An adaptive model matching fillet weld perception method and storage medium

Abstract

The application discloses a kind of self-adapting model matching fillet weld perception method, it is related to the technical field of automated welding.It can be applicable to standard base component, splicing component, fillet weld identification scene of mixed component, self-adapting matching length and height size variation component, intelligent perception mixed component weld by base component.The technical scheme of the application is: step 1, base component model weld information identification;Step 2, base component model is matched with actual workpiece point cloud;Step 3, the calculation of all weld information of actual workpiece.This case can effectively adapt to workpiece assembly error, simplify workpiece model database, while reducing camera imaging quality requirement and shooting frequency, simplify intelligent welding robot system weld identification workflow.

Description

Technical Field

[0001] This invention relates to the field of automated welding technology, and more particularly to the field of weld seam sensing and identification. Background Technology

[0002] Welding, as one of the most important processing technologies in modern manufacturing, is widely used in mechanical engineering, construction, shipbuilding, and automobile manufacturing. Common welding methods include manual welding and robotic welding. Manual welding is widely used in industrial production due to its high flexibility and adaptability; however, it suffers from low efficiency, inconsistent operator skill levels, and inconsistent weld quality, making it difficult to meet the demands of high-volume, high-speed production. Furthermore, the high temperatures, fumes, and arc light generated during welding can cause significant harm to welders. With the development of Industry 4.0 and intelligent manufacturing, industrial robots are gradually playing a crucial role in welding. By precisely controlling welding parameters and paths, they can ensure the consistency and stability of weld quality. Robots can also perform continuous, high-speed welding operations, improving production efficiency and capacity. Simultaneously, robotic welding reduces the safety risks for welders, such as exposure to high temperatures, harmful substances, and sparks, thus improving the safety of the working environment. However, robotic welding requires pre-programming and setup. For workpieces with complex shapes and varying sizes, significant adjustments and adaptations may be necessary. This results in problems such as long manual teaching times and inaccurate teaching positions. Furthermore, it cannot make real-time adjustments based on the welding environment and the workpiece being welded, making it difficult to guarantee welding accuracy. Therefore, intelligent welding robot systems equipped with machine vision have emerged. The key to achieving automated welding operations in industrial robots lies in how to quickly and accurately identify weld seams on workpieces automatically.

[0003] Based on the weld joint, common weld types can be divided into three main categories: butt joints, corner joints, and lap joints. Based on the weld shape, common weld types can be divided into straight lines, circles, and arcs. Among them, straight corner joints are the most common and are widely used in the welding of steel structural components in shipbuilding, automobile manufacturing, and other fields. They are usually the process of assembling plate parts or welding two or more plate and profile parts into sheet-like structural components. Generally, a large field of view and long viewing distance are required to identify multiple welds. Weld identification methods for workpieces can be divided into two types: model-based and model-based. Model-based methods directly calculate weld information from the actual point cloud. Due to field obstruction, the point cloud at the weld is not visible or only partially visible. Considering the principle of recognition accuracy and focusing on the weld, model-based methods require high camera imaging accuracy and anti-reflection capabilities, or require multiple multi-angle shots of the point cloud at the weld. Model-based methods, on the other hand, calculate the transformation matrix between the workpiece model and the actual point cloud to map the weld information of the model. However, it is necessary to consider the assembly errors that may occur when splicing basic components.

