Double-layer feedback grinding system and method based on image recognition and welding seam force control grinding

Abstract

The application is based on a double feedback grinding system and method for image recognition and welding seam force control grinding. The double feedback grinding system comprises a robot grinding main body, a servo displacement machine for grinding structure clamping and displacement, and a grinding process expert system. The robot grinding main body realizes overall translation through the robot grinding main body sliding guide. The grinding process expert system is used for logical control of the double feedback grinding system, planning of the robot grinding track, and welding seam recognition, centralized control, force control, and process sequence control. The application guarantees the welding seam grinding in a constant force state through the feedback of the software and the force control system end, and simulates the artificial grinding state. The image recognition and force control grinding are taken as a whole linkage system, which has double feedback of welding seam grinding, and the double machining precision guarantee is more suitable for efficient precision grinding.

Description

Technical Field

[0001] This invention relates to the field of weld seam grinding technology, and in particular to a dual-layer feedback grinding system and method based on image recognition and weld seam force control grinding. Background Technology

[0002] Construction machinery structural components are subjected to alternating stresses during construction, thus requiring the highest welding quality. Manual fine grinding is required for dovetail welds, cross welds, joint welds, and transition welds after robotic welding to reduce stress concentration. Construction machinery structural components have numerous weld types, complex structures, poor consistency, and high grinding quality requirements. Existing grinding robots cannot meet the needs of weld grinding in construction machinery. For example, existing grinding robots often result in inconsistent weld morphology, leading to substandard grinding quality; inconsistent weld height causes the grinding wheel to grind without contact with the workpiece, resulting in high wear; and they cannot identify the location or perform real-time calculation and compensation for the grinding position. Therefore, manual grinding of welds in construction machinery has always been the preferred method.

[0003] Manual grinding is labor-intensive and requires highly skilled workers, and the harsh working environment qualifies it as a special type of operation. Therefore, grinding robots are needed to replace manual grinding, performing weld grinding according to different weld types, requiring smooth transitions and high surface quality after grinding. Summary of the Invention

[0004] The purpose of this invention is to overcome the defects in the existing technologies and provide a dual-feedback grinding system and method based on image recognition and weld seam force control grinding. Through the calibration of the robot's "hand" and "eye", it can achieve high-efficiency, high-precision and high-degree-of-freedom weld seam grinding.

[0005] The technical solution adopted to achieve the purpose of this invention is:

[0006] A dual-feedback grinding system based on image recognition and weld seam force control includes a robot grinding body, a sliding guide rail for the robot grinding body, a servo positioner for clamping and repositioning the grinding structural parts, and a grinding process expert system. The robot grinding body achieves overall translation through the sliding guide rail. The grinding process expert system is used for the logic control of the dual-feedback grinding system, the planning of the robot grinding trajectory, and weld seam recognition, centralized control, force control, and process sequence control.

[0007] The system uses a line laser to acquire images of the weld seam. After 3D image matching and position and orientation calculation, the optimal grinding starting point is calculated in real time and compared with the image ending point. Finally, a 3D camera is used to take pictures for inspection and feedback, realizing self-learning double-layer feedback and double-layer processing accuracy of weld seam grinding. Through feedback from the software and force control system at the end, the system simulates the manual grinding state to ensure that the weld seam grinding is in a constant force state.

[0008] The robot grinding body includes a robot front-end grinding system and a force-controlled grinding flexible shaft. The robot front-end grinding system is connected to the force-controlled grinding flexible shaft and is powered by the force-controlled grinding flexible shaft.

[0009] The robot grinding body includes a quick-change placement rack for placing the robot's front-end grinding system and a fixed force-controlled grinding flexible shaft.

[0010] The robot grinding body includes an industrial robot system and a grinding front-end execution system connected to the industrial robot system. The industrial robot system identifies and grinds the weld seam through the grinding front-end execution system.

[0011] The grinding front-end execution system includes a laser line scanner, a quick-change interface, and a force control feedback mechanism. The laser line scanner is used for weld seam identification, the quick-change interface is used to achieve quick connection with the front-end grinding system, and the force control feedback mechanism is used to detect the force magnitude of the grinding front-end execution system, and provide timely feedback and correction.

[0012] The robot front-end grinding system includes a force-controlled grinding flexible shaft front end, a quick-change interface, a sanding belt, a grinding wheel, and a power wheel. The force-controlled grinding flexible shaft front end is connected to the force-controlled grinding flexible shaft, which provides power and then transmits the power to the power wheel, which in turn transmits the power to the sanding belt. The grinding wheel and the sanding belt form a grinding arc to achieve weld seam grinding.

