Method and device for detecting dead angles at intersections of grid-shaped welded components

Abstract

This invention relates to the fields of welding automation and 3D machine vision technology, providing a method and apparatus for detecting intersection dead angles in crisscross welded parts. The method includes: extracting multiple reference planes from an acquired 3D point cloud; using preset multi-plane joint distance constraints to screen candidate intersection dead angles that are close to multiple planes from the point cloud; combining this with centroid fitting of the clustering of weld line segment endpoints to obtain weld dead angle detection points; and finally, comprehensively obtaining the intersection dead angle detection points. The method provided by this invention overcomes the inefficiency and errors of manual visual positioning, and overcomes the shortcomings of traditional 2D vision, such as loss of depth information and susceptibility to environmental interference. Through a dual mechanism of joint distance constraints and endpoint clustering in 3D space, it achieves anti-interference, high-precision automatic detection of narrow, enclosed dead angles, greatly improving the consistency and product yield of robotic automated welding operations.

Description

Technical Field

[0001] This invention relates to the fields of welding automation and three-dimensional machine vision technology, and in particular to a method and device for detecting dead angles at intersections of grid-shaped weldments. Background Technology

[0002] In the fields of steel structure welding, grid-shaped frame welding, and multi-faceted bevel welding, the three-plane intersection dead angle is the most difficult welding location and the most prone to defects such as incomplete penetration, slag inclusion, and porosity. Accurate identification of this dead angle is a core prerequisite technology for achieving automated robotic welding, grinding, and root cleaning. Currently, there are two main methods for dead angle detection in the industry: The first is manual visual positioning combined with ruler measurement. Operators rely on visual observation of the grid-shaped weldment structure, manually marking the three-plane intersection dead angle positions with a tape measure and calipers, and then manually teaching the robot's movement points into the system. This method relies entirely on human experience and lacks automated perception capabilities. The second method is two-dimensional visual contour intersection detection technology. This uses an industrial camera to acquire two-dimensional images, determines weld intersection points through edge detection, line fitting, and line intersection calculation, and then maps the two-dimensional pixel intersections to three-dimensional coordinates to achieve coarse positioning of the dead angle. This type of solution is based solely on the geometric intersection judgment of two-dimensional contours.

[0003] However, manual inspection methods are extremely inefficient, inconsistent, and dangerous, while two-dimensional visual intersection detection suffers from problems such as missing dimensions, extremely poor accuracy, and a high risk of false detection. Therefore, none of the aforementioned existing technologies are suitable for the high-precision inspection of the three-plane closed dead corners of grid-shaped weldments. They suffer from the low efficiency and poor consistency of manual dead corner inspection, and the problems of missing dimensions and susceptibility to reflection and spatter interference in two-dimensional visual intersection detection, resulting in low accuracy in dead corner detection. Summary of the Invention

[0004] This invention provides a method and apparatus for detecting dead corners at intersections of grid-shaped welded parts, which solves the problems of low efficiency and poor consistency in manual detection of dead corners in the prior art, as well as the lack of dimension and susceptibility to interference from reflection and spatter in two-dimensional visual intersection detection, which leads to low accuracy of dead corner detection.

[0005] This invention provides a method for detecting dead angles at intersections of grid-shaped welded components, comprising: Obtain the 3D point cloud of the grid-shaped weldment to be inspected; The three-dimensional point cloud to be detected is subjected to planar extraction to obtain multiple reference planes corresponding to the grid-shaped weldment; For each spatial point in the three-dimensional point cloud to be detected, the spatial distance from each spatial point to the multiple reference planes is calculated. Based on multiple spatial distances corresponding to each spatial point and preset multi-plane joint distance constraints, candidate intersection dead angle points that are simultaneously close to the multiple reference planes are selected from the spatial points. Clustering and filtering are performed on all endpoints of the weld line segments corresponding to the multiple reference planes to obtain endpoint intersection clusters. The centroid of the endpoint intersection clusters is calculated to obtain the weld dead angle detection points. By integrating the candidate intersection dead angle points and the weld dead angle detection points, the intersection dead angle detection points of the grid-shaped weldment are obtained.

[0006] According to the present invention, a method for detecting intersection dead angles of a grid-shaped weldment is provided, wherein, based on multiple spatial distances corresponding to each spatial point and preset multi-plane joint distance constraints, candidate intersection dead angle points that are simultaneously close to the multiple reference planes are selected from the spatial points, including: For each spatial point, the plurality of spatial distances are compared with a single-plane distance threshold. If all the spatial distances satisfy the single-plane distance threshold, the corresponding spatial point is determined as the candidate point of the intersection dead angle.

[0007] According to the present invention, a method for detecting dead angles at intersections of grid-shaped weldments is provided, the method further comprising: If there are no spatial points that satisfy the single-plane distance threshold, the multi-plane joint distance constraint is relaxed to the single-plane distance relaxation threshold. For each spatial point, the multiple spatial distances are compared with the single-plane distance relaxation threshold. If all the multiple spatial distances meet the single-plane distance relaxation threshold, the corresponding spatial point is determined as the intersection dead angle candidate point.

[0008] According to the present invention, a method for detecting intersection dead angles of a grid-shaped weldment is provided, wherein the step of extracting a plane from the three-dimensional point cloud to be detected to obtain multiple reference planes corresponding to the grid-shaped weldment includes: The three-dimensional point cloud to be detected is subjected to multiple iterative plane segmentation to obtain multiple initial planes; For any two initial planes among the plurality of initial planes, calculate the angle between the normal vectors of the two initial planes, and the minimum Euclidean distance between the point clouds contained in the two initial planes; If the angle between the normal vectors and the minimum Euclidean distance both satisfy the preset coplanarity condition, it is determined that the two initial planes are the same physical plane, and the point clouds of the two initial planes are merged to refit and obtain the merged plane. Based on all merged planes and unmerged initial planes, the multiple reference planes corresponding to the grid-shaped weldment are obtained.

[0009] According to the present invention, a method for detecting intersection dead angles of a grid-shaped weldment is provided, wherein the three-dimensional point cloud to be detected is subjected to multiple iterative planar segmentation to obtain multiple initial planes, including: Using the 3D point cloud to be detected as the current point cloud to be segmented, perform the following planar extraction operation: Perform planar segmentation on the current point cloud to be segmented to obtain the current segmented inner point cloud; Calculate the distance between each point in the inner point cloud, perform cluster analysis on the inner point cloud based on distance connectivity, and take the connected domain point cloud with the largest number of connected points as the main plane for the current extraction. If the principal plane is successfully extracted, the remaining point cloud in the current point cloud to be segmented after removing the principal plane is updated to the current point cloud to be segmented; if the principal plane is not successfully extracted, the number of consecutive failures is accumulated. If the number of consecutive failures does not reach the preset threshold, return to perform the planar extraction operation; If the number of consecutive failures reaches the preset threshold, the iteration ends, and the main planes successfully extracted in each iteration are used as the multiple initial planes.

