Method and system for identifying burrs on surface of high-pressure casting through mechanical arm laser scanning

Abstract

The invention provides a high-pressure casting surface burr identification method and system through mechanical arm laser scanning, and relates to the technical field of high-pressure casting quality detection.The mechanical arm drives a laser scanner to conduct multi-angle three-dimensional scanning on a high-pressure casting, and complete and continuous surface point cloud information is obtained; furthermore, curvature change and normal vector mutation characteristics of the point cloud are combined, and a characteristic normalization and dynamic threshold judgment method is adopted, so that automatic identification and marking of edge mutation areas (namely burrs) are realized, and the identification rate of the burrs of the complex curved surface and the system robustness are effectively improved. The technical problems that the burrs of the high-pressure casting are difficult to quantify, detection is unstable, and integration is difficult are solved, a complete detection scheme with engineering practicability is constructed, and efficient and accurate recognition of the burr defects on the surface of the high-pressure casting is achieved.

Description

Technical Field

[0001] This invention relates to the field of high-pressure casting quality inspection technology, belonging to the cross-technology fields of intelligent manufacturing, industrial vision inspection, and three-dimensional data processing. In particular, it relates to a method and system for identifying surface burrs of high-pressure castings using laser scanning with a robotic arm. Background Technology

[0002] High-pressure casting, as a highly efficient metal forming process, relies on high pressure to rapidly fill the mold cavity with molten metal. This process boasts significant advantages such as high production efficiency and good dimensional accuracy, making it particularly suitable for producing thin-walled parts with complex structures. However, due to the high-pressure, high-speed filling characteristics, defects such as flash and burrs inevitably occur on the surface of the casting. These defects not only affect the product's appearance but can also become stress concentration points, reducing the fatigue life of structural components.

[0003] Currently, the industry mainly uses two inspection methods to address the above problems: traditional manual inspection methods and emerging machine vision technology. Manual inspection relies on operators using tools such as calipers and plug gauges for sampling measurements. Although the equipment cost is low, it has inherent drawbacks such as low inspection efficiency and strong subjectivity. While machine vision systems based on two-dimensional images achieve automated inspection, their two-dimensional projection characteristics make it impossible to accurately identify spatial defect features when dealing with castings with complex three-dimensional features.

[0004] With the development of 3D sensor and robotic arm technologies, integrating laser scanners with six-axis robots and using point cloud data for part size detection and defect analysis has gradually become a mainstream approach. The underlying principle is as follows: the laser scanner performs non-contact scanning of the target surface using structured light, laser stripes, or flying spot methods, outputting a 3D coordinate point cloud; the robotic arm performs multi-angle sampling based on the workpiece structure to ensure data coverage; and computer vision technology is used to reconstruct, filter, register, and perform difference analysis on the point cloud data to extract defects. Point cloud defect detection methods mainly include: height difference-based, depth map-based edge detection, CAD difference analysis-based, and voxel sparsity analysis-based methods. However, these methods still have many limitations in practical applications, particularly: they are prone to misjudgment when the workpiece surface structure is complex and the edges are sharp; detection is unstable when point cloud sampling is uneven or noisy; and threshold adjustment is difficult, making it hard to adapt to different workpieces and working conditions. Summary of the Invention

[0005] To overcome the problems of low accuracy, poor automation, and insufficient adaptability to complex geometric surfaces in existing high-pressure casting burr detection technologies, this invention proposes a method and system for identifying surface burrs in high-pressure castings using robotic arm laser scanning. By employing a multi-dimensional geometric feature fusion detection mechanism, combined with a dynamic threshold and edge point derivation mechanism, this invention solves the technical pain points of difficulty in quantifying burrs in high-pressure castings, unstable detection, and integration difficulties. It constructs a complete detection scheme with engineering practicality, achieving efficient and accurate identification of surface burr defects in high-pressure castings.

[0006] Therefore, the present invention provides the following technical solution: On one hand, the present invention provides a method for identifying surface burrs on high-pressure castings using laser scanning with a robotic arm, comprising the following steps: A six-axis industrial robotic arm drives a laser scanner to perform multi-angle 3D scanning of high-pressure castings and obtain complete surface point cloud data. For the surface point cloud data, the normal vector of each point is calculated using a point cloud normal vector estimation method based on principal component analysis, and the local curvature is calculated by eigenvalue decomposition of the covariance matrix. The rate of curvature change and the rate of change of normal vector at each point are calculated based on the local curvature and then normalized. The normalized rate of curvature change and the rate of normal vector change are weighted and fused to obtain the fused feature score; Using a dynamic threshold method, the jagged edges region is automatically determined based on the statistical distribution of the fused feature scores, and the detection results are output.

