Welding seam milling guide method based on edge characteristics of base metal

Abstract

The invention provides a weld joint milling guide method based on base metal edge characteristics. The method comprises the steps that base metal connected with weld joints is marked as base metal I and base metal II; selecting an edge line I and an edge line II on the base metal I / base metal II; obtaining a welding part point cloud, and segmenting local point clouds of the edge line I and the edge line II from the point clouds; obtaining a space straight line I / space straight line II corresponding to each local point cloud; solving all intersection points between the space straight line I and the space straight line II, and storing the intersection points in an actually measured feature point set; an actual measurement normal vector P of the current welding part is obtained; solving an offset matrix by using the actually measured feature point set and the actually measured normal vector P; the robot corrects the milling track according to the offset matrix; according to the method, the weld joint is accurately positioned through the base metal edge point cloud information, the offset of the current welding part relative to the teaching welding part is calculated, the robot is guided to accurately and efficiently mill the weld joint based on the offset, excessive milling of the base metal is avoided, and the method is suitable for the automatic milling process of various types of weld joints.

Description

Technical Field

[0001] The invention relates to the field of image processing, and in particular to a weld milling guidance method based on parent material edge features. Background Art

[0002] Welding is a process that heats and pressurizes two or more metal or non-metal parts to achieve atomic bonding, thus forming an inseparable whole. It is one of the most important processing methods in the manufacturing industry and is widely used in many fields such as construction, shipbuilding, automobiles, aviation, and aerospace. Weld milling is a post-weld processing process used to improve the appearance and dimensional accuracy of the weld, reduce stress concentration, or prepare for subsequent processing or testing. It is a high-precision and high-efficiency process.

[0003] With the development of automation technology, the current mainstream weld milling process is: mount a milling cutter on the end of the robot, and program and teach the robot so that a track for milling the weld is obtained at the end of the milling cutter. Whenever a new workpiece is placed, the robot repeats this track to mill the weld on the workpiece surface. This method can effectively reduce labor costs and improve milling efficiency, but due to certain manufacturing errors, placement errors, and fixture clamping errors in the workpiece, it cannot guarantee that the weld state of the workpiece surface is consistent after each milling, and there will often be problems such as excessive residue on the weld surface or over-milling of the parent material. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a weld milling guidance method based on the edge features of the base material. The method describes the current position of the weld part and its own processing and welding status based on the surface point cloud information of the base material of the workpiece, and uses the base material edge point cloud information to accurately locate the weld. Based on the positioning coordinate information, the offset of the current weld part compared to the teaching weld part is calculated, and the robot is guided to accurately and efficiently mill the weld based on the offset. The method can improve the accuracy of robot weld milling and effectively avoid the problem of over-milling of the base material, and is suitable for the automated milling process of various types of welds.

[0005] The technical solution is as follows:

[0006] A weld milling guidance method based on the edge features of a base material, wherein the weld milling process is as follows: a milling cutter is installed at the end of a robot, batch welded parts reach a specified position in sequence, and the robot mills the weld on the welded parts according to the taught milling trajectory;

[0007] The parent materials connected on both sides of the weld are respectively marked as parent material I and parent material II; the edge line where the weld is located is selected on parent material I / parent material II as edge line I, and at least two edge lines are selected on the other parent material and marked as edge lines II respectively;

[0008] Guide the robot to complete milling using:

[0009] ① Obtain the surface point cloud of the current weldment, and sequentially segment the edge line I and the local point cloud corresponding to each edge line II from the point cloud;

[0010] ② Obtain the spatial straight line corresponding to each local point cloud respectively;

[0011] ③ According to the correspondence between the local point cloud and the edge line I and edge line II, the space line is marked as space line I / space line II;

[0012] Find all the intersection points between space straight line I and space straight line II;

[0013] ④Store each intersection point in the measured feature point set;

[0014] Based on the surface point cloud of the current weldment, the plane normal vector corresponding to the base material I / base material II of the current weldment is obtained, which is recorded as the measured normal vector P; the standard feature point set and the standard normal vector Q of the weldment pre-stored in the milling trajectory of the teaching robot are retrieved; the offset matrix is ​​obtained by using the correspondence between the standard feature point set and each point in the measured feature point set and the correspondence between the standard normal vector Q and the measured normal vector P;

[0015] ⑤ The robot corrects the milling trajectory according to the offset matrix, and mills the weld on the current weldment based on the corrected trajectory.

