kernel. This reflects that there may be no microcracks in the inner and outer rings of the bearing, or the microcracks are very fine and have low contrast, thus indicating that the possibility of such spalling damage in the bearing is also reduced.
[0070] (4) Based on the linear response intensity of each bearing image and in combination with the directional characteristic coefficient, determine the microcrack evaluation index of each bearing image.
[0071] In this embodiment, the result of linearly and positively fusing the directional feature coefficients and linear response intensity of each bearing image is used as the microcrack evaluation index for each bearing image.
[0072] It should be understood that positive fusion refers to combining two or more indicators through addition or multiplication to obtain a comprehensive indicator, thereby more comprehensively and accurately assessing a phenomenon or problem. This fusion method is not limited to simple arithmetic operations, but can also include more complex statistical models and analytical methods. Implementers can choose according to specific circumstances, and this embodiment does not impose any special restrictions.
[0073] Preferably, as one implementation method, in this embodiment, the expression for the microcrack evaluation index of each bearing image is: In the formula, The microcrack evaluation index represents the k-th bearing image. , These represent the directional characteristic coefficients and linear response intensity of the k-th bearing image, respectively. , These represent the preset first weighting factor and the preset second weighting factor, respectively. , In practical applications, as other implementation methods, implementers may also adopt other positive fusion methods such as product or sum, depending on the specific circumstances. This embodiment does not impose any special restrictions.
[0074] It should be noted that the values of the preset first weighting factor and the preset second weighting factor are both set manually. In this embodiment, the values of the preset first weighting factor and the preset second weighting factor are 0.6 and 0.4, respectively. The directional characteristic coefficient reflects the directional disorder of microcracks and is the core feature of spalling damage, so it is given a higher weight. The linear response intensity quantifies the significance of the linear structure, but it may be affected by noise interference, so the weight is slightly lower. In practical applications, as other implementation methods, implementers can also set their own weights according to specific circumstances. This embodiment does not impose any special restrictions.
[0075] Based on the microcrack evaluation index of each bearing image, it can be understood that the microcrack evaluation index reflects the comprehensive performance of pixel values in two dimensions: directional disorder and linear significance. It is used to characterize the overall intensity of microcracks, the core feature of spalling damage. If the directional feature coefficient of the current bearing image is larger, it indicates that the direction of microcracks in the current bearing image is more disordered, which is consistent with the true morphology of microcracks at the edge of spalling damage. Therefore, the corresponding microcrack evaluation index is larger. At the same time, if the linear response intensity of the current bearing image is larger, it indicates that the microcracks in the current bearing image are more obvious. Therefore, the corresponding microcrack evaluation index is larger, indicating that the severity of spalling damage in the bearing is higher.
[0076] Conversely, if the directional consistency coefficient of the current bearing image is smaller, it indicates that the direction of its microcracks is more uniform, which is inconsistent with the usually chaotic morphology of microcracks at the edge of spalling damage. Therefore, it will reduce the microcrack evaluation index. At the same time, if the linear response intensity of the current bearing image is also smaller, it indicates that its microcrack features are vague, few in number, or insignificant. This will also reduce the microcrack evaluation index. Overall, it indicates that the severity of spalling damage in the bearing is low, or there may not even be such damage.
[0077] Thus, this embodiment defines local regions by adaptively clustering edge pixels. Furthermore, it quantifies microcrack features from two dimensions: directional disorder and linear significance. Finally, it constructs a microcrack evaluation index that reflects the severity of spalling damage. This effectively overcomes the problem that traditional algorithms are insensitive to the randomness of microcrack direction and length, and helps to improve the accuracy and stability of synchronous flaw detection of bearing inner and outer rings.
[0078] S2.2: Based on the degree of disorder of pixel values of all pixels in each local region and the distribution of gradient direction angles of all pixels, determine the saliency of pitting features of each bearing image; based on the brightness distribution and gradient amplitude dispersion of pixel values of all pixels in each local region, and in combination with the saliency of pitting features, determine the pitting evaluation index of each bearing image; based on the microcrack evaluation index and the pitting evaluation index, determine the confidence level of bearing spalling damage of each bearing image.
[0079] (1) In this embodiment, the salience of pit features in each bearing image is determined based on the degree of disorder of pixel values of all pixels in each local area and the distribution of gradient direction angles of all pixels.
