Method and system for processing circular weld joint based on point cloud data

Abstract

The invention discloses a method and system for processing a circular weld joint based on point cloud data, and belongs to the technical field of machine identification of weld joints, and the method comprises the following steps: collecting the point cloud data of the surface of a workpiece through a visual sensor; preprocessing the collected point cloud data; calculating a normal vector of each point in the point cloud data; segmenting the workpiece surface point cloud and the reflective interference points based on a normal vector, and filtering the interference points deviating from a preset range in the normal direction to obtain a target point cloud and a reference point cloud; projecting the target point cloud to a reference plane obtained by fitting the reference point cloud, generating a target projection point cloud, detecting boundary points of the target projection point cloud, screening candidate weld joints based on distance distribution from the boundary points to the center of mass, and extracting weld information. According to the method, the workpiece made of the material with high light reflection is treated, the finally obtained result can still keep good precision, and the quality of the workpiece is improved.

Description

Technical Field

[0001] This invention belongs to the field of machine recognition of weld seams, specifically a method and system for processing circular weld seams based on point cloud data. Background Technology

[0002] In the welding field, circular welds are very common. Currently, there are two main methods for welding circular welds: one is manual welding, which is prone to uneven welds, unattractive finishes, and poor weld quality; the other is robot-taught welding, where the worker pre-sets the welding path. Robot-taught welding can ensure aesthetically pleasing welds and good quality, but it becomes ineffective when there are assembly errors or workpiece misalignments. In recent years, scanning welding for circular welds has emerged. However, current robotic scanning methods for handling circular welds typically involve first using sensors to acquire the weld's position, or having technicians pre-set the scanning path for the robot. The weld's position is then transmitted to the robot, which completes the welding along the path. However, direct scanning is sensitive to the reflective properties of the workpiece, especially for materials like aluminum and stainless steel, which can easily lead to inaccuracies in the detected weld position. Summary of the Invention

[0003] To address the above issues, this application proposes a method and system for processing circular weld seams based on point cloud data. Specifically, a vision sensor is used to acquire point cloud data of the entire surface of the circular tube workpiece, the position of the circular weld seam is accurately extracted based on the point cloud data, and finally the weld seam position is sent to a welding robot to complete the welding.

[0004] A first aspect of the present invention: a method for processing circular weld seams based on point cloud data, comprising the following steps: S1: Collect point cloud data of the surface of the workpiece to be welded using a vision sensor; S2: Preprocess the point cloud data collected in step S1, including downsampling and statistical filtering for noise reduction; S3: Calculate the normal vector of each point in the point cloud data after preprocessing in step S2; S4: Based on the normal vector obtained in step S3, the point cloud on the workpiece surface and the reflection interference points are first classified and distinguished, and interference points whose normal directions deviate from the preset range are filtered out; finally, the target point cloud and reference point cloud are further extracted from the remaining effective point cloud through a clustering algorithm. S5: After projecting the target point cloud obtained in step S4, detect the boundary point cloud and obtain candidate weld points; S6: Perform circle fitting on the candidate weld points detected in step S5, thus completing the extraction of weld information.

[0005] The point cloud data preprocessing in step S2 includes: Statistical filtering is performed on the downsampled point cloud data. The average distance between each point and its K neighboring points is calculated, and outliers are removed based on statistical analysis.

[0006] In step S3, the normal vector is calculated using the function for calculating the normal vector from the open-source point cloud processing library PCL.

[0007] The point cloud data classification steps in step S4 include: Based on the calculated normal vectors, the point cloud is classified using the region growing algorithm in the PCL library, separating the interference points caused by reflection from the point cloud on the actual workpiece surface. Based on the statistical point cloud normal information, the normal vector of the point cloud on the workpiece surface is obtained. The world coordinate axis parallel to the normal vector is found, and an angle threshold is set to filter out points with large angles between the normal vector and the target world coordinate axis. Two sets of surface point cloud data are obtained from the workpiece surface point cloud data through a clustering algorithm. The centroids of the two sets of point cloud data are calculated. The point cloud with fewer points and whose centroid coordinates are smaller than the selected world coordinate axis is the target point cloud. The other set of point cloud data is the reference point cloud.

[0008] Step S5, obtaining candidate weld points, includes the following steps: A reference plane is obtained by performing plane fitting on the reference point cloud, and the circular weld seam formed by the circular tube is located on the plane where the reference point cloud is located. The target point cloud is projected onto the reference plane to obtain the target projected point cloud; Detect the boundary point cloud of the target's projected point cloud; The actual weld location is closer to the boundary point of the outer ring, and candidate weld points are extracted from the boundary point cloud.

[0009] Extracting weld points from the boundary point cloud includes the following steps: Calculate the centroid of the boundary point cloud; Calculate the distance from each boundary point to the centroid and save it to an array, which is regarded as a distance histogram; Using the mean of the two largest peaks in the histogram as the threshold DistThresh, points with a distance greater than this threshold are extracted as candidate weld points.

[0010] In step S6, the circle fitting uses the spatial circle model fitting algorithm from the PCL library.

