A laser-tracking-based two-dimensional weld start point location method

Abstract

This invention belongs to the field of robotic welding, specifically relating to a laser-tracked method for locating the starting point of a two-dimensional weld. The method includes: scanning the workpiece to be welded frame-by-frame along the weld's extension direction using a laser sensor; converting the laser stripes obtained from each scan into point cloud data; inputting the laser stripes obtained from each scan into a trained convolutional neural network classifier to identify the type of weld bevel, and extracting feature points of the weld bevel using either the median method or Hough transform based on the type of weld bevel and the converted point cloud data; stopping the laser sensor from scanning the workpiece when the number of feature points of the weld bevel reaches a set threshold; fitting all extracted feature points of the weld bevel into a straight line using an orthogonal regression method based on random sampling consistency; and calculating the starting point coordinates of the weld in the laser sensor coordinate system based on the fitted straight line. This invention can accurately and efficiently automatically identify the starting points of various two-dimensional welds and welds with different bevel forms, greatly improving the automatic welding efficiency of welding robots.

Description

Technical Field

[0001] This invention belongs to the field of intelligent automation of robotic welding, and in particular relates to a two-dimensional weld start point positioning method based on laser tracking. Background Technology

[0002] To address the inevitable trend of high-quality, high-efficiency development in the manufacturing industry and the deepening reform of intelligent manufacturing, developing intelligent and automated welding technologies has become a key means to improve welding quality and efficiency. Among these, adaptive weld tracking is a crucial technology for achieving intelligent and automated welding. Given the diverse forms of welds on workpieces today, including straight welds, right-angle welds, S-shaped welds, and curved welds, accurately sensing the weld start point for V-shaped, I-shaped, Y-shaped, and lap-shaped weld bevels has become a significant factor restricting high-quality welding. Therefore, breakthroughs in automatic weld start point positioning and adaptive welding trajectory tracking technologies are urgently needed.

[0003] Currently, for conventional industrial production scenarios, the main methods for applying welding robots to locate the weld start point include: First, using a teach-and-playback approach. If the size or position of the workpiece to be welded changes, re-teaching is required, resulting in high labor intensity and low efficiency. Second, locating the weld start point through the design of tooling fixtures. However, clamping errors during workpiece processing can also lead to inaccurate starting point positioning. Finally, using visual sensors such as CCD cameras to capture weld images and employing related algorithms to find the weld start point. However, due to the interactive nature of welding work in the production environment, phenomena such as arc light, molten pool spatter, and sparks occur, resulting in various noises in CCD camera imaging and low welding efficiency. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a laser-tracking-based method for locating the starting point of a two-dimensional weld, applicable to all two-dimensional welds and bevel types. It is unaffected by arc light, molten pool spatter, sparks, etc., generated by the interactive working environment, and can be used in harsh working conditions and scenarios where manual contact is inaccessible, thereby improving the efficiency, robustness, and accuracy of welding.

[0005] This invention provides a two-dimensional weld start point location method based on laser tracking, comprising:

[0006] S1: Use a laser sensor to scan the workpiece to be welded frame by frame along the extension direction of the weld seam;

[0007] S2: Convert the laser stripes obtained from each scan into point cloud data;

[0008] S3: Input the laser stripes obtained from each scan into the trained convolutional neural network classifier to identify the type of weld groove, and select the median method or Hough transform to extract the feature points of the weld groove based on the type of weld groove and the converted point cloud data.

[0009] S4: When the number of feature points on the weld bevel reaches the set threshold, control the laser sensor to stop scanning the workpiece to be welded.

[0010] S5: Use the orthogonal regression method based on random sampling consistency to fit all the feature points of the extracted weld grooves into a straight line; calculate the coordinates of the starting point of the weld in the laser sensor coordinate system based on the fitted straight line.

[0011] Furthermore, the weld bevel types include: continuous and discontinuous; for continuous weld bevels, Hough transform is used to extract feature points; for discontinuous weld bevels, median method is used to extract feature points.

