Multi-angle scanning and curved surface unfolding weld defect detection method, system and device

Abstract

The invention discloses a multi-angle scanning and curved surface unfolding weld defect detection method, system and device, and the method comprises the steps: controlling each group of triangular laser contour sensors to scan a weld region of a target welding object, and obtaining a multi-view weld point cloud; point cloud fusion processing is carried out on the multi-view welding seam point cloud to obtain a three-dimensional surface point cloud of the welding seam area; performing curved surface expansion processing on the three-dimensional surface point cloud to obtain a two-dimensional expansion graph; performing fusion processing on the three-dimensional surface point cloud and the two-dimensional expansion graph according to a preset fusion algorithm to obtain a fused image; and inputting the fused image into a preset defect detection model to obtain a defect detection result. According to the method, the multi-view weld point cloud of the weld area is obtained, the problems that data are incomplete and detection blind areas are prone to occurring are avoided, curved surface expansion is conducted on the three-dimensional surface point cloud to obtain the two-dimensional expansion view, then the three-dimensional surface point cloud and the two-dimensional expansion view are fused, and the geometric feature utilization rate of the two-dimensional expansion view to the three-dimensional surface point cloud can be increased.

Description

Methods, systems and devices for weld defect detection using multi-angle scanning and surface unfolding Technical Field

[0001] This application relates to the field of weld inspection technology, and in particular to a method, system and apparatus for detecting weld defects using multi-angle scanning and surface unfolding. Background Technology

[0002] In the process of automated welding, weld inspection is a crucial step in ensuring product quality and structural reliability. Welding is typically achieved through methods such as laser welding and brazing, and the quality of the weld directly determines the sealing performance and overall strength of the workpiece. Therefore, defect inspection of the weld must be performed during the production process.

[0003] Currently, existing weld defect detection methods typically employ a single triangular laser profile sensor to inspect the weld appearance. This involves using a single triangular laser profile sensor to scan the weld cross-sectional profile to obtain depth data, which is then combined with image analysis algorithms to identify weld morphological anomalies. However, a single triangular laser profile sensor emits a single-angle laser, making it susceptible to interference from weld reflections, shadows, and curved surface obstructions during scanning, leading to data loss in certain areas and creating detection blind spots.

[0004] Therefore, existing weld defect detection methods suffer from incomplete data and are prone to detection blind spots. Summary of the Invention

[0005] The main purpose of this application is to propose a method, system and device for detecting weld defects by multi-angle scanning and surface unfolding, which aims to solve the problems of incomplete data and blind spots in weld defect detection.

[0006] To achieve the above objectives, this application proposes a multi-angle scanning and surface unfolding method for weld defect detection. This method is used in a multi-angle scanning and surface unfolding weld defect detection system, which includes several sets of triangular laser profile sensors with different incident angles. The method comprises: controlling each set of triangular laser profile sensors to scan the weld area of ​​the target welded object to obtain multi-view weld point clouds; performing point cloud fusion processing on the multi-view weld point clouds to obtain a three-dimensional surface point cloud of the weld area; performing surface unfolding processing on the three-dimensional surface point cloud to obtain a two-dimensional unfolded image; fusing the three-dimensional surface point cloud and the two-dimensional unfolded image according to a preset fusion algorithm to obtain a fused image; and inputting the fused image into a preset defect detection model to obtain a defect detection result.

[0007] In some embodiments, before controlling each group of triangular laser contour sensors to scan the weld area of ​​the target welded object to obtain a multi-view weld point cloud, the method further includes: controlling each group of triangular laser contour sensors to scan a preset calibration plate to obtain standard three-dimensional point clouds, wherein the preset calibration plate includes multiple feature points with known spatial positions; extracting feature points from each standard three-dimensional point cloud to obtain each feature point of each standard three-dimensional point cloud; establishing a correspondence between each feature point of each standard three-dimensional point cloud based on the spatial position of each feature point on the preset calibration plate to obtain a spatial position correspondence; constructing a world coordinate system based on the preset calibration plate, and calculating the rigid body transformation matrix of each group of triangular laser contour sensors relative to the world coordinate system based on the spatial position correspondence to obtain the rigid body transformation matrix of each group of triangular laser contour sensors.

[0008] In some embodiments, the multi-angle scanning and surface unfolding weld defect detection system further includes a motion unit, which is used to drive the target welded object to move; controlling each group of triangular laser contour sensors to scan the weld area of ​​the target welded object to obtain a multi-view weld point cloud includes: controlling the motion unit to drive the target welded object to move at a constant speed according to a preset motion trajectory, and synchronously controlling each group of triangular laser contour sensors to continuously sample the weld area; acquiring the displacement information of the motion unit and multiple frames of contour data continuously output by each group of triangular laser contour sensors; and sorting the multiple frames of contour data from each group of triangular laser contour sensors according to the displacement information to obtain the multi-view weld point cloud.

[0009] In some embodiments, the step of performing point cloud fusion processing on the multi-view weld point cloud to obtain the three-dimensional surface point cloud of the weld region includes: mapping the multi-view weld point cloud to the world coordinate system according to the rigid body transformation matrix of each group of triangular laser contour sensors to obtain a mapped point cloud; calling a nearest neighbor search algorithm to perform neighborhood analysis on the mapped point cloud to obtain overlapping and non-overlapping regions; fusing the points in the overlapping region according to a weighted average strategy to obtain an overlapping fused point cloud; and integrating the overlapping fused point cloud and the point cloud of the non-overlapping region to obtain the three-dimensional surface point cloud.

[0010] In some embodiments, the step of performing surface unfolding processing on the three-dimensional surface point cloud to obtain a two-dimensional unfolded diagram includes: extracting the three-dimensional surface point cloud based on the centerline of the weld region to obtain a discrete point sequence of the centerline; constructing a local coordinate system for each discrete point according to the discrete point sequence; for any surface point of the three-dimensional surface point cloud, querying the discrete point sequence to determine the target discrete point closest to the surface point; calculating the cumulative arc length from the surface point to the target discrete point as the abscissa; calculating the projection distance of the surface point in the local coordinate system of the target discrete point as the ordinate; generating the two-dimensional unfolded diagram based on the abscissa and ordinate of all the surface points, and establishing an index table of each point in the two-dimensional unfolded diagram and each surface point in the three-dimensional surface point cloud.