[0004] Chinese invention patent CN114372981B discloses a method, device, electronic device, and storage medium for identifying weld seams on T-shaped workpieces. It acquires initial point cloud data obtained by a lidar sensor moving and scanning along a direction perpendicular to the length of the T-shaped workpiece. The initial point cloud data is then processed by removing background noise based on the first height difference between adjacent points on each point cloud line. Abnormal point clouds caused by occlusion are then removed based on the second height difference and distance between adjacent points on the point cloud line. Height abrupt change points are then extracted from the point cloud line and grouped according to the distance between them. Linear fitting is then performed on each group of height abrupt change points to obtain the corresponding weld seam line. This method improves the accuracy of weld seam identification on T-shaped workpieces during ship assembly with a large line-of-sight and wide-field-of-view. However, this method directly calculates weld seam information by scanning the workpiece point cloud with a high-precision line-scan camera. High-precision line-scan cameras are expensive, and their imaging quality is unstable for lines perpendicular or nearly perpendicular to the camera's movement direction, potentially leading to glare and poor point cloud accuracy, which can interfere with the accuracy and robustness of the weld seam identification algorithm.

[0005] Chinese invention patent CN113177983B discloses a method for locating fillet welds based on point cloud geometric features. The method involves presetting the initial position of the camera, reconstructing the point cloud model by taking photos from multiple angles, calculating the transformation matrix between the overall point cloud model of the workpiece and the digital model of the workpiece, and then combining the camera and robot calibration matrix to map each weld seam of the digital model to the robot coordinate system. The method calculates the observation position and posture of each weld seam, controls the robot to move to the corresponding posture, and takes multiple more photos of the point cloud. Point cloud preprocessing and stitching are then performed to generate a relatively complete welded workpiece point cloud. Further processing, such as point cloud clustering and segmentation, calculation of point cloud surface intersection lines, and calculation of the start and end points of the lines, ultimately generates weld information. This method comprises three parts: weld observation pose generation, point cloud model reconstruction, and weld intersection line positioning and recognition. It automatically generates the actual point cloud model of the welded workpiece and the weld observation pose. Based on the actual point cloud model, it calculates the actual weld information, which to some extent overcomes the shortcomings of traditional model matching that cannot adapt to assembly errors. However, it requires two multi-angle shots to stitch the point cloud and control the robot's movement, making the process relatively complex. Furthermore, the accuracy of weld recognition largely depends on the error of the point cloud stitching. In addition, different complete models are required for components with different stitching shapes, which is not user-friendly.

[0006] In summary, how to utilize relatively low-cost vision acquisition mechanisms, such as 3D industrial cameras, to perform weld seam recognition workflows in intelligent welding robot systems, and to overcome the shortcomings of traditional model matching in adapting to assembly errors and the large data volume of traditional workpiece model databases, has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] To address the above problems, this invention proposes an adaptive model matching fillet weld perception method, which is applicable to fillet weld recognition scenarios for standard base components, spliced ​​components, and hybrid components. It adaptively matches components with varying length and height dimensions and intelligently perceives welds in hybrid components composed of base components.

[0008] The technical solution of this invention is as follows:

[0009] Follow these steps:

[0010] Step 1: Identify weld information of the basic component model;

[0011] Identify the weld information of the basic component model to obtain the welds of the basic component model;

[0012] Step 2: Match the basic component model with the actual workpiece point cloud;

[0013] First, the basic component model is upsampled to generate a model point cloud, and then geometric features are extracted from the surface point cloud of the basic component model in step 1.

[0014] Then, a 3D industrial camera is used to scan the actual workpiece point cloud. The workpiece point cloud is segmented multiple times to obtain the workpiece base plate plane and the surface planes of all components. All point clouds corresponding to the surface planes of all components are extracted. Then, clustering and segmentation are performed to obtain the point cloud of each component surface. Geometric features are extracted from the surface point cloud of each component.

[0015] The coarse registration matrix is ​​calculated based on the geometric features of the surface point cloud of the basic component model and the geometric features of the surface point cloud of each component on the actual workpiece. The point cloud position is initialized based on the coarse registration matrix, and then ICP fine registration is performed to optimize the position transformation and obtain the fine registration transformation matrix.