[0080] As one implementation method, in this embodiment, the saliency of the pit-like features in the k-th bearing image is... The expression is: ; The information entropy represents the histogram of pixel values of all pixels in local region a in the k-th bearing image; The response index of local region a in the k-th bearing image is represented. This represents the number of all local regions in the k-th bearing image; represents the preset attenuation coefficient; exp[ ] represents the exponential function with the natural constant as the base.
[0081] It should be noted that the preset attenuation coefficient is set manually. In this embodiment, the preset attenuation coefficient is 0.5. In actual applications, as other implementation methods, implementers can also set it according to specific circumstances. This embodiment does not impose any special restrictions.
[0082] The methods for obtaining the histogram of pixel values and calculating the information entropy are well-known techniques. In the histogram, the horizontal axis represents the range of pixel values, and the vertical axis represents the probability of a pixel value falling into each range. The specific process of obtaining the histogram and calculating the information entropy will not be elaborated here.
[0083] Based on the saliency of the pit-like features in each bearing image, it can be understood that the saliency of the pit-like features reflects the randomness and unpredictability of the pixel gray-level distribution in a local area after excluding the interference of linear structures (such as microcracks). It is used to characterize nonlinear spalling damage features such as small pits. If the information entropy of the histogram of pixel values of all pixels in local area a in the k-th bearing image is larger, and if the response index of local area a in the k-th bearing image is smaller, it indicates that the pixel gray-level distribution in local area a is more chaotic and irregular, the linear feature is weaker, and local area a is more likely to have spalling damage features such as small pits. Therefore, the corresponding saliency of the pit-like features is larger.
[0084] Conversely, if the information entropy of the histogram of pixel values of all pixels in local region a in the k-th bearing image is smaller, and if the response index of local region a in the k-th bearing image is larger, it indicates that the pixel gray-level distribution in local region a is more regular and uniform, the linear feature is stronger, and the local region a is less likely to have small pit-like peeling damage features. Instead, it is more likely to be a clear micro-crack or a smooth background. Therefore, the significance of the corresponding pit-like feature is smaller.
[0085] (2) In this embodiment, the evaluation index of pitting for each bearing image is determined based on the brightness distribution and gradient amplitude of all pixels in each local area, combined with the salience of the pitting feature.
[0086] In this embodiment, in each bearing image, the gray-level co-occurrence matrix contrast of all pixels in each local region is calculated and recorded as the contrast feature value of each local region. The mean of the contrast feature values of all local regions is calculated and recorded as the contrast index of each bearing image.
[0087] The degree of dispersion of the gradient magnitude of all pixels in each local region is denoted as gradient dispersion. The mean of the gradient dispersion of all local regions in each bearing image is calculated and denoted as the gradient magnitude exponent of each bearing image.
[0088] The contrast index, gradient magnitude index, and saliency of pit features of each bearing image are positively fused to form the pit evaluation index for each bearing image.
[0089] Preferably, as a specific implementation method, in this embodiment, the expression for the pitting evaluation index of each bearing image is: In the formula, The pitting evaluation index represents the k-th bearing image; , , These represent the contrast index, gradient magnitude index, and saliency of the pitting feature in the k-th bearing image, respectively. , , These represent the preset first weight value, the preset second weight value, and the preset third weight value, respectively. In practical applications, as other implementation methods, implementers may also adopt positive fusion methods such as product or sum, depending on the specific circumstances. This embodiment does not impose any special restrictions on the selection of positive fusion methods.
[0090] It should be noted that there are many methods to measure the dispersion of data. In this embodiment, the variance of the gradient magnitude of all pixels in each local region is used as the dispersion of the gradient magnitude of all pixels in each local region. In practical applications, as other implementation methods, implementers may also use other methods such as standard deviation or coefficient of variation to measure the dispersion of data, depending on the specific circumstances. This embodiment does not impose any special restrictions on the selection of methods to measure the dispersion of data.
[0091] The process of obtaining the gradient magnitude and the process of calculating the contrast using the gray-level co-occurrence matrix are well-known techniques and will not be elaborated further.
