Large-curvature thin-wall component cladding layer feature identification method based on three-dimensional point cloud information

Abstract

The invention discloses a large-curvature thin-wall component cladding layer feature recognition method based on three-dimensional point cloud information, and the method comprises the steps: carrying out the filtering and simplification preprocessing of the collected point cloud data of a repaired blade profile at each visual angle, and obtaining the blade point cloud data at each visual angle; performing multi-view point cloud splicing on the blade point cloud data to obtain point cloud data of a repaired blade profile; calculating an initial point cloud normal vector and an initial point cloud curvature of each point position, and redirecting the initial point cloud normal vector to obtain a repaired blade initial point cloud normal vector consistent in direction; re-calculating the normal vector of the repaired blade point cloud to perform feature enhancement on the normal vector of the initial point cloud so as to obtain the normal vector feature enhanced repaired blade profile point cloud data; clustering the cladding layer features in each region to obtain point cloud data of the cladding layer in each region; and carrying out classification and combination on the cladding layer point cloud of each region by using Euclidean distance clustering to obtain completely spliced repaired blade cladding layer point cloud data.

Description

Technical Field

[0001] This invention relates to, but is not limited to, the field of aerospace component repair technology, and in particular to a method for identifying the cladding layer features of thin-walled components with large curvature based on three-dimensional point cloud information. Background Technology

[0002] For blade damage incurred during service, additive manufacturing technology is typically used for repair. After additive repair, an irregularly shaped cladding layer remains on the blade surface. Due to the presence of this cladding layer, the repaired blade cannot meet the requirements for secondary service, necessitating reshaping to improve its surface quality and profile accuracy. Accurate measurement and identification of the cladding layer characteristics are prerequisites for reshaping. Therefore, the accuracy of 3D blade inspection affects the quality of cladding layer detection and surface reconstruction of the repaired blade, thus influencing the quality of blade reshaping.

[0003] Repairing the cladding layer on the blade surface is challenging due to its non-fixed location, complex geometry, and uneven residual height, making precise measurement and identification difficult. Furthermore, the leading and trailing edges and tips of aero-engine blades exhibit high curvature, leading to difficulties in identifying cladding features and blurred cladding layer boundaries. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying the cladding layer features of thin-walled components with large curvature based on three-dimensional point cloud information. This method addresses the challenges of accurately measuring and identifying the cladding layer on repair blades, which is characterized by its non-fixed position, complex geometry, and uneven residual height. It also addresses the difficulties in identifying cladding features and the ambiguity in identifying cladding layer boundaries.

[0005] The technical solution of this invention is as follows: To solve the above-mentioned technical problems, this invention provides a method for identifying the cladding layer features of thin-walled components with large curvature based on three-dimensional point cloud information, comprising the following steps: Step 1: Collect point cloud data of the actual repair blade profile from various viewpoints; Step 2: Filter and simplify the point cloud data of the real repair blade profile obtained in Step 1 from various perspectives to obtain filtered and simplified point cloud data from various perspectives. Step 3: Perform multi-view point cloud stitching on the point cloud data from each view after filtering and simplification in Step 2 to obtain the point cloud data for repairing the blade profile. Step 4: For the point cloud data of the repair blade profile obtained in Step 3, calculate the initial point cloud normal vector and initial point cloud curvature at each point position, and redirect the calculated initial point cloud normal vector at each point position to obtain the initial point cloud normal vector of the repair blade with the same direction. Step 5: Based on the initial point cloud normal vector of the repaired blade, the point cloud normal vector of the repaired blade is recalculated by introducing spatial distance weight and normal vector direction weight, so as to enhance the features of the initial point cloud normal vector in Step 4 and obtain the repaired blade profile point cloud data with enhanced normal vector features. Step 6: Divide the repair blade profile point cloud data enhanced with normal vector features in Step 5 into regions based on the repair blade curvature and normal vector information. Step 7: For the point cloud data in each region obtained in Step 6, apply the region growth factor algorithm to cluster the cladding layer features in each region to obtain the point cloud data of the cladding layer in each region. Step 8: Use Euclidean distance clustering to classify and merge the point clouds of the cladding layer of each region obtained in Step 7 to obtain the complete point cloud data of the cladding layer of the repair blade.

