Self-adaptive parameter optimization method for robot welding seam grinding based on laser vision sensor

Abstract

The invention discloses a robot welding seam grinding self-adaptive parameter optimization method based on a laser vision sensor, and belongs to the technical field of industrial robot intelligent grinding. According to the method, weld point cloud features are obtained through real-time scanning of a laser vision sensor; establishing a contact model of a welding seam grinding process based on a macroscopic Hertz contact theory, and establishing a material removal model of a welding seam contour based on a Prayton equation; the method comprises the following steps: configuring an XML (Extensible Markup Language) framework for interaction, and establishing an Xpath framework to describe and read data of the XML framework; finally, the file is sent to a KRC (Kuka Robot Controller), and an inverter is controlled to control the grinding speed to obtain a consistent grinding surface. According to the method, high-precision self-adaptive grinding of the welding seam can be completed, the residual height of the welding seam can be controlled within 0.2 mm finally, and the maximum removal height is controlled within 0.05 mm. The device is suitable for high-quality automatic grinding operation of large components in the industry.

Description

Technical Field

[0001] The present invention relates to a technology related to the field of intelligent grinding of industrial robots, and in particular to an adaptive parameter optimization method for robot weld grinding based on a laser vision sensor. Background Art

[0002] Welding technology is widely used in the manufacturing industry to connect various structural components due to its advantages, including excellent joint strength, strong sealing, low production costs, and low metal consumption. However, cracks, pores, and slag inclusions on the weld surface after welding significantly reduce the machining accuracy and dimensional stability of the joint, easily leading to cracking of the joint surface, thereby reducing the overall workability of the workpiece. To improve the joint strength of welded workpieces, grinding the weld seam after welding is crucial. Currently, the most efficient and cost-effective method is rolling belt grinding. However, traditional rolling belt grinding mostly uses a predetermined path for grinding, requiring the existing workpiece to be pre-processed. This process is not only cumbersome but also requires a certain level of technical skill from the operator, which affects processing efficiency. Furthermore, due to the soft contact and uneven weld height, the material removal depth is inconsistent with the feed depth, affecting the uniformity of the weld surface, resulting in a decrease in the actual workpiece strength. Furthermore, the actual weld direction does not conform to the taught path, making it impossible to guarantee grinding accuracy. Therefore, there is an urgent need to research and implement robotic weld grinding. Summary of the Invention