[0011] In some embodiments, the step of fusing the three-dimensional surface point cloud and the two-dimensional unfolded map according to a preset fusion algorithm to obtain a fused image includes: rasterizing the two-dimensional unfolded map according to a preset grid resolution to obtain multiple two-dimensional grids; for any two-dimensional grid, determining the surface point in the three-dimensional surface point cloud corresponding to each grid point of the two-dimensional grid according to the index table; performing geometric feature calculation on all the surface points corresponding to each two-dimensional grid to obtain the geometric feature channels of each two-dimensional grid, wherein the geometric feature channels include an average depth channel, a surface normal vector channel, and an average curvature channel; mapping each grid point of each two-dimensional grid to an image channel; concatenating the geometric feature channels of each surface point with the image channels of each grid point to generate a multi-channel feature image, and using the multi-channel feature image as the fused image.

[0012] In some embodiments, the step of inputting the fused image into a preset defect detection model to obtain a defect detection result includes: inputting the fused image into the preset defect detection model; obtaining a plurality of candidate defect boxes output by the preset defect detection model and the defect category and defect confidence level corresponding to each candidate defect box; determining whether the defect confidence level corresponding to each candidate defect box is greater than a preset confidence threshold; when it is greater than the preset confidence threshold, determining the candidate defect box as a target defect box; and generating the defect detection result based on all the target defect boxes and the defect category and defect confidence level corresponding to each target defect box.

[0013] This application further proposes a weld defect detection system based on multi-angle scanning and surface unfolding. The multi-angle scanning and surface unfolding weld defect detection system includes a data processor and several sets of triangular laser contour sensors with different incident angles. The multi-angle scanning and surface unfolding weld defect detection system is capable of performing the multi-angle scanning and surface unfolding weld defect detection method described above.

[0014] In some embodiments, the multi-angle scanning and surface unfolding weld defect detection system further includes a motion unit for moving the target welded object. Each set of triangular laser profile sensors is fixedly configured, and the scanning range of each set of triangular laser profile sensors includes part or all of the surface area of ​​the motion unit.

[0015] This application further proposes a weld defect detection device for multi-angle scanning and surface unfolding, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions to be executed by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the weld defect detection method for multi-angle scanning and surface unfolding described above.

[0016] This application's technical solution uses several sets of triangular laser contour sensors with different incident angles to scan the weld area and obtain multi-view weld point clouds. These multi-view weld point clouds are then fused to obtain a three-dimensional surface point cloud. The three-dimensional surface point cloud is then unfolded to obtain a two-dimensional unfolded image. Finally, the three-dimensional surface point cloud and the two-dimensional unfolded image are fused to obtain a fused image. Finally, defect detection is performed on the fused image to obtain the defect detection result. Using several sets of triangular laser contour sensors with different incident angles to scan the weld area avoids the problems of incomplete data and the potential for blind spots. Furthermore, unfolding the three-dimensional surface point cloud to obtain a two-dimensional unfolded image, and then fusing the three-dimensional surface point cloud and the two-dimensional unfolded image, can improve the utilization rate of the geometric features of the three-dimensional surface point cloud in the two-dimensional unfolded image. Attached Figure Description

[0017] Figure 1 is a flowchart illustrating one embodiment of the weld defect detection method using multi-angle scanning and surface unfolding according to this application; Figure 2 is a flowchart illustrating another embodiment of the weld defect detection method using multi-angle scanning and surface unfolding according to this application; Figure 3 is a flowchart illustrating another embodiment of the weld defect detection method using multi-angle scanning and surface unfolding according to this application; Figure 4 is a flowchart illustrating another embodiment of the weld defect detection method using multi-angle scanning and surface unfolding according to this application; Figure 5 is a flowchart illustrating another embodiment of the weld defect detection method using multi-angle scanning and surface unfolding according to this application; Figure 6 is a flowchart illustrating another embodiment of the weld defect detection method using multi-angle scanning and surface unfolding according to this application; Figure 7 is a flowchart illustrating another embodiment of the weld defect detection method using multi-angle scanning and surface unfolding according to this application; Figure 8 is a structural schematic diagram of the weld defect detection system using multi-angle scanning and surface unfolding according to the embodiments of this application; Figure 9 is a structural schematic diagram of the weld defect detection device using multi-angle scanning and surface unfolding according to the embodiments of this application. Detailed Implementation

[0018] The solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments in this application, and not all of the 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.

[0019] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0020] It should also be noted that when a component is referred to as "fixed to" or "set on" another component, it can be directly on the other component or an intervening component can be present simultaneously. When a component is referred to as "connected to" another component, it can be directly connected to the other component or an intervening component can be present simultaneously.

[0021] Furthermore, the use of terms such as "first" and "second" in this application is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined as "first" or "second" explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. If the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed in this application.

[0022] This application proposes a multi-angle scanning and surface unfolding method for weld defect detection. Referring to Figures 1 and 8, Figure 1 is a flowchart illustrating an embodiment of the multi-angle scanning and surface unfolding method for weld defect detection, and Figure 8 is a structural schematic diagram of the multi-angle scanning and surface unfolding weld defect detection system according to an embodiment of this application. In some embodiments, the multi-angle scanning and surface unfolding weld defect detection method is used in a multi-angle scanning and surface unfolding weld defect detection system, which includes several sets of triangular laser contour sensors with different incident angles; the multi-angle... The weld defect detection method based on scanning and surface unfolding includes: step S110, controlling each group of triangular laser contour sensors to scan the weld area of ​​the target welded object to obtain multi-view weld point clouds; step S120, performing point cloud fusion processing on the multi-view weld point clouds to obtain a three-dimensional surface point cloud of the weld area; step S130, performing surface unfolding processing on the three-dimensional surface point cloud to obtain a two-dimensional unfolded image; step S140, performing fusion processing on the three-dimensional surface point cloud and the two-dimensional unfolded image according to a preset fusion algorithm to obtain a fused image; step S150, inputting the fused image into a preset defect detection model to obtain the defect detection result.

[0023] In this embodiment, as shown in Figures 1 and 8, the multi-angle scanning and surface unfolding weld defect detection method can be applied to a multi-angle scanning and surface unfolding weld defect detection system. The multi-angle scanning and surface unfolding weld defect detection system includes a data processor and several sets of triangular laser profile sensors with different incident angles. The data processor is connected to each set of triangular laser profile sensors. In this embodiment, the data processor is the main entity executing the method steps.