[0016] Step 3: Calculate all weld information for the actual workpiece;

[0017] Based on the transformation matrix finally obtained in step 2, the initial reference weld of each component on the actual workpiece is mapped. The initial reference weld and the point cloud of each component are then adaptively corrected to complete the coarse positioning of the weld information of each basic component. Then, the start and end points of the weld are precisely positioned based on the actual point cloud of the component to achieve flexible matching of the weld. Finally, the weld segments of complex components with splicing relationships are calculated and restored.

[0018] Step 1 is as follows:

[0019] Step 1.1: Perform two RANSAC plane splits on the mesh vertices of the basic component model to obtain the base plate surface srcPlane0 and the component surface plane srcPlane1, and calculate the height difference between the two planes. ;

[0020] Step 1.2: Extract the vertex coordinates corresponding to the boundary of the surface plane srcPlane1 of the component. And project it onto the base plate surface srcPlane0, the projection point is denoted as ;

[0021] Step 1.3: Randomly read in one of the above... Coordinates as the first point And find the point closest to it among the remaining points as the second point. , and That is, one of the ends of the two weld sections;

[0022] Step 1.4, respectively using and Starting point The remaining points are the endpoints; find the points that are the same as the endpoints. The perpendicular vector corresponds to the other endpoint of the two weld seams, resulting in two weld seam segments. , , recorded as .

[0023] Step 2 is as follows:

[0024] Step 2.1: Upsample the basic component model to generate a model point cloud, denoted as source;

[0025] Step 2.2: Extract the corresponding point cloud from the source based on the planar model coefficients of srcPlane1, denoted as... ;

[0026] Step 2.3: Analyze the surface point cloud of the basic component model from Step 1. Extracting geometric features Including the center of mass External normal direction Principal component direction ;

[0027] Step 2.4: Scan the workpiece point cloud with a 3D industrial camera and record it as the target;

[0028] Step 2.5: Perform multiple planar segmentations on the workpiece point cloud target to obtain the workpiece base plate plane dstPlane0 and all component surface planes, and extract all point clouds corresponding to the surface planes dstPlane of all components.

[0029] Step 2.6: Cluster and segment the point cloud extracted in the above steps to obtain the point cloud of each component surface, denoted as . , The number of parts is given, and the calculation is performed. The distance between the plane containing the workpiece and the plane containing the workpiece base plate, dstPlane0, is denoted as... ;

[0030] Step 2.7: Create a point cloud on the surface of each of the above components. Extracting geometric features Including the center of mass External normal direction Principal component direction ;

[0031] Step 2.8: Based on the surface point cloud of the basic component model and the point cloud on the surface of each component Calculating the coarse registration matrix from the geometric features of point clouds ;

[0032] Step 2.8.1: Correct the principal component direction;

[0033] ;

[0034] ;

[0035] Step 2.8.2: Calculate the target rotation vector;

[0036] ;

[0037] ;

[0038] Step 2.8.3: Calculate the axis of rotation;

[0039] ;

[0040] Step 2.8.4: Calculate the rotation angle;

[0041] ;

[0042] Step 2.8.5: Calculate the rotation matrix;

[0043] ,in It is the identity matrix. for Transpose of;

[0044] Step 2.8.6: Calculate the translation vector;

[0045] ;

[0046] Then the coarse registration matrix ;

[0047] Step 2.9: Initialize based on coarse registration matrix Point cloud location, and then with Point clouds are subjected to ICP fine registration, and position transformation is optimized to obtain the fine registration transformation matrix. That is, through The basic component model and the corresponding component surface point cloud in the actual workpiece point cloud are flexibly aligned.

