Point cloud data-based welding seam track identification method for impeller wrap angle welding

Abstract

The invention discloses a welding seam track identification method for impeller wrap angle welding based on point cloud data, and belongs to the field of machine welding seam identification, and the method comprises the following steps: carrying out filtering, down-sampling and clustering on original scanning point cloud; classifying the point cloud data through a region growing clustering algorithm; determining an optimal interception plane based on principal component analysis, and performing curve fitting on section data to obtain a preliminary main welding seam track; the main welding seam track is projected to a local approximate plane in a segmented mode, and accurate extraction of the main welding seam track is completed; and performing plane fitting and thickness calculation on the wrap angle surface, solving an intersection line in combination with an adjacent local approximate plane, calculating an intersection point of the intersection line and the main welding seam track, and finally generating a wrap angle welding seam track in combination with the direction of the intersection line and the wrap angle thickness. Through cooperative calculation of local plane projection and spatial intersection line solving, the whole process operation from recognition, planning to welding can be automatically completed, and unmanned intelligent welding in the true sense is achieved.

Description

Technical Field

[0001] This invention belongs to the field of machine recognition of weld seams, specifically a method for identifying the weld seam trajectory of impeller fillet welds based on point cloud data. Background Technology

[0002] Enclosed corner welding significantly improves structural strength, stress distribution, and fatigue resistance, making it a crucial process for key load-bearing components such as wind turbine impellers and power towers. However, this process currently relies heavily on manual welding, resulting in inherent drawbacks such as low efficiency, high labor intensity, and difficulty in ensuring consistent quality. Especially when the adjacent surfaces of the weld seam are complex spatial curved surfaces (such as the hyperboloid structure of wind turbine impellers), the difficulty of three-dimensional contour recognition and trajectory detection of the weld seam increases dramatically, further hindering the application of automated welding technology and making welding such workpieces a production bottleneck.

[0003] Currently, fillet welding typically employs two methods: 1. When dealing with workpieces with complex curved surfaces or irregular shapes, it generally relies on skilled workers for the entire manual process. This mode not only requires manual identification and welding of all curved or straight weld seams but also necessitates additional fillet weld repair. Its quality depends entirely on the worker's experience and condition, resulting in significant drawbacks such as low efficiency, high labor intensity, poor consistency, and high production costs. 2. For scenarios with relatively simple geometry, robotic automation systems can be used to scan and weld the main curved or straight weld seams. However, this system is ineffective for fillet welds outside the robot's pre-programmed parameters, requiring manual repair. While this method partially improves efficiency, it remains essentially a fragmented process. It also introduces new challenges such as robot programming and debugging, manual operation coordination, and high overall process coordination complexity, failing to achieve true full-process automation. Summary of the Invention

[0004] To address the above issues, this application proposes a method for identifying the weld trajectory of impeller fillet welds based on point cloud data. Specifically, it proposes and implements an algorithm mechanism based on point cloud data that can automatically analyze and reconstruct the complex connection relationship between the fillet weld and the main weld trajectory in three-dimensional space. This mechanism, through the collaborative calculation of "local plane projection" and "spatial intersection line acquisition," decomposes the complex spatial problem on the global surface into a series of locally solvable planar geometric problems, thereby accurately locating inflection points and generating continuous, smooth three-dimensional weld trajectories.

[0005] A method for identifying the weld trajectory of impeller fillet welds based on point cloud data includes the following steps: S1: Point cloud data preprocessing: Filtering, downsampling and clustering of the original scanned point cloud to effectively remove noise and reduce data volume while retaining key geometric features; S2: Point cloud clustering and segmentation: The preprocessed point cloud data is divided into three key parts by a region growing clustering algorithm with adjustable parameters: front cover surface, blade surface and wrap angle surface; S3: Main weld trajectory extraction: Based on principal component analysis, the intercepting plane is determined, and curve fitting is performed on the cross-sectional data of the blade surface to obtain the preliminary main weld trajectory; then the main weld trajectory is segmented and projected onto the local approximate plane of the front disc cover to complete the accurate extraction of the main weld trajectory; S4: Fillet weld trajectory generation: Perform plane fitting and thickness calculation on the fillet surface to obtain the intersection line of the plane with the adjacent local approximate plane; by calculating the intersection point of the intersection line with the main weld trajectory, and combining the direction of the intersection line with the fillet thickness, the complete fillet weld trajectory is finally generated.

