A method for online monitoring of grinding of shaft forgings

Abstract

The present invention relates to the field of image processing, and more specifically, the present invention relates to an online monitoring method for grinding of shaft forgings, the method comprising: obtaining a grayscale image of the grinding image of the shaft forgings, performing edge detection on the grayscale image, extracting edge connected domains, and calculating edge features of the edge connected domains; clustering the edge features to obtain a number of clusters, calculating the abnormality of the clusters, and calculating the abnormal performance of each edge connected domain in the cluster according to the abnormality; calculating the suspected defect degree of the edge connected domain of the cluster according to the abnormal performance and the edge features, and judging the grinding degree of the shaft forgings based on the suspected defect degree. The present invention improves the recognition accuracy of surface defects through statistical analysis of edge features and abnormal performance in clusters.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and more specifically, to an online monitoring method for grinding shaft forgings. Background Art

[0002] In modern manufacturing, shaft forgings are used as key components in the automotive, aviation, machinery and other industries, and their quality directly affects the performance and safety of the equipment. Forgings usually go through multiple processes during the production process, including forging, heat treatment, machining and grinding. During the forging and machining process, some defects will appear on the surface of shaft forgings, such as burrs, scratches and uneven surface roughness. These defects will cause stress concentration points on the forgings, increasing the risk of cracking or breaking, so shaft forgings need to be polished.

[0003] The existing Chinese patent application document with publication number CN118096582A discloses an intelligent metal forging quality inspection method, which includes: obtaining the grayscale mutation degree of each pixel point in the transmission gear image according to the grayscale value difference between the neighboring points of each pixel point, and then obtaining a number of pixel points to be filtered; dividing the number of pixel points to be filtered into a number of texture sets; according to the periodicity of the distribution of the texture sets, obtaining the possibility that each texture set is the texture of the gear itself; according to the possibility that each texture set is the texture of the gear itself, obtaining the initial forgetting factor of each pixel point to be filtered in each texture set, and then obtaining the forgetting factor of each pixel point to be filtered in each texture set, filtering the transmission gear image, and obtaining the filtered transmission gear image.

[0004] The above application documents solve the influence of the gear texture on the crack defect recognition by adjusting the filter in the image processing. In addition, the existing machine vision-based edge features in the images of shaft forgings are recognized. Among them, the commonly used Canny edge detection algorithm can effectively recognize the edge features in the images of shaft forgings. These edges are usually important signs of surface defects (burrs, scratches). However, due to the existence of certain metal textures in metal forgings, when using the Canny edge detection algorithm to detect the edges of shaft forgings, the Canny edge detection algorithm cannot distinguish between the texture edges and surface defects of the metal itself, resulting in inaccurate monitoring results of the grinding quality. Summary of the invention

[0005] In order to solve the problem that the Canny edge detection algorithm cannot distinguish the texture edges and surface defects existing in the metal itself, thereby causing inaccurate monitoring results of the grinding quality, the present invention provides solutions in the following aspects.

[0006] A method for online monitoring of shaft forging grinding, comprising: obtaining a grayscale image of a shaft forging grinding image, performing edge detection on the grayscale image, extracting edge connected domains, and calculating edge features of the edge connected domains; clustering the edge features to obtain a plurality of clusters, calculating the abnormality degree of the cluster clusters, and calculating the abnormal performance of each edge connected domain in the cluster cluster according to the abnormality degree; calculating the suspected defect degree of the edge connected domain of the cluster cluster according to the abnormal performance and the edge features, and judging the grinding degree of the shaft forging based on the suspected defect degree; wherein the suspected defect degree satisfies the following relationship: , where Indicates The first The suspected defect degree of the edge connected domain, Indicates The first The abnormal performance of the edge connected domain, Indicates The edge features of the edge connected domain, Represents the standard normalization function.

[0007] By calculating the edge features of edge connected domains and performing cluster analysis, this method can distinguish between metal textures and surface defects. The calculation of clustering and abnormality levels helps to identify edge connected domains that have significant differences in features from other edge connected domains in the cluster. These may be surface defects. By calculating the abnormal performance and suspected defect level of each edge connected domain, the grinding decision can be determined, the actual grinding operation can be guided, and the surface quality of the forgings can be ensured.