[0024] Understandably, the data processor can be a computer device or an embedded device. The multi-angle scanning and surface unfolding weld defect detection system also includes a motion unit and a mounting frame; the motion unit can be mounted at the bottom of the mounting frame, and each set of triangular laser profile sensors can be mounted at the top of the mounting frame. Each set of triangular laser profile sensors is fixedly configured, and each set has a different incident angle (i.e., different scanning angles); the scanning range of each set of triangular laser profile sensors includes part or all of the surface area of ​​the motion unit. Both the motion unit and each set of triangular laser profile sensors are connected to the data processor, enabling the data processor to control the motion unit and each set of triangular laser profile sensors.

[0025] When a user needs to detect weld defects in the weld area of ​​a target welded object, the target welded object can be moved to a motion unit. The motion unit then moves the target welded object within the scanning range of each set of triangular laser profile sensors. At this point, the data processor can control several sets of triangular laser profile sensors with different incident angles to scan the weld area of ​​the target welded object, thereby obtaining a multi-view weld point cloud. For example, the several sets of triangular laser profile sensors with different incident angles can be two sets. One set of triangular laser profile sensors can be fixedly configured at the upper left of the motion unit, and the other set can be fixedly configured at the upper right of the motion unit. Thus, one set of triangular laser profile sensors can scan the weld area from the upper left, obtaining a weld point cloud from the upper left perspective; the other set of triangular laser profile sensors can scan the weld area from the upper right, obtaining a weld point cloud from the upper right perspective. The combination of the weld point clouds from the upper left and upper right perspectives constitutes the multi-view weld point cloud.

[0026] After obtaining multi-view weld point clouds, the data processor can perform point cloud fusion processing on these clouds. This fusion process yields a three-dimensional surface point cloud of the weld region. For example, the data processor can fuse the weld point clouds from the upper left and upper right views into a single, complete point cloud, thus obtaining the three-dimensional surface point cloud of the weld region.

[0027] After obtaining a 3D surface point cloud, the data processor can perform surface unfolding on the 3D surface point cloud. The data processor performs surface unfolding on the 3D surface point cloud to obtain a 2D unfolded image. For example, the weld in the weld area can be a slender continuous curved surface. The data processor can flatten the curved surface of the 3D surface point cloud along the center line of the weld, thereby unfolding the surface and obtaining a 2D unfolded image.

[0028] After obtaining the 2D unfolded image, the data processor can fuse the 3D surface point cloud and the 2D unfolded image according to a preset fusion algorithm to obtain a fused image. The data processor can fuse the 3D surface point cloud and the 2D unfolded image so that the final fused image has both 2D texture features and 3D geometric features.

[0029] After obtaining the fused image, the data processor can input it into a preset defect detection model. This preset defect detection model can be pre-trained by the user and then configured on the data processor. Once the data processor inputs the fused image into the preset defect detection model, the model can perform defect detection on the fused image, thereby obtaining the defect detection results.

[0030] This application's technical solution uses several sets of triangular laser contour sensors with different incident angles to scan the weld area and obtain multi-view weld point clouds. These multi-view weld point clouds are then fused to obtain a three-dimensional surface point cloud. The three-dimensional surface point cloud is then unfolded to obtain a two-dimensional unfolded image. Finally, the three-dimensional surface point cloud and the two-dimensional unfolded image are fused to obtain a fused image. Finally, defect detection is performed on the fused image to obtain the defect detection result. Using several sets of triangular laser contour sensors with different incident angles to scan the weld area avoids the problems of incomplete data and the potential for blind spots. Furthermore, unfolding the three-dimensional surface point cloud to obtain a two-dimensional unfolded image, and then fusing the three-dimensional surface point cloud and the two-dimensional unfolded image, can improve the utilization rate of the geometric features of the three-dimensional surface point cloud in the two-dimensional unfolded image.

[0031] Referring to Figure 2, which is a flowchart illustrating another embodiment of the weld defect detection method based on multi-angle scanning and surface unfolding of this application, in some embodiments, before controlling each group of triangular laser contour sensors to scan the weld area of ​​the target welded object to obtain a multi-view weld point cloud, the method further includes: step S160, controlling each group of triangular laser contour sensors to scan a preset calibration plate to obtain each standard three-dimensional point cloud, wherein the preset calibration plate includes multiple feature points with known spatial positions; step S161, extracting feature points from each standard three-dimensional point cloud to obtain each feature point of each standard three-dimensional point cloud; step S162, establishing a correspondence between each feature point of each standard three-dimensional point cloud based on the spatial position of each feature point on the preset calibration plate to obtain a spatial position correspondence; step S163, constructing a world coordinate system based on the preset calibration plate, and calculating the rigid transformation matrix of each group of triangular laser contour sensors relative to the world coordinate system based on the spatial position correspondence to obtain the rigid transformation matrix of each group of triangular laser contour sensors.

[0032] In this embodiment, as shown in Figure 2, before executing step S110, the data processor can first scan the preset calibration plate. The preset calibration plate can be a high-precision laser 3D scanner magnetic marker calibration plate, which includes multiple feature points (magnetic markers) with known spatial locations. The data processor can control each set of triangular laser contour sensors to scan the preset calibration plate, thereby obtaining standard 3D point clouds. For example, several sets of triangular laser contour sensors with different incident angles can be two sets of triangular laser contour sensors with different incident angles. One set of triangular laser contour sensors can be fixedly configured at the upper left of the motion unit, and the other set can be fixedly configured at the upper right of the motion unit. Thus, one set of triangular laser contour sensors can scan the preset calibration plate from the upper left to obtain a standard 3D point cloud from the upper left perspective; the other set of triangular laser contour sensors can scan the preset calibration plate from the upper right to obtain a standard 3D point cloud from the upper right perspective.

[0033] After obtaining the standard 3D point clouds, the data processor can extract feature points from each standard 3D point cloud separately. The preset calibration board includes multiple feature points with known spatial locations. For example, since the standard 3D point clouds are obtained by scanning the preset calibration board, the feature points on the preset calibration board will also be scanned. The data processor can extract all the scanned feature points from the standard 3D point cloud at the upper left viewpoint, thus obtaining the feature points of the standard 3D point cloud at the upper left viewpoint; it can also extract all the scanned feature points from the standard 3D point cloud at the upper right viewpoint, thus obtaining the feature points of the standard 3D point cloud at the upper right viewpoint.

