Quality detection and analysis method used in grinding wheel production process

Abstract

The invention relates to the technical field of quality detection, in particular to a quality detection analysis method used in the grinding wheel production process. The method comprises the following steps: acquiring a target surface image of a grinding wheel to be detected and a surface feature value vector corresponding to each pixel point on the target surface image; according to a principal component analysis algorithm and the surface feature value vector, obtaining scatter points in a mapping space corresponding to the target surface image and pixel points corresponding to the scatter points; obtaining an initial bandwidth of a mean shift clustering algorithm, and clustering scattered points in the mapping space according to the initial bandwidth to obtain each target cluster; and obtaining an abnormal cluster according to the distance between the scatter points in the target cluster, obtaining an uneven area on the target surface image according to the pixel points corresponding to the scatter points in the abnormal cluster, and obtaining a quality detection result of the grinding wheel to be detected according to the uneven area. According to the invention, the accuracy of quality detection of the to-be-detected grinding wheel can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of quality detection, and in particular to a quality detection and analysis method used in a grinding wheel production process. Background Art

[0002] Grinding wheel is the core tool in the grinding process of various hard and brittle materials such as stone, glass, ceramics, semiconductor materials, etc., and its surface morphology has an important influence on the subsequent processing accuracy, surface quality and grinding performance. The surface morphology of the grinding wheel refers to the flatness of the grinding wheel surface, and the flatness of the grinding wheel surface can reflect the quality of the grinding wheel in the production process or the quality of the grinding wheel. If the area of ​​the uneven area on the grinding wheel surface does not meet the requirements, it also indicates that the quality of the grinding wheel does not meet the requirements. When the quality of the grinding wheel does not meet the requirements, it will lead to a decrease in the accuracy of the workpiece being processed, an increase in surface roughness, and even affect the stable operation of the equipment and the overall quality of the product. Therefore, it is very important to identify the uneven area on the grinding wheel surface for the quality detection accuracy of the grinding wheel.

[0003] At present, when detecting uneven areas on the surface of a grinding wheel, an edge detection algorithm is usually used to achieve it, and the edge detection algorithm is generally used to extract the contour and edge features of the grinding wheel surface, and the purpose of identifying the uneven areas on the grinding wheel surface is achieved by analyzing the extracted contour and edge features, such as analyzing edge features such as edge continuity and curvature change; however, since the surface of the grinding wheel is a non-mirror plane and its surface is granular, that is, there is a large amount of detail information on its surface, when the uneven features of the grinding wheel surface are relatively subtle or the unevenness is low, the uneven areas cannot be accurately detected, resulting in inaccurate problems when evaluating the quality of the grinding wheel or determining the quality grade of the grinding wheel. Therefore, it is urgent to solve the problem of identifying or detecting the uneven areas on the surface of the grinding wheel to improve the accuracy of evaluating the quality of the grinding wheel or determining the quality grade of the grinding wheel. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a quality detection and analysis method for a grinding wheel production process, and the technical solution adopted is as follows: An embodiment of the present invention provides a quality detection and analysis method for a grinding wheel production process, comprising the following steps: Acquire a target surface image of the grinding wheel to be detected and a surface eigenvalue vector corresponding to each pixel point on the target surface image; Obtaining scattered points in the mapping space corresponding to the target surface image and pixel points corresponding to the scattered points according to the principal component analysis algorithm and the surface eigenvalue vector; Acquire an initial bandwidth of a mean shift clustering algorithm, and cluster the scattered points in the mapping space according to the initial bandwidth to obtain target clusters; Obtaining abnormal clusters according to the distances between scattered points in the target clusters; According to the pixel points corresponding to the scattered points in the abnormal cluster, the uneven area on the target surface image is obtained, and the quality inspection result of the grinding wheel to be inspected is obtained according to the uneven area.

[0005] Beneficial effects: The present invention first obtains the target surface image of the grinding wheel to be detected and the surface eigenvalue vector corresponding to each pixel point on the target surface image; then, according to the principal component analysis algorithm and the surface eigenvalue vector, the scattered points in the mapping space corresponding to the target surface image and the pixel points corresponding to the scattered points are obtained; then, the initial bandwidth of the mean shift clustering algorithm is obtained, and according to the initial bandwidth, the scattered points in the mapping space are clustered to obtain each target clustering cluster; then, according to the distance between the scattered points in the target clustering cluster, the abnormal clustering cluster is obtained, and according to the pixel points corresponding to each scattered point in the abnormal clustering cluster, the uneven area on the target surface image is obtained, and the quality detection result of the grinding wheel to be detected is obtained according to the uneven area. The present invention can more accurately and reliably identify the uneven area on the target surface image based on the more accurate and reliable clustering result, and then can accurately determine the quality detection result of the grinding wheel to be detected, that is, the present invention can improve the accuracy of quality detection of the grinding wheel to be detected. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0007] Figure 1 A flow chart of a quality inspection and analysis method for a grinding wheel production process according to the present invention; Figure 2 It is the top view surface image of the grinding wheel to be inspected. DETAILED DESCRIPTION

[0008] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the embodiments of the present invention.

[0009] Unless defined otherwise, 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 invention pertains.

[0010] This embodiment provides a quality detection and analysis method for a grinding wheel production process, which is described in detail as follows: like Figure 1 As shown, the quality detection and analysis method used in the grinding wheel production process includes the following steps: Step S001, obtaining a target surface image of a grinding wheel to be inspected and a surface eigenvalue vector corresponding to each pixel point on the target surface image.

[0011] Since the surface of the grinding wheel is a non-mirror plane and has a granular feel, that is, there is a large amount of detail information on the surface, when the uneven features of the grinding wheel surface are relatively subtle or the unevenness is low, the uneven area cannot be accurately detected. That is, when the uneven area on the grinding wheel surface is currently detected or identified, there is a problem of inaccurate detection or identification, which will lead to the inability to accurately detect the quality of the grinding wheel surface. In order to ensure or improve the accuracy of identifying or detecting the uneven area on the grinding wheel surface and to achieve the purpose of accurately detecting the surface quality of the grinding wheel, this embodiment will first accurately identify the abnormal clusters, and then accurately identify the abnormal clusters on the target surface image based on the acquired abnormal clusters. The purpose of the flat area is to determine the quality inspection result of the grinding wheel to be inspected according to the identified uneven area, that is, the present embodiment will subsequently achieve the purpose of accurately inspecting the quality of the grinding wheel to be inspected through abnormal clustering clusters, and in this embodiment, before obtaining the abnormal clustering clusters, it is necessary to first obtain the target surface image of the grinding wheel to be inspected and the surface eigenvalue vector corresponding to each pixel point on the target surface image, and then obtain the target clustering cluster, and then determine the abnormal clustering cluster by analyzing the obtained target clustering cluster; and this step is mainly to obtain the target surface image of the grinding wheel to be inspected and the surface eigenvalue vector corresponding to each pixel point on the target surface image, that is, the specific acquisition process of the target surface image and the surface eigenvalue vector corresponding to each pixel point on the target surface image is: This embodiment mainly detects the grinding wheels that have been produced but have not been tested for unevenness; and after the grinding wheels are produced, they will be transported by a conveyor belt. Therefore, in this embodiment, a CCD camera is arranged directly above the conveyor belt, and the grinding wheels that are transported on the conveyor belt are recorded as grinding wheels to be detected; and in the process of the grinding wheels to be detected being transported, the CCD camera arranged directly above the conveyor belt will capture each grinding wheel to be detected, and the captured image will be recorded as a top view surface image of the grinding wheel to be detected, such as Figure 2 As shown, Figure 2It is a top-view surface image of any grinding wheel to be detected; then the top-view surface image is preprocessed, and the preprocessed image is recorded as the target surface image of the corresponding grinding wheel to be detected, and in this embodiment, the preprocessing includes but is not limited to denoising and grayscale processing of the image; and this embodiment requires that the acquisition frequency of the set camera can completely acquire the target surface image of each grinding wheel to be detected.

