Self-adaptive welding method for sheet bending fillet weld

Abstract

The invention discloses a self-adaptive welding method for a sheet bending angle welding seam. The method comprises the steps that a 2D image and a 3D depth map of a bending angle position are obtained through scanning; the trained YOLOv5n-seg network is used to output a bending cavity probability mask at the bending angle position; processing the bending cavity probability mask by using a BLOB model to obtain a binary mask; based on a Zhang-Suen refinement algorithm, skeleton extraction is carried out on the binarized mask to obtain a weld joint center line pixel point set; performing B spline curve fitting on the center line of the welding seam; determining a corresponding ROI frame according to the points of the B-spline curve, extracting the boundary of the binarized mask, obtaining the 2D coordinates of boundary representative points of the boundaries on the two sides of the points of the B-spline curve in the ROI frame, and obtaining the 3D coordinates of the boundary representative points based on the corresponding relationship between the 2D image and the 3D depth map; and constructing a 3D welding line path according to the midpoint coordinates of the 3D coordinates of the boundary representative points, and welding according to the 3D welding line path. The device can adapt to welding of bent fillet weld joints with holes.

Description

Technical Field

[0001] The invention relates to a bending angle welding method, in particular to a self-adaptive welding method for a thin plate bending angle weld. Background Art

[0002] Existing automatic welding devices for thin plates mostly target straight welds, using teaching robots to obtain a relatively fixed welding path for welding. However, in actual industrial production, the specific implementation of automated welding of thin plates is still challenged by the geometric changes of some complex weld structures. For example, in the manufacturing process of stainless steel water dispenser water tank products, the thin plate materials must be shaped by a bending machine, which leads to a relatively complex weld structure at each bending angle. Due to the characteristics of the material and the bending process, the welds in the bending angle area will form hollow gaps of different shapes, resulting in significant uncertainty in the spacing between the parent materials on both sides. This diversity of geometric structures makes it difficult for traditional teaching robots to adapt to the fixed welding path mode. Summary of the Invention

[0003] In view of the above-mentioned defects of the prior art, the object of the present invention is to provide a method for adaptive welding of thin plate bent corner welds to achieve adaptive welding of different shapes of bent corner areas.

[0004] The technical solution of the present invention is as follows: A method for adaptive welding of thin plate bent fillet welds, comprising the steps of:

[0005] Scan to obtain 2D images and 3D depth maps of the bending angle position;

[0006] The trained YOLOv5n-seg network outputs the bending hole probability mask of the bending angle position;

[0007] Processing the bending hole probability mask with a BLOB model to obtain a binary mask;

[0008] Performing skeleton extraction on the binary mask based on the Zhang-Suen thinning algorithm to obtain a weld centerline pixel point set;

[0009] Perform B-spline curve fitting on the weld centerline;

[0010] Determine the corresponding ROI frame using the points of the B-spline curve, extract the boundary of the binary mask, obtain the 2D coordinates of the boundary representative points on both sides of the boundary of the points of the B-spline curve within the ROI frame, and obtain the 3D coordinates of the boundary representative points based on the correspondence between the 2D image and the 3D depth map;

[0011] A 3D weld path is constructed using the midpoint coordinates of the 3D coordinates of the boundary representative points, and welding is performed according to the 3D weld path.

[0012] Furthermore, since the results of the bending angle forming process before welding are difficult to maintain consistency, the size of the void gap generated by the bending angle has a large deviation, and an excessively large void gap may make automatic welding impossible. Even if automatic welding can be implemented, it is difficult to ensure that bending angles with different void gap sizes can obtain better welding quality through the same welding process parameters. In order to solve this problem, the adaptive welding method for thin plate bending angle welds also includes calculating the maximum gap between the boundaries of the points on both sides of the B-spline curve based on the 3D coordinates of the boundary representative points, setting several groups of welding process parameters based on the gap value segments, determining the welding process parameters based on the gap value segments where the maximum gap is located, and welding with the determined welding process parameters.

[0013] Furthermore, the welding process parameters include pre-gas supply time, arc starting current, base current, duty cycle, constant current, arc end decay time, arc end current, post-weld gas supply time, and welding speed.

[0014] Furthermore, the gap value segment includes a segment with a gap greater than 3 mm, and welding is abandoned when the maximum gap is greater than 3 mm.