[0016] Furthermore, in step ②, the method of obtaining the spatial straight line corresponding to each local point cloud is as follows:

[0017] For a single local point cloud, the main direction of the local point cloud is obtained, and then a spatial straight line is constructed based on the main direction and the three-dimensional point S, where the three-dimensional point S is one of the projection points of the local point cloud after projection along the main direction, or is a three-dimensional point in the local point cloud.

[0018] Furthermore, the main direction of the local point cloud is obtained as follows:

[0019] The principal direction is obtained by using the principal component analysis method (PCA), or the cross section of the current local point cloud is obtained, and the normal vector n of the cross section is recorded as the principal direction.

[0020] Furthermore, the method of obtaining the cross section of the current local point cloud is as follows:

[0021] Preset the initial normal vector and construct the initial plane;

[0022] After projecting each 3D point in the local point cloud onto the plane, the area occupied by the projected points is counted on the current plane;

[0023] Iteratively update the normal vector, construct a plane based on the updated normal vector, and recalculate the area of ​​the projection point;

[0024] When the area value is less than the threshold or the area value tends to be stable, the iteration is stopped, and the current normal vector is recorded as the normal vector n and the current plane is recorded as the cross section.

[0025] Furthermore, the three-dimensional point S is one of the projection points of the local point cloud after being projected along the main direction. The three-dimensional point S is obtained as follows:

[0026] Step S1, record any coordinate axis of the camera coordinate system as axis A;

[0027] Calculate the angle θ between the normal vector n and the A axis, then use the cross product of the normal vector n and the A axis to obtain the vector C, use the vector C as the rotation axis, rotate the normal vector n by the angle θ to be parallel to the A axis, and obtain the rotation matrix R through Rodrigues transformation;

[0028] Step S2, using the rotation matrix R to rotate each projection point in the cross section to obtain the rotated projection point;

[0029] Extract the other two coordinates except the A-axis coordinate from the coordinates of each projection point after rotation, and store them in a two-dimensional point set;

[0030] Step S3, fitting a curve using a two-dimensional point set, selecting a feature point on the curve, multiplying the coordinates of the rotated projection point corresponding to the feature point by the inverse matrix of the rotation matrix R to obtain the projection point before rotation, and recording the projection point before rotation as the three-dimensional point S;

[0031] Preferably, the characteristic point is a point on the curve with the largest curvature change, a point with a slope of 0, a point with a maximum ordinate value, or a preset distance point, wherein the ordinate is a coordinate component perpendicular to the direction of the parent material I / parent material II, and the preset distance point is: a tangent is drawn at a point with a slope of 0, the tangent is translated by a preset distance to intersect with the curve, and the intersection is recorded as the preset distance point.

[0032] Furthermore, the edge line I and the edge line II intersect or are not in the same plane;

[0033] In step ③, the intersection points include real intersection points and virtual intersection points (intersection points of non-coplanar spatial lines);

[0034] Preferably, the virtual intersection is: point I, point II or the midpoint of point I and point II; point I is the point on spatial straight line I that is closest to spatial straight line II, and point II is the point on spatial straight line II that is closest to spatial straight line I.

[0035] Furthermore, there are two ways to obtain the virtual intersection:

[0036] Method 1:

[0037] The direction vector b1 of the space line I is cross-producted with the direction vector b2 of the space line II to obtain the vector b3;

[0038] Use vectors b1 and b3 to construct a first plane, use b2 and b3 to construct a second plane, and calculate the straight line where the first plane and the second plane intersect, which is recorded as space line III;

[0039] Calculate the intersection point between space line I and space line III, recorded as point I;

[0040] Calculate the intersection point between space line II and space line III, and record it as point II;

[0041] Point I, point II, or the midpoint between point I and point II is recorded as a virtual intersection point.