[0013] The dual-layer feedback grinding method based on image recognition and weld force-controlled grinding is characterized by being implemented using the dual-layer feedback grinding system based on image recognition and weld force-controlled grinding, and includes the following steps:

[0014] S1. After the AGV transports the grinding structural parts to the designated location, the servo positioner clamps the grinding structural parts for installation.

[0015] S2. After clamping, the industrial robot system controls the laser line scanner to perform 3D line laser weld seam scanning, collect the three-dimensional contour information of the weld seam, form point cloud data, and after processing, form the starting point for weld seam grinding.

[0016] S3. After confirming the starting point of weld seam grinding, calculate the number of grinding layers based on the three-dimensional contour information. Compare the coordinate point after each grinding with the end point of the three-dimensional contour to calculate the grinding amount for the next step. The industrial robot system compares the starting point position with the weld seam of the grinding template and updates the compensation amount to the original robot-calibrated coordinate system.

[0017] S4. The industrial robot system uses the front-end grinding system to grind the structural parts according to the grinding process. During the grinding process, the force-controlled grinding coordinates are compared with the image recognition coordinates to confirm the number of grinding layers and the amount of grinding. The system also controls the laser line scanner to take pictures and provide feedback to inspect the welds.

[0018] In step S2, when acquiring the three-dimensional contour information of the weld, a coordinate construction step is included, which involves constructing a transformation relationship based on the camera system to the base coordinate system. It is used to transform the point position from the camera coordinate system to the robot's base coordinate system.

[0019] In step S2, when scanning a regular weld seam using 3D line laser, the starting point of the weld seam is found by utilizing the slope change feature when scanning a regular weld seam; when scanning a complex curved surface weld seam, the normal vector of the tangent plane of the complex curved surface weld seam is found to be collinear with the radius of curvature of the complex curved surface weld seam. When the curved surface is divided into a sufficiently small predetermined state, the differential micro-curved surface is the tangent plane of the point; the starting and ending positions of the weld seam are located by the surface smoothness.

[0020] In step S2, the starting point for weld grinding is determined based on the different image characteristics of smooth surfaces and uneven surfaces at the weld location; the height difference information of the weld is scanned using a 3D line laser scanner, and the smoothness of the weld height difference is calculated using an operator, including:

[0021] Processing of undisplayed points in weld images after 3D laser scanning:

[0022] Image processing after processing without displaying points: Set the size of the image after processing without displaying points to X×Y, add one row / column of image boundaries to the top, bottom, left and right sides of the image, with a pixel value of 0, and the size of the new image after the addition is (X+2)×(Y+2);

[0023] The image after adding rows / columns is enhanced by operators. Starting from X1 and Y1 of the new image, the absolute values ​​of the eight pixels around the coordinate point and the included angle are taken. Then, the eight pixels around the coordinate point and the included angle are subtracted from the coordinate point. The absolute values ​​are then taken as the new coordinate points.

[0024] Starting from X1 and Y1 of the new image, calculate the entire image in the X and Y directions respectively, moving by 1 pixel value each time; finally, subtract one row / column of the image boundary from the top, bottom, left, and right sides to restore the image size X×Y.

[0025] By adjusting the threshold, the characteristics of the image texture are preserved, resulting in an image with clear welding texture, thus separating the welded area from the non-welded area.

[0026] Image filtering and outlier handling: Standard deviation is used, and the calculation model is as follows:

[0027]

[0028] In the formula, X i Let r be the height of the i-th data point, r be the average height value, and r be the number of points in each row. Outliers are equal to the average value of each row of data. Calculate the standard deviation or variance of each row of data. Calculate the standard deviation or variance of each row of height data in turn, and fit the above data to find the location with the largest change, thus obtaining the boundary value of the welding area.

[0029] This invention relates to a dual-feedback system and method for weld seam force-controlled grinding, based on image recognition and weld seam grinding. It can calculate the optimal grinding starting point in real time according to the weld seam morphology of different workpieces, compare it with the image's ending point, and finally use a 3D camera to take a picture for verification and feedback. This achieves self-learning dual-feedback and dual-layer processing accuracy weld seam grinding. Through the cooperation of a force control system and a robot, this invention ensures that the grinding process is controllable and that the grinding belt has a long lifespan. Attached Figure Description

[0030] Figure 1 This is a three-dimensional view of the dual-layer feedback grinding system based on image recognition and weld seam force control grinding of the present invention.