[0092] It should be noted that the values of the preset first weight, preset second weight, and preset third weight are all set manually. In this embodiment, the preset first weight is set to 0.4, the preset second weight is set to 0.4, and the preset third weight is set to 0.2. The contrast index and gradient amplitude index directly reflect the edge sharpness and grayscale change of the pits, which are key features, and each accounts for 0.4. The salience of the pit feature measures the pixel disorder through information entropy, but it may be affected by lighting. The weight is set to 0.2 to reduce interference. In actual applications, as other implementation methods, implementers can also set their own weights according to the situation. This embodiment does not impose any special restrictions.
[0093] Based on the pitting evaluation index for each bearing image, it can be understood that the pitting evaluation index reflects the comprehensive performance of small pits in three dimensions: texture contrast, gradient discretization, and texture disorder after suppression. It is used to characterize the overall strength of the core feature of small pits in spalling damage. If the contrast index of the current bearing image is larger, it means that the edges of the small pits in the current bearing image are clearer. Therefore, the corresponding pitting evaluation index is larger, indicating that there is a greater possibility of spalling damage in the current bearing image. At the same time, if the gradient amplitude index of the current bearing image is larger, it means that the gray-level changes in the current bearing image are more uneven. It indicates that there is a greater possibility of small pit spalling damage. Therefore, the corresponding pitting evaluation index is larger. In addition, if the saliency of the pitting feature in the current bearing image is larger, it means that there are a large number of chaotic textures without linear features in the bearing image. It indicates that there is a greater possibility of granular small pit spalling damage in the current bearing image. Therefore, the corresponding pitting evaluation index is larger.
[0094] Conversely, a smaller contrast index in the current bearing image indicates that the edges of the small pits in the current bearing image are blurred and the contrast with the background is not strong. Therefore, the corresponding pitting evaluation index is smaller. At the same time, a smaller gradient amplitude index in the current bearing image indicates that the gray-level changes in the current bearing image are more uniform and the surface is smoother, indicating that the possibility of small pit-like peeling damage is smaller. Therefore, the corresponding pitting evaluation index is also smaller. In addition, a smaller saliency of the pitting features in the current bearing image indicates that the texture regularity in the bearing image is strong, indicating that the possibility of granular small pit peeling damage in the current bearing image is smaller. Therefore, the corresponding pitting evaluation index also decreases.
[0095] (3) In this embodiment, the confidence level of bearing spalling damage for each bearing image is determined based on the microcrack evaluation index and the pitting evaluation index.
[0096] As one implementation method, in this embodiment, the confidence level of bearing peeling damage in the k-th bearing image is... The expression is: In the formula, , These represent the microcrack evaluation index and the pitting evaluation index for the k-th bearing image, respectively.
[0097] Preferably, the schematic diagram of the bearing spalling damage confidence extraction process provided in this embodiment is as follows: Figure 2 As shown.
[0098] Based on the bearing spalling damage confidence score for each bearing image, it can be understood that the bearing spalling damage confidence score reflects the combined evidentiary strength of two key features: microcracks and pitting damage. It is used to make a final judgment on whether the bearing has suffered spalling damage. If the microcrack evaluation index of the current bearing image is larger, it indicates that there is a greater possibility of microcrack damage in the current bearing image, and the corresponding bearing spalling damage confidence score is also larger. At the same time, if the pitting evaluation index of the current bearing image is larger, it indicates that there is a greater possibility of small pitting spalling damage in the current bearing image, and the corresponding bearing spalling damage confidence score is also larger.
[0099] Conversely, the smaller the microcrack evaluation index of the current bearing image, the lower the probability of microcrack damage in the current bearing image, and the lower the confidence level of bearing spalling damage. At the same time, the smaller the pitting evaluation index of the current bearing image, the lower the probability of small pit spalling damage in the current bearing image, and the lower the confidence level of bearing spalling damage. Overall, this indicates that the probability of the bearing experiencing spalling damage is extremely low.
[0100] Thus, this embodiment quantifies the characteristics of small pits by analyzing the pixel disorder and directional distribution in local areas, further combines brightness and gradient dispersion to construct a pit-like evaluation index, and finally integrates it with the microcrack evaluation index to obtain the confidence level of bearing spalling damage. This helps to effectively identify bearing spalling damage, reduce misjudgments and missed detections, and improve the accuracy and reliability of bearing spalling damage determination.