[0003] In response to the above-mentioned problems, the present invention proposes an adaptive parameter optimization method for robot weld grinding based on a laser vision sensor. A flexible and movable laser vision sensor and a grinding belt wheel are installed at the end of the robot. The laser scans the weld to obtain weld feature points, establishes a macro material removal model, determines the weld height and grinding trajectory, and exchanges structural data with the robot and the vision server in real time. The grinding parameters are continuously changed according to the weld surface, and the grinding wheel belt speed is adjusted to ensure uniform grinding of the weld surface. This overcomes the problem of low surface precision of traditional grinding and improves the accuracy of the grinding work. The present invention is achieved through the following technical solutions: The adaptive parameter optimization method for robot weld grinding based on laser vision sensor specifically includes the following steps: Step 1: Perform system calibration on the robot, including hand-eye calibration, workpiece coordinate system calibration, and tool coordinate system calibration; Step 2: The laser vision sensor fixed at the end of the robot scans and obtains the weld point cloud data in real time based on triangulation method to reconstruct the weld morphology; Step 3: In the PC vision server, a contact model of the weld grinding process is established based on the macro-Hertz contact theory, and a material removal model of the weld profile is established based on the Preston equation; Step 4: Establish the robot offline planning grinding trajectory based on the isometric method, and the robot moves to the processing starting point; Step 5: Extract the weld feature points and calculate the weld removal depth, which is then removed by grinding with a grinding wheel; Step 6: Before reaching each grinding point, the robot sends a response to the vision server. The vision server obtains real-time information from the laser vision sensor and calculates the weld removal depth. Step 7: Calculate the optimal grinding wheel linear speed value based on the removal depth according to the model established in the visual server; Step 8: Configure the XML (Extensible Markup Language) framework for interaction and establish an XPath framework to describe and read the data of the XML framework; Step 9: The XML file is sent to KRC, which then controls the inverter to vary the speed to achieve a consistent grinding surface. Step 10: The robot completes the grinding of the entire weld and stops working; In a preferred embodiment of the present invention, a robotic system performs hand-eye calibration, workpiece coordinate system calibration, and tool coordinate system calibration. A 25.4mm diameter ceramic standard sphere serves as the hand-eye calibration target. The rotation matrix is ​​calculated by scanning the non-rotating ceramic sphere with a translational motion, and the translation matrix is ​​calculated by scanning the ceramic sphere in an arbitrary pose. In a preferred embodiment of the present invention, the elastic deformation characteristics of belt grinding are taken into account. A contact model of the weld grinding process is established based on the macroscopic Hertz contact theory. The contact area of ​​the grinding wheel is considered to be an ellipse. The pressure distribution model can be obtained through calculation: Where (x, y) is the coordinate of any point in the contact area. is the normal grinding force, a and b are the major and minor axes of the ellipse, respectively. At the same time, the material removal model of the weld profile is established based on the Preston equation. The grinding path is aligned with the Y axis of the ellipse. The maximum resection depth of the weld can be obtained by calculation: in, is the Preston coefficient related to the abrasive particle velocity, represents the second kind of elliptic integral, k represents the ratio of the major axis to the minor axis of the contact profile, It represents the nominal contact elastic modulus between the grinding wheel and the weld, A and B represent the relative principal curvatures of the grinding, Indicates the linear speed of the grinding wheel. represents the workpiece feed rate, and finally the Preston coefficient is corrected through orthogonal experiments to obtain an accurate weld removal model; In a preferred embodiment of the present invention, the grinding path planning is based on grinding experience and the equidistant non-uniform rational B-Splines (NURBS) algorithm. The interval distance between the grinding points is defined as 20 mm. The path of the long straight weld is planned based on the one-section polynomial method. The grinding path of the curved weld is planned using a NURBS curve based on the equidistant method. The planning mainly includes curve construction and grinding point calculation. By adding weight coefficients corresponding to the control points, the center line of the weld contour is obtained as the grinding path: The function is a NURBS curve in a rectangular coordinate system, which is a NURBS curve in a homogeneous coordinate system and a constructed n+1 dimensional vector The ratio of for Control point, definition , is the weight factor, is the k-order spline curve basis function of the node vector u; In a preferred embodiment of the present invention, the laser vision sensor is mounted on a multi-degree-of-freedom sensor bracket, so that the laser beam can be projected onto the weld contour at multiple angles and positions. During robot grinding, the weld height at any position can be calculated for data exchange. According to similar triangles, it can be seen that: Among them, M represents the projection point of the laser on the weld surface, and N represents the detection point of the laser on the parent material surface. The images of points M and N in the laser vision sensor are M' and N', β is the angle between the normal line of the weld surface and NN', α is the angle between the laser line and the normal line of the weld surface, is the angle between M'N' and NN', L1 is the distance from the MN line to the vision sensor lens, and L2 is the distance from the M'N' line to the vision sensor lens. The welding height z at any position during robot grinding can then be calculated using the following formula: ; In