Head abnormal detection device

Abstract

The present invention provides a pier abnormal detection device, which includes a pier detection system, a cooling system, and a data processing system; the cooling system cools down the pier detection system and the pier working environment; the pier detection system includes a vision unit for taking images of the pier; the vision unit is communicatively connected to the data processing system; the data processing system includes a data preprocessing unit, a pier dropping detection unit, a pier surface flaw detection unit, and an alarm unit, which are used for data acquisition, data processing, pier dropping and surface flaw detection during the operation of the pier. The present invention preprocesses the original point cloud data to obtain preprocessed point cloud data, and divides the pier data based on this; then performs pier dropping detection. If the pier has not dropped, pier surface flaw detection is performed. If the pier has dropped, the steps of data preprocessing and pier dropping detection are repeated.

Description

Technical Field

[0001] The present invention relates to the field of industrial detection technology, and particularly to a piercing plug abnormal detection device. Background Art

[0002] The seamless steel pipe production method is a process for producing seamless steel pipes or other metal pipes and alloy pipes by methods such as piercing. This process generally includes processes such as heating, piercing, rolling, sizing, sizing, reducing, and finishing. Among them, piercing is the main process. The piercing plug used in the piercing process is a crucial tool in seamless steel pipe production. During the actual production process, the situation of the plug falling off may occur. The falling off of the plug will bring problems such as inner surface defects of the steel pipe, deviations in the inner diameter and wall thickness of the steel pipe, damage to the piercing mill and rolling mill, and a large amount of heat, dust, and flue gas will be generated during the hot rolling process, and a lot of water vapor will appear during the rapid cooling process. The harsh environment affects the operating state of the equipment and the accuracy of detection. When the piercing plug falls off, the production line must stop running for maintenance, which will lead to production interruption, affecting the production efficiency and delivery date of the factory. In addition, the fallen plug may cause further damage to the equipment, increasing the maintenance cost and time. At the same time, the falling off of the plug will also lead to waste of raw materials and semi-finished products, increasing the cost burden of the enterprise. During the production process of the piercing process, defects such as melting tips and meat loss will occur on the plug, which will cause defects in the pierced and rolled steel pipes, such as guide plate marks, seriously affecting the production quality of hot-rolled steel pipes.

[0003] For example, in the application number: CN202210984474.9, a piercing plug detection system for a steel mill, according to its structural design diagram, uses two groups of high-speed industrial cameras and a group of radars. Through deep learning and machine vision, 6 images collected by the high-speed cameras are processed and predicted to obtain the detection result of the plug falling off.

[0004] For example, in the application number: CN201520420680.2, a piercing plug detection device for a piercing mill, the detection basis comes from the pressure sensor in the funnel-shaped cavity opened inside the device, and relevant information about the plug falling off is obtained through the sensor.

[0005] For example, in the application number: CN202310827620.1, a seamless steel pipe piercing plug detection system based on machine vision, the structural design on which the detection system depends is similar to the structure of the application number CN202210984474.9, and a deep learning neural network is used to judge whether the plug has fallen off by collecting the plug image with a 2D camera group.

[0006] The above methods mostly use a sensing module or a traditional camera to judge whether the plug has fallen off. The self-setting of the sensing module is limited, such as the package size, power consumption requirements, etc. It is difficult for a traditional CCD camera to resist the interference of high-temperature water vapor, fine soot, etc. during the plug piercing process. The above solutions cannot reduce the noise caused by high-temperature water vapor and fine soot during camera shooting, nor can they detect surface defects such as plug tip melting and plug meat loss existing in the hot rolling piercing process. Summary of the Invention

[0007] The technical problem to be solved by the present invention is: how to design a detection device that is less affected by the environment and can detect both plug falling and surface defects at the same time.

[0008] The present invention solves the above technical problems through the following technical means:

[0009] A plug abnormality detection device, including a plug detection system, a cooling system, and a data processing system; the cooling system cools the plug detection system and the plug working environment; the plug detection system includes a vision unit for taking images of the plug; the vision unit is communicatively connected to the data processing system; the data processing system includes a data preprocessing unit, a plug falling detection unit, a plug surface defect detection unit, and an alarm unit;

[0010] The data preprocessing unit receives the point cloud data sent by the vision unit and segments the point cloud data to obtain the plug data;

[0011] The plug falling detection unit detects whether the plug has fallen off based on the segmented plug data. The specific detection process is as follows: perform lateral projection on the plug data to reduce the data dimension and obtain the plug lateral projection image; perform inference on the plug lateral projection image based on the trained deep learning detection model to judge the plug state and obtain the falling determination information; according to the falling determination information, send a falling alarm signal to the alarm unit;

[0012] If the plug has not fallen off, the plug surface defect detection unit is started. The detection process is as follows: perform ray projection on the plug data, project it as a depth image with the plug surface as the reference plane, and obtain the plug cross-section contour projection image; perform binary processing on the depth image according to the depth threshold to generate a binary image; extract the connected regions of the binary image and judge whether the area meets the conditions. If so, perform a retention operation and send an alarm signal to the alarm unit. If not, perform a discard operation.

[0013] Further, the top head detection system includes a support frame for the top head detection, which includes columns and a rotating bracket located at the top of the columns. At the top of the rotating bracket, it is connected to the I-beam bracket through a rear roller member and can drive the I-beam bracket to move. The front end of the I-beam bracket is connected to an angle control mechanism through a front roller member. The angle control mechanism clamps the top head detection system and can drive the top head detection system to adjust omnidirectionally. The cooling system is located on one side of the support frame for the top head detection, and it can act on the top head below the top head detection system with cold air flow through the top head detection system.

[0014] Further, the rear roller member can indirectly push the vision unit to move in the "X" axis direction;

[0015] The angle control mechanism includes a "Z" axis rotation component for adjusting the vision unit to rotate in the "Z" axis, an "X" axis rotation component for adjusting the vision unit to rotate in the "X" axis, and a "Y" axis rotation component for adjusting the vision unit to rotate in the "Y" axis.

[0016] Further, the "Z" axis rotation component includes a first rotating support seat. At the top of the first rotating support seat, it is connected to the front roller member through a support column. At the middle position of the first rotating support seat, a rotating member is provided, and the bottom of the rotating member is connected to the "X" axis rotation component through a rotating column.

[0017] Further, the "X" axis rotation component includes a second rotating support seat and a support seat located below the first rotating support seat. At the middle positions on both sides of the support seat, vertical rods are rotatably connected. Rotating plates are installed on both sides of the vertical rods, and arc-shaped track grooves are provided on the rotating plates. Protrusions are provided at the corresponding positions on both sides of the second rotating support seat;

[0018] When the vertical rod drives the rotating plate to rotate along the "X" axis, the protrusion can move in the corresponding arc-shaped track groove.