[0010] According to the present invention, a method for detecting dead angles at intersections in a grid-shaped weldment includes: clustering and screening all endpoints of weld line segments corresponding to the plurality of reference planes to obtain endpoint intersection clusters; calculating the centroid of the endpoint intersection clusters to obtain weld dead angle detection points; and including: Obtain all endpoints of the weld line segment; Based on a preset clustering radius, spatial distance clustering is performed on all endpoints of the weld line segment to filter out the endpoint intersection clusters containing multiple endpoints; Calculate the average spatial coordinates of multiple endpoints within the endpoint intersection cluster, and use the average spatial coordinates as the centroid to obtain the weld dead angle detection point.

[0011] According to the present invention, a method for detecting intersection dead angles of a grid-shaped weldment is provided, wherein obtaining the three-dimensional point cloud of the grid-shaped weldment to be detected includes: Obtain the initial three-dimensional point cloud of the grid-shaped weldment to be inspected; A region of interest is constructed based on a preset spatial coordinate range. Background point clouds located outside the region of interest in the initial 3D point cloud are removed to obtain the retained point cloud. The retained point cloud is uniformly downsampled according to a preset voxel size to obtain a sampled point cloud; The average neighborhood distance of each point in the sampled point cloud is calculated. If the average neighborhood distance of the target point does not meet the preset standard deviation multiple constraint, the target point is determined to be an outlier noise point and is removed to obtain the three-dimensional point cloud to be detected.

[0012] According to the present invention, a method for detecting dead angles at intersections of grid-shaped weldments is provided, wherein the grid-shaped weldment to be detected includes a three-plane grid-shaped weldment.

[0013] The present invention also provides a device for detecting dead angles at intersections of grid-shaped welded parts, comprising: The acquisition unit acquires the three-dimensional point cloud of the grid-shaped weldment to be inspected; The planar extraction unit performs planar extraction on the three-dimensional point cloud to be detected, and obtains multiple reference planes corresponding to the grid-shaped weldment; The distance calculation unit calculates the spatial distance from each spatial point in the three-dimensional point cloud to be detected to the multiple reference planes. The blind spot candidate point screening unit filters out intersection blind spot candidate points that are simultaneously close to the multiple reference planes from the spatial points based on multiple spatial distances corresponding to each spatial point and preset multi-plane joint distance constraints. The weld inspection point calculation unit performs clustering and filtering on all endpoints of the weld line segments corresponding to the multiple reference planes to obtain endpoint intersection clusters, calculates the centroid of the endpoint intersection clusters, and obtains weld dead angle inspection points. The integration unit integrates the candidate intersection dead angle points and the weld dead angle detection points to obtain the intersection dead angle detection points of the grid-shaped weldment.

[0014] According to the present invention, a dead angle detection device for intersecting weldments in a grid pattern is provided, wherein the dead angle candidate point screening unit is specifically used for: For each spatial point, the plurality of spatial distances are compared with a single-plane distance threshold. If all the spatial distances satisfy the single-plane distance threshold, the corresponding spatial point is determined as the candidate point of the intersection dead angle.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the intersection dead angle detection method for the grid-shaped weldment as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intersection dead angle detection method for grid-shaped weldments as described above.

[0017] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the intersection dead angle detection method for the grid-shaped weldment as described above.

[0018] The present invention provides a method and apparatus for detecting intersection dead angles of grid-shaped welded parts. By extracting multiple reference planes from the acquired three-dimensional point cloud, candidate intersection dead angles that are close to multiple planes are selected from the point cloud using preset multi-plane joint distance constraints. The detection points of weld dead angles are obtained by fitting the centroid of clustering the endpoints of weld line segments. Finally, the intersection dead angle detection points are obtained by combining the results. This method eliminates the inefficiency and error of manual visual positioning and overcomes the defects of traditional two-dimensional vision, such as loss of depth information and susceptibility to environmental interference. Through the dual mechanism of joint distance constraints and endpoint clustering in three-dimensional space, it achieves anti-interference and high-precision automatic detection of narrow and closed dead angles, which greatly improves the consistency and product yield of robotic automated welding operations. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the method for detecting dead angles at intersections of grid-shaped welded parts provided by the present invention. Figure 2 This is a schematic diagram of the intersection dead angle detection device for grid-shaped welded parts provided by the present invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] It should be understood that, in the description of the embodiments of the present invention, "multiple" means two or more, unless otherwise explicitly specified. Furthermore, all actions involving the acquisition of signals, information, or data in this invention are performed in accordance with the applicable data protection laws and regulations of the locality and with permission granted by the owner of the relevant device.

[0023] It should be noted that the narrow, dimly lit, grid-shaped blind spots result in significant errors and time-consuming manual positioning, making them unsuitable for automated production lines. Furthermore, the manual marking of points varies from person to person, leading to poor consistency in the robot's welding trajectory and a high defect rate in mass production. Additionally, two-dimensional visual intersection detection suffers from a lack of dimension, extremely poor accuracy, and a high risk of false positives. Moreover, the loss of depth information in two-dimensional images makes it impossible to distinguish between spatially skewed lines and true three-dimensional intersection blind spots; surface reflections, weld slag, and fumes can cause contour breaks, leading to intersection calculation offsets, missed detections, and false positives, completely failing to meet the requirements for precise welding blind spot positioning.

[0024] Therefore, in view of the problems of low efficiency and poor consistency of manual detection of dead corners in the existing technology, and the problems of dimension loss, susceptibility to reflection and splash interference in two-dimensional visual intersection detection, which result in low accuracy of dead corner detection, the present invention provides a method for detecting intersection dead corners of grid-shaped weldments, so as to achieve high-precision and automated positioning of multi-plane closed dead corners. This method can be applied to corresponding automated detection equipment or control systems. Figure 1 This is a flowchart illustrating the method for detecting dead angles at intersections of grid-shaped weldments provided by the present invention, as shown below. Figure 1 As shown, the method includes: Step 110: Obtain the three-dimensional point cloud of the grid-shaped weldment to be inspected.

[0025] Here, the "grid-shaped weldment to be inspected" refers to a workpiece in a multi-faceted welding scenario, such as a steel structure or a grid-shaped frame, where multiple planes intersect and form a closed spatial dead angle. The weldment to be inspected is a three-plane grid-shaped weldment or a four-plane grid-shaped weldment, obtained by welding a base plate and multiple vertical plates. Furthermore, the "3D point cloud to be inspected" refers to a 3D dataset acquired through acquisition devices such as 3D vision sensors, which reflects the surface spatial structure and depth information of the grid-shaped weldment to be inspected.

[0026] Specifically, firstly, the grid-shaped weldment to be inspected can be scanned and acquired using equipment such as a 3D laser scanner or 3D camera, and the acquired point cloud data (PCD) file of the workpiece can be loaded. To improve the accuracy and efficiency of subsequent processing, after acquiring the initial point cloud data, preprocessing operations can be performed on the point cloud data as needed, such as spatial region cropping to remove background interference, or removing discrete noise points such as welding spatter and dust through downsampling and filtering operations, thereby obtaining a clean and structured 3D point cloud to be inspected.

[0027] Understandably, by acquiring 3D point clouds containing depth information, we can overcome the defect of 2D images losing spatial depth and provide a reliable data foundation for subsequent precise 3D spatial positioning.

[0028] Step 120: Perform planar extraction on the three-dimensional point cloud to be detected to obtain multiple reference planes corresponding to the grid-shaped weldment.