[0012] Furthermore, the extraction of feature points of the weld groove using the median method includes:

[0013] S311: Calculate the distance ΔI between any two adjacent points in the point cloud data. i =X i -X i-1 ; where X i Let X be the coordinate value of the i-th point in the point cloud data. i-1 Let be the coordinates of the (i-1)th point in the point cloud data, where i ∈ {2, 3, ..., N}, and N represents the number of points in the point cloud data.

[0014] S312: Traverse all points in the point cloud data and define ΔI i The maximum value is ΔI Max And let ΔI Max The two adjacent points are feature point C and feature point A;

[0015] S313: Based on the coordinate values ​​of feature points A and C, the feature point B of the weld groove is calculated using the median method. 终 Therefore, characteristic point B of the weld bevel. 终 The coordinates are (X B Z B )in

[0016] Furthermore, the extraction of feature points of the weld groove using Hough transform includes:

[0017] S321: Select the point with the largest ordinate in the point cloud data as point B. 始 ;

[0018] S322: Take the first point in the point cloud data as the left endpoint L, and the last point in the point cloud data as the right endpoint R; connect point R and point B. 始 The connected straight lines serve as the right reference line; calculate points R and B. 始 Find the shortest distance from each point to the right reference line, and select the point with the largest distance as point C;

[0019] S323: Connect points L and B 始 The connected straight lines serve as the left reference line; calculate points L and B. 始 Find the shortest distance from each point between them to the left reference line, and select the point with the largest distance as point A;

[0020] S324: Divide the point cloud data into LA segment and AB segment. 始 Section B 始 Segments C and CR are transformed using Hough transform to convert AB... 始 Section and B 始 Segment C is fitted to a straight line, AB 始 Segment of line and B 始 The intersection of the straight lines in segment C is the characteristic point B of the weld bevel. 终 .

[0021] Furthermore, the step of fitting all extracted weld bevel feature points into a straight line using the orthogonal regression method based on random sampling consistency includes: using the random sampling consistency algorithm to filter out some feature points of noisy weld bevels; and using the orthogonal regression method to fit the remaining feature points of weld bevels into a straight line.

[0022] Furthermore, the feature points for filtering out some noisy weld grooves using the random sampling consensus algorithm include:

[0023] S511: Randomly select two feature points from all extracted weld bevel feature points and connect them to form a straight line L; calculate the shortest distance from the remaining weld bevel feature points to the straight line L. If the shortest distance is less than a set threshold d, define the point as an in-place point and store it in dataset B. jin Save it in the middle;

[0024] S512: Determine if the current iteration count is greater than the maximum iteration count m. If yes, proceed to step S514; otherwise, proceed to step S513.

[0025] S513: Repeat steps S511-S512;

[0026] S514: Extract the dataset B with the largest number of local points during the iteration process. jin As the optimal dataset B max This will not belong to dataset B. maxThe characteristic points of the weld bevel are used as noise points for filtering.

[0027] Furthermore, the maximum number of iterations m includes:

[0028]

[0029] Where η0 is the confidence level, ε is the proportion of inliers to all sample points, e = 2 is the minimum number of points estimated by the model, and log represents the natural logarithm.

[0030] Furthermore, the step of fitting the remaining weld groove feature points into a straight line using orthogonal regression includes:

[0031] S521: Define the set of feature points of the remaining weld groove as P, and the target line as z = a + bx. Construct the objective function with the goal of minimizing the sum of the shortest distances from all points in set P to the line z = a + bx. Where n represents the number of points in set P; X j and Z j Let represent the coordinates of point j in set P; b represent the slope of the line; and a represent the intercept of the line.

[0032] S522: Regarding the objective function Find the partial derivative, that is, find and And order:

[0033]

[0034] Find the intercept of the final fitted line. slope in,

[0035] Further, the step of calculating the starting point coordinates of the weld in the laser sensor coordinate system based on the fitted straight line includes: taking the X coordinate of the first feature point in the feature point set P as the X coordinate of the starting point, substituting x equal to the X value into the fitted target straight line z = a + bx to obtain the Z coordinate of the starting point, then the two-dimensional coordinates of the starting point in the laser sensor coordinate system are (X, Z); then calculating the Y coordinate of the starting point based on the number of scanning frames of the laser sensor when the feature points of the weld groove are first extracted, thus obtaining the three-dimensional coordinates (X, Y, Z) of the starting point of the weld in the laser sensor coordinate system.