[0008] Preferably, the performing edge detection on the grayscale image to extract edge connected domains includes:

[0009] The Canny edge detection algorithm is used to perform edge detection on the grinding image of the shaft forgings to obtain a binary image, wherein in the binary image, the edge is marked as 1 and the non-edge is marked as 0, and all the edges in the binary image are formed into multiple connected domains as edge connected domains.

[0010] Preferably, the edge features include:

[0011] Obtain the position coordinates of all pixel points in the edge connected domain, and use the least squares method to fit straight lines to all pixel points to obtain the fitting straight line of each edge connected domain;

[0012] Take any edge connected domain as the target connected domain, obtain the number of pixels in the target connected domain, calculate the average value of the Euclidean distance between each pixel and the straight line where the connected domain is located, and perform normalization to obtain the edge features of the target edge connected domain.

[0013] By fitting straight lines, the main direction and trend of the pixels in each edge connected domain can be accurately described. Since metal textures are usually arranged consistently along a specific direction, while defects are irregular, these two features can be distinguished by analyzing the distance between the pixels and the fitting straight lines.

[0014] Preferably, the edge feature further includes:

[0015] Take any edge connected domain as the target connected domain, obtain the number of pixels in the target connected domain, take any pixel in the target connected domain as the target pixel, calculate the average of the absolute values ​​of the slope differences between the adjacent pixels before and after the target pixel, and obtain the average slope change. Take the negative value of the average slope change as an exponent, and use an exponential function to calculate the edge features.

[0016] By analyzing the average slope change, the local slope change in the edge connected domain can be captured. At the same time, the average value of the slope change can reduce the impact of noise on edge features to a certain extent, because it considers the overall trend rather than the fluctuation of a single point; even tiny defects can cause significant changes in the local slope, which helps to improve the accuracy of defect detection.

[0017] Preferably, the abnormality degree of the cluster includes:

[0018] The average value of the sum of the absolute values ​​of the differences between the edge features of all clusters and the corresponding cluster centers is calculated to obtain the degree of dispersion of the clusters, and the ratio of the degree of dispersion to the total number of edge features in the clusters is normalized to obtain the degree of abnormality of the clusters;

[0019] The cluster center is the mean of all edge features in the cluster.

[0020] By analyzing the abnormality degree of the clusters, clusters with high abnormality degrees may contain outliers or noise, which helps to identify clusters affected by defects.

[0021] Preferably, the calculating of the abnormal performance of each edge connected domain in the cluster includes:

[0022] Take any cluster as the target cluster, take any edge connected domain in the target cluster as the marked connected domain, calculate the average of the absolute value of the difference between the edge features of the marked connected domain in the target cluster and all other edge connected domains, and perform normalization to obtain the similarity of the edge features between the marked connected domain and all other edge connected domains;

[0023] In response to the similarity being greater than or equal to the hyperparameter, the product of the similarity and the abnormality degree of the target cluster is taken as the abnormal performance;

[0024] In response to the similarity being less than the hyperparameter, the inverse of the product of the similarity and the abnormality degree of the target cluster is taken as the abnormal performance.

[0025] By calculating the difference in edge features between the marked connected domain and other connected domains, we can accurately identify which connected domains show abnormalities in the clusters. Combined with the degree of abnormality of the clusters, we can improve the detection sensitivity of abnormal performance of edge connected domains.

[0026] Preferably, the calculating of the abnormal performance of each edge connected domain in the clustering cluster further includes:

[0027] Take any cluster as the target cluster, take any edge connected domain in the target cluster as the marked connected domain, calculate the average of the absolute values ​​of the differences between the edge features of the marked connected domain in the target cluster and all other edge connected domains, and divide it by the standard deviation of the target cluster to obtain the standardized difference;

[0028] In response to the inverse of the square of the standardized difference plus 1 being greater than or equal to the hyperparameter, the inverse of the square of the standardized difference plus 1 is multiplied by the abnormality degree of the target cluster as the abnormal performance;

[0029] In response to the inverse of the square of the standardized difference plus 1 being less than the hyperparameter, the square of the standardized difference plus 1 multiplied by the inverse of the abnormality degree of the target cluster is taken as the abnormal performance.

[0030] Preferably, the method of judging the grinding degree of the shaft forging based on the suspected defect degree includes:

[0031] In response to the suspected defect level being greater than a preset defect threshold , the edge connected domain is a defective connected domain, otherwise, it is a normal area; in response to the defective connected domain being greater than the preset number of connected domains, the defective connected domain still has defects and needs to be further polished.

