A method of welding path planning of steel mesh based on point cloud for welding robot

At present, the operators needs to carry out complicated teaching and programming work on the welding path planning for the welding robot before welding the steel mesh. In this work, an automatic welding path planning method of steel mesh based on point cloud is proposed to simplify the complicated teaching and programming work in welding path planning. The point cloud model of steel mesh is obtained by three-dimensional vision structured light camera. Then, we use the relevant point cloud processing algorithm to calculate the welding path of the steel mesh, and obtain the 3D information of the welding path for the welding localization of the robot welding process. Experimental results show that the method can accurately realize the welding path planning of the steel mesh and accomplish the welding task without teaching and programming before welding, which improves the production efficiency.


Introduction
With the rapid development of automation and robot technologies, the welding robots are widely applied into the welding environment to replace human work. The teachingplayback mode of welding robots still plays an important role in the current industrial production. However, the operator needs to carry out complicated teaching and programming work on the welding path of this mode of welding robot before welding. Meanwhile, this work has high requirements for the operator's operation level and the accuracy. To conquer the above problems, many researchers study on weld extraction and welding planning using different sensors for different workpieces.
In the application and research of welding robot, the mainly used sensor include infrared sensors [1], RGB-D sensors [2,3], and vision sensors [4,5]. The vision sensors of welding robots could be divided into two-dimensional (2D) vision sensors and three-dimensional (3D) vision sensors. The use of 2D vision sensor in welding process mostly needs to cooperate with laser sensor. For example, Wang et al. [6] proposed a method of combining laser and vision sensor to identify V-shaped welds of oil pipelines through image processing, which can be used for subsequent trajectory planning. Xu et al. [7] designed a set of real-time welding seam tracking system based on laser and vision sensor, through an improved Canny algorithm to detect the edges of seam and pool, which could better overcome the deficiencies of the welding seam tracking control of the teaching-playback mode during welding process.
Compared with the 2D vision sensor, the 3D vision sensor can obtain 3D coordinate information of the workpiece and accurate completely the welding task. At present, linear structured light vision sensors and stereo vision sensors are the commonly used 3D vision sensors in robot welding task. For the use of linear structured light vision sensors, Zeng et al. [8] proposed a narrow butt 3D off-line welding path planning method based on a laser structured light sensor. Hou et al. [9] proposed non-instructional welding method of robotic gas metal arc welding (GMAW) based on laser structured light vision sensing system (LVSS), and experiments on V-grooves and fillet welds were performed. The 3D structured light could obtain the global information of the welding environment. However, the laser structured light vision could only obtain the local information. Therefore, the linear structured light is mostly used for the online identification and tracking of the weld seam. It is not suitable for the off-line 3D path planning of the welding robot.
To overcome the deficiency of linear structured light sensor and realize accurate and efficient off-line 3D path planning, using the stereo structured light sensor to generate point cloud and using the point cloud processing method to process, it has become a new scheme to solve the path planning of welding robot without teaching and programming. Lei et al. [10] proposed a novel 3D path extraction method of weld seams based on point cloud, which could well serve for the 3D path teaching task before welding. Zhang et al. [11] proposed point cloud based approach to recognize working environment and locate welding initial position using laser stripe sensor.
Few researches hammer at automatic welding path planning method of welding robot under the condition of without teaching and programming. With the development of society and the infrastructure construction, structural parts, such as steel cage and steel mesh, are widely used. At the same time, the welding of steel mesh faces scenes with many crossing points, which results in the cumbersome teaching process. Therefore, using the point cloud processing method to plan the welding path of steel mesh is an important part of solving the complex teaching and programming problems of welding robot before welding. Secondly, at present, most of workpiece point cloud is obtained by scanning the entire workpiece with a structured light sensor. This process takes a long time and we need to calculate the coordinates of the point cloud. In response to this problem, this paper introduces a 3D surface structured light camera, which can quickly obtain the point cloud of the workpiece and obtain the three-dimensional coordinates of the point cloud at same time. This process takes a short time and greatly improves the welding efficiency.
In this paper, an automatic welding path planning method of steel mesh based on point cloud is proposed, which realizes the welding path planning of welding robot without teaching and programming. Section 2 introduces the configuration of the experiment system; Section 3 illustrates the steps of point cloud preprocessing; Section 4 illustrates the step of welding path planning; Section 5 shows about the analysis of the experimental results, and finally, the conclusion and prospect of this paper are described.