[0012] Therefore, this embodiment can obtain the target surface image of the grinding wheel to be detected through the above process; in addition, in order to facilitate subsequent analysis, this embodiment will take any grinding wheel to be detected as an example for analysis, that is, this embodiment is mainly for obtaining the quality inspection result of the grinding wheel to be detected.

[0013] After obtaining the target surface image of the grinding wheel to be detected, the present embodiment obtains the surface eigenvalue vector corresponding to each pixel point on the target surface image, and in the present embodiment, the surface eigenvalue vector corresponding to each pixel point is composed of the roughness, gradient value, grayscale value and texture eigenvalue of the corresponding pixel point, and the roughness, gradient value, grayscale value and texture eigenvalue are all standardized data; as another real-time method, the surface eigenvalue vector corresponding to each pixel point can also be composed of the gradient value, grayscale value and texture eigenvalue of the corresponding pixel point; the method for obtaining the gradient value of the pixel point is a well-known technology, so it will not be described in detail.

[0014] In addition, the method for obtaining the roughness of each pixel point is as follows: first, the local area of ​​each pixel point is obtained, such as the local area of ​​each pixel point is centered on each pixel point. Window area, and then obtain the three-dimensional contour data of the local area of ​​each pixel point through a high-precision measuring device (such as a laser scanner, an optical microscope, etc.), and then calculate the roughness parameters of the local area of ​​each pixel point according to the three-dimensional contour data of the local area of ​​each pixel point, such as calculating the arithmetic mean roughness corresponding to the local area of ​​each pixel point according to the three-dimensional contour data of the local area of ​​each pixel point, that is, an arithmetic mean roughness can be obtained based on the three-dimensional contour data of the local area of ​​a pixel point, and then the arithmetic mean roughness corresponding to the local area of ​​each pixel point is standardized, and the value after the standardized processing is used as the roughness of the corresponding pixel point; and for the case where the three-dimensional contour data cannot be directly obtained, the image processing technology can be used to indirectly evaluate the roughness of the surface, such as the surface roughness can be estimated by calculating the local standard deviation or variance of the grayscale image; and some detailed processes in the process of obtaining the roughness are well-known technologies, so this embodiment will not be described in detail.

[0015] And the method for obtaining the texture eigenvalue of each pixel point is: constructing a grayscale co-occurrence matrix according to the local area of ​​each pixel point, then obtaining the eigenvalue of the grayscale co-occurrence matrix, and standardizing the eigenvalue of the grayscale co-occurrence matrix, and using the value after the standardization as the texture eigenvalue of the corresponding pixel point, such as using the contrast of the grayscale co-occurrence matrix as the texture eigenvalue of the corresponding pixel point; and the method for obtaining the grayscale co-occurrence matrix and the eigenvalue of the grayscale co-occurrence matrix is ​​a well-known technology, so it will not be described in detail in this embodiment.

[0016] Therefore, this embodiment obtains the surface eigenvalue vector corresponding to each pixel point through the above process; and the purpose of obtaining the surface eigenvalue vector is to characterize the properties of the grinding wheel surface from more angles, so as to facilitate the subsequent acquisition of accurate non-planar areas.

[0017] Step S002, obtaining scattered points in a mapping space and pixel points corresponding to the scattered points according to a principal component analysis algorithm and the surface eigenvalue vector.

[0018] Since the dimensional features in the surface eigenvalue vector of the uneven area are different from those in the normal area, this embodiment will subsequently construct a mapping space by acquiring dimensional features with large distinguishing features, and acquire scattered points in the mapping space and pixel points corresponding to the scattered points. Subsequently, adaptive clustering is performed in the constructed mapping space to obtain the target cluster cluster. Therefore, this embodiment will then acquire scattered points in the mapping space and pixel points corresponding to the scattered points. The specific acquisition process is as follows: First, a principal component analysis is performed on the surface eigenvalue vectors corresponding to each pixel point on the target surface image to obtain all eigenvalues ​​and the eigenvectors corresponding to each eigenvalue, and the eigenvector corresponding to the maximum eigenvalue and the eigenvector corresponding to the eigenvalue next to the maximum eigenvalue are selected, and are recorded as the first eigenvector and the second eigenvector respectively; the space formed by the first eigenvector and the second eigenvector is recorded as the mapping space, the surface eigenvalue vectors corresponding to each pixel point are projected into the mapping space, and all points in the mapping space after the projection are completed are recorded as scattered points; and the process of obtaining the first eigenvector and the second eigenvector based on the principal component analysis algorithm is a well-known technology, so some specific steps in the acquisition process will not be described in detail in this embodiment.

[0019] For any scattered point a0 in the mapping space: if the scattered point a0 is obtained by projecting the surface eigenvalue vector corresponding to the pixel point a1 on the target surface image, then the pixel point a1 is recorded as the pixel point corresponding to the scattered point a0.

[0020] Therefore, through the above process, each scattered point in the mapping space and the pixel point corresponding to each scattered point can be obtained.

[0021] Step S003: obtaining an initial bandwidth of the mean shift clustering algorithm, and clustering the scattered points in the mapping space according to the initial bandwidth to obtain target clusters.

[0022] This embodiment will then use the mean shift clustering algorithm to cluster the scattered points in the mapping space to determine the target clustering cluster. However, in the clustering process, the size of the bandwidth directly affects the quality of the clustering result or directly affects the accuracy and reliability of the clustering result. For example, if the bandwidth is too large, multiple clusters will be clustered into one cluster, and if the bandwidth is too small, one cluster will be clustered into multiple clusters. Both of these phenomena will make the accuracy and reliability of the clustering result poor, and the clustering result will directly affect the subsequent accurate identification of uneven areas, and the identification of uneven areas will directly affect the quality detection results of the grinding wheel to be detected. In order to make the clustering result more accurate and reliable, this embodiment will next determine the target clustering cluster by adaptively adjusting the bandwidth. The specific acquisition process of the target clustering cluster is as follows: First, the initial bandwidth in the mean shift clustering algorithm is obtained. In this embodiment, an empirical value is selected as the initial bandwidth, such as setting the initial bandwidth to 0.3.

[0023] After obtaining the initial bandwidth of the mean shift clustering algorithm, the initial bandwidth is adaptively adjusted and iterated according to the initial bandwidth and the scattered points in the mapping space to determine each target clustering cluster, that is, according to the initial bandwidth, the scattered points in the mapping space are clustered to obtain each target clustering cluster, and each target clustering cluster is specifically obtained as follows: first, according to the initial bandwidth, all scattered points in the mapping space are clustered to obtain each initial clustering cluster, the center point of each initial clustering cluster and the access frequency of each scattered point in each initial clustering cluster; then, according to the distance between the center points of any two initial clustering clusters and the access frequency of each scattered point in each initial clustering cluster, each target clustering cluster is obtained; and according to the distance between the center points of any two initial clustering clusters and the access frequency of each scattered point in each initial clustering cluster, the process of obtaining each target clustering cluster is as follows: if the distance between the center points of any two initial clustering clusters is less than the preset distance, the two initial clustering clusters are merged, and also The access frequencies of the scattered points in the two initial clustering clusters are also merged, and the clusters after merging the initial clustering clusters that meet the merging conditions are recorded as merged clusters, and the unmerged initial clustering clusters and merged clusters are recorded as clusters to be analyzed. Since after the clusters to be analyzed are obtained, there may be a situation where a scattered point belongs to multiple clusters to be analyzed, then it is necessary to determine the belonging of this scattered point according to the access frequencies of each scattered point in each cluster to be analyzed. For example, if a scattered point belongs to two clusters to be analyzed at the same time, namely cluster L1 to be analyzed and cluster L2 to be analyzed, and the access frequency of the scattered point in cluster L1 to be analyzed is 3, and the access frequency in cluster L2 to be analyzed is 6, then it is determined that the scattered point belongs to cluster L2 to be analyzed. After finally determining the clusters to be analyzed to which all scattered points belong, the corresponding scattered points are divided into the corresponding clusters to be analyzed, and the clusters after the division are recorded as target clusters; and the preset distance needs to be set by the implementer according to the actual situation. For example, in this embodiment, the preset distance can be set to 0.2.

