Bearing cage pocket inspection method and device

Abstract

The present invention belongs to the technical field of bearing cage inspection, and discloses a method and device for inspecting pockets of bearing cage in order to solve the technical problem that the pockets of conventional bearing cages are mostly inspected manually, which is prone to high labor costs and low inspection efficiency, and affects the straight-through rate of bearing cages. The method includes: comparing and matching each pocket image in an initial pocket image set with each other, screening some pocket images with abnormal matching in the comparison matching result to obtain some abnormal pocket images, recognizing and marking abnormal area images in the abnormal pocket images to obtain abnormal mark areas, classifying pocket abnormalities for the abnormal mark areas to obtain the types of pocket abnormalities, taking pocket images for the abnormal pockets corresponding to the types of pocket abnormalities a second time to obtain images of real pocket abnormalities, and performing beam positioning in a two-dimensional coordinate system to obtain position information of the abnormal pockets.

Description

[Technical field]

[0001] The present application relates to the related field of vision inspection of bearing cages, and more particularly to a method and apparatus for inspecting pockets in bearing cages. [Background technology]

[0002] A retainer (i.e., bearing retainer, also called a bearing cage) is a bearing part that partially surrounds all or some of the rolling elements and moves with them, is used to isolate the rolling elements, and generally guides the rolling elements and holds them within the bearing.

[0003] Currently, most of the quality inspection methods for bearing cages are manual inspections, and workers must carefully inspect each bearing cage step by step, especially for some bearing cages that require special pocket processing, which requires manual inspection of the pocket size, pocket size, column inclination, and whether holes are drilled, which not only consumes a lot of manpower but also reduces the efficiency of quality inspection. Because it is manual inspection that relies on the experience and seriousness of workers, it is likely to introduce a certain amount of human error, which will affect the quality of bearing cage products and make it difficult to ensure the straight rate of bearing cages. Summary of the Invention [Problem to be solved by the invention]

[0004] The embodiments of the present application provide a method and device for inspecting pockets of a bearing cage in order to solve the technical problems that the conventional bearing cage pockets are mostly inspected manually, which results in high labor costs and a decrease in inspection efficiency, and that it is difficult to perform a comprehensive and thorough inspection of the pockets, which affects the straight-through rate of the bearing cage. [Means for solving the problem]

[0005] The embodiments of the present application use the following technical means. In one aspect of the present application, an embodiment of the present application includes taking continuous pocket images of a bearing cage on a rotary table with a camera to obtain an initial pocket image set; comparing and matching each pocket image in the initial pocket image set with each other; screening some pocket images in the comparison matching result that have abnormal matching based on the comparison matching result to obtain some abnormal pocket images; recognizing and marking abnormal area images in the abnormal pocket images to obtain abnormal mark areas in the abnormal pocket images; and classifying the abnormal mark areas into pocket abnormalities to obtain a type of pocket abnormality of the bearing cage, the type of pocket abnormality being at least one of pocket abnormalities. and determining a pocket inspection result for the bearing cage from the abnormal pocket position information based on the type of pocket abnormality and the image of the true pocket abnormality, and transmitting the pocket inspection result to a terminal of an operator.

[0006] In the embodiment of the present application, a rotary table and a camera are used to inspect pockets on a bearing cage, and all the recognized pockets are compared with each other, and then abnormal pockets are screened out to determine the type of pocket abnormality. Then, the position and abnormal type of the abnormal pocket can be accurately recognized, and the image of the actual pocket abnormality can be sent to back-end operators, which can help the operators to greatly reduce the amount of pocket inspection work, and the accuracy rate is high. Combined with laser positioning, it can help the operators to quickly find pockets with abnormal conditions and determine the corresponding abnormal fault type, which greatly reduces labor costs and improves inspection efficiency. Furthermore, the pockets can be comprehensively and thoroughly inspected, reducing human errors and improving the straight-through rate of the bearing cage.

[0007] In one possible embodiment, taking successive pocket images of the bearing cage on the rotary table with a camera to obtain an initial pocket image set specifically includes obtaining drawing information of the bearing cage to be manufactured, the drawing information including at least the number of pockets, the size of the pockets, the radius of the bearing cage, and the processing parameters of the columns; inputting the number of pockets and the radius of the bearing cage in the drawing information into a control unit of a rotary table control system, the rotary table control system including the control unit, a servo motor unit, and a laser emission unit; determining an angular velocity of the rotary table from the interval of the camera based on the number of pockets to take images of each pocket of the bearing cage; controlling the operation of the rotary table based on the angular velocity of the rotary table; and taking successive pocket images of the bearing cage with the camera. the step of detecting an area of ​​a pillar pixel for a grayscale image set of the partial pocket image set based on a preset Canny operator to obtain an actual pillar pixel area, and comparing an area value of the actual pillar pixel area based on a reference pixel area of ​​the pillar, and determining an actual pillar pixel area that is successfully compared as a complete pillar pixel value; determining a pocket image corresponding to the complete pillar pixel value as a first pocket image, and feeding back the first pocket image and the corresponding camera's photographing time interval back to a control unit of the rotary table control system, thereby completely photographing each pocket of the bearing cage to obtain the initial pocket image set.

[0008] In the embodiment of the present application, the servo motor of the rotary table and the shooting time interval of the camera are controlled so that they can be matched with each other, and the image edge detection of the Canny operator ensures that each pocket of each different type of bearing cage can be completely captured, that is, the rotation of the servo motor and the shooting of the camera are ensured to be compatible with each other, and an initial pocket image set of all pockets of the bearing cage is obtained.

[0009] In one possible embodiment, comparing and matching each pocket image in the initial pocket image set with each other specifically includes: performing grayscale pre-processing on each pocket image to obtain several grayscale pocket images; dividing the grayscale pocket image into grayscale pixel regions by a preset dividing line based on the actual pixel size of the grayscale pocket image, where each of the grayscale pocket images is divided into nine grayscale pixel regions; scanning pixel features in each of the grayscale pixel regions, where the pixel features include a light-dark gradient feature of each pixel point and a corresponding pixel coordinate position feature; and performing similarity comparison matching between each grayscale pocket image for several pixel features that correspond one-to-one with each grayscale pixel region in the several grayscale pocket images based on the pixel features in each of the grayscale pixel regions to obtain the comparative matching result, where the comparative matching result includes normal matching and abnormal matching, and the similarity comparison matching is matching the similarity of corresponding pixel features between any two groups of grayscale pocket images.