[0042] Method 2:

[0043] The direction vector b1 of the space line I is cross-producted with the direction vector b2 of the space line II to obtain the vector b3;

[0044] One of the spatial straight lines is translated along the direction of vector b3 until it is coplanar with the other spatial straight line, and the intersection point between the two spatial straight lines at this time is calculated and recorded as the virtual intersection point.

[0045] Furthermore, in step ④, the offset matrix is ​​obtained as follows:

[0046] Any intersection point in the measured feature point set is recorded as the measured intersection point H, and the intersection point corresponding to the position of point H is found in the standard feature point set and recorded as the standard intersection point K;

[0047] The measured intersection point H is translated to the origin of the camera coordinate system, and a translation matrix T1 is obtained. The coordinates of other intersection points in the measured feature point set are translated using the translation matrix TI, and the translated intersection coordinates and the translated measured intersection point H are stored in the measured translation point set;

[0048] The standard intersection point K is translated to the origin of the camera coordinate system to obtain a translation matrix T2; the coordinates of other intersection points in the standard feature point set are translated using the translation matrix T2, and the translated intersection coordinates and the translated standard intersection point K are stored in the standard translation point set;

[0049] 3) Taking the standard intersection point K as the rotation center, use the rotation matrix to rotate the measured translation point set and the measured normal vector P, and use the optimization method to find the corresponding rotation matrix R' when the difference between the measured translation point set and the standard translation point set, the measured normal vector P and the standard normal vector Q is minimized;

[0050] 4) Then translate the standard intersection point K from the origin of the coordinate system to the coordinate position of the original measured intersection point H to obtain the translation matrix T3;

[0051] Offset matrix = T3 x R' x T2.

[0052] Preferably, the process of obtaining the pre-stored standard feature point set and the standard normal vector Q of the weldment is as follows:

[0053] Before teaching the robot milling trajectory, select a welded part from the batch of welded parts, record it as the teaching welded part, and place it at the specified position;

[0054] Based on the teaching weldment, teaching the robot a milling trajectory so that it can drive the milling cutter to accurately mill the weld on the teaching weldment;

[0055] Performing steps ① to ③ on the teaching weldment; storing each intersection point in a standard feature point set;

[0056] Based on the surface point cloud of the teaching weldment, the plane normal vector corresponding to the parent material I / parent material II of the teaching weldment is obtained and recorded as the standard normal vector Q.

[0057] Preferably, in step ①, the method of sequentially segmenting the local point clouds corresponding to the edge line I and the edge line II from the point cloud is as follows:

[0058] Select one or more edge lines from the base material I and the base material II of the teaching weldment, respectively, and mark them as edge line I and edge line II respectively;

[0059] In the surface point cloud of the teaching weldment, manually select the local point cloud corresponding to each edge line I and each edge line II, and record the position and size of each selection box;

[0060] In step ①, based on the recorded positions and sizes of each selection box, the local point cloud corresponding to each edge line I and each edge line II is selected from the point cloud.

[0061] This method solves the problem of relying on robots to mill welded parts and over-relying on the repeatability and manufacturing accuracy of the workpieces. The edge features of the workpiece base material are located through point cloud data, and the edge features are used to represent the state of the two overlapping base materials. By obtaining the coordinates of the intersection, the workpiece feature changes of the current welded part (current state) and the teaching welded part (teaching state) are calculated based on the coordinates, and the robot is guided for milling. This can greatly improve the accuracy of weld milling while ensuring efficiency and avoid excessive milling of the base material. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A schematic diagram for selecting edge lines on a welded part;

[0063] Figure 2 It is a schematic diagram of the effect of a single local point cloud and the projection points on the cross section;

[0064] Figure 3 It is a curve effect diagram fitted by using a two-dimensional point set in step S3. DETAILED DESCRIPTION

[0065] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0066] A weld milling guidance method based on the edge features of the parent material, the weld milling process is: a milling cutter is installed at the end of the robot, batch welded parts reach the specified position in sequence, and at the specified position, the robot mills the weld on the welded parts according to the taught milling trajectory;

[0067] like Figure 1 As shown, the parent materials connected on both sides of the weld are respectively marked as parent material I and parent material II; the edge line where the weld is located is selected on parent material I / parent material II as edge line I, and at least two edge lines are selected on the other parent material and marked as edge lines II; edge line I and edge line II intersect or are not in the same plane;

[0068] In detail, the edge line can be understood to include the convex / concave contour line of the parent material surface and the edge line of the parent material itself.