[0034] After obtaining the feature points of each standard 3D point cloud, the data processor can establish the correspondence between these feature points. The data processor can establish the correspondence between the feature points of each standard 3D point cloud based on the spatial positions of the feature points on a preset calibration board, thus obtaining the spatial position correspondence. For example, the data processor can establish the correspondence between the feature points of the standard 3D point cloud viewed from the upper left and the feature points of the standard 3D point cloud viewed from the upper right, based on the spatial positions of the feature points on the preset calibration board (e.g., if a feature point on both sides falls on a single feature point on the preset calibration board, then the feature points on both sides can be marked as having the same relationship; if a feature point on both sides falls on two adjacent feature points on the preset calibration board, then the feature points on both sides can be marked as having an adjacent relationship, etc.). After establishing the correspondence between all feature points, the spatial position correspondence can be obtained.

[0035] The data processor can also construct a world coordinate system based on a preset calibration plate, and then calculate the rigid transformation matrix of each group of triangular laser contour sensors relative to the world coordinate system according to the spatial position correspondence, thereby obtaining the rigid transformation matrix of each group of triangular laser contour sensors, thus realizing the coordinate alignment of multi-view point clouds under the same spatial frame.

[0036] For example: When the weld defect detection system with multi-angle scanning and surface unfolding is used, the data processor can first verify the integrity of the calibration parameter file. After the verification is passed, a high-precision laser 3D scanner magnetic marker calibration plate is used to establish a unified world coordinate system according to the following steps: (1) Place the preset calibration plate in the common visual range of each group of triangular laser contour sensors. Each group of triangular laser contour sensors collects the preset calibration plate data to obtain each standard 3D point cloud; (2) Extract the feature points on the preset calibration plate on each standard 3D point cloud and establish the corresponding relationship; (3) The unified world coordinate system takes the preset calibration plate coordinates as a reference and uses the least squares method to solve the rigid body transformation matrix. By optimizing, the rigid body transformation matrix of each group of triangular laser contour sensors relative to the world coordinate system is obtained, and the coordinate alignment of multi-view point clouds under the same spatial frame is realized.

[0037] Referring to Figure 3, which is a flowchart illustrating another embodiment of the weld defect detection method based on multi-angle scanning and surface unfolding of this application, in some embodiments, the weld defect detection system based on multi-angle scanning and surface unfolding further includes a motion unit, which is used to drive the target welded object to move; the aforementioned control of each group of triangular laser contour sensors to scan the weld area of ​​the target welded object to obtain a multi-view weld point cloud includes: step S170, controlling the motion unit to drive the target welded object to move at a constant speed according to a preset motion trajectory, and synchronously controlling each group of triangular laser contour sensors to continuously sample the weld area; step S171, acquiring the displacement information of the motion unit and the multi-frame contour data frames continuously output by each group of triangular laser contour sensors; step S172, sorting the multi-frame contour data frames of each group of triangular laser contour sensors according to the displacement information to obtain a multi-view weld point cloud.

[0038] In this embodiment, as shown in Figure 3, when the data processor executes step S110, it can first control the motion unit to drive the target welding object to move at a constant speed according to the preset motion trajectory. The multi-angle scanning and surface unfolding weld defect detection system also includes a motion unit, which is used to drive the target welding object to move. For example, the motion unit may include a conveyor belt and a fixture tray, and the conveyor belt can drive the fixture tray to move. The user can first place the target welding object on the fixture tray. The preset motion trajectory is set based on the fact that the target welding object can pass through the scanning range of each set of triangular laser contour sensors at a constant speed from head to tail (from one end to the other).

[0039] When the data processor detects a target welding object on the fixture tray, it controls the motion unit to move the target welding object at a constant speed according to a preset motion trajectory. This allows the target welding object to pass through the scanning range of each set of triangular laser contour sensors from beginning to end (one end to the other) at a uniform speed. This enables each set of triangular laser contour sensors to perform a complete scan of the target welding object from beginning to end (one end to the other). When the data processor controls the motion unit to move the target welding object at a constant speed according to the preset motion trajectory and moves the target welding object into the scanning range of each set of triangular laser contour sensors, the data processor can synchronously control each set of triangular laser contour sensors to continuously sample the weld area. For example, each set of triangular laser contour sensors samples a segment of the target welding object as it moves at a uniform speed.

[0040] The data processor acquires the displacement information of the motion unit and multiple contour data frames continuously output by each set of triangular laser contour sensors. For example, when the motion unit moves, it generates displacement information, which the data processor can acquire. Simultaneously, each set of triangular laser contour sensors outputs sampled contour data frames during sampling. Continuous sampling by each set of triangular laser contour sensors results in continuous output, forming multiple contour data frames. The data processor can then acquire these multiple contour data frames continuously output by each set of triangular laser contour sensors. In this embodiment, only the middle portion of the scanning range of each set of triangular laser contour sensors can be used to sample the weld area to prevent angular offset in the acquired contour data frames.

[0041] The data processor can also sort multiple frames of contour data from each group of triangular laser contour sensors according to displacement information to obtain multi-view weld point clouds. For example, several groups of triangular laser contour sensors with different incident angles can be two groups of triangular laser contour sensors with different incident angles. One group of triangular laser contour sensors can be fixedly configured at the upper left of the motion unit, and the other group can be fixedly configured at the upper right of the motion unit. In this way, one group of triangular laser contour sensors can sample multiple frames of contour data from the upper left perspective; the other group of triangular laser contour sensors can sample multiple frames of contour data from the upper right perspective. The data processor sorts the multiple frames of contour data from the upper left perspective according to displacement information to obtain the weld point cloud from the upper left perspective; sorting the multiple frames of contour data from the upper right perspective according to displacement information to obtain the weld point cloud from the upper right perspective; the set of the weld point clouds from the upper left and upper right perspectives constitutes the multi-view weld point cloud. Each group of triangular laser contour sensors can also output a two-dimensional grayscale image.

[0042] For example, the target object to be welded can be placed on a jig tray to fix the product in place, while ensuring that the weld area is within the scanning range of each set of triangular laser contour sensors. Then, the data processor controls the motion unit to move at a uniform speed along a predetermined trajectory, simultaneously driving each set of laser contour sensors to synchronously acquire contour data, ensuring that each sampling position has a contour cross-section from a different viewpoint. During data acquisition, each set of sensors is synchronously sampled by a unified hardware trigger signal, and the data processor records the encoder displacement information of the motion control unit in real time to ensure a one-to-one correspondence between the sampling frame and the spatial position. Each acquisition output consists of a synchronized laser contour data frame and a camera data frame. The contour data output is a three-dimensional point set (X, Y, Z), and the camera data output is a two-dimensional grayscale image. All acquired data is saved in the original sensor coordinate system.