[0010] In the embodiment of the present application, after comparing and matching any two grayscale pocket images in the initial pocket image set with each other, the pocket images with abnormal conditions can be quickly found out, so as to obtain the comparison group, thereby reducing the difficulty of image processing, and shortening the inspection time of the entire bearing cage.

[0011] In one possible embodiment, the step of screening some pocket images with abnormal matching in the comparison matching result based on the comparison matching result to obtain some abnormal pocket images includes the steps of: if the comparison matching result is abnormal matching, obtaining some groups of grayscale pocket images with abnormal matching; extracting distinguishing pixel features and corresponding distinguishing grayscale pixel regions in the several groups of grayscale pocket images; obtaining normal grayscale pixel regions based on some normal pocket images with normal matching in the comparison matching result; and calculating related intersections for the distinguishing grayscale pixel regions and the normal grayscale pixel regions based on a convolutional neural network and a sigmoid function. The present invention includes: learning and training pixel feature similarities to obtain a cross-classification network model, the cross-classification network model using a Siamese network as a basic structure; assigning weights according to differences between the distinguishing grayscale pixel regions and the normal grayscale pixel regions by the cross-classification network model, and performing cross-classification on the several groups of grayscale pocket images to remove normal pixel features from the distinguishing pixel features and obtain abnormal pixel features from the distinguishing pixel features, the abnormal pixel features being pixel features corresponding to abnormal pocket images in the grayscale pocket images of each group; and screening several grayscale pixel regions corresponding to the abnormal pixel features to obtain the several corresponding abnormal pocket images.

[0012] The embodiment of the present application cross-classifies several groups of grayscale pocket images by a cross-classification network model, thereby screening pocket images with truly abnormal pixel features in the comparison matching group, i.e., abnormal pocket images.

[0013] In one possible embodiment, the method further includes, before recognizing and marking anomalous area images in the anomalous pocket images to obtain anomalous marked areas in the anomalous pocket images, performing an image sequence encoding process on all pocket images in the initial pocket image set to obtain an encoded data set, and marking anomalies against encoding numbers corresponding to the anomalous pocket images in the encoded data set based on positions of some of the anomalous pocket images corresponding to the initial pocket image set to obtain encoding numbers of the anomalies.

[0014] The embodiment of the present application first encodes all pocket images in the initial pocket image set, and then individually marks the recognized abnormal pocket images to obtain an abnormality encoding number, which is helpful to accurately locate each pocket with abnormality, and also helps workers to find the corresponding abnormal pockets later based on the marked abnormal pocket images.

[0015] In one possible embodiment, recognizing and marking an abnormal area image in the abnormal pocket image to obtain an abnormal mark area in the abnormal pocket image specifically includes: performing image area segmentation on the abnormal pocket image corresponding to the coding number of the abnormality based on the grayscale pixel area corresponding to the abnormal pixel feature to obtain an original abnormal area image; performing a basic edge image feature constraint on the original abnormal area image after the grayscale processing by second differentiation to obtain a candidate abnormal area; calculating a minimum pixel distance for the center of gravity of the candidate abnormal area and the edge image feature area to obtain a distance constraint, where the edge image feature area is the edge pixel area of ​​the candidate abnormal area; constructing an edge slope plane coordinate system from the candidate abnormal area based on the image corner and the center of gravity in the candidate abnormal area; locating a slope angle direction with respect to the image slope plane pixels in the candidate abnormal area based on the edge slope plane coordinate system to obtain a direction constraint; performing a second edge image feature constraint on the candidate abnormal area based on the distance constraint and the direction constraint to obtain the abnormal mark area.

[0016] In the embodiment of the present application, based on the original abnormal area images in some abnormal pocket images, the candidate abnormal areas are located by edge constraint processing on the grayscale image by second-order differentiation, and then the image edge feature constraint is performed for the grayscale pixels in the candidate abnormal areas for a second time based on the distance constraint and the direction constraint to locate and mark the abnormal grayscale pixels with more accurate constraint, thereby obtaining the abnormal mark area corresponding to the abnormal grayscale pixels.

[0017] In one possible embodiment, performing pocket abnormality classification for the abnormal mark area to obtain the type of pocket abnormality of the bearing cage specifically includes performing pocket abnormality classification matching for the abnormal mark area with a pocket abnormality template in a past pocket abnormality database, and if matching is successful, determining the type of pocket abnormality corresponding to the pocket abnormality template as the type of pocket abnormality of the abnormal mark area, and if matching is not successful, determining the type of pocket abnormality for the abnormal mark area; and performing line detection of pixel point sets for horizontal column pixels and vertical column pixels of the pocket in the abnormal mark area using a line detection algorithm based on Hough transformation to obtain pocket size information. calculating the slope of a straight line for a set of pixels with the same brightness gradient in the abnormal mark area through a Hough transform to obtain pillar slope information; recognizing non-intersecting curves in the abnormal mark area through a circle detection algorithm using a Hough transform to obtain a circular outline area; calculating areas for the same pixels in the circular outline area based on adjacent connected areas in the abnormal mark area to obtain hole area information; calculating pixel lengths for uneven areas in the abnormal mark area to obtain groove depth information; and comparing drawing information with information on the pocket size information, the pillar slope information, the hole area information, and the groove depth information to determine a pocket anomaly type in the abnormal mark area.

[0018] In the embodiment of the present application, some abnormal mark areas are recognized and first compared with a past pocket anomaly database, thereby realizing rapid matching recognition of some common anomaly types. Then, various types of judgments are made for the mark areas where no abnormality is recognized, and the Hough detection algorithm is used to recognize and judge the pixel curve, pixel area, consecutive same pixel length, adjacent connected area of ​​pixels, etc. in the abnormal mark areas to obtain the corresponding pocket size information, pillar inclination information, hole area information and groove depth information. Then, these information are judged and compared with the corresponding reference range and range information of each size in the uploaded drawing information, thereby screening and recognizing the corresponding pocket anomaly types in the abnormal mark areas.