[0069] Guide the robot to complete milling using:

[0070] ① Obtain the surface point cloud of the current weldment, and sequentially segment the local point clouds corresponding to the edge line I and each edge line II from the point cloud; that is, obtain M pieces of local point clouds, where M represents the total number of edge lines I and edge lines II;

[0071] ② Obtain the spatial straight line corresponding to each local point cloud respectively;

[0072] ③ According to the correspondence between the local point cloud and the edge line I and edge line II, the space line is marked as space line I / space line II;

[0073] Find all the intersection points between the spatial straight line I and the spatial straight line II; the intersection points include real intersection points and virtual intersection points; among which the virtual intersection points are: point I, point II, or the midpoint between point I and point II; point I is the point on the spatial straight line I that is closest to the spatial straight line II, and point II is the point on the spatial straight line II that is closest to the spatial straight line I;

[0074] As a preferred implementation, in order to prevent over-milling of the base material, the coordinate component values ​​of point I and point II in the direction perpendicular to base material I / base material II are compared, and the point with the larger coordinate component value is recorded as the virtual intersection point. ④ Store each intersection point in the measured feature point set;

[0075] Based on the surface point cloud of the current weldment, the plane normal vector corresponding to the base material I / base material II of the current weldment is obtained, which is recorded as the measured normal vector P; when the milling trajectory of the teaching robot is retrieved, the standard feature point set and the standard normal vector Q of the weldment stored in advance are used; the offset matrix is ​​obtained by using the correspondence between the standard feature point set and each point in the measured feature point set and the correspondence between the standard normal vector Q and the measured normal vector P;

[0076] As a preferred implementation, in order to prevent over-milling of the base material, the point clouds corresponding to base material I and base material II can be compared, and the base material with a higher mean value of point cloud depth is recorded as the reference base material. The normal vector of the high surface is used to obtain the normal vector of the reference base material.

[0077] ⑤The robot corrects the milling trajectory according to the offset matrix, and mills the weld on the current weldment based on the corrected trajectory.

[0078] In detail, in step ②, the method of obtaining the spatial straight line corresponding to each local point cloud is as follows:

[0079] For a single local point cloud, the main direction of the local point cloud is obtained, and then a spatial straight line is constructed based on the main direction and the three-dimensional point S, where the three-dimensional point S is one of the projection points of the local point cloud after projection along the main direction, or is a three-dimensional point in the local point cloud.

[0080] Among them, the method of obtaining the main direction of the local point cloud is:

[0081] Using the principal component analysis method (PCA), or obtaining the cross section of the current local point cloud, the normal vector n of the cross section is recorded as the principal direction.

[0082] Specifically, the method for obtaining the cross section of the current local point cloud is as follows:

[0083] Preset the initial normal vector and construct the initial plane;

[0084] like Figure 2 , after projecting each 3D point in the local point cloud onto the plane, the area occupied by the projection points is counted on the current plane;

[0085] Iteratively update the normal vector, construct a plane based on the updated normal vector, and recalculate the area of ​​the projection point;

[0086] When the area value is less than the threshold or the area value tends to be stable, the iteration is stopped, and the current normal vector is recorded as the normal vector n and the current plane is recorded as the cross section.

[0087] In specific implementation, the normal vector is iteratively updated based on the optimization method, and the objective function can be set as:

[0088]

[0089] Among them, p 1 、p 2 ……p n are the coordinates of n three-dimensional points in the local point cloud, p is the coordinate of any point on the plane, and n represents the normal vector that needs to be iterated.

[0090] The method for counting the area occupied by the projection point is the grid method. Specifically, define the side length a, fill the current plane with grids, and each grid is a square with a side length of a; count the total number of grids occupied by the projection point, and then multiply the total number by the area of ​​a single grid to get the area occupied by the projection point.