[0043] Referring to Figure 4, which is a flowchart of another embodiment of the weld defect detection method of multi-angle scanning and surface unfolding according to this application, in some embodiments, the aforementioned point cloud fusion processing of multi-view weld point clouds to obtain a three-dimensional surface point cloud of the weld region includes: step S180, mapping the multi-view weld point clouds to the world coordinate system according to the rigid body transformation matrix of each group of triangular laser contour sensors to obtain a mapped point cloud; step S181, calling the nearest neighbor search algorithm to perform neighborhood analysis on the mapped point cloud to obtain overlapping and non-overlapping regions; step S182, fusing the points in the overlapping region according to the weighted average strategy to obtain an overlapping fused point cloud; and step S183, integrating the overlapping fused point cloud and the point clouds of the non-overlapping regions to obtain a three-dimensional surface point cloud.

[0044] In this embodiment, as shown in Figure 4, when the data processor executes step S120, it can first map the multi-view weld point cloud to the world coordinate system according to the rigid body transformation matrix of each group of triangular laser contour sensors. The data processor can map the multi-view weld point cloud to the world coordinate system according to the rigid body transformation matrix of each group of triangular laser contour sensors, thereby obtaining the mapped point cloud. For example, the multi-view weld point cloud includes the weld point cloud from the upper left view and the weld point cloud from the upper right view. The data processor maps both the weld point cloud from the upper left view and the weld point cloud from the upper right view to the world coordinate system according to the corresponding rigid body transformation matrix, integrating them into one world coordinate system, thereby obtaining the mapped point cloud.

[0045] The data processor can call the nearest neighbor search algorithm to perform neighborhood analysis on the mapped point cloud to obtain overlapping and non-overlapping regions; then, it can fuse the points in the overlapping region according to the weighted average strategy to obtain the overlapping fused point cloud; finally, it can integrate the overlapping fused point cloud and the point cloud of the non-overlapping region to obtain the three-dimensional surface point cloud.

[0046] For example, after the data processor obtains the multi-view weld point cloud, it can map the weld point cloud of each group of triangular laser contour sensors to the world coordinate system through the rigid body transformation matrix of each group of triangular laser contour sensors. After mapping to the world coordinate system, the normal vector of each mapped point cloud is calculated. The nearest neighbor search algorithm based on KD-tree (KDimensional-tree) is used to select neighborhood points for each point. The local normal vector is calculated through the neighborhood point covariance matrix. This normal vector is used for the judgment of the normal angle and the calculation of the weighted fusion weight in the subsequent establishment of overlapping point correspondence.

[0047] A KD-tree-based nearest neighbor search algorithm is used to identify overlapping and non-overlapping regions in the mapped point cloud. For overlapping regions, a spatial distance threshold δ and a normal angle threshold θ are set. For each candidate point pair, a valid corresponding point pair is determined when the Euclidean distance between the two points is less than δ and the normal angle is less than θ. Point pairs that do not meet the threshold conditions are eliminated to ensure the uniqueness and stability of the correspondence.

[0048] For corresponding point pairs in overlapping regions, a weighted average strategy is used for fusion. The weights are determined by the cosine of the incident angle and the consistency of the normal vector, and then normalized. This yields the fused point positions and additional attributes. For non-overlapping regions, the original point data is directly retained. After integrating the overlapping and non-overlapping point clouds, filtering can be performed.

[0049] The integrated point cloud is subjected to voxel-based downsampling. The voxel-based side length is set to v to ensure that the point density is reduced while preserving the geometric details of the weld, thereby reducing data redundancy and improving computational efficiency. Subsequently, a statistical filtering algorithm is used for noise removal, with the number of neighborhood points set to k and the standard deviation factor set to σ. Based on this, the point cloud normal vectors are re-estimated to improve the stability of subsequent surface unfolding and feature calculation.

[0050] Finally, a three-dimensional surface point cloud is obtained after unified coordinate transformation, registration, integration, and filtering.

[0051] Referring to Figure 5, which is a flowchart illustrating another embodiment of the weld defect detection method based on multi-angle scanning and surface unfolding of this application, in some embodiments, the aforementioned surface unfolding process of the three-dimensional surface point cloud to obtain a two-dimensional unfolded image includes: step S190, extracting the three-dimensional surface point cloud based on the centerline of the weld region to obtain a discrete point sequence of the centerline; step S191, constructing a local coordinate system for each discrete point based on the discrete point sequence; step S192, for any surface point in the three-dimensional surface point cloud, querying the discrete point sequence to determine the target discrete point closest to the surface point; step S193, calculating the cumulative arc length from the surface point to the target discrete point as the abscissa; step S194, calculating the projection distance of the surface point in the local coordinate system of the target discrete point as the ordinate; step S195, generating a two-dimensional unfolded image based on the abscissa and ordinate of all surface points, and establishing an index table of each point in the two-dimensional unfolded image and each surface point in the three-dimensional surface point cloud.

[0052] In this embodiment, as shown in Figure 5, when the data processor executes step S130, it can first extract the three-dimensional surface point cloud based on the centerline of the weld area. The data processor can extract the three-dimensional surface point cloud based on the centerline of the weld area, thereby obtaining a discrete point sequence of the centerline. After obtaining the discrete point sequence, the data processor constructs a local coordinate system for each discrete point based on the discrete point sequence; that is, a local coordinate system is constructed for each discrete point. For any surface point in the three-dimensional surface point cloud, the data processor can determine the target discrete point closest to the surface point by querying the discrete point sequence, where the target discrete point is the discrete point closest to the surface point in the discrete point sequence. The data processor can calculate the cumulative arc length from the surface point to the target discrete point as the abscissa and calculate the projection distance of the surface point in the local coordinate system of the target discrete point as the ordinate. The data processor can generate a two-dimensional unfolded diagram based on the abscissa and ordinate of all surface points and establish an index table between each point in the two-dimensional unfolded diagram and each surface point in the three-dimensional surface point cloud.

[0053] For example, the centerline of the weld area can be the central skeleton of the weld area. The data processor performs central skeleton extraction on the 3D surface point cloud in the weld area, using a distance field-based central skeleton extraction method to obtain a discrete point sequence of the centerline. Then, the centerline is smoothly fitted and resampled into an equidistant point sequence for subsequent calculations.