[0019] In one possible embodiment, taking a pocket image a second time with the rotary table and the camera for an abnormal pocket corresponding to the type of the pocket abnormality to obtain an image of a true pocket abnormality specifically includes: obtaining an abnormality coding number of the abnormal pocket corresponding to the type of the pocket abnormality; performing rotational feedback control on the servo motor of the rotary table based on the sequence position of the anomaly coding number in the coding data set to face the abnormal pocket corresponding to the anomaly coding number to the camera, wherein the rotational feedback control controls the rotational operation of the servo motor based on the shooting time interval between the anomaly coding number and the target coding number; and taking pocket images a second time with the camera for several abnormal pockets corresponding to the anomaly coding numbers to obtain images of several of the true pocket abnormalities.

[0020] In the embodiment of the present application, the rotation of the motor in the rotary table is controlled based on the abnormality code number of the abnormal pocket, and the abnormal pocket is photographed a second time to obtain an image of the true pocket abnormality, thereby realizing a second inspection of the abnormal pocket, which may also provide the operator with a criterion for judging whether the pocket is abnormal, and combined with the type of abnormal pocket detected by the system, the operator can make a better judgment.

[0021] In one possible embodiment, using a laser beam to perform beam positioning in a two-dimensional coordinate system relative to pocket positions corresponding to the images of the real pocket anomalies to obtain position information of the abnormal pockets, specifically includes: after obtaining all the images of the real pocket anomalies and the coding numbers of the corresponding several anomalies, using the laser beam to irradiate a single-line laser to several abnormal pockets corresponding to the images of the real pocket anomalies to obtain several laser beams; and positioning beam inclination angles for the several laser beams according to a two-dimensional coordinate system preset on a rotating table to obtain several beam inclination angles, and determining the position information of the abnormal pockets of the several anomalous pockets based on the several beam inclination angles and the coding numbers of the several anomalies.

[0022] In the embodiment of the present application, after obtaining the images of all the real pocket abnormalities and the corresponding several abnormality coding numbers, the laser beam emitted from the laser emitter at the center of the rotary table is irradiated onto the abnormal pockets to form several laser beams, and then the inclination angle of each abnormal pocket is determined according to the two-dimensional coordinate system preset on the rotary table, and then the position information of the abnormal pocket is finally determined according to the corresponding abnormality coding number, which helps the operator to timely find the corresponding abnormal pocket according to the abnormal pocket position information obtained from the inclination angle and the abnormality coding number, and saves the operator's searching time, thereby further improving the efficiency of bearing cage pocket inspection.

[0023] In another aspect, an embodiment of the present application also provides a bearing retainer pocket inspection device comprising at least one processor and a memory communicatively connected to the at least one processor, the memory storing instructions executable by the at least one processor, thereby enabling the at least one processor to execute the bearing retainer pocket inspection method described in any of the above embodiments. Effect of the Invention

[0024] In the present application, a rotary table and a camera are used to inspect the pockets of the bearing cage, and all the recognized pockets are compared with each other, and then the abnormal pockets are screened out to determine the type of pocket abnormality. Then, the position and abnormal type of the abnormal pocket can be accurately recognized, and the images of the actual pocket abnormality are sent to the back-end operators, which can help the operators to greatly reduce the workload of pocket inspection, and the accuracy rate is high. Combined with laser positioning, it can help the operators to quickly find the pockets with abnormal conditions and determine the corresponding abnormal fault type, which greatly reduces labor costs and improves inspection efficiency. Furthermore, the pockets can be inspected comprehensively and thoroughly, reducing human errors and improving the straight-through rate of the bearing cage. [Brief description of the drawings]

[0025] In order to more clearly describe the technical solutions of the embodiments of the present application or the prior art, the following briefly introduces drawings to be used in the description of the embodiments or the prior art. It goes without saying that the drawings in the following description are only some of the embodiments described in the present application, and those skilled in the art can also obtain other drawings based on these drawings without performing any novel work. In the drawings,

[0026] [Figure 1] FIG. 1 is a flowchart of a pocket inspection method for a bearing cage provided by an embodiment of the present application. [Diagram 2]FIG. 2 is an overall structural diagram of a pocket inspection of a bearing cage provided by an embodiment of the present application. [Diagram 3] FIG. 3 is a plan view of a two-dimensional coordinate system of a rotary table provided by an embodiment of the present application. [Figure 4] FIG. 4 is a schematic diagram of a pocket structure of a bearing cage provided by an embodiment of the present application. [Diagram 5] FIG. 5 is a schematic diagram of a pocket and column structure of a bearing cage provided by an embodiment of the present application. [Figure 6] FIG. 6 is a structural schematic diagram of a pocket inspection device for a bearing cage provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0027] In order to allow those skilled in the art to better understand the technical solution of the present application, the technical solution of the embodiments of the present application will be described below clearly and completely with reference to the drawings of the embodiments of the present application, it goes without saying that the described embodiments are only a part of the embodiments of the present application, not all of the embodiments, and all other embodiments that those skilled in the art can obtain from the embodiments of the present application without making any novel work shall fall within the protection scope of the present application.

[0028] An embodiment of the present application provides a pocket inspection method for a bearing cage, and as shown in FIG. 1, the pocket inspection method for a bearing cage specifically includes steps S101 to S106. S101: A camera takes successive pocket images of a bearing cage on a rotary table to obtain an initial pocket image set.

[0029] Specifically, the drawing information of the bearing cage to be manufactured is obtained, and the drawing information includes at least the number of pockets, the size of the pockets, the radius of the bearing cage, and the processing parameters of the columns. The number of pockets and the radius of the bearing cage in the drawing information are input to a control unit of a rotary table control system. The rotary table control system includes a control unit, a servo motor unit, and a laser emission unit.

[0030] Furthermore, the angular velocity of the rotary table is determined from the distance between the cameras based on the number of pockets, and each pocket of the bearing cage is photographed.