[0091] More specifically, the 3D point S is one of the projection points of the local point cloud after being projected along the main direction. The 3D point S is obtained as follows:

[0092] Step S1, record any coordinate axis (X axis, Y axis or Z axis) of the camera coordinate system as A axis;

[0093] Calculate the angle θ between the normal vector n and the A axis, and then use the cross product of the normal vector n and the A axis to get the vector C. Use the vector C as the rotation axis to rotate the normal vector n by the angle θ to be parallel to the A axis. Use the Rodrigues transformation to get the rotation matrix R.

[0094] Step S2, using the rotation matrix R to rotate each projection point in the cross section to obtain the rotated projection point;

[0095] Extract the other two coordinates except the A-axis coordinate from the coordinates of each projection point after rotation, and store them in a two-dimensional point set;

[0096] Step S3: Fitting a curve using a two-dimensional point set (e.g. Figure 3 ), select a feature point on the curve, multiply the coordinates of the projection point after rotation corresponding to the feature point by the inverse matrix of the rotation matrix R to obtain the projection point before rotation, and record the projection point before rotation as the three-dimensional point S;

[0097] Among them, the characteristic point can be the point on the curve with the largest curvature change, the point with a slope of 0, the point with the largest ordinate value, or a preset distance point. The ordinate is a coordinate component perpendicular to the direction of the parent material I / parent material II. The preset distance point is: draw a tangent at the point with a slope of 0, translate the tangent by a preset distance to intersect with the curve, and record the intersection as the preset distance point.

[0098] In this scheme, the three-dimensional point S is a point with stability and consistency in the point cloud features of the edge line. In specific implementation, a three-dimensional point can also be directly extracted from the local point cloud. The specific steps are as follows: denoising the local point cloud, and then calculating the mean of the X-axis coordinates of all three-dimensional points in the local point cloud The mean of the Y-axis coordinates The mean Z-axis coordinate In the local point cloud, The point closest to the point is recorded as three-dimensional point S.

[0099] In specific implementation, in step ①, the method of sequentially segmenting the local point clouds corresponding to the edge line I and the edge line II from the point cloud is as follows:

[0100] Select one or more edge lines from the base material I and base material II of the teaching weldment, respectively, and mark them as edge line I and edge line II respectively;

[0101] In the surface point cloud of the teaching weldment, manually select the local point cloud corresponding to each edge line I and each edge line II, and record the position and size of each selection box;

[0102] In step ①, based on the recorded positions and sizes of each selection box, the local point cloud corresponding to each edge line I and each edge line II is selected from the point cloud.

[0103] Furthermore, in step ③, the following two methods are provided in this embodiment to obtain the virtual intersection points between each space straight line I and each space straight line II:

[0104] Method 1:

[0105] The direction vector b1 of the space line I is cross-producted with the direction vector b2 of the space line II to obtain the vector b3;

[0106] Use vectors b1 and b3 to construct the first plane, use b2 and b3 to construct the second plane, and solve the straight line where the first plane and the second plane intersect, which is recorded as space line III;

[0107] Calculate the intersection point between space line I and space line III, recorded as point I;

[0108] Calculate the intersection point between space line II and space line III, and record it as point II;

[0109] Point I, point II, or the midpoint between point I and point II is recorded as a virtual intersection point.

[0110] Method 2:

[0111] The direction vector b1 of the space line I is cross-producted with the direction vector b2 of the space line II to obtain the vector b3;

[0112] One of the spatial straight lines is translated along the direction of vector b3 until it is coplanar with the other spatial straight line, and the intersection point between the two spatial straight lines at this time is calculated and recorded as the virtual intersection point.