[0048] Step 3 specifically involves:

[0049] Step 3.1: Based on the transformation matrix obtained from the point cloud matching above, map the initial reference weld of each actual component. ,in ; ;

[0050] Step 3.2: Map the initial reference weld as described above. and point cloud of each component Perform height adaptive correction, denoted as follows: , , where the correction vector is ;

[0051] Step 3.3, for Perform RANSAC line segmentation fitting and extract the starting point of the fitted line. and the end point ;

[0052] Step 3.4: Place the above starting point and the end point Along the matching weld seams respectively Obtain the foot of the perpendicular by projecting. , , , Update to the actual start and end points of the weld, and record the new weld as follows: ;

[0053] Step 3.5: Weld the above-mentioned components. Perform pairwise line segment intersection detection and calculate the coordinates of the intersection points, denoted as . And mark the intersection point as being in or In the middle section, assuming all components are joined vertically, then when the mark... At that time, there are two intersection points. and At this point, it is necessary to Split into two parts and update to , Keep it unchanged, and repeat the calculation until all flat welds have been updated;

[0054] Step 3.6: Vertical welds exist at the intersections of the spliced ​​components. This marks the end point of the vertical weld, through... Obtain the indexes of the two components with the smaller height relative to the base plate related to the intersection point, and the starting point of the vertical weld. Obtain vertical weld seam Repeat this step until the vertical weld calculation at all intersections is complete.

[0055] In addition, this case also proposes a storage medium storing a computer program that, when executed by a processor, runs the steps in the adaptive model matching fillet weld sensing method.

[0056] The beneficial effects of this invention are as follows:

[0057] 1. This system can automatically identify the weld information of the basic components, eliminating the need for manual selection of the welds.

[0058] Second, this solution can flexibly match components with varying lengths and heights, dynamically extract the start and end points of welds in each basic component of the actual point cloud, and achieve adaptive weld matching for multi-target point clouds.

[0059] Third, for actual complex components with splicing relationships, this case can estimate and restore the information of the real weld segment, and intelligently perceive the weld of the mixed component composed of basic components;

[0060] Fourth, this solution can effectively adapt to workpiece assembly errors, simplify the workpiece model database, reduce camera imaging quality requirements and the number of shots, and simplify the weld seam recognition workflow of the intelligent welding robot system.

[0061] Overall, this invention can effectively simplify the workpiece model database, intelligently perceive the weld seams of hybrid components composed of basic components, adapt to workpiece assembly errors, improve the flexibility of the model matching weld seam recognition method, reduce camera imaging quality requirements and the number of shots, and simplify the weld seam recognition workflow of the intelligent welding robot system. Attached Figure Description

[0062] Figure 1 This is a structural diagram of the welding system.

[0063] In the figure, 1 is the workpiece to be welded, 2 is the conveyor belt device, 3 is the laser 3D industrial camera (or structured light 3D industrial camera), 4 is the vision processing system, 5 is the industrial robot cabinet, 6 is the industrial robot body, 7 is the welding torch, and 8 is the welding machine.

[0064] Figure 2 This is a diagram showing the type of workpiece to be welded.

[0065] Figure 3 This is the flowchart of the algorithm framework for this case.

[0066] Figure 4 This is a schematic diagram of planar segmentation during weld identification of a basic component model.

[0067] Figure 5 This is a schematic diagram of weld extraction during weld identification in the basic component model.

[0068] Figure 6 This is a schematic diagram of the workpiece point cloud weld recognition results. Detailed Implementation

[0069] To clearly illustrate the technical features of this patent, the following detailed description is provided through specific embodiments and in conjunction with the accompanying drawings.

[0070] like Figure 1 As shown, the workpiece 1 to be welded moves at a constant speed through the conveyor belt 2. The laser 3D industrial camera 3 collects the workpiece point cloud in real time. The vision processing system 4 intelligently senses the weld seam information and plans a complete operation path that meets the welding requirements. It generates a welding executable program and sends it to the industrial robot cabinet (containing the controller) 5. The robot body 6 cooperates with the welding gun 7 and welding machine 8 according to the instructions to complete the automated welding operation.

[0071] like Figure 2 As shown, the types of workpieces 1 to be welded include single base components, multiple base components, single splicing components, multiple splicing components, and mixed components.