[0031] In one embodiment, Fig. 2 is an overall structure diagram of the pocket inspection of the bearing cage provided by the embodiment of the present application. As shown in Fig. 2, first, the drawing information of the bearing cage to be manufactured is uploaded to the pocket inspection system of the bearing cage. Fig. 4 is a schematic diagram of the pocket structure of the bearing cage provided by the embodiment of the present application, and Fig. 5 is a schematic diagram of the pocket and column structure of the bearing cage provided by the embodiment of the present application. As shown in Figs. 4 and 5, the pockets and columns of the bearing cage have various complex types of structures that are suitable for the operating characteristics of various bearings, so the pass inspection of the pockets of the bearing cage is a condition for the stable operation of the bearing. For example, the number of pockets to be manufactured is 36, the length and width of the pockets are 4 x 6 cm, the thickness is 1 cm, the radius is 35 cm, the column has four inclined surfaces, the angle of each inclined surface is 45°, the front of each column has three holes with a radius of 2 mm, and the bottom end of each pocket has two grooves with a width of 1 cm, all of which are provided for the smooth flow of lubricating liquid. Next, the pocket number information in the drawing information and the radius information of the bearing cage are input into the rotary table control system by the pocket inspection system, and the rotational angular velocity of the servo motor is calculated from the camera shooting interval, so that the camera and the rotary table can be linked to each other to photograph each pocket of the bearing cage.

[0032] Furthermore, the operation of the rotary table is controlled based on the angular velocity of the rotary table, and the camera continuously captures pocket images of the bearing cage to obtain a partial pocket image set, which is an image set captured by the camera at a first plurality of time intervals, for example, the camera captures an image set at a first five time intervals.

[0033] Furthermore, based on a preset Canny operator, the area of ​​the pillar pixel is detected for the grayscale image set of the partial pocket image set to obtain an actual pillar pixel area. Based on the reference pixel area of ​​the pillar, the area value is matched for the actual pillar pixel area. The actual pillar pixel area that is successfully matched is determined as the complete pillar pixel value. The pocket image corresponding to the complete pillar pixel value is determined as the first pocket image. Based on the first pocket image and the corresponding camera shooting time interval, each pocket of the bearing cage is completely photographed by re-feeding back to the control unit of the rotary table control system to obtain an initial pocket image set.

[0034] In one embodiment, as shown in FIG. 2, first perform partial shooting pre-processing to determine the running time of the servo motor in the rotary table so that the camera can be fully matched to the rotary table, i.e., each image captured by the camera contains one complete bearing retaining pocket, and then roughly extract using the Canny operator to detect the area of ​​the pillar pixel for the grayscale image set of the partial pocket image set to obtain the actual pillar pixel area, i.e., calculate the pixel area of ​​the pillar in each image one by one, then convert it into the reference pixel area of ​​the pillar based on the reference area of ​​the pillar in the drawing information, and further match the actual pillar pixel area with the area value to determine the actual pillar pixel area that is successfully matched as the complete pillar pixel value, and define the pocket image corresponding to the recognized complete pillar pixel value as the first pocket image, and then feed back the first pocket image and the corresponding camera time interval to the control unit to control the rotation operation of the servo motor, so that the angular velocity of the servo motor and the shooting time of the camera are matched with each other, and ensure that each pocket of the bearing retainer is completely photographed.

[0035] S102: comparing and matching each pocket image in the initial pocket image set with each other; and based on the comparison matching result, screening some pocket images with abnormal matching in the comparison matching result to obtain some abnormal pocket images.

[0036] Specifically, grayscale pre-processing is performed on each pocket image to obtain several grayscale pocket images. According to the actual pixel size of the grayscale pocket image, the grayscale pocket image is divided into regions by a preset division line to obtain grayscale pixel regions. Each grayscale pocket image is divided into nine grayscale pixel regions.

[0037] Furthermore, the pixel features in each grayscale pixel region are scanned. The pixel features include a light-dark gradient feature of each pixel point and a corresponding pixel coordinate location feature. Based on the pixel features in each grayscale pixel region, a similarity comparison matching is performed between each grayscale pocket image for several pixel features that correspond one-to-one with each grayscale pixel region in several grayscale pocket images to obtain a comparison matching result. The comparison matching result includes normal matching and abnormal matching. The similarity comparison matching is to match the similarity of corresponding pixel features between any two groups of grayscale pocket images.

[0038] In one embodiment, first, grayscale pre-processing is performed on each pocket image in the acquired initial pocket image set, then, based on the actual pixel size of the grayscale pocket image, each grayscale pocket image is divided by two horizontal and two vertical dividing lines to obtain nine grayscale pixel regions, and then, based on the light-dark gradient feature and corresponding pixel coordinate position feature of each scanned pixel point, similarity comparison matching between each grayscale pocket image is performed for several pixel features that correspond one-to-one with each grayscale pixel region in several grayscale pocket images, i.e., one-by-one matching of the light-dark gradient feature and corresponding pixel coordinate position feature of related pixel point for the same grayscale pixel region in any two grayscale images in the initial pocket image set, and finally obtain several groups of comparison matching results.

[0039] In addition, if the comparison matching result is a matching abnormality, several groups of gray-scale pocket images with matching abnormality are obtained; distinguishing pixel features and corresponding distinguishing gray-scale pixel regions in the several groups of gray-scale pocket images are extracted; and, according to several normal pocket images with the comparison matching result being a matching normality, a normal gray-scale pixel region is obtained.

[0040] Furthermore, based on the convolutional neural network and the sigmoid function, the similarity of the cross pixel features related to the distinct grayscale pixel regions and the normal grayscale pixel regions is learned and trained to obtain a cross-classification network model. The cross-classification network model uses a Siamese network as a basic structure, and the cross-classification network model assigns weights according to the differences between the distinct grayscale pixel regions and the normal grayscale pixel regions, cross-classifies several groups of grayscale pocket images, removes the normal pixel features in the distinct pixel features, and obtains abnormal pixel features in the distinct pixel features. The abnormal pixel features are pixel features corresponding to the abnormal pocket images in the grayscale pocket images of each group. The normal pixel features are pixel features corresponding to the normal pocket images in the grayscale pocket images of each group. Then, several grayscale pixel regions corresponding to the abnormal pixel features are screened to obtain several corresponding abnormal pocket images.