[0006] The point cloud data preprocessing in step S1 includes: The robot is equipped with a 3D vision sensor to scan and obtain the initial 3D point cloud of the workpiece. The original point cloud is filtered and voxel downsampled. Based on the structural characteristics of the workpiece, a region growing clustering algorithm is used to retain the cluster with the most points.

[0007] The point cloud clustering and segmentation in step S2 includes: First fine clustering: The preprocessed data is clustered a second time using the region growing method; the two largest data categories are extracted, corresponding to the front disc cover data and the impeller composite data, respectively; at the same time, the data that were not classified in the clusters are extracted and recorded as connected data.

[0008] Secondary fine clustering: The clustering parameters of the region growing algorithm are adjusted again, and the impeller composite data obtained above is clustered for the third time, thereby successfully separating the wrap angle surface data from the blade surface data.

[0009] The acquisition of the intercepting plane in step S3 includes: Principal component analysis is performed on the data obtained in step S2, and a cross product operation is performed on the directions of the two principal components to obtain a vector with an approximate height direction. This vector is used as the normal vector of the intercept plane. At the same time, the centroid connecting the data is taken as a point on the intercept plane, thus uniquely determining the intercept plane.

[0010] The local projection in step S3 includes: According to the preset grouping threshold, the initial trajectory is divided into multiple line segments; the centroid of each line segment is used as the query point, and the data of the front cover surface is extracted in the neighborhood range. The local data points are then fitted to a plane to obtain a local approximate plane; the line segments are then projected onto their respective local approximate planes.

[0011] In step S4, the intersection of the intersection line and the main weld trajectory is the inflection point of the fillet weld and the main weld trajectory.

[0012] The impeller composite data is an undivided set of the wrap angle surface and the blade surface, and the connection data is the connection transition area between the front disc cover surface and the impeller body.

[0013] Beneficial effects: This invention provides a high-precision, high-efficiency trajectory detection foundation for robotic automated welding on complex workpieces requiring corner welding, such as impellers and power towers. Practical verification shows that the algorithm's processing time is consistently within 500 milliseconds, and the spatial error of the generated weld trajectory is less than 1 millimeter. This method requires no separate processing or manual intervention for the corner weld; it can simultaneously generate the main weld trajectory and the corner weld trajectory in a single scan planning, achieving seamless connection and one-time welding between the two types of welds. These performance indicators fully meet the requirements of industrial sites for welding accuracy and production cycle time. Attached Figure Description

[0014] Figure 1 It is an algorithm flowchart; Figure 2 This is a drawing of a physical hyperboloid impeller. Figure 3 It is a scanned point cloud image of a hyperboloid impeller; Figure 4 This is a clustering result of the point cloud of the hyperboloid impeller; Figure 5 This is a schematic diagram of data extraction from the plane of the corner face; Figure 6 , Figure 7 This is a diagram showing the results of impeller weld identification. Detailed Implementation

[0015] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described in this application are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0016] This invention proposes a method for identifying the weld trajectory of a fillet weld on an impeller based on point cloud data. The fillet weld trajectory includes the main weld trajectory and the fillet weld trajectory. The following example uses a hyperboloid impeller, which is both challenging and representative. Figure 2 Taking the type shown as an example, the implementation steps of this method will be explained in detail.

[0017] S1: Point cloud data preprocessing.

[0018] The robot is equipped with a 3D vision sensor to scan and acquire the initial 3D point cloud of the impeller (e.g., ...). Figure 3 (As shown).

[0019] The original point cloud is subjected to voxel downsampling. The original point cloud data is mapped to a pre-defined voxel mesh. The centroid of all point cloud data within a voxel is calculated, and this centroid is used as the representative point of that voxel. Other redundant point clouds within the voxel are discarded. This significantly reduces the point cloud density while ensuring that the key geometric information of the workpiece is not lost.

[0020] Based on the structural characteristics of the impeller, a region growing clustering algorithm is used on the downsampled point cloud to count the number of points contained in each cluster and retain the cluster with the most points. This step can remove isolated noise points and interference data surfaces formed by impurities that are far away from the main working surface of the impeller, while completely preserving the continuous point cloud structure of the main working surface of the impeller, ensuring the integrity and accuracy of the point cloud after denoising.

[0021] S2: Point cloud clustering and segmentation (see...) Figure 4 ).

