Rapid batch online detection method and system for miniature bearings

Abstract

The invention discloses a rapid batch online detection method and system for miniature bearings, and relates to the technical field of bearing relevance, and the method comprises the steps: carrying out the cooperative collection of miniature bearings, and generating a multi-dimensional detection data set; traversing the key detection area to carry out dynamic space registration, and positioning the key detection area of the miniature bearing; traversing the key detection area to carry out defect feature extraction, generating a defect probability distribution diagram, and carrying out dynamic segmentation on the defect probability distribution diagram to generate a defect type label group; and on-line sorting is conducted on the miniature bearings based on the defect type label group, a sorting instruction set is obtained, and a detection quality report is generated and fed back to a production control system. The technical problems that in the prior art, the miniature bearing detection efficiency is low, the sampling inspection precision is insufficient, and the quality of the bearing is difficult to detect and control online in real time are solved, rapid batch online detection of the miniature bearing is achieved, and the technical effect of improving the miniature bearing detection efficiency and precision is achieved.

Description

Technical Field

[0001] The present application relates to the technical field related to bearings, and in particular to a rapid batch online detection method and system for micro bearings. Background Art

[0002] Miniature bearings are widely used in many high-end manufacturing fields such as aerospace, medical equipment, electronic equipment, and precision instruments. The performance and reliability of the equipment place extremely high demands on the accuracy, quality, and stability of miniature bearings. Traditional miniature bearing testing methods mostly rely on manual sampling or offline laboratory testing equipment. However, manual sampling is not only inefficient and difficult to meet the pace of large-scale production, but the test results are also easily affected by the subjective factors of the testers, with a high probability of misjudgment and missed judgment. Although offline laboratory testing can provide more accurate test results, the testing process is time-consuming and cannot achieve real-time monitoring and feedback of products on the production line. It is difficult to adapt to the demand for efficient and rapid quality testing in modern production, thus affecting the detection efficiency and accuracy of miniature bearing production.

[0003] Therefore, in the current relevant technologies, there are technical problems such as low efficiency in micro bearing detection, insufficient sampling accuracy, and difficulty in real-time online detection and control of bearing quality. Summary of the Invention

[0004] This application solves the technical problems of low micro-bearing detection efficiency, insufficient sampling accuracy, and difficulty in real-time online detection and control of bearing quality in the existing technology by providing a rapid batch online detection method and system for micro-bearings. It realizes rapid batch online detection of micro-bearings and achieves the technical effect of improving the efficiency and accuracy of micro-bearing detection.

[0005] The present application provides a rapid batch online detection method for micro bearings, which includes: collaboratively collecting micro bearings to generate a multi-dimensional detection data set; dynamically spatially aligning the multi-dimensional detection data set to locate key detection areas of micro bearings; traversing the key detection areas to extract defect features and generate a defect probability distribution map; dynamically segmenting the defect probability distribution map to generate a defect type label group; online sorting of micro bearings based on the defect type label group, obtaining a sorting instruction set, and generating an inspection quality report to feed back to the production control system.

[0006] The present application also provides a rapid batch online detection system for micro bearings, which includes: a collaborative acquisition module for collaboratively acquiring micro bearings and generating a multi-dimensional detection data set; a dynamic spatial registration module for dynamically spatially registering the multi-dimensional detection data set to locate the key detection area of the micro bearing; a dynamic segmentation module for traversing the key detection area to extract defect features, generate a defect probability distribution map, dynamically segment the defect probability distribution map, and generate a defect type label group; an online sorting module for online sorting of micro bearings based on the defect type label group, obtaining a sorting instruction set, and generating an inspection quality report to feed back to the production control system.

[0007] The proposed rapid batch online inspection method and system for micro-bearings will collaboratively collect micro-bearings and generate a multi-dimensional inspection data set. Key inspection areas will be traversed for dynamic spatial registration to locate the key inspection areas for micro-bearings. Key inspection areas will be traversed for defect feature extraction to generate a defect probability distribution map. The defect probability distribution map will be dynamically segmented to generate a defect type label group. Based on the defect type label group, micro-bearings will be sorted online to obtain a sorting instruction set, and an inspection quality report will be generated and fed back to the production control system. This solves the technical problems of low micro-bearing inspection efficiency, insufficient sampling accuracy, and difficulty in real-time online inspection and control of bearing quality in the prior art. Rapid batch online inspection of micro-bearings will be achieved, achieving the technical effect of improving the efficiency and accuracy of micro-bearing inspection. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the accompanying drawings of the embodiments of the present disclosure are briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in precise order. Instead, various steps may be processed in reverse order or simultaneously as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0009] Figure 1 A schematic flow chart of a rapid batch online detection method for miniature bearings provided in an embodiment of the present application.

[0010] Figure 2 Schematic diagram of the structure of a rapid batch online detection system for micro bearings provided in an embodiment of the present application.

[0011] Description of the accompanying drawings: collaborative acquisition module 10, dynamic spatial registration module 20, dynamic segmentation module 30, online sorting module 40. DETAILED DESCRIPTION

[0012] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0013] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0014] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict, and the terms “first\second” involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. The terms “including” and “having” and any variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used herein are for the purpose of describing the embodiments of this application only.

[0015] The present application provides a rapid batch online detection method for micro bearings, such as Figure 1 As shown, the method includes: Step S100 : collaboratively collect data on the micro bearings to generate a multi-dimensional detection data set.

[0016] Preferably, a high-speed line scan camera and an annular multispectral light source are used to collaboratively capture images of microbearings, specifically capturing images of the surface and internal structure of the microbearings on the assembly line, generating a multi-dimensional inspection dataset. Specifically, the high-speed line scan camera can quickly acquire images. When inspecting microbearings on the assembly line, since they are in a dynamic state, the high-speed line scan camera can quickly scan and image their surfaces, capturing images at a very high frame rate, ensuring that various surface features, such as surface texture and the presence of scratches, wear, cracks, and other defects, are fully and clearly recorded during the movement of the microbearings. The annular multispectral light source can emit light of different wavelengths with different penetration capabilities. When inspecting microbearings, light of different wavelengths can penetrate the surface of the microbearing to reach different depths inside the bearing, and be reflected, refracted, or scattered when encountering internal structural features (such as internal geometry, material inhomogeneities, and internal defects). The annular multispectral light source surrounds the microbearing, illuminating it from all angles. This allows the high-speed line scan camera to capture images containing information about the internal structure of the microbearing, reflecting the internal conditions of the microbearing and facilitating the detection of potential internal defects. The collected surface images and internal structure images provide information about the micro bearing from different dimensions, forming a multi-dimensional detection data set, which enables the quality of the micro bearing to be evaluated from multiple angles.