[0015] Furthermore, if there is a sudden change in the robot welding trajectory during actual welding, a large instantaneous speed and acceleration will be introduced, which may easily cause the robot to produce an overspeed shock. In order to solve this problem, the adaptive welding method for thin plate bending fillet welds also includes calculating the robot welding trajectory T={T i |i=1...K},

[0016]

[0017] in is the weld direction vector, is the weld normal vector, V i is the 3D weld path point, K is the number of 3D weld path points in the 3D weld path,

[0018] Define T i The rotation matrix is And converted into Euler angle sequence Ω={ω i |i=1...K}, the rotation speed ω′ in Euler angle i and acceleration ω″ i When any value in is greater than the preset threshold, ω i Construct a smoothing interval window with a filter size of 2L+1 for the center, and calculate the smoothed Euler angle value

[0019]

[0020] Then ω iConvert back to the rotation matrix R i And update T i A final robot welding trajectory is obtained, and welding is performed using the final robot welding trajectory.

[0021] Furthermore, in order to extract an area closer to the true contour of the bending corner hole and optimize the continuity and smoothness of edge extraction, the bending hole probability mask is processed by the BLOB model to obtain a binary mask, which includes first performing softness processing on the bending hole probability mask and then binarizing it using the BLOB model to obtain a binary mask. The function of the softness processing is expressed as

[0022]

[0023] Where f(I) is the output pixel value, t is the input pixel value, L is the set low threshold, H is the set high threshold, W = (HL) / N, and N is the set number of steps.

[0024] Furthermore, the value of L is 80, the value of H is 190, and the value of N is 5.

[0025] Furthermore, performing B-spline curve fitting on the weld centerline includes:

[0026] The B-spline curve can be expressed as C(t)=(x(t),y(t)) is the curve point under parameter t, c n =(x n ,y n ) is the control point under the B-spline, B n,d(t) is the B-spline basis function corresponding to the n-th control point, and d is the B-spline order;

[0027] Substitute the weld centerline pixel point set as the control point Get the B-spline description of the weld centerline.

[0028] Furthermore, the step of obtaining the 2D coordinates of the boundary representative points of the points on both sides of the B-spline curve within the ROI frame includes: assuming that the point set of the boundaries of the points on both sides of the B-spline curve within the ROI frame is and Then take and is the representative point of the boundary, and <> is a rounding operation.

[0029] Furthermore, after the B-spline curve is fitted to the weld centerline, the obtained B-spline curve is downsampled with equal intervals to avoid data redundancy and resource waste.

[0030] Compared with the prior art, the advantages of the technical solution provided by the present invention are:

[0031] By manipulating the geometry of the bend, the invention enhances the adaptability of the welding method in the actual welding process, ultimately achieving a welding success rate of 92.86%. This inventive method can effectively handle bends of varying shapes, exhibiting excellent process robustness and widespread application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 Schematic diagram of the process of the adaptive welding method for thin plate bending fillet welds according to an embodiment.

[0033] Figure 2 Schematic diagram of robot laser vision parametric scanning of a rectangular box.

[0034] Figure 3 The segmentation map of the 2D image, the bending hole probability mask, and the superposition effect of the bending hole probability mask and the brightness map.

[0035] Figure 4 The binary mask result map output by BLOB.

[0036] Figure 5 This is the weld centerline result image extracted from the binary mask.

[0037] Figure 6 Schematic diagram of boundary extraction of binary mask.

[0038] Figure 7 A schematic diagram of the ROI region is generated by using a B-spline curve with 5 points obtained by downsampling.

[0039] Figure 8 A schematic diagram of the ROI region is generated by using a B-spline curve with 10 points obtained by downsampling.

[0040] Figure 9 Schematic diagram of extracting boundary points based on ROI.

[0041] Figure 10 Schematic diagram for obtaining the posture of weld trajectory points. DETAILED DESCRIPTION

[0042] The present invention will be further described below with reference to the following examples. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading this description, various equivalent modifications to this description by those skilled in the art fall within the scope defined by the claims appended to this application.

[0043] Please combine Figure 1 As shown, a method for adaptive welding of a thin plate bent fillet weld in this embodiment includes:

[0044] 1. Scan to obtain 2D images and 3D depth maps of the bending angle position.

[0045] Taking the common rectangular box welding as an example, the size change is the length, width and height of the rectangular box. The robot laser vision parametric scanning method is used to scan and obtain the 2D image and 3D depth map of the bending angle position. Figure 2 As shown in the figure, after the rectangular box is loaded, the tooling will ensure that one corner point on its bottom side coincides with the origin of the preset welding robot working Base coordinate system {B}. The four bent fillet welds on its upper side are numbered in counterclockwise order as shown in the figure, and the L, W, and H parameters represent the length, width, and height specifications of the corresponding rectangular box respectively.