[0024] Therefore, this embodiment can obtain each target cluster through the above process; in addition, this embodiment only changes the process of obtaining the initial cluster, and after obtaining the initial cluster, the process of obtaining each target cluster according to the distance between the center points of any two initial clusters and the access frequency of each scattered point in each initial cluster is consistent with the process in the traditional mean shift clustering algorithm, and the target cluster is the final clustering result.

[0025] This embodiment will now specifically describe the process of clustering all scattered points in the mapping space according to the initial bandwidth to obtain each initial clustering cluster, the center point of each initial clustering cluster, and the access frequency of each scattered point in each initial clustering cluster, which is specifically: First, in the mapping space corresponding to the target surface image, any scattered point is selected as the starting center point Q0, and the initial bandwidth is recorded as the target bandwidth of the starting center point Q0. Then, according to the target bandwidth of the starting center point Q0, the neighborhood set C0 and the average offset vector D0 of the starting center point Q0 are obtained; then, it is determined whether the modulus of D0 is less than the preset offset threshold. If so, the cluster formed by the neighborhood set C0 is taken as the initial clustering cluster corresponding to the starting center point Q0, and the starting center point Q0 is taken as the center point of the initial clustering cluster corresponding to the starting center point Q0, and the access frequency of each scattered point in the initial clustering cluster corresponding to the starting center point Q0 is recorded as 1; otherwise, the starting center point Q0 is moved according to the size and direction in the average offset vector D0, and the position where the movement stops is recorded as the center point Q1 after the first move, and the target bandwidth of the starting center point Q0 is recorded as the bandwidth to be adjusted of the center point Q1, that is, if the modulus of D0 is not less than If the preset offset threshold is exceeded, the cluster formed by the neighborhood set C0 is not used as the initial clustering cluster corresponding to the starting center point Q0. Then, the local density of the center point Q1 is obtained according to the bandwidth to be adjusted of the center point Q1. The adjustment factor corresponding to the center point Q1 is obtained according to the local density of the center point Q1. The bandwidth to be adjusted of the center point Q1 is adjusted according to the adjustment factor corresponding to the center point Q1 to obtain the target bandwidth of the center point Q1. According to the target bandwidth of the center point Q1, the neighborhood set C1 and the average offset vector D1 of the center point Q1 are obtained, and it is determined whether the modulus length of D1 is less than the preset offset threshold. If so, the cluster formed by the neighborhood set C0 and the neighborhood set C1 is used as the initial clustering cluster corresponding to the starting center point Q0, and the center point Q1 is used as the center point of the initial clustering cluster corresponding to the starting center point Q0. According to the neighborhood set C0 and the neighborhood set C1, the access frequency of each scattered point in the initial clustering cluster corresponding to the starting center point Q0 is obtained.Otherwise, the center point Q1 is moved according to the size and direction of the average offset vector D1, and the position where the movement stops is recorded as the center point Q2 after the second movement, and the target bandwidth of the center point Q1 is used as the bandwidth to be adjusted of the center point Q2, that is, if the modulus of D1 is not less than the preset offset threshold, the cluster formed by the neighborhood set C0 and the neighborhood set C1 is not used as the initial clustering cluster corresponding to the starting center point Q0, and the local density of the center point Q2 is obtained according to the bandwidth to be adjusted of the center point Q2, and the adjustment factor corresponding to the center point Q2 is obtained according to the local density of the center point Q2. 2, adjust the bandwidth to be adjusted of the center point Q2, and obtain the target bandwidth of the center point Q2. According to the target bandwidth of the center point Q2, obtain the neighborhood set C2 and the average offset vector D2 of the center point Q2, and continue to judge whether the modulus of D2 is less than the preset offset threshold. If so, take the cluster formed by the neighborhood set C0, the neighborhood set C1 and the neighborhood set C2 as the initial clustering cluster corresponding to the starting center point Q0, take the center point Q2 as the center point of the initial clustering cluster corresponding to the starting center point Q0, and obtain the starting center according to the neighborhood set C0, the neighborhood set C1 and the neighborhood set C2. The access frequency of each scattered point in the initial clustering cluster corresponding to point Q0 is calculated; otherwise, the center point Q2 is moved according to the size and direction in the average offset vector D2, and the position where the movement stops is recorded as the center point Q3 after the third move, and the target bandwidth of the center point Q2 is used as the bandwidth to be adjusted of the center point Q3, that is, if the modulus length of D2 is not less than the preset offset threshold, the cluster formed by the neighborhood set C0, the neighborhood set C1 and the neighborhood set C2 is not used as the initial clustering cluster corresponding to the starting center point Q0, and the local density of the center point Q3 is obtained according to the bandwidth to be adjusted of the center point Q3. The local density of Q3 obtains the adjustment factor corresponding to the center point Q3. According to the adjustment factor corresponding to the center point Q3, the bandwidth to be adjusted of the center point Q3 is adjusted to obtain the target bandwidth of the center point Q3. According to the target bandwidth of the center point Q3, the neighborhood set C3 and the average offset vector D3 of the center point Q3 are obtained. By analogy, the initial cluster corresponding to the starting center point Q0, the center point of the initial cluster corresponding to the starting center point Q0, and the access frequency of each scattered point in the initial cluster corresponding to the starting center point Q0 are obtained, and the scattered points in the initial cluster corresponding to the starting center point Q0 are marked as visited. ;

[0026] Then, among all the scattered points that do not belong to the initial cluster corresponding to the starting center point Q0, select any scattered point as the new starting center point , and obtain the new starting center point according to the method of obtaining the initial cluster corresponding to the starting center point Q0, the center point of the initial cluster corresponding to the starting center point Q0, and the access frequency of each scattered point in the initial cluster corresponding to the starting center point Q0. The corresponding initial cluster, the new starting center point The corresponding initial cluster center point and the new starting center point The access frequency of each scattered point in the corresponding initial cluster will be located at the new starting center point The scattered points in the corresponding initial clusters are marked as visited, and so on, until there are no scattered points that are not marked as visited in the mapping space corresponding to the target surface image, and all the initial clusters, the center points of all the initial clusters, and the visit frequencies of each scattered point in all the initial clusters are obtained.

[0027] Therefore, this embodiment can obtain all the initial clusters, the center points of all the initial clusters and the access frequencies of each scattered point in all the initial clusters through the above process. In addition, in specific applications, the implementer needs to set the preset offset threshold according to the situation. For example, this embodiment can set the offset threshold to 0.5.