[0007] Furthermore, the multi-angle three-dimensional scanning includes: The robotic arm rotates and scans around the high-pressure casting according to a preset trajectory to ensure that all surfaces of the casting are covered. The point cloud data output from the scan is downsampled to a 0.3mm interval using voxel filtering, and noise points are removed using a statistical outlier removal method.

[0008] Furthermore, the normal vector estimation method includes: For each point, construct its nearest neighbor set and perform decentralized processing; Calculate the neighborhood covariance matrix and perform eigenvalue decomposition. The eigenvector corresponding to the smallest eigenvalue is the normal vector of that point. The magnitude of local normal variation is measured by the average of the included angles of the neighboring normals.

[0009] Furthermore, the calculation of the local curvature includes: The curvature estimate is calculated based on the ratio of the smallest eigenvalue of the covariance matrix to the sum of its eigenvalues. Calculate the difference between the curvature of the current point and the curvature of its neighbors, and use it as the rate of change of curvature; The greater the rate of change of curvature, the sharper the local surface at that point.

[0010] Furthermore, the calculation of the rate of change of the normal vector includes: Calculate the average angle between the current point and the neighborhood normals; The rate of change of the normal vector is used to measure the severity of local directional changes.

[0011] Furthermore, the feature fusion includes: The rate of change of curvature and the rate of change of normal vector are respectively subjected to minimum-maximum normalization. The fusion feature score is calculated using a weighted fusion method, where the curvature change rate has a weight of 0.8 and the normal vector change rate has a weight of 0.2.

[0012] Furthermore, the dynamic threshold method includes: The threshold is calculated using the percentile method based on the global statistical distribution of the fused feature scores. The condition for determining edge points is that the fusion feature score is higher than the 90th percentile threshold; The threshold is adjusted adaptively without manual intervention.

[0013] Furthermore, it also includes a verification step for the burr detection results: The detection results are compared with manually labeled areas to calculate accuracy and recall.

[0014] In another aspect, the present invention also provides a system for identifying surface burrs on high-pressure castings using laser scanning with a robotic arm, comprising: A six-axis industrial robotic arm is used to drive a laser scanner to perform multi-angle scanning. Laser scanners are used to acquire three-dimensional point cloud data of high-pressure castings; The data processing module is used to perform the following operations on the point cloud data: First, it calculates the normal vector of each point using a point cloud normal vector estimation method based on principal component analysis, and calculates the local curvature through eigenvalue decomposition of the covariance matrix. Based on the local curvature, it calculates the rate of change of curvature and the rate of change of normal vector for each point and performs normalization. Then, it performs weighted fusion of the normalized rate of change of curvature and the rate of change of normal vector to obtain a fused feature score. Finally, it uses a dynamic threshold method to automatically determine the jagged edge region based on the statistical distribution of the fused feature score and outputs the detection result. The results display module is used to visualize the results of burr detection.

[0015] Furthermore, it also includes: A custom-designed connection fixture between the robotic arm and the laser scanner ensures stability during the scanning process; Workpiece clamping platform, used to fix high-pressure castings and adaptable to workpieces of different sizes and shapes.

[0016] Advantages and positive effects of the present invention: The present invention uses a robotic arm to drive a laser scanner to perform multi-angle three-dimensional scanning of high-pressure castings to obtain complete and continuous surface point cloud information; further, by combining the curvature change and normal vector mutation characteristics of the point cloud, a feature normalization and dynamic threshold judgment method is adopted to realize the automatic identification and marking of edge mutation regions (i.e. burrs), which effectively improves the recognition rate of burrs on complex curved surfaces and the robustness of the system. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a structural block diagram of a high-pressure casting surface burr detection system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the robotic arm structure and its trajectory in an embodiment of the present invention; Figure 3 This is a schematic diagram of the scanner installation in an embodiment of the present invention; Figure 4 This is a structural diagram of a customized connecting clamp in an embodiment of the present invention; Figure 5 This is a structural diagram of a high-pressure casting in an embodiment of the present invention; Figure 6 This is a flowchart of a method for detecting burrs on the surface of high-pressure castings according to an embodiment of the present invention; Figure 7 This is a first schematic diagram of the detection results in an embodiment of the present invention (the red point cloud represents the detected edge points); Figure 8 This is a second schematic diagram of the detection results in an embodiment of the present invention (the red point cloud represents the detected edge points); Figure 9 This is a third schematic diagram of the detection results in an embodiment of the present invention (the red point cloud represents the detection edge points). Detailed Implementation