[0031] Figure 2 This is a three-dimensional view of the main body of the weld seam grinding robot of the dual-layer feedback grinding system based on image recognition and weld seam force control of the present invention.

[0032] Figure 3 This is a three-dimensional diagram of the visual recognition and force control system of the dual-layer feedback grinding system based on image recognition and weld seam force control of the present invention.

[0033] Figure 4 This is a three-dimensional diagram of the grinding mechanism of the dual-layer feedback grinding system based on image recognition and weld seam force control grinding of the present invention.

[0034] Figure 5 This is a schematic diagram of the grinding starting point calculated by the dual-layer feedback grinding system based on image recognition and weld seam force control of the present invention.

[0035] Figure 6 This is a schematic diagram of the dual-layer feedback control of the dual-layer feedback grinding system based on image recognition and weld seam force control grinding according to the present invention. Detailed Implementation

[0036] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0037] This invention employs a linear laser and a grinding device installed at the end effector of a robot to achieve accurate positioning of the "hand" and "eye." Weld seam recognition utilizes the linear laser to acquire weld seam images. After 3D image matching and position / pose calculation, the number of force-controlled grinding layers is calculated and compared with the image's endpoint position. Finally, a 3D camera captures an image for verification and feedback. This invention ensures constant force during weld seam grinding through software and feedback from the end effector of the force control system, mimicking manual grinding. By integrating image recognition and force-controlled grinding into a unified system, it provides dual-layer feedback for weld seam grinding, ensuring higher processing accuracy and making it more suitable for efficient precision grinding.

[0038] like Figure 1 As shown in the figure, the dual-feedback grinding system based on image recognition and weld seam force control in this embodiment of the invention uses a line laser to acquire images of the weld seam. After 3D image matching and position and posture calculation, the optimal grinding starting point is calculated in real time and compared with the image ending point. Finally, a 3D camera takes a picture for verification and feedback, realizing self-learning dual-layer feedback and dual-layer processing accuracy weld seam grinding. That is, after each picture is taken, the sample is updated for the first layer of feedback. During the grinding process, scanning and grinding are synchronized, and the subsequent grinding layers and grinding parameters are updated simultaneously for the second layer of feedback. Through the installed software and the feedback at the end of the force control system, the weld seam grinding is kept in a constant force state, mimicking the state of manual grinding.

[0039] The dual-feedback grinding system based on image recognition and weld seam force control includes a robot grinding body 1, which achieves overall translation through a robot grinding body sliding guide rail 2. The grinding structural component 3 is clamped and repositioned through a servo positioner 4. The grinding process expert system 5 is used for the system's logic control, robot grinding trajectory planning, and the implementation of recognition algorithms, centralized control, force control, process sequence, etc.

[0040] like Figure 2As shown, the robot grinding body 1 includes a robot front-end grinding system 6, a quick-change placement frame 8, a force-controlled grinding flexible shaft 7, a robot base 9, an industrial robot system 10, and a front-end execution system 11. The robot front-end grinding system 6 is connected to the force-controlled grinding flexible shaft 7 for transmission, and is powered by the force-controlled grinding flexible shaft 7. The quick-change placement frame 8 is used to place the robot front-end grinding system 6 and fix the force-controlled grinding flexible shaft 7. The robot base 9 is used to connect the grinding body sliding guide rail 2 and the industrial robot system 10 (i.e., the industrial robot). The industrial robot system 10 realizes the identification and grinding of weld seams through the grinding front-end execution system 11. The quick-change placement frame 8 and the robot base 9 are mounted together on the robot grinding body sliding guide rail 2.

[0041] See Figure 3 As shown, the grinding front-end execution system 11 includes a laser line scanner 111, a quick-change interface 112, and a force control feedback mechanism 113. The laser line scanner 111 is used for weld seam identification, and the quick-change interface 112 is used to realize the quick connection between the grinding front-end execution system 11 and the front-end grinding system 6. The front-end grinding system 6 grinds the workpiece. After the grinding front-end execution system 11 is connected to the front-end grinding system 6, during the grinding process, the force control feedback mechanism 113 detects the magnitude of the force of the grinding front-end execution system 11, provides timely feedback, and makes timely corrections. The robot is equipped with a dynamic force position actuator, which provides feedback force.