[0041] In one embodiment, for some groups of grayscale pocket images with abnormal matching in the comparison matching result, distinguishing pixel features and corresponding distinguishing grayscale pixel regions are extracted, and also normal grayscale pixel regions are obtained from some normal pocket images with normal matching; then, by utilizing the image similarity comparison effect of the convolutional neural network and the sigmoid function, a cross-classification network model with distinguishing cross-classification effect for the grayscale pixel region image is obtained through learning and training, and the some groups of grayscale pocket images with abnormal matching are cross-classified from the normal grayscale pixel region image template to recognize the normal grayscale pocket images in each group of grayscale pocket images; then, the normal pixel features in the distinguishing pixel features of each group are removed, i.e., the corresponding normal grayscale pocket images are removed, and only the abnormal pocket images corresponding to the abnormal pixel features are left, thereby completing the classification of some groups of grayscale pocket images with abnormal matching, and finally obtaining each abnormal pocket image after classification.

[0042] In S103, an abnormal area image in the abnormal pocket image is recognized and marked to obtain an abnormal marked area in the abnormal pocket image.

[0043] Specifically, an image sequence encoding process is performed for all pocket images in the initial pocket image set to obtain an encoded data set, and an anomaly is marked for the encoding numbers corresponding to the abnormal pocket images in the encoded data set based on the positions corresponding to the initial pocket image set of some abnormal pocket images to obtain an anomaly encoding number.

[0044] In one embodiment, after obtaining an initial pocket image set, each pocket image in the initial pocket image set is encoded with an image sequence to construct an encoded data set, and then, after recognizing an abnormal pocket image, the abnormality is marked on the encoding number corresponding to the abnormal pocket image, for example, by marking with a color mark or a special code, and finally, the encoding number of the marked abnormality is obtained.

[0045] Further, image segmentation is performed on the abnormal pocket image corresponding to the coding number of the abnormality based on the grayscale pixel area corresponding to the abnormal pixel feature to obtain the original abnormal area image. Grayscale processing is performed on the original abnormal area image.

[0046] Furthermore, a rudimentary edge image feature constraint is performed on the original abnormal region image after grayscale processing by second-order differentiation to obtain a candidate abnormal region. A minimum pixel distance is calculated for the centroid and edge image feature region of the candidate abnormal region to obtain a distance constraint. The edge image feature region is the edge pixel region of the candidate abnormal region. An edge slope coordinate system is constructed from the candidate abnormal region based on the image corner and centroid in the candidate abnormal region. Based on the edge slope coordinate system, a slope angle direction is located with respect to the image slope pixels in the candidate abnormal region to obtain a direction constraint. Based on the distance constraint and the direction constraint, a second edge image feature constraint is performed on the candidate abnormal region to obtain an abnormal mark region.

[0047] In one embodiment, first obtain the grayscale pixel area corresponding to the abnormal pixel feature of the input end by the cross-classification network model, i.e. determine the abnormal pocket image corresponding to the coding number of the abnormality, then realize image area division for the abnormal pocket image by the dividing line, and define it as the original abnormal area image. Next, perform edge image feature constraint on the original abnormal area image by using second-order differentiation to obtain a rough candidate abnormal area, then use the image corner and center of gravity in the abnormal area to locate the direction of the inclination angle for the image slope pixel in the candidate abnormal area, and calculate the minimum pixel distance for the center of gravity and edge image feature area of ​​the candidate abnormal area, respectively perform a second feature constraint of the image edge feature on the candidate abnormal area, i.e. use the direction constraint and distance constraint to further perform image edge feature constraint on the candidate area to obtain the exact abnormal area of ​​the abnormal pocket image, then perform box selection mark on the exact abnormal area to box select the abnormal mark area.

[0048] In one possible embodiment, the calculation of the minimum pixel distance in the distance constraint is d ={√((x i -x 平均 ) 2 -(y i -y 平均 ) 2 ),(x i ,y i )∈T}, where T is the upper boundary area of ​​the texture image of the pocket anomaly in the candidate anomaly region, and (x 平均 ,y 平均 ) is the centroid of the abnormal candidate edge of the texture image of the pocket anomaly, and x i and y i are the coordinates of each pixel point in the candidate abnormal region, and D d is the minimum pixel distance. Locating the direction of the tilt angle relative to the image tilt plane pixel in the direction constraint is θ=0.5argtan((2M 11 ) / (M 20 -M 02 )) to obtain a relative angle θ of the image inclination pixel based on M 11 , M 20, M 02 respectively represent different image corners in the candidate abnormal region, where M ij =Σ (r,c)∈R (r0-r) i (c0-c) j is an image corner, R is a candidate anomaly region, (r0,c0) is the centroid of the candidate region, and (r,c) is the intersection coordinate of the texture image of a pocket anomaly in the candidate anomaly region and its edge.

[0049] S104, performing pocket abnormality classification for the abnormal mark area based on the past pocket abnormality database to obtain a pocket abnormality type of the bearing cage, The pocket abnormality type includes at least one of pocket size abnormality, column inclination abnormality, hole penetration abnormality, and groove depth abnormality.

[0050] Specifically, a pocket anomaly classification matching is performed for the abnormal mark area with a pocket anomaly template in the past pocket anomaly database. If the matching is successful, the type of pocket anomaly corresponding to the pocket anomaly template is determined as the type of pocket anomaly of the abnormal mark area.

[0051] In one embodiment, image template matching is performed between the box-selected abnormal mark area and the pocket abnormality template in the past pocket abnormality database. For example, information is collated and matched between parameter information such as pocket size information, pillar inclination information, hole area information, and groove depth information in the pocket abnormality template and corresponding information in the abnormal mark area. If the collation and matching is successful, the type of pocket abnormality corresponding to the pocket abnormality template is determined as the type of pocket abnormality corresponding to the abnormal mark area, that is, the type of pocket abnormality corresponding to the abnormal pocket.