[0072] This case first identifies the weld information of the basic component model, then cloudifies the sampled points of the model, performs multi-target matching with the actual workpiece point cloud, adapts to changes in component length and height, coarsely locates the weld information of each basic component in the actual point cloud, then extracts and precisely locates the start and end points of the weld based on the actual point cloud, achieving flexible weld matching, and finally calculates and reconstructs the weld segments of complex components with splicing relationships. The specific steps are as follows:

[0073] Step 1: Identification of weld information in the basic component model;

[0074] Step 1.1: Perform two RANSAC plane partitions on the mesh vertices of the basic component model to obtain the plane model coefficients for the base plate surface srcPlane0 and the component surface srcPlane1, respectively. Figure 4 And calculate the height difference between the two planes. In mathematics, a plane can be represented by Ax+By+Cz+D=0, and the plane model coefficients here are {A,B,C,D}; the basic components in this embodiment include parts placed vertically on the base plate, and a fillet weld is formed at the junction between the bottom edge of the parts and the base plate;

[0075] Step 1.2: Extract the coordinates of the point corresponding to srcPlane1. And project it onto srcPlane0, the projection point is denoted as ;

[0076] Step 1.3: Randomly read in one of the above... Coordinates as the first point And find the point closest to it among the remaining points as the second point. , and That is, one of the ends of the two weld sections;

[0077] Step 1.4, respectively using and Starting point The remaining points are the endpoints; find the points that are the same as the endpoints. A perpendicular vector corresponds to the other endpoint of the two weld seams, such as... Figure 5 As shown, two weld segments were obtained. , , recorded as .

[0078] Step 2: Match the basic component model with the actual workpiece point cloud;

[0079] Step 2.1: Upsample the basic component model to generate a model point cloud, denoted as source;

[0080] Step 2.2: Extract the corresponding point cloud from the source based on the coefficients of the srcPlane1 planar model, denoted as... ;

[0081] Step 2.3: Analyze the surface point cloud of the above basic component model. Extracting geometric features Including the center of mass External normal direction Principal component direction (All vectors are normalized vectors);

[0082] Step 2.4: Scan the workpiece point cloud with a 3D industrial camera and record it as the target;

[0083] Step 2.5: Perform multiple planar segmentations on the workpiece point cloud target to obtain the workpiece base plate plane dstPlane0 and all component surface planes, and extract all point clouds corresponding to the surface planes dstPlane of all components.

[0084] Step 2.6: Cluster and segment the point cloud extracted in the above steps to obtain the point cloud of each component surface, denoted as . , The number of parts is given, and the calculation is performed. The distance between the plane containing the workpiece and the plane containing the workpiece base plate, dstPlane0, is denoted as... ;

[0085] Step 2.7: Create a point cloud on the surface of each of the above components. Extracting geometric features Including the center of mass External normal direction Principal component direction (All vectors are normalized vectors);

[0086] Step 2.8, according to and Calculating the coarse registration matrix from the geometric features of point clouds ;

[0087] Step 2.8.1: Correct the principal component direction;

[0088] ;

[0089] ;

[0090] Step 2.8.2: Calculate the target rotation vector;

[0091] (Normalization);

[0092] (Normalization);

[0093] Step 2.8.3: Calculate the axis of rotation;

[0094] ;

[0095] Step 2.8.4: Calculate the rotation angle;

[0096] ;

[0097] Step 2.8.5: Calculate the rotation matrix;

[0098] ,in It is the identity matrix. for Transpose of;

[0099] Step 2.8.6: Calculate the translation vector;

[0100] ;

[0101] Then the coarse registration matrix

[0102] Step 2.9: Initialize based on coarse registration matrix Point cloud location, and then with Point clouds are subjected to ICP (Iterative Closest Point) fine registration, and position transformation is optimized to obtain the fine registration transformation matrix. Thus, the complete matching transformation matrix can be obtained. That is, through The basic component model and the corresponding component surface point cloud in the actual workpiece point cloud can be flexibly aligned.