[0054] For any sampling point in the equidistant point sequence, a local coordinate system is constructed along the centerline at the sampling point, with tangent vector t, normal vector n, and binormal vector b, to define the horizontal normal and vertical arc length directions.

[0055] For any surface point in a 3D surface point cloud, find its nearest centerline point, calculate the cumulative arc length along the centerline as the x-coordinate, and calculate the projection distance of the point onto the local normal plane as the y-coordinate. This maps each surface point onto a 2D unfolded coordinate system to obtain a 2D unfolded diagram. The relationship between each surface point and its corresponding point on the 2D unfolded diagram is recorded to obtain an index table.

[0056] Referring to Figure 6, which is a flowchart illustrating another embodiment of the weld defect detection method based on multi-angle scanning and surface unfolding of this application, in some embodiments, the aforementioned process of fusing the three-dimensional surface point cloud and the two-dimensional unfolded image according to a preset fusion algorithm to obtain a fused image includes: step S200, rasterizing the two-dimensional unfolded image according to a preset grid resolution to obtain multiple two-dimensional grids; step S201, for any two-dimensional grid, determining the surface points corresponding to each grid point in the three-dimensional surface point cloud according to an index table; step S202, performing geometric feature calculations on all surface points corresponding to each two-dimensional grid to obtain geometric feature channels for each two-dimensional grid, wherein the geometric feature channels include an average depth channel, a surface normal vector channel, and an average curvature channel; step S203, mapping each grid point of each two-dimensional grid to an image channel; and step S204, stitching the geometric feature channels of each surface point with the image channels of each grid point to generate a multi-channel feature image, and using the multi-channel feature image as the fused image.

[0057] In this embodiment, as shown in FIG6, when the data processor executes step S140, it can first rasterize the two-dimensional unfolded image according to a preset grid resolution to obtain multiple two-dimensional grids. The preset grid resolution can be customized by the user. For example, the preset grid resolution can be set according to the size of the defect to be detected, so that the side length of the two-dimensional grid is less than or equal to half of the defect size, ensuring that the defect details (such as the edge of a small crack, the outline of a pore) can be accurately represented by the rasterized feature channels, and avoiding the loss of defect information due to the two-dimensional grid being too large.

[0058] The data processor can rasterize the 2D unfolded map according to a preset grid resolution, thereby obtaining multiple 2D grids. For any 2D grid, the data processor can determine the surface points in the 3D surface point cloud corresponding to each grid point of the 2D grid according to an index table. The data processor can perform geometric feature calculations on all surface points corresponding to each 2D grid to obtain the geometric feature channels of each 2D grid, where the geometric feature channels include the average depth channel, the surface normal vector channel, and the average curvature channel. The data processor can also map each grid point of each 2D grid to an image channel. Finally, the data processor concatenates the geometric feature channels of each surface point with the image channels of each grid point to generate a multi-channel feature image, and uses the multi-channel feature image as a fused image.

[0059] For example, a data processor can rasterize the point cloud on a 2D unfolded image at a preset grid resolution, and collect all 3D points (surface points) corresponding to each grid point of each 2D grid according to an index table. Simultaneously, it statistically analyzes and saves information such as the depth value and point count of each 2D grid. For multiple grid points falling into the same 2D grid, an aggregation strategy of averaging or weighted averaging can be used. The average depth value is calculated using the depth values ​​of each 2D grid, and an average depth channel is constructed based on the average depth value. The average depth channel is then normalized to enhance the dynamic range. For each grid point, a local plane is fitted using its corresponding 3D point (surface point) and neighboring points. A surface normal vector is calculated, and a surface normal vector channel is constructed based on the surface normal vector. The surface normal vector is then mapped onto the aforementioned 2D grayscale image using an angle-encoded method to obtain the image channel.

[0060] The average curvature is calculated based on the neighborhood covariance matrix method, and an average curvature channel is constructed based on the average curvature. Median filtering is then used to suppress outliers. The average depth channel, surface normal vector channel, average curvature channel, and image channels are concatenated channel by channel to generate a multi-channel feature image, which is then used as the fused image.

[0061] Referring to Figure 7, which is a flowchart illustrating another embodiment of the weld defect detection method based on multi-angle scanning and surface unfolding of this application, in some embodiments, the aforementioned step of inputting the fused image into a preset defect detection model to obtain defect detection results includes: step S210, inputting the fused image into the preset defect detection model; step S211, obtaining several candidate defect boxes output by the preset defect detection model and the defect category and defect confidence level corresponding to each candidate defect box; step S212, determining whether the defect confidence level corresponding to each candidate defect box is greater than a preset confidence threshold; step S213, when it is greater than the preset confidence threshold, determining the candidate defect box as the target defect box; step S214, generating defect detection results based on all target defect boxes and the defect category and defect confidence level corresponding to each target defect box.

[0062] In this embodiment, as shown in Figure 7, when the data processor executes step S150, it can first input the fused image into a preset defect detection model. The preset defect detection model can be pre-trained and configured by the user on the data processor, and can be an improved YOLOv8 defect detection model. The data processor can first input the fused image into the preset defect detection model. After receiving the fused image, the preset defect detection model can perform defect detection on the fused image, thereby obtaining several candidate defect boxes and the corresponding defect category and defect confidence score for each candidate defect box, and then outputting them. At this time, the data processor can obtain the several candidate defect boxes and the corresponding defect category and defect confidence score output by the preset defect detection model.

[0063] After obtaining several candidate defect boxes and their corresponding defect categories and confidence levels, the data processor can determine whether the defect confidence level of each candidate defect box is greater than a preset confidence threshold. If the confidence level is less than or equal to the preset confidence threshold, the candidate defect box is identified as a non-defect box. If the confidence level is greater than the preset confidence threshold, the candidate defect box is identified as a target defect box. The data processor can then generate defect detection results based on all target defect boxes and their corresponding defect categories and confidence levels.

[0064] For example, the preset defect detection model can be an improved YOLOv8 defect detection model; the preset defect detection model is a detection model that uses a dual-branch backbone network (texture branch backbone network and geometric branch backbone network) + multi-scale cross-attention fusion + FPN detection head.

[0065] The data processor can extract features from the fused image to obtain texture branch features and geometric branch features; then, the texture branch features and geometric branch features are input into a preset defect detection model. The preset defect detection model processes the texture branch features (grayscale and depth) and the geometric branch features (normal vector and curvature) separately. The primary layers of the two branches independently extract modal features while maintaining feature scale alignment.