[0017] Furthermore, step S100 also includes step S110, arranging a focusable annular light source array on both sides of the bearing transmission track and setting an initial spectral combination; step S120, acquiring the bearing speed in real time, and when the micro-bearing enters the imaging area, triggering the high-speed linear array camera to collect the micro-bearing according to the bearing rotation period in combination with the bearing speed, to obtain an axial image data set of the micro-bearing; step S130, collaboratively collecting the internal structure of the micro-bearing based on the initial spectral combination, to obtain an internal structure image data set; step S140, temporally and spatially aligning the axial image data set with the internal structure image data set to obtain the multi-dimensional detection data set.

[0018] Preferably, an annular light source array with adjustable focal length is arranged on both sides of the bearing transmission track to provide uniform lighting. The initial spectral combination is set to alternate irradiation of visible light and near-infrared light. Visible light can clearly display the color, texture and other features of the bearing surface, and near-infrared light is used to detect structural features at a certain depth under the bearing surface; a rotary encoder is then used to monitor the rotation speed of the bearing in real time, wherein the rotary encoder can convert information such as the angle and speed of the rotating object into an electrical signal. Specifically, when the micro-bearing enters a pre-set imaging area, based on the bearing speed obtained in real time, the high-speed linear array camera performs 360° image capture according to the rotation period of the bearing to ensure that image information of the entire circumferential surface of the micro-bearing is captured, thereby obtaining an axial image data set containing surface features of the bearing at various axial angles, which is used to detect defects, wear and the like on the bearing surface.

[0019] Preferably, based on the initial spectral combination of alternating visible light and near-infrared light, X-ray transmission is used to collaboratively collect the internal structure of the micro-bearing, and an internal structure image dataset containing detailed information on the internal structure of the micro-bearing is obtained, which is then used to detect whether there are defects such as cracks, pores, and material unevenness; then the axial image dataset (reflecting the bearing surface information) and the internal structure image dataset (reflecting the bearing internal information) are aligned in time and space, even if the two sets of data correspond to each other in time and space, to ensure data consistency and accuracy; and then integrated to form a multi-dimensional detection dataset, thereby ensuring accurate detection of micro-bearings.

[0020] Step S200 , performing dynamic spatial registration on the multi-dimensional detection data set to locate the key detection area of the micro bearing.

[0021] Preferably, dynamic spatial registration is performed on the multi-dimensional detection data set, that is, the miniature bearing image data collected from different sources and different perspectives are spatially aligned to eliminate the differences under different perspectives and coordinate systems. For example, feature point registration is performed, including extracting feature points from different images with unique positions and feature descriptions (such as the edges of the inner and outer rings, the contours of the balls, etc.), and then calculating the transformation relationship between the images by matching the feature points, such as translation, rotation, scaling, etc., so as to accurately align the detection images. Different parts of miniature bearings have different impacts on performance and quality. For example, the bearing raceway, the contact area between the ball and the raceway, and the sealing ring installation area are prone to defects such as wear and cracks. Then, by analyzing the aligned multi-dimensional detection data set, the key detection areas are determined. For example, a three-dimensional model of the miniature bearing is established, the aligned image data is matched with the three-dimensional model, and the corresponding area is found in the image data according to the set key area; or a recognition model is built based on machine learning (such as target detection), and a large amount of annotated miniature bearing image data is trained to learn the characteristics of the bearing's key detection area, so as to automatically identify and locate the key detection area of the miniature bearing, while ensuring the accuracy and efficiency of batch detection of miniature bearings.

[0022] Furthermore, step S200 also includes step S210, performing point cloud reconstruction based on the multi-dimensional detection data set to generate bearing three-dimensional point cloud parameters; step S220, setting a set of geometric constraint conditions for the micro-bearing, dynamically spatially aligning the bearing three-dimensional point cloud parameters according to the set of geometric constraint conditions, and constructing a spatial transformation matrix; step S230, performing alignment residual analysis according to the spatial transformation matrix to identify multiple local deformation difference areas, wherein the multiple local deformation difference areas include multiple deformation gradient amplitudes, and the multiple deformation gradient amplitudes correspond to the multiple local deformation difference areas; step S240, assigning weights to the multiple local deformation difference areas according to the multiple deformation gradient amplitudes, constructing a regional weight distribution map, performing detection and positioning according to the regional weight distribution map, and determining the key detection area.

[0023] Preferably, a point cloud reconstruction is performed on the multi-dimensional detection data set. Specifically, the feature points of the axial image of the micro-bearing surface are projected into three-dimensional space, and the internal structure image is projected and transformed according to the penetration characteristics of X-rays and the imaging geometry to obtain preliminary point cloud data. The point cloud data of the axial image and the internal structure image are then fused to generate complete three-dimensional point cloud data of the micro-bearing. The basic parameters of the point cloud are calculated, including calculating the average coordinates of all points to obtain the coordinates of the center of mass of the point cloud, and calculating the minimum and maximum values of the boundary to obtain the bounding box parameters of the point cloud; the geometric characteristic parameters of the micro-bearing are calculated based on the point cloud data, such as determining the radius of the inner and outer rings of the bearing and the diameter of the ball by fitting the circular or spherical area in the point cloud data; the ball distribution angle is calculated, and the angular relationship is determined by analyzing the position distribution of the balls in the point cloud; the surface characteristic parameters are calculated using the point cloud data, such as estimating the surface roughness parameters by calculating the height change of the point cloud in a local area, and describing the surface texture characteristics by analyzing the local density change and curvature change of the point cloud.

[0024] Preferably, the collective constraints of the micro bearing are set, including the ball distribution angle tolerance, the inner and outer ring concentricity threshold and the assembly stress distribution characteristics, wherein the ball distribution angle tolerance specifies the allowable deviation range of the ball distribution angle in the bearing, the inner and outer ring concentricity threshold represents the standard of the degree of concentricity that should be maintained between the inner and outer rings, and the assembly stress distribution characteristics describe the reasonable distribution of stresses borne by various parts of the bearing during the assembly process; then, according to the set geometric constraints, the generated three-dimensional point cloud parameters of the bearing are dynamically spatially aligned, that is, the point cloud parameter points are spatially transformed so that the point cloud data meets the geometric constraints, and then a spatial transformation matrix is constructed, which contains transformation information such as translation, rotation, and scaling, and is used to transform the point cloud data to the correct spatial position and posture.