[0019] Further, the "Y" axis rotation component includes connecting plates installed on both sides of the support seat. Inside both of the connecting plates, third rotating support seats are provided, and the two third rotating support seats clamp the vision unit.

[0020] Further, the front roller member includes a bottom plate. On both sides of the top of the bottom plate, side plates are provided. Inside the side plates, I-beam front rollers for driving the I-beam bracket are installed. Between the two side plates and below the I-beam front rollers, guide wheels for contacting the lower part of the I-beam bracket are provided.

[0021] Further, the rear roller member includes a bracket base mounted on the top of the rotating bracket. On both sides of the top of the bracket base, support plates are installed. On the inner sides of the two support plates, rear I-beam rollers for driving the I-beam bracket are installed. On both sides of the top of the bracket base and in front of the two support plates, guiding optical axes are also provided.

[0022] Further, the process of the data processing system for processing the original point cloud data is as follows:

[0023] S11. Collect and obtain the original point cloud data;

[0024] S12. Through voxel filtering, downsample the original point cloud data to obtain the downsampled point cloud;

[0025] S13. Density-based point cloud segmentation. Based on the density of points within the neighborhood range, group the points with reachable density into the same cluster, and divide the points with unreachable density into new clusters. Traverse all points until all points have corresponding cluster divisions to obtain the segmented point cloud and segmented clusters;

[0026] S14. For the segmented point cloud, introduce preset conditions, and filter the segmented clusters according to the preset conditions to obtain the target clusters. Among them, for each cluster , calculate the coordinates of the center point ;

[0027] S15. Judge whether the cluster meets the preset point number condition and preset center condition; among them, compare the coordinates of the center point of each cluster with the preset coordinate range;

[0028] S16. If so, perform a retention operation on the current cluster ;

[0029] S17. If not, perform a discard operation on the current cluster.

[0030] Further, the specific method of lateral projection in the top head drop detection unit is as follows:

[0031] S211. Input the point cloud data;

[0032] S212. According to the point cloud data, calculate the coordinate range of the projected point cloud. Among them, calculate the boundaries of the coordinates and coordinates in the point cloud data, define the physical boundaries of the two-dimensional image, and determine the size of the two-dimensional projection image:

[0033]

[0034] S213. Calculate, according to the boundary values of the physical boundary of the two-dimensional image and the size of the two-dimensional projection image, the resolution of the image in the direction and

[0035]

[0036]

[0037] S214. Calculate the coordinate projection;

[0038] S215. Obtain the lateral projection image and obtain the image pixel positions through coordinate conversion operations.

[0039] Further, in the step S215, specifically, using the following mapping logic, convert the coordinates and coordinates of each point in the point cloud data to the image pixel positions :

[0040]

[0041] .

[0042] Further, the specific method of ray projection of the head surface defect detection unit is as follows: Use the least squares method to perform ellipse fitting on each contour of the processed head data to obtain the center of the ellipse major axis , minor axis , rotation angle ; Perform plane projection on the head point cloud data in a ray projection manner. The point cloud data is projected as a depth image with its surface as the reference plane. The ray projection steps include:

[0043] S311. Emit a ray from the center of the circle to the starting phase angle as the starting direction, and emit a ray from the center of the circle to the ending phase angle as the ending direction, and emit multiple rays at a fixed step size between them, satisfying ;

[0044] S312. Calculate the phase and Euclidean distance of each point to the center of the ellipse. Specifically, for each point on the contour line, the calculation methods of the phase and Euclidean distance are , ;

[0045] S313. Calculate the contour point with the closest phase distance on each ray within the given boundary threshold range as the representative point of the ray. Specifically, for ray its closest contour point satisfies If , then this contour point is the representative point of ray , otherwise ray has no representative point;

[0046] S314. Calculate the distance from the representative point to the elliptical surface. According to the elliptical parameters and the phase of the representative point , the elliptical radius in the direction of the representative point can be calculated by Equation . Then the distance from the representative point to the elliptical surface can be calculated by . If >0, it means that this point is above the boundary. If <0, it means that this point is below the boundary;

[0047] S315. Calculate the grayscale value of the depth map. The grayscale value can be calculated by Equation , where is the resolution in the z direction, is the median grayscale value; The width of the finally generated depth image is the number of rays , and the height is the number of contours in the point cloud data; Since the radii of the contours in the head data are inconsistent, resulting in different resolutions of each row of data in the width direction of the depth map, separate calculations are performed. The calculation formula is tep。

[0048] Furthermore, the binary image generation method is: Based on the given depth threshold , calculate the corresponding grayscale threshold through the formula . Compare the grayscale value of the depth map with , and generate a binary image through the formula , where represents the grayscale value of .

[0049] Furthermore, the calculation method for the connected regions of the binary image is: For any connected region , its area is , is the resolution in the x direction, is the resolution in the y direction; If satisfies the condition , then this connected region is considered a defect.

[0050] The advantages of the present invention are:

[0051] 1. The present invention is provided with a rotating bracket, a rear roller member, a front roller member and an angle control mechanism on the plug detection support frame body. Through the cooperation of the above structures, it can be used for horizontal, vertical and omnidirectional rotation adjustment of the plug detection system, continuously adjust the detection perspective of the plug detection system, improve the detection accuracy, better fit the existing production line, and there is no need to change the installation position; this application is flexible and light in installation, convenient to combine with the manufacturer's production line, and does not affect the normal operation of the production line;

[0052] 2. Existing hot rolling production lines are already in production and operation, and the installation space left for new equipment is very limited. Therefore, this application has very high requirements for the flexibility of equipment installation. The superiority of the angle control mechanism lies in that it can conveniently adjust the positions in the X, Y, and Z directions, thereby indirectly driving the omnidirectional adjustment of the plug detection member, with higher flexibility;

[0053] 3. The present invention is provided with a cooling system. Under the high-temperature conditions of the hot rolling production line, without a perfect cooling system, it is very difficult to ensure the long-term stable operation of the detection. The cold air flow of this application can remove the smoke and water vapor below the detection box perspective, providing a good external environment for the detection;

[0054] 4. The heat insulation board frame outside the plug detection member of the present invention adopts high-temperature resistant and heat-insulating materials, which can cope with the high-temperature and corrosive environment in the hot rolling area. The design of the internal detection device is different from that of traditional cameras. Traditional cameras mostly have a high integration of lasers and cameras and the depth of field of the cameras is too small to adapt to the hot rolling environment; in order to meet the detection requirements, this application separates the camera and the laser, realizes free combination, can better meet the depth of field range of hot detection, and ensures that the camera has the best perspective; a cooling device is added to the laser to ensure the normal operation of the laser. In summary, the equipment of this application has the advantages of high temperature resistance, corrosion resistance, anti-interference, easy installation, etc.