[0011] Second aspect of the invention: A welding system includes a vision sensor, a processor, and a welding robot, characterized in that the processor is configured to perform the method described in any of the first aspects above and to transmit weld position information to the welding robot.

[0012] Beneficial effects: This invention directly acquires point cloud data of the workpiece surface to be welded. The scanning path is a straight line determined by the scanning start and end points, making operation simple and convenient. Even with an angle between the sensor scanning angle and the workpiece surface, accurate scanning is still possible, improving work efficiency. Using this method to process highly reflective aluminum workpieces, the final result can still be controlled within 0.7mm. For materials with low reflectivity, the accuracy can be controlled within 0.5mm, fully meeting welding requirements. Compared to using AI technology to process point cloud data and classify target point clouds, reference point clouds, and interference points, this invention reduces equipment costs, as AI technology generally requires industrial computers with dedicated NVIDIA graphics cards. Attached Figure Description

[0013] Figure 1 This is a flowchart of the algorithm for processing circular welds in this invention; Figure 2 This is a schematic diagram of the workpiece to be welded according to the present invention; Figure 3 This is a point cloud data map of the workpiece acquired using a line structured light sensor in this invention; Figure 4 This is a point cloud result image of the classification based on normal vectors according to the present invention; Figure 5 This is a result image of the filtered point cloud data of the workpiece surface according to the present invention; Figure 6 This is a point cloud result image of the boundary points detected from the target projection point cloud according to the present invention; Figure 7 This is a diagram showing the processing result obtained by circle fitting in this invention. Detailed Implementation

[0014] 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.

[0015] Example 1 A method for processing circular weld seams based on point cloud data includes the following steps: S1: Collect point cloud data of the surface of the workpiece to be welded using a vision sensor; S2: Preprocess the point cloud data collected in step S1, including downsampling and statistical filtering for noise reduction; S3: Calculate the normal vector of each point in the point cloud data after preprocessing in step S2; S4: Based on the normal vector obtained in step S3, the point cloud on the workpiece surface and the reflection interference points are first classified and distinguished, and interference points whose normal directions deviate from the preset range are filtered out; finally, the target point cloud and reference point cloud are further extracted from the remaining effective point cloud through a clustering algorithm. S5: After projecting the target point cloud obtained in step S4, detect the boundary point cloud and obtain candidate weld points; S6: Perform circle fitting on the candidate weld points detected in step S5, thus completing the extraction of weld information.

[0016] The detailed algorithm steps are explained using the circular weld of a heat exchanger tube as an example. Figure 2 The image shows the heat exchanger workpiece to be welded, specifically: S1: Data Acquisition Generally, a vision sensor completes the point cloud data acquisition of the workpiece to be welded. In this embodiment, the vision sensor used is a 3D line structured light sensor. The acquired point cloud data is as follows: Figure 3 As shown; The scanning method can also use a 3D surface structured light sensor to collect point cloud data of the workpiece surface at fixed points; S2: The point cloud data collected in step S1 includes not only the point cloud data of the workpiece surface, but also a lot of interference data. This interference data will affect the processing efficiency and accuracy. Therefore, the point cloud data needs to be preprocessed. The process is as follows: First, downsampling of point cloud data is mainly aimed at reducing the amount of data and improving processing efficiency.

[0017] Then, the downsampled point cloud data is filtered. In this embodiment, statistical filtering is used. The average distance between each point and its K neighboring points is calculated, and then statistical analysis is used to determine which points are outliers. Finally, the outliers are removed from the point cloud data. In this embodiment, the value of K is 25 and the standard deviation factor is 1.

[0018] S3: Compute the normal vector of the point cloud after preprocessing in step S2. The normal vector is calculated using the function for calculating the normal vector in the open-source point cloud processing library PCL (Point Cloud Library). S4: Classify the point cloud data, the steps are as follows: First, based on the point cloud normals calculated in step S3, the point cloud is classified using the region growing algorithm in the PCL library. This primarily involves separating the interference points caused by reflections from the point cloud on the actual workpiece surface. Figure 4 As shown.

[0019] Secondly, based on the statistical point cloud normal information, the normal vectors of the workpiece surface point cloud are basically consistent. In this embodiment, when processing the heat exchanger circular tube, the normal vectors of the workpiece surface point cloud are basically parallel to the Z-axis in the world coordinate system. Points with large angles between the normal and the Z-axis of the world coordinate system are filtered out. In this embodiment, the angle threshold is set to 20 degrees, resulting in point cloud data containing only the workpiece surface. Figure 5 As shown.

[0020] Finally, by using a clustering algorithm on the point cloud data containing only the workpiece surface, two sets of surface point cloud data were obtained. The centroids of the two sets of point cloud data were calculated. The one with fewer point clouds and a smaller Z-value in the centroid coordinates is the target point cloud (ObjectCloud), because it represents the upper surface data of the circular pipe to be welded. The other set of surface point cloud data is the reference point cloud (ReferenceCloud). Figure 5 The pink area at the top represents the target point cloud, while the blue area at the bottom represents the reference point cloud.