[0022] First fine clustering: The parameters of the region growing algorithm were adjusted, and a second clustering was performed on the preprocessed data. The two largest categories in this cluster were extracted, corresponding to the front disc cover data and the impeller composite data (the unsegmented set of the corner surface and blade surface). At the same time, the unclassified point clouds in this cluster were extracted and denoted as connectivity data, which represents the transition region connecting the front disc cover and the impeller body.

[0023] Secondary fine clustering: The parameters of the region growing algorithm are adjusted again to perform a third clustering on the obtained impeller composite data. The impeller composite data includes the wrap-around surface and the blade surface. Although both belong to the main body of the impeller, their surface features are significantly different. Therefore, by adjusting the parameters, the wrap-around surface and the blade surface can be accurately separated.

[0024] S3: Extraction of main weld seam trajectory.

[0025] Determine the intercept plane: Principal component analysis (PCA) is performed on the blade surface data and connection data obtained in step S2. The core purpose of PCA is to extract the main distribution characteristic directions of the two types of data: the distribution of blade surface data mainly spreads along its surface extension direction, and its minimum principal component direction corresponds to the direction with the smallest data distribution variance. This direction is perpendicular to the extension plane of the blade surface, and therefore approximates the thickness direction of the blade; the distribution of connection data is concentrated along the extension path of the main weld trajectory, and its maximum principal component direction corresponds to the direction with the largest data distribution variance, and therefore approximates the extension direction of the main weld trajectory.

[0026] A cross product is performed between the minimum principal component direction of the blade surface data and the maximum principal component direction of the connecting data to obtain a vector approximating the height direction. This vector serves as the normal vector of the intercepting plane. Simultaneously, the centroid of the connecting data is calculated and used as a fixed point on the intercepting plane. Since, according to the point normal form of a plane, the position of a plane can be uniquely determined by its normal vector and a fixed point on the plane, the intercepting plane can be uniquely determined using the obtained normal vector and fixed point.

[0027] Curve fitting and local projection: The cross-sectional data of the blade surface is captured using a cutting plane (this cross-sectional data directly reflects the contour shape of the blade surface in the connection area), and curve fitting is performed. The fitted curve can best fit the distribution trend of the cross-sectional data, so it is used as the preliminary main weld trajectory, providing a basic framework for subsequent local precise adjustments.

[0028] Based on a preset grouping threshold, the initial trajectory is divided into multiple line segments. The preset grouping threshold is a configurable parameter, which can be set according to the blade model, the rate of curvature change in the connecting area, or welding process requirements. The core purpose of segmentation is that the front disc cover surface may have slight curvature variations or local uneven structures, making it difficult for a single global trajectory to adapt to the actual geometry of all areas.

[0029] For each segmented line, its centroid is calculated and used as a query point. Front cover surface data is extracted within its neighborhood, and plane fitting is performed on this local data point to obtain a local approximate plane. Finally, the corresponding line segments are projected onto their respective local approximate planes to obtain the projected line segments of the line segments on the local approximate planes. All projected line segments are connected sequentially according to the original segmentation order to finally form the accurate main weld trajectory.

[0030] S4: Extraction of fillet weld.

[0031] Calculation of wrap angle thickness: Perform multiple planar cuts at different positions of the corner surface data obtained in step S2, and calculate the thickness of each section (see...). Figure 5 ).in Figure 5 The cutting plane in step S3 is parallel to the cutting plane obtained in step S3, which is equivalent to extracting data at different heights from the cutting plane in step S3. The extracted data is then fitted with a straight line to obtain the thickness. After removing outliers from the thickness of all sections, the arithmetic mean of the remaining thicknesses is calculated, and this average value is used as the final edge-wrapping thickness.

[0032] Corner trajectory generation: Plane fitting is performed on the complete corner surface point cloud data from step S2 to obtain the corner plane. From the local approximation planes obtained in step S3, local approximation planes adjacent to the corner plane are selected. For the adjacent local approximation planes, the intersection line with the corner plane is calculated according to the spatial intersection formula of two planes, and the direction of the intersection line is further determined. This direction is the extension direction of the corner weld trajectory. Two consecutive trajectory points at the beginning of the main weld trajectory are selected, and the line segments they contain are spatially intersected with the intersection line. The resulting intersection point is the inflection point. Since the intersection line is jointly determined by the corner plane and the adjacent local approximation planes, it can effectively correct the length deviation at the beginning of the main weld trajectory. Finally, the final corner weld trajectory is generated using the intersection line direction as the trajectory extension reference and combined with the pre-calculated edge thickness parameters (see...). Figure 6 and Figure 7 ).