[0019] This invention proposes a method and system for detecting burrs on high-pressure castings based on point cloud data. The method uses a robotic arm to control a laser scanner to acquire three-dimensional point cloud information of the high-pressure casting and utilizes curvature differences and normal vector changes for burr detection. This approach combines the geometric gradient and spatial orientation changes of point cloud features, effectively identifying minute defects such as burrs and flash on the surface of complex high-pressure castings.

[0020] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0021] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] like Figure 1 As shown, a high-pressure casting surface burr identification system utilizing a robotic arm and laser scanning comprises the following core components: a six-axis industrial robotic arm, a laser scanner, a robotic arm-scanner connection assembly, and a workpiece clamping and support platform. Among them, such as Figure 2 As shown, the robotic arm is a six-degree-of-freedom serial structure, possessing multi-angle positioning and attitude control capabilities in space. Its trajectory planning can achieve comprehensive scanning of any position on the casting surface. The robotic arm's end effector is equipped with a standard flange interface, which connects to a laser scanner via a custom connector to ensure assembly accuracy and stability.

[0023] The laser scanner is an integrated 3D measurement device with high sampling frequency and sub-millimeter accuracy. It is mounted on the end effector of a robotic arm and synchronously acquires data as the arm moves. The scanner has mounting holes and a data cable interface, and can communicate with the control unit via Ethernet. A schematic diagram of the scanner installation is shown below. Figure 3 As shown. Its scanning method is area array or linear array structured light, which forms a point cloud data set on the surface of the workpiece within the scanning path. The output data format supports standard .pcd, .ply and other common formats.

[0024] The robotic arm and scanner connection assembly is a set of custom-designed connection fixtures used to achieve stable assembly between a standard robotic arm and a non-standard scanner, such as... Figure 4As shown, the connecting fixture includes: a flange transition plate that matches the flange hole positions at the end of the robotic arm; a fixed bracket adapted to the scanner housing structure; and screw-on fasteners for eliminating assembly gaps and preventing vibration-induced displacement. After installation, the connecting assembly has a low overall center of gravity, ensuring stable equipment posture and eliminating dynamic vibration during scanning.

[0025] The workpiece (high-pressure casting) is placed on a fixed platform or adaptive fixture. The three-dimensional structural model of the alloy die casting is as follows: Figure 5 As shown, these types of parts often exhibit typical defects such as burrs and flash at the mold joint lines or around internal openings, requiring rapid screening after demolding to ensure consistent quality. The platform has a certain degree of adjustability to accommodate workpieces of different sizes and shapes. The typical surface of the workpiece has complex undulations and sharp angles, making it highly susceptible to minute defects such as burrs and flash, necessitating high-density point cloud scanning and high-resolution detection.

[0026] The above system adopts a robotic arm + laser scanning linkage structure to achieve high-precision and high-coverage casting scanning, which is suitable for casting inspection scenarios with various shapes and complex curved surfaces.

[0027] The system described above utilizes a robotic arm carrying a laser scanner to scan surfaces. Combined with point cloud geometric feature calculations and edge extraction algorithms, it successfully identified and marked the minute burrs on the surfaces of multiple aluminum alloy die-cast parts. For example... Figure 5 As shown, the surface complexity of the object being inspected includes porous, multi-faceted, and deep cavity structures; the target defect size is 1mm to 5mm (burrs / flashes). Figure 6 As shown, the method for identifying rough edges specifically includes the following steps: Step 1: Use a six-axis industrial robotic arm to drive a laser scanner to perform multi-angle 3D scanning on the high-pressure casting to obtain complete surface point cloud data.