[0055] 5. The present invention projects the processed point cloud data into a 2D image by means of lateral projection, and detects the 2D projection image based on the method of deep learning pattern recognition to judge the state of the plug. This method not only improves the reliability and convenience of plug detection, but also provides new ideas for subsequent tasks such as plug melting tip detection and plug meat loss detection, and has a reference role for the research in related fields.

[0056] Since the plug has obvious longitudinal ( direction) and height ( direction) characteristics, the Oz plane can well represent these characteristics and clearly show the contour and size of the plug. And the data in the The plane not only retains the main shape features of the top head, but also makes the features after data projection more intuitive, providing a basis for subsequent morphological recognition.

[0057] 6. The present invention uses ray projection to convert the preprocessed point cloud into a depth image, and then applies image binarization and connected region analysis technology to screen out defect areas that meet the depth threshold and area requirements. This method not only improves the accuracy and efficiency of plug defect detection, but also provides a reference for further research in related fields.

[0058] In the point cloud filtering and point cloud segmentation operations of the present invention, the main purpose of filtering is to downsample the original point cloud, remove redundant data while ensuring data semantics and detail information, and reduce the amount of calculation of the entire processing flow. Point cloud segmentation is to distinguish the top data from the noise data caused by external factors such as water vapor and borax, so as to improve data quality.

[0059] The density-based point cloud segmentation of the present invention can segment the point cloud into different clusters according to the connectivity between the point clouds, thereby improving the segmentation effect of block noise points and reducing false detection or missed detection caused by a large amount of noise data. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic diagram of a device for detecting plug drop and surface defects according to an embodiment of the present invention;

[0061] Figure 2 A schematic diagram of another viewing angle of the device for detecting plug drop and surface defects according to an embodiment of the present invention;

[0062] Figure 3 This is a schematic structural diagram of a plug detection support frame according to an embodiment of the present invention;

[0063] Figure 4 This is a schematic diagram of the structure of a rotating bracket according to an embodiment of the present invention;

[0064] Figure 5 It is a schematic structural diagram of a rear roller member and an I-beam support according to an embodiment of the present invention;

[0065] Figure 6 It is a schematic structural diagram of a rear roller member according to an embodiment of the present invention;

[0066] Figure 7 It is a schematic structural diagram of a front roller member according to an embodiment of the present invention;

[0067] Figure 8 It is a structural schematic diagram of the angle control mechanism according to an embodiment of the present invention;

[0068] Figure 9 This is a schematic structural diagram of a plug detection member according to an embodiment of the present invention;

[0069] Figure 10 Structural schematic diagram of the electric control cabinet in the embodiment of the present invention;

[0070] Figure 11 Trend diagram of the cold air flow direction of the cooling system in the embodiment of the present invention;

[0071] Figure 12 Overall flow block diagram of the top head abnormality detection in the embodiment of the present invention.

[0072] Figure 13 Schematic diagram of the basic steps of the top head dropping detection method based on lateral projection in the embodiment of the present invention;

[0073] Figure 14 Schematic diagram of the specific steps of the preprocessing of the original point cloud in the embodiment of the present invention;

[0074] Figure 15 Schematic diagram of the specific steps of performing lateral projection in the embodiment of the present invention;

[0075] Figure 16 Schematic diagram of the basic steps of the top head surface defect detection method based on ray projection in the embodiment of the present invention;

[0076] Figure 17 Ray projection schematic diagram in the embodiment of the present invention;

[0077] Figure 18 、 Figure 19 Test diagram of the top head dropping detection in the embodiment of the present invention.

[0078] Explanation of reference numerals:

[0079] 1, air cooler; 2, air duct; 3, fan;

[0080] 4, top head detection support frame; 41, pillar;

[0081] 42, rotating bracket; 421, positioning pillar; 422, adjusting friction screw; 423, adjusting frame; 424, rotating rod; 425, bearing;

[0082] 43, rear roller member; 431, bracket base; 432, support plate; 433, I-beam rear roller; 434, support; 435, friction force adjusting screw; 436, locking screw; 437, guiding optical axis;

[0083] 44, I-beam bracket; 441, reinforcing rib; 442, longitudinal beam; 443, diagonal beam;

[0084] 45, front roller member; 451, bottom plate; 452, limiting plate; 453, guiding wheel; 454, side plate; 455, I-beam front roller;

[0085] 46. Angle control mechanism; 461. "Z"-axis rotation assembly; 4611. First rotation support base; 4612. Rotating part; 4613. Supporting column; 4614. Rotating column; 462. "X"-axis rotation assembly; 4621. Second rotation support base; 4622. Rotating plate; 4623. Vertical rod; 4624. Support base; 463. "Y"-axis rotation assembly; 4631. Connecting plate; 4632. Third rotation support base

[0086] 5. Heat insulation board frame

[0087] 6. Electric control cabinet; 601. Touch screen; 602. Cabinet air conditioner; 603. Condensate evaporator

[0088] 7. Steel pipe; 8. Three-jaw pipe clamping machine; 9. Plug

[0089] 10. Vision unit; 101. Laser sensor; 102. Refrigeration chip; 103. Camera Specific embodiments

[0090] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention

[0091] This embodiment provides a plug abnormal detection device, including a plug detection system, a cooling system, and a data processing system; the cooling system cools down the plug detection system and the plug working environment; the plug detection system includes a vision unit for capturing images of the plug; the vision unit is communicatively connected to the data processing system; the data processing system includes a data preprocessing unit, a plug drop detection unit, a plug surface defect detection unit, and an alarm unit

[0092] The data preprocessing unit receives the point cloud data sent by the vision unit and segments the point cloud data to obtain the plug data

[0093] The plug drop detection unit detects whether the plug has dropped based on the segmented plug data. The specific detection process is as follows: perform lateral projection on the plug data to reduce the data dimension and obtain the lateral projection image of the plug; perform inference on the lateral projection image of the plug based on the trained deep learning detection model to judge the state of the plug and obtain the drop determination information; according to the drop determination information, send a drop alarm signal to the alarm unit

[0094] If the tip does not fall off, the surface defect detection unit of the tip is activated, and the detection process is as follows: Perform ray projection on the tip data, project it onto the surface of the tip as a reference plane to obtain a depth image, and acquire the projection image of the cross-sectional contour of the tip; Binarize the depth image according to the depth threshold to generate a binary image; Extract the connected regions of the binary image and determine whether the area meets the conditions. If so, perform a retention operation and send an alarm signal to the alarm unit. If not, perform a discard operation.