[0036] The present invention has at least the following beneficial effects:

[0037] This invention utilizes a laser sensor to precisely scan the workpiece to be welded, efficiently acquiring its two-dimensional point cloud data. A pre-trained convolutional neural network classifier identifies the weld bevel type and extracts its feature points. A straight line fitted using orthogonal regression with random sampling consistency ensures the accuracy and timeliness of weld start-point extraction, avoiding inaccurate start-point positioning due to errors. The point cloud data obtained through laser radar scanning allows for start-point location adaptation to welds of different sizes and dimensions. This invention not only improves the robustness and generalization of automatic weld start-point location, eliminating the need for repetitive teaching and reproduction for different welds, but also overcomes the influence of arc light, molten pool spatter, and sparks on start-point positioning through laser sensor scanning. This significantly improves the efficiency of automated welding robots, making it particularly suitable for common two-dimensional welds and various weld bevels in engineering, unaffected by harsh working conditions or industrial scenarios inaccessible to humans. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of scanning a workpiece to be welded using a laser sensor.

[0039] Figure 2 A step-by-step diagram for locating the starting point of a two-dimensional weld using laser tracking;

[0040] Figure 3 A flowchart for laser-tracked two-dimensional weld seam starting point location;

[0041] Figure 4 A schematic diagram for extracting weld bevel feature points for various weld bevel types;

[0042] Figure 5 This is a schematic diagram showing the results of filtering out noise points using the Ransac algorithm in MATLAB software;

[0043] Figure 6 This is a schematic diagram of the results of fitting a straight line through orthogonal regression;

[0044] Figure 7 This is a schematic diagram of the result of fitting a straight line using the least squares method;

[0045] Figure 8 This is a schematic diagram of the result of fitting a straight line using the iterative weighted least squares method;

[0046] Figure 9 A diagram illustrating eye-in-hand alignment. Detailed Implementation

[0047] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. In particular, the present invention is described using the coordinate system indicated in the accompanying drawings, which is intended to facilitate understanding by those skilled in the art and does not represent a sole limitation of the present invention. Those skilled in the art can convert the coordinate system indicated in the accompanying drawings based on relevant common knowledge.

[0048] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0049] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0050] Please see Figures 1-9 This invention provides a two-dimensional weld start point location method based on laser tracking, comprising:

[0051] S1: Use a laser sensor to scan the workpiece to be welded frame by frame along the extension direction of the weld seam;

[0052] S2: Convert the laser stripes obtained from each scan into point cloud data;

[0053] S3: Input the laser stripes obtained from each scan into the trained convolutional neural network classifier to identify the type of weld groove, and select the median method or Hough transform to extract the feature points of the weld groove based on the type of weld groove and the converted point cloud data.

[0054] S4: When the number of feature points on the weld bevel reaches the set threshold, control the laser sensor to stop scanning the workpiece to be welded.

[0055] S5: Use the orthogonal regression method based on random sampling consistency to fit all the feature points of the extracted weld grooves into a straight line; calculate the coordinates of the starting point of the weld in the laser sensor coordinate system based on the fitted straight line.

[0056] This embodiment takes an S-type weld with a continuous V-type weld groove as an example. Figure 1 As shown, the robotic welding torch 1 and the laser sensor 4 are fixedly connected together via a connecting bracket 3. The laser sensor 4 is kept parallel to the robotic welding torch 1 and located close to the workpiece to be welded. Simultaneously, the height of the laser sensor 4 and the workpiece to be welded are pre-adjusted to ensure that the scanning range of the laser sensor 4 covers the entire longitudinal distance of the workpiece. During the positioning process, the actuator drives the robotic welding torch 1 and the laser sensor 4 to scan the workpiece frame by frame along the weld seam extension direction, ensuring that the range of the laser stripes covers the entire longitudinal distance of the workpiece. Figure 3 The flowchart shows the process of finding the starting point of a 2D weld using laser tracking. The entire process includes scanning, extracting feature point cloud data, filtering out noise points in the point cloud data, fitting a straight line to the dataset after filtering out noise points, and calculating the starting point of the weld.