[0006] Optionally, in the above-described method for identifying the cladding layer features of a large-curvature thin-walled component based on three-dimensional point cloud information, step 2 involves preprocessing the point cloud data from various viewpoints, such as filtering and simplification, including: Step 21: Use a combined filtering method based on direct filtering and statistical filtering to remove irrelevant points and noise from the point cloud data at each viewpoint; Step 22: The point cloud data from each viewpoint after filtering is simplified using the voxel grid method; Step 23: The holes in the simplified point cloud data from each viewpoint are filled using a triangular mesh-based method, and finally the complete point cloud data of the repaired blade from each viewpoint is obtained.

[0007] Optionally, in the above-described method for identifying the cladding layer features of a large-curvature thin-walled component based on three-dimensional point cloud information, step 5 involves introducing spatial distance weights and normal vector direction weights to enhance the features of the initial point cloud normal vector of the repair blade obtained in step 4, resulting in repair blade profile point cloud data with enhanced normal vector features; wherein, the calculation formula for enhancing the features of the initial point cloud normal vector is as follows: ; ; ; in, This represents the current point in the point cloud dataset; Point , neighborhood points; Indicates the current point The normal vector; Representing neighborhood points The normal vector; Indicates the number of point clouds; Represents the normal vector of the current point. normal vectors of neighboring points Differences; This represents the spatial distance weight; points that are closer within the neighborhood have a higher weight. This represents the weight of the similarity in the direction of the normal vectors; the closer the directions of the normal vectors are, the greater the weight. The objective function representing the adjustment of the normal vector direction; This represents the parameter that controls the decay of spatial weights; Point and its neighboring points Euclidean distance; Represents the control parameters for normal vector similarity; Normal vector and Cosine similarity between them.

[0008] Optionally, in the above-described method for identifying the cladding layer features of a large-curvature thin-walled component based on three-dimensional point cloud information, In step 6, the repaired leaf surface point cloud data enhanced by the normal vector feature in step 5 is divided into four regions: leading edge point cloud region, trailing edge point cloud region, leaf basin point cloud region, and leaf back point cloud region.

[0009] Optionally, in the above-described method for identifying the cladding layer features of a large-curvature thin-walled component based on three-dimensional point cloud information, step 6 includes: Step 61: Based on the fact that the trailing edge is the region with the greatest curvature in the blade, the trailing edge point cloud region is segmented first by setting the trailing edge curvature threshold and the normal vector threshold. Step 62: Set the leading edge curvature threshold and normal vector threshold for the remaining point cloud data to segment the leading edge point cloud region; Step 63: Set a distance threshold in the remaining point cloud data after segmenting the leading edge point cloud region and the trailing edge point cloud region, and segment the leaf basin point cloud region and the leaf back point cloud region based on the distance threshold.

[0010] Optionally, in the method for identifying the cladding layer features of a large curvature thin-walled component based on three-dimensional point cloud information as described above, after step 63, the method further includes: Step 64: For the leading edge point cloud region and trailing edge point cloud region segmented in step 63, determine whether the direction of the point cloud normal vector of each point cloud data in the two point cloud regions is in the same direction or opposite to the reference direction of the point cloud region. If they are in the same direction, it is determined to be the current point cloud region; if they are opposite, it is confirmed to be another point cloud region, thereby accurately dividing the point cloud data of the leaf basin point cloud region and the leaf back point cloud region.

[0011] Optionally, in the method for identifying the cladding layer features of a large curvature thin-walled component based on three-dimensional point cloud information as described above, before step 61, the method further includes: Given the irregular geometry of the cladding layer on the repaired blade, the point cloud data of the cladding layer in each region satisfies both leading-edge and trailing-edge clustering conditions. To improve the segmentation accuracy of the leading and trailing-edge point cloud regions, the point cloud clusters with the largest number of cladding layer point clouds in each region are selected and marked as leading and trailing-edge point cloud data, while the remaining point cloud data are marked as point clouds to be segmented, thereby avoiding interference from the cladding layer point cloud data of other regions on the segmentation results of the leading and trailing-edge point clouds. The leading and trailing-edge point cloud regions are segmented using the marked leading and trailing-edge point cloud data.