[0028] In this embodiment, according to the neighborhood set C0 and the neighborhood set C1, the method for obtaining the access frequency of each scattered point in the initial clustering cluster corresponding to the starting center point Q0 is: for any scattered point b in the initial clustering cluster corresponding to the starting center point Q0, if both the neighborhood set C0 and the neighborhood set C1 contain the scattered point b, then the access frequency of the scattered point b is recorded as 2; if only one of the neighborhood sets C0 and the neighborhood set C1 contains the scattered point b, then the access frequency of the scattered point b is recorded as 1.

[0029] In this embodiment, according to the neighborhood set C0, the neighborhood set C1 and the neighborhood set C2, the method for obtaining the access frequency of each scattered point in the initial clustering cluster corresponding to the starting center point Q0 is: for any scattered point b in the initial clustering cluster corresponding to the starting center point Q0, if the three sets of neighborhood set C0, neighborhood set C1 and neighborhood set C3 all contain the scattered point b, then the access frequency of scattered point b is recorded as 3; if only two of the three sets of neighborhood set C0, neighborhood set C1 and neighborhood set C3 contain the scattered point b, then the access frequency of scattered point b is recorded as 2; if only one of the three sets of neighborhood set C0, neighborhood set C1 and neighborhood set C3 contains the scattered point b, then the access frequency of scattered point b is recorded as 1.

[0030] In this embodiment, according to the target bandwidth of the starting center point Q0, the method of obtaining the neighborhood set C0 and the average offset vector D0 of the starting center point Q0 is as follows: first, the target bandwidth of the starting center point Q0 is recorded as , and then in the mapping space corresponding to the target surface image, the image at the starting center point Q0 is The set constructed by all scattered points within the neighborhood range is recorded as the neighborhood set C0 of the starting center point Q0; then the mean of the offset vectors of each scattered point in the neighborhood set C0 is obtained and recorded as the average offset vector D0 of the starting center point Q0; and in this embodiment, the size of the offset vector of any scattered point g in the neighborhood set C0 is the distance required for the starting center point Q0 to move to the scattered point g, and the direction of the offset vector of the scattered point g is the direction when the starting center point Q0 moves to the scattered point g.

[0031] In addition, in this embodiment, the method for obtaining the neighborhood set C1 and the average offset vector D1 of the center point Q1 according to the target bandwidth of the center point Q1, the method for obtaining the neighborhood set C2 and the average offset vector D2 of the center point Q2 according to the target bandwidth of the center point Q2, and the method for obtaining the neighborhood set C3 and the average offset vector D3 of the center point Q3 according to the target bandwidth of the center point Q3 are all the same as the method for obtaining the neighborhood set C0 and the average offset vector D0 of the starting center point Q0 according to the target bandwidth of the starting center point Q0, and therefore, this embodiment will not be described in detail.

[0032] In this embodiment, the method for obtaining the local density is as follows: for the center point Q1, the kernel density obtained according to the kernel function and the bandwidth to be adjusted of the center point Q1 is recorded as the local density corresponding to the center point Q1; for the center point Q2, the kernel density obtained according to the kernel function and the bandwidth to be adjusted of the center point Q2 is recorded as the local density corresponding to the center point Q2; for the center point Q3, the kernel density obtained according to the kernel function and the bandwidth to be adjusted of the center point Q3 is recorded as the local density corresponding to the center point Q3; in addition, for ease of understanding, this embodiment will be described by taking the calculation formula of the local density corresponding to the center point Q1 as an example, that is, the calculation formula for obtaining the local density corresponding to the center point Q1 is: in, is the local density corresponding to the center point Q1, is the bandwidth to be adjusted of the center point Q1, M0 is the total number of scattered points in the initial scattered point set of the center point Q1, and the initial scattered point set of the center point Q1 refers to the points located at the center point Q1. The set constructed by all scattered points in the neighborhood, is the Euclidean distance between the jth scattered point in the initial scattered point set of the center point Q1 and the center point Q1, K() is a kernel function, and in this embodiment, K() is a Gaussian kernel function; in addition, the local density calculated above is essentially a kernel density.

[0033] If the local density corresponding to the center point Q2 needs to be obtained, then the bandwidth to be adjusted of the center point Q1 in the above local density calculation formula needs to be replaced with the bandwidth to be adjusted of the center point Q2, and the value of M0 needs to be replaced with the total number of scattered points in the initial scattered point set of the center point Q2. At this time, the initial scattered point set of the center point Q2 is also determined based on the bandwidth to be adjusted of the center point Q2, that is, the initial scattered point set of the center point Q2 is located at the center point Q2. The set constructed by all scattered points in the neighborhood, is the bandwidth to be adjusted at the center point Q2. is the Euclidean distance between the jth scattered point in the initial scattered point set of the center point Q2 and the center point Q2; similarly, the calculation method of the local density corresponding to the center point Q3 is the same as the calculation method of the local density corresponding to the center point Q1, so it will not be described in detail.

[0034] In this embodiment, the method for obtaining the adjustment factor is: first, obtain the to-be-mapped value corresponding to the center point after each movement, and then perform negative correlation mapping on the to-be-mapped value corresponding to the center point after each movement, and record the value obtained after the negative correlation mapping as the adjustment factor corresponding to the corresponding center point; in addition, this embodiment mainly adapts the adjustment factor based on the stability of the local density change, and the more uniform the local density change, the larger the adjustment factor, and the more uneven the local density change, the smaller the adjustment factor, which can avoid the existence of adjustment abnormalities; and in this embodiment, the method for obtaining the to-be-mapped value corresponding to the center point after the i-th movement is: first, obtain the first local density sequence and the corresponding second local density sequence corresponding to the center point after the i-th movement, and record the mean value of the first local density sequence corresponding to the center point after the i-th movement as the value after the i-th movement. The mean of the first density corresponding to the center point, the mean of the second local density sequence corresponding to the center point after the i-th move is recorded as the mean of the second density corresponding to the center point after the i-th move, and the first square difference sequence of the first local density sequence and the second square difference sequence of the second local density sequence are obtained. The absolute value of the difference between the mean of the first square difference set and the mean of the second square difference set is used as the value to be mapped corresponding to the center point after the i-th move, and the x1-th first square difference value in the first square difference sequence of the first local density sequence is the square value of the difference between the x1-th first local density in the first local density sequence and the mean of the first density, and the x2-th second square difference value in the second square difference sequence of the second local density sequence is the square value of the difference between the x2-th second local density in the second local density sequence and the mean of the second density.

[0035] And the method for obtaining the first local density sequence and the second local density sequence corresponding to the center point after the i-th movement is: record the sequence constructed from the local density corresponding to the center point after the 1st movement to the local density corresponding to the center point after the i-th movement as the first local density sequence corresponding to the center point after the i-th movement, that is, the number of data in the first local density sequence at this time is i; record the sequence constructed from the local density corresponding to the center point after the 1st movement to the local density corresponding to the center point after the i-1th movement as the second local density sequence corresponding to the center point after the i-th movement, that is, the number of data in the second local density sequence at this time is i-1; in addition, the value of i is not equal to 1, that is, the above process cannot calculate the value to be mapped corresponding to the center point after the first movement, so it is impossible to obtain the adjustment factor corresponding to the center point Q1 after the first movement, and in order to facilitate subsequent calculations, this embodiment uses the preset first constant as the adjustment factor corresponding to the center point Q1 after the first movement.

[0036] For example, in this embodiment, the first local density sequence corresponding to the center point Q2 after the second movement is composed of the local density of the center point Q2 and the local density of the center point Q1, and the second local density sequence corresponding to the center point Q2 after the second movement is composed of the local density of the center point Q1.