[0046] When H=0, the robot reference scanning trajectory will be set by teaching, and the starting position Ts of the reference scanning trajectory will be r and the end position Te r composition:

[0047] Ts r ={Xs r ,Ys r ,Zs r ,A r ,B r ,C r}, Te r ={Xe r ,Ye r ,Ze r ,A r ,B r ,C r},

[0048] The corresponding robot parametric scanning trajectory starting position Ts i With Te i They are:

[0049] Ts i ={Xs r +α i ·L,Ys r +β i ·W,Zs r +H,A r ,B r ,C r},

[0050] Te r ={Xe r +α i ·L,Ye r +β r ·W,Ze r +H,A r ,B r ,C r},

[0051] where the parameters α are taken i ={0,1,1,0|i=1,2,3,4} and β i ={0,0,1,1|i=1,2,3,4}, and finally the laser vision camera at the end of the robot scans along the trajectory to obtain the 2D images and 3D depth maps of the four bending corners on the top of the rectangular box.

[0052] 2. The trained YOLOv5n-seg network outputs the bending hole probability mask of the bending angle position.

[0053] The complete bend corner part is extracted based on the YOLOv5n-seg network. The network model uses the SGD optimization algorithm, sets the weight decay to 0.0001, and the initial learning rate to 0.01. The dataset used includes a total of 430 weld seam images, of which 380 are used as training sets and the remaining 50 are used as validation sets for the model. The number of training rounds is set to 30,000 rounds, and the training weight is saved every 1,000 rounds. At the same time, the training batch size is set to 4. The grayscale value of the bend hole probability mask represents the model's recognition that the pixel at this point belongs to the target category. The recognition results are as follows: Figure 3 As shown in the figure, from left to right are the segmentation map of the 2D image, the bending hole probability mask, and the superposition effect of the bending hole probability mask and the brightness map.

[0054] 3. Use the BLOB model to process the bending hole probability mask to obtain a binary mask.

[0055] Since the initial segmentation result not only includes the target void gap, but also retains some pixels in the base material area, resulting in discontinuity and blurring of the boundary. In order to further extract the area closer to the true contour of the bend corner void and optimize the continuity and smoothness of edge extraction, the bend void probability mask is first softened and then binarized using the BLOB model to obtain a binary mask. The function of softness processing is expressed as

[0056]

[0057] Where f(I) is the output pixel value, t is the input pixel value, L is the set low threshold, H is the set high threshold, W = (HL) / N, N is the set number of steps. Where L is 80, H is 190, and N is 5. The output result of the binary mask is as follows Figure 4 shown.

[0058] 4. Based on the Zhang-Suen thinning algorithm, the skeleton of the binary mask is extracted to obtain the weld centerline pixel point set {s(x i ,y i )|i=1…k}, the result is as follows Figure 5 As shown in the figure, from left to right are the original image, the binary mask image, and the weld centerline extraction result image.

[0059] 5. Perform B-spline curve fitting on the weld centerline.

[0060] The B-spline curve can be expressed as C(t)=(x(t),y(t)) is the curve point under parameter t, c n =(x n ,y n ) is the control point under the B-spline, B n,d(t) is the B-spline basis function corresponding to the n-th control point, and d is the B-spline order.

[0061] The weld centerline pixel point set {s(x i ,y i )|i=1…k} as the control point The weld centerline B-spline description weld centerline SC(t), that is, the weld centerline B-spline description, is obtained.

[0062] 6. Determine the corresponding ROI frame using the points of the B-spline curve, extract the boundary of the binary mask, obtain the 2D coordinates of the boundary representative points on both sides of the B-spline curve point within the ROI frame, and obtain the 3D coordinates of the boundary representative points based on the correspondence between the 2D image and the 3D depth map. Specifically:

[0063] The obtained B-spline curve is downsampled with equal spacing.

[0064] Perform edge traversal on the binary mask to obtain the ordered image boundary E, such as Figure 6 shown.

[0065] For point sC on the center line of the weld i , a fixed region of interest (ROI) is generated at this point, such as Figure 7 、 Figure 8 As shown, the length and width of the ROI are fixed values, and the tangent direction at this point determines the rotation angle α of the ROI. Through this ROI frame, the image boundary E can be intersected and the ordered boundary point set in the ROI area on both sides can be obtained. and And further take and is a representative 2D boundary intersection point and Where <> is a rounding operation, which avoids the situation where the 2D boundary point corresponds to an invalid 3D boundary point. Finally, the 3D coordinates of the two boundary points can be obtained by using the corresponding mapping relationship between the 2D image pixels and the 3D depth map and point cloud in the 3D camera. and like Figure 9 As shown. Finally, the 3D point set of the weld double-side boundary is obtained.