[0029] The multiple reference planes here refer to the main physical planes that constitute the spatial intersection structure of the grid-shaped weldment to be inspected, such as the three main solid plate surfaces corresponding to the dead angle scenario of the three plane intersection.

[0030] Specifically, a planar segmentation algorithm can be used to mathematically fit and extract the 3D point cloud to be detected. For example, algorithms such as Random Sample Consensus (RANSAC) can be used to iteratively find a planar model that contains the most interior points in the 3D point cloud to be detected. In this process, cluster analysis and other connected component filtering methods can be used to filter out fragmented and scattered regions, retaining only the real principal planes with a large number of points and good connectivity, and finally outputting multiple reference planes corresponding to the grid-shaped weldment.

[0031] Step 130: For each spatial point in the three-dimensional point cloud to be detected, calculate the spatial distance from each spatial point to the multiple reference planes.

[0032] The spatial distance here refers to the shortest vertical Euclidean distance from discrete points in the point cloud to each fitted reference plane. This distance reflects the degree to which a specific spatial point deviates from each entity plane.

[0033] Specifically, after obtaining the mathematical equations for multiple reference planes, the various spatial points in the 3D point cloud to be detected can be traversed. To accelerate the retrieval process, a spatial partitioning data structure, such as a KdTree, can be constructed for the extracted reference planes, thereby efficiently calculating the minimum spatial distance from each spatial point to these multiple reference planes. In this step, each spatial point will obtain multiple distance values ​​corresponding to each reference plane.

[0034] Step 140: Based on the multiple spatial distances corresponding to each spatial point and the preset multi-plane joint distance constraints, select candidate intersection dead angle points that are simultaneously close to the multiple reference planes from the spatial points.

[0035] The multi-plane joint distance constraint here refers to the distance threshold rule used to determine whether a spatial point is truly located at the intersection of multiple planes. Here, the intersection dead zone candidate point refers to a three-dimensional spatial point that is simultaneously adjacent to these multiple reference planes and is highly likely to be a real physical intersection dead zone.

[0036] Specifically, the multiple spatial distances corresponding to each spatial point calculated above can be substituted into the preset multi-plane joint distance constraints for judgment. Only when the distances from a spatial point to all relevant reference planes meet the set proximity conditions will it be extracted as a candidate point for intersection dead angle.

[0037] It should be noted that this multi-plane joint constraint mechanism directly utilizes the three-dimensional geometric feature that the dead angle of the grid will inevitably be close to multiple solid planes at the same time, effectively eliminating the misjudgment caused by simply relying on the intersection of a single plane or two planes, and greatly improving the accuracy and anti-interference ability of candidate point selection.

[0038] Step 150: Cluster and filter all endpoints of the weld line segments corresponding to the multiple reference planes to obtain endpoint intersection clusters, calculate the centroid of the endpoint intersection clusters, and obtain the weld dead angle detection points.

[0039] Here, weld seam segments refer to linear structures extracted or detected in the intersection area of ​​the reference plane. All endpoints refer to the coordinates of the spatially extending ends of these weld seam segments. An endpoint cluster refers to a set of endpoints that are spatially close to each other and belong to the same intersection area; the centroid refers to the geometric center or weighted average center of this endpoint set in spatial coordinates. The weld dead angle detection point is the fitted point of the dead angle center calculated from the endpoint features.

[0040] Specifically, the coordinates of all endpoints of each weld line segment on the weldment can be extracted first. Then, distance clustering analysis is performed on these endpoints based on a specific spatial radius range to find the set of endpoints that are spatially clustered together, i.e., the endpoint intersection cluster is selected. Next, the average spatial coordinates of multiple endpoints contained in the cluster are calculated to determine its centroid position, which is used as the weld dead angle detection point.

[0041] It should be noted that by using endpoint clustering and centroid fitting, sub-pixel-level accurate fitting of dead-angle positions can be achieved, effectively avoiding local noise errors that may be caused by a single endpoint.

[0042] Step 160: Integrate the candidate intersection dead angle points and the weld dead angle detection points to obtain the intersection dead angle detection points of the grid-shaped weldment.

[0043] The intersection blind spot detection point here is the final output, a high-precision three-dimensional coordinate point that can be used by the robot to perform automatic welding or grinding operations.

[0044] Specifically, candidate intersection dead angles selected based on multi-plane joint distance constraints can be comprehensively compared, verified, or fused with weld dead angle detection points obtained based on endpoint clustering centroid fitting. For example, these two types of points can be used together as the final set of detected dead angles. Finally, the obtained intersection dead angle detection points can be 3D visualized and labeled or directly sent to the execution agency to complete the entire inspection process.

[0045] The method provided in this invention extracts multiple reference planes from the acquired 3D point cloud, selects candidate points for intersection dead angles that are close to multiple planes from the point cloud using preset multi-plane joint distance constraints, and obtains weld dead angle detection points by fitting the centroids of clustering the endpoints of weld line segments. Finally, the intersection dead angle detection points are obtained by combining the results. This method eliminates the inefficiency and error of manual visual positioning, overcomes the shortcomings of traditional 2D vision such as loss of depth information and susceptibility to environmental interference, and achieves anti-interference and high-precision automatic detection of narrow closed dead angles through the dual mechanism of joint distance constraints and endpoint clustering in 3D space. This greatly improves the consistency and product yield of robotic automated welding operations.

[0046] Based on any of the above embodiments, step 140 includes: For each spatial point, the plurality of spatial distances are compared with a single-plane distance threshold. If all the spatial distances satisfy the single-plane distance threshold, the corresponding spatial point is determined as the candidate point of the intersection dead angle.

[0047] The single-plane distance threshold here refers to a pre-set, strict distance limit value used to measure whether a single spatial point is close enough to a single reference plane.

[0048] Specifically, after obtaining the shortest vertical distances from any spatial point to multiple reference planes, these distance values ​​can be determined one by one. For example, in a three-plane grid-shaped welding scenario, three spatial distances will be obtained. These three spatial distances will be compared with the set single-plane distance threshold to examine the tightness of the spatial point's fit to the corresponding physical plane in each dimension.

[0049] A spatial point is considered to be in close contact with multiple physical surfaces in space only when the spatial distance from the point to all relevant reference planes is less than or equal to the strict single-plane distance threshold. In this case, the corresponding spatial point satisfying this stringent condition is extracted and identified as a candidate point for intersection dead zones. If any of the spatial distances exceeds the threshold, the point is directly discarded.

[0050] In one embodiment, candidate intersection dead zones can be obtained by filtering using the following multi-plane joint distance constraint bar, as shown in the following formula: ; In the formula, This represents the spatial distance between any spatial point and each reference plane. There are a total of 3 reference planes. 2.0mm is the single-plane distance threshold. As can be seen from the formula, a spatial point is considered as a candidate point for intersection dead angle only when the three spatial distances corresponding to any spatial point are all less than the preset single-plane distance threshold of 2.0mm, and the sum of the three spatial distances is less than the preset distance threshold of 5.0mm.