Experiment system
The robot welding system of the experimental platform is shown in Fig. 1. It consists of two parts: the welding execution system and the 3D vision system [12]. The welding execution system includes the welding torch, the wire feeder, the manipulator, the robot controller and the cross steel mesh, which is used to complete the welding of the intersections of steel mesh. The 3D vision system includes a 3D surface scanning structured light camera and an industrial personal computer (IPC), which is used to obtain the 3D information of the cross steel mesh in the camera field of vision.
As shown in Fig. 2, the manipulator used in the experimental platform is universal robots 5 (UR5), and the 3D surface scanning industrial camera is chishine surface 120. It should be noted that the installation position of 3D surface scanning structured light camera needs to ensure that the welding torch does not enter the field of vision of the camera. The characteristics of 3D surface scanning structured light camera are shown in Table 1.

Steel mesh model
In order to introduce the method of planning the welding path of steel mesh more clearly, the steel mesh model used in this paper is shown in Fig. 3a. For the convenience of robot welding, there is a gap at the intersection of the upper and lower steel bars of the steel mesh model. The upper steel bars are supported by two fixed-size support plates on the left and right sides, and the relevant dimensions of the model are marked in the top view of the steel mesh with the main optical axis of the camera as the main viewing direction. Figure 3b shows the size of the steel mesh from the top view.

System framework
During the process of welding path planning of steel mesh, first of all, we should move the manipulator to steel mesh intersection position and adjust the camera field of view to a reasonable range through the industrial robot controller. Simultaneously, we record the shooting points P i |P 1, P 2, ... , P n and generate the shooting path [13]. Then, we form a point cloud image of the steel mesh through 3D surface scanning structured light camera at shooting point. The image is transferred into the IPC through the MicroB. Finally, the IPC obtains the welding path of the steel mesh within the vision of the 3D camera through the relevant point cloud processing method and sends the welding path to the robot controller [14]. After, the robot completes the welding task of the current shooting point according to the robot controller instructions and then continues the welding task of the next shooting point [15]. The specific operation process is shown in Fig. 4.
It should be noted that, since the shooting range of the 3D surface scanning industrial camera is the coverage range of the structured light, if the angle between the main optical axis of the camera and the plane of the steel mesh is too large or too small, as shown in Fig. 5a, the taken point cloud image will be the side surface of the steel bar, which will affect the accuracy of subsequent welding path planning of the steel mesh. Therefore, as shown in Fig. 5b, when recording the shooting point position, it is necessary to adjust the shooting posture of the camera and record to ensure that the angle between the main optical axis and the steel mesh plane is reasonable. We conducted multiple experiments on the selection of the reasonable angle range of the main optical axis and the steel mesh plane. On the premise of ensuring the accuracy of the results of each experiment, we choose 90 • ± 5 • as the best angle range according to the results of the experiment. The red area in the Fig. 5 is the area that the camera can capture. The position of the shooting point and the posture of the camera can be determined through adjustment based on the number of display points of the steel mesh intersection in the camera's field of view.

Point cloud preprocessing
After the 3D surface structured light camera takes pictures of the steel mesh at the shooting point, it will form a point cloud of the steel mesh within the camera's field of view at the shooting point. It is the initial point cloud of the steel mesh without any processing, as shown in Fig. 6. Compared with the preprocessed steel mesh point cloud, it has the characteristics of complex background, many irrelevant features and high density of point cloud. Therefore, in order to obtain a high-quality point cloud of the steel mesh, it is necessary to preprocess the initial point cloud.