[0037] In this embodiment, the method for obtaining the target bandwidth of the center point Q1 is: first obtain the global density, and then record the ratio of the local density of the center point Q1 to the global density as the relative density corresponding to the center point Q1; then determine whether the relative density corresponding to the center point Q1 belongs to the preset relative density range. If so, it indicates that there is no need to adjust the bandwidth, that is, the bandwidth to be adjusted of the center point Q1 is directly recorded as the target bandwidth of the center point Q1. Otherwise, it indicates that the bandwidth to be adjusted of the center point Q1 needs to be adjusted according to the bandwidth to be adjusted, the adjustment factor and the relative density corresponding to the center point Q1 to obtain the target bandwidth of the center point Q1.

[0038] And the method for obtaining the global density is: first, the set constructed by all scattered points in the mapping space corresponding to the target surface image is recorded as a comprehensive set, and then the first distance set corresponding to the comprehensive set is obtained, and the mean of the first distance set is recorded as the first feature mean, and then the first feature mean is negatively correlated and mapped, and the value obtained after the negative correlation mapping is recorded as the global density.

[0039] In addition, the method for obtaining the first distance set corresponding to the above-mentioned comprehensive set is as follows: first, the total number of scattered points in the comprehensive set is obtained, and recorded as N0; then, any scattered point in the comprehensive set is selected as the first scattered point d1, and the set constructed by all scattered points in the comprehensive set except the first scattered point is recorded as the first set corresponding to the first scattered point d1; then, in the first set, the scattered point closest to the first scattered point is obtained, and the obtained scattered point closest to the first scattered point is recorded as the nearest neighbor scattered point of the first scattered point, and the Euclidean distance between the first scattered point and the nearest neighbor scattered point of the first scattered point is recorded as the first distance of the first scattered point, and the nearest neighbor scattered point of the first scattered point is taken as the second scattered point, and the set constructed by all scattered points in the first set except the second scattered point is recorded as the second set corresponding to the second scattered point; and continue to obtain the scattered point closest to the second scattered point in the second set, and record the obtained distance The scattered point closest to the second scattered point is recorded as the nearest neighbor scattered point of the second scattered point, the Euclidean distance between the second scattered point and the nearest neighbor scattered point of the second scattered point is recorded as the first distance of the second scattered point, the nearest neighbor scattered point of the second scattered point is recorded as the third scatter point, and the set constructed by all scattered points in the second set except the second scattered point is recorded as the third set corresponding to the third scatter point, and so on, until the (N0-1)th set corresponding to the (N0-1)th scatter point is obtained, and the Euclidean distance between the (N0-1)th scatter point and the scatter points in the (N0-1)th set is recorded as the first distance of the (N0-1)th scatter point, that is, the number of scattered points in the (N0-1)th set corresponding to the (N0-1)th scatter point is 1; then the set constructed by the first distances corresponding to all scatter points is recorded as the first distance set corresponding to the comprehensive set, that is, the number of first distances in the first distance set is N0-1.

[0040] In this embodiment, the bandwidth to be adjusted of the center point Q1 is adjusted according to the bandwidth to be adjusted, the adjustment factor and the relative density corresponding to the center point Q1, and the method for obtaining the target bandwidth of the center point Q1 is: first, the minimum value in the preset relative density interval is recorded as the minimum preset value, and the maximum value in the preset relative density interval is recorded as the maximum preset value, and the value obtained by subtracting the adjustment factor corresponding to the center point Q1 from the preset first constant is recorded as the first indicator value corresponding to the center point Q1. Then, it is determined whether the relative density corresponding to the center point Q1 is less than the minimum preset value. If so, it indicates that the local density of the center point Q1 is less than the global density or the density of the area where the data points are located is low. In this case, the bandwidth needs to be increased to avoid forming too many small clusters or isolated points. That is, the bandwidth to be adjusted corresponding to the center point Q1 needs to be increased or the increased characterization value of the bandwidth to be adjusted corresponding to the center point Q1 needs to be obtained, and the sum of the bandwidth to be adjusted corresponding to the center point Q1 and the increased characterization value is recorded as the target bandwidth of the center point Q1; and if it is determined that the relative density corresponding to the center point Q1 is greater than the maximum preset value, it indicates that the local density of the center point Q1 is greater than the global density or the density of the area where the data points are located is high. In this case, the bandwidth needs to be reduced in order to divide the clusters more finely, that is, the bandwidth to be adjusted corresponding to the center point Q1 needs to be reduced or the reduced characterization value of the bandwidth to be adjusted corresponding to the center point Q1 needs to be obtained, and the value obtained by subtracting the reduced characterization value from the bandwidth to be adjusted corresponding to the center point Q1 is recorded as the target bandwidth of the center point Q1.

[0041] In specific applications, the implementer needs to set the preset first constant and the preset relative density interval according to actual conditions. For example, in this embodiment, the preset first constant is set to 1, and the interval [0.8, 1.2] is used as the preset relative density interval.

[0042] In addition, the above process of obtaining the increased characterization value of the bandwidth to be adjusted corresponding to the center point Q1 is as follows: the ratio of the relative density corresponding to the center point Q1 to the minimum preset value is recorded as the density ratio, the value obtained by subtracting the preset first constant from the density ratio and then multiplying the value obtained by the adjustment factor corresponding to the center point Q1 is recorded as the first product value, and the sum of the first product value and the first index value is recorded as the second index value corresponding to the center point Q1; the product of the second index value corresponding to the center point Q1 and the bandwidth to be adjusted corresponding to the center point Q1 is recorded as the increased characterization value of the bandwidth to be adjusted corresponding to the center point Q1; and in this embodiment, the specific expression for calculating the increased characterization value corresponding to the center point Q1 is: ,in, is the increased representation value of the bandwidth to be adjusted corresponding to the center point Q1, is the bandwidth to be adjusted corresponding to the center point Q1, is the adjustment factor corresponding to the center point Q1, is the relative density corresponding to the center point Q1, is the minimum preset value, and the minimum preset value in this embodiment is 0.8; is the density ratio, is the first product value, is the first index value corresponding to the center point Q1, is the second index value corresponding to the center point Q1; and when The bigger it is, The larger the value of Add adjustment factor The purpose is to limit the value obtained by adding the characterization value to the bandwidth to be adjusted corresponding to the center point Q1 to be no greater than twice the bandwidth to be adjusted corresponding to the center point Q1.

[0043] The above process of obtaining the reduced characterization value of the bandwidth to be adjusted corresponding to the center point Q1 is as follows: the value obtained after negative correlation mapping of the relative density corresponding to the center point Q1 is recorded as the mapped relative density, the value obtained after subtracting the preset first constant from the mapped relative density is multiplied by the adjustment factor corresponding to the center point Q1, and recorded as the second product value, and then the sum of the second product value and the first index value is recorded as the third index value; the value obtained by multiplying the third index value by the bandwidth to be adjusted corresponding to the center point Q1 and then dividing it by the preset second constant is recorded as the reduced characterization value of the bandwidth to be adjusted corresponding to the center point Q1. In this embodiment, the specific expression for calculating the reduced characterization value corresponding to the center point Q1 is: ,in, is the reduced characterization value of the bandwidth to be adjusted corresponding to the center point Q1, w2 is a preset second constant, the value of the preset second constant in this embodiment is 2, exp() is an exponential function with the natural constant e as the base, and the method adopted for all negative correlation mappings in this embodiment is the same as the method of performing negative correlation mapping on the relative density corresponding to the center point Q1; is the relative density map, is the second product value, is the third index value corresponding to the center point Q1; and when The bigger it is, The larger the value of Add adjustment factor The purpose is to limit the maximum value of the characterization value to be less than or equal to half of the bandwidth to be adjusted corresponding to the center point Q1, that is, to calculate Also add the adjustment factor The purpose is to require that the value of the bandwidth to be adjusted corresponding to the center point Q1 minus the reduced characterization value is not less than half of the bandwidth to be adjusted corresponding to the center point Q1.