[0028] The robotic arm rotates and scans around the workpiece according to a preset trajectory, collecting complete point cloud data. The total point cloud volume is approximately 1,500,000 points. The point cloud data is saved in .pcd format and input into the point cloud processing module. The collected 3D point cloud data can be represented as follows: ; in, Indicates the first The three-dimensional coordinate vector of a point They are respectively in Coordinate values ​​on the axis This represents the total number of points in the point cloud.

[0029] The original point cloud was downsampled to a voxel interval of 0.3 mm using a Voxel filter; isolated noise points were removed using statistical outlier removal (nb_neighbors=50, std_ratio=2.0).

[0030] Step 2: For the surface point cloud data, the normal vector of each point is calculated using the point cloud normal vector estimation method based on principal component analysis, and the local curvature is calculated by eigenvalue decomposition of the covariance matrix. The core method used in this invention is based on the principle of geometric feature extraction of point clouds, especially curvature changes and abrupt changes in normal vector direction.

[0031] The normal vector of each point in the point cloud is obtained by calculating the principal axes of the neighborhood covariance matrix using principal component analysis (PCA). The variation in normal vectors in edge regions is typically greater than that in smooth surface regions, which can be reflected by the average of the neighborhood normal vector angles. ; This method is rotationally invariant and can identify edge features at different angles.

[0032] In practice, the neighborhood search is performed first: For each point , construct its Nearest neighbor set: .

[0033] Secondly, the neighborhood points are decentralized, and the covariance matrix is ​​calculated: ; in It is the center point of the neighborhood point set.

[0034] Finally, regarding the covariance matrix Eigenvalue decomposition yields three eigenvalues. and the corresponding eigenvectors, where the smallest eigenvalue is... The corresponding feature vector is that point. normal vector .

[0035] Curvature is a geometric quantity that describes the degree of bending of a local surface at a point in three dimensions. For a neighborhood of a point, its curvature estimate can be calculated using the eigenvalues ​​of the covariance matrix. ; in The minimum eigenvalue represents the change in the direction of the normal. The greater the curvature, the sharper the surface. It is often used to identify areas such as edges, openings, and protrusions.

[0036] This value reflects the smoothness of the local surface at that point; rough areas typically exhibit larger curvature changes.

[0037] Step 3: Calculate the rate of change of curvature and the rate of change of normal vector at each point based on the local curvature, and then perform normalization processing; (1) Calculation of the rate of change of curvature: at each point Calculate the curvature jump value (i.e., the difference from the neighborhood mean): ; This metric measures the consistency of the curvature of the current point with its neighborhood; jagged areas typically have a larger curvature. .

[0038] (2) Calculation of the rate of change of the normal vector: using the estimated normal vector Calculate the average angle between the current point and the neighborhood normals as the rate of change of the normal: ; in This represents a dot product operation. All angles are measured in degrees or radians (units are consistent). The larger the value, the more drastic the change in local direction.

[0039] More specifically, the neighborhood parameter is set as follows: Curvature estimation neighborhood: ; Normal radius estimation: Maximum number of neighbors: 10.

[0040] Step 4: Calculate the rate of change of curvature and the rate of change of normal vector at each point based on the local curvature, and perform normalization processing; perform weighted fusion of the normalized rate of change of curvature and rate of change of normal vector to obtain the fused feature score; To comprehensively consider the effects of curvature and normal changes, both are normalized and weighted and then fused: (1) Normalization process: Using the minimum-maximum scaling method: ; (2) Weighted fusion: Use parameters Set the weights (usually adjustable to 0.8) and calculate the fusion score: .

[0041] Step 5: Using the dynamic threshold method, the rough edge region is automatically determined based on the statistical distribution of the fused feature scores, and the detection results are output.

[0042] Dynamic thresholding for edge extraction: Thresholds are calculated using percentile methods (e.g., the 90th percentile). The set of edge points is determined as follows: ; Edge points are shown in red, while the remaining points are shown in gray, for use in 3D visualization.

[0043] This embodiment employs a multi-dimensional point cloud feature fusion algorithm, which integrates curvature differences and normal vector angle changes, making it more robust than traditional methods based solely on height or color differences. This embodiment also utilizes a dynamic threshold determination mechanism to avoid false positives and false negatives caused by using fixed thresholds, and can adaptively adjust according to the characteristics of the point cloud itself.