[0012] Optionally, in the method for identifying the cladding layer features of a large-curvature thin-walled component based on three-dimensional point cloud information as described above, step 7 includes: The average curvature and average normal angle of the non-cladding layer region in each region are used as the curvature threshold and normal threshold for clustering, and the threshold selection of the growth factor in each region is optimized. In order to avoid the interference of the point cloud data of the cladding layer region in each region on the clustering threshold, the points participating in the calculation of the average value in each region are screened by setting the mean and standard deviation, and the points with large deviations in the cladding layer region in each region are filtered out to avoid their influence on the mean, and the average value is recalculated.

[0013] Optionally, in the method for identifying the cladding layer features of a large-curvature thin-walled component based on three-dimensional point cloud information as described above, in step 7, The method to optimize the selection of growth factor thresholds for each region is to replace the growth factor threshold for that region with the average curvature and average normal vector angle of the non-cladding layer region within each region. The average values ​​used in the calculation for each region are: the average curvature and average normal vector of all points within each region; all points include the cladding layer region and the non-cladding layer region within the region.

[0014] The beneficial effects of this invention are as follows: This invention provides a method for identifying the cladding layer features of thin-walled components with large curvature based on three-dimensional point cloud information. The method involves filtering and simplifying the point cloud data of the collected repair blade profile from various viewpoints to obtain blade point cloud data from each viewpoint; stitching the blade point cloud data together from multiple viewpoints to obtain point cloud data of the repair blade profile; calculating the initial point cloud normal vector and initial point cloud curvature at each point position, and redirecting the initial point cloud normal vector to obtain initial point cloud normal vectors of the repair blade with consistent directions; recalculating the repair blade point cloud normal vector to enhance the features of the initial point cloud normal vector, resulting in point cloud data of the repair blade profile with enhanced normal vector features; clustering the cladding layer features in each region to obtain point cloud data of the cladding layer in each region; and classifying and merging the point clouds of the cladding layer in each region using Euclidean distance clustering to obtain stitched complete point cloud data of the repair blade cladding layer. The cladding layer feature recognition method for thin-walled components with large curvature provided by this invention can effectively solve the problems of difficult cladding layer feature recognition and blurred cladding layer boundary recognition in thin-walled components with large curvature, improve the accuracy of cladding layer feature recognition for repaired components, and provide support for their subsequent processing. The technical solution provided by this invention has the following beneficial effects: First, this invention enhances the normal vector features of the blade point cloud by introducing spatial distance weights and normal vector direction weights, which can better identify the features of the blade cladding layer.

[0015] Secondly, this invention divides the blade into four regions based on its geometric features: the leading edge, trailing edge, leaf base, and leaf back. This provides a foundation for accurate identification of the cladding features of the entire blade.

[0016] Third, this invention uses the average curvature and average normal angle of the non-cladding region in each area of ​​the blade as the curvature threshold and normal threshold for clustering, thereby optimizing the selection of the threshold for the regional growth factor and improving the accuracy of cladding layer identification. Attached Figure Description

[0017] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of the present invention and do not constitute a limitation on the technical solutions of the present invention.

[0018] Figure 1 The flowchart illustrates a method for identifying the cladding layer features of a large curvature thin-walled component based on three-dimensional point cloud information, as provided in an embodiment of the present invention.

[0019] Figure 2 This is an example image of point cloud data of the actual repair blade profile obtained by scanning with a robotic line laser 3D measurement platform in an embodiment of the present invention, viewed from a specific perspective. Figure 3This is an example diagram of the repair of point cloud data of the blade from various perspectives in an embodiment of the present invention; Figure 4 This is an example diagram of point cloud data of the repaired blade profile after stitching together point cloud data from various perspectives in an embodiment of the present invention; Figure 5 This is an example diagram of the initial normal vector of the blade point cloud data in an embodiment of the present invention; Figure 6 This is an example diagram of the normal vector result after the initial point cloud normal vector is redirected in an embodiment of the present invention; Figure 7 This is an example diagram of the normal vector result after feature enhancement of the initial point cloud normal vector in an embodiment of the present invention; Figure 8 This is an example diagram showing the region partitioning results of the repair blade profile point cloud data enhanced by normal vector feature enhancement in an embodiment of the present invention; Figure 9 This is an example diagram showing the clustering results of the cladding layer features in each region in an embodiment of the present invention; Figure 10 This is an example diagram of complete point cloud data for repairing the cladding layer of a blade in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

[0021] As explained in the background section, the cladding layer on the repaired blade surface is characterized by its non-fixed location, complex geometry, and uneven residual height, making it difficult to accurately measure and identify. Furthermore, the leading and trailing edges and tips of aero-engine blades exhibit high curvature, leading to difficulties in identifying cladding features and blurred cladding layer boundaries. Therefore, achieving accurate identification of the cladding layer features across the entire blade is a critical technical problem that urgently needs to be solved.