[0051] The method provided in this invention rigorously compares the distances from spatial points to multiple reference planes with single-plane distance thresholds to ensure that the selected candidate points are simultaneously attached to all intersecting planes. This effectively eliminates wandering interference points in non-intersecting areas, ensuring the rigor of blind spot detection and extremely high spatial positioning accuracy.

[0052] Based on any of the above embodiments, the method further includes: If there are no spatial points that satisfy the single-plane distance threshold, the multi-plane joint distance constraint is relaxed to the single-plane distance relaxation threshold. For each spatial point, the multiple spatial distances are compared with the single-plane distance relaxation threshold. If all the multiple spatial distances meet the single-plane distance relaxation threshold, the corresponding spatial point is determined as the intersection dead angle candidate point.

[0053] The single-plane distance relaxation threshold here refers to a spare distance limit value that is larger and has relatively looser constraints than the initially set strict distance threshold value.

[0054] Specifically, after the initial rigorous screening of all spatial points, the system automatically checks whether the selected set of candidate intersection blind spots is empty. If no spatial points satisfying the aforementioned strict single-plane distance threshold are found, it indicates that the current working condition is relatively complex, such as significant local deformation of the weldment leading to imperfections in the intersection area, or blind spots in the scanned point cloud. In this case, an adaptive fallback mechanism can be automatically triggered to dynamically adjust and relax the originally stringent multi-plane joint distance constraints to a relaxed single-plane distance threshold.

[0055] After the trigger threshold is relaxed, each spatial point in the 3D point cloud to be detected is traversed again, and a secondary search is performed using the adjusted single-plane distance relaxation threshold. As long as the spatial distances from a spatial point to each reference plane are not greater than the relaxation threshold, the corresponding spatial point can be identified as a candidate point for intersection dead angles, thus ensuring stable output of candidate results under any working condition.

[0056] The method provided in this invention introduces a dual-threshold adaptive fallback mechanism, which automatically relaxes the distance constraint conditions for secondary retrieval when there are no candidate points. This cleverly solves the problem of missed detection caused by local missing points in the point cloud or workpiece deformation under harsh industrial conditions, and greatly improves the robustness and detection rate of the algorithm in complex scenarios.

[0057] Based on any of the above embodiments, in order to solve the problem that multiple plane segmentation easily fragments the same physical plane, step 120 includes: The three-dimensional point cloud to be detected is subjected to multiple iterative plane segmentation to obtain multiple initial planes; For any two initial planes among the plurality of initial planes, calculate the angle between the normal vectors of the two initial planes, and the minimum Euclidean distance between the point clouds contained in the two initial planes; If the angle between the normal vectors and the minimum Euclidean distance both satisfy the preset coplanarity condition, it is determined that the two initial planes are the same physical plane, and the point clouds of the two initial planes are merged to refit and obtain the merged plane. Based on all merged planes and unmerged initial planes, the multiple reference planes corresponding to the grid-shaped weldment are obtained.

[0058] Specifically, firstly, the 3D point cloud to be detected is subjected to multiple iterative planar segmentation to obtain multiple initial planes. Here, the initial plane refers to the original planar model extracted independently by a single segmentation algorithm in the point cloud segmentation process, without deduplication and merging processing.

[0059] In practice, an iterative segmentation algorithm can be used to repeatedly extract the principal plane from the remaining point cloud in the 3D point cloud to be detected. Each segmentation can extract a planar region and its corresponding point cloud data. After a finite number of iterations, multiple initial planes containing various surface features of the workpiece can be obtained.

[0060] Then, for any two initial planes among the multiple initial planes, calculate the angle between the normal vectors of the two initial planes, and the minimum Euclidean distance between the point clouds contained in the two initial planes.

[0061] The angle between the normal vectors here refers to the angle of intersection of the normal vectors of two planes in three-dimensional space, used to measure the angular similarity or parallelism of the two planes. The minimum Euclidean distance refers to the shortest straight-line absolute distance between the point cloud sets belonging to these two planes, used to measure the degree of spatial fit between the two planes.

[0062] In practice, to identify whether the same physical surface exists within these initial planes that have been fragmented by the algorithm, each newly segmented plane can be compared pairwise with the previously extracted planes. The angle between the normal vectors is obtained by calculating the inverse cosine of the dot product between their respective plane equation normal vectors. Simultaneously, the minimum Euclidean distance in space is calculated by determining the nearest distance between the point cloud data points contained within each of the two initial planes.

[0063] Furthermore, if the angle between the normal vectors and the minimum Euclidean distance both satisfy the preset coplanarity condition, the two initial planes are determined to be the same physical plane, and the point clouds of the two initial planes are merged to refit and obtain the merged plane.

[0064] The pre-defined coplanarity condition here refers to the dual criteria of similar angles and spatial fit. The merged plane here refers to the unified plane equation generated by integrating discrete point clouds belonging to the same physical surface and recalculating them.

[0065] In practice, the calculated angle between the normal vectors is compared with an angle threshold, and the minimum Euclidean distance is compared with a distance threshold. When the angle between the normal vectors is extremely small and the minimum Euclidean distance approaches zero, that is, when the preset coplanarity condition is met simultaneously, it can be determined that the two initial planes physically belong to the same weldment reference plane.

[0066] For example, it can be expressed by the following formula, as shown in the following equation: ; ; In the formula, Represents the angle between normal vectors cosine value, and Represents the normal vectors of the two initial planes; Indicates the minimum Euclidean distance; , These represent the points corresponding to the two initial planes used to calculate the minimum Euclidean distance. As shown in the formula, the angle between the normal vectors... The cosine value is greater than And the minimum Euclidean distance is less than In this case, the two initial planes are merged to refit and obtain a merged plane.

[0067] Then, the point cloud data in the two initial planes are stitched together and merged. Using the merged massive point cloud set, the plane coefficients are re-solved using mathematical fitting methods such as the least squares method, thereby obtaining a more accurate and complete merged plane.

[0068] Finally, based on all merged planes and unmerged initial planes, multiple reference planes corresponding to the grid-shaped weldment are obtained.

[0069] In practice, after completing comprehensive similarity judgment and automatic merging and deduplication, invalid and scattered small planes are filtered out. Finally, the large-sized merged planes and the original large-sized initial planes that were complete but not merged are used as multiple reference planes that truly correspond to the grid-shaped weldment.

[0070] The method provided in this invention uses the angle between normal vectors and the minimum Euclidean distance as dual criteria to determine the similarity of the initial planes generated by multiple iterations of segmentation and to automatically merge and remove duplicates. This solves the problem of multiple fragments of the same real plane being incorrectly split due to splash occlusion or algorithm iteration, ensuring that the final output reference planes are non-repetitive and highly restore the real workpiece surface, providing an extremely reliable and unified three-dimensional geometric reference for subsequent dead angle distance calculation.