[0113] In step ④, the method for obtaining the offset matrix is:

[0114] Any intersection point in the measured feature point set is recorded as the measured intersection point H, and the intersection point corresponding to the position of point H is found in the standard feature point set and recorded as the standard intersection point K;

[0115] The measured intersection point H is translated to the origin of the camera coordinate system, and the translation matrix T1 is obtained. The coordinates of other intersection points in the measured feature point set are translated using the translation matrix TI, and the translated intersection coordinates and the translated measured intersection point H are stored in the measured translation point set;

[0116] The standard intersection point K is translated to the origin of the camera coordinate system to obtain the translation matrix T2; the coordinates of other intersection points in the standard feature point set are translated using the translation matrix T2, and the translated intersection coordinates and the translated standard intersection point K are stored in the standard translation point set;

[0117] 3) Taking the standard intersection point K as the rotation center, use the rotation matrix to rotate the measured translation point set and the measured normal vector P, and use the optimization method to find the corresponding rotation matrix R' when the difference between the measured translation point set and the standard translation point set, the measured normal vector P and the standard normal vector Q is minimized;

[0118] 4) Then translate the standard intersection point K from the origin of the coordinate system to the coordinate position of the original measured intersection point H to obtain the translation matrix T3;

[0119] Offset matrix = T3 x R' x T2.

[0120] More specifically, in step 3), an objective function can be constructed, and an optimization method can be used to obtain the corresponding rotation matrix R' when the difference between the measured translation point set and the standard translation point set, the measured normal vector P and the standard normal vector Q is minimized:

[0121] For example, two objective function expressions are given: see objective function 1 and objective function 2:

[0122] Among them, the objective function is:

[0123] Among them, α and β are preset weight coefficients, N represents the number of intersections, N ≥ 2, a i represents the coordinates of the i-th intersection point in the measured translation point set, b i represents the coordinates of the i-th intersection point in the standard translation point set;

[0124] The optimization method is used to obtain the value of the Euler angle θ(rx, ry, rz) corresponding to the minimum residual between the measured translation point set and the standard translation point set, the measured normal vector P and the standard normal vector Q; and then the rotation matrix R' in the offset matrix is ​​obtained based on the value of the Euler angle θ(rx, ry, rz).

[0125] Objective function 2:

[0126] Among them, α and β are preset weight coefficients, N represents the number of intersections, N ≥ 2, a i represents the coordinates of the i-th intersection point in the measured translation point set, b i represents the coordinates of the i-th intersection point in the standard translation point set;

[0127] The optimization method is used to obtain the value of the Euler angle θ(rx, ry, rz) corresponding to the minimum residual between the measured translation point set and the standard translation point set, and the minimum angle between the measured normal vector P and the standard normal vector Q; then, the rotation matrix R' in the offset matrix is ​​obtained based on the value of the Euler angle θ(rx, ry, rz).

[0128] In the specific implementation, the following process is performed in advance to store the standard feature point set and the standard normal vector Q of the weldment:

[0129] Before teaching the robot milling trajectory, select a welded part from the batch of welded parts, record it as the teaching welded part, and place it at the specified position;

[0130] Based on the teaching weldment, teach the robot milling trajectory so that it can drive the milling cutter to accurately mill the weld on the teaching weldment;

[0131] Perform steps ① to ③ on the teaching weldment; store each intersection point into the standard feature point set;

[0132] Based on the surface point cloud of the teaching weldment, the plane normal vector corresponding to the parent material I / parent material II of the teaching weldment is obtained and recorded as the standard normal vector Q.

[0133] This solution uses point cloud data to locate the edge features of the workpiece base material, and uses the edge features to represent the state of the two overlapping base materials. By obtaining the coordinates of the intersection, the workpiece feature changes of the current welded part (current state) and the taught welded part (teaching state) are calculated based on the coordinates, and the robot is guided for milling. This can greatly improve the accuracy of weld milling while ensuring efficiency and avoid excessive milling of the base material.

[0134] The foregoing description of the specific exemplary embodiments of the present invention is presented for the purpose of illustration and description. The foregoing description is not intended to be exhaustive, nor is it intended to limit the present invention to the precise form disclosed, and it is apparent that many changes and variations are possible in light of the above teachings. The exemplary embodiments are selected and described to explain the specific principles of the present invention and its practical application, so that other technicians in the field can realize and utilize the various exemplary embodiments of the present invention and its different selected forms and modified forms. The scope of the present invention is intended to be limited by the appended claims and their equivalents.