[0022] To address the aforementioned issues, embodiments of the present invention provide a method for identifying the cladding layer features of thin-walled components with high curvature based on three-dimensional point cloud information.

[0023] The present invention provides the following specific embodiments, which can be combined with each other. For the same or similar concepts or processes, they may not be described again in some embodiments.

[0024] Figure 1 This is a flowchart illustrating a method for identifying the cladding layer features of a full-blade according to an embodiment of the present invention. Figure 1 As shown in the figure, this invention provides a method for identifying the cladding layer features of thin-walled components with large curvature based on three-dimensional point cloud information, comprising the following steps: Step 1: Collect point cloud data of the actual repair blade profile from various viewpoints; Step 2: Filter and simplify the point cloud data of the real repair blade profile obtained in Step 1 from various perspectives to obtain filtered and simplified point cloud data from various perspectives. Step 3: Perform multi-view point cloud stitching on the point cloud data from each view after filtering and simplification in Step 2 to obtain the point cloud data for repairing the blade profile. Step 4: For the point cloud data of the repair blade profile obtained in Step 3, calculate the initial point cloud normal vector and initial point cloud curvature at each point position, and redirect the calculated initial point cloud normal vector at each point position to obtain the initial point cloud normal vector of the repair blade with the same direction. Step 5: Based on the initial point cloud normal vector of the repaired blade, the point cloud normal vector of the repaired blade is recalculated by introducing spatial distance weight and normal vector direction weight, so as to enhance the features of the initial point cloud normal vector in Step 4 and obtain the repaired blade profile point cloud data with enhanced normal vector features. Step 6: Divide the repair blade profile point cloud data enhanced with normal vector features in Step 5 into regions based on the repair blade curvature and normal vector information. Step 7: For the point cloud data in each region obtained in Step 6, apply the region growth factor algorithm to cluster the cladding layer features in each region to obtain the point cloud data of the cladding layer in each region. Step 8: Use Euclidean distance clustering to classify and merge the point clouds of the cladding layer of each region obtained in Step 7 to obtain the complete point cloud data of the cladding layer of the repair blade.

[0025] This invention focuses on thin-walled components with high curvature, such as blades, and proposes a method for identifying additive cladding features of these components based on 3D point cloud information. The method includes the following steps: using a robot carrying a high-precision line laser and a motorized precision turntable to perform 3D scanning of the blade profile; preprocessing the scanned point cloud data by filtering, simplification, and stitching to obtain a complete blade profile point cloud; calculating the normal vector and curvature of the blade point cloud, and enhancing the blade's normal features by setting spatial distance weights and normal vector angle weights; dividing the blade into regions based on the curvature and normal vectors of each region; using the average curvature and average normal vector angle of the non-cladding regions within each region as the threshold for region growth factor clustering to identify cladding features in different regions; and fusing the cladding features of each region using Euclidean clustering to obtain the full blade cladding feature information. This invention effectively solves the problems of difficult cladding feature identification and ambiguous cladding boundary identification in thin-walled components with high curvature, improving the accuracy of cladding feature identification in repaired components and providing support for their subsequent processing.

[0026] The following is a specific embodiment illustrating the detailed implementation of each step in the method for identifying the cladding layer features of large curvature thin-walled components based on three-dimensional point cloud information provided by the present invention.

[0027] See Figure 1 As shown, to address the challenges of accurately measuring and identifying repaired blades due to the non-fixed position, complex geometry, and uneven residual height of the cladding layer, as well as difficulties in identifying cladding features and blurred cladding layer boundaries, the method includes the following steps: Step 1: A 3D measurement platform for the blade is built using a robot-assisted line laser and an electric precision turntable. Hand-eye calibration is performed on the robot and line laser, and the axis of the electric precision turntable is calibrated. Then, by setting a series of scanning perspectives, the actual blade profile is scanned in 3D from each perspective to obtain point cloud data of the actual blade profile at each perspective, such as... Figure 2 The image shown is an example of point cloud data of the actual repair blade profile obtained by scanning with a robotic line laser 3D measurement platform in an embodiment of the present invention.