[0044] In this embodiment, the methods for obtaining the target bandwidths of other center points are the same as the method for obtaining the target bandwidth of the center point Q1. For example, when obtaining the target bandwidth of the center point Q2, the relative density corresponding to the center point Q2 is first obtained. The relative density corresponding to the center point Q2 is the ratio of the local density of the center point Q2 to the global density. It is then determined whether the relative density corresponding to the center point Q2 belongs to a preset relative density interval. If so, the bandwidth to be adjusted of the center point Q2 is recorded as the target bandwidth of the center point Q2. Otherwise, the bandwidth to be adjusted of the center point Q2 is adjusted according to the bandwidth to be adjusted, the adjustment factor, and the relative density corresponding to the center point Q2. The target bandwidth of the center point Q2 is obtained by adjusting the bandwidth to be adjusted of the center point Q2 according to the bandwidth to be adjusted, the adjustment factor, and the relative density corresponding to the center point Q2 to obtain the target bandwidth of the center point Q2. The method is the same as the method of adjusting the bandwidth to be adjusted of the center point Q1 according to the bandwidth to be adjusted, the adjustment factor, and the relative density corresponding to the center point Q1 to obtain the target bandwidth of the center point Q1, except that the increase characterization value or the decrease characterization value corresponding to the center point Q1 is calculated by using the bandwidth to be adjusted, the adjustment factor, and the relative density corresponding to the center point Q2. Therefore, this embodiment will not be described in detail.

[0045] Therefore, this embodiment can obtain each target cluster more reliably and accurately through the above process.

[0046] Step S004, obtaining an abnormal cluster according to the distance between the scattered points in the target cluster; obtaining an uneven area on the target surface image according to the pixel points corresponding to each scattered point in the abnormal cluster, and obtaining a quality inspection result of the grinding wheel to be inspected according to the uneven area.

[0047] After obtaining the target cluster, this embodiment will obtain the abnormal cluster density according to the distance between the scattered points in the target cluster. The abnormal cluster density is the key to accurately obtain the uneven area in the subsequent step. The specific process of obtaining the abnormal cluster density is as follows: For any target cluster V: the set constructed by all scattered points in the target cluster V is recorded as the set to be analyzed corresponding to the target cluster V, and then the first distance set corresponding to the set to be analyzed is obtained, and the method for obtaining the first distance set corresponding to the set to be analyzed is the same as the method for obtaining the first distance set corresponding to the comprehensive set, so it is not described in detail; then the statistical histogram of the first distance set corresponding to the set to be analyzed is obtained, and according to the value interval of each group on the statistical histogram and the frequency of each group, the points to be connected corresponding to each group are obtained, and the points to be connected corresponding to each group are connected in turn, and the connected curve is recorded as the statistical curve, and the horizontal coordinate value of the points to be connected corresponding to each group is the middle value of the value interval of the corresponding group, the horizontal coordinate value of the points to be connected corresponding to each group is the middle value of the value interval of the corresponding group, and the horizontal coordinate value of the points to be connected corresponding to each group is the middle value of the value interval of the corresponding group. The vertical axis value is the frequency of the corresponding group; then the normal distribution curve of the first distance set corresponding to the set to be analyzed is obtained, and the process of obtaining the statistical histogram and the normal distribution curve is a well-known technology, so it is not described in detail; then the similarity between the statistical curve and the normal distribution curve is calculated, and it is determined whether the similarity between the statistical curve and the normal distribution curve is greater than a preset similarity threshold. If so, the target cluster cluster V is recorded as a normal cluster cluster, otherwise, the target cluster cluster V is recorded as an abnormal cluster cluster; and in this embodiment, the implementer needs to determine the method of obtaining the similarity between the statistical curve and the normal distribution curve according to the actual situation. For example, in this embodiment, the Pearson correlation coefficient between the statistical curve and the normal distribution curve is used as the similarity between the statistical curve and the normal distribution curve.

[0048] In addition, in a specific application, the implementer needs to set a preset similarity threshold according to actual conditions. For example, in this embodiment, the preset similarity threshold is set to 0.75.

[0049] Therefore, this embodiment can obtain various abnormal clusters through the above process, and if there is no abnormal cluster in the target cluster, it is determined that there is no uneven area on the target surface image of the grinding wheel to be detected, that is, it is determined that the grinding wheel to be detected meets the quality requirements; and if there is an abnormal cluster, then according to the pixel points corresponding to each scattered point in the abnormal cluster, the various uneven areas on the target surface image are obtained. The specific acquisition process is: for any abnormal cluster P, the area formed by the pixel points corresponding to each scattered point in the abnormal cluster P is recorded as an uneven area on the target surface image, or the convex hull detection is performed on the pixel point set composed of the pixel points corresponding to all scattered points in the abnormal cluster P, and the area obtained by the detection is used as an uneven area on the target surface image.

[0050] Therefore, all uneven areas on the target surface image can be obtained through the above process. Next, the quality inspection result of the grinding wheel to be inspected is obtained according to each uneven area on the target surface image, specifically: the area of ​​each uneven area on the target surface image is obtained, the maximum area is selected as the area to be determined, and then the quality grade value of the grinding wheel to be inspected is obtained according to the area to be determined. The larger the value of the area to be determined, the larger the quality grade value of the grinding wheel to be inspected. The larger the quality grade value of the grinding wheel to be inspected, the more it indicates that the grinding wheel to be inspected does not meet the quality requirements, that is, the larger the quality grade value of the grinding wheel to be inspected, the greater the probability that the grinding wheel to be inspected does not meet the quality requirements or the worse the morphology quality of the grinding wheel to be inspected.

[0051] In summary, the present embodiment first obtains the target surface image of the grinding wheel to be detected and the surface eigenvalue vector corresponding to each pixel point on the target surface image; then, according to the principal component analysis algorithm and the surface eigenvalue vector, the scattered points in the mapping space corresponding to the target surface image and the pixel points corresponding to the scattered points are obtained; then, the initial bandwidth of the mean shift clustering algorithm is obtained, and according to the initial bandwidth, the scattered points in the mapping space are clustered to obtain each target clustering cluster; then, according to the distance between the scattered points in the target clustering cluster, the abnormal clustering cluster is obtained, and according to the pixel points corresponding to each scattered point in the abnormal clustering cluster, the uneven area on the target surface image is obtained, and the quality detection result of the grinding wheel to be detected is obtained according to the uneven area. Based on the relatively accurate and reliable clustering results, the present embodiment can more accurately and reliably identify the uneven area on the target surface image, and then can accurately determine the quality detection result of the grinding wheel to be detected, that is, the present embodiment can improve the accuracy of quality detection of the grinding wheel to be detected.

[0052] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A quality detection and analysis method for a grinding wheel production process, characterized in that: The method comprises the following steps: Acquire a target surface image of the grinding wheel to be detected and a surface eigenvalue vector corresponding to each pixel point on the target surface image; Obtaining scattered points in the mapping space corresponding to the target surface image and pixel points corresponding to the scattered points according to the principal component analysis algorithm and the surface eigenvalue vector; Acquire an initial bandwidth of a mean shift clustering algorithm, and cluster the scattered points in the mapping space according to the initial bandwidth to obtain target clusters; Obtaining abnormal clusters according to the distances between scattered points in the target clusters; According to the pixel points corresponding to the scattered points in the abnormal cluster, the uneven area on the target surface image is obtained, and the quality inspection result of the grinding wheel to be inspected is obtained according to the uneven area.