[0066] A cross-attention mechanism is introduced in the feature fusion layer. Texture branch features are used as queries and geometric branch features are used as keys. After calculating attention weights, geometric information is weighted and converged onto texture features. Cross-attention units are inserted at each candidate scale (e.g., P3, P4, P5). A multi-head attention mechanism is adopted and residual connections and learnable scaling factors (γ) are added to stabilize training. The output contains both "enhanced texture features" and "geometric compensation features", realizing dynamic weighted fusion and information interaction between different modalities.

[0067] The data is then fed into the FPN detection head for multi-scale integration. The FPN detection head uses a lightweight FPN / PAFPN to aggregate multi-scale information. The detection head includes multiple branches such as classification, regression (boundary), and geometric consistency. Multi-scale detection is performed through a feature pyramid network structure. Finally, the detection box, category, and confidence score are output (several candidate defect boxes and the defect category and defect confidence score corresponding to each candidate defect box).

[0068] Finally, the candidate boxes output by the model are filtered using confidence thresholds and subjected to non-maximum suppression to obtain the target defect boxes. Based on all the target defect boxes and their corresponding defect categories and confidence levels, defect detection results are generated. The data processor can also perform geometric inversion and localization on the defect detection results. Using an index table, the data processor inverts the position of each target defect box in the two-dimensional unfolded map to a three-dimensional surface point cloud, obtaining the three-dimensional spatial localization information of the defect.

[0069] This application's technical solution uses several sets of triangular laser contour sensors with different incident angles to scan the weld area and obtain multi-view weld point clouds. These multi-view weld point clouds are then fused to obtain a three-dimensional surface point cloud. The three-dimensional surface point cloud is then unfolded to obtain a two-dimensional unfolded image. Finally, the three-dimensional surface point cloud and the two-dimensional unfolded image are fused to obtain a fused image. Finally, defect detection is performed on the fused image to obtain the defect detection result. Using several sets of triangular laser contour sensors with different incident angles to scan the weld area avoids the problems of incomplete data and the potential for blind spots. Furthermore, unfolding the three-dimensional surface point cloud to obtain a two-dimensional unfolded image, and then fusing the three-dimensional surface point cloud and the two-dimensional unfolded image, can improve the utilization rate of the geometric features of the three-dimensional surface point cloud in the two-dimensional unfolded image.

[0070] This application further proposes a multi-angle scanning and surface unfolding weld defect detection system. Referring to Figure 8, which is a schematic diagram of the structure of the multi-angle scanning and surface unfolding weld defect detection system according to an embodiment of this application, in some embodiments, the multi-angle scanning and surface unfolding weld defect detection system includes a data processor and several sets of triangular laser contour sensors with different incident angles; the multi-angle scanning and surface unfolding weld defect detection system can perform the above-described multi-angle scanning and surface unfolding weld defect detection method. In some embodiments, the multi-angle scanning and surface unfolding weld defect detection system further includes a motion unit, which is used to drive the target welding object to move. Each set of triangular laser contour sensors is fixedly configured, and the scanning range of each set of triangular laser contour sensors includes part or all of the surface area of ​​the motion unit.

[0071] In this embodiment, as shown in Figure 8, the multi-angle scanning and surface unfolding weld defect detection system includes a data processor and several sets of triangular laser profile sensors with different incident angles. The data processor is connected to each set of triangular laser profile sensors. The data processor can be a computer device or an embedded device. The multi-angle scanning and surface unfolding weld defect detection system also includes a motion unit and a mounting frame. The motion unit can be mounted at the bottom of the mounting frame, and each set of triangular laser profile sensors can be mounted at the top of the mounting frame. Each set of triangular laser profile sensors is fixedly configured, and the incident angles of each set of triangular laser profile sensors are different (i.e., the scanning angles of each set of triangular laser profile sensors are different). The scanning range of each set of triangular laser profile sensors includes part or all of the surface area of ​​the motion unit. Both the motion unit and each set of triangular laser profile sensors are connected to the data processor, enabling the data processor to control the motion unit and each set of triangular laser profile sensors.

[0072] This application further proposes a weld defect detection device based on multi-angle scanning and surface unfolding. Referring to Figure 9, which is a structural schematic diagram of the weld defect detection device based on an embodiment of this application, in some embodiments, the weld defect detection device based on multi-angle scanning and surface unfolding includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executed by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the weld defect detection method based on multi-angle scanning and surface unfolding as described above.

[0073] In this embodiment, referring to FIG9, the weld defect detection device for multi-angle scanning and surface unfolding according to this application embodiment can be a processor capable of running the weld defect detection method for multi-angle scanning and surface unfolding; there is at least one processor. As shown in FIG9, the weld defect detection device for multi-angle scanning and surface unfolding can include: a processor 1001 (e.g., CPU), a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 can include a display screen and an input unit, such as a keyboard. Optionally, the user interface 1003 can also include a standard wired interface or a wireless interface. The network interface 1004 can optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 can be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk storage device. Optionally, the memory 1005 can also be a storage device independent of the aforementioned processor 1001.

[0074] Those skilled in the art will understand that the multi-angle scanning and surface unfolding weld defect detection device structure shown in Figure 9 does not constitute a limitation on the multi-angle scanning and surface unfolding weld defect detection device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0075] As shown in Figure 9, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and computer programs.

[0076] In the weld defect detection device with multi-angle scanning and surface unfolding shown in Figure 9, the network interface 1004 is mainly used to connect to the back-end server and communicate with the back-end server; the user interface 1003 is mainly used to connect to the client (user end) and communicate with the client; and the processor 1001 can be used to call the computer program stored in the memory 1005. When the computer program is called and executed by the processor 1001, it implements the steps of the above-mentioned weld defect detection method with multi-angle scanning and surface unfolding.

[0077] The above description is only a part or preferred embodiment of this application. Neither the text nor the drawings should limit the scope of protection of this application. All equivalent structural transformations made using the content of this application's specification and drawings under the overall concept of this application, or direct / indirect applications in other related technical fields, are included within the scope of protection of this application.