[0028] Step 2: Perform preprocessing such as filtering and simplification on the point cloud data of the real repair blade profile obtained in Step 1 from various perspectives.

[0029] The specific processing method for this step is as follows: A combined filtering method based on direct-pass filtering and statistical filtering is used to remove irrelevant points and noise from the point cloud data at each viewpoint. Then, the voxel mesh method is used to simplify the filtered point cloud data at each viewpoint. Next, a triangular mesh method is used to fill the holes in the simplified point cloud data at each viewpoint, finally obtaining the point cloud data of the repaired blade at each viewpoint, such as... Figure 3 The image shown is an example diagram of the repair of point cloud data of the blade from various perspectives in an embodiment of the present invention.

[0030] Step 3: Perform multi-view point cloud stitching on the filtered and simplified point cloud data obtained in Step 2. Use coarse stitching based on the turntable axis and fine stitching using the ICP algorithm to obtain complete point cloud data for the repaired blade profile; such as Figure 4 The image shown is an example of point cloud data of the repaired blade profile after point cloud data splicing from various perspectives in an embodiment of the present invention.

[0031] Step 4: For the point cloud data of the repaired blade profile obtained in Step 3, local fitting is used to calculate the initial point cloud normal vector and initial point cloud curvature at each point location. However, the directions of the calculated initial point cloud normal vectors at each point location are not consistent, such as... Figure 5The image shown is an example diagram of the initial normal vector of the blade point cloud data in an embodiment of the present invention. A normal vector retargeting method based on minimum spanning tree is used to retarget the initial point cloud normal vectors at each point location, resulting in repaired blade initial point cloud normal vectors with consistent directions, as shown below. Figure 6 The image shown is an example of the normal vector result after the initial point cloud normal vector is redirected in an embodiment of the present invention.

[0032] Step 5: Due to the irregular geometry of the cladding layer on the repaired blade, the normal vectors of the cladding layer point cloud are chaotic, and the difference between the point cloud normal vectors at the cladding layer boundary and the normal vectors of the repaired blade surface is blurred. To address this issue, based on the initial point cloud normal vectors of the repaired blade obtained in Step 4, spatial distance weights and normal vector direction weights are introduced to enhance the features of the initial point cloud normal vectors in Step 4, resulting in point cloud data of the repaired blade profile with enhanced normal vector features. The specific method for enhancing the features of the initial point cloud normal vectors in Step 4 is as follows: ; ; ; in, This represents the current point in the point cloud dataset; Point , neighborhood points; Indicates the current point The normal vector; Representing neighborhood points The normal vector; Indicates the number of point clouds; Represents the normal vector of the current point. normal vectors of neighboring points Differences; This represents the spatial distance weight; points that are closer within the neighborhood have a higher weight. This represents the weight of the similarity in the direction of the normal vectors; the closer the directions of the normal vectors are, the greater the weight. The objective function representing the adjustment of the normal vector direction; This represents the parameter that controls the decay of spatial weights; Point and its neighboring points Euclidean distance; This represents the control parameters for normal vector similarity; Normal vector and Cosine similarity between them.

[0033] After introducing spatial distance weights and normal vector direction weights to enhance the features of the initial point cloud normal vectors in step 4, the resulting repaired blade profile point cloud data has more obvious normal vector features, i.e., repaired blade profile point cloud data with enhanced normal vector features, such as... Figure 7 The image shown is an example of the normal vector result after feature enhancement of the initial point cloud normal vector in an embodiment of the present invention.

[0034] Step 6: Based on the curvature and normal vector information of the repaired blade, the point cloud data of the repaired blade profile enhanced in Step 5 is divided into four regions: leading edge point cloud region, trailing edge point cloud region, leaf base point cloud region, and leaf underside point cloud region. Figure 8 The image shown is an example of the region partitioning results of the repair blade profile point cloud data enhanced by normal vector feature enhancement in an embodiment of the present invention.