[0095] The following describes each system in detail.

[0096] I. Tip detection system and cooling system

[0097] The tip detection system and the cooling system constitute the hardware part of the tip detection, so they are described as a whole. Refer to Figure 1 and Figure 2 , the hardware part of the tip abnormality detection device includes a cold air blower 1, an air duct 2, a fan 3, a tip detection support frame 4, a heat insulation board frame 5, an electric control cabinet 6, and a vision unit 10 installed on the same operation platform or the ground. In this embodiment, the vision unit uses a 3D profiler. Among them, the cold air blower 1, the air duct 2, and the fan 3 are installed on one side of the tip detection support frame 4, and the electric control cabinet 6 is located on the other side, used to control the intelligent control of the cold air blower 1, the fan 3, the tip detection support frame 4, and the vision unit 10.

[0098] Refer to Figure 1 , Figure 2 and Figure 11 , the cold air blower 1 is connected to the fan 3 through the air duct 2, and the fan 3 is connected to the heat insulation board frame 5 and the vision unit 10 through the air duct 2, and then acts on the working area position of the lower tip 9 and the steel pipe 7 from the bottom of the heat insulation board frame 5. It should be noted that the bottom of the steel pipe 7 is supported by a three-jaw pipe clamping machine 8, the tip 9 and the steel pipe 7 are on the same horizontal plane, and the tip 9 is located behind the steel pipe 7.

[0099] Refer to Figure 1 and Figure 2, the top head detection support frame 4 includes a pillar 41, a rotary bracket 42, a rear roller member 43, an I-beam bracket 44, a front roller member 45, and an angle control mechanism 46. Among them, the pillar 41 is vertically fixed, the top of the pillar 41 is provided with a rotary bracket 42, the top of the rotary bracket 42 is provided with a rear roller member 43, the rear roller member 43 is installed at the rear end of the I-beam bracket 44, and a front roller member 45 is also installed at the front end of the I-beam bracket 44. The bottom of the front roller member 45 is connected to the vision unit 10 through the angle control mechanism 46, and a heat insulation plate frame 5 is provided outside the vision unit 10; In summary, through the rotary bracket 42, the I-beam bracket 44 and the vision unit 10 can be driven to rotate along the central axis of the rotary bracket 42 in a large trajectory; through the rear roller member 43 and the front roller member 45, the I-beam bracket 44 and the vision unit 10 can be driven to move horizontally in the "X" axis direction along the I-beam direction; through the angle control mechanism 46, the vision unit 10 can be driven to rotate in all directions, and the vision unit 10 can be adjusted according to the specific working area of the top head 9.

[0100] Refer to Figure 3 , the pillar 41 is in an overall cylindrical structure, and its bottom is provided with threaded holes for fixing on the operating platform or the ground. The top of the pillar 41 is fixedly installed with a rotary bracket 42 through bolts, screws, etc. The height and diameter of the pillar 41 are determined according to the actual operating environment, and this application does not make any limitations.

[0101] Refer to Figure 4 , the rotary bracket 42 includes a positioning pillar 421 fixedly connected to the top of the pillar 41 through bolts, screws, etc. A rotary rod 424 is rotatably connected to the middle position inside the positioning pillar 421. A sector-shaped adjusting frame 423 that can be fixed to the positioning pillar 421 through a pin or a screw, and an adjusting friction top screw 422 that contacts the positioning pillar 421 are arranged outside the rotary rod 424. The top of the rotary rod 424 is connected to a positioning plate fixed to the rear roller member 43 through a bearing 425; Therefore, when controlling the rotation of the rotary rod 424 inside the positioning pillar 421, at this time, both the sector-shaped adjusting frame 423 and the adjusting friction top screw 422 can rotate together with the rotary rod 424 on the positioning pillar 421, thereby indirectly driving the I-beam bracket 44 and the vision unit 10 to rotate around the positioning pillar 421 as the axis. When rotating to the required position, the sector-shaped adjusting frame 423 can be locked and fixed to the positioning pillar 421 by using a pin or a screw. During this process, the rear roller member 43 will not affect the normal rotation of the rotary rod 424.

[0102] Refer to Figure 5 and Figure 6, the rear roller member 43 includes a bracket base 431 installed on the top of the rotating bracket 42. On both sides of the top of the bracket base 431, support plates 432 are installed. On the inner sides of the two support plates 432, rear I-beam rollers 433 for driving the I-beam bracket 44 to move horizontally along the "X" axis are installed. On both sides of the top of the bracket base 431 and in front of the two support plates 432, two guiding optical axes 437 for positioning the horizontal movement of the I-beam bracket 44 are symmetrically provided. On the top of the bracket base 431 and behind the two support plates 432, a support 434 is provided. On the top of the support 434, a locking screw 436 and a friction force adjusting screw 435 are installed. The top of the support 434 can contact the bottom surface of the I-beam bracket 44. By vertically adjusting the vertical position of the friction force adjusting screw 435, the distance from its bottom surface to the bottom surface of the I-beam bracket 44 can be adjusted, thereby adjusting the friction force between it and the I-beam bracket 44. The locking screw 436 can fix the support 434 and the I-beam bracket 44. When the I-beam bracket 44 needs to be positioned, the locking screw 436 can be used to lock the support 434 and the I-beam bracket 44.

[0103] It should be noted that the rear I-beam roller 433 of the present application can be driven by an external motor or can be manually driven.

[0104] Refer to Figure 5 , the I-beam bracket 44 is horizontally arranged as a whole. At the top near the rear end, that is, near one end of the rear roller member 43, a longitudinal beam 442 is provided. The longitudinal beam 442 is perpendicularly arranged with the I-beam bracket 44. The rear side of the longitudinal beam 442 is fixed to the top of the I-beam bracket 44 through an inclined beam 443, and the front side of the longitudinal beam 442 is fixed to the top of the I-beam bracket 44 through a reinforcing rib 441.

[0105] Refer to Figure 3 and Figure 7 , the front roller member 45 is installed at the bottom of the front end of the I-beam bracket 44 and can cooperate with the rear roller member 43 to drive the horizontal movement of the I-beam bracket 44; the front roller member 45 includes a bottom plate 451. On both sides of the top of the bottom plate 451, side plates 454 are provided. On the inner sides of the side plates 454, front I-beam rollers 455 for driving the I-beam bracket 44 are installed. Between the two side plates 454 and below the front I-beam rollers 455, a guiding wheel 453 for contacting the lower part of the I-beam bracket 44 is provided. On the top of the bottom plate 451 and in front of the two groups of side plates 454, two groups of symmetric limiting plates 452 are provided. A gap for passing through the middle vertical plate of the I-beam bracket 44 is left between the two groups of limiting plates 452, which can play a certain limiting role in the horizontal movement of the I-beam bracket 44.