[0042] See Figure 4 As shown, the robot front-end grinding system 6 includes a force-controlled grinding flexible shaft front end 601, a quick-change interface 602, an abrasive belt 603, a grinding wheel 604, and a power wheel 605. The power wheel is connected to the force-controlled grinding flexible shaft front end 601, the grinding wheel is connected to the abrasive belt, and the abrasive belt is connected to the power wheel and the grinding wheel via a transmission connection. The quick-change interface is arranged above the force-controlled grinding flexible shaft front end 601. The force-controlled grinding flexible shaft front end 601 is powered by the force-controlled grinding flexible shaft 7, which then transmits the power to the power wheel 605. The power wheel 605 transmits the power to the abrasive belt 603. The grinding wheel 604 and the abrasive belt 603 form a grinding arc to achieve the grinding of the weld seam.

[0043] During the grinding process, grinding is achieved by the grinding wheel contacting the grinding structure 3. During the grinding process, the force control feedback mechanism 113 of the grinding front-end execution system 11 detects the magnitude of the force of the grinding front-end execution system 11. The position or posture of the robot front-end grinding system 6 is controlled according to the force feedback, thereby controlling the contact position and posture of its grinding wheel with the grinding structure 3. This continuous feedback and adjustment ensures that the grinding process is handled in a constant force state. Through real-time monitoring and timely feedback, such as through the installed software and the feedback from the end of the force control system, the weld grinding process is kept in a constant force state, mimicking the manual grinding state, thereby ensuring the quality of the grinding structure.

[0044] The grinding steps of the dual-layer feedback grinding system based on image recognition and weld force control grinding according to an embodiment of the present invention are as follows:

[0045] S1: The grinding structural component 3 is transported by AGV. After the servo positioner 4 receives the wireless signal, the guide rail expansion clamp is released, the grinding structural component 3 is in place, the servo positioner 4 tightens, and the clamping is completed.

[0046] S2: The industrial robot system 10 controls the laser line scanner 111 to scan the weld seam, collect the three-dimensional contour information of the weld seam, form point cloud data, and after processing by interpolation algorithms, finally form the starting point for weld seam grinding.

[0047] S3: The starting point of the formed weld repair is compared with the weld of the grinding template, and the compensation amount is updated to the original robot-calibrated coordinate system. For example... Figure 5 As shown; after confirming the starting point of weld seam grinding, the number of grinding layers is calculated based on the three-dimensional contour information. The coordinate point after each grinding is compared with the end point of the three-dimensional contour to calculate the grinding amount for the next step.

[0048] S4: The industrial robot system 10 grinds the grinding structure 3 sequentially through the front-end grinding system 6 according to the grinding process, and inspects the weld seam by controlling the laser line scanner 111; during the grinding process, the force-controlled grinding coordinates are compared with the image recognition coordinates to confirm the starting point, number of layers and grinding amount of the grinding.

[0049] During the grinding process, after obtaining the three-dimensional contour information point cloud data, a self-learning sample is established and stored in the weld expert database before grinding. During the grinding process, when the weld is inspected using a laser line scanner, a self-learning sample is established and stored in the weld expert database after grinding. The grinding process is controlled by comparing with the weld expert database.

[0050] In step S2, before acquiring the three-dimensional contour information of the weld, the process includes the construction of a coordinate system: based on the transformation relationship from the camera system to the base coordinate system. This enables the conversion of the point location from the camera coordinate system to the robot's base coordinate system.

[0051] Since the camera's position relative to the base coordinate system changes dynamically, obtaining dynamic... An intermediate tool is needed to calculate this information, which can be read from the robot controller at any time. A fixed relationship is required, so the auxiliary tool is fixed to the camera, and this relationship is obtained through calibration. The camera origin is near the intersection of the laser direction and the lens axis direction. The relationship required for camera calibration is the positional relationship between the camera origin and the calibration auxiliary tool.

[0052] In step S2, during 3D line laser scanning, when scanning regular weld seams, the starting point for grinding the weld seam is found using the slope abrupt change feature. When scanning complex curved surface weld seams, the collinearity of the normal vector of the tangent plane of the radius of curvature of the weld seam is found. When the surface is divided sufficiently, the differential micro-surface is the tangent plane at the point. The start and end positions of the weld seam can be located by the smoothness of the surface.