[0057] In this embodiment, a laser sensor is used to precisely scan the workpiece to be welded, which can efficiently acquire the two-dimensional point cloud data of the workpiece. A pre-trained convolutional neural network classifier is used to obtain the type of weld bevel and extract the feature points of the weld bevel. The straight line fitted by the orthogonal regression method of random sampling consistency ensures the accuracy and timeliness of the extraction of the weld start point, avoiding problems such as inaccurate start point positioning due to errors. The present invention obtains point cloud data by laser radar scanning, which can adapt to welds of different sizes and shapes for start point positioning. The present invention not only improves the robustness and generalization of automatic weld start point positioning, eliminating the need for repeated teaching and reproduction for different welds, but also overcomes the influence of arc light, molten pool spatter, sparks and other factors on start point positioning by using laser sensor scanning positioning, which significantly improves the efficiency of automatic welding by welding robots. It is especially suitable for common two-dimensional welds and various weld bevels in engineering and is not affected by harsh working conditions or industrial scenarios inaccessible to humans.

[0058] In this embodiment, appropriate methods are used to extract weld bevel feature points for different weld bevel types. For example... Figure 4 As shown in (a), for weld bevels with continuous point cloud data such as fillet joints or V-shaped welds, Hough transform is used to extract weld bevel feature points; such as Figure 4As shown in (b), the median method is used to extract weld bevel feature points for point cloud data with discontinuous types such as Y-shaped or I-shaped weld bevels. By introducing a pre-trained convolutional neural network classifier, the different types of weld bevels can be accurately identified, laying a good foundation for subsequent weld bevel feature point extraction.

[0059] Furthermore, the weld bevel types include: continuous and discontinuous; for continuous weld bevels, Hough transform is used to extract feature points; for discontinuous weld bevels, median method is used to extract feature points.

[0060] Furthermore, the extraction of feature points of the weld groove using the median method includes:

[0061] S311: Calculate the distance ΔI between any two adjacent points in the point cloud data. i =X i -X i-1 ; where X i Let X be the coordinate value of the i-th point in the point cloud data. i-1 Let be the coordinates of the (i-1)th point in the point cloud data, where i ∈ {2, 3, ..., N}, and N represents the number of points in the point cloud data.

[0062] S312: Traverse all points in the point cloud data and define ΔI i The maximum value is ΔI Max And let ΔI Max The two adjacent points are feature point C and feature point A;

[0063] S313: Based on the coordinate values ​​of feature points A and C, the feature point B of the weld groove is calculated using the median method. 终 Therefore, characteristic point B of the weld bevel. 终 The coordinates are (X B Z B )in

[0064] Furthermore, the extraction of feature points of the weld groove using Hough transform includes:

[0065] S321: Select the point with the largest ordinate in the point cloud data as point B. 始 ;

[0066] S322: Take the first point in the point cloud data as the left endpoint L, and the last point in the point cloud data as the right endpoint R; connect point R and point B. 始 The connected straight lines serve as the right reference line; calculate points R and B. 始 Find the shortest distance from each point to the right reference line, and select the point with the largest distance as point C;

[0067] S323: Connect points L and B 始 The connected straight lines serve as the left reference line; calculate points L and B. 始 Find the shortest distance from each point between them to the left reference line, and select the point with the largest distance as point A;

[0068] S324: Divide the point cloud data into LA segment and AB segment. 始 Section B 始 Segments C and CR are transformed using Hough transform to convert AB... 始 Section and B 始 Segment C is fitted to a straight line, AB 始 Segment of line and B 始 The intersection of the straight lines in segment C is the characteristic point B of the weld bevel. 终 .