[0106] It should be noted that the front I-beam roller 455 of the present application can be driven by an external motor or can be manually driven.

[0107] Reference Figure 8 , the angle control mechanism 46 includes a "Z" - axis rotation assembly 461 for adjusting the "Z" - axis rotation of the vision unit 10, an "X" - axis rotation assembly 462 for adjusting the "X" - axis rotation of the vision unit 10, and a "Y" - axis rotation assembly 463 for adjusting the "Y" - axis rotation of the vision unit 10.

[0108] Furthermore, the "Z" - axis rotation assembly 461 includes a first rotary support base 4611 located directly below the bottom plate 451. At the top of the first rotary support base 4611, the bottom plate 451 of the front roller member 45 is connected through a support column 4613. A rotating member 4612 is arranged at the middle position of the first rotary support base 4611. The bottom of the rotating member 4612 is connected to the "X" - axis rotation assembly 462 through a rotating column 4614. It should be noted that the rotating member 4612 can be driven to rotate by an external motor. Through holes leading to the bottom are opened in the rotating member 4612 and the rotating column 4614. When cooling is carried out later, the air duct 2 can be inserted here to act the cold air on the vision unit 10. When air cooling is not carried out, the air duct 2 can be taken out to prevent affecting the normal rotation of the rotating member 4612 and the rotating column 4614.

[0109] Furthermore, the "X" - axis rotation assembly 462 includes a second rotary support base 4621 located directly below the first rotary support base 4611. A support base 4624 is also provided below the second rotary support base 4621. Vertical rods 4623 are rotatably connected to the middle positions on both sides of the support base 4624. Rotating plates 4622 are installed on both sides of the bottom of the vertical rods 4623. Arc - shaped track grooves are opened on the rotating plates 4622. Protrusions are arranged at the corresponding positions on both sides of the second rotary support base 4621. When the vertical rods 4623 drive the rotating plates 4622 to rotate along the "X" - axis, the protrusions can move in the corresponding arc - shaped track grooves. The contact position between the vertical rods 4623 and the second rotary support base 4621 is the rotating part, and this rotating part can be driven by an external motor.

[0110] Furthermore, the "Y" - axis rotation assembly 463 includes connecting plates 4631 installed on the other two sides of the support base 4624. Inner sides of the two connecting plates 4631 are respectively provided with third rotary support bases 4632. The two third rotary support bases 4632 clamp the vision unit 10, and the two third rotary support bases 4632 can be driven by an external motor or manually driven by a handle connected externally.

[0111] Reference Figure 9, the heat insulation board frame 5 is an overall rectangular frame with an upper opening, and a vision unit 10 is installed inside it. A laser sensor 101 and a camera 103 are arranged inside the vision unit 10. Both the laser sensor 101 and the camera 103 can act on the operation position of the lower top head 9 through the vision unit 10 and the external heat insulation board frame 5; a refrigeration sheet 102 is arranged above the laser sensor 101; after the equipment runs normally, first calibrate the detection module camera 103 and the laser sensor 101. During the hot rolling process, the temperature of the pipe is very high. It is necessary to ensure that the distance between the detected laser and the camera and the top head pipe is about 1.5 m, and the laser line falls in the direction perpendicular to the pipe. Then adjust the angle of the camera to ensure that the brightest laser line can be seen within the field of view of the camera. After that, start the fan 3 to test the intensity of the air flow and the refrigeration effect.

[0112] Refer to Figure 10 , a touch screen 601 is arranged at the front end of the electric control cabinet 6, and a cabinet air conditioner 602 and a condensate evaporator 603 are also arranged on one side of the electric control cabinet 6. The electric control cabinet 6 is used for intelligent control of each component of the present application, such as intelligent control of the cold air blower 1, the fan 3, the vision unit 10, the rear roller member 43, the front roller member 45, the "Z" - axis rotation assembly 461, the "X" - axis rotation assembly 462, and the "Y" - axis rotation assembly 463.

[0113] The specific operation principle of the present application is as follows:

[0114] Equipment installation overview ( Figure 1 ), mainly including a rotating bracket 42, an I - beam bracket 44, a rear roller member 43, a front roller member 45, an angle control mechanism 46, a heat insulation board frame 5, a vision unit 10, a cold air path, an electric control cabinet 6, etc. Select a suitable installation position and fix the support column 41. After assembly, then through the angle control mechanism 46, adjust the X, Y, and Z axes of the vision unit 10 so that the viewing angles of the laser and the camera in the vision unit 10 box are in the best position. Finally, fix the I - beam bracket 44, lock the rear roller member 43 and the front roller member 45. Then start assembling the cold air line. After completion, power on to verify whether the functions of each module are normal;

[0115] After the equipment is powered on and running, the cold air blower 1 will continuously output high - pressure air flow. After the high - pressure air flow passes through the high - power fan 3, the air flow continues to increase. The strong air flow passes through the air duct and reaches the vision unit 10 box body. The air flow will quickly take away the heat in the box body, playing a role in cooling the equipment. The air flow that takes away the heat continues to blow out from the air outlet, clearing the water vapor and dust below the detection box, ensuring a good environment within the detection viewing range. The camera 103 and the laser sensor 101 in the detection box body will be linked with the PLC of the production line. The laser and the camera will be turned on before the top head passes through to start data acquisition.

[0116] II. Detection of abnormal top head

[0117] Based on the above first point, the visual unit captures the image of the plugging station and sends it to the data processing system. After denoising, the plug image data is obtained, and then it is detected by the plug dropping detection unit and the plug surface abnormality detection system. The specific detection process is as Figure 12 shown in the following, including the following steps:

[0118] Step S1: Preprocess the original point cloud data to obtain preprocessed point cloud data, and segment the plug data therefrom;

[0119] Step S2: Plug dropping detection. If the plug has not dropped, execute Step S3. If the plug has dropped, repeat Steps S1 and S2; The plug dropping detection includes the following steps:

[0120] S21: Perform lateral projection on the plug data to reduce the data dimension and obtain the plug lateral projection image;

[0121] S22: Perform data annotation on the plug lateral projection image to construct a deep learning detection model;

[0122] S23: Based on the deep learning detection model, perform inference on the plug lateral projection image to judge the plug state and obtain the dropping judgment information;

[0123] S24: According to the dropping judgment information, perform dropping alarm;

[0124] Step S3: Plug surface defect detection, including the following steps:

[0125] S31: Perform ray projection on the plug data, project it onto the depth image with the plug surface as the reference plane, and obtain the plug cross-section contour projection image;

[0126] S32: Perform binary processing on the depth image according to the depth threshold to generate a binary image;

[0127] S33: Extract the connected regions of the binary image and judge whether the area meets the conditions. If so, perform the retention operation and alarm. If not, perform the discard operation.