[0071] Based on any of the above embodiments, the three-dimensional point cloud to be detected is subjected to multiple iterative plane segmentation to obtain multiple initial planes, including: Using the 3D point cloud to be detected as the current point cloud to be segmented, perform the following planar extraction operation: Perform planar segmentation on the current point cloud to be segmented to obtain the current segmented inner point cloud; Calculate the distance between each point in the inner point cloud, perform cluster analysis on the inner point cloud based on distance connectivity, and take the connected domain point cloud with the largest number of connected points as the main plane for the current extraction. If the main plane is successfully extracted, the remaining point cloud in the current point cloud to be segmented after removing the main plane is updated to the current point cloud to be segmented; if the main plane is not successfully extracted, the number of consecutive failures is accumulated. If the number of consecutive failures does not reach the preset threshold, return to perform the planar extraction operation; If the number of consecutive failures reaches the preset threshold, the iteration ends, and the main planes successfully extracted in each iteration are used as the multiple initial planes.

[0072] Specifically, the 3D point cloud to be detected is used as the current point cloud to be segmented, and the following planar extraction operation is performed: First, the current point cloud to be segmented is subjected to planar segmentation to obtain the inner point cloud of the current sub-segmentation. Here, the current point cloud to be segmented refers to the set of remaining point clouds that have not yet been successfully extracted into the main plane at the beginning of each iteration. The inner point cloud refers to the set of point clouds that, in the current sub-plane fitting algorithm, conform to the mathematical model of the fitted plane and its distance tolerance range.

[0073] In practice, during the initialization phase, the complete 3D point cloud to be detected is input, and the iterative planar extraction operation is initiated. In a single extraction operation, algorithms such as random sampling consistency can be used to fit the current point cloud to be segmented, calculate and extract the coplanar point cloud that meets the current segmentation distance threshold.

[0074] Then, the distance between points in the inner point cloud is calculated, and cluster analysis is performed on the inner point cloud based on distance connectivity. The connected component point cloud with the largest number of connected points is selected as the principal plane for the current extraction. Here, the connected component point cloud refers to a set of points that are spatially close to each other, with a distance less than a set threshold, and clustered together.

[0075] In practice, considering that the extracted interior point cloud may contain coplanar noise distributed in different spatial locations, it is necessary to calculate the spatial distance between each discrete point in the interior point cloud. Subsequently, based on the Euclidean clustering algorithm, distance connectivity analysis is performed on these points, dividing the clustered points into the same connected domain, thereby filtering out fragmented and scattered regions, and retaining only the largest connected domain point cloud with the most points and the best connectivity, which is then extracted as the true principal plane.

[0076] Furthermore, if the principal plane is successfully extracted, the remaining point cloud after removing the principal plane is updated to the current point cloud to be segmented. If the principal plane is not successfully extracted, the number of consecutive failures is accumulated. The number of consecutive failures here is used to record the number of times the plane extraction algorithm has failed to effectively extract a principal plane that meets the requirements, and is an important parameter for controlling the termination of the iteration.

[0077] In practice, if a main plane with good connectivity is successfully extracted in the current round, the points contained in the main plane are removed from the current point cloud to be segmented, and the remaining point cloud is used as the new input for the next iteration; if a valid main plane is not extracted in the current round due to the point cloud being too sparse or lacking obvious planar features, the number of consecutive failures is incremented by one.

[0078] If the number of consecutive failures does not reach the preset threshold, return to the plane extraction operation. If the number of consecutive failures reaches the preset threshold, end the iteration, and use the previously successfully extracted principal planes as multiple initial planes. The preset threshold here refers to the maximum allowable consecutive failure tolerance limit set to prevent the algorithm from getting stuck in an infinite loop.

[0079] In practice, at the end of each loop, the number of consecutive failures is checked. If the number of consecutive failures does not reach a preset threshold (e.g., less than a set value), the updated remaining point cloud is used to return to the process of plane segmentation and connected component analysis. Once the cumulative number of consecutive failures reaches the preset threshold, it indicates that there are no longer any valid large-size planar structures in the remaining point cloud, and the iteration loop ends immediately. Finally, all successfully extracted principal planes during the entire iteration process are summarized as multiple initial planes for output. Here, any of the extracted initial planes can be represented by the following general plane equation. ,in, Represents the three-dimensional coordinates of any point. Represents the coefficient.

[0080] The method provided in this invention extracts the principal plane through multiple iterations and combines it with connected component clustering for filtering. It also terminates the iteration with a constraint mechanism based on the number of consecutive failures. This method overcomes the defect of traditional single-segmentation that easily misses planes. It not only eliminates scattered noisy planes but also avoids invalid segmentation iterations, thus significantly improving the stability and extraction accuracy of the plane segmentation algorithm.

[0081] To further pinpoint the dead angle via the weld endpoint, based on any of the above embodiments, step 150 includes: Obtain all endpoints of the weld line segment; Based on a preset clustering radius, spatial distance clustering is performed on all endpoints of the weld line segment to filter out the endpoint intersection clusters containing multiple endpoints; Calculate the average spatial coordinates of multiple endpoints within the endpoint intersection cluster, and use the average spatial coordinates as the centroid to obtain the weld dead angle detection point.

[0082] Specifically, first, obtain all endpoints of the weld line segment.

[0083] In practice, edge extraction or line segment detection algorithms can be used to process the features of the intersecting areas of each reference plane, extract each weld line segment, and obtain the spatial coordinates of the start and end points of these weld line segments in three-dimensional space, which are then summarized into a set of all endpoints.

[0084] Then, based on a preset clustering radius, spatial distance clustering is performed on all endpoints of the weld line segment to filter out endpoint intersection clusters containing multiple endpoints. Here, the preset clustering radius refers to the distance range threshold used to define which endpoints belong to the same spatial clustering area.

[0085] In practical implementation, a Euclidean distance clustering algorithm can be used, with a preset cluster radius as the search range, to perform a spatial neighborhood search on all aggregated endpoints. Endpoints whose spatial distance is within the cluster radius are grouped into one class, thereby accurately filtering out endpoint clusters containing multiple endpoints that intersect in complex intersection regions.

[0086] Then, the average spatial coordinates of multiple endpoints within the endpoint intersection cluster are calculated, and the average spatial coordinates are used as the centroid to obtain the weld dead angle detection point.

[0087] In practice, after identifying an intersection cluster containing multiple endpoints, the centroid coordinates can be calculated by mathematically averaging the coordinates of the 3D points within the cluster. For example, this can be achieved using the following formula, as shown below: ; In the formula, Indicates the coordinates of the centroid; This indicates multiple endpoints within an endpoint intersection cluster.

[0088] Then, the centroid coordinates are used to represent the center position of the intersection area, and the output is used as a high-precision weld dead angle detection point.

[0089] The method provided in this invention achieves sub-pixel-level accurate fitting of intersection dead angles by performing radius clustering on all endpoints of the weld line segment and calculating the centroid coordinates of the intersection clusters of the clustered endpoints as dead angle detection points. This effectively avoids the local noise and calculation errors that are easily generated when relying on a single endpoint for positioning.

[0090] To ensure the quality of the input point cloud data, based on any of the above embodiments, step 110 includes: Obtain the initial three-dimensional point cloud of the grid-shaped weldment to be inspected; A region of interest is constructed based on a preset spatial coordinate range. Background point clouds located outside the region of interest in the initial 3D point cloud are removed to obtain the retained point cloud. The retained point cloud is uniformly downsampled according to a preset voxel size to obtain a sampled point cloud; The average neighborhood distance of each point in the sampled point cloud is calculated. If the average neighborhood distance of the target point does not meet the preset standard deviation multiple constraint, the target point is determined to be an outlier noise point and is removed to obtain the three-dimensional point cloud to be detected.