Claims

1. A weld milling guidance method based on the edge features of the parent material, wherein the weld milling process is as follows: a milling cutter is installed at the end of the robot, batch welded parts reach the specified position in sequence, and the robot mills the weld on the welded parts according to the taught milling trajectory; Features: The parent materials connected on both sides of the weld are marked as parent material I and parent material II respectively; the edge line where the weld is located is selected on parent material I / parent material II as edge line I, and at least two edge lines are selected on the other parent material and marked as edge lines II respectively; Guide the robot to complete milling using: ① Obtain the surface point cloud of the current weldment, and sequentially segment the edge line I and the local point cloud corresponding to each edge line II from the point cloud; ② Obtain the spatial straight line corresponding to each local point cloud respectively; ③ According to the correspondence between the local point cloud and the edge line I and edge line II, the space line is marked as space line I / space line II; Find all the intersection points between space straight line I and space straight line II; ④Store each intersection point in the measured feature point set; Based on the surface point cloud of the current weldment, the plane normal vector corresponding to the base material I / base material II of the current weldment is obtained, which is recorded as the measured normal vector P; the standard feature point set and the standard normal vector Q of the weldment pre-stored in the milling trajectory of the teaching robot are retrieved; the offset matrix is ​​obtained by using the correspondence between the standard feature point set and each point in the measured feature point set and the correspondence between the standard normal vector Q and the measured normal vector P; ⑤ The robot corrects the milling trajectory according to the offset matrix, and mills the weld on the current weldment based on the corrected trajectory.

2. The weld milling guidance method based on the edge features of the parent material according to claim 1, characterized in that: In step ②, the method of obtaining the spatial straight line corresponding to each local point cloud is as follows: For a single local point cloud, the main direction of the local point cloud is obtained, and then a spatial straight line is constructed based on the main direction and the three-dimensional point S, where the three-dimensional point S is one of the projection points of the local point cloud after projection along the main direction, or is a three-dimensional point in the local point cloud.

3. The weld milling guidance method based on the edge features of the parent material according to claim 2, characterized in that: The way to obtain the main direction of the local point cloud is: The principal direction is obtained by using the principal component analysis method, or the cross section of the current local point cloud is obtained, and the normal vector n of the cross section is recorded as the principal direction.

4. The weld milling guidance method based on the edge feature of the parent material according to claim 3, characterized in that: The way to obtain the cross section of the current local point cloud is as follows: Preset the initial normal vector and construct the initial plane; After projecting each 3D point in the local point cloud onto the plane, the area occupied by the projected points is counted on the current plane; Iteratively update the normal vector, construct a plane based on the updated normal vector, and recalculate the area of ​​the projection point; When the area value is less than the threshold or the area value tends to be stable, the iteration is stopped, and the current normal vector is recorded as the normal vector n and the current plane is recorded as the cross section.

5. The weld milling guidance method based on the edge feature of the parent material according to claim 4, characterized in that: The 3D point S is one of the projection points of the local point cloud after being projected along the main direction. The 3D point S is obtained as follows: Step S1, record any coordinate axis of the camera coordinate system as axis A; Calculate the angle θ between the normal vector n and the A axis, then use the cross product of the normal vector n and the A axis to obtain the vector C, use the vector C as the rotation axis, rotate the normal vector n by the angle θ to be parallel to the A axis, and obtain the rotation matrix R through Rodrigues transformation; Step S2, using the rotation matrix R to rotate each projection point in the cross section to obtain the rotated projection point; Extract the other two coordinates except the A-axis coordinate from the coordinates of each projection point after rotation, and store them in a two-dimensional point set; Step S3, fitting a curve using a two-dimensional point set, selecting a feature point on the curve, multiplying the coordinates of the rotated projection point corresponding to the feature point by the inverse matrix of the rotation matrix R to obtain the projection point before rotation, and recording the projection point before rotation as the three-dimensional point S; Preferably, the characteristic point is a point on the curve with the largest curvature change, a point with a slope of 0, a point with a maximum ordinate value, or a preset distance point, wherein the ordinate is a coordinate component perpendicular to the direction of the parent material I / parent material II, and the preset distance point is: a tangent is drawn at a point with a slope of 0, the tangent is translated by a preset distance to intersect with the curve, and the intersection is recorded as the preset distance point.