[0103] Step 3: Calculate all weld information for the actual workpiece;

[0104] Step 3.1: Based on the transformation matrix obtained from the point cloud matching above... Mapping the initial reference weld of each actual component ,in ; ;

[0105] Step 3.2: Map the initial reference weld as described above. and point cloud of each component Perform height adaptive correction, denoted as follows: , , where the correction vector is ;

[0106] Step 3.3, for Perform RANSAC line segmentation fitting and extract the starting point of the fitted line. and the end point ;

[0107] Step 3.4: Place the above starting point and the end point Along the matching weld seams respectively Obtain the foot of the perpendicular by projecting. , , , Update to the actual start and end points of the weld, and record the new weld as follows: ;

[0108] Step 3.5: Weld the above-mentioned components. Perform pairwise line segment intersection detection and calculate the coordinates of the intersection points, denoted as . And mark the intersection point as being in or In the middle section, assuming all components are joined vertically, then when the mark... At that time, there are two intersection points. and At this point, it is necessary to Split into two parts and update to , Keep it unchanged, repeat the calculation until all flat welds have been updated, such as... Figure 6 As shown;

[0109] Step 3.6: Vertical welds exist at the intersections of the spliced ​​components. This marks the end point of the vertical weld, through... Obtain the indexes of the two components with the smaller height relative to the base plate related to the intersection point, and the starting point of the vertical weld. Obtain vertical weld seam Repeat this step until the vertical weld calculations at all intersections are complete, such as... Figure 6 As shown.

[0110] There are many specific ways to implement this invention. The above description is only a preferred embodiment of this invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of this invention, and these improvements should also be considered within the scope of protection of this invention.

Claims

1. A method for sensing fillet welds using adaptive model matching, characterized in that, Follow these steps: Step 1: Identify weld information of the basic component model; Identify the weld information of the basic component model to obtain the welds of the basic component model; Step 1.1: Perform two RANSAC plane splits on the mesh vertices of the basic component model to obtain the base plate surface srcPlane0 and the component surface plane srcPlane1, and calculate the height difference between the two planes. ; Step 1.2: Extract the vertex coordinates corresponding to the boundary of the surface plane srcPlane1 of the component. And project it onto the base plate surface srcPlane0, the projection point is denoted as ; Step 1.3: Randomly read in one of the above... Coordinates as the first point And find the point closest to it among the remaining points as the second point. , and That is, one of the ends of the two weld sections; Step 1.4, respectively using and Starting point The remaining points are the endpoints; find the points that are the same as the endpoints. The perpendicular vector corresponds to the other endpoint of the two weld seams, resulting in two weld seam segments. , , recorded as ; Step 2: Match the basic component model with the actual workpiece point cloud; First, the basic component model is upsampled to generate a model point cloud, and then geometric features are extracted from the surface point cloud of the basic component model in step 1. Then, a 3D industrial camera is used to scan the actual workpiece point cloud. The workpiece point cloud is segmented multiple times to obtain the workpiece base plate plane and the surface planes of all components. All point clouds corresponding to the surface planes of all components are extracted. Then, clustering and segmentation are performed to obtain the point cloud of each component surface. Geometric features are extracted from the surface point cloud of each component. The coarse registration matrix is ​​calculated based on the geometric features of the surface point cloud of the basic component model and the geometric features of the surface point cloud of each component on the actual workpiece. The point cloud position is initialized based on the coarse registration matrix, and then ICP fine registration is performed to optimize the position transformation and obtain the fine registration transformation matrix. Step 3: Calculate all weld information for the actual workpiece; Based on the transformation matrix finally obtained in step 2, the initial reference weld of each component on the actual workpiece is mapped. The initial reference weld and the point cloud of each component are then adaptively corrected to complete the coarse positioning of the weld information of each basic component. Then, the start and end points of the weld are precisely positioned based on the actual point cloud of the component to achieve flexible matching of the weld. Finally, the weld segments of complex components with splicing relationships are calculated and restored.