a preferred embodiment of the present invention, the adaptive parameter optimization process is as follows: the robot sends a response to the server before reaching each grinding point. After the server receives the signal, the visual sensor extracts the weld removal depth based on the triangulation principle and calculates the optimal speed of the grinding wheel: ; In a preferred embodiment of the present invention, XML files are used as the data transmission medium between the robot and the vision server. Due to their data description capabilities and consistent data description format, they can be used for data exchange between heterogeneous systems. An XPath framework is established to describe and read the data in the XML framework, including parameter configuration, data reception, and data transmission. The grinding wheel speed calculated in the vision server is sent to the KRC. The KRC and the grinding system interact in real time through the data structure to adaptively change the linear grinding speed. Based on this data, the inverter is controlled to increase the grinding wheel speed in the convex part and reduce it in the concave part, thereby obtaining a uniform weld surface. Compared with the prior art, the adaptive parameter optimization method for robot weld grinding based on laser vision sensor of the present invention has the following advantages: 1. The previous simple "teach-and-go" method to determine the path cannot guarantee grinding accuracy due to human error and method limitations. This paper combines a laser vision sensor to obtain the weld surface in real time and extract feature points. The NURBS curve based on the equidistant method is used to plan the grinding path. The obtained grinding path fitting curve has higher accuracy. 2. Compared with the two optimization methods of real-time control force and offline grinding parameters, this paper establishes a material removal model and combines it with the adaptive parameter optimization method to grind the weld surface. It not only overcomes the limitations of offline grinding parameters, but also controls the weld residual height within 0.2 mm and the maximum removal height within 0.05 mm, ensuring the uniformity of the weld surface after grinding. 3. Currently, an adaptive intelligent human-machine collaboration method can adaptively compensate the grinding trajectory to promote the optimization of the robot belt grinding trajectory of complex parts, but it does not consider the influence of process parameters. The adaptive optimization method of laser vision sensor and material removal model proposed in this paper can improve the grinding quality by optimizing the grinding process parameters. 4. Intelligent robots replace complex manual operations, reducing labor input, while reducing processing errors and improving standard quality; The present invention will be further described below with reference to the accompanying drawings and specific implementation methods: like Figures 1 to 5 As shown, the adaptive parameter optimization method for robot weld grinding based on laser vision sensor of the present invention comprises the following specific steps: Step 1: A Keyence LJ-G500 laser vision sensor is mounted on the end-of-robot end using a multi-degree-of-freedom sensor bracket along with a grinding system including an abrasive belt wheel. The robot system is calibrated using the teach pendant for hand-eye, workpiece, and tool coordinate systems. This determines the conversion between robot and vision coordinates. Specifically, a mark is placed at a fixed position on the workpiece. The end-of-robot sensor is controlled to position the mark at nine locations within the camera's field of view in a predetermined order. The pixel coordinates of these nine points and the robot's arm coordinates are then automatically generated using the findHomography function. This homography is generated by capturing the mark points on the weld in a predefined virtual nine-square grid format. This enables the correlation between the pixel coordinate system within the image and the robot's coordinate system. By detecting the target position using the CCD, the weld's coordinates in the robot coordinate system are determined, allowing the robot to accurately locate the weld grinding point. The robot coordinates of the weld are obtained by parsing the module using the OpenCV internal function perspectiveTransform, and the robot offset is obtained by subtracting the origin. The homography matrix is ​​as follows: ; Step 2: Use Enthercat to establish communication between the KUKA robot, Keyence LJ-G500 laser vision sensor, and the PC, and create send and receive signal ports on the robot laser vision sensor. The communication implementation procedure is: RET=EKI_init(“XmlCallBack”);RET=EKI_Open(“XmlCallBack”) The XmlCallBack is the name of the port that receives the signal; Step 3: Designing an online data transmission method based on the KUKA robot communication software package, combining XML and C#, this method uses XML files as the data transmission medium between the robot and the vision server. Its data description capabilities and consistent data description format make it suitable for data exchange between heterogeneous systems. When data is exchanged between the KRC and the vision sensor, the exchanged XML file framework must remain consistent. Therefore, an XPath framework was established to describe and read XML framework data, including parameter configuration, data reception, and data transmission. The sanding wheel speed calculated in the vision server is sent to the KRC. This method establishes an XPath framework to describe and read XML data, including parameter configuration, data reception, and data transmission. The EKI_ instructions in the KUKA Ethernet KRC communication software package use program call instructions specific to the interaction process. First, define the communication parameter configuration between KRC and laser vision sensor, including port number and IP number: <configuration> <external> <ip> ***< / ip> <port> ***< / port> < / external> < / configuration> Next, the receiving structure of the robot controller receiving data is specified: <receive> <xml> <element type=""STRING”Tag="***” / "> <element type=""INT”Tag="***” / "> <element type=""BOOL”Tag="***” / "> <element type=""FRAME”Tag="***” / "> <element type=""Sensor / Fre”Set_Flag="***”"> < / element> < / element> < / element> < / element> < / element> < / xml> Finally, the sending structure of the robot controller to send data is specified: <send> <xml> <element Tag=""Robot / Date / last" Pos @X” / > <element Tag=""Robot / Date / last" Pos @Y” / > <element Tag=""Robot / Date / last" Pos @Z” / > <element Tag=""Robot / Date / last" Pos @A” / > <element Tag=""Robot / Date / last" Pos @B” / > <element Tag=""Robot / Date / last" Pos @C” / > <element tag=""Robot / Status” / "> <element tag=""Robot / mode” / "> < / element> < / element> < / xml> < / send> Step 4: The laser vision sensor is mounted on a multi-degree-of-freedom sensor bracket, allowing the laser beam to be projected onto the weld contour at multiple angles and positions. The vision sensor scans the entire weld surface, and the vision server processes and extracts weld feature points. The weld height at the grinding position during robot grinding is calculated based on the principle of triangulation. Based on similar triangles, we know that: Among them, M represents the projection point of the laser on the weld surface, and N represents the detection point of the laser on the parent material surface. The images of points M and N in the laser vision sensor are M' and N', β is the angle between the normal line of the weld surface and NN', α is the angle between the laser line and the normal line of the weld surface, is the angle between M'N' and NN', L1 is the distance from the MN line to the vision sensor lens, and L2 is the distance from the M'N' line to the vision sensor lens; Then, the welding height z at any position during robot grinding can be calculated according to the following formula to achieve real-time data interaction: ; Step 5: Then plan the grinding trajectory points of the robot according to the extracted feature points. The grinding trajectory points of the robot are planned as follows: the grinding path adopts the equidistant method, and the interval distance of the grinding points is defined as 20 mm. The long straight weld adopts the trajectory planning method based on the first-order polynomial, and for the curved weld, the trajectory is planned by the NURBS curve of the equidistant method. The addition and control points based on the equidistant method are The corresponding weight coefficient To plan the grinding path, obtain the NURBS curve under the homogeneous coordinate system, divide it by the constructed n+1 dimensional vector, and obtain the NURBS construction curve under the rectangular coordinate system, so as to extract the coordinates of the weld center point and determine the robot grinding trajectory. :The function is a NURBS curve in a rectangular coordinate system, which is a NURBS curve in a homogeneous coordinate system and a constructed n+1 dimensional vector The ratio of for Control point, definition , is the weight factor, is the k-order spline curve basis function of the node vector u; Step 6: In the visual server, two models are created using a C# program to process the acquired weld feature points. The first model is a contact pressure model for the weld grinding process based on the macro-Hertz contact theory: Where (x, y) is the coordinate of any point in the contact area. is the normal grinding force, a and b are the major and minor axes of the ellipse, respectively. At the same time, the material removal model of the weld profile is established based on the Preston equation. The grinding path is aligned with the Y axis of the ellipse. The maximum resection depth of the weld can be obtained by calculation: in, is the Preston coefficient related to the abrasive particle velocity, represents the second kind of elliptic integral, k represents the ratio of the major axis to the minor axis of the contact profile, It represents the nominal contact elastic modulus between the grinding wheel and the weld, A and B represent the relative principal curvatures of the grinding, Indicates the linear speed of the grinding wheel. represents the workpiece feed rate, and finally the Preston coefficient is corrected through orthogonal experiments to obtain an accurate weld removal model; Step 7: Before reaching each grinding point, the robot sends a response to the vision server. The vision server then acquires the weld image from the laser vision sensor and calculates the weld removal depth based on the triangulation principle. The vision server then calculates the optimal grinding wheel speed based on the parameters of the established Preston equation: ,Use XML file as the data transmission medium between KRC and visual server, Xpath framework is used to describe and read the data of XML frame, and the grinding wheel speed calculated in the server is sent to KRC, and then KRC changes the grinding wheel speed through the inverter; The experimental object of the present invention is a metal welded pipe. Through adaptive parameter optimization between the visual sensor and the robot, it is finally possible to control the residual height of the weld within 0.2 mm and the maximum removal height within 0.05 mm. Compared to existing technologies, this method uses a laser vision sensor to establish a material removal model, enabling real-time interactive data structures between the robot controller and the grinding system. This method continuously adjusts grinding parameters based on the weld surface and adjusts the grinding wheel speed to achieve a uniform weld surface. This method adaptively optimizes grinding process parameters, not only overcoming the problem of low surface uniformity associated with traditional grinding, but also improving grinding efficiency. The above-mentioned specific implementation can be partially replaced in different ways by those skilled in the art without departing from the method and principle of the present invention. The scope of protection of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Other implementation schemes within the scope of its method shall be subject to the constraints of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 Schematic diagram of the robot weld grinding experimental platform Figure 2 It is an isometric method based on non-uniform B-Splines (NURBS) splines Figure 3 Calculation of weld removal height Figure 4 Flowchart of the adaptive parameter optimization method Figure 5 Flowchart for online adjustment of grinding speed based on laser vision sensor.< / receive>