[0101] Step S3: Based on the confidence level of the bearing spalling damage, determine the spalling damage of the inner and outer rings of the bearing.
[0102] Based on the confidence level of bearing spalling damage obtained in step S2, spalling damage is determined for both the inner and outer rings of the bearing. Specifically:
[0103] In this embodiment, a threshold segmentation algorithm is used to select all bearing images other than those with only foreground or only background in a preset number of bearing images as bearing images to be detected.
[0104] It should be noted that there are many commonly used threshold segmentation algorithms. In this embodiment, the Otsu threshold segmentation algorithm is used to segment the bearing image. In practical applications, as other implementation methods, implementers may also choose other threshold segmentation algorithms according to specific circumstances. This embodiment does not impose any special restrictions on the selection of threshold segmentation algorithms.
[0105] Among them, the Otsu threshold segmentation algorithm is a well-known technology, and the specific process of using it to segment the bearing image to be detected will not be described in detail.
[0106] It should be noted that the image of the bearing to be tested may contain peeling damage. This is because if there is peeling damage on the inner and outer rings of the bearing, the peeling damage will exist in the shaft image as connected regions. Therefore, if a bearing image only has a foreground or only has a background, it indicates that the bearing image does not have color partitioning, and therefore, there is no peeling damage in the bearing image.
[0107] Furthermore, in this embodiment, the k-means clustering algorithm is used to cluster all the bearing images to be detected. The number of clusters is set to 2, and 2 clusters are output. Based on the mean of the bearing peeling damage confidence scores of all bearing images to be detected in each cluster, the peeling damage range and the non-peeling damage range are determined, specifically as follows:
[0108] In this embodiment, the mean confidence score of bearing spalling damage for all bearing images to be detected in each cluster is calculated. The higher the mean confidence score of bearing spalling damage, the greater the probability of spalling damage on the inner and outer rings of the bearing; conversely, the lower the mean confidence score, the less likely spalling damage is to exist on the inner and outer rings of the bearing. Therefore, this embodiment defines the spalling damage range, i.e., the spalling damage range and the non-spalling damage range, based on the mean confidence score of bearing spalling damage for each cluster. The spalling damage range is […]. The non-stripping damage range is [ ), where V represents the sum of the mean confidence scores of bearing spalling damage in the two clusters and the result of normalization.
[0109] If the confidence level of the bearing peeling damage in the current peeling image to be detected is within the peeling damage range, then there is peeling damage in the current peeling image to be detected; conversely, if the confidence level of the bearing peeling damage in the current peeling image to be detected is within the non-peeling damage range, then there is peeling damage in the current peeling image to be detected.
[0110] Thus, this embodiment, through adaptive clustering analysis and combined with the confidence level of bearing spalling damage, automatically learns and determines the dynamic threshold for damage judgment, objectively classifying the bearing images to be detected into two categories: those with spalling damage and those without spalling damage. This achieves the final automated diagnosis of the bearing spalling state, effectively avoiding the subjectivity of manually setting thresholds and improving the accuracy and reliability of judging spalling damage of the inner and outer rings of the bearing.
[0111] Based on the same inventive concept as the above method, this application embodiment also provides a bearing inner and outer ring synchronous flaw detection system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described bearing inner and outer ring synchronous flaw detection methods.
[0112] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.
[0113] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.
[0114] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.