2. A quality detection and analysis method for grinding wheel production according to claim 1, characterized in that: The surface eigenvalue vector corresponding to the pixel point is composed of the roughness, gradient value, grayscale value, and texture eigenvalue of the corresponding pixel point.

3. A quality detection and analysis method for grinding wheel production process as claimed in claim 1, characterized in that: The method for acquiring scattered points in the mapping space and pixel points corresponding to the scattered points includes: Performing principal component analysis on the surface eigenvalue vectors corresponding to each pixel point on the target surface image to obtain all eigenvalues ​​and eigenvectors corresponding to each eigenvalue, selecting the eigenvector corresponding to the maximum eigenvalue and the eigenvector corresponding to the eigenvalue next to the maximum eigenvalue, and recording them as the first eigenvector and the second eigenvector, respectively; recording the space formed by the first eigenvector and the second eigenvector as a mapping space, projecting the surface eigenvalue vectors corresponding to each pixel point into the mapping space, and recording all points in the mapping space after the projection is completed as scattered points; For any scattered point a0 in the mapping space: if the scattered point a0 is obtained by projecting the surface eigenvalue vector corresponding to the pixel point a1 on the target surface image, then the pixel point a1 is recorded as the pixel point corresponding to the scattered point a0.

4. A quality detection and analysis method for grinding wheel production process as claimed in claim 1, characterized in that: The method of clustering the scattered points in the mapping space according to the initial bandwidth to obtain each target cluster cluster includes: According to the initial bandwidth, all scattered points in the mapping space are clustered to obtain each initial clustering cluster, the center point of each initial clustering cluster and the access frequency of each scattered point in each initial clustering cluster; according to the distance between the center points of any two initial clustering clusters and the access frequency of each scattered point in each initial clustering cluster, each target clustering cluster is obtained.

5. A quality detection and analysis method for grinding wheel production as claimed in claim 4, characterized in that: The method for acquiring the access frequency of each initial cluster, the center point of each initial cluster, and each scattered point in each initial cluster includes: In the mapping space, any scattered point is selected as the starting center point Q0, and the initial bandwidth is recorded as the target bandwidth of the starting center point Q0. According to the target bandwidth of the starting center point Q0, the neighborhood set C0 and the average offset vector D0 of the starting center point Q0 are obtained; it is determined whether the modulus of D0 is less than the preset offset threshold value. If so, the cluster formed by the neighborhood set C0 is used as the initial clustering cluster corresponding to the starting center point Q0, and the starting center point Q0 is used as the center point of the initial clustering cluster corresponding to the starting center point Q0, and the access frequency of each scattered point in the initial clustering cluster corresponding to the starting center point Q0 is recorded as 1; otherwise, the starting center point Q0 is converted into a large cluster according to the average offset vector D0. The center point Q1 moves in the direction of the first movement, and the position where the movement stops is recorded as the center point Q1 after the first movement. The target bandwidth of the starting center point Q0 is recorded as the bandwidth to be adjusted of the center point Q1. The local density of the center point Q1 is obtained according to the bandwidth to be adjusted of the center point Q1. The adjustment factor corresponding to the center point Q1 is obtained according to the local density of the center point Q1. The bandwidth to be adjusted of the center point Q1 is adjusted according to the adjustment factor corresponding to the center point Q1 to obtain the target bandwidth of the center point Q1. According to the target bandwidth of the center point Q1, the neighborhood set C1 and the average offset vector D1 of the center point Q1 are obtained, and it is determined whether the modulus length of D1 is less than the preset offset threshold. If so, the neighborhood set C1 and the average offset vector D1 are adjusted. The cluster formed by the domain set C0 and the neighborhood set C1 is used as the initial cluster cluster corresponding to the starting center point Q0, and the center point Q1 is used as the center point of the initial cluster cluster corresponding to the starting center point Q0, and the access frequency of each scattered point in the initial cluster cluster corresponding to the starting center point Q0 is obtained according to the neighborhood set C0 and the neighborhood set C1; otherwise, the center point Q1 is moved according to the size and direction in the average offset vector D1, and the position where the movement stops is recorded as the center point Q2 after the second movement, and the target bandwidth of the center point Q1 is used as the bandwidth to be adjusted of the center point Q2, and the local density of the center point Q2 is obtained according to the bandwidth to be adjusted of the center point Q2, and the local density of the center point Q2 is obtained according to the bandwidth to be adjusted of the center point Q2. The local density of the center point Q2 obtains the adjustment factor corresponding to the center point Q2. According to the adjustment factor corresponding to the center point Q2, the bandwidth to be adjusted of the center point Q2 is adjusted to obtain the target bandwidth of the center point Q2. According to the target bandwidth of the center point Q2, the neighborhood set C2 and the average offset vector D2 of the center point Q2 are obtained, and it is further determined whether the modulus of D2 is less than the preset offset threshold. By analogy, the initial cluster corresponding to the starting center point Q0, the center point of the initial cluster corresponding to the starting center point Q0, and the access frequency of each scattered point in the initial cluster corresponding to the starting center point Q0 are obtained, and the scattered points in the initial cluster corresponding to the starting center point Q0 are marked as visited; Continue to select a scattered point from all the scattered points that do not belong to the initial cluster corresponding to the starting center point Q0 as the new starting center point , get the starting center point The corresponding initial cluster, the starting center point The center point of the corresponding initial cluster and the starting center point The access frequency of each scattered point in the corresponding initial cluster will be located at the starting center point The scattered points in the corresponding initial clusters are marked as visited; and so on, the process stops when there are no scattered points in the mapping space that are not marked as visited, and all the initial clusters, the center points of all the initial clusters, and the visit frequencies of the scattered points in all the initial clusters are obtained; The method for obtaining the access frequency of each scattered point in the initial cluster corresponding to the starting center point Q0 according to the neighborhood set C0 and the neighborhood set C1 includes: for any scattered point b in the initial cluster corresponding to the starting center point Q0, if the scattered point b is included in both the neighborhood set C0 and the neighborhood set C1, the access frequency of the scattered point b is recorded as 2; if only one of the neighborhood sets C0 and the neighborhood set C1 contains the scattered point b, the access frequency of the scattered point b is recorded as 1; According to the target bandwidth of the starting center point Q0, a method for obtaining a neighborhood set C0 and an average offset vector D0 of the starting center point Q0 includes: recording the target bandwidth of the starting center point Q0 as , the starting center point Q0 A set constructed by all scattered points within the neighborhood range is recorded as the neighborhood set C0 of the starting center point Q0, and the mean of the offset vectors of each scattered point in the neighborhood set C0 is obtained and recorded as the average offset vector D0 of the starting center point Q0; the size of the offset vector of any scattered point g in the neighborhood set C0 is the distance required for the starting center point Q0 to move to the scattered point g, and the direction of the offset vector of the scattered point g is the direction when the starting center point Q0 moves to the scattered point g; The method for obtaining the local density includes: for the center point Q1, the kernel density obtained according to the kernel function and the bandwidth to be adjusted of the center point Q1 is recorded as the local density corresponding to the center point Q1; for the center point Q2, the kernel density obtained according to the kernel function and the bandwidth to be adjusted of the center point Q2 is recorded as the local density corresponding to the center point Q2.