Claims

1. An adaptive parameter optimization method for robot weld grinding based on laser vision sensor, characterized by: The specific steps include: Step 1: Perform system calibration on the robot, including hand-eye calibration, workpiece coordinate system calibration, and tool coordinate system calibration; Step 2: The laser vision sensor fixed at the end of the robot scans and obtains weld point cloud data in real time based on triangulation. Step 3: Establish a contact model for the weld grinding process based on the macro-Hertz contact theory in the PC vision server, and establish a material removal model for the weld profile based on the Preston equation; Step 4: Establish the robot offline planning grinding trajectory based on the isometric method, and the robot moves to the processing starting point; Step 5: Extract the weld feature points and calculate the weld removal depth, which is then removed by grinding with a grinding wheel; Step 6: The robot sends a response to the vision server before reaching each grinding point. The vision server obtains real-time information from the laser vision sensor and calculates the removal depth of the weld. Step 7: Calculate the optimal grinding wheel speed value from the removal depth based on the model established in the visual server; Step 8: Configure the XML framework for interaction and establish an XPath framework to describe and read the data of the XML framework; Step 9: Send the XML file to KRC, which then controls the inverter to change the speed to achieve a uniform grinding surface. Step 10: The robot completes the grinding of the entire weld and stops working.

2. The adaptive parameter optimization method for robot weld grinding based on laser vision sensor according to claim 1, characterized in that: The contact state during the grinding process is regarded as the elastic contact between the cylinder and the sphere. The shape of the contact area is approximately elliptical. A contact simulation model based on Hertz contact theory is established to calculate its pressure distribution model: Where (x, y) is the coordinate of any point in the contact area. is the normal grinding force, a and b are the major and minor axes of the ellipse, respectively.

3. The adaptive parameter optimization method for robot weld grinding based on laser vision sensor according to claim 1, characterized in that: The grinding path is aligned with the minor axis of the ellipse, and the material removal model of the weld profile is established according to the Preston equation, which is expressed as follows: in, is the Preston coefficient related to the abrasive particle velocity, represents the second kind of elliptic integral, k represents the ratio of the major axis to the minor axis of the contact profile, It represents the nominal contact elastic modulus between the grinding wheel and the weld, A and B represent the relative principal curvatures of the grinding, Indicates the linear speed of the grinding wheel. Indicates the workpiece feed speed; When x=0, the maximum weld removal height Zmax can be expressed as: Please note that It is the Preston coefficient related to the abrasive particle velocity. It depends on the size, shape, hardness and grinding angle of the abrasive particles. Except for the wall pressure, the influence of abrasive particles on other factors is usually regarded as a constant. Therefore, the Preston coefficient is finally corrected through orthogonal experiments to obtain an accurate weld removal model.

4. The adaptive parameter optimization method for robot weld grinding based on laser vision sensor according to claim 1, characterized in that: The laser vision sensor is mounted on a multi-degree-of-freedom sensor bracket, allowing the laser beam to be projected onto the weld contour at multiple angles and positions. The weld height at the grinding position during robot grinding is calculated based on the triangulation principle, which is used to implement data interaction. Based on similar triangles, we can know that: Among them, M represents the projection point of the laser on the weld surface, and N represents the detection point of the laser on the parent material surface. The images of points M and N in the laser vision sensor are M' and N', β is the angle between the normal line of the weld surface and NN', α is the angle between the laser line and the normal line of the weld surface, is the angle between M'N' and NN', L1 is the distance from the MN line to the vision sensor lens, and L2 is the distance from the M'N' line to the vision sensor lens; Then the welding height z at any position during robot grinding can be calculated according to the following formula:

5. The adaptive parameter optimization method for robot weld grinding based on laser vision sensor according to claim 1, characterized in that: Before reaching each grinding point, the robot sends a signal to the vision server in response. The vision server then obtains real-time information from the laser vision sensor and calculates the weld removal depth. Finally, the vision server calculates the optimal speed of the grinding wheel based on the parameters of the Preston equation established in Claim 3:

6. The adaptive parameter optimization method for robot weld grinding based on laser vision sensor according to claim 1, characterized in that: XML files are used as the data transmission medium between the robot and the vision server. Due to its ability to describe data and the consistency of data description formats, it can be used for data exchange between heterogeneous systems. An XPath framework is established to describe and read the data of the XML framework, including parameter configuration, data reception and data transmission. The grinding wheel speed calculated in the vision server is sent to the KRC, and then the KRC changes the grinding wheel speed through the inverter. The three parts of the XPath framework for data interaction are: First, define the communication parameter configuration between KRC and laser vision sensor, including port number and IP number: <configuration>< / configuration> <external>< / external> <ip> ***< / ip> <port> ***< / port> Next, the receiving structure of the robot controller receiving data is specified: <receive>< / receive> <xml>< / xml> <element type=""STRING”Tag="***” / ">< / element> <element type=""INT”Tag="***” / ">< / element> <element type=""BOOL”Tag="***” / ">< / element> <element type=""FRAME”Tag="***” / ">< / element> <element type=""Sensor / Fre”Set_Flag="***”">< / element> Finally, the sending structure of the robot controller to send data is specified: <send>< / send> <xml>< / xml> <element Tag=""Robot / Date / last" Pos @X” / > <element Tag=""Robot / Date / last" Pos @Y” / > <element Tag=""Robot / Date / last" Pos @Z” / > <element Tag=""Robot / Date / last" Pos @A” / > <element Tag=""Robot / Date / last" Pos @B” / > <element Tag=""Robot / Date / last" Pos @C” / > <element tag=""Robot / Status” / ">< / element> <element tag=""Robot / mode” / ">< / element> 。

Images