[0052] Furthermore, if the matching is not successful, the type of pocket abnormality is determined for the abnormal mark area. A line detection algorithm based on Hough transform is used to perform line detection of pixel point sets for horizontal and vertical column pixels of the pocket in the abnormal mark area to obtain pocket size information. A Hough transform is used to calculate the slope of a line for a set of pixels with the same luminance gradient in the abnormal mark area to obtain column slope information. A circle detection algorithm based on Hough transform is used to recognize non-intersecting curves in the abnormal mark area to obtain a circular outline area. Based on adjacent connected areas in the abnormal mark area, the area is calculated for the same pixels in the circular outline area to obtain hole area information. A pixel length is calculated for the concave and convex areas in the abnormal mark area to obtain concave groove depth information.

[0053] Furthermore, the drawing information is collated with the pocket size information, pillar inclination information, hole area information, and groove depth information to determine the type of pocket abnormality in the abnormal mark area.

[0054] In one embodiment, as shown in Figures 5 and 6, the type of pocket abnormality associated with the abnormal mark area image that does not match the pocket abnormality template is determined, and the pixel arrangement features in the box-selected abnormal mark area are used to recognize and obtain information on the brightness features of various pixel point sets of the pocket, the same pixel arrangement, the distribution of adjacent connected areas, and the size of the same pixel area through the Hough transform line detection algorithm and the Hough transform circle detection algorithm, respectively obtaining pocket size information, pillar inclination information, hole area information, and groove depth information, etc., and then comparing the error range with the drawing information uploaded to the pocket inspection system, for example, if the error between the actual area of ​​a single hole after conversion in the hole area information and the reference hole area in the drawing information is within 1%, the hole area information is regarded as pass information, otherwise it is regarded as hole penetration abnormality information. Finally, the size information, pillar inclination information, and groove depth information of all unsuccessful pockets are determined as pocket size abnormality information, pillar inclination abnormality information, groove depth abnormality information, etc., respectively, to obtain the pocket abnormality type corresponding to each abnormal pocket.

[0055] S105: using the rotary table and the camera to capture a pocket image for the abnormal pocket corresponding to the type of the pocket abnormality for a second time to obtain an image of the true pocket abnormality; and using the laser beam to perform beam positioning in a two-dimensional coordinate system relative to the pocket position corresponding to the image of the true pocket abnormality to obtain position information of the abnormal pocket.

[0056] Specifically, obtain anomaly coding numbers of anomalous pockets corresponding to the types of pocket anomalies. Based on the sequence position of the anomaly coding numbers in the coding data set, perform rotation feedback control on the servo motor of the rotary table to make the anomalous pockets corresponding to the anomaly coding numbers face the camera. In the rotation feedback control, control the rotation operation of the servo motor based on the shooting time interval between the anomaly coding number and the target coding number. Then, the camera shoots pocket images for some anomalous pockets corresponding to the anomaly coding numbers for a second time to obtain some real pocket anomaly images.

[0057] In one embodiment, after all the types of pocket anomalies are recognized, the servo motor in the rotary table is controlled to be turned off, and then, based on the position in the entire encoding data set of the anomaly code number corresponding to the type of pocket anomaly and the position of the camera, the abnormal pocket corresponding to each anomaly code number is rotated again and moved to the front of the camera, so that the abnormal pocket corresponding to the anomaly code number faces the camera. The rotation time of the servo motor is determined by the anomaly code number, the target code number in front of the camera and the shooting time interval of the camera, and further combined with the constant angular speed of the rotary table, each abnormal pocket of the bearing cage on the rotary table is photographed by the camera for the second time, which helps the operator to view the true situation of the pocket of the bearing cage in the back-end system.

[0058] Furthermore, after obtaining the images of all the real pocket anomalies and the coding numbers of the corresponding several anomalies, a single-line laser is irradiated to several abnormal pockets corresponding to the images of the real pocket anomalies with a laser beam to obtain several laser beams.

[0059] Furthermore, the beam tilt angles are positioned for several laser beams according to a two-dimensional coordinate system preset on the rotary table to obtain several beam tilt angles. The position information of several abnormal pockets is determined according to the several beam tilt angles and the coding numbers of several abnormal pockets.

[0060] In one embodiment, FIG. 3 is a plan view of a two-dimensional coordinate system of a rotary table provided by an embodiment of the present application. As shown in FIG. 3 and FIG. 2, the two-dimensional coordinate system on the rotary table is used to accurately position the abnormal pocket with the laser beam emitted from the laser emitter, and the beam tilt angle formed by the laser beam at this time is obtained. Then, based on the corresponding abnormality coding number, the position information of the abnormal pocket of the related beam tilt angle-abnormality coding number is obtained.

[0061] In S106, a pocket inspection result of the bearing cage is determined based on the position information of the abnormal pocket, the type of the pocket abnormality, and an image of the true pocket abnormality, and the pocket inspection result is transmitted to a terminal of an operator.

[0062] Specifically, according to some related beam tilt angle-abnormality coding numbers, the abnormal pocket position information and some pocket abnormality types are sent to the terminal of the pocket inspection system together with the real pocket abnormality images obtained by the second shooting after feedback control of the rotary table, and the pocket inspection result content is a table-form data content, the header includes the abnormal pocket position information, the pocket abnormality type and the real pocket abnormality images, and then according to the abnormality coding numbers, the data is displayed one by one, and finally the pocket inspection result of the bearing cage is formed. Then the tabular pocket inspection result data is displayed to the workers, who can manually review according to the automatic pocket inspection result of the bearing cage, further helping to ensure the straight rate of the bearing cage, reducing human error and improving the efficiency of the bearing cage pocket inspection.