6. A quality detection and analysis method for grinding wheel production as claimed in claim 5, characterized in that: The method for obtaining the adjustment factor comprises: Obtain the to-be-mapped value corresponding to the center point after each movement, perform negative correlation mapping on the to-be-mapped value corresponding to the center point after each movement, and record the value obtained after the negative correlation mapping as the adjustment factor corresponding to the corresponding center point; For the center point after the i-th movement, i is not equal to 1, the method for obtaining the to-be-mapped value corresponding to the center point after the i-th movement includes: obtaining a first local density sequence and a second local density sequence corresponding to the center point after the i-th movement, and recording the mean of the first local density sequence as the first density mean, and recording the mean of the second local density sequence as the second density mean, obtaining a first square difference sequence of the first local density sequence and a second square difference sequence of the second local density sequence, and taking the absolute value of the difference between the mean of the first square difference set and the mean of the second square difference set as the to-be-mapped value corresponding to the center point after the i-th movement; the x1th first square difference value in the first square difference sequence of the first local density sequence is the square value of the difference between the x1th first local density in the first local density sequence and the first density mean, and the x2th second square difference value in the second square difference sequence of the second local density sequence is the square value of the difference between the x2th second local density in the second local density sequence and the second density mean; The first local density sequence and the second local density sequence are obtained by: a sequence constructed from the local density corresponding to the center point after the first movement to the local density corresponding to the center point after the i-th movement is recorded as the first local density sequence, and a sequence constructed from the local density corresponding to the center point after the first movement to the local density corresponding to the center point after the i-1-th movement is recorded as the second local density sequence; The preset first constant is used as the adjustment factor corresponding to the center point Q1 after the first movement.

7. A quality detection and analysis method for grinding wheel production process as claimed in claim 6, characterized in that: The method for acquiring the target bandwidth of the center point Q1 includes: Recording a set constructed by all scattered points in the mapping space as a comprehensive set, obtaining a first distance set corresponding to the comprehensive set, recording a mean of the first distance set as a first feature mean, performing negative correlation mapping on the first feature mean, and recording a value obtained by the negative correlation mapping as a global density; The ratio of the local density of the center point Q1 to the global density is recorded as the relative density corresponding to the center point Q1; whether the relative density corresponding to the center point Q1 belongs to the preset relative density interval is determined; if so, the bandwidth to be adjusted of the center point Q1 is recorded as the target bandwidth of the center point Q1; otherwise, the bandwidth to be adjusted of the center point Q1 is adjusted according to the bandwidth to be adjusted, the adjustment factor and the relative density corresponding to the center point Q1 to obtain the target bandwidth of the center point Q1; The method for obtaining the first distance set corresponding to the comprehensive set includes: recording the number of scattered points in the comprehensive set as N0; recording any scattered point in the comprehensive set as the first scattered point, and recording the set constructed by all scattered points in the comprehensive set except the first scattered point as the first set corresponding to the first scattered point; in the first set, obtaining the nearest neighbor scattered point of the first scattered point, and recording the Euclidean distance between the first scattered point and the nearest neighbor scattered point of the first scattered point as the first distance of the first scattered point, recording the nearest neighbor scattered point of the first scattered point as the second scattered point, and recording the set constructed by all scattered points in the first set except the second scattered point as the second set corresponding to the second scattered point; and continuing to obtain in the second set, The nearest neighbor scatter point of the second scatter point, and the Euclidean distance between the second scatter point and the nearest neighbor scatter point of the second scatter point is recorded as the first distance of the second scatter point, the nearest neighbor scatter point of the second scatter point is recorded as the third scatter point, and the set constructed by all scatter points in the second set except the third scatter point is recorded as the third set corresponding to the third scatter point, and so on, until the (N0-1)th set corresponding to the (N0-1)th scatter point is obtained, and the Euclidean distance between the (N0-1)th scatter point and the scatter points in the (N0-1)th set is recorded as the first distance of the (N0-1)th scatter point; the set constructed by the first distances corresponding to all the obtained scatter points is recorded as the first distance set corresponding to the comprehensive set.

8. A quality detection and analysis method for grinding wheel production process as claimed in claim 7, characterized in that: The method for obtaining the target bandwidth of the center point Q1 according to the bandwidth to be adjusted, the adjustment factor, and the relative density corresponding to the center point Q1 includes: The minimum value in the preset relative density interval is recorded as the minimum preset value, and the maximum value in the preset relative density interval is recorded as the maximum preset value; the difference between the preset first constant and the adjustment factor corresponding to the center point Q1 is recorded as the first index value corresponding to the center point Q1; If it is determined that the relative density corresponding to the center point Q1 is less than the minimum preset value, the increase characterization value of the bandwidth to be adjusted corresponding to the center point Q1 is obtained, and the sum of the bandwidth to be adjusted corresponding to the center point Q1 and the increase characterization value is recorded as the target bandwidth of the center point Q1; if it is determined that the relative density corresponding to the center point Q1 is greater than the maximum preset value, the decrease characterization value of the bandwidth to be adjusted corresponding to the center point Q1 is obtained, and the value obtained by subtracting the decrease characterization value from the bandwidth to be adjusted corresponding to the center point Q1 is recorded as the target bandwidth of the center point Q1; The method for obtaining the increased characterization value of the bandwidth to be adjusted corresponding to the center point Q1 includes: recording the ratio of the relative density corresponding to the center point Q1 to the minimum preset value as the density ratio, multiplying the value obtained by subtracting the preset first constant from the density ratio by the adjustment factor corresponding to the center point Q1 as the first product value, recording the sum of the first product value and the first index value as the second index value corresponding to the center point Q1; recording the product of the second index value and the bandwidth to be adjusted corresponding to the center point Q1 as the increased characterization value of the bandwidth to be adjusted corresponding to the center point Q1; The method for obtaining the reduced characterization value of the bandwidth to be adjusted corresponding to the center point Q1 includes: performing negative correlation mapping on the relative density corresponding to the center point Q1 to obtain a value, which is recorded as the mapped relative density; subtracting a preset first constant from the mapped relative density to obtain a value, and then multiplying the value by the adjustment factor corresponding to the center point Q1 to obtain a value, which is recorded as the second product value; and recording the sum of the second product value and the first index value as the third index value; multiplying the third index value by the bandwidth to be adjusted corresponding to the center point Q1 and dividing the value by the preset second constant, which is recorded as the reduced characterization value of the bandwidth to be adjusted corresponding to the center point Q1.

9. A quality detection and analysis method for grinding wheel production process as claimed in claim 7, characterized in that: The method for obtaining abnormal clusters includes: For any target cluster V: record the set constructed by all scattered points in the target cluster V as the set to be analyzed corresponding to the target cluster V, obtain the first distance set corresponding to the set to be analyzed, and the method for obtaining the first distance set corresponding to the set to be analyzed is the same as the method for obtaining the first distance set corresponding to the comprehensive set; obtain the statistical histogram of the first distance set corresponding to the set to be analyzed; according to the value interval of each group on the statistical histogram and the frequency of each group, obtain the points to be connected corresponding to each group, and connect the points to be connected corresponding to each group in turn, The connected curve is recorded as a statistical curve; the normal distribution curve of the first distance set corresponding to the set to be analyzed is obtained; the similarity between the statistical curve and the normal distribution curve is obtained, and it is determined whether the similarity between the statistical curve and the normal distribution curve is greater than a preset similarity threshold. If so, the target cluster cluster V is recorded as a normal cluster cluster, otherwise, the target cluster cluster V is recorded as an abnormal cluster cluster; the horizontal coordinate value of the to-be-connected point corresponding to each group is the middle value of the value range of the corresponding group, and the horizontal and vertical coordinate values ​​of the to-be-connected point corresponding to each group are the frequencies of the corresponding groups.

10. A quality detection and analysis method for grinding wheel production process according to claim 1, characterized in that: The method for acquiring the uneven area on the target surface image comprises: For any abnormal cluster P, the area formed by the pixel points corresponding to each scattered point in the abnormal cluster P is recorded as an uneven area on the target surface image.

Images