[0069] Specifically, the extraction of weld groove feature points using Hough transform includes:

[0070] Step S321: As Figure 4 As shown in (a), the point with the largest ordinate in the point cloud data is taken as point B. 始 ;

[0071] Step S322: As Figure 4 As shown in (c), the first point in the point cloud data is taken as the left endpoint L, and the last point in the point cloud data is taken as the right endpoint R; point R and point B are... 始 The connected straight lines serve as the right reference line; calculate points R and B. 始 Find the shortest distance from each point to the right reference line, and select the point with the largest distance as point C;

[0072] Step S323: Connect point L and point B 始 The connected straight lines serve as the left reference line; calculate points L and B. 始 Find the shortest distance from each point between them to the left reference line, and select the point with the largest distance as point A;

[0073] Step S324: As Figure 4 As shown in (d), the point cloud data is divided into LA segment and AB segment. 始 Section B 始 Segments C and CR are transformed using Hough transform to convert AB... 始 Section and B 始 Segment C is fitted to a straight line, AB 始 Segment of line and B 始 The intersection of the straight lines in segment C is the characteristic point B of the weld bevel. 终 .

[0074] Further, specifically, step S4, where the number of feature points on the weld bevel reaches a set threshold, involves controlling the laser sensor to stop scanning the workpiece to be welded. This specifically includes:

[0075] The laser sensor is equipped with a counter M, and the initial value of counter M is 0;

[0076] Depend on Figure 1 , Figure 2 and Figure 3 It can be known that:

[0077] Step S401: The laser sensor scans the workpiece to be welded along the extension direction of the weld seam.

[0078] Step S402: If the laser sensor acquires point cloud data and successfully extracts weld bevel feature points, then counter M = M + 1 and the feature point is stored in point set B. j Otherwise, return to step S401:

[0079] Step S403: Determine if counter M ≥ 30. If so, the laser sensor stops scanning the workpiece to be welded and obtains the complete set of weld groove feature points B. j Otherwise, return to step S401.

[0080] Furthermore, the step of fitting all extracted weld bevel feature points into a straight line using the orthogonal regression method based on random sampling consistency includes: using the random sampling consistency algorithm to filter out some feature points of noisy weld bevels; and using the orthogonal regression method to fit the remaining feature points of weld bevels into a straight line.

[0081] Furthermore, the feature points for filtering out some noisy weld grooves using the random sampling consensus algorithm include:

[0082] S511: In the feature point set B of all extracted weld bevels j Two feature points are randomly selected from the data and connected to form a straight line L. The shortest distance from the remaining feature points of the weld bevel to the straight line L is calculated. If the shortest distance is less than a set threshold d, the point is defined as an inlier and stored in dataset B. jin Save it in the middle;

[0083] S512: Determine if the current iteration count is greater than the maximum iteration count m. If yes, proceed to step S514; otherwise, proceed to step S513.

[0084] S513: Repeat steps S511-S512;

[0085] S514: Extract the dataset B with the largest number of local points during the iteration process. jin As the optimal dataset B max This will not belong to dataset B.max The characteristic points of the weld bevel are used as noise points for filtering.

[0086] Furthermore, the maximum number of iterations m includes:

[0087]

[0088] Where η0 is the confidence level (typically 0.95-0.99), ε is the proportion of inliers to all sample points, e = 2 is the minimum number of points estimated by the model, and log represents the natural logarithm. Figure 5 This is a schematic diagram showing the results of using the Ransac algorithm in MATLAB to filter out noise points. The results show that the Ransac algorithm is effective and successfully filters out noise outliers.

[0089] Furthermore, the step of fitting the remaining weld groove feature points into a straight line using orthogonal regression includes:

[0090] S521: Define the set P of feature points for the remaining weld groove, fit the target line z = a + bx, and construct the objective function with the goal of minimizing the sum of the shortest distances from all points in set P to the line z = a + bx. Where n represents the number of points in set P; X j and Z j Let represent the coordinates of point j in set P; b represent the slope of the line; and a represent the intercept of the line.