[0091] Specifically, the first step is to acquire the initial 3D point cloud of the grid-shaped weldment to be inspected. For example, the initial 3D point cloud data of the weldment can be acquired using a 3D laser scanning device. Here, the initial 3D point cloud refers to the raw 3D data that has just been output by the 3D acquisition device and has not undergone any noise reduction or cropping processing. It often contains a large amount of environmental background and welding spatter.

[0092] Then, a region of interest (ROI) is constructed based on a preset spatial coordinate range. Background point clouds outside the ROI in the initial 3D point cloud are removed to obtain the retained point cloud. Here, the region of interest (ROI) refers to the effective observation bounding box defined in 3D space based on the actual physical size and placement of the workpiece.

[0093] In practice, the upper and lower limits of spatial coordinates are set to construct the region of interest. Cluttered data that falls outside the region in the initial 3D point cloud, such as background interference point clouds like the ground and equipment fixtures, are directly deleted, leaving only the point cloud data within the effective workpiece space, thus obtaining the retained point cloud.

[0094] Then, the retained point cloud is uniformly downsampled according to a preset voxel size to obtain a sampled point cloud. Here, the preset voxel size refers to the side length of the grid when the three-dimensional space is divided into tiny cubic grids, which is used to control the degree of sparsity of the point cloud.

[0095] In practical implementation, a voxel grid filter can be used to process the retained point cloud. The 3D space is divided into multiple grids of a preset voxel size, and the centroid of all points within each voxel grid is used to replace all points within that grid. This significantly reduces the density of the point cloud data while maintaining the macroscopic structural features of the workpiece, resulting in a smaller, more uniformly distributed sampled point cloud. Here, the voxel downsampling mean filter formula is: ; in, The coordinates of the representative point of a voxel are obtained by replacing the original N points with the positions of the centroids of all points within the voxel. Represents the voxel unit of the first The three-dimensional coordinates of the points, where N represents the number of original point clouds contained in the voxel, that is, the total number of points falling into this voxel grid.

[0096] Finally, the average neighborhood distance of each point in the sampled point cloud is calculated. If the average neighborhood distance of the target point does not meet the preset standard deviation constraint, the target point is determined to be an outlier and is removed to obtain the 3D point cloud to be detected. Here, the average neighborhood distance refers to the average spatial distance between a target point and a predetermined number of its nearest neighbors. The preset standard deviation constraint is a statistical threshold, based on global distance distribution, used to determine whether a point significantly deviates from the main point group.

[0097] In practice, a statistical outlier filtering algorithm is used to calculate the average distance from each target point in the sampled point cloud to its K nearest neighbors, and the mean and standard deviation of the distances across the entire point cloud are calculated. Here, the average distance from each target point in the sampled point cloud to its K nearest neighbors is... It can be calculated using the following formula, as shown below: ; As can be seen from the formula, if the average distance of the neighborhood of a certain target point is... Greater than the global mean Add several times the standard deviation If the preset standard deviation multiple constraint is not met, the point is considered too isolated in space and is determined to be an outlier noise point caused by welding spatter or dust, and is removed from the sampling point cloud.

[0098] It should be noted that after the above multi-level preprocessing, the final output is a clean and structured 3D point cloud to be detected, which is used for plane extraction and blind spot detection.

[0099] The method provided in this invention, through region of interest clipping, voxel uniform downsampling, and statistical filtering based on neighborhood average distance in a two-layer noise reduction process, not only limits the effective workpiece space, but also removes discrete noise such as spatter and dust commonly found in welding conditions, thus significantly improving point cloud quality and processing speed while preserving planar features.

[0100] Based on any of the above embodiments, the grid-shaped weldment to be tested includes a three-plane grid-shaped weldment.

[0101] Here, the three-plane grid-shaped weldment can be constructed by welding three steel plates perpendicularly to each other. The base plate is a horizontally placed first plane steel plate; the first vertical plate can be a second plane steel plate extending perpendicularly to the base plate along the X-direction; and the second vertical plate can be a third plane steel plate extending perpendicularly to the base plate along the Y-direction. It can be understood that when the three steel plates are welded perpendicularly to each other, their top-view projection forms a grid shape, creating three three-plane closed dead angles at the points where the three steel plates intersect. In actual welding, the three closed dead corners of the three-plane grid-shaped weldment are the most critical locations where incomplete penetration, slag inclusion, and porosity are most likely to occur. Due to the limited space and restricted accessibility of the welding torch, conventional welding processes cannot guarantee the welding quality of this location. Therefore, it is necessary to automatically repair the dead corner, clean the root, or grind it after welding.

[0102] By employing the intersection dead angle detection method provided in this embodiment of the invention, three-dimensional point cloud data of the three-plane grid-shaped weldment is acquired using a three-dimensional laser scanning device. After preprocessing, multi-iteration plane segmentation, and plane similarity merging and deduplication, three reference planes corresponding to the base plate, the first upright plate, and the second upright plate are obtained. Subsequently, the distance from each spatial point to the three reference planes is calculated. Candidate points for intersection dead angles are selected through the joint distance constraint of the three planes. Then, endpoint clustering and centroid fitting are performed on the candidate points, and finally, the three-dimensional coordinates of the three three-plane closed dead angles are output to guide the robot to perform automatic welding operations at these locations.

[0103] It should be noted that, compared to existing manual inspection and two-dimensional visual intersection detection, the method provided in this invention solves the problem of the inability to automatically identify dead corners in polyhedral welding by establishing a dedicated algorithm model for multi-plane iterative segmentation, plane deduplication and merging, and three-distance joint constraints. Specifically, through ROI spatial constraints, double-layer noise reduction, connected component filtering, plane similarity deduplication, and a dual-threshold adaptive mechanism, it shields against interference from welding slag, spatter, dust, sparse point clouds, and local deformation. The detection stability is far superior to traditional visual and general point cloud algorithms, making it highly applicable in industrial settings.

[0104] No, the entire process requires no manual marking, no manual teaching, and no manual parameter adjustment. The inspection time for a single workpiece is in the hundreds of milliseconds, which is dozens of times more efficient than manual inspection. It can be directly integrated into a robotic production line to achieve intelligent unmanned welding production.

[0105] Based on any of the above embodiments Figure 2 This is a schematic diagram of the intersection dead angle detection device for grid-shaped welded parts provided by the present invention, as shown in the figure. Figure 2 As shown, the device includes: Acquisition unit 210 acquires the three-dimensional point cloud of the grid-shaped weldment to be inspected; The planar extraction unit 220 performs planar extraction on the three-dimensional point cloud to be detected to obtain multiple reference planes corresponding to the grid-shaped weldment. The distance calculation unit 230 calculates the spatial distance from each spatial point in the three-dimensional point cloud to be detected to the plurality of reference planes. The blind spot candidate point screening unit 240, based on multiple spatial distances corresponding to each spatial point and preset multi-plane joint distance constraints, filters out intersection blind spot candidate points that are simultaneously close to the multiple reference planes from the spatial points. The weld inspection point calculation unit 250 performs clustering and filtering on all endpoints of the weld line segments corresponding to the multiple reference planes to obtain endpoint intersection clusters, calculates the centroid of the endpoint intersection clusters, and obtains weld dead angle inspection points. Integration unit 260 integrates the candidate intersection dead angle points and the weld dead angle detection points to obtain the intersection dead angle detection points of the grid-shaped weldment.