[0025] Preferably, after the three-dimensional point cloud parameters are registered using the spatial transformation matrix, there may be differences between the actual point cloud data and the ideal state that meets the geometric constraints, namely, the registration residual. The registration residual is then analyzed, that is, the difference between the transformed and ideal positions of each point in the point cloud is calculated using the Euclidean distance, and then the registration residual threshold is set according to the bearing manufacturing tolerance, detection accuracy, etc. Points with registration residuals greater than the registration residual threshold are marked as deformed points, and areas with local deformation on the micro-bearing (i.e., local deformation difference areas) are identified. The degree of deformation in each local deformation difference area is not uniform, and includes multiple deformation gradient amplitudes, wherein the deformation gradient amplitude describes the rate of change of deformation in the area, and different local deformation difference areas correspond to different deformation gradient amplitudes.

[0026] Preferably, since the deformation gradient amplitudes of different local deformation difference areas are different, the degree of their influence on the performance and quality of the micro-bearing is also different. Therefore, a weight is assigned to each local deformation difference area according to the size of the deformation gradient amplitude. The larger the amplitude, the higher the weight, which means that the possibility of problems in the area is greater and the impact on the bearing quality is greater. Then, a regional weight distribution map is constructed to intuitively display the risk level of each area on the micro-bearing; finally, detection and positioning are carried out according to the regional weight distribution map, and high-risk detection areas, that is, key detection areas, are determined to facilitate more accurate detection, thereby ensuring that potential quality problems can be discovered in a timely manner.

[0027] Furthermore, step S210 also includes step S211, performing edge enhancement processing on the axial image data set to obtain three-dimensional spatial coordinate parameters, performing interlayer interpolation processing on the internal structure image data set, and extracting internal crack feature point cloud parameters based on the continuum data; step S212, retrieving the assembly gap measurement value of the micro bearing to construct a gap distribution matrix, and constructing gap space point cloud parameters from the gap distribution matrix; step S213, fusing the three-dimensional spatial coordinate parameters, the internal crack feature point cloud parameters, and the gap space point cloud parameters to generate the bearing three-dimensional point cloud parameters.

[0028] Preferably, edge enhancement processing is performed on the axial image data set, including using an image processing algorithm (such as the Sobel operator, the Canny operator, etc.) to detect areas with large grayscale changes in the image, i.e., edges, thereby highlighting the edge information of the micro-bearing in the image, making the outline of the bearing clearer and more obvious, and then combining the camera's imaging model (such as the pinhole camera model) and camera parameters (such as focal length, shooting angle, etc.) to convert the edge points in the two-dimensional image into three-dimensional space to obtain three-dimensional spatial coordinates. By performing spatial transformation on all points in the axial image data set, the three-dimensional spatial coordinate parameters of the micro-bearing surface are obtained, which describes the geometric shape and position information of the bearing surface. Interlayer interpolation processing is performed on the internal structure image data, including using linear interpolation, spline interpolation, etc. to generate more intermediate layer data between discrete layered internal structure images, filling the blank area in the middle, making the data more continuous and smooth, and thus obtaining more complete internal structure data; then image analysis processing (such as threshold segmentation) is used to identify and extract internal crack information, and the relevant points in the crack area are converted into point cloud data, and the characteristic parameters of the point cloud are extracted, such as the position, shape, size, etc. of the point cloud, which describe the characteristics of the internal crack, namely the internal crack characteristic point cloud parameters.

[0029] Preferably, a laser interferometer is used to measure the assembly clearance of the micro bearing. By measuring the assembly clearance at different positions, multiple measurement values are obtained, and the measurement points are arranged in order of their spatial positions to construct a gap distribution matrix to reflect the distribution of the assembly clearance at different positions of the micro bearing. Then, based on the information in the gap distribution matrix, each measurement point and its corresponding gap value are converted into point cloud data in three-dimensional space. Each point has three-dimensional coordinates (indicating the position of the measurement point) and an attribute representing the gap value, thereby constructing gap space point cloud parameters to describe the distribution characteristics of the assembly clearance in space. Finally, the obtained three-dimensional spatial coordinate parameters (describing the bearing surface geometry), internal crack feature point cloud parameters (describing internal crack characteristics), and gap space point cloud parameters (describing the assembly clearance distribution) are integrated to form a complete three-dimensional point cloud parameter set containing information on the micro bearing surface, internal structure, and assembly clearance, i.e., the final bearing three-dimensional point cloud parameters, which are used for efficient and accurate detection and analysis of micro bearings.

[0030] Furthermore, step S220 also includes step S221, constructing a benchmark parameter feature tree of the micro bearing, performing multidimensional analysis through the benchmark parameter feature tree, performing assembly stress analysis based on the multidimensional analysis data, and constructing the geometric constraint condition set; step S222, performing feature analysis on the three-dimensional point cloud parameters of the bearing to obtain a surface feature data set and a structural feature data set; step S223, extracting edge contour feature points of the micro bearing based on the surface feature data set; step S224, rigidly aligning the edge contour feature points with the structural feature data set according to the geometric constraint condition set, and constructing the spatial transformation matrix.

[0031] Preferably, a benchmark parameter feature tree of the micro bearing is constructed, that is, various benchmark parameters of the micro bearing (such as size parameters, structural parameters, etc.) are organized into a tree structure in a tree structure. For example, the basic size of the bearing (inner and outer ring diameters, ball diameters, etc.), the number of balls, the distribution angle and other parameters are used as nodes of the tree. The relationship between the nodes reflects the hierarchical structure of the parameters. Then, through the benchmark parameter feature tree, a comprehensive analysis is performed on multiple parameters of the micro bearing, and the mutual influence of the bearing parameters is analyzed from different dimensions (such as size dimensions, structural dimensions, etc.). For example, the influence of the number of balls and the distribution circumferential angle on the bearing load capacity, and the influence of the inner and outer ring diameter tolerance band on the assembly accuracy are analyzed, and then multi-dimensional analysis data, including the number of balls N, the distribution circumferential angle θ and the inner and outer ring diameter tolerance band ΔD, are obtained. Then, the assembly stress of the micro bearing is analyzed using finite element simulation. By discretizing the bearing model into multiple small units, the stress distribution of each unit during the assembly process is calculated, thereby obtaining the assembly stress state of the entire bearing.