Point cloud filtering
The initial steel mesh point cloud contains all the features in the camera's field of view. In order to prevent the interference of irrelevant features on the welding path planning of the steel mesh and to reduce the number of points to increase the calculation speed, the initial steel The principle of a pass-through filter is to perform a simple filtering along a specified dimension, that is, cut off values that are either inside or outside a given user range.
We use the graphic display method to determine the filtering range. The 3D coordinate system of the point cloud Fig. 4 The specific operation process is based on the camera position with the coordinate origin (0, 0, 0). Therefore, we display the steel mesh point cloud in this coordinate system through the software, and determine the filtering range of irrelevant features point cloud according to the display area of the point cloud on the x-, y-, and z-axes.
The filtering of the initial point cloud in this method is mainly to remove the point cloud of the support platform. Therefore, there is no need to filter along the z-axis. According to the display of the steel mesh point cloud on the right of the Fig. 7, we can determine that the initial point cloud only needs to be filtered along the x-axis. The accepted interval values of x-axis are set to (−10, 140), and the rest are all removed.
The filtering point cloud of the steel mesh using a passthrough filter is shown in Fig. 8. The number of points in the point cloud is reduced from 176,843 to 110,777.
In the welding of steel mesh with many intersections, it is impossible to display all the intersections in the camera's field of view at the same time. Therefore, it is necessary to record the shooting point and camera pose for multipoint shooting. Since, the shooting distance and posture basically remain unchanged, this filtering parameter can be used in subsequent shooting, which is determined in the first shooting. If the steel mesh point cloud is shown as Fig. 8 without any support platform or other irrelevant features point cloud, the next step of point cloud plane segmentation can be operated directly without using point cloud filtering.

Background point cloud removal
After point cloud filtering, irrelevant features in the steel mesh point cloud have been removed, and only the steel mesh point cloud and the background point cloud are retained. At this time, the independent steel mesh point cloud can be obtained by removing the background point cloud through point cloud segmentation. In this experiment, the placement of the steel mesh is a plane. The background When using the point cloud segmentation algorithm in the point cloud library, the first step is to create an object in the program. The second step is to define the model type. In this experiment, the model type is defined as a plane model (SACMODEL PLANE) because the placement of the steel mesh is a plane. Plane model contains four parameters, shown in Table 2, which determine the plane ax + by + cz + d = 0. The plane is obtained by using the Random Sample Consensus (RANSAC) method (SAC RANSAC) as Finally, the plane model point cloud and outlier point cloud are classified by setting the distance threshold. All points with a distance less than the threshold are regarded as interior points, and others are regarded as outlier points. The retention of internal or outlier points can be achieved through the setting program.
The selection of the distance threshold uses the coordinate graphic display method. According to the point cloud along the z-axis in Fig. 9, it can be found that the thickness of the background plane point cloud is about 6 mm, so set the distance threshold to 6 and keep the outlier point cloud.  It should be noted that the background point cloud has the same thickness when shooting at each shooting point. Therefore, the distance threshold can be reused. The steel mesh point cloud after background point cloud removal is shown in Fig. 10.

Independent steel bar point cloud acquisition
In the process of welding path planning of steel mesh, it is necessary to carry out linear fitting of steel bar point cloud and other operations. Therefore, point clouds belonging to the same steel bar can be grouped into same a class to form an independent steel bar point cloud by point cloud clustering method, which can facilitate subsequent operation.
The method of steel bars point cloud clustering is similar to the point cloud plane segmentation. After creating the object, the first step is to define the model type as a linear model(SACMODEL LINE). The linear model contains six parameters, shown in Table 3, which jointly determine the straight line. The straight line is also obtained using the Table 2 The specific meaning of the parameters in plane model

Parameter
Meaning normal x The x coordinate of the plane's normal normal y The y coordinate of the plane's normal normal y The z coordinate of the plane's normal d The fourth Hessian component of the plane's equation According to the point cloud after the segmentation in the camera x-O-y coordinate system in Fig. 11, it can be seen that the diameter of the steel bar is about 10 mm. Compared with the actual size 9.8 mm, it can be determined that the maximum distance from a point in the same bar to the straight line fitted by the RANSAC method will not exceed 10 mm. Hence, the distance threshold is set to 10, to ensure that all points belonging to the same steel bar are regarded as interior points. Through program setting, the interior points of each line are reserved and stored separately to obtain four independent steel bar point clouds.
Clustered points cloud are shown in Fig. 12, where four colors represent four clustered steel bars.  The y coordinate of a line's direction line direction.z The z coordinate of a line's direction  It should be noted that the straight line fitting based on SVD method is suitable for independent steel bar point cloud [16]. However, the straight line fitting based on RANSAC method is suitable for clustering and segmentation of steel mesh point cloud. The main differences are listed as follows: 1) The RANCAC method principle is shown on the left side of Fig. 13, which determines a line by two points on the basis of all sample points, and obtain the line model based on interior points by setting the distance threshold. The algorithm is simple and the calculation speed is fast. However the straight line obtained by the fitting must pass through two of the sample points. It slightly reduces the fitting accuracy.
2) The SVD method principle is shown on the right side of Fig. 13, which determines the line based on sample points in the way of minimizing the distance between all sample points and the fitting line. The fitting process does not need pass through any sample point, leading to a high-accurate fitting results.
In order to ensure the rigor of this method, we used SVD and RANSAC methods to perform straight line fitting for the same set of points respectively, and use the root mean square error (rmse) as the comparison standard for the fitting results. The specific results are shown in Table 4.
The purpose of clustering and segmentation of steel mesh point cloud is to obtain independent steel bars point cloud. This process has low requirements on the accuracy of the straight line fitting method and high requirements on the fitting speed. Therefore, the straight line fitting based on   According to the principle of radius outlier removal in Fig. 16, calculate the number of other points within the radius d of each point. When the number of other points within the radius is less than the set number, the point will be removed.
After repeated debugging in this experiment, the search radius is set to 2.5 and the number of points is set to 12. The point cloud after filtering based on radius outlier removal is shown in Fig. 17.