[0106] The device provided in this invention extracts multiple reference planes from the acquired 3D point cloud, selects candidate points for intersection dead angles that are close to multiple planes from the point cloud using preset multi-plane joint distance constraints, and obtains weld dead angle detection points by fitting the centroids of clustering the endpoints of weld line segments. Finally, it obtains the intersection dead angle detection points, thus eliminating the inefficiency and error of manual visual positioning and overcoming the shortcomings of traditional 2D vision in losing depth information and being susceptible to environmental interference. Through the dual mechanism of joint distance constraints and endpoint clustering in 3D space, it achieves anti-interference and high-precision automatic detection of narrow closed dead angles, greatly improving the consistency and product yield of robotic automated welding operations.

[0107] Based on any of the above embodiments, the blind spot candidate point screening unit is specifically used for: For each spatial point, the plurality of spatial distances are compared with a single-plane distance threshold. If all the spatial distances satisfy the single-plane distance threshold, the corresponding spatial point is determined as the candidate point of the intersection dead angle.

[0108] Based on any of the above embodiments, the blind spot candidate point screening unit is further specifically used for: If there are no spatial points that satisfy the single-plane distance threshold, the multi-plane joint distance constraint is relaxed to the single-plane distance relaxation threshold. For each spatial point, the multiple spatial distances are compared with the single-plane distance relaxation threshold. If all the multiple spatial distances meet the single-plane distance relaxation threshold, the corresponding spatial point is determined as the intersection dead angle candidate point.

[0109] Based on any of the above embodiments, the planar extraction unit is specifically used for: The three-dimensional point cloud to be detected is subjected to multiple iterative plane segmentation to obtain multiple initial planes; For any two initial planes among the plurality of initial planes, calculate the angle between the normal vectors of the two initial planes, and the minimum Euclidean distance between the point clouds contained in the two initial planes; If the angle between the normal vectors and the minimum Euclidean distance both satisfy the preset coplanarity condition, it is determined that the two initial planes are the same physical plane, and the point clouds of the two initial planes are merged to refit and obtain the merged plane. Based on all merged planes and unmerged initial planes, the multiple reference planes corresponding to the grid-shaped weldment are obtained.

[0110] Based on any of the above embodiments, the planar extraction unit is further specifically used for: Using the 3D point cloud to be detected as the current point cloud to be segmented, perform the following planar extraction operation: Perform planar segmentation on the current point cloud to be segmented to obtain the current segmented inner point cloud; Calculate the distance between each point in the inner point cloud, perform cluster analysis on the inner point cloud based on distance connectivity, and take the connected domain point cloud with the largest number of connected points as the main plane for the current extraction. If the principal plane is successfully extracted, the remaining point cloud in the current point cloud to be segmented after removing the principal plane is updated to the current point cloud to be segmented; if the principal plane is not successfully extracted, the number of consecutive failures is accumulated. If the number of consecutive failures does not reach the preset threshold, return to perform the planar extraction operation; If the number of consecutive failures reaches the preset threshold, the iteration ends, and the main planes successfully extracted in each iteration are used as the multiple initial planes.

[0111] Based on any of the above embodiments, the weld inspection point calculation unit is specifically used for: Obtain all endpoints of the weld line segment; Based on a preset clustering radius, spatial distance clustering is performed on all endpoints of the weld line segment to filter out the endpoint intersection clusters containing multiple endpoints; Calculate the average spatial coordinates of multiple endpoints within the endpoint intersection cluster, and use the average spatial coordinates as the centroid to obtain the weld dead angle detection point.

[0112] Based on any of the above embodiments, the acquisition unit is specifically used for: Obtain the initial three-dimensional point cloud of the grid-shaped weldment to be inspected; A region of interest is constructed based on a preset spatial coordinate range. Background point clouds located outside the region of interest in the initial 3D point cloud are removed to obtain the retained point cloud. The retained point cloud is uniformly downsampled according to a preset voxel size to obtain a sampled point cloud; The average neighborhood distance of each point in the sampled point cloud is calculated. If the average neighborhood distance of the target point does not meet the preset standard deviation multiple constraint, the target point is determined to be an outlier noise point and is removed to obtain the three-dimensional point cloud to be detected.

[0113] Based on any of the above embodiments, the grid-shaped weldment to be tested includes a three-plane grid-shaped weldment.

[0114] Figure 3 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 3 As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 can call logic instructions in the memory 330 to execute a method for detecting intersection dead angles of a grid-shaped weldment. This method includes: acquiring a three-dimensional point cloud of the grid-shaped weldment to be detected; performing planar extraction on the three-dimensional point cloud to obtain multiple reference planes corresponding to the grid-shaped weldment; calculating the spatial distance from each spatial point in the three-dimensional point cloud to the multiple reference planes; based on the multiple spatial distances corresponding to each spatial point and a preset multi-plane joint distance constraint, selecting candidate intersection dead angle points from the spatial points that are simultaneously close to the multiple reference planes; performing clustering and filtering on all endpoints of the weld line segments corresponding to the multiple reference planes to obtain endpoint intersection clusters; calculating the centroid of the endpoint intersection clusters to obtain weld dead angle detection points; and integrating the candidate intersection dead angle points and the weld dead angle detection points to obtain the intersection dead angle detection points of the grid-shaped weldment.

[0115] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the intersection dead angle detection method for the grid-shaped weldment provided by the above methods. The method includes: acquiring a three-dimensional point cloud of the grid-shaped weldment to be detected; performing planar extraction on the three-dimensional point cloud to obtain multiple reference planes corresponding to the grid-shaped weldment; calculating the spatial distance from each spatial point in the three-dimensional point cloud to the multiple reference planes; selecting candidate intersection dead angles that are close to the multiple reference planes from the spatial points based on the multiple spatial distances corresponding to each spatial point and a preset multi-plane joint distance constraint; performing clustering screening on all endpoints of the weld line segments corresponding to the multiple reference planes to obtain endpoint intersection clusters; calculating the centroid of the endpoint intersection clusters to obtain weld dead angle detection points; and integrating the candidate intersection dead angles and the weld dead angle detection points to obtain the intersection dead angle detection points of the grid-shaped weldment.

[0117] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a method for detecting intersection dead angles of a grid-shaped weldment provided by the above methods. The method includes: acquiring a three-dimensional point cloud of the grid-shaped weldment to be detected; performing planar extraction on the three-dimensional point cloud to obtain multiple reference planes corresponding to the grid-shaped weldment; calculating the spatial distance from each spatial point in the three-dimensional point cloud to the multiple reference planes; selecting candidate intersection dead angles that are simultaneously close to the multiple reference planes from the spatial points based on the multiple spatial distances corresponding to each spatial point and a preset multi-plane joint distance constraint; performing clustering screening on all endpoints of the weld line segments corresponding to the multiple reference planes to obtain endpoint intersection clusters; calculating the centroid of the endpoint intersection clusters to obtain weld dead angle detection points; and integrating the candidate intersection dead angles and the weld dead angle detection points to obtain the intersection dead angle detection points of the grid-shaped weldment.