[0032] Preferably, finally, based on the analysis results, a set of geometric constraints is determined, including the allowable range of ball distribution angle deviation. The absolute value of the distribution angle deviation of each ball should be less than or equal to the distribution circumferential angle θ divided by 2N to ensure uniform distribution of the balls on the circumference, thereby ensuring the normal operation and load-bearing capacity of the bearing; the center offset of the inner and outer rings should be less than or equal to one tenth of the inner and outer ring diameter tolerance band ΔD to ensure the concentricity of the inner and outer rings and avoid bearing wear and performance degradation due to insufficient concentricity; the stress concentration coefficient of the contact area is less than or equal to δ, where δ is the allowable stress concentration coefficient threshold determined according to the bearing material and design requirements to prevent material fatigue and damage caused by excessive stress concentration.

[0033] Preferably, feature analysis is performed on the bearing's three-dimensional point cloud parameters. Specifically, by calculating geometric properties such as the point cloud's normal vector and curvature, information related to surface features is extracted from the point cloud parameters to form a surface feature dataset (such as surface roughness, texture, and edge contours). Simultaneously, structural features are extracted by analyzing the point cloud's depth information and spatial distribution, and information related to internal structural features is extracted to form a structural feature dataset (such as internal cracks and material density distribution). Edge detection algorithms (such as the Canny operator and the Sobel operator) and contour extraction are used within the surface feature dataset to identify and extract the edge contour of the miniature bearing. Specifically, regions in the point cloud data where the surface normal vector varies significantly, i.e., edge regions, are detected, and feature points on the edge contour are extracted. Then, according to the set of geometric constraints, the extracted edge contour feature points are rigidly aligned with the structural feature dataset. That is, the shape and size of the object do not change during the alignment process, and only translation and rotation operations are performed. By adjusting the position and posture of the edge contour feature points, they are made to conform to the geometric constraints with the relevant information in the structural feature dataset; at the same time, the translation and rotation parameters required for matching alignment are calculated to form a spatial transformation matrix, which is used to transform the edge contour feature points from their current position and posture to a position and posture that conforms to the geometric constraints, and is also used to perform a unified transformation on the entire point cloud data.

[0034] Step S300 , traversing the key detection area to extract defect features, generating a defect probability distribution map, dynamically segmenting the defect probability distribution map, and generating a defect type label group.

[0035] Preferably, defect features are extracted from the key inspection areas. Specifically, feature information representing defects is extracted from the point cloud data of the key inspection areas. Specifically, for surface defects, the height difference between each point and its neighboring points is calculated, and the surface roughness is obtained through statistical analysis. Surface curvature is estimated by fitting local surfaces or calculating the normal vector of the point cloud. Areas of high curvature may indicate edges, corners, or potential defect locations. Edge detection algorithms (such as the Canny operator and the Sobel operator) are used to identify edge points in the point cloud and determine edge irregularities. For internal defects, point cloud density is extracted. This means that the point cloud density within each small area, such as pores and looseness, is counted. The point cloud data is divided into several voxels (3D pixels), and the number or mass distribution of points within each pixel is calculated. A defect probability model is established based on a support vector machine (SVM) and a naive Bayes classifier. The extracted defect features are used as input to train and predict the probability of each point or area being a defect. The point cloud data of the key inspection areas is then input into the defect probability model to calculate the probability of defects. A defect probability distribution map is then generated to more clearly display the trends and characteristics of the probability distribution.

[0036] Preferably, the defect probability distribution map is dynamically segmented. Specifically, the grayscale value distribution of the defect probability distribution map is analyzed, and a suitable threshold is determined by empirical methods, the maximum between-class variance method (OTSU), etc. For example, if it is found that the defect probability is mainly concentrated in the area with higher grayscale values, a grayscale value of 0.6 is used as the initial threshold, and the areas with grayscale values greater than 0.6 in the probability distribution map are regarded as possible defect areas, and the areas with grayscale values less than this value are regarded as normal areas; then, according to the determined threshold, the image is segmented into foreground (possible defect area) and background (normal area). For the foreground area after preliminary segmentation, its relevant features are extracted, such as the area, perimeter, centroid coordinates, grayscale mean, etc., and an appropriate clustering algorithm, such as the K-means clustering algorithm, is selected to cluster the foreground area according to the features. Assuming that based on the morphology and probability distribution of the defects, three different types of defects are expected, the K value is set to 3, that is, the foreground area is divided into three different clusters, each cluster representing a possible defect type; and defect type labels are assigned to the corresponding areas. For example, the label "D1" is assigned to the area representing surface scratches, "D2" is assigned to the area representing internal cracks, and "D3" is assigned to the area representing wear, etc. Finally, the defect type label group is output to intuitively display the defect distribution of the micro bearing.

[0037] Furthermore, step S300 also includes step S310, performing cross-modal feature fusion on the multi-dimensional detection data set to obtain multi-scale semantic features; step S320, traversing the key detection area based on the multi-scale semantic features to perform defect analysis, dynamically weighting the defect analysis results to generate a defect feature tensor; step S330, spatially upsampling the key detection area based on the defect feature tensor to generate the defect probability distribution map.

[0038] Preferably, the data features in the multi-dimensional detection data set are integrated, for example, the texture features in the optical image are combined with the three-dimensional geometric features obtained by laser scanning, and then the fused features are processed using a convolutional neural network to extract multi-scale semantic features, that is, semantic features of different scales. Specifically, small-scale features can capture detailed information on the bearing surface, such as tiny scratches and wear; large-scale features can reflect the overall structure and morphological information of the bearing, such as the bearing's contour, overall deformation, etc., thereby more comprehensively describing the bearing's status. Then, based on multi-scale semantic features, key detection areas (such as the inner and outer ring surfaces of the bearing, the contact area between the ball and the raceway, etc.) are analyzed one by one. By comparing the features with the defect pattern or feature library, it is determined whether there are defects in each area, as well as the possible defect type and severity. Based on the results of the defect analysis, different weights are assigned to different features. Among them, if the features of a certain area indicate the possibility of serious crack defects, the weight of the features related to the crack will increase; for features that are unlikely to be related to defects, their weights will be reduced. Through dynamic weighting, features that play an important role in defect judgment are highlighted, and the accuracy of defect detection is improved. Finally, a defect feature tensor is generated, which contains various feature information of the key detection areas and the corresponding weights, and can describe the defect characteristics of the bearing more compactly and accurately.