[0035] The region segmentation method in this step is as follows: Since the trailing edge is the region with the greatest curvature in the blade, the trailing edge point cloud region is segmented first by setting a trailing edge curvature threshold and a normal vector threshold. Then, the leading edge point cloud region is segmented by setting a leading edge curvature threshold and a normal vector threshold for the remaining point cloud data. It should be noted that the curvature threshold and normal vector threshold for the leading and trailing edges are manually set based on the geometric characteristics of the repaired blade. Finally, in the remaining point cloud after segmenting the leading and trailing edge point cloud regions, a distance threshold is set to segment the leaf base point cloud region and the leaf back point cloud region based on the distance threshold, and the direction of the point cloud normal vector is determined to be the same as or opposite to the established reference direction to distinguish between the leaf base and the leaf back.

[0036] It should be noted that the leaf basin point cloud region and the leaf back point cloud region are also based on the method of manually set distance thresholds. Since the radius of curvature of the trailing edge point cloud region is the smallest in the entire leaf point cloud, the leaf basin point cloud region and the leaf back point cloud region can be separated as long as the set distance threshold is greater than the distance threshold of the trailing edge point cloud region.

[0037] Step 64: For the leading edge point cloud region and trailing edge point cloud region segmented in step 63, determine whether the direction of the point cloud normal vector of each point cloud data in the two point cloud regions is in the same direction or opposite to the reference direction of the point cloud region. If they are in the same direction, it is determined to be the current point cloud region; if they are opposite, it is confirmed to be another point cloud region, thereby accurately dividing the point cloud data of the leaf basin point cloud region and the leaf back point cloud region.

[0038] In this step, for the leaf basin point cloud region and leaf back point cloud region segmented in step 63, it is determined whether the direction of the point cloud normal vector is in the same direction or opposite to the artificially established reference direction to distinguish between the leaf basin and the leaf back. If they are in the same direction, it is determined to be this point cloud region; if they are opposite, it is confirmed to be another point cloud region, thereby accurately dividing the point cloud data of the leaf basin point cloud region and the leaf back point cloud region.

[0039] It should also be noted that, due to the irregular geometry of the cladding layer on the repaired blade, the point cloud data of the cladding layer in each region may satisfy both the leading-edge clustering and trailing-edge clustering conditions. Therefore, in order to improve the segmentation accuracy of the leading and trailing-edge regions, the following steps can be performed before step 61 above: select the point cloud cluster with the largest number of cladding layer point clouds in each region and mark it as the leading and trailing-edge point cloud data, and mark the remaining point cloud data as the point clouds to be segmented, thereby avoiding interference from the cladding layer point cloud data of other regions on the segmentation results of the leading and trailing-edge point clouds; and use the marked leading and trailing-edge point cloud data to segment the leading and trailing-edge point cloud regions.

[0040] Step 7: For the point cloud data within the four regions (leading edge, trailing edge, leaf base, and leaf underside) obtained in Step 6, apply the region growth factor algorithm to cluster the cladding layer features within each region, obtaining the point cloud data of the cladding layer in each region, such as... Figure 9 The image shown is an example diagram illustrating the clustering results of the cladding layer features in each region in an embodiment of the present invention.

[0041] The specific implementation of this step is as follows: The average curvature and average normal angle of the non-cladding layer regions within each region are used as the curvature and normal vector thresholds for clustering. The selection of the region growth factor threshold is optimized by replacing the region growth factor threshold with the average curvature and average normal angle of the non-cladding layer regions within each region. Furthermore, to avoid interference from the cladding layer region point cloud on the clustering threshold, the points participating in the average calculation are filtered by setting the mean and standard deviation. Points with large deviations, such as those in the cladding layer region, are filtered out, and the average value is recalculated. The average value participating in the calculation within each region is: the average curvature and average normal vector of all points within each region; all points include both cladding layer and non-cladding layer regions within the region.

[0042] Step 8: The point clouds of the cladding layer in each region obtained from the clustering in Step 7 are classified and merged using Euclidean distance clustering to obtain a complete, stitched-together point cloud data of the cladding layer of the repaired blade; For example... Figure 10 The image shown is an example diagram of complete point cloud data for repairing the cladding layer of a blade in an embodiment of the present invention.