[0044] In another embodiment, it may also include: Step 6: Result verification and performance evaluation; If labeled point cloud data is provided, the detection accuracy can be further evaluated. The red areas are compared with manually labeled data, and the following performance metrics are calculated: Accuracy: ; Recall: ; in:

[0045] Visualization results as follows Figure 7 , 8 As shown in Figure 9, the red area represents the successfully detected jagged point cloud; compared with manually labeled jagged areas, the detection accuracy reached 93.7%, and the recall rate was 88.1%.

[0046] The key technical parameters in the above embodiments are shown in Table 1.

[0047] Table 1

[0048] Compared with existing point cloud detection methods based on single height difference, image edge transformation, or CAD model comparison, the point cloud detection method in the above embodiments introduces the fusion feature of curvature difference and normal vector change as the basis for identifying burrs on high-pressure castings. Combined with a robotic arm-controlled laser scanning acquisition of full-coverage point clouds, this method achieves significant improvements in detection accuracy, adaptability, efficiency, and data structuring. Specific advantages and their formation mechanisms are as follows: (1) It has the ability to identify jagged edges with multi-dimensional geometric features, thereby improving detection accuracy and robustness. Traditional detection methods based on height difference or depth map edges can only demonstrate detection capability under specific directions or lighting conditions. When faced with high-pressure castings with complex structures or severely curved surfaces, they are prone to missed detections and false detections.

[0049] This invention uses two types of three-dimensional geometric features as indicators for burr detection: Curvature jump value (Δκ): reflects sharp changes in local surface morphology; Normal angle variation (Δθ): reflects the abrupt change in the uniformity of the surface orientation.

[0050] These two types of indicators correspond to the "first-order" and "second-order" geometric change characteristics of a point, respectively, and their combination provides a more stable basis for edge detection.

[0051] (2) It adopts a dynamic threshold mechanism, which has adaptive capability and reduces dependence on manual parameter adjustment. Traditional algorithms often rely on manually setting fixed thresholds (such as height difference > x mm), which require frequent parameter adjustments under different point cloud densities, scanning angles, and workpiece sizes, resulting in poor versatility.

[0052] This invention introduces a percentile threshold mechanism, based on the fused feature score. The global statistical distribution is used to adaptively set the threshold for edge point determination.

[0053] Edge points are defined as:

[0054] When applied to different sizes, scanning densities, and casting types, no parameters need to be modified, and the F1 index fluctuates within a range of less than ±3%.

[0055] By dynamically identifying "abnormally high" points using statistical methods, without relying on specific numerical settings, the system avoids false positives or false negatives caused by biased threshold settings, thus enhancing the universality and batch deployment capability of the detection solution.

[0056] (3) Applicable to multi-angle scanning inspection of workpieces with complex structures, improving inspection coverage. Image detection methods rely on a single viewpoint and suffer from occlusion problems. CAD comparison methods require high alignment accuracy and struggle to handle workpieces with multiple free-form surfaces and internal cavities.

[0057] This invention employs a six-axis robotic arm to control a laser scanner to move and sample around the workpiece at multiple angles, achieving complete point cloud acquisition of the entire surface and blind spots of the high-pressure casting, and then performing inspection. Combined with robotic arm trajectory planning and fixture angle switching, it can achieve full-coverage inspection of castings with complex structures.

[0058] Total time for scanning and inspecting a single workpiece: 8~15 seconds Point cloud coverage: ≥98% (an improvement of >40% compared to manual single-sided scanning) (4) The method and equipment are highly adaptable and applicable to a variety of castings and related process parts. Since the algorithm relies on the geometric features of the point cloud (which are independent of color, material, texture, etc.), the method of this invention is not only applicable to high-pressure aluminum castings, but also to other process parts with edge defect features, such as: burr detection of parting line of injection molded parts; weld bead detection of welded parts; burr analysis of powder metallurgy parts; and support residue identification of metal printed parts.

[0059] The system described in the above embodiments is suitable for online quality inspection of high-pressure cast aluminum parts such as automobiles, motorcycles, power tools, and pump bodies. The system supports the replacement of scanners and adaptation to different arm types, such as ABB and KUKA. The inspection process can be extended to similar surface anomaly analysis tasks such as weld spatter detection, coating boundary analysis, and burr identification of injection molded parts. If the target is a small or medium-sized precision part, a desktop six-axis collaborative robot equipped with a micro scanning module can be used to reduce costs.