[0039] Preferably, the key detection area is spatially upsampled based on the defect feature tensor, that is, the defect feature tensor is amplified or restored in the spatial dimension so that it can match the spatial resolution of the original key detection area to more accurately locate the position of the defect on the surface or inside the bearing. Then, based on the upsampled defect feature tensor, the probability of the defect occurring at each spatial position is calculated. Finally, the probability value is expressed in the form of an image to generate a defect probability distribution map. Different colors or grayscales in the map represent different probability values, thereby intuitively showing the possible distribution of defects in the key detection area of the bearing.

[0040] Furthermore, step S300 also includes step S340, performing graph feature analysis based on the defect probability distribution graph to obtain histogram features, performing boundary demarcation based on the histogram features combined with the defect feature tensor, and generating a dynamic segmentation threshold; step S350, performing morphological analysis on the defect probability distribution graph, and calculating multiple adjacent defect areas based on the graph morphological analysis results; step S360, dynamically segmenting and identifying the defect probability distribution graph according to the dynamic segmentation threshold combined with the multiple adjacent defect areas, and generating geometric feature parameters of multiple adjacent areas; step S370, matching the geometric feature parameters with the preset defect type to generate the defect type label group.

[0041] Preferably, a graph feature analysis is performed on the defect probability distribution graph, that is, the probability value distribution of the pixels in the graph is analyzed, and a histogram feature is generated by counting the frequency of occurrence of different probability values. For example, the number of pixels with probability values in each interval of 0 to 0.1, 0.1 to 0.2, etc. is counted to obtain a histogram of the probability distribution graph to reflect the distribution of probability values in the entire graph; then, the defect feature tensor is combined with the feature information of different regions, and the boundary between the defect and non-defective regions is delineated according to the histogram features. Specifically, by analyzing the distribution peaks and valleys of the probability values in the histogram, a suitable probability value is determined as the dynamic segmentation threshold. Then, morphological image processing methods such as corrosion, dilation, opening and closing operations are used to perform morphological analysis on the defect probability distribution map, mainly to analyze the morphological characteristics of the defect area in the map, such as shape, size, connectivity, etc. For example, the corrosion operation can remove isolated points on the boundary of the defect area, and the dilation operation can fill small holes in the defect area, thereby optimizing the morphological representation of the defect area; finally, based on the results of the morphological analysis, multiple adjacent defect areas are identified and calculated, and by judging which pixel areas have probability values higher than the dynamic segmentation threshold and are spatially adjacent, they are divided into a defect area, and finally multiple different defect areas are obtained, each of which represents a possible defect location or range.

[0042] Preferably, the defect probability distribution map is converted into a binary image based on a dynamic segmentation threshold, wherein pixels with probability values above the threshold are marked as 1 (indicating a defective area), and pixels below the threshold are marked as 0 (indicating a non-defective area). Connected domains, i.e., interconnected defective areas, are marked in the binary image. For each marked connected domain, its geometric feature parameters, such as area, perimeter, centroid coordinates, major axis length, minor axis length, etc., are calculated to describe the size, shape, and location of each adjacent defective area. Finally, the geometric feature parameters of each adjacent defective area are matched with pre-set feature templates of different defect types. Pre-set surface scratch defects may have specific length and width ratio characteristics, and internal cracks may have specific shape and position characteristics. Through comparison and matching, it is determined which pre-set defect type each defective area belongs to, and a label group containing the defect location (such as centroid coordinates) and type is generated.

[0043] Step S400: sort the miniature bearings online based on the defect type label group, obtain a sorting instruction set, generate an inspection quality report and feed it back to the production control system.

[0044] Preferably, the micro bearings on the production line are sorted online using defect type label groups. Specifically, the sorting device grabs, moves and places the micro bearings according to the defect type label groups, and sorts each bearing to the corresponding area. For example, if a bearing's label shows "qualified", the sorting device will guide it to the qualified product storage area; if it is a bearing with "surface scratch" defects, it will be guided to the repair or defective product processing area; for bearings with serious defects such as "internal cracks", they will be directly led to the scrap area. In the process of online sorting, the control system generates multiple sorting instructions to clarify how the sorting device operates the miniature bearings. For example, the instructions may include "move bearing numbered 1 to the scrap area" and "place qualified bearing numbered 4 on the third layer of the qualified product storage rack". All sorting instructions for different bearings constitute a sorting instruction set, which is used to guide the sorting device to process each bearing in the correct order and manner to ensure the efficiency and accuracy of the sorting work; and generate a miniature bearing inspection quality report, which may include the total number of bearings inspected, the number of qualified products, the number of bearings with different types of defects and their proportions, etc. It may also include an analysis of the defect distribution, such as the frequency of defects in different parts of the bearing, etc. Finally, the generated inspection quality report is sent to the production control system to adjust the production process parameters in a timely manner, such as checking the quality of raw materials, adjusting the accuracy of processing equipment, etc., to improve product quality and reduce the production of defective products.

[0045] Furthermore, step S400 also includes step S410, establishing a defect type-severity mapping table according to the defect type label group, and constructing an initial sorting decision tree according to the defect type-severity mapping table; step S420, introducing the batch inspection speed of miniature bearings to determine the sorting time extreme value, performing inspection timing analysis on the initial sorting decision tree according to the sorting time extreme value, and generating a detection descending queue; step S430, formulating a sorting priority for the initial sorting decision tree according to the detection descending queue, updating the initial sorting decision tree according to the sorting priority, and obtaining a sorting decision tree; step S440, mapping the defect type label group to the sorting decision tree for multi-level classification, and generating the sorting instruction set.

[0046] Preferably, based on the defect type label group of the miniature bearing, grade judgment rules are defined for different defect types such as cracks, scratches, and rust, so as to establish a defect type-severity mapping table. For example, for cracks, they are divided into three levels: slight, moderate, and severe according to factors such as their length and depth; scratches are graded according to their width, length, and density; rust is graded according to the area and degree of rust, and then an initial sorting decision tree is constructed based on the defect type-severity mapping table. Specifically, the defect type is the tree node, the branch corresponds to different severity levels, and the leaf node can be the corresponding sorting decision, such as sending bearings with slight defects to the repair area, those with moderate defects to the defective area, and those with severe defects to the scrap area.