[0032] The present invention has the following effects:

[0033] 1. The present invention can effectively distinguish metal texture and surface defects by comprehensive edge feature extraction, cluster analysis and abnormal performance calculation, and combines the statistical analysis of edge features and clustering results, thereby improving the accuracy of surface defect recognition.

[0034] 2. The present invention can more accurately evaluate the grinding quality of shaft forgings by calculating the abnormal performance and suspected defect degree of each edge connected domain, and facilitate individualized analysis of each edge connected domain, thereby identifying areas that need further grinding and ensuring that the forging surface meets the required quality standards. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0036] Figure 1 It is a method flow chart of steps S1 to S3 in an online monitoring method for grinding shaft forgings according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0038] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0039] It should be noted that the implementation scenario of the present invention is: by performing edge detection on the shaft forgings during the grinding process, the grinding quality is judged based on the edge detection results. At the same time, the shaft forgings in the present invention are used in mechanical professional fields such as automobiles, aerospace, etc., so the quality requirements of shaft forgings are high. Small scratches and burrs will become stress concentration points. If small defects cannot be identified, they will break under high load or repeated stress conditions, reducing the strength of the shaft forgings.

[0040] Reference Figure 1 A method for online monitoring of shaft forging grinding includes steps S1 to S3, which are as follows:

[0041] S1: Obtain a grayscale image of the grinding image of a shaft forging, perform edge detection on the grayscale image, extract the edge connected domain, and calculate the edge features of the edge connected domain.

[0042] The Canny edge detection algorithm is used to perform edge detection on the grinding image of shaft forgings to obtain a binary image, in which the edge is marked as 1 and the non-edge is marked as 0. All the edges in the binary image are formed into multiple connected domains as edge connected domains.

[0043] It should be noted that the edge detection algorithm can effectively extract important features in the grayscale image by identifying the area with significant grayscale changes in the image and locating the edge.

[0044] The specific steps for calculating the edge features of the edge connected domain are as follows:

[0045] Obtain the position coordinates of all pixel points in the edge connected domain, and use the least squares method to fit straight lines to all pixel points to obtain the fitting straight line of each edge connected domain;

[0046] Take any edge connected domain as the target connected domain, obtain the number of pixels in the target connected domain, calculate the average value of the Euclidean distance between each pixel and the straight line where the connected domain is located, and perform normalization to obtain the edge features of the target edge connected domain.

[0047] It should be noted that all edge connected domains in the binary image, including defect edges and texture edges existing in the forging metal itself, have different edge characteristics. Specifically, the texture edges existing in the forging metal itself are caused by the fluidity of the material due to the pressure applied to the metal at high temperature during the forging process, which causes plastic deformation. Therefore, the texture edges existing in the metal itself are more regular, while the defects are caused by the processing process, and the regularity of the defect edges is poor. Therefore, the present invention calculates the edge characteristics of all edge connected domains to facilitate the subsequent distinction between the two.

[0048] It should be further noted that, during the metal processing of forgings, a consistent metal texture is often formed on the metal surface, and these textures usually show linear features along the processing direction, reflecting the flow direction of the material. Therefore, the present invention calculates the linear features of the edge connected domain as edge features.

[0049] Specifically, the edge features satisfy the following relationship:

[0050] ;

[0051] In the formula, Indicates The edge features of the edge connected domain, Indicates The total number of pixels in the edge connected domain, Represents the pixel number, Indicates The edge connected domain The Euclidean distance between the pixel point and the fitted straight line in the connected domain, Expressed as a natural constant The exponential function of base .

[0052] It is further explained that the smaller the Euclidean distance between all pixels in the edge connected domain and the fitted straight line of the edge connected domain, the closer all pixels are to the fitted straight line, and the more linear the distribution of the edge connected domain is, that is, the stronger the edge feature is.

[0053] In addition, in another embodiment, any edge connected domain is taken as the target connected domain, the number of pixel points in the target connected domain is obtained, and any pixel point in the target connected domain is taken as the target pixel point. The average value of the sum of the absolute values ​​of the slope differences between the adjacent pixel points before and after the target pixel point is calculated to obtain the average slope change, and the negative value of the average slope change is used as an exponent to calculate the edge features using an exponential function.