[0021] S5: Obtain candidate weld points. The main process is as follows: First, the circular weld formed by the circular tube is located on the plane of the reference point cloud ReferenceCloud. Therefore, the reference plane ReferencePlane is obtained by performing plane fitting on the reference point cloud ReferenceCloud. Secondly, the target point cloud ObjectCloud is projected onto the reference plane ReferencePlane to obtain the target projected point cloud ObjectProjectCloud; Then, detect the boundary point cloud (BoundaryCloud) of the target projection point cloud (ObjectProjectCloud), such as... Figure 6 As shown.

[0022] Finally, since the actual weld location is known to be closer to the outer boundary points, candidate weld points (CandidateWeldSeam) need to be extracted from the boundary point cloud (BoundaryCloud). The method is as follows: Calculate the centroid of the boundary point cloud (Boundary Cloud).

[0023] The distance from each boundary point to the centroid (Centroid) is calculated and stored in an array called DistanceArray. DistanceArray is treated as a histogram, showing two distinct peaks. The points at the larger peaks represent the more concentrated points on the outer edge. The average of the two peaks is used as the threshold DistThresh. Points in the point cloud exceeding the DistThresh threshold are considered candidate weld seams (CandidateWeldSeam). Figure 7The green dots.

[0024] S6: Perform circle fitting on the candidate weld points (CandidateWeldSeam) obtained in S5, such as... Figure 7 The red curve in the middle uses a spatial circle model fitting method from the PCL library.

[0025] Example 2 Another embodiment of this application proposes a welding system including a vision sensor, a processor, and a welding robot. The processor is configured to implement the steps of any of the methods described in Embodiment 1 of this application and to transmit weld position information to the welding robot.

Claims

1. A method for processing circular weld seams based on point cloud data, characterized in that, Includes the following steps: S1: Collect point cloud data of the surface of the workpiece to be welded using a vision sensor; S2: Preprocess the point cloud data collected in step S1, including downsampling and statistical filtering for noise reduction; S3: Calculate the normal vector of each point in the point cloud data after preprocessing in step S2; S4: Based on the normal vector obtained in step S3, the point cloud on the workpiece surface and the reflection interference points are first classified and distinguished, and interference points whose normal directions deviate from the preset range are filtered out; finally, the target point cloud and reference point cloud are further extracted from the remaining effective point cloud through a clustering algorithm. S5: After projecting the target point cloud obtained in step S4, detect the boundary point cloud and obtain candidate weld points; S6: Perform circle fitting on the candidate weld points detected in step S5, thus completing the extraction of weld information.

2. The method for processing circular weld seams based on point cloud data as described in claim 1, characterized in that, The point cloud data preprocessing in step S2 includes: Statistical filtering is performed on the downsampled point cloud data. The average distance between each point and its K neighboring points is calculated, and outliers are removed based on statistical analysis.

3. The method for processing circular weld seams based on point cloud data as described in claim 1, characterized in that, In step S3, the normal vector is calculated using the function for calculating the normal vector from the open-source point cloud processing library PCL.

4. The method for processing circular weld seams based on point cloud data as described in claim 1, characterized in that, The point cloud data classification steps in step S4 include: Based on the calculated normal vectors, the point cloud is classified using the region growing algorithm in the PCL library, separating the interference points caused by reflection from the point cloud on the actual workpiece surface. Based on the statistical point cloud normal information, the normal vector of the point cloud on the workpiece surface is obtained. The world coordinate axis parallel to the normal vector is found, and an angle threshold is set to filter out points with large angles between the normal vector and the target world coordinate axis. Two sets of surface point cloud data are obtained from the workpiece surface point cloud data through a clustering algorithm. The centroids of the two sets of point cloud data are calculated. The point cloud with fewer points and whose centroid coordinates are smaller than the selected world coordinate axis is the target point cloud. The other set of point cloud data is the reference point cloud.

5. The method for processing circular weld seams based on point cloud data as described in claim 1, characterized in that, Step S5, obtaining candidate weld points, includes the following steps: A reference plane is obtained by performing plane fitting on the reference point cloud, and the circular weld seam formed by the circular tube is located on the plane where the reference point cloud is located. The target point cloud is projected onto the reference plane to obtain the target projected point cloud; Detect the boundary point cloud of the target's projected point cloud; Extract candidate weld points from the boundary point cloud.

6. The method for processing circular weld seams based on point cloud data as described in claim 5, characterized in that, Extracting candidate weld points from the boundary point cloud includes the following steps: Calculate the centroid of the boundary point cloud; Calculate the distance from each boundary point to the centroid and save it to an array, which is regarded as a distance histogram; Using the mean of the two largest peaks in the histogram as the threshold DistThresh, point clouds with a distance greater than this threshold are extracted as candidate weld points.

7. The method for processing circular welds based on point cloud data as described in claim 1, characterized in that, In step S6, the circle fitting uses the spatial circle model fitting algorithm from the PCL library.

8. A welding system comprising a vision sensor, a processor, and a welding robot, characterized in that, The processor is configured to perform the method described in any one of claims 1-7 and transmit weld position information to the welding robot.

Images