[0066] 7. Construct a 3D weld path using the midpoint coordinates of the 3D coordinates of the boundary representative point and perform welding according to the 3D weld path, specifically including:

[0067] The robot weld path point is calculated as the middle point of the 3D points on both sides of the weld.

[0068]

[0069] Get the 3D weld path point set:

[0070]

[0071] It is further necessary to calculate the robot welding posture, such as Figure 10 As shown, point V i , V i+1 are two consecutive points on the weld path. and is the boundary point corresponding to the feature point.

[0072] Calculate the welding direction vector is the unit vector of the weld direction:

[0073]

[0074] Among them, → is the calculation method of the vector between two points in space, || is the Euclidean distance value of the space vector, and the pseudo lateral vector is further calculated. for:

[0075]

[0076] According to the right-hand rule, the weld normal vector It can be calculated as:

[0077]

[0078] Further follow the weld direction and normal To set the process constraints, the welding lateral vector for:

[0079]

[0080] By setting the welding gun end TCP for the robot, each weld path point V i The robot welding posture at is:

[0081]

[0082] By traversing the entire welding path, the robot welding trajectory can be obtained as follows:

[0083] T={T i |i=1...K}.

[0084] During actual welding, if the robot's welding trajectory experiences sudden changes in value, it will introduce large instantaneous velocities and accelerations, which can easily cause the robot to experience overspeed shocks. Therefore, the posture trajectory can be smoothed to improve motion continuity and stability. The specific method is to convert the welding gun posture into Euler angles and use a moving average smoother to process the rotation trajectory.

[0085] The robot used in this embodiment adopts the Euler angle form of ZYX, defining T i The rotation matrix in is Then for any rotation matrix we can set:

[0086]

[0087] Assuming the Euler angles are A, B, and C, the conversion relationship between them is:

[0088]

[0089] The planned pose sequence is: R i ={R i |i=1...K}.

[0090] Then go through the following steps to obtain the final robot weld posture trajectory:

[0091] Step 1: Based on the above transformation relationship, R i ={R i |i=1...K} is converted into the Euler angle sequence Ω={ω i |i=1...K}, where ω i =[A i B i C i ];

[0092] Step 2: Calculate the Euler angle rotation speed ω respectively i ′ and acceleration ω i ", if all components of are less than the preset threshold H ω1 and H ω2 , then go to step 4, otherwise go to step 3;

[0093] Step 3: Take ω i Construct a smoothing interval window with a filter size of 2L+1 for the center, and calculate the smoothed Euler angle value as follows: Go to step 2;

[0094] Step 4: Add ω i Convert to R i , and update T i The rotation part in the robot obtains the new welding posture TF i , and the final welding pose sequence is obtained as TF = {TF i |i=1...K}.

[0095] Because it's difficult to consistently control the results of the pre-process for bent fillet welds, robotic welding is often unfeasible for welds with large gaps exceeding a certain threshold. Furthermore, even if the gap is controlled within the process's feasible range, process adaptability for different gaps still requires adaptation. Therefore, the present invention conducts test welds for welds with the appropriate materials and working conditions, then determines the appropriate process parameters based on the calculated gap value.

[0096] First calculate the gap of each weld as

[0097]

[0098] Among them, max(g) is the maximum value in all sequences, || is the Euclidean distance between two point vectors in space, and the unit is mm.

[0099]

[0100] Among them, JOB is the welding process parameter number adapted for the welding robot, which includes relevant parameters such as pre-gas supply time, arc starting current, base current, duty cycle, constant current, arc decay time, arc ending current, post-gas supply time, and welding speed. This embodiment sets the corresponding welding process parameter sets under three different working conditions through experiments. When the gap is greater than 3mm, it is determined that the gap is too large and it is an abnormal weld, and automatic welding is abandoned.

[0101] This embodiment verified and summarized the parameter combinations under different path types through multiple manual welding tests and teaching welding experiments, and finally determined the parameter configuration strategy, which is detailed in the table below.

[0102]

[0103]

[0104] The effectiveness of the method was verified by using a subjective evaluation method, with welding quality as the core evaluation criterion. This method primarily relies on the experience of welding experts to judge and evaluate welding results. The method was used to weld three rectangular water tanks of different sizes. Excluding factors such as abnormalities in the intermediate welding process equipment and abnormalities in the preceding preparation, a total of 98 sets of valid data were obtained. Of these, 91 sets were rated as successful by welding experts, for a welding success rate of 92.86%.

[0105] From the overall evaluation, the present invention can effectively complete the forming quality of the welded joints, the weld surface is flat and the filling is uniform, no obvious typical welding defects such as pores, cracks or slag inclusions are found, and the weld metal is well bonded to the base material.

Claims

1. A method for adaptive welding of thin plate bending fillet welds, characterized in that: Including steps: Scan to obtain 2D images and 3D depth maps of the bending angle position; The trained YOLOv5n-seg network outputs the bending hole probability mask of the bending angle position; Processing the bending hole probability mask with a BLOB model to obtain a binary mask; Performing skeleton extraction on the binary mask based on the Zhang-Suen thinning algorithm to obtain a weld centerline pixel point set; Perform B-spline curve fitting on the weld centerline; Determine the corresponding ROI frame using the points of the B-spline curve, extract the boundary of the binary mask, obtain the 2D coordinates of the boundary representative points on both sides of the boundary of the points of the B-spline curve within the ROI frame, and obtain the 3D coordinates of the boundary representative points based on the correspondence between the 2D image and the 3D depth map; A 3D weld path is constructed using the midpoint coordinates of the 3D coordinates of the boundary representative points, and welding is performed according to the 3D weld path.

2. The adaptive welding method for thin plate bending fillet weld according to claim 1, characterized in that: The method includes calculating the maximum gap between the boundaries of the points on both sides of the B-spline curve based on the 3D coordinates of the boundary representative points, setting several groups of welding process parameters based on the gap value segments, determining the welding process parameters based on the gap value segments where the maximum gap is located, and performing welding with the determined welding process parameters.

3. The adaptive welding method for thin plate bending fillet weld according to claim 2, characterized in that: The welding process parameters include pre-gas supply time, arc starting current, base current, duty cycle, constant current, arc end decay time, arc end current, post-weld gas supply time, and welding speed.

4. The adaptive welding method for thin plate bending fillet weld according to claim 2, characterized in that: The gap value segmentation includes a segmentation with a gap greater than 3 mm, and welding is abandoned when the maximum gap is greater than 3 mm.

5. The self-adaptive welding method for thin plate bending fillet weld according to claim 1, characterized in that: Including calculating the robot welding trajectory T={T based on the 3D weld path i |i=1...K}, in is the weld direction vector, is the weld normal vector, V i is the 3D weld path point, K is the number of 3D weld path points in the 3D weld path, Define T i The rotation matrix is And converted into Euler angle sequence Ω={ω i |i=1...K}, the rotation speed in Euler angle ω i ′ and acceleration ω″ i When any value in is greater than the preset threshold, ω i Construct a smoothing interval window with a filter size of 2L+1 for the center, and calculate the smoothed Euler angle value Then ω i Convert back to the rotation matrix R i And update T i A final robot welding trajectory is obtained, and welding is performed using the final robot welding trajectory.

6. The self-adaptive welding method for thin plate bending fillet welds according to claim 1, 2 or 5, characterized in that: The method of processing the bending hole probability mask with the BLOB model to obtain a binary mask includes first performing softness processing on the bending hole probability mask, and then binarizing the mask with the BLOB model to obtain a binary mask. The function of the softness processing is expressed as Where f(I) is the output pixel value, t is the input pixel value, L is the set low threshold, H is the set high threshold, W = (HL) / N, and N is the set number of steps.

7. The self-adaptive welding method for thin plate bending fillet welds according to claim 1, 2 or 5, characterized in that: The L value is 80, the H value is 190, and the N value is 5.

8. The self-adaptive welding method for thin plate bending fillet welds according to claim 1, 2 or 5, characterized in that: The B-spline curve fitting of the weld centerline comprises: The B-spline curve can be expressed as C(t)=(x(t),y(t)) is the curve point under parameter t, c n =(x n ,y n ) is the control point under the B-spline, B n,d(t) is the B-spline basis function corresponding to the n-th control point, and d is the B-spline order; Substitute the weld centerline pixel point set as the control point Get the B-spline description of the weld centerline.

9. The self-adaptive welding method for thin plate bending fillet welds according to claim 1, 2 or 5, characterized in that: The step of obtaining the 2D coordinates of the boundary representative points of the two sides of the point of the B-spline curve in the ROI frame includes: assuming that the point set of the two sides of the boundary of the point of the B-spline curve in the ROI frame is and Then take and is the representative point of the boundary, and <> is a rounding operation.

10. The self-adaptive welding method for thin plate bending fillet welds according to claim 1, 2 or 5, characterized in that: After the weld centerline is fitted with a B-spline curve, the obtained B-spline curve is downsampled with equal distance intervals.

Images