[0053] In step S2, the starting point for weld grinding is determined based on the characteristics of the weld and the different image features of smooth surfaces and uneven planes. This is mainly achieved by using a 3D laser scanner to scan the height difference of the weld and using an operator to calculate the smoothness of the weld height difference to determine the starting point. Taking a dovetail weld in an engineering machinery structural component as an example, the surfaces of the ear plate and the top cover plate are smooth. After welding the transition dovetail weld, the weld surface becomes an uneven transition curved surface, and welding spatter exists on the surface.

[0054] The specific calculation process for determining the smoothness of weld height difference using operators includes:

[0055] The initial scan of the weld seam image after 3D laser scanning revealed that some points were not displayed due to irregular welding and ambient lighting, resulting in an incomplete image. These missing points needed to be processed and marked. After processing these missing points, the image was further processed. The image size was initially set to X×Y. The first step was to add a row / column border to the top, bottom, left, and right sides of the image, with a pixel value of 0. The resulting new image size was (X+2)×(Y+2). Next, operator enhancement was applied to the added image. Starting from X1 and Y1 of the new image, the absolute values ​​of eight pixels surrounding and within the included angle of the coordinate point were taken. Then, the absolute values ​​of these eight pixels were subtracted from the original coordinate point, and the new coordinate point was obtained. Using X1 and Y1, the entire image was calculated in the X and Y directions, shifting by one pixel each time. Finally, the top, bottom, left, and right sides of the image were subtracted from the original image size to restore the X×Y dimensions. By adjusting the threshold, the image texture features were preserved. Through the above algorithm, the welding texture was displayed well. The welded and non-welded areas were well displayed.

[0056] Next, image filtering and outlier processing are performed. To eliminate the impact of outliers on the image, the standard deviation of the data needs to be considered, and the calculation model is as follows:

[0057]

[0058] In the formula, X i Let r be the height of the r-th data point, where r is the average height and r is the number of points per row. The principle is that outliers are equal to the average of each row of data: calculate the standard deviation or variance of each row of data, and calculate the standard deviation or variance of the height data for each row sequentially. Fit the above data to find the locations with large abrupt changes and obtain the boundary values ​​of the welding area.

[0059] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention.

[0060] Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of the equivalents of the claims be included within the invention.

[0061] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A dual-layer feedback grinding system based on image recognition and weld seam force control grinding, characterized in that, The system includes a robot grinding body, a sliding guide rail for the robot grinding body, a servo positioner for clamping and repositioning structural components for grinding, and a grinding process expert system. The robot grinding body achieves overall translation through the sliding guide rail. The grinding process expert system is used for logic control of the dual-layer feedback grinding system, planning of robot grinding trajectory, weld identification, centralized control, force control, and process sequence control. A laser line scanner is used to acquire images of the weld seam. After 3D image matching and position / pose calculation, the number of layers for force-controlled grinding is calculated. The optimal grinding starting point is calculated in real time and compared with the image's ending point. Finally, a 3D camera is used to take a picture for verification and feedback, achieving self-learning, dual-layer feedback, and dual-layer processing accuracy weld seam grinding. Through software and feedback from the end of the force control system, the manual grinding state is simulated to ensure the weld seam grinding is under constant force. The dual-layer feedback grinding method of the image recognition and weld seam force-controlled grinding system includes: S1. After the AGV transports the grinding structural parts to the designated location, the servo positioner clamps the grinding structural parts for installation. S2. After clamping, the industrial robot system controls the laser line scanner to perform 3D line laser weld seam scanning, collect the three-dimensional contour information of the weld seam, form point cloud data, and after processing, form the starting point for weld seam grinding. S3. After confirming the starting point of weld grinding, calculate the number of grinding layers based on the three-dimensional contour information. Compare the coordinate point after each grinding with the end point of the three-dimensional contour to calculate the grinding amount for the next step. The industrial robot system compares the starting point position with the weld of the grinding template and updates the compensation amount to the original robot-calibrated coordinate system. S4. The industrial robot system uses the front-end grinding system to grind the structural parts according to the grinding process. During the grinding process, the force-controlled grinding coordinates are compared with the image recognition coordinates to confirm the number of grinding layers and the amount of grinding. The system also controls the laser line scanner to take pictures and provide feedback for weld inspection. During the grinding process, after obtaining the three-dimensional contour information point cloud data, a self-learning sample is established and stored in the weld expert database before grinding. During the grinding process, when the weld is inspected using a laser line scanner, a self-learning sample is established and stored in the weld expert database after grinding. The grinding is controlled by comparing with the weld expert database. In step S2, the starting point for weld grinding is determined based on the different image characteristics of smooth surfaces and uneven planes at the weld location. This includes: when scanning regular welds using 3D line laser scanning, the starting point for grinding is found using the slope abrupt change feature; when scanning complex curved surface welds, the collinearity between the radius of curvature of the complex curved surface weld and the normal vector of the tangent plane of that radius of curvature is found; when the surface is divided to a sufficiently small predetermined state, the differential micro-surface is the tangent plane of the point; the starting and ending positions of the weld are located by the surface smoothness. This process involves using a laser line scanner to scan the height difference information of the weld seam, using an operator to calculate the smoothness of the weld seam height difference, and determining the welded and non-welded areas, including: Processing of undisplayed points in weld images after 3D line laser scanning: Image processing after processing without displaying points: Set the size of the image after processing without displaying points to X×Y, add one row / column of image boundaries to the top, bottom, left and right sides of the image, with a pixel value of 0, and the size of the new image after the addition is (X+2)×(Y+2). The image after adding rows / columns is enhanced by operators. Starting from X1 and Y1 of the new image, the absolute values ​​of the eight pixels around the coordinate point and the included angle are taken. Then, the eight pixels around the coordinate point and the included angle are subtracted from the coordinate point. The absolute values ​​are then taken as the new coordinate points. Starting from X1 and Y1 of the new image, calculate the entire image in the X and Y directions respectively, moving by 1 pixel value each time; finally, subtract one row / column of the image boundary from the top, bottom, left, and right sides to restore the image size X×Y. By using threshold adjustment, the features of the image texture are preserved, resulting in an image with clear welding texture, thus separating the welding area from the non-welding area.