[0054] Specifically, the edge features satisfy the following relationship:

[0055] ;

[0056] In the formula, Indicates The edge features of the edge connected domain, Indicates The total number of pixels in the edge connected domain, Represents the pixel number, Respectively represent , , The vertical coordinate of the pixel point, Respectively represent , , The horizontal coordinate of the pixel point, Expressed as a natural constant The exponential function of base .

[0057] It should be noted that It represents the change of the slope of three adjacent pixels. The smaller the value, the smaller the average slope change, the closer the result of the exponential function is to 1, and the stronger the edge feature is. Conversely, the larger the average slope change is, the closer the result of the exponential function is to 0, and the weaker the edge feature is.

[0058] S2: Cluster the edge features to obtain several clusters, calculate the abnormality degree of the clusters, and use the abnormality degree to calculate the abnormal performance of each edge connected domain in the cluster.

[0059] It should be noted that the edge features of all edge connected domains are used as clustering samples, and clustering is performed using an iterative self-organizing clustering algorithm to obtain clustering results. Exemplarily, the clustering method can be a K-means clustering algorithm or an iterative self-organizing clustering algorithm; wherein, in the K-means clustering algorithm, K=10, and in the iterative self-organizing clustering algorithm, the number of iterations=50; it should be noted that in the iterative self-organizing clustering algorithm, there is no need to preset the number of cluster centers, but there is an iteration termination condition, namely: the cluster centers of each cluster cluster no longer change or reach the preset iteration rounds.

[0060] The average value of the sum of the absolute values ​​of the differences between the edge features of all clusters and the corresponding cluster centers is calculated to obtain the degree of dispersion of the clusters. The ratio of the degree of dispersion to the total number of edge features in the clusters is normalized to obtain the degree of abnormality of the clusters.

[0061] The cluster center is the mean of all edge features in the cluster.

[0062] Specifically, the abnormality degree of the clusters satisfies the following relationship:

[0063] ;

[0064] In the formula, Indicates The abnormality of the clusters, Indicates The total number of edge features in the edge-connected domains of the clusters, represents the ordinal number of the edge connected domain, Indicates The first The edge characteristics of the edge connected domain are The absolute value of the difference between the cluster centers of the clusters, Represents the normalization function.

[0065] It should be noted that It indicates the average level between the edge features of all edge connected domains in the cluster and the cluster center of the cluster. The larger its value is, the greater the difference in edge features of the edge connected domains in the cluster, that is, the more abnormal the cluster is.

[0066] It should also be noted that It indicates the number of edge connected domains in the cluster. The smaller its value is, the smaller the number of edge connected domains similar to the cluster is, that is, the more abnormal the cluster is.

[0067] In addition, in another embodiment, the average value of the sum of the absolute values ​​of the differences between the edge features of all clusters and the corresponding cluster centers is calculated and normalized to obtain the abnormality degree of each cluster.

[0068] Specifically, the abnormality degree satisfies the following relationship:

[0069] ;

[0070] In the formula, Indicates The abnormality of the clusters, Indicates The total number of edge features in the edge-connected domains of the clusters, represents the ordinal number of the edge connected domain, Indicates The first The edge characteristics of the edge connected domain are The Euclidean distance between the cluster centers of the clusters, Represents the standard normalization function.

[0071] Take any cluster as the target cluster, take any edge connected domain in the target cluster as the marked connected domain, calculate the average of the absolute value of the difference between the edge features of the marked connected domain in the target cluster and all other edge connected domains, and perform normalization to obtain the similarity of the edge features between the marked connected domain and all other edge connected domains;

[0072] In response to the similarity being greater than or equal to the hyperparameter, the product of the similarity and the abnormality degree of the target cluster is taken as the abnormal performance;

[0073] In response to the similarity being less than the hyperparameter, the inverse of the product of the similarity and the abnormality degree of the target cluster is taken as the abnormal performance.

[0074] Specifically, the abnormal performance satisfies the following relationship:

[0075] ;

[0076] In the formula, Indicates The first The abnormal performance of the edge connected domain, Indicates The first The average value of the absolute value of the difference between the edge features of the edge connected domain and other edge connected domains in the cluster, Indicates The abnormality of the clusters, Expressed as a natural constant is the exponential function of the base, Represents preset hyperparameters.