[0047] Preferably, the batch inspection speed of miniature bearings is considered, and the maximum allowable sorting decision time, i.e., the sorting time extreme value, is calculated based on the production line speed. For example, if the production line produces 10 bearings per minute, the inspection and sorting time for each bearing cannot exceed 6 seconds, which is the sorting time extreme value. The initial sorting decision tree is then subjected to a detection timing analysis based on the sorting time extreme value, analyzing the time required for each decision path. Paths with longer inspection times are then prioritized to generate a descending inspection queue. For example, the detection and judgment of crack defects may require more time, while scratch defects are relatively less time-consuming. Therefore, crack-related inspection paths are prioritized to the front of the descending inspection queue. The initial sorting decision tree is then prioritized based on the descending inspection queue. Specifically, the defect types and severity corresponding to inspection paths at the front of the descending inspection queue have higher sorting priorities. The decision tree depth and branching factor are optimized based on the sorting priority. This may include increasing the depth of the decision tree to optimize high-priority branches, thereby making more detailed classification decisions, simplifying and merging low-priority branches to reduce decision time, and thus obtaining a more efficient sorting decision tree. Finally, the defect type label group is mapped to a sorting decision tree for multi-level classification. This involves judging the defect type and severity of each bearing based on the branches and nodes of the decision tree, and ultimately generating a specific sorting instruction set. For example, bearings with moderate scratch defects are mapped to the sorting decision tree. According to the corresponding branches and decision rules, sorting instructions are generated to send them to the defective area. Each bearing corresponds to a sorting instruction, thus forming a sorting instruction set to guide the sorting of miniature bearings.

[0048] Furthermore, step S300 also includes step S350, matching the defect types in the historical sorting cycle according to the sorting instruction set, obtaining the occurrence frequencies of multiple defect types, and constructing a defect distribution heat map according to the occurrence frequencies of the multiple defect types; step S360, performing time domain correlation identification with the production process parameters of the micro bearing according to the defect distribution heat map, and determining the parameter abnormality correlation pattern; step S370, performing defect prediction based on the parameter abnormality correlation pattern, drawing a defect prediction trend graph, compensating the production process parameters according to the defect prediction trend graph, and generating process optimization suggestions; step S380, correlating and integrating the parameter abnormality correlation pattern, the defect prediction trend graph, and the process optimization suggestions to generate the inspection quality report.

[0049] Preferably, a sorting instruction set is used to match the defect type corresponding to each instruction in the historical sorting cycle, and then the number of times different defect types appear is counted, and the frequency of occurrence of multiple defect types is calculated, that is, the frequency of occurrence of multiple defect types is obtained, and then a defect distribution heat map is constructed in a graphical manner. Specifically, different colors in the heat map represent different defect occurrence frequencies. The darker the color, the higher the frequency of occurrence of the defect type. The red area may indicate that a certain defect has a high frequency of occurrence, while the blue area indicates a low frequency of occurrence, thereby intuitively displaying the distribution of various defects in historical sorting. The defect distribution heat map is associated with the production process parameters of the miniature bearing (such as temperature, pressure, speed, processing time, etc.) in the time dimension, and the relationship between the changes in the production process parameters and the frequency of occurrence of the defect type in different time periods is analyzed, thereby determining the abnormal correlation pattern between the production process parameters and the defect type. For example, when the pressure exceeds a certain set value, the frequency of occurrence of internal crack defects increases significantly, thereby determining the abnormal parameter correlation pattern, which helps to identify which changes in production process parameters may cause defects.

[0050] Preferably, based on the determined parameter anomaly association pattern, future defect conditions are predicted, that is, based on the current and historical production process parameters, the frequency and trend of different defect types that may occur in the future are predicted, and then a defect prediction trend graph is drawn to intuitively display the prediction results. Then, based on the defect prediction trend graph, compensation adjustments are made to the production process parameters that may cause an increase in defects. For example, if it is predicted that a temperature increase will lead to an increase in a certain defect, the temperature setting value is appropriately lowered; at the same time, process optimization suggestions are generated based on the compensation adjustment, including specific parameter adjustment values, adjustment time, etc., to guide the optimization of the production process. Finally, the parameter anomaly association pattern, defect prediction trend graph, and process optimization suggestions are correlated and integrated to generate an inspection quality report, including the historical defect distribution, the correlation between production process parameters and defects, future defect predictions, and corresponding process optimization measures, thereby improving the production efficiency and product quality of miniature bearings.

[0051] In the above, refer to Figure 1 The rapid batch online detection method for micro bearings according to the embodiment of the present invention is described in detail. Figure 2 A rapid batch online detection system for micro bearings according to an embodiment of the present invention is described.

[0052] The rapid batch online detection system for micro bearings according to the embodiment of the present invention is used to solve the technical problems of low micro bearing detection efficiency, insufficient sampling accuracy, and difficulty in real-time online detection and control of bearing quality in the prior art, and realizes rapid batch online detection of micro bearings, achieving the technical effect of improving the efficiency and accuracy of micro bearing detection. Figure 2As shown, the rapid batch online detection system for micro bearings includes: a collaborative acquisition module 10, a dynamic spatial registration module 20, a dynamic segmentation module 30, and an online sorting module 40.

[0053] The collaborative acquisition module 10 is used to collaboratively acquire micro bearings and generate a multi-dimensional detection data set; the dynamic spatial alignment module 20 is used to dynamically spatially align the multi-dimensional detection data set to locate the key detection area of the micro bearing; the dynamic segmentation module 30 is used to traverse the key detection area to extract defect features, generate a defect probability distribution map, dynamically segment the defect probability distribution map, and generate a defect type label group; the online sorting module 40 is used to perform online sorting of micro bearings based on the defect type label group, obtain a sorting instruction set, and generate an inspection quality report to feed back to the production control system.