2. The dual-layer feedback grinding system based on image recognition and weld seam force control grinding according to claim 1, characterized in that, The robot grinding body includes a robot front-end grinding system and a force-controlled grinding flexible shaft. The robot front-end grinding system is connected to the force-controlled grinding flexible shaft and is powered by the force-controlled grinding flexible shaft.

3. The dual-layer feedback grinding system based on image recognition and weld force control grinding according to claim 2, characterized in that, The robot grinding body includes a quick-change placement rack for placing the robot's front-end grinding system and a fixed force-controlled grinding flexible shaft.

4. The dual-layer feedback grinding system based on image recognition and weld force control grinding according to claim 3, characterized in that, The robot grinding body includes an industrial robot system and a grinding front-end execution system connected to the industrial robot system. The industrial robot system identifies and grinds the weld seam through the grinding front-end execution system.

5. The dual-layer feedback grinding system based on image recognition and weld force control grinding according to claim 4, characterized in that, The grinding front-end execution system includes a laser line scanner, a quick-change interface, and a force control feedback mechanism. The laser line scanner is used for weld seam identification, the quick-change interface is used to achieve quick connection with the front-end grinding system, and the force control feedback mechanism is used to detect the force magnitude of the grinding front-end execution system, and provide timely feedback and correction.

6. The dual-layer feedback grinding system based on image recognition and weld force control grinding according to claim 5, characterized in that, The robot front-end grinding system includes a force-controlled grinding flexible shaft front end, a quick-change interface, a sanding belt, a grinding wheel, and a power wheel. The force-controlled grinding flexible shaft front end is connected to the force-controlled grinding flexible shaft, which provides power and then transmits the power to the power wheel, which in turn transmits the power to the sanding belt. The grinding wheel and the sanding belt form a grinding arc to achieve weld seam grinding.

7. The dual-layer feedback grinding system based on image recognition and weld force control grinding according to claim 1, characterized in that, In step S2, when acquiring the three-dimensional contour information of the weld, the process includes coordinate construction, which involves establishing a transformation relationship from the camera system to the base coordinate system. It is used to transform the point position from the camera coordinate system to the robot's base coordinate system.

8. The dual-layer feedback grinding system based on image recognition and weld force control grinding according to claim 1, characterized in that, After separating the welded and non-welded areas, the process also includes determining the boundary values ​​of the welded area by eliminating the influence of abnormal data on the image; Image filtering and outlier handling: Standard deviation is used, and the calculation model is as follows: ； In the formula, Let i be the height of the i-th data point. This is the average height value. The number of dots in each row; Outlier data is equal to the average of each row of data. Calculate the standard deviation or variance of each row of data. Calculate the standard deviation or variance of each row of height data sequentially, row by row. Fit the data to find the location with the largest abrupt change and obtain the boundary value of the welding area.

Images