[0060] This invention provides a highly efficient, stable, and automated technical solution for detecting burrs on high-pressure castings, which provides strong support for realizing a closed-loop intelligent quality inspection and manufacturing process for high-pressure casting products. It has significant engineering practical value and promising prospects for industrial promotion.

[0061] 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 or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying surface burrs on high-pressure castings using laser scanning with a robotic arm, characterized in that, Includes the following steps: A six-axis industrial robotic arm drives a laser scanner to perform multi-angle 3D scanning of high-pressure castings and obtain complete surface point cloud data. For the surface point cloud data, the normal vector of each point is calculated using a point cloud normal vector estimation method based on principal component analysis, and the local curvature is calculated by eigenvalue decomposition of the covariance matrix. The rate of curvature change and the rate of change of normal vector at each point are calculated based on the local curvature and then normalized. The normalized rate of curvature change and the rate of normal vector change are weighted and fused to obtain the fused feature score; Using a dynamic threshold method, the jagged edges region is automatically determined based on the statistical distribution of the fused feature scores, and the detection results are output.

2. The method according to claim 1, characterized in that, The multi-angle three-dimensional scan includes: The robotic arm rotates and scans around the high-pressure casting according to a preset trajectory to ensure that all surfaces of the casting are covered. The point cloud data output from the scan is downsampled to a 0.3mm interval using voxel filtering, and noise points are removed using a statistical outlier removal method.

3. The method according to claim 1, characterized in that, The normal vector estimation method includes: For each point, construct its nearest neighbor set and perform decentralized processing; Calculate the neighborhood covariance matrix and perform eigenvalue decomposition. The eigenvector corresponding to the smallest eigenvalue is the normal vector of that point. The magnitude of local normal variation is measured by the average of the included angles of the neighboring normals.

4. The method according to claim 1, characterized in that, The calculation of the local curvature includes: The curvature estimate is calculated based on the ratio of the smallest eigenvalue of the covariance matrix to the sum of its eigenvalues. Calculate the difference between the curvature of the current point and the curvature of its neighbors, and use it as the rate of change of curvature; The greater the rate of change of curvature, the sharper the local surface at that point.

5. The method according to claim 1, characterized in that, The calculation of the rate of change of the normal vector includes: Calculate the average angle between the current point and the neighborhood normals; The rate of change of the normal vector is used to measure the severity of local directional changes.

6. The method according to claim 1, characterized in that, The feature fusion includes: The rate of change of curvature and the rate of change of normal vector are respectively subjected to minimum-maximum normalization. The fusion feature score is calculated using a weighted fusion method, where the curvature change rate has a weight of 0.8 and the normal vector change rate has a weight of 0.

2.

7. The method according to claim 1, characterized in that, The dynamic threshold method includes: The threshold is calculated using the percentile method based on the global statistical distribution of the fused feature scores. The condition for determining edge points is that the fusion feature score is higher than the 90th percentile threshold; The threshold is adjusted adaptively without manual intervention.

8. The method according to claim 1, characterized in that, It also includes a verification step for the burr detection results: The detection results are compared with manually labeled areas to calculate accuracy and recall.

9. A system for identifying surface burrs on high-pressure castings using laser scanning with a robotic arm, characterized in that, include: A six-axis industrial robotic arm is used to drive a laser scanner to perform multi-angle scanning. Laser scanners are used to acquire three-dimensional point cloud data of high-pressure castings; The data processing module is used to perform the following operations on the point cloud data: calculate the normal vector of each point using a point cloud normal vector estimation method based on principal component analysis, and calculate the local curvature through eigenvalue decomposition of the covariance matrix; calculate the rate of curvature change and the rate of change of normal vector of each point based on the local curvature, and perform normalization processing. The normalized rate of curvature change and the rate of normal vector change are weighted and fused to obtain the fused feature score; The dynamic threshold method is used to automatically determine the rough edge region based on the statistical distribution of the fused feature scores, and the detection results are output. The results display module is used to visualize the results of burr detection.

10. The system according to claim 9, characterized in that, Also includes: A custom-designed connection fixture between the robotic arm and the laser scanner ensures stability during the scanning process; Workpiece clamping platform, used to fix high-pressure castings and adaptable to workpieces of different sizes and shapes.

Images