2. The adaptive model matching fillet weld sensing method according to claim 1, characterized in that, Step 2 is as follows: Step 2.1: Upsample the basic component model to generate a model point cloud, denoted as source; Step 2.2: Extract the corresponding point cloud from the source based on the planar model coefficients of srcPlane1, denoted as... ; Step 2.3: Analyze the surface point cloud of the basic component model from Step 1. Extracting geometric features Including the center of mass External normal direction Principal component direction ; Step 2.4: Scan the workpiece point cloud with a 3D industrial camera and record it as the target; Step 2.5: Perform multiple planar segmentations on the workpiece point cloud target to obtain the workpiece base plate plane dstPlane0 and all component surface planes, and extract all point clouds corresponding to the surface planes dstPlane of all components. Step 2.6: Cluster and segment the point cloud extracted in the above steps to obtain the point cloud of each component surface, denoted as . , The number of parts is given, and the calculation is performed. The distance between the plane containing the workpiece and the plane containing the workpiece base plate, dstPlane0, is denoted as... ; Step 2.7: Create a point cloud on the surface of each of the above components. Extracting geometric features Including the center of mass External normal direction Principal component direction ; Step 2.8: Based on the surface point cloud of the basic component model and the point cloud on the surface of each component Calculating the coarse registration matrix from the geometric features of point clouds ; Step 2.8.1: Correct the principal component direction; ； ； Step 2.8.2: Calculate the target rotation vector; ； ； Step 2.8.3: Calculate the axis of rotation; ； Step 2.8.4: Calculate the rotation angle; ； Step 2.8.5: Calculate the rotation matrix; ,in It is the identity matrix. for Transpose of; Step 2.8.6: Calculate the translation vector; ； Then the coarse registration matrix ; Step 2.9: Initialize based on coarse registration matrix Point cloud location, and then with Point clouds are subjected to ICP fine registration, and position transformation is optimized to obtain the fine registration transformation matrix. Thus, the complete matching transformation matrix can be obtained. That is, through The basic component model and the corresponding component surface point cloud in the actual workpiece point cloud can be flexibly aligned.

3. The adaptive model matching fillet weld sensing method according to claim 2, characterized in that, Step 3 specifically involves: Step 3.1: Based on the transformation matrix obtained from the point cloud matching above... Mapping the initial reference weld of each actual component ,in ; ; Step 3.2: Map the initial reference weld as described above. and point cloud of each component Perform height adaptive correction, denoted as follows: , , where the correction vector is ; Step 3.3, for Perform RANSAC line segmentation fitting and extract the starting point of the fitted line. and the end point ; Step 3.4: Place the above starting point and the end point Along the matching weld seams respectively Obtain the foot of the perpendicular by projecting. , , , Update to the actual start and end points of the weld, and record the new weld as follows: ; Step 3.5: Weld the above-mentioned components. Perform pairwise line segment intersection detection and calculate the coordinates of the intersection points, denoted as . And mark the intersection point as being in or In the middle section, assuming all components are joined vertically, then when the mark... At that time, there are two intersection points. and At this point, it is necessary to Split into two parts and update to , Keep it unchanged, and repeat the calculation until all flat welds have been updated; Step 3.6: Vertical welds exist at the intersections of the spliced ​​components. This marks the end point of the vertical weld, through... By obtaining the indices of the two components with the smaller height relative to the base plate, the starting point of the vertical weld can be determined. Obtain vertical weld seam Repeat this step until the vertical weld calculation at all intersections is complete.

4. The storage medium of the adaptive model matching fillet weld sensing method according to any one of claims 1-3, characterized in that, The storage medium stores a computer program that, when executed by a processor, runs the steps of the adaptive model-matching fillet weld sensing method.

Images