[0043] This invention provides a method for identifying the cladding layer features of thin-walled components with large curvature based on three-dimensional point cloud information. The method involves filtering and simplifying the point cloud data of the collected repair blade profile from various viewpoints to obtain blade point cloud data from each viewpoint; stitching the blade point cloud data together from multiple viewpoints to obtain the point cloud data of the repair blade profile; calculating the initial point cloud normal vector and initial point cloud curvature at each point position, and redirecting the initial point cloud normal vector to obtain initial point cloud normal vectors of the repair blade with consistent direction; recalculating the repair blade point cloud normal vector to enhance the features of the initial point cloud normal vector, resulting in repair blade profile point cloud data with enhanced normal vector features; clustering the cladding layer features in each region to obtain the point cloud data of the cladding layer in each region; and classifying and merging the cladding layer point clouds in each region using Euclidean distance clustering to obtain the stitched complete point cloud data of the repair blade cladding layer. The cladding layer feature recognition method for thin-walled components with large curvature provided by this invention can effectively solve the problems of difficult cladding layer feature recognition and blurred cladding layer boundary recognition in thin-walled components with large curvature, improve the accuracy of cladding layer feature recognition for repaired components, and provide support for their subsequent processing. The technical solution provided by this invention has the following beneficial effects: First, this invention enhances the normal vector features of the blade point cloud by introducing spatial distance weights and normal vector direction weights, which can better identify the features of the blade cladding layer.

[0044] Secondly, this invention divides the blade into four regions based on its geometric features: the leading edge, trailing edge, leaf base, and leaf back. This provides a foundation for accurate identification of the cladding features of the entire blade.

[0045] Third, this invention uses the average curvature and average normal angle of the non-cladding region in each area of ​​the blade as the curvature threshold and normal threshold for clustering, thereby optimizing the selection of the threshold for the regional growth factor and improving the accuracy of cladding layer identification.

[0046] While the embodiments disclosed in this invention are as described above, they are merely illustrative of the embodiments to facilitate understanding of the invention and are not intended to limit the invention. Any person skilled in the art to which this invention pertains may make any modifications and variations in the form and details of the implementation without departing from the spirit and scope disclosed herein; however, the scope of patent protection for this invention shall still be determined by the scope defined in the appended claims.

Claims

1. A method for identifying the cladding layer features of thin-walled components with large curvature based on three-dimensional point cloud information, comprising: Step 1: Collect point cloud data of the actual repair blade profile from various viewpoints; Step 2: Filter and simplify the point cloud data of the real repair blade profile obtained in Step 1 from various perspectives to obtain filtered and simplified point cloud data from various perspectives. Step 3: Perform multi-view point cloud stitching on the point cloud data from each view after filtering and simplification in Step 2 to obtain the point cloud data for repairing the blade profile. Step 4: For the point cloud data of the repair blade profile obtained in Step 3, calculate the initial point cloud normal vector and initial point cloud curvature at each point position, and redirect the calculated initial point cloud normal vector at each point position to obtain the initial point cloud normal vector of the repair blade with the same direction. Step 5: Based on the initial point cloud normal vector of the repaired blade, the point cloud normal vector of the repaired blade is recalculated by introducing spatial distance weight and normal vector direction weight, so as to enhance the features of the initial point cloud normal vector in Step 4 and obtain the repaired blade profile point cloud data with enhanced normal vector features. Step 6: Divide the repair blade profile point cloud data enhanced with normal vector features in Step 5 into regions based on the repair blade curvature and normal vector information. Step 7: For the point cloud data in each region obtained in Step 6, apply the region growth factor algorithm to cluster the cladding layer features in each region to obtain the point cloud data of the cladding layer in each region. Step 8: Use Euclidean distance clustering to classify and merge the point clouds of the cladding layer of each region obtained in Step 7 to obtain the complete point cloud data of the cladding layer of the repair blade.

2. The method for identifying the cladding layer features of a large curvature thin-walled component based on three-dimensional point cloud information according to claim 1, characterized in that, Step 2 involves preprocessing the point cloud data from various viewpoints, including filtering and simplification. Step 21: Use a combined filtering method based on direct filtering and statistical filtering to remove irrelevant points and noise from the point cloud data at each viewpoint; Step 22: The point cloud data from each viewpoint after filtering is simplified using the voxel grid method; Step 23: The holes in the simplified point cloud data from each viewpoint are filled using a triangular mesh-based method, and finally the complete point cloud data of the repaired blade from each viewpoint is obtained.