[0091] S522: Regarding the objective function Find the partial derivative, that is, find and And order:

[0092]

[0093] Find the intercept of the final fitted line. slope in,

[0094] Figure 6 , Figure 7 and Figure 8 This diagram illustrates the results of fitting the set of feature points P of the remaining weld groove using different common engineering straight line fitting methods in MATLAB. After calculating the sum of squared errors (SSE) and root mean square errors (RMSE), the final conclusion is that the orthogonal regression method for fitting the straight line has smaller sum of squared errors (SSE) and root mean square errors (RMSE) compared to other methods. Furthermore, its fitting speed is comparable to least squares and faster than iterative weighted least squares. Therefore, it has the highest goodness of fit and better fitting accuracy, laying a solid foundation for the subsequent calculation of the weld initiation point.

[0095] Further, the step of calculating the starting point coordinates of the weld in the laser sensor coordinate system based on the fitted straight line includes: taking the X coordinate of the first feature point in the feature point set P as the X coordinate of the weld starting point, substituting x equal to the X value into the fitted target straight line z = a + bx to obtain the Z coordinate of the weld starting point, then the two-dimensional coordinates of the weld starting point in the laser sensor coordinate system are (X, Z); then calculating the Y coordinate of the weld starting point based on the number of scanning frames of the laser sensor when the feature points of the weld bevel are first extracted, thus obtaining the three-dimensional coordinates (X, Y, Z) of the weld starting point in the laser sensor coordinate system.

[0096] Furthermore, such as Figure 9 The hand-eye calibration shown adopts the eye-in-hand method. Let the robot base coordinate system be {B}, the robot tool coordinate system be {T}, the laser sensor coordinate system be {C}, and the coordinate system of the robot end effector flange center be {E}. The transformation matrix between the robot tool coordinate system and the robot end effector flange center coordinate system is: The transformation matrix between the laser sensor coordinate system and the robot tool coordinate system is: The transformation matrix between the robot's end effector flange center coordinate system and the robot's base coordinate system is: Let the coordinates of any point P in space in the laser sensor coordinate system be P( c X c Y c Z c The coordinates of P in the robot's base coordinate system are B (X B Y B Z B Therefore, based on the fact that any point P in space is equal in position under the same coordinate system, the following equation can be constructed: The coordinates of the weld start point in the laser sensor coordinate system are transformed into coordinates in the robot's base coordinate system. Finally, the welding robot is controlled to move to the weld start point, ensuring the stability and consistency of the welding process and completing the adaptive search for and positioning of the weld start point.

[0097] In summary, this invention uses a laser sensor to precisely scan the workpiece to be welded, efficiently acquiring its two-dimensional point cloud data. A pre-trained convolutional neural network classifier identifies the weld bevel type and extracts its feature points. A straight line fitted using orthogonal regression with random sampling consistency ensures the accuracy and timeliness of weld start-point extraction, avoiding inaccurate start-point positioning due to errors. This invention uses laser radar scanning to obtain point cloud data, enabling it to locate start-points for welds of different sizes and dimensions. This invention not only improves the robustness and generalization of automatic weld start-point positioning, eliminating the need for repetitive teaching and reproduction for different welds, but also overcomes the influence of arc light, molten pool spatter, and sparks on start-point positioning through laser sensor scanning. This significantly improves the efficiency of automated welding robots, making it particularly suitable for common two-dimensional welds and various weld bevels in engineering, unaffected by harsh working conditions or industrial scenarios inaccessible to humans.