[0054] The specific configuration of the collaborative acquisition module 10 will be described in detail below. The collaborative acquisition module 10 further includes: arranging an array of focusable annular light sources on both sides of the bearing transmission track and setting an initial spectral combination; acquiring the bearing rotational speed in real time. When the micro-bearing enters the imaging area, triggering a high-speed linear array camera to capture the micro-bearing according to the bearing rotation period in combination with the bearing rotational speed, thereby obtaining an axial image dataset of the micro-bearing; collaboratively acquiring the internal structure of the micro-bearing based on the initial spectral combination, thereby obtaining an internal structure image dataset; and spatially and temporally aligning the axial image dataset with the internal structure image dataset to obtain the multi-dimensional detection dataset.

[0055] The specific configuration of the dynamic spatial registration module 20 will be described in detail below. The dynamic spatial registration module 20 further includes: performing point cloud reconstruction based on the multi-dimensional detection data set to generate bearing three-dimensional point cloud parameters; setting a set of geometric constraints for the micro-bearing, performing dynamic spatial registration on the bearing three-dimensional point cloud parameters according to the set of geometric constraints, and constructing a spatial transformation matrix; performing registration residual analysis based on the spatial transformation matrix to identify multiple local deformation difference areas, wherein the multiple local deformation difference areas include multiple deformation gradient amplitudes, and the multiple deformation gradient amplitudes correspond to the multiple local deformation difference areas; assigning weights to the multiple local deformation difference areas based on the multiple deformation gradient amplitudes, constructing a regional weight distribution map, performing detection and positioning according to the regional weight distribution map, and determining the key detection area.

[0056] The specific configuration of the dynamic spatial registration module 20 will be described in detail below. The dynamic spatial registration module 20 further includes: performing edge enhancement processing on the axial image dataset to obtain three-dimensional spatial coordinate parameters; performing interlayer interpolation processing on the internal structure image dataset to extract internal crack feature point cloud parameters based on the continuum data; retrieving the assembly gap measurement values of the micro-bearing to construct a gap distribution matrix, and constructing gap space point cloud parameters from the gap distribution matrix; and fusing the three-dimensional spatial coordinate parameters, the internal crack feature point cloud parameters, and the gap space point cloud parameters to generate the bearing three-dimensional point cloud parameters.

[0057] The specific configuration of the dynamic spatial registration module 20 will be described in detail below. The dynamic spatial registration module 20 further includes: constructing a baseline parameter feature tree for the micro-bearing, performing multidimensional analysis using the baseline parameter feature tree, performing assembly stress analysis based on the multidimensional analysis data, and constructing the geometric constraint condition set; performing feature parsing on the bearing's three-dimensional point cloud parameters to obtain a surface feature dataset and a structural feature dataset; extracting edge profile feature points of the micro-bearing based on the surface feature dataset; and performing rigid registration of the edge profile feature points with the structural feature dataset according to the geometric constraint condition set to construct the spatial transformation matrix.

[0058] The specific configuration of the dynamic segmentation module 30 will be described in detail below. The dynamic segmentation module 30 further includes: performing cross-modal feature fusion on the multi-dimensional detection dataset to obtain multi-scale semantic features; traversing key detection areas to perform defect analysis based on the multi-scale semantic features, dynamically weighting the defect analysis results to generate a defect feature tensor; and spatially upsampling the key detection areas based on the defect feature tensor to generate the defect probability distribution map.

[0059] The specific configuration of the dynamic segmentation module 30 will be described in detail below. The dynamic segmentation module 30 further includes: performing a graph feature analysis based on the defect probability distribution graph to obtain histogram features, demarcating boundaries based on the histogram features combined with the defect feature tensor, and generating a dynamic segmentation threshold; performing a morphological analysis on the defect probability distribution graph, and calculating multiple adjacent defect regions based on the graph morphological analysis results; dynamically segmenting and identifying the defect probability distribution graph according to the dynamic segmentation threshold combined with the multiple adjacent defect regions, generating geometric feature parameters for multiple adjacent regions; and matching the geometric feature parameters with preset defect types to generate the defect type label group.

[0060] The specific configuration of the online sorting module is described in detail below. The online sorting module further includes: establishing a defect type-severity mapping table based on the defect type label group, constructing an initial sorting decision tree based on the defect type-severity mapping table; determining a sorting time extreme value based on the batch inspection speed of miniature bearings; performing a detection timing analysis on the initial sorting decision tree based on the sorting time extreme value to generate a detection descending queue; assigning a sorting priority to the initial sorting decision tree based on the detection descending queue; updating the initial sorting decision tree based on the sorting priority to obtain a sorting decision tree; and mapping the defect type label group to the sorting decision tree for multi-level classification to generate the sorting instruction set.

[0061] The specific configuration of the online sorting module will be described in detail below. The online sorting module further includes: matching defect types within the historical sorting cycle according to the sorting instruction set, obtaining the occurrence frequencies of multiple defect types, and constructing a defect distribution heat map according to the occurrence frequencies of the multiple defect types; performing time domain correlation identification with the production process parameters of the micro-bearing according to the defect distribution heat map to determine the parameter abnormality correlation pattern; performing defect prediction based on the parameter abnormality correlation pattern, drawing a defect prediction trend map, compensating the production process parameters according to the defect prediction trend map, and generating process optimization suggestions; and correlating and integrating the parameter abnormality correlation pattern, the defect prediction trend map, and the process optimization suggestions to generate the inspection quality report.

[0062] The rapid batch online detection system for micro bearings provided by the embodiment of the present invention can execute the rapid batch online detection method for micro bearings provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0063] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules may be used and run on the user terminal and / or server, and the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other and are not used to limit the scope of protection of the present invention.

[0064] The above specific embodiments do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application shall be included within the scope of protection of this application.

Claims

1. A rapid batch online detection method for miniature bearings, characterized in that: The method comprises: Collaboratively collect micro bearings to generate multi-dimensional detection data sets; Performing dynamic spatial registration on the multi-dimensional detection data set to locate key detection areas of the micro-bearing; Traversing the key detection area to extract defect features, generate a defect probability distribution map, dynamically segment the defect probability distribution map, and generate a defect type label group; Based on the defect type label group, the miniature bearings are sorted online, a sorting instruction set is obtained, and an inspection quality report is generated and fed back to the production control system.

2. The rapid batch online detection method for miniature bearings according to claim 1, characterized in that: Collaboratively collect micro bearings to generate a multi-dimensional detection dataset. The method includes: Arrange an array of focusable annular light sources on both sides of the bearing transmission track and set the initial spectrum combination; Acquire the bearing speed in real time. When the micro-bearing enters the imaging area, trigger the high-speed linear array camera to capture the micro-bearing according to the bearing rotation period in combination with the bearing speed, and obtain an axial image dataset of the micro-bearing. collaboratively collecting the internal structure of the micro-bearing based on the initial spectral combination to obtain an internal structure image dataset; The axial image dataset and the internal structure image dataset are temporally and spatially aligned to obtain the multi-dimensional detection dataset.