[0128] The following will describe in detail the data preprocessing, plug dropping detection, and plug surface defect detection one by one.

[0129] I) Data preprocessing

[0130] As Figure 2As shown, in this embodiment, the point cloud data is collected by the 3D profiler arranged at the top working position. In the working environment, a cooling device has been arranged to cool the top working position environment to reduce the influence of water vapor on the collected data, but there is still a small amount of water vapor and other influencing factors in the environment, resulting in large noise in the point cloud data.

[0131] Voxel filtering downsampling;

[0132] In this embodiment, there are a large number of dense points in the original 3D point cloud data, which contains redundant information. Downsampling the point cloud on the basis of meeting the task accuracy requirements is a prerequisite for ensuring the real-time performance of the algorithm. Voxel filtering is a commonly used downsampling method. Its core idea is to divide the point cloud into 3D grids (i.e., voxels) and replace all points in the voxel with the representative points in each voxel, thereby reducing the amount of point cloud data. For the input point cloud, its boundary range is The side length of the voxel grid is defined as , then each point , the index of the voxel is calculated as:

[0133]

[0134] For each voxel grid, the geometric center of all the points it contains is used as the representative point of the voxel, which is calculated as follows:

[0135] ;

[0136] Obtain the main cluster based on density segmentation;

[0137] In this embodiment, density-based point cloud segmentation can segment point clouds into different clusters according to the connectivity between point clouds, thereby improving the segmentation effect of blocky noise points and reducing false detection or missed detection caused by a large amount of noise data. First, the status of all points is set to unvisited, and then for each point If the point has not been visited, mark it as visited. And calculate the number of points within the domain range ε , which can be expressed as:

[0138]

[0139] in, is a point cloud dataset, express and For the preset point threshold ,like , then mark the point as a noise point. , then mark the point as a core point and create a new cluster. All points within it and its domain are added to this cluster. Then, expand this cluster. For each point in the cluster , calculate the ε-neighborhood of . If , then add all unvisited points within the neighborhood of to the current cluster. In this way, gradually traverse each density-reachable point until no new core points can be found. Repeat the above steps until all points are visited. For the segmented point cloud, there may be some small clusters that do not meet the target requirements or noise clusters that deviate from the specified area. To further extract the target clusters that meet the requirements, the following conditions are introduced to screen the segmented clusters to ensure that the retained clusters meet specific spatial range and density requirements. For each cluster , calculate the coordinates of its center point , and compare it with the set coordinate range. If the cluster center is not within the predefined range, discard this cluster. The specific calculation is as follows:

[0140]

[0141] In this embodiment, the range of the cluster center is set as , , . If and and , then the cluster center meets the conditions and proceeds to the next judgment. Otherwise, discard this cluster. To exclude small clusters generated by noise or sparse areas of the point cloud, retain clusters with the number of points greater than a specific value. Set a minimum number of points threshold . If the number of points of cluster satisfies:

[0142]

[0143] then retain this cluster; otherwise, discard this cluster.

[0144] II) Plug Drop Detection

[0145] Perform lateral projection on the plug data segmented in the foregoing step S1 to reduce the data dimension and obtain the plug lateral projection image;

[0146] As Figure 4 shown, in this embodiment, the step S2 of performing lateral projection further includes the following specific steps:

[0147] S21. Input point cloud data;

[0148] In this embodiment, due to the large amount and high dimension of the 3D point cloud data, direct processing often requires a large amount of computing resources and time. Projection is an effective means to simplify the calculation. Projecting the point cloud data onto a two-dimensional plane can significantly reduce the amount of data, facilitating subsequent analysis operations to be completed quickly and at low cost. Through projection, although some spatial information is discarded, this processing method can greatly shorten the calculation time while ensuring the key features;

[0149] S22. Calculate the coordinate range of the projected point cloud;

[0150] In this embodiment, calculate the coordinates and coordinates in the point cloud data. The boundary values define the physical boundaries of the two-dimensional image, so as to determine the size and resolution of the image. The specific calculation formula is as follows:

[0151]

[0152] S23. Calculate the image resolution;

[0153] In this embodiment, according to the boundary values and the size of the two-dimensional projection image, calculate the resolutions in the and directions:

[0154]

[0155]

[0156] S24. Calculate the coordinate projection;

[0157] S25. Obtain the lateral projection image;

[0158] In this embodiment, convert the and coordinates of each point in the point cloud to the pixel position in the image , and the mapping formula is as follows:

[0159]

[0160] ;

[0161] S26. Perform data annotation on the projection image to construct a deep learning detection model;

[0162] S27. Based on the deep learning detection model, perform inference on the projection image to judge the state of the top head and determine whether a drop occurs;

[0163] S28. Carry out a drop alarm.

[0164] In this embodiment, the head feature recognition model of the present invention is implemented based on a convolutional neural network. The equipment continuously collects and enriches intact and dropped images of the head under different working conditions, and gradually expands the training data set, thereby steadily improving the system's recognition accuracy and detection accuracy of the head features. The system adopts an online detection method, which effectively combines deep learning with machine vision functions: specifically, the machine vision module is responsible for image capture and processing, and obtains image information in the production process in real time; the deep learning module uses the GPU of the same host to analyze and classify image features and distinguish the head status. While ensuring the detection effect, this design greatly improves the overall computing efficiency and real-time response speed of the system, further ensuring the rapidity and reliability of production detection.

[0165] In summary, the present invention is based on the point cloud data collected by a 3D camera, and realizes data downsampling and segmentation of top data and water vapor noise data through point cloud filtering and a density-based clustering algorithm, which makes full use of the spatial three-dimensional information of the point cloud and improves the denoising effect.

[0166] In the point cloud filtering and point cloud segmentation operations of the present invention, the main purpose of filtering is to downsample the original point cloud, remove redundant data while ensuring data semantics and detail information, and reduce the amount of calculation of the entire processing flow. Point cloud segmentation is to distinguish the top data from the noise data caused by external factors such as water vapor and borax, so as to improve data quality.

[0167] The density-based point cloud segmentation of the present invention can segment the point cloud into different clusters according to the connectivity between the point clouds, thereby improving the segmentation effect of block noise points and reducing false detection or missed detection caused by a large amount of noise data.

[0168] The present invention projects the processed point cloud data into a 2D image by lateral projection, and detects the 2D projection image based on the deep learning pattern recognition method to determine the state of the plug. This method not only improves the reliability and convenience of plug detection, but also provides new ideas for subsequent tasks such as plug melt tip detection and plug meat loss detection, and has a reference role in research in related fields.