Claims
1. A method for simultaneous flaw detection of the inner and outer rings of a bearing, characterized in that, The method includes the following steps: Acquire a preset number of bearing images of the inner and outer rings of the same bearing; Extract all edge pixels from each bearing image, cluster all edge pixels, and denote the region containing the minimum bounding rectangle of each cluster as a local region; determine the directional feature coefficients of each bearing image based on the distribution of gradient direction angles of all pixels in each local region; determine the linear response intensity of each bearing image based on the variation characteristics of pixel values of all pixels in each local region in different gradient directions, and determine the microcrack evaluation index of each bearing image in combination with the directional feature coefficients. Based on the degree of disorder of pixel values of all pixels in each local region and the distribution of gradient direction angles of all pixels, the saliency of pitting features in each bearing image is determined; based on the brightness distribution and gradient amplitude dispersion of pixel values of all pixels in each local region, and in combination with the saliency of pitting features, the pitting evaluation index of each bearing image is determined; based on the microcrack evaluation index and the pitting evaluation index, the confidence level of bearing spalling damage in each bearing image is determined. Based on the confidence level of the bearing spalling damage, spalling damage is determined for the inner and outer rings of the bearing. The method for determining the linear response intensity of each bearing image is as follows: The maximum response intensity of each pixel in each local region is calculated using convolution kernels across all gradient directions. The average of the maximum response intensities of all pixels in each local region is denoted as the response index of each local region. The mean of the response indices of all local regions in each bearing image is taken as the linear response intensity of each bearing image. The expression for the saliency of the pit-like features in each bearing image is as follows: ; The significance of the pitting features in the k-th bearing image is represented. The information entropy represents the histogram of pixel values of all pixels in local region a in the k-th bearing image; The response index of local region a in the k-th bearing image is represented. This represents the number of all local regions in the k-th bearing image; represents the preset attenuation coefficient; exp[ ] represents the exponential function with the natural constant as the base.
2. The method for simultaneous flaw detection of the inner and outer rings of a bearing as described in claim 1, characterized in that, The method for determining the intra-cluster distance metric during the clustering process is as follows: The distance between edge pixels i and j in the k-th bearing image is measured. The expression is: In the formula, This represents the coordinate distance between edge pixel i and edge pixel j in the k-th bearing image; This represents the distance between edge pixel i and edge pixel j in the k-th bearing image, using the unit gradient vector.
3. The method for simultaneous flaw detection of the inner and outer rings of a bearing as described in claim 1, characterized in that, The method for determining the directional feature coefficients of each bearing image is as follows: In each local region of each bearing image, the cosine and sine values of the gradient direction angle of each pixel are calculated. The mean of the cosine value and the mean of the sine value of all pixels are calculated and denoted as the cosine mean and the sine mean, respectively. The square root of the sum of the squares of the cosine mean and the sine mean is denoted as the direction value of each local region. The result of subtracting the normalized value of the direction value of each local region from 1 is recorded as the direction index of each local region. The mean of the direction indices of all local regions in each bearing image is used as the direction feature coefficient of each bearing image.
4. The method for simultaneous flaw detection of the inner and outer rings of a bearing as described in claim 1, characterized in that, The microcrack evaluation index for each bearing image is the result of a linear positive fusion of the directional characteristic coefficient and the linear response intensity of each bearing image.
5. The method for simultaneous flaw detection of the inner and outer rings of a bearing as described in claim 1, characterized in that, The method for determining the pitting evaluation index for each bearing image is as follows: In each bearing image, the gray-level co-occurrence matrix contrast of all pixels in each local region is calculated and denoted as the contrast feature value of each local region. The mean of the contrast feature values of all local regions is calculated and denoted as the contrast index of each bearing image. The degree of dispersion of the gradient magnitude of all pixels in each local region is denoted as gradient dispersion. The mean of the gradient dispersion of all local regions in each bearing image is calculated and denoted as the gradient magnitude exponent of each bearing image. The contrast index, gradient magnitude index, and saliency of pit features of each bearing image are positively fused to form the pit evaluation index for each bearing image.
6. The method for simultaneous flaw detection of the inner and outer rings of a bearing as described in claim 1, characterized in that, The expression for the confidence level of bearing spalling damage in each bearing image is as follows: In the formula, This represents the confidence level of bearing spalling damage in the k-th bearing image; , These represent the microcrack evaluation index and the pitting evaluation index for the k-th bearing image, respectively.
7. The method for simultaneous flaw detection of the inner and outer rings of a bearing as described in claim 1, characterized in that, The determination of spalling damage on the inner and outer rings of the bearing includes: A threshold segmentation algorithm is used to select all bearing images other than those with only foreground or only background from a preset number of bearing images as bearing images to be detected. Cluster all bearing images to be detected, and determine the range of spalling damage and non-spalling damage based on the mean of the bearing spalling damage confidence scores of all bearing images to be detected in each cluster. Based on the range to which the bearing spalling damage confidence scores of the bearing images to be detected belong, determine whether spalling damage exists in the bearing images to be detected.
8. A bearing inner and outer ring synchronous flaw detection system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the bearing inner and outer ring synchronous flaw detection method as described in any one of claims 1-7.