3. The method for identifying the cladding layer features of a large curvature thin-walled component based on three-dimensional point cloud information according to claim 1, characterized in that, In step 5, spatial distance weights and normal vector direction weights are introduced to enhance the features of the initial point cloud normal vectors of the repaired blade obtained in step 4, resulting in repaired blade profile point cloud data with enhanced normal vector features. The calculation formula for enhancing the features of the initial point cloud normal vectors is as follows: ； ； ； in, This represents the current point in the point cloud dataset; Point , neighborhood points; Indicates the current point The normal vector; Representing neighborhood points The normal vector; Indicates the number of point clouds; Represents the normal vector of the current point. normal vectors of neighboring points Differences; This represents the spatial distance weight; points that are closer within the neighborhood have a higher weight. This represents the weight of the similarity in the direction of the normal vectors; the closer the directions of the normal vectors are, the greater the weight. The objective function representing the adjustment of the normal vector direction; This represents the parameter that controls the decay of spatial weights; Point and its neighboring points Euclidean distance; This represents the control parameters for normal vector similarity; Normal vector and Cosine similarity between them.

4. The method for identifying the cladding layer features of a large curvature thin-walled component based on three-dimensional point cloud information according to claim 3, characterized in that, In step 6, the repaired leaf surface point cloud data enhanced by the normal vector feature in step 5 is divided into four regions: leading edge point cloud region, trailing edge point cloud region, leaf basin point cloud region, and leaf back point cloud region.

5. The method for identifying the cladding layer features of a large curvature thin-walled component based on three-dimensional point cloud information according to claim 4, characterized in that, Step 6 includes: Step 61: Based on the fact that the trailing edge is the region with the greatest curvature in the blade, the trailing edge point cloud region is segmented first by setting the trailing edge curvature threshold and the normal vector threshold. Step 62: Set the leading edge curvature threshold and normal vector threshold for the remaining point cloud data to segment the leading edge point cloud region; Step 63: Set a distance threshold in the remaining point cloud data after segmenting the leading edge point cloud region and the trailing edge point cloud region, and segment the leaf basin point cloud region and the leaf back point cloud region based on the distance threshold.

6. The method for identifying the cladding layer features of a large curvature thin-walled component based on three-dimensional point cloud information according to claim 5, characterized in that, After step 63, the method further includes: Step 64: For the leading edge point cloud region and trailing edge point cloud region segmented in step 63, determine whether the direction of the point cloud normal vector of each point cloud data in the two point cloud regions is in the same direction or opposite to the reference direction of the point cloud region. If they are in the same direction, it is determined to be the current point cloud region; if they are opposite, it is confirmed to be another point cloud region, thereby accurately dividing the point cloud data of the leaf basin point cloud region and the leaf back point cloud region.

7. The method for identifying the cladding layer features of a large curvature thin-walled component based on three-dimensional point cloud information according to claim 5, characterized in that, Before step 61, the method also includes: Given the irregular geometry of the cladding layer on the repaired blade, the point cloud data of the cladding layer in each region satisfies both leading-edge and trailing-edge clustering conditions. To improve the segmentation accuracy of the leading and trailing-edge point cloud regions, the point cloud clusters with the largest number of cladding layer point clouds in each region are selected and marked as leading and trailing-edge point cloud data, while the remaining point cloud data are marked as point clouds to be segmented, thereby avoiding interference from the cladding layer point cloud data of other regions on the segmentation results of the leading and trailing-edge point clouds. The leading and trailing-edge point cloud regions are segmented using the marked leading and trailing-edge point cloud data.

8. A method for identifying the cladding layer features of a large curvature thin-walled component based on three-dimensional point cloud information according to any one of claims 1 to 7, characterized in that, Step 7 includes: The average curvature and average normal angle of the non-cladding layer region in each region are used as the curvature threshold and normal threshold for clustering, and the threshold selection of the growth factor in each region is optimized. In order to avoid the interference of the point cloud data of the cladding layer region in each region on the clustering threshold, the points participating in the calculation of the average value in each region are screened by setting the mean and standard deviation, and the points with large deviations in the cladding layer region in each region are filtered out to avoid their influence on the mean, and the average value is recalculated.

9. The method for identifying the cladding layer features of a large curvature thin-walled component based on three-dimensional point cloud information according to claim 8, characterized in that, In step 7 The method to optimize the selection of growth factor thresholds for each region is to replace the growth factor threshold for that region with the average curvature and average normal vector angle of the non-cladding layer region within each region. The average values ​​used in the calculation for each region are: the average curvature and average normal vector of all points within each region; all points include the cladding layer region and the non-cladding layer region within the region.

Images