[0169] Since the top has a distinct longitudinal ( direction, i.e. the length direction of the head) and height ( Direction) characteristics, The Oz plane can well show these features and clearly show the outline and size of the head. The data of direction is relatively redundant and contributes little to the judgment of the head status. The plane not only retains the main shape features of the top head, but also makes the features after data projection more intuitive, providing a basis for subsequent morphological recognition.

[0170] As Figure 7 Figure 8 shown, Figure 7 Figure 8 it is a test diagram. The upper left image in Figure 7 is the image captured by the 3D camera, and the lower left image is the image after lateral projection. It can be seen from the lower left image that there is a top head, indicating that the top head has not fallen off. Figure 8 The displayed are the detection results at different time points. Since the 3D camera captures the top head from top to bottom and lacks the data of the lower half of the top head, the image of the top head after lateral projection is in a downward state. It should be noted here that Figure 7 、 Figure 8 the time in

[0171] is the test time, not the technology disclosure time.

[0172] When it is detected that the top head has not fallen off, continue to detect whether there are defects on the surface of the top head. The specific steps are as follows:

[0173] (1) Calculate the ray projection depth image;

[0174] In this embodiment, for the processed point cloud data, the least squares method is used to perform elliptical fitting on each contour of the top head data to obtain the center of the ellipse major axis minor axis and rotation angle . In order to ensure that the relative positions of the point clouds before and after projection between each contour remain unchanged, a ray projection-based method is used to perform planar projection on the top head point cloud data. The point cloud data is projected as a depth image with its surface as the reference plane for defect detection. See Figure 6 for the projection schematic diagram of a single contour.

[0175] During the projection operation in this embodiment, a ray is emitted from the center of the circle to the starting phase angle as the starting direction, and a ray is emitted from the center of the circle to the ending phase angle as the ending direction. and emit multiple rays at a fixed step size between them, satisfying ;

[0176] As Figure 5 shown, calculate the phase and Euclidean distance from each point to the center of the ellipse. Specifically, for each point on the contour line, the calculation methods of the phase and Euclidean distance are , ;

[0177] Calculate the contour point with the closest phase distance on each ray within the given boundary threshold range as the representative point of the ray. Specifically, for the ray its closest contour point satisfies , if , then this contour point is the representative point of the ray , otherwise the ray has no representative point;

[0178] Calculate the distance from the representative point to the elliptical surface. According to the elliptical parameters and the phase of the representative point , from equation the elliptical radius in the direction of the representative point can be calculated. Then the distance from the representative point to the elliptical surface can be calculated by . If >0, it means this point is above the boundary. If <0, it means this point is below the boundary;

[0179] Calculation of the gray value of the depth map. The gray value in the depth map reflects the undulation of the contour surface compared to the reference plane. The gray value can be calculated by the equation , where is the resolution in the z direction, is the median gray value. The width of the finally generated depth image is the number of rays , and the height is the number of contours in the point cloud data. Since the radii of the contours in the head data are inconsistent, resulting in different resolutions of each row of data in the width direction of the depth map, separate calculations must be performed. The calculation formula is tep。

[0180] (2) Perform image binarization processing according to the depth threshold;

[0181] In this embodiment, based on the given depth threshold , calculate the corresponding gray threshold through the formula , compare the gray value of the depth map with , and generate a binary image through the formula , where represents the gray value of .

[0182] (3) Extract the connected regions;

[0183] (4) Judge whether the area meets the conditions;

[0184] (5) If so, perform the retention operation and give an alarm;

[0185] (6) If not, perform the discard operation.

[0186] In this embodiment, connected component analysis is performed on foreground pixels. Specifically, for each foreground pixel, if there are foreground pixels within the eight-neighborhood range of its top, bottom, left, right, and four diagonal pixel points, these foreground pixels belong to a group of connected components. All foreground pixels are traversed until all connected components are found. For any connected component , its area is , is the resolution in the x direction, is the resolution in the y direction. If meets the condition , then this connected component is considered a defect.

[0187] In summary, based on the point cloud data collected by the 3D camera, combined with point cloud filtering and density-based clustering algorithms, this method realizes downsampling of the data and effective segmentation of the head data and water vapor noise data. By utilizing the three-dimensional spatial information of the point cloud, the effect of noise removal is significantly improved. On this basis, ray projection is used to convert the preprocessed point cloud into a depth image, and then image binarization and connected component analysis techniques are applied to screen out the defect areas that meet the depth threshold and area requirements. This method not only improves the accuracy and efficiency of head defect detection but also provides a reference for further research in related fields.

[0188] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. Head abnormal detection device, characterized in that, It includes a plug detection system, a cooling system, and a data processing system; the cooling system cools down the plug detection system and the plug working environment; the plug detection system includes a vision unit for capturing images of the plug; the vision unit is communicatively connected to the data processing system; the data processing system includes a data preprocessing unit, a plug drop detection unit, a plug surface defect detection unit, and an alarm unit; The data preprocessing unit receives the point cloud data sent by the vision unit and segments the point cloud data to obtain plug data; The plug drop detection unit detects whether the plug has dropped based on the segmented plug data. The specific detection process is as follows: perform lateral projection on the plug data to reduce the data dimension and obtain a lateral projection image of the plug; perform inference on the lateral projection image of the plug based on the trained deep learning detection model to judge the plug state and obtain a drop determination information; according to the drop determination information, send a drop alarm signal to the alarm unit; If the plug has not dropped, then start the plug surface defect detection unit. The detection process is as follows: perform ray projection on the plug data, project it with the plug surface as the reference plane into a depth image, and obtain a cross-sectional contour projection image of the plug; perform binary processing on the depth image according to a depth threshold to generate a binary image; extract the connected regions of the binary image and judge whether the area meets the conditions. If so, perform a retention operation and send an alarm signal to the alarm unit. If not, perform a discard operation.

2. The plug abnormal detection device according to claim 1, characterized in that The plug detection system includes a plug detection support frame (4), which includes a pillar (41) and a rotating bracket (42) located at the top of the pillar (41). At the top of the rotating bracket (42), it is connected to an I-beam bracket (44) through a rear roller member (43) and can drive the I-beam bracket (44) to move. The front end of the I-beam bracket (44) is connected to an angle control mechanism (46) through a front roller member (45). The angle control mechanism (46) clamps the plug detection system and can drive the plug detection system to adjust omnidirectionally; the cooling system is located on one side of the plug detection support frame (4), and it can act on the plug (9) below the plug detection system with cold air flow through the plug detection system.