[0098] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A two-dimensional weld start point location method based on laser tracking, characterized in that, include: S1: Use a laser sensor to scan the workpiece to be welded frame by frame along the extension direction of the weld seam; S2: Convert the laser stripes obtained from each scan into point cloud data; S3: Input the laser stripes obtained from each scan into the trained convolutional neural network classifier to identify the type of weld groove, and select the median method or Hough transform to extract the feature points of the weld groove based on the type of weld groove and the converted point cloud data. The extraction of feature points of weld groove using Hough transform includes: S321: Select the point with the largest ordinate in the point cloud data as the point. ; S322: Take the first point in the point cloud data as the left endpoint L, and the last point in the point cloud data as the right endpoint R; connect point R and point... The connected straight lines serve as the right reference line; calculate point R and point... Find the shortest distance from each point to the right reference line, and select the point with the largest distance as point C; S323: Connect point L and point... The connected straight lines serve as the left reference line; calculate point L and point... Find the shortest distance from each point between them to the left reference line, and select the point with the largest distance as point A; S324: Divide the point cloud data into LA segment and A segment. part, Segments C and CR are transformed using Hough transform to convert A Duan He Segment C is fitted to a straight line, and A Segment of a straight line and The intersection of the straight lines in segment C is the characteristic point of the weld bevel. ; S4: When the number of feature points on the weld bevel reaches the set threshold, control the laser sensor to stop scanning the workpiece to be welded. S5: Use the orthogonal regression method based on random sampling consistency to fit all the feature points of the extracted weld grooves into a straight line; calculate the coordinates of the starting point of the weld in the laser sensor coordinate system based on the fitted straight line.

2. The laser-tracking-based two-dimensional weld start point location method according to claim 1, characterized in that, The weld bevel types include continuous and discontinuous types. For continuous weld bevels, Hough transform is used to extract feature points; for discontinuous weld bevels, median method is used to extract feature points.

3. The two-dimensional weld start point location method based on laser tracking according to claim 1, characterized in that, The step of fitting all extracted weld bevel feature points into a straight line using the orthogonal regression method based on random sampling consistency includes: using the random sampling consistency algorithm to filter out some feature points of noisy weld bevels; and using the orthogonal regression method to fit the remaining feature points of weld bevels into a straight line.

4. The two-dimensional weld start point location method based on laser tracking according to claim 3, characterized in that, The feature points for filtering out some noise weld grooves using the random sampling consensus algorithm include: S511: Randomly select two feature points from all extracted weld bevel feature points and connect them to form a straight line L; calculate the shortest distance from the remaining weld bevel feature points to the straight line L. If the shortest distance is less than a set threshold d, define the point as an in-place point and store it in the dataset. Save it in the middle; S512: Determine if the current iteration count is greater than the maximum iteration count m. If yes, proceed to step S514; otherwise, proceed to step S513. S513: Repeat steps S511-S512; S514: Extract the dataset with the largest number of local points during the iteration process. As the optimal dataset This will not belong to the dataset. The characteristic points of the weld bevel are used as noise points for filtering.

5. The two-dimensional weld start point location method based on laser tracking according to claim 4, characterized in that, The maximum number of iterations m includes: in, For credibility, The proportion of in-place points to all sample points. This represents the minimum number of points estimated by the model. It represents the natural logarithm.

6. The laser-tracking-based two-dimensional weld start point location method according to claim 5, characterized in that: The method of fitting the remaining feature points of the weld groove into a straight line using orthogonal regression includes: S521: Define the set of feature points for the remaining weld groove as follows The target line is From all points in set P to the line The objective function is to minimize the sum of the shortest distances. ,in, Represents a set The number of midpoints; and Denotes the midpoint of set P The coordinates; Represents the slope of a straight line; Represents the intercept of a line; S522: Regarding the objective function Find the partial derivative, that is, find and And order: Find the intercept of the final fitted line. slope ;in, , , , , , , , .

7. The two-dimensional weld start point location method based on laser tracking according to claim 6, characterized in that, The calculation of the starting point coordinates of the weld in the laser sensor coordinate system based on the fitted straight line includes: [The process involves] setting up a set of feature points... Let the X-coordinate of the first feature point be taken as the X-coordinate of the starting point, and let... Substitute the X value into the target line of the fit. The Z coordinate of the starting point is obtained, and the two-dimensional coordinates of the starting point in the laser sensor coordinate system are (X, Z). Then, the Y coordinate of the starting point is calculated based on the number of scanning frames of the laser sensor when the feature points of the weld groove are first extracted, so as to obtain the three-dimensional coordinates (X, Y, Z) of the weld starting point in the laser sensor coordinate system.

Images