6. The weld milling guidance method based on the edge feature of the parent material according to claim 1, characterized in that: The edge line I and the edge line II intersect or are not in the same plane; In step ③, the intersection points include real intersection points and virtual intersection points; Preferably, the virtual intersection is: point I, point II or the midpoint of point I and point II; point I is the point on spatial straight line I that is closest to spatial straight line II, and point II is the point on spatial straight line II that is closest to spatial straight line I.

7. The weld milling guidance method based on the edge feature of the parent material according to claim 6, characterized in that: There are two ways to obtain the virtual intersection point: Method 1: The direction vector b1 of the space line I is cross-producted with the direction vector b2 of the space line II to obtain the vector b3; Use vectors b1 and b3 to construct a first plane, use b2 and b3 to construct a second plane, and calculate the straight line where the first plane and the second plane intersect, which is recorded as space line III; Calculate the intersection point between space line I and space line III, recorded as point I; Calculate the intersection point between space line II and space line III, and record it as point II; Point I, point II, or the midpoint between point I and point II is recorded as a virtual intersection point; Method 2: The direction vector b1 of the space line I is cross-producted with the direction vector b2 of the space line II to obtain the vector b3; One of the spatial straight lines is translated along the direction of vector b3 until it is coplanar with the other spatial straight line, and the intersection point between the two spatial straight lines at this time is calculated and recorded as the virtual intersection point.

8. The weld milling guidance method based on the edge feature of the parent material according to claim 1, characterized in that: In step ④, the method for obtaining the offset matrix is: Any intersection point in the measured feature point set is recorded as the measured intersection point H, and the intersection point corresponding to the position of point H is found in the standard feature point set and recorded as the standard intersection point K; The measured intersection point H is translated to the origin of the camera coordinate system, and a translation matrix T1 is obtained. The coordinates of other intersection points in the measured feature point set are translated using the translation matrix TI, and the translated intersection coordinates and the translated measured intersection point H are stored in the measured translation point set; The standard intersection point K is translated to the origin of the camera coordinate system to obtain a translation matrix T2; the coordinates of other intersection points in the standard feature point set are translated using the translation matrix T2, and the translated intersection coordinates and the translated standard intersection point K are stored in the standard translation point set; 3) Taking the standard intersection point K as the rotation center, use the rotation matrix to rotate the measured translation point set and the measured normal vector P, and use the optimization method to find the corresponding rotation matrix R' when the difference between the measured translation point set and the standard translation point set, the measured normal vector P and the standard normal vector Q is minimized; 4) Then translate the standard intersection point K from the origin of the coordinate system to the coordinate position of the original measured intersection point H to obtain the translation matrix T3; Offset matrix = T3 x R' x T2.

9. The weld milling guidance method based on the edge feature of the parent material according to claim 1, characterized in that: The process of obtaining the pre-stored standard feature point set of the weldment and the standard normal vector Q is as follows: Before teaching the robot milling trajectory, select a welded part from the batch of welded parts, record it as the teaching welded part, and place it at the specified position; Based on the teaching weldment, teaching the robot a milling trajectory so that it can drive the milling cutter to accurately mill the weld on the teaching weldment; Performing steps ① to ③ on the teaching weldment; storing each intersection point in a standard feature point set; Based on the surface point cloud of the teaching weldment, the plane normal vector corresponding to the parent material I / parent material II of the teaching weldment is obtained and recorded as the standard normal vector Q.

10. The weld milling guidance method based on the edge feature of the parent material according to claim 1, characterized in that: In step ①, the method of sequentially segmenting the local point clouds corresponding to the edge line I and the edge line II from the point cloud is as follows: Select one or more edge lines from the base material I and the base material II of the teaching weldment, respectively, and mark them as edge line I and edge line II respectively; In the surface point cloud of the teaching weldment, manually select the local point cloud corresponding to each edge line I and each edge line II, and record the position and size of each selection box; In step ①, based on the recorded positions and sizes of each selection box, the local point cloud corresponding to each edge line I and each edge line II is selected from the point cloud.

Images