Claims

1. A method for detecting weld defects using multi-angle scanning and surface unfolding, characterized in that, The multi-angle scanning and surface unfolding weld defect detection method is used in a multi-angle scanning and surface unfolding weld defect detection system. The system includes several sets of triangular laser contour sensors with different incident angles. The method comprises: controlling each set of triangular laser contour sensors to scan the weld area of ​​the target welded object to obtain multi-view weld point clouds; performing point cloud fusion processing on the multi-view weld point clouds to obtain a three-dimensional surface point cloud of the weld area; performing surface unfolding processing on the three-dimensional surface point cloud to obtain a two-dimensional unfolded image; fusing the three-dimensional surface point cloud and the two-dimensional unfolded image according to a preset fusion algorithm to obtain a fused image; and inputting the fused image into a preset defect detection model to obtain a defect detection result.

2. The weld defect detection method based on multi-angle scanning and surface unfolding according to claim 1, characterized in that, Before controlling each group of triangular laser contour sensors to scan the weld area of ​​the target welded object to obtain a multi-view weld point cloud, the method further includes: controlling each group of triangular laser contour sensors to scan a preset calibration plate to obtain standard three-dimensional point clouds, wherein the preset calibration plate includes multiple feature points with known spatial positions; extracting feature points from each standard three-dimensional point cloud to obtain each feature point of each standard three-dimensional point cloud; establishing a correspondence between each feature point of each standard three-dimensional point cloud based on the spatial position of each feature point on the preset calibration plate to obtain a spatial position correspondence; constructing a world coordinate system based on the preset calibration plate, and calculating the rigid body transformation matrix of each group of triangular laser contour sensors relative to the world coordinate system based on the spatial position correspondence to obtain the rigid body transformation matrix of each group of triangular laser contour sensors.

3. The weld defect detection method based on multi-angle scanning and surface unfolding according to claim 2, characterized in that, The multi-angle scanning and surface unfolding weld defect detection system also includes a motion unit, which is used to move the target welded object. Controlling each group of triangular laser contour sensors to scan the weld area of ​​the target welded object to obtain a multi-view weld point cloud includes: controlling the motion unit to move the target welded object at a constant speed according to a preset motion trajectory, and simultaneously controlling each group of triangular laser contour sensors to continuously sample the weld area; acquiring the displacement information of the motion unit and multiple frames of contour data continuously output by each group of triangular laser contour sensors; and sorting the multiple frames of contour data from each group of triangular laser contour sensors according to the displacement information to obtain the multi-view weld point cloud.

4. The weld defect detection method based on multi-angle scanning and surface unfolding according to claim 3, characterized in that, The step of performing point cloud fusion processing on the multi-view weld point cloud to obtain the three-dimensional surface point cloud of the weld region includes: mapping the multi-view weld point cloud to the world coordinate system according to the rigid body transformation matrix of each group of triangular laser contour sensors to obtain the mapped point cloud; calling the nearest neighbor search algorithm to perform neighborhood analysis on the mapped point cloud to obtain overlapping and non-overlapping regions; fusing the points in the overlapping region according to the weighted average strategy to obtain the overlapping fused point cloud; and integrating the overlapping fused point cloud and the point cloud of the non-overlapping region to obtain the three-dimensional surface point cloud.

5. The weld defect detection method based on multi-angle scanning and surface unfolding according to claim 4, characterized in that, The step of performing surface unfolding processing on the three-dimensional surface point cloud to obtain a two-dimensional unfolded diagram includes: extracting the three-dimensional surface point cloud based on the centerline of the weld region to obtain a discrete point sequence of the centerline; constructing a local coordinate system for each discrete point according to the discrete point sequence; for any surface point in the three-dimensional surface point cloud, querying the discrete point sequence to determine the target discrete point closest to the surface point; calculating the cumulative arc length from the surface point to the target discrete point as the abscissa; calculating the projection distance of the surface point in the local coordinate system of the target discrete point as the ordinate; generating the two-dimensional unfolded diagram based on the abscissa and ordinate of all the surface points, and establishing an index table of each point in the two-dimensional unfolded diagram and each surface point in the three-dimensional surface point cloud.

6. The weld defect detection method based on multi-angle scanning and surface unfolding according to claim 5, characterized in that, The step of fusing the three-dimensional surface point cloud and the two-dimensional unfolded map according to a preset fusion algorithm to obtain a fused image includes: rasterizing the two-dimensional unfolded map according to a preset grid resolution to obtain multiple two-dimensional grids; for any two-dimensional grid, determining the surface point corresponding to each grid point of the two-dimensional grid in the three-dimensional surface point cloud according to the index table; performing geometric feature calculation on all the surface points corresponding to each two-dimensional grid to obtain the geometric feature channels of each two-dimensional grid, wherein the geometric feature channels include an average depth channel, a surface normal vector channel, and an average curvature channel; mapping each grid point of each two-dimensional grid to an image channel; concatenating the geometric feature channels of each surface point with the image channels of each grid point to generate a multi-channel feature image, and using the multi-channel feature image as the fused image.

7. The weld defect detection method based on multi-angle scanning and surface unfolding according to claim 6, characterized in that, The step of inputting the fused image into a preset defect detection model to obtain a defect detection result includes: inputting the fused image into the preset defect detection model; obtaining a plurality of candidate defect boxes output by the preset defect detection model and the defect category and defect confidence level corresponding to each candidate defect box; determining whether the defect confidence level corresponding to each candidate defect box is greater than a preset confidence threshold; when it is greater than the preset confidence threshold, determining the candidate defect box as a target defect box; and generating the defect detection result based on all the target defect boxes and the defect category and defect confidence level corresponding to each target defect box.

8. A weld defect detection system with multi-angle scanning and surface unfolding, characterized in that, The multi-angle scanning and surface unfolding weld defect detection system includes a data processor and several sets of triangular laser contour sensors with different incident angles; the multi-angle scanning and surface unfolding weld defect detection system is capable of performing the multi-angle scanning and surface unfolding weld defect detection method according to any one of claims 1 to 7.

9. The weld defect detection system based on multi-angle scanning and surface unfolding according to claim 8, characterized in that, The multi-angle scanning and surface unfolding weld defect detection system also includes a motion unit, which is used to drive the target welded object to move. Each set of triangular laser contour sensors is fixedly configured, and the scanning range of each set of triangular laser contour sensors includes part or all of the surface area of ​​the motion unit.

10. A weld defect detection device using multi-angle scanning and curved surface unfolding, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executed by the at least one processor to enable the at least one processor to perform the weld defect detection method of multi-angle scanning and surface unfolding as described in any one of claims 1 to 7.

Images