The straight line fitting based on SVD method
The idea of fitting a space line based on SVD method is straightforward, that is, minimizing to minimize the distance from all sample points to the straight line. Firstly, we calculate the arithmetic average of all sample points coordinates (x, y, z) according to Eq. 1. The straight line must pass through the position of the arithmetic average of all sample points coordinates. The difference matrix A between the coordinates of each sample point and the arithmetic average of all sample points coordinates (x, y, z) is calculated by The singular value decomposition of matrix A is performed by U is an n*n orthogonal matrix. S is a square matrix composed of r singular values from large to small along the arranged diagonally, and r is the rank of matrix A. V is a 3*3 singular vector matrix arranged from large to small along the column direction. The direction of the obtained straight line is the same as the singular vector corresponding to the maximum singular value. Therefore, the first column of the V matrix is selected as the direction of the fitted straight line, and the singular vector matrix with three rows and one column is expressed as V d .
We can define a straight line by the known direction and one point. The coordinates (x l , y l , z l ) of all points on this straight line satisfy Eq. 4. t is the relation variable between the point coordinates on the straight line and the arithmetic average of all sample point coordinates.
The length of the fitting line can be determined according to Eq. 4. The direction of the steel bar is divided into two types: extending along the x-axis and extending along the yaxis. When the steel bar extending along the x-axis, firstly, we select the x coordinate of the outermost points on both ends of the steel bar point cloud extension direction as the x coordinate on the two endpoints of the fitting line. Then, we bring the x coordinate of each endpoint into the x coordinate expression in Eq. 4, and find the corresponding t of this endpoint, which are t 1 and t 2 [17]. Finally, we bring t 1 and t 2 into the expression of y coordinate and z coordinate in Eq. 4. At this point, we can determine the coordinates of the two endpoints and length of the fitting line. The relation variable t corresponding to all the points between the two endpoints of the line are all the numbers between t 1 and t 2 . The length of the fitting straight line of the steel bar point cloud extending along the y-axis is solved in the same way as extending along the x-axis.The straight line fitting based on SVD method is shown in Fig. 18.

Find the common vertical line
After obtaining the fitting straight line of the steel bar point cloud, we need to find the common vertical line of the fitting line. As shown in Fig. 19, the fitting straight lines of steel bar 1 and steel bar 3 are AB and CD. The common vertical line of AB and CD is P Q. M is the foot point of AB and N is the foot point of CD.
We can calculate the four points coordinates of A(x a ,  y a , z a ), B(x b , y b , z b ), C(x c , y c , z c ), and D(x d , y d , z d ) according to Eq. 4 [18]. The relationship between straight line AM and AB, CN and CD is By solving Eq. 5, it can be known that the coordinates of foot point M (x m , y m , z m ) and foot point N (x n , y n , z n ) are expressed as Substituting Eq. 6 into 7, we can obtain k 1 and k 2 according to Eq. 7 which denotes the inner product of two perpendicular vectors is 0. Therefore, we can obtain the coordinates of the foot point M and the foot point N .
Through the foot points M and N, we can determine a straight line, which is the common vertical line of the straight lines AB and CD. The coordinates of the points between MN on the common vertical line conform to Eq. 8,   Fig. 20.