3. The plug abnormal detection device according to claim 2, characterized in that, The rear roller member (43) can indirectly push the vision unit (10) to move in the "X" axis direction; The angle control mechanism (46) includes a "Z" axis rotation component (461) for adjusting the vision unit (10) to rotate in the "Z" axis, an "X" axis rotation component (462) for adjusting the vision unit (10) to rotate in the "X" axis, and a "Y" axis rotation component (463) for adjusting the vision unit (10) to rotate in the "Y" axis.

4. The plug abnormal detection device according to claim 3, characterized in that: The "Z"-axis rotation assembly (461) includes a first rotation support base (4611). At the top of the first rotation support base (4611), a front roller member (45) is connected through a support column (4613). A rotating member (4612) is provided at the middle position of the first rotation support base (4611). The bottom of the rotating member (4612) is connected to the "X"-axis rotation assembly (462) through a rotating column (4614).

5. The head abnormal detection device according to claim 3, wherein: The "X"-axis rotation assembly (462) includes a second rotation support base (4621) and a support base (4624) located below the first rotation support base (4611). Vertical rods (4623) are rotatably connected to the middle positions on both sides of the support base (4624). Rotating plates (4622) are installed on both sides of the vertical rods (4623). Arc-shaped track grooves are formed in the rotating plates (4622). Protrusions are provided at the corresponding positions on both sides of the second rotation support base (4621). When the vertical rods (4623) drive the rotating plates (4622) to rotate along the "X"-axis, the protrusions can travel in the corresponding arc-shaped track grooves.

6. The plug abnormal detection device according to claim 3, wherein: The "Y"-axis rotation assembly (463) includes connecting plates (4631) installed on both sides of the support base (4624). Third rotation support bases (4632) are provided on the inner sides of two of the connecting plates (4631). The two third rotation support bases (4632) clamp the vision unit (10).

7. The head abnormal detection device according to claim 2, wherein: The front roller member (45) includes a bottom plate (451). Side plates (454) are provided on both sides of the top of the bottom plate (451). Front I-beam rollers (455) for driving the I-beam bracket (44) are installed on the inner sides of the side plates (454). Guide wheels (453) for contacting the lower part of the I-beam bracket (44) are provided between the two side plates (454) and below the front I-beam rollers (455).

8. The plug abnormal detection device according to claim 2, characterized in that: The rear roller member (43) includes a bracket base (431) installed on the top of the rotating bracket (42). Support plates (432) are installed on both sides of the top of the bracket base (431). Rear I-beam rollers (433) for driving the I-beam bracket (44) are installed on the inner sides of the two support plates (432). Guide optical axes (437) are further provided on both sides of the top of the bracket base (431) and in front of the two support plates (432).

9. The head abnormal detection device according to claim 1, wherein, The processing process of the data processing system for the original point cloud data is as follows: S11. Collect and obtain the original point cloud data; S12. Through voxel filtering, downsample the original point cloud data to obtain the downsampled point cloud; S13. Density-based point cloud segmentation. Based on the density of points within the neighborhood range, group the points with reachable density into the same cluster, and divide the points with unreachable density into new clusters. Traverse all points until all points have corresponding cluster divisions to obtain the segmented point cloud and segmented clusters; S14. For the segmented point cloud, introduce preset conditions, and screen the segmented clusters according to the preset conditions to obtain target clusters. For each cluster , calculate the coordinates of the center point ; S15. Determine the cluster to check whether it meets the preset point number condition and preset center condition; where the coordinates of the center point of each cluster are compared with the preset coordinate range S16. If so, perform a retention operation on the current cluster ; S17. If not, perform a discard operation on the current cluster.

10. The plug abnormal detection device according to any one of claims 1 to 9, characterized in that, The specific method of lateral projection in the top head drop detection unit is as follows: S211. Input the point cloud data; S212. Calculate the coordinate range of the projected point cloud based on the point cloud data, where the boundaries of the coordinates and coordinates in the point cloud data are calculated to define the physical boundaries of the two-dimensional image and determine the size of the two-dimensional projected image: S213. Calculate the image resolutions in the direction and the direction according to the boundary values of the physical boundary of the two-dimensional image and the size of the two-dimensional projection image: direction and direction: S214. Calculate the coordinate projection; S215. Obtain a lateral projection image and obtain the image pixel positions through coordinate conversion operations.

11. The plug abnormal detection device according to claim 10, characterized in that, In S215, each point in the point cloud data is coordinate, coordinate is converted to the image pixel position : 。 12. The head abnormal detection device according to claim 1, wherein, The specific method of ray projection of the head surface defect detection unit is as follows: The least squares method is used to perform elliptical fitting on each contour of the processed head data to obtain the center of the ellipse. major axis , minor axis , rotation angle ; Perform a planar projection on the plug point cloud data in a ray projection-based manner. The point cloud data is projected as a depth image with its surface as the reference plane. The ray projection steps include: S311. From the center of the circle to the starting phase angle Emit a ray As the starting direction, from the center of the circle to the ending phase angle Emit a ray As the ending direction, and Between them, emit multiple rays at a fixed step size , satisfying ; S312. Calculate the phase and Euclidean distance from each point to the center of the ellipse. Specifically, for each point on the contour line , the calculation methods of the phase and the Euclidean distance are , ; S313. Calculate the contour point with the closest phase distance on each ray within the given boundary threshold range as the representative point of the ray. Specifically, for ray its closest contour point satisfies . If , then this contour point is the representative point of ray . Otherwise, ray has no representative point; S314. Calculate the distance from the representative point to the elliptical surface, based on the elliptical parameters and the phase of the representative point , from the formula the elliptical radius in the direction of the representative point can be calculated, and then the distance from the representative point to the elliptical surface can be calculated by . If > 0, it means that the point is above the boundary. If < 0, it means that the point is below the boundary; S315. Calculating the grayscale value of the depth map. The grayscale value can be calculated by the formula where is the resolution in the z direction, and is the median grayscale value. The width of the finally generated depth image is the number of rays , and the height is the number of contours in the point cloud data. Since the radii of the contours in the head data are inconsistent, resulting in different resolutions of each row of data in the width direction of the depth map, separate calculations are performed. The calculation formula is tep.

13. The head abnormal detection device according to claim 12, characterized in that, The binary image generation method is as follows: based on a given depth threshold , calculate the corresponding grayscale threshold through the formula , compare the grayscale value of the depth map with , and generate a binary image through the formula , where , and represents 's grayscale value.

14. The plug abnormal detection device according to claim 13, characterized in that, The method for calculating the connected region of a binary image is as follows: For any connected region , its area is , is the resolution in the x direction, is the resolution in the y direction; if satisfies the condition , then this connected region is considered a defect.

Images