Aero-engine blade complex component three-dimensional curved surface reconstruction method and blade crown contour extraction method

Abstract

The invention discloses an aero-engine blade complex component three-dimensional curved surface reconstruction method and a blade crown contour extraction method, and belongs to the technical field of three-dimensional point cloud data processing. The problems that an existing blade three-dimensional curved surface reconstruction method cannot adjust and control the scanning height in a self-adaptive mode according to needs, and reconstruction and contour extraction precision is poor are solved. The curved surface reconstruction method comprises the steps of obtaining a corresponding relation of calibration circle pixel numbers under different known height differences; determining the height of the blade relative to the workbench under the condition that the number of the pixels of the two calibration circles is known, determining the scanning height of the line laser measuring instrument, and scanning to obtain original point cloud data of the blade; preprocessing the original point cloud data of the blade to obtain a preprocessed blade point cloud; and carrying out three-dimensional curved surface reconstruction on the preprocessed blade point cloud based on an optimized leading edge propulsion algorithm, and carrying out reconstruction optimization to obtain a complete and continuous blade three-dimensional curved surface. The method is used for blade complex component three-dimensional curved surface reconstruction and blade crown contour extraction.

Description

Technical Field

[0001] This invention relates to a method for reconstructing the three-dimensional curved surface of complex components of aero-engine blades and a method for extracting the blade crown contour, belonging to the field of three-dimensional point cloud data processing technology. Background Technology

[0002] As a crucial component of aero-engines, blades endure various complex operating conditions during operation, including high-load aerodynamic pressure, high-temperature combustion gas scouring, airflow vibration, and foreign object impacts. Ensuring their structural strength and wear resistance is vital for the safe operation of aero-engines. In such circumstances, regional welding repair of the blades can improve the surface hardness and wear resistance of the affected area, reduce wear, enhance sealing performance, and decrease blade vibration amplitude and noise levels. This is of great significance for ensuring the safe operation of aero-engine blades.

[0003] In the repair of aero-engine blades, the large volume of welding repairs, high quality requirements, and the small and complex surface of the blade crown welding repair area, coupled with the inherent challenges of manual repair (high workload, poor accuracy, and inconsistent quality), necessitate an automated welding repair method. The primary task in implementing an automated welding repair system is the automatic planning of 3D surface reconstruction and welding trajectories for various complex blade components. While 3D surface reconstruction technology is widely used in blade repair across aviation, wind power, and shipbuilding industries, it suffers from insufficient reconstruction accuracy, low data processing efficiency, inadequate automation, and poor adaptability. Due to the diverse morphologies of complex aero-engine blade components, adaptive adjustments to the scanning height are necessary for different components. Furthermore, the large angle at the connection between the blade crown and the blade body leads to excessively sparse point cloud data under various scanning strategies, significantly impacting surface reconstruction and requiring optimization. Finally, to achieve path planning for the blade crown welding repair area, high-precision blade crown contour extraction is required, laying the foundation for selecting appropriate path points during the welding path planning process. Summary of the Invention

[0004] To address the problems of existing three-dimensional surface reconstruction methods for airfoils, which cannot adaptively adjust the scanning height as needed and have poor reconstruction and contour extraction accuracy, this invention provides a method for three-dimensional surface reconstruction of complex components of aero-engine airfoils and a method for extracting the airfoil crown contour.

[0005] The present invention provides a method for reconstructing the three-dimensional curved surface of a complex component of an aero-engine blade, comprising:

[0006] By setting calibration circles on the worktable and the blade respectively, the corresponding relationship of the number of pixels of the calibration circle under different known height differences is obtained; then, given the known number of pixels of the two calibration circles, the height of the blade relative to the worktable is determined, thereby determining the scanning height of the line laser measuring instrument, and scanning to obtain the original point cloud data of the blade;

[0007] Preprocess the original point cloud data of the blade to obtain the preprocessed point cloud of the blade;

[0008] An optimized frontier propulsion algorithm is used to reconstruct the three-dimensional surface of the preprocessed blade point cloud, and the reconstruction is optimized to obtain a complete and continuous three-dimensional surface of the blade.

[0009] According to the three-dimensional surface reconstruction method for complex components of aero-engine blades of the present invention, the method for determining the scanning height of the line laser measuring instrument is as follows:

[0010] Images of the calibration circles on the worktable and blades at different height differences are acquired using a camera; after grayscale transformation and Gaussian filtering of the images, Canny edge detection is performed to obtain the pixel contours of the calibration circles on the worktable and the blades.

[0011] Define the gradient operator Gx0 in the horizontal direction and the gradient operator Gy0 in the vertical direction, and calculate the gradient magnitude threshold M(x0, y0) and the direction θ0(x0, y0):

[0012] ,

[0013] ,

[0014] In the formula The x-coordinate of the pixel is The ordinate of the pixel;

[0015] Calculate the gradient magnitude of adjacent pixels in the θ0 (x0, y0) direction of the outline of the calibration circle pixel point on the worktable or the outline of the calibration circle pixel point on the blade. If the gradient magnitude exceeds the gradient magnitude threshold M (x0, y0), the corresponding pixel point is retained as the edge point of the calibration circle; otherwise, it is deleted. The outline of the edge point of the calibration circle on the worktable and the outline of the edge point of the calibration circle on the blade are obtained.

[0016] The least squares method is used to extract the circular contours of the calibration circle edge points on the worktable and the calibration circle edge points on the blade to obtain the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade.

[0017] A BP neural network optimized by random forest is used to establish a mapping model of pixel number-object distance. The mapping model is trained by using the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade under different known height differences. Based on the number of pixels of the two calibration circles obtained under the current height difference, the height of the blade relative to the worktable is determined by the mapping model of pixel number-object distance, and the scanning height of the line laser measuring instrument is determined so that the blade is within the scanning height range of the line laser measuring instrument.

[0018] The method for reconstructing the three-dimensional curved surface of complex components of an aero-engine blade according to the present invention includes a method for preprocessing the original point cloud data of the blade, comprising:

[0019] The original point cloud data of the blade is transformed from the point spacing on the laser line to the point spacing distributed along the scanning direction, and then subjected to pass-through filtering and statistical filtering. Statistical filtering uses the K-nearest neighbor algorithm to select the n nearest neighbors of the center point, calculates the average distance, and calculates the standard deviation of the distance. A distance threshold is set based on the average distance and the standard deviation. The average distance between the center point and the n nearest neighbors is calculated, and center points with an average distance greater than the set distance threshold are considered outliers and removed