[0063] In addition, an embodiment of the present application also provides a pocket inspection apparatus for a bearing cage. As shown in FIG. 6, the pocket inspection apparatus 600 for a bearing cage specifically includes: The at least one processor 601 and a memory 602 communicatively coupled to the at least one processor 601, the memory 602 storing instructions executable by the at least one processor 601, such that the at least one processor 601: taking successive pocket images with a camera relative to a bearing cage on a rotary table to obtain an initial set of pocket images; Compare and match each pocket image in the initial pocket image set with each other, and based on the comparison and matching result, screen some pocket images in the comparison and matching result that have abnormal matching to obtain some abnormal pocket images; Recognizing and marking an abnormal area image in the abnormal pocket image to obtain an abnormal marked area in the abnormal pocket image; Classifying pocket abnormalities for the abnormal mark area to obtain a type of pocket abnormality of the bearing cage, the type of pocket abnormality including at least one of pocket size abnormality, column inclination abnormality, hole penetration abnormality, and groove depth abnormality; Using the rotary table and the camera to take a second pocket image for the abnormal pocket corresponding to the type of the pocket abnormality to obtain an image of a real pocket abnormality, and using the laser beam to perform beam positioning in a two-dimensional coordinate system relative to the pocket position corresponding to the image of the real pocket abnormality to obtain position information of the abnormal pocket; From the position information of the abnormal pocket, the pocket inspection result of the bearing retainer can be determined based on the type of pocket abnormality and an image of the actual pocket abnormality, and the pocket inspection result can be transmitted to an operator's terminal.

[0064] In the present application, a rotary table and a camera are used to inspect the pockets of the bearing cage, and all the recognized pockets are compared with each other, and then the abnormal pockets are screened out to determine the type of pocket abnormality. Then, the position and abnormal type of the abnormal pocket can be accurately recognized, and the images of the actual pocket abnormality are sent to the back-end operators, which can help the operators to greatly reduce the workload of pocket inspection, and the accuracy rate is high. Combined with laser positioning, it can help the operators to quickly find the pockets with abnormal conditions and determine the corresponding abnormal fault type, which greatly reduces labor costs and improves inspection efficiency. Furthermore, the pockets can be inspected comprehensively and thoroughly, reducing human errors and improving the straight-through rate of the bearing cage.

[0065] In this application, each embodiment is described in a chain-like manner, and the same or similar parts between the embodiments can be referred to, and the description of each embodiment focuses on the differences from other embodiments. In particular, in the case of the device and non-volatile computer storage medium embodiments, they are similar to the method embodiments, so they are briefly described, and the relevant parts can be referred to the description of the method embodiments.

[0066] Above, specific examples of the present application have been described. Other examples are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the examples and still achieve the desired results. Also, the processes depicted in the figures do not necessarily achieve the desired results in the particular order or sequential order shown. In some embodiments, multitasking or parallel processing may be possible or beneficial.

[0067] The above is merely an embodiment of the present application, and is not intended to limit the present application. Those skilled in the art may have various modifications and changes to the embodiments of the present application. Any amendments, equivalent replacements, improvements, etc. made within the spirit and principle of the embodiments of the present application shall be included in the scope of the claims of the present application.

Claims

1. A method for inspecting pockets of a bearing cage, comprising the steps of: taking successive pocket images with a camera relative to a bearing cage on a rotary table to obtain an initial set of pocket images; Comparing and matching each pocket image in the initial pocket image set with each other, and screening some pocket images with abnormal matching in the comparison matching result according to the comparison matching result to obtain some abnormal pocket images, specifically, when the comparison matching result is matching abnormal, obtaining some groups of grayscale pocket images with matching abnormality; Recognizing and marking an abnormal area image in the abnormal pocket image to obtain an abnormal marked area in the abnormal pocket image, specifically, performing an image sequence encoding process on all pocket images in the initial pocket image set to obtain an encoded data set; marking anomalies against coding numbers corresponding to the anomalous pocket images in the encoded data set based on positions of some of the anomalous pocket images relative to the initial set of pocket images to obtain coding numbers for the anomalies; Segmenting the image into regions for the abnormal pocket image corresponding to the coding number of the anomaly based on the grayscale pixel regions corresponding to the abnormal pixel features to obtain an original abnormal region image, and performing grayscale processing on the original abnormal region image, where the abnormal pixel features are pixel features corresponding to the abnormal pocket images in the grayscale pocket images of each group, and each grayscale pocket image is divided into 9 grayscale pixel regions; By second-order differentiation, elementary edge image feature constraint is performed on the original abnormal region image after grayscale processing to obtain candidate abnormal regions; Calculating a minimum pixel distance for a centroid of the candidate abnormal region and an edge image feature region to obtain a distance constraint, the edge image feature region being an edge pixel region of the candidate abnormal region; constructing an edge slope coordinate system from the candidate anomaly region based on image corners in the candidate anomaly region and the centroid; locating a slope angle direction relative to image slope pixels in the candidate abnormal region based on the edge slope coordinate system to obtain a direction constraint; performing edge image feature constraint on the candidate abnormal region for a second time according to the distance constraint and the direction constraint to obtain the abnormal mark region; Classifying the pocket abnormality for the abnormal mark area to obtain a type of pocket abnormality of the bearing cage, the type of pocket abnormality including at least one of a pocket size abnormality, a column inclination abnormality, a hole penetration abnormality, and a groove depth abnormality; Using the rotary table and the camera, take a pocket image for the abnormal pocket corresponding to the type of the pocket abnormality for a second time to obtain an image of a real pocket abnormality, and perform beam positioning of a laser beam in a two-dimensional coordinate system relative to a pocket position corresponding to the image of the real pocket abnormality to obtain position information of the abnormal pocket; determining a pocket inspection result for the bearing retainer based on the position information of the abnormal pocket, the type of the pocket abnormality, and an image of the actual pocket abnormality, and transmitting the pocket inspection result to an operator's terminal.