Find the gap width of crossed steel bars
The welding path planning needs to be determined according to the gap width of crossed steel bars. When the gap width is less than 2 mm, the steel mesh welding adopts spot welding. When the gap width is greater than 2 mm, the steel mesh welding adopts arc welding. Therefore, before determining the welding path, we should calculate the gap width of the crossed steel bars on first.
After obtaining the coordinates of the two endpoints of the common vertical line MN, we can get the length of the common vertical line MN according to The length of the common vertical line MN refers to the distance between the upper surface of the lower steel bar and the upper surface of the upper steel bar. Therefore, we can get the gap width between the crossed bars by subtracting the diameter d up of the upper steel bar from the length d up of  (10) After calculating the gap width of the crossed steel bars, we began to plan the welding path. As shown in Fig. 21, point P is the midpoint of the gap between the upper and lower steel bars, point E 1 is obtained by offsetting point P along the −→ NC direction by the distance of the upper steel bar radius, and the length of is − − → P E 1 the radius d up 2 of the upper steel bar.
Point is located on the common vertical line MN, so the coordinate of point P conforms to Eq. 8. At this time, we could find l corresponding to point P according to Eq. 11, and then substituting it into Eq. 8 to find the coordinate of point p x p , y p , z p The coordinates of point E 1 (x E1 , y E1 , z E1 ) can be calculated according to Eq. 12. It denotes the coordinates of two parallel vectors are proportional to each other.
According to Eq. 12, the relationship between y E1 and x E1 , z E1 and x E1 can be summarized as ⎧ ⎨ ⎩ y E1 = (xE1−xp)×(yc−yn) Substituting Eq. 13 into the − − → P E 1 distance calculation Eq. 14, we can get the x coordinate x E1 of point E 1 . Fig. 21 The schematic diagram of welding path planning Substituting x E1 into Eq. 12, we can get the proportional relationship, and then acquire the coordinate (x E1 , y E1 , z E1 ) of point E 1 .
Point F 1 is obtained by offsetting point P along the − − → MB direction by the distance of the lower steel bar radius. The way of calculating the coordinates of point F 1 is the same as point E 1 . Similar to Eq. 12, we can get Substituting Eq. 15 into the − − → P F 1 distance calculation Eq. 16, we can get the x coordinate x F 1 of point F 1 . Then, substituting x F 1 into vector parallel formula, we can get the Point G 1 is obtained by offsetting point P along the −→ NC direction by the distance of the upper steel bar radius and then along the − − → MB direction by the distance of the lower steel bar radius. We can get Eq. 17 according to − −− → E 1 G 1 is parallel to − − → P F 1 and has the same length. Then, the coordinates (x G1 , y G1 , z G1 ) of point G 1 can be obtained by Eq. 17. Similarly, we can get the coordinates of point F 2 and point H 2

Welding path planning
The final step is to plan the welding path, which can be divided into the following two situations.
1) When the gap width D F N between crossed bars is less than 2mm, the steel mesh welding adopts spot welding, and the welding point of the welding torch is E 1 . 2) When the gap width D F N between crossed bars is greater than 2mm, the steel mesh welding adopts arc welding. The initial point of the welding path is H 1 . Then, the welding torch passes through point E 1 along the direction of − −− → H 1 G 1 in a straight line and finally reaches the endpoint G 1 of the welding path to accomplish the primary welding [19]. If the gap width D F N between crossed bars is too large, it is necessary to weld the welding path several times according to the actual situation. For example, after completing a welding task, the welding torch reaches point G 1 , then the welding torch passes through point E 1 along the direction of − −− → G 1 H 1 in a straight line, and finally reaches the endpoint H 1 and to complete the secondary welding. Actual welding times, can be set freely according to the gap width D F N . The positions of welding points required for welding path planning are shown in Fig. 22.

Experiments and results
After obtaining the welding path of the steel mesh, we verify the feasibility, efficiency, and accuracy of the welding path planning method of steel mesh based on point cloud through error analysis, method efficiency, and welding platform experiment results.