3. The rapid batch online detection method for micro bearings according to claim 2, characterized in that: Dynamically registering the multi-dimensional detection data set to locate a key detection area of the micro bearing includes: Performing point cloud reconstruction based on the multi-dimensional detection data set to generate bearing three-dimensional point cloud parameters; Setting a set of geometric constraints for the micro bearing, performing dynamic spatial registration on the three-dimensional point cloud parameters of the bearing according to the set of geometric constraints, and constructing a spatial transformation matrix; Performing registration residual analysis according to the spatial transformation matrix to identify a plurality of local deformation difference regions, wherein the plurality of local deformation difference regions include a plurality of deformation gradient amplitudes, and the plurality of deformation gradient amplitudes correspond to the plurality of local deformation difference regions; The multiple local deformation difference regions are weighted according to the multiple deformation gradient amplitudes, a regional weight distribution map is constructed, and detection and positioning are performed according to the regional weight distribution map to determine the key detection region.

4. The rapid batch online detection method for micro bearings according to claim 3, characterized in that: Point cloud reconstruction is performed based on the multi-dimensional detection data set to generate bearing three-dimensional point cloud parameters, the method comprising: Performing edge enhancement processing on the axial image dataset to obtain three-dimensional space coordinate parameters, performing interlayer interpolation processing on the internal structure image dataset, and extracting internal crack feature point cloud parameters based on the continuum data; Retrieving the assembly clearance measurement value of the miniature bearing to construct a clearance distribution matrix, and constructing the clearance space point cloud parameters from the clearance distribution matrix; The three-dimensional space coordinate parameters, the internal crack feature point cloud parameters, and the gap space point cloud parameters are fused to generate the bearing three-dimensional point cloud parameters.

5. The rapid batch online detection method for micro bearings according to claim 3, characterized in that: A set of geometric constraints for the micro bearing is set, and dynamic spatial registration of the three-dimensional point cloud parameters of the bearing is performed according to the set of geometric constraints to construct a spatial transformation matrix. The method includes: Constructing a reference parameter feature tree of the miniature bearing, performing multidimensional analysis through the reference parameter feature tree, performing assembly stress analysis based on the multidimensional analysis data, and constructing the geometric constraint condition set; Performing feature analysis on the three-dimensional point cloud parameters of the bearing to obtain a surface feature data set and a structural feature data set; Extracting edge contour feature points of the micro bearing based on the surface feature data set; The edge contour feature points are rigidly registered with the structural feature data set according to the geometric constraint condition set to construct the spatial transformation matrix.

6. The rapid batch online detection method for micro bearings according to claim 1, characterized in that: Traversing the key detection area to extract defect features and generate a defect probability distribution map, the method includes: Performing cross-modal feature fusion on the multi-dimensional detection dataset to obtain multi-scale semantic features; Based on the multi-scale semantic features, the key detection areas are traversed to perform defect analysis, and dynamic weighting is performed according to the defect analysis results to generate a defect feature tensor; The key detection area is spatially upsampled based on the defect feature tensor to generate the defect probability distribution map.

7. The rapid batch online detection method for micro bearings according to claim 6, characterized in that: Dynamically segmenting the defect probability distribution map to generate a defect type label group includes: Performing graph feature analysis based on the defect probability distribution graph to obtain histogram features, performing boundary delineation based on the histogram features combined with the defect feature tensor to generate a dynamic segmentation threshold; Performing morphological analysis on the defect probability distribution map, and calculating a plurality of adjacent defect regions according to the morphological analysis results; Dynamically segmenting and marking the defect probability distribution map according to the dynamic segmentation threshold and the multiple adjacent defect areas to generate geometric feature parameters of the multiple adjacent areas; The defect type label group is generated by matching the geometric feature parameters with the preset defect type.

8. The rapid batch online detection method for micro bearings according to claim 1, characterized in that: Online sorting of miniature bearings is performed based on the defect type label group to obtain a sorting instruction set, the method comprising: Establishing a defect type-severity mapping table according to the defect type label group, and constructing an initial sorting decision tree according to the defect type-severity mapping table; Introducing the batch detection speed of miniature bearings to determine the extreme value of sorting time, performing detection timing analysis on the initial sorting decision tree according to the extreme value of sorting time, and generating a detection descending queue; Formulate a sorting priority for the initial sorting decision tree according to the detected descending queue, and update the initial sorting decision tree according to the sorting priority to obtain a sorting decision tree; The defect type label group is mapped to the sorting decision tree for multi-level classification to generate the sorting instruction set.

9. The rapid batch online detection method for miniature bearings according to claim 1, characterized in that: The process of generating the detection quality report includes: Matching defect types in historical sorting cycles according to the sorting instruction set to obtain occurrence frequencies of multiple defect types, and constructing a defect distribution heat map according to the occurrence frequencies of the multiple defect types; Perform time domain correlation identification based on the defect distribution heat map and the production process parameters of the miniature bearing to determine the parameter abnormality correlation pattern; Defect prediction is performed based on the parameter abnormality correlation pattern, a defect prediction trend graph is drawn, production process parameters are compensated according to the defect prediction trend graph, and process optimization suggestions are generated; The parameter anomaly correlation pattern, the defect prediction trend graph, and the process optimization suggestion are correlated and integrated to generate the inspection quality report.

10. Rapid batch online detection system for micro bearings, characterized by: The system is used to implement the rapid batch online detection method for miniature bearings according to any one of claims 1 to 9, and the system comprises: Collaborative acquisition module, used to collaboratively acquire micro bearings and generate multi-dimensional detection data sets; A dynamic spatial registration module, configured to perform dynamic spatial registration on the multi-dimensional detection data set to locate key detection areas of the micro-bearing; A dynamic segmentation module is used to traverse the key detection area to extract defect features, generate a defect probability distribution map, dynamically segment the defect probability distribution map, and generate a defect type label group; The online sorting module is used to sort the miniature bearings online based on the defect type label group, obtain the sorting instruction set, generate the inspection quality report and feed it back to the production control system.

Images