[0077] It should be noted that if The first The edge connected domain and The smaller the average absolute value of the difference between the edge features of other edge connected domains in the cluster, the more similar the data point is to other data points in the cluster, and the The abnormality of the clusters indicates The higher the confidence level of abnormal performance of the edge connected domain of the clusters;

[0078] It needs to be explained that , Indicates the similarity between the edge features of the marked edge connected domain and other edge connected domains in the target cluster. If its value is small, it means that the similarity between the edge features of the marked edge connected domain and other edge connected domains in the cluster is small, indicating that the abnormal performance of the marked edge connected domain is quite different from the abnormal degree of the cluster. Therefore, it is necessary Operation to make Smaller, When smaller, Larger; at the same time, If the value is large, the edge features of the marked edge connected domain are more similar to those of other edge connected domains in the target cluster, which means that the abnormal performance of the marked edge connected domain is less different from the abnormal degree of the target cluster.

[0079] In addition, in another embodiment, calculating the abnormal performance of each edge connected region in the clustering cluster further includes:

[0080] Take any cluster as the target cluster, take any edge connected domain in the target cluster as the marked connected domain, calculate the average of the absolute values ​​of the differences between the edge features of the marked connected domain in the target cluster and all other edge connected domains, and divide it by the standard deviation of the target cluster to obtain the standardized difference;

[0081] In response to the inverse of the square of the standardized difference plus 1 being greater than or equal to the hyperparameter, the inverse of the square of the standardized difference plus 1 is multiplied by the abnormality degree of the target cluster as the abnormal performance;

[0082] In response to the inverse of the square of the standardized difference plus 1 being less than the hyperparameter, the square of the standardized difference plus 1 multiplied by the inverse of the abnormality degree of the target cluster is taken as the abnormal performance.

[0083] Specifically, the abnormal performance satisfies the following relationship:

[0084] ;

[0085] In the formula, Indicates The first The abnormal performance of the edge connected domain, Indicates The first The average value of the absolute value of the difference between the edge features of the edge connected domain and other edge connected domains in the cluster, Indicates The standard deviation of the edge features of all edge-connected regions in the clusters, Indicates The abnormality of the clusters, represents the preset hyperparameters, .

[0086] That is, calculation Indicates The ratio of the average absolute value of the difference between the edge connected domains to the standard deviation of the cluster clusters is a standardized measure. The standardization is squared to increase the sensitivity of the difference. is smaller, indicating that the edge features of the marked edge connected domain are similar to those of other edge connected domains in the target cluster. Smaller, approaches 1. On the contrary, when is larger, indicating that the edge features of the annotated edge connected domain are significantly different from those of other edge connected domains in the target cluster. Larger, Approaching 0.

[0087] S3: Based on the abnormal performance and edge features, the suspected defect degree of the edge connected domain of the cluster is calculated, and the grinding degree of the shaft forgings is judged based on the suspected defect degree.

[0088] Specifically, the suspected defect degree satisfies the following relationship:

[0089] ;

[0090] In the formula, Indicates The first The suspected defect degree of the edge connected domain, Indicates The first The abnormal performance of the edge connected domain, Indicates The edge features of the edge connected domain, Represents the standard normalization function.

[0091] In response to the suspected defect level being greater than a preset defect threshold , the edge connected domain is a defective connected domain, otherwise, it is a normal area; in response to the defective connected domain being greater than the preset number of connected domains, the defective connected domain still has defects and needs to be further polished.

[0092] Further explanation, preset defect threshold , preset number of connected domains .

[0093] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly defined.

[0094] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for online monitoring of shaft forging grinding, characterized in that: include: Obtain a grayscale image of the grinding image of the shaft forging, perform edge detection on the grayscale image, extract the edge connected domain, and calculate the edge features of the edge connected domain; Clustering edge features to obtain a number of clusters, calculating the abnormality degree of the clusters, and using the abnormality degree to calculate the abnormal performance of each edge connected domain in the cluster; According to the abnormal performance and the edge features, the suspected defect degree of the edge connected domain of the cluster is calculated, and the grinding degree of the shaft forging is judged based on the suspected defect degree; The suspected defect degree satisfies the following relationship: , where Indicates The first The suspected defect degree of the edge connected domain, Indicates The first The abnormal performance of the edge connected domain, Indicates The edge features of the edge connected domain, represents the standard normalization function; Specifically, the abnormal performance satisfies the following relationship: ； In the formula, Indicates The first The abnormal performance of the edge connected domain, Indicates The first The average value of the absolute value of the difference between the edge features of the edge connected domain and other edge connected domains in the cluster, Indicates The abnormality of the clusters, Indicated by natural constant is the exponential function of the base, represents the preset hyperparameters, .