[0033] The application scenarios of this invention are not limited to the hyperboloid impellers described above. Its core technical solution is universally applicable to complex workpieces requiring corner welding, such as impellers and power towers. Similarly, for simple scenarios where the adjacent surfaces of the welds are planar, the above process is still applicable, but a simpler implementation process can also be adopted: In the main weld trajectory extraction process in step S3, the entire adjacent surface can be fitted with a plane as the projection plane, replacing the local approximate plane; the cross-sectional data can be fitted with a straight line, replacing the curve fitting, to obtain a straight weld.

Claims

1. A method for identifying a weld seam trajectory for a shrouded impeller fillet weld based on point cloud data, characterized by, The method comprises the following steps: S1: point cloud data preprocessing: filtering, downsampling and clustering of the original scanning point cloud, effectively removing noise and reducing data volume while retaining key geometric features; S2: point cloud clustering segmentation: through a parameter-adjustable region growing clustering algorithm, the preprocessed point cloud data is divided into three key parts: front disc cover surface, blade surface and wrap angle surface; S3: main weld seam trajectory extraction: based on principal component analysis to determine the intercepting plane, curve fitting is performed on the cross-sectional data of the blade surface to obtain the preliminary main weld seam trajectory; then the main weld seam trajectory is projected onto the local approximate plane of the front disc cover surface in segments to complete the accurate extraction of the main weld seam trajectory; S4: wrap angle weld seam trajectory generation: plane fitting and thickness calculation are performed on the wrap angle surface to obtain the intersection line of the plane and the adjacent local approximate plane; the intersection point of the intersection line and the main weld seam trajectory is calculated, and finally the complete wrap angle weld seam trajectory is generated in combination with the intersection line direction and the wrap angle thickness.

2. The impeller fillet weld based on point cloud data of the weld seam trajectory identification method as claimed in claim 1, characterized by, The point cloud data preprocessing in step S1 comprises: A 3D vision sensor is mounted on a robot to scan and obtain the initial three-dimensional point cloud of the workpiece, the original point cloud is filtered and voxel downsampling is performed, and a region growing clustering algorithm is used according to the structural features of the workpiece to retain the largest number of clusters.

3. The impeller fillet weld girth weld based on point cloud data based weld seam trajectory identification method as claimed in claim 1, characterized in that, The point cloud clustering segmentation in step S2 comprises: First fine clustering: the region growing method is used for the second clustering of the preprocessed data; the largest two types of data are extracted, which correspond to the front disc cover surface data and the impeller composite data respectively; at the same time, the data not classified in the clustering is extracted and recorded as connection data; Second fine clustering: the clustering parameters of the region growing algorithm are adjusted again, and the third clustering is performed on the impeller composite data obtained above, so that the wrap angle surface data and the blade surface data are successfully separated.

4. The impeller fillet weld based on point cloud data of the weld seam trajectory identification method as claimed in claim 1, characterized by, The acquisition of the intercepting plane in step S3 comprises: The data obtained in step S2 is subjected to principal component analysis, and the two principal component directions are subjected to cross multiplication operation to obtain a vector in the approximate height direction, which is used as the normal vector of the intercepting plane. At the same time, the centroid of the connection data is used as a point on the intercepting plane, so as to uniquely determine the intercepting plane.

5. The impeller fillet weld girth weld based on point cloud data based weld seam trajectory identification method as claimed in claim 1, wherein, The local projection in step S3 comprises: According to a preset grouping threshold, the preliminary trajectory is divided into multiple line ends; the centroid of each line end is used as a query point to extract the data of the front disc cover surface in the neighborhood range, and the local data points are subjected to plane fitting to obtain a local approximate plane; the line segment is projected onto the respective local approximate plane.

6. The impeller fillet weld based on point cloud data based weld seam trajectory identification method as claimed in claim 1, characterized by, The intersection point of the intersection line and the main weld seam trajectory in step S4 is the inflection point of the wrap angle weld seam and the main weld seam trajectory.

7. The impeller fillet weld based on point cloud data of the weld seam trajectory identification method as claimed in claim 3, characterized by, The impeller composite data is the unsegmented set of the wrap angle surface and the blade surface, and the connection data is the connection transition zone of the front disc cover surface and the impeller body.

Images