[0118] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0119] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting dead angles at intersections of grid-shaped welded components, characterized in that, include: Obtain the 3D point cloud of the grid-shaped weldment to be inspected; The three-dimensional point cloud to be detected is subjected to planar extraction to obtain multiple reference planes corresponding to the grid-shaped weldment; For each spatial point in the three-dimensional point cloud to be detected, the spatial distance from each spatial point to the multiple reference planes is calculated. Based on multiple spatial distances corresponding to each spatial point and preset multi-plane joint distance constraints, candidate intersection dead angle points that are simultaneously close to the multiple reference planes are selected from the spatial points. Clustering and filtering are performed on all endpoints of the weld line segments corresponding to the multiple reference planes to obtain endpoint intersection clusters. The centroid of the endpoint intersection clusters is calculated to obtain the weld dead angle detection points. By integrating the candidate intersection dead angle points and the weld dead angle detection points, the intersection dead angle detection points of the grid-shaped weldment are obtained.

2. The method for detecting dead angles at intersections of grid-shaped weldments according to claim 1, characterized in that, The method, based on multiple spatial distances corresponding to each spatial point and preset multi-plane joint distance constraints, filters out candidate intersection dead angles from the spatial points that are simultaneously close to the multiple reference planes, including: For each spatial point, the plurality of spatial distances are compared with a single-plane distance threshold. If all the spatial distances satisfy the single-plane distance threshold, the corresponding spatial point is determined as the candidate point of the intersection dead angle.

3. The method for detecting dead angles at intersections of grid-shaped weldments according to claim 2, characterized in that, The method further includes: If there are no spatial points that satisfy the single-plane distance threshold, the multi-plane joint distance constraint is relaxed to the single-plane distance relaxation threshold. For each spatial point, the multiple spatial distances are compared with the single-plane distance relaxation threshold. If all the multiple spatial distances meet the single-plane distance relaxation threshold, the corresponding spatial point is determined as the intersection dead angle candidate point.

4. The method for detecting dead angles at intersections of grid-shaped weldments according to any one of claims 1 to 3, characterized in that, The step of extracting planar data from the three-dimensional point cloud to be detected, resulting in multiple reference planes corresponding to the grid-shaped weldment, includes: The three-dimensional point cloud to be detected is subjected to multiple iterative plane segmentation to obtain multiple initial planes; For any two initial planes among the plurality of initial planes, calculate the angle between the normal vectors of the two initial planes, and the minimum Euclidean distance between the point clouds contained in the two initial planes; If the angle between the normal vectors and the minimum Euclidean distance both satisfy the preset coplanarity condition, it is determined that the two initial planes are the same physical plane, and the point clouds of the two initial planes are merged to refit and obtain the merged plane. Based on all merged planes and unmerged initial planes, the multiple reference planes corresponding to the grid-shaped weldment are obtained.

5. The method for detecting dead angles at intersections of grid-shaped weldments according to claim 4, characterized in that, The process of performing multiple iterative plane segmentation on the 3D point cloud to be detected yields multiple initial planes, including: Using the 3D point cloud to be detected as the current point cloud to be segmented, perform the following planar extraction operation: Perform planar segmentation on the current point cloud to be segmented to obtain the current segmented inner point cloud; Calculate the distance between each point in the inner point cloud, perform cluster analysis on the inner point cloud based on distance connectivity, and take the connected domain point cloud with the largest number of connected points as the main plane for the current extraction. If the main plane is successfully extracted, the remaining point cloud in the current point cloud to be segmented after removing the main plane is updated to the current point cloud to be segmented; if the main plane is not successfully extracted, the number of consecutive failures is accumulated. If the number of consecutive failures does not reach the preset threshold, return to perform the planar extraction operation; If the number of consecutive failures reaches the preset threshold, the iteration ends, and the main planes successfully extracted in each iteration are used as the multiple initial planes.

6. The method for detecting dead angles at intersections of grid-shaped weldments according to any one of claims 1 to 3, characterized in that, The process of clustering and filtering all endpoints of the weld line segments corresponding to the multiple reference planes to obtain endpoint intersection clusters, calculating the centroid of the endpoint intersection clusters, and obtaining weld dead angle detection points includes: Obtain all endpoints of the weld line segment; Based on a preset clustering radius, spatial distance clustering is performed on all endpoints of the weld line segment to filter out the endpoint intersection clusters containing multiple endpoints; Calculate the average spatial coordinates of multiple endpoints within the endpoint intersection cluster, and use the average spatial coordinates as the centroid to obtain the weld dead angle detection point.

7. The method for detecting dead angles at intersections of grid-shaped weldments according to any one of claims 1 to 3, characterized in that, The process of acquiring the three-dimensional point cloud of the grid-shaped weldment to be inspected includes: Obtain the initial three-dimensional point cloud of the grid-shaped weldment to be inspected; A region of interest is constructed based on a preset spatial coordinate range. Background point clouds located outside the region of interest in the initial 3D point cloud are removed to obtain the retained point cloud. The retained point cloud is uniformly downsampled according to a preset voxel size to obtain a sampled point cloud; The average neighborhood distance of each point in the sampled point cloud is calculated. If the average neighborhood distance of the target point does not meet the preset standard deviation multiple constraint, the target point is determined to be an outlier noise point and is removed to obtain the three-dimensional point cloud to be detected.

8. The method for detecting dead angles at intersections of grid-shaped weldments according to any one of claims 1 to 3, characterized in that, The grid-shaped weldment to be tested includes a three-plane grid-shaped weldment.

9. A device for detecting dead angles at intersections of grid-shaped welded parts, characterized in that, include: The acquisition unit acquires the three-dimensional point cloud of the grid-shaped weldment to be inspected; The planar extraction unit performs planar extraction on the three-dimensional point cloud to be detected, and obtains multiple reference planes corresponding to the grid-shaped weldment; The distance calculation unit calculates the spatial distance from each spatial point in the three-dimensional point cloud to be detected to the multiple reference planes. The blind spot candidate point screening unit filters out intersection blind spot candidate points that are simultaneously close to the multiple reference planes from the spatial points based on multiple spatial distances corresponding to each spatial point and preset multi-plane joint distance constraints. The weld inspection point calculation unit performs clustering and filtering on all endpoints of the weld line segments corresponding to the multiple reference planes to obtain endpoint intersection clusters, calculates the centroid of the endpoint intersection clusters, and obtains weld dead angle inspection points. The integration unit integrates the candidate intersection dead angle points and the weld dead angle detection points to obtain the intersection dead angle detection points of the grid-shaped weldment.

10. The intersection dead angle detection device for grid-shaped welded parts according to claim 9, characterized in that, The blind spot candidate point filtering unit is specifically used for: For each spatial point, the plurality of spatial distances are compared with a single-plane distance threshold. If all the spatial distances satisfy the single-plane distance threshold, the corresponding spatial point is determined as the candidate point of the intersection dead angle.

Images