2. Specifically, the step of taking successive pocket images of the bearing cage on the rotary table with a camera to obtain an initial pocket image set includes the steps of: Obtaining drawing information of a bearing cage to be manufactured, the drawing information including at least the number of pockets, the size of the pockets, the radius of the bearing cage, and processing parameters of the columns; inputting the number of pockets and the radius of the bearing cage in the drawing information into a control unit of a rotary table control system, the rotary table control system including the control unit, a servo motor unit, and a laser emission unit; determining an angular velocity of the rotary table from an image capturing interval of the camera based on the number of pockets, and capturing images of each pocket of the bearing cage; controlling the operation of the rotary table based on the angular velocity of the rotary table, and continuously photographing pockets with the camera relative to the bearing cage to obtain a partial pocket image set, the partial pocket image set being an image set photographed by the camera at an initial plurality of time intervals; According to a preset Canny operator, detect the area of ​​pillar pixels for the grayscale image set of the partial pocket image set to obtain an actual pillar pixel area, and according to a reference pixel area of ​​the pillar, match the area value of the actual pillar pixel area, and determine the actual pillar pixel area that is successfully matched as a complete pillar pixel value; The method for inspecting pockets of a bearing cage as described in claim 1, characterized in that it includes determining a pocket image corresponding to the complete column pixel value as a first pocket image, and re-feeding back to a control unit of the rotary table control system based on the first pocket image and the shooting time interval of the corresponding camera, thereby completely photographing each pocket of the bearing cage to obtain the initial pocket image set.

3. Specifically, comparing and matching each pocket image in the initial pocket image set with each other includes: performing grayscale preprocessing on each pocket image to obtain several grayscale pocket images; Dividing the grayscale pocket image into grayscale pixel regions by a preset dividing line based on an actual pixel size of the grayscale pocket image, each of the grayscale pocket images being divided into nine grayscale pixel regions; scanning pixel features in each of the grayscale pixel regions, the pixel features including a light-dark gradient feature and a corresponding pixel coordinate location feature of each pixel point; 2. The method for inspecting pockets of a bearing cage according to claim 1, further comprising: performing a similarity comparison matching between each of the grayscale pocket images for several pixel features that correspond one-to-one with each of the grayscale pixel areas in the several grayscale pocket images based on pixel features in each of the grayscale pixel areas to obtain the comparative matching result; the comparative matching result includes normal matching and abnormal matching; and the similarity comparison matching is matching the similarity of corresponding pixel features between any two groups of grayscale pocket images.

4. According to the comparison matching result, screening some pocket images having abnormal matching in the comparison matching result to obtain some abnormal pocket images is specifically Extracting distinguishing pixel features and corresponding distinguishing grayscale pixel regions in the several groups of grayscale pocket images, and obtaining normal grayscale pixel regions according to the several normal pocket images that are normal matched by the comparison matching result; According to a convolutional neural network and a sigmoid function, learn and train the similarity of cross-pixel features related to the distinguished grayscale pixel region and the normal grayscale pixel region to obtain a cross-classification network model, the cross-classification network model is based on a Siamese network; Using the cross-classification network model, weights are assigned according to the differences between the distinguishing grayscale pixel regions and the normal grayscale pixel regions, and cross-classification is performed on the grayscale pocket images of the several groups to remove normal pixel features from the distinguishing pixel features and obtain abnormal pixel features from the distinguishing pixel features, the abnormal pixel features being pixel features corresponding to abnormal pocket images in the grayscale pocket images of each group; 2. The method of claim 1, further comprising: screening a number of grayscale pixel regions corresponding to the abnormal pixel features to obtain the corresponding number of abnormal pocket images.

5. Specifically, the method of classifying the pocket abnormality for the abnormal mark area to obtain the type of the pocket abnormality of the bearing cage includes: Performing classification matching of pocket anomalies for the abnormal mark area with a pocket anomaly template in a past pocket anomaly database, and if matching is successful, determining the type of pocket anomaly corresponding to the pocket anomaly template as the type of pocket anomaly for the abnormal mark area; if the matching is not successful, determining a type of pocket anomaly for the abnormal mark region; A line detection algorithm based on Hough transform is used to perform line detection of pixel point sets for horizontal and vertical column pixels of pockets in the abnormal mark area to obtain pocket size information; Calculate the slope of a line for a set of pixels with the same luminance gradient in the abnormal mark area through Hough transform to obtain slope information of the pillar; Using a circle detection algorithm based on Hough transform, the non-intersecting curves in the abnormal mark area are recognized to obtain a circular outline area, and the area of ​​the same pixel in the circular outline area is calculated according to adjacent connected areas in the abnormal mark area to obtain hole area information; Calculating pixel lengths for the irregular regions in the abnormal mark region to obtain groove depth information; The method for inspecting pockets of a bearing retainer as described in claim 1, characterized in that it includes comparing drawing information with information on the pocket size, the pillar inclination, the hole area, and the groove depth to determine the type of pocket abnormality in the abnormal mark area.

6. Specifically, the method of taking a pocket image for the abnormal pocket corresponding to the type of the pocket abnormality for the second time using the rotary table and the camera to obtain an image of the true pocket abnormality includes: Obtaining an anomaly code number of the abnormal pocket corresponding to the type of the pocket anomaly; performing a rotation feedback control on a servo motor of the rotary table based on a sequence position in the encoded data set of the encoded number of the anomaly, to orient the anomaly pocket corresponding to the encoded number of the anomaly toward the camera, the rotation feedback control controlling a rotational movement of the servo motor based on a photographing time interval between the encoded number of the anomaly and the encoded number of the target; The method for inspecting pockets of a bearing retainer as described in claim 1, further comprising using the camera to take pocket images a second time for some abnormal pockets corresponding to the coded numbers of the abnormality, thereby obtaining images of some of the true pocket abnormalities.

7. Specifically, the method includes using a laser beam to perform beam positioning in a two-dimensional coordinate system relative to a pocket position corresponding to the image of the true pocket abnormality to obtain position information of the abnormal pocket, After obtaining all the images of the real pocket anomalies and the corresponding coding numbers of some anomalies, irradiate some abnormal pockets corresponding to the images of the real pocket anomalies with a single line laser with the laser beam to obtain some laser beams; The method for inspecting pockets of a bearing cage as described in claim 1, characterized in that it includes positioning beam inclination angles for the several laser beams according to a two-dimensional coordinate system preset on a rotary table to obtain several beam inclination angles, and determining position information of several abnormal pockets based on the several beam inclination angles and the several abnormality coding numbers.

8. At least one processor; a memory communicatively connected to the at least one processor; A bearing retainer pocket inspection device characterized in that the memory stores instructions that can be executed by the at least one processor, thereby enabling the at least one processor to execute the bearing retainer pocket inspection method described in any one of claims 1 to 7.

Images