2. The method for online monitoring of shaft forging grinding according to claim 1, characterized in that: The step of performing edge detection on the grayscale image and extracting edge connected domains includes: The Canny edge detection algorithm is used to perform edge detection on the grinding image of the shaft forgings to obtain a binary image, wherein in the binary image, the edge is marked as 1 and the non-edge is marked as 0, and all the edges in the binary image are formed into multiple connected domains as edge connected domains.

3. The method for online monitoring of shaft forging grinding according to claim 1, characterized in that: The edge features include: Obtain the position coordinates of all pixel points in the edge connected domain, and use the least squares method to fit straight lines to all pixel points to obtain the fitting straight line of each edge connected domain; Take any edge connected domain as the target connected domain, obtain the number of pixels in the target connected domain, calculate the average value of the Euclidean distance between each pixel and the straight line where the connected domain is located, and perform normalization to obtain the edge features of the target edge connected domain.

4. The method for online monitoring of shaft forging grinding according to claim 1, characterized in that: The edge features further include: Take any edge connected domain as the target connected domain, obtain the number of pixels in the target connected domain, take any pixel in the target connected domain as the target pixel, calculate the average of the absolute values ​​of the slope differences between the adjacent pixels before and after the target pixel, and obtain the average slope change. Take the negative value of the average slope change as an exponent, and use an exponential function to calculate the edge features.

5. The method for online monitoring of shaft forging grinding according to claim 1, characterized in that: The abnormality degree of the cluster includes: The average value of the sum of the absolute values ​​of the differences between the edge features of all clusters and the corresponding cluster centers is calculated to obtain the degree of dispersion of the clusters, and the ratio of the degree of dispersion to the total number of edge features in the clusters is normalized to obtain the degree of abnormality of the clusters; The cluster center is the mean of all edge features in the cluster.

6. The method for online monitoring of shaft forging grinding according to claim 1, characterized in that: The calculating of the abnormal performance of each edge connected domain in the clustering cluster includes: Take any cluster as the target cluster, take any edge connected domain in the target cluster as the marked connected domain, calculate the average of the absolute value of the difference between the edge features of the marked connected domain in the target cluster and all other edge connected domains, and perform normalization to obtain the similarity of the edge features between the marked connected domain and all other edge connected domains; In response to the similarity being greater than or equal to the hyperparameter, the product of the similarity and the abnormality degree of the target cluster is taken as the abnormal performance; In response to the similarity being less than the hyperparameter, the inverse of the product of the similarity and the abnormality degree of the target cluster is taken as the abnormal performance.

7. The method for online monitoring of shaft forging grinding according to claim 1, characterized in that: The calculating of the abnormal performance of each edge connected domain in the clustering cluster also includes: Take any cluster as the target cluster, take any edge connected domain in the target cluster as the marked connected domain, calculate the average of the absolute values ​​of the differences between the edge features of the marked connected domain in the target cluster and all other edge connected domains, and divide it by the standard deviation of the target cluster to obtain the standardized difference; In response to the inverse of the square of the standardized difference plus 1 being greater than or equal to the hyperparameter, the inverse of the square of the standardized difference plus 1 is multiplied by the abnormality degree of the target cluster as the abnormal performance; In response to the inverse of the square of the standardized difference plus 1 being less than the hyperparameter, the square of the standardized difference plus 1 multiplied by the inverse of the abnormality degree of the target cluster is taken as the abnormal performance.

8. The method for online monitoring of shaft forging grinding according to claim 1, characterized in that: The method of judging the grinding degree of shaft forgings based on the degree of suspected defects includes: In response to the suspected defect level being greater than a preset defect threshold , the edge connected domain is a defective connected domain, otherwise, it is a normal area; in response to the defective connected domain being greater than the preset number of connected domains, the defective connected domain still has defects and needs to be further polished.

Images