Method error analysis
The error analysis is to verify whether the welding path planning method of the steel mesh based on point cloud meets the welding accuracy requirements, and to validate its feasibility. Through the welding path planning, we determine the coordinates of the points required for the welding path planning. We use the relevant points to calculate the gap width of the crossed steel bars, then compare with the actual gap width of the crossed steel bars and calculate its error [20]. Finally, we determine the Fig. 22 The positions of welding points required for welding path planning Through the calculation of the point cloud taken by the 3D camera, the coordinates of the two endpoints of the gap between crossed steel bars at four welding positions are calculated as shown in Table 5.
According to the coordinates of the two endpoints of the gap, we get the gap width and error of the four welding positions. The results are shown in Table 6 [20]. Through the comparison between the calculated gap width of the crossed bars and the actual gap width, it can be found that the maximum error is 0.75 mm, the minimum error is 0.49 mm, and the average error is 0.635 mm. According to Table 1, the repeatability of the camera is ±0.5 mm [21]. Taking into account the shooting error of the 3D camera, it can be concluded that the error between the calculated gap width and the actual gap width of the crossed steel bars meets the accuracy requirements. Therefore, the welding path planning method of the steel mesh based on point cloud s feasible, which meets the welding accuracy requirements.

Method efficiency analysis
Running time is a key factor to reflect the method performance. Because there are large number of shooting points in  The whole process 2927 industrial field, there are certain requirements for the time of point cloud generation and welding path planning for each shooting. Through many experiments, we record the total time for the welding path planning at four welding positions, as shown in Table 7.
The process of welding path planning at four welding positions in this method takes about 3000 ms. Therefore, the efficiency of this method can fully adapt to the needs of industrial production. Table 8 summarizes the results of the welding point coordinates required for the four welding positions in the 3D camera field of view obtained by the welding path planning method based on point cloud.

Welding platform experiment results analysis
The welding robot needs to carry out reasonable and accurate hand-eye calibration experiments to realize the accurate positioning of the front end of the welding torch to the position of the welding point. The principle of hand-eye calibration in this experiment is shown in Fig. 23.  In the eye-in-hand calibration method, Eq. 18 is applicable to any two postures of the robot in the process of moving.
According to Eq. 18, the external matrix End TCamera with the smallest error is selected as follows after multiple calibrations [20].
We record the shooting posture (−272.08, 647.67, 419.28, −174.76, 3.37, 20.09) of the 3D camera. According to the shooting posture and the external matrix, we acquire the coordinates of the front end of welding torch corresponding to the welding points at four welding positions in the robot base coordinate system. The results are shown in Table 9.
We acquire the coordinates of the front end of welding torch corresponding to the welding points at four welding positions in the robot base coordinate system by manual teaching [22]. The results are shown in Table 10.
After reasonable planning of the robot posture, the coordinates of the welding point in Table 9 are sent to the welding robot through the controller. Then, the welding robot drives the front end of the welding torch to accurately find the welding point and execute the welding task according to the planned welding path.
Finally, we analyze the error of the corresponding welding points coordinates in Tables 9 and 10. Figure 24a, b, and c respectively show the x, y, and z axes errors between the welding points coordinates acquired by the article method and manual teaching, which the error is within ±0.6 mm. Figure 24d shows the distance error between the welding points acquired by the article method and manual teaching, which the error is within 1 mm. Through the actual operation of the experiments and the error analysis, we found that all the errors are not more than 1 mm and within the allowable reasonable range, which does not affect the welding effect. It is verified that the method in this paper can realize the accuracy of welding path planning without teaching and programming.

Fig. 24
The error analysis between the welding point coordinates acquired by the article method and manual teaching

Conclusion
This paper studies a method of welding path planning of steel mesh based on point cloud for welding robot, which lays the foundation for the accurate planning of the steel mesh welding path and independent welding while eliminating the complicated teaching and programming work in welding path planning. The main contributions of this paper are summarized as follows.
1) The application of the 3D surface scanning structured light camera to the industrial welding scene can quickly and conveniently obtain the point cloud of the workpiece, which improves the welding efficiency. 2) This method solved the complicated teaching and programming problem of the welding robot before welding the steel mesh. Through the combination of point cloud library and mathematical theory, we can accurately plan the welding path of the steel mesh and complete the welding task without teaching and programming before welding. 3) We verify the feasibility, efficiency, and accuracy of the welding path planning method of steel mesh based on point cloud through analysis of method error, method efficiency and welding platform experiment results.
In the future work, we will improve and complete our work. Meanwhile, the proposed method also has some weaknesses. For example, the proposed method in this article is only suitable for the steel mesh workpieces. We will improve our method to adapt to different welding scenarios.
Author contribution Yusen Geng was a major contributor in writing the manuscript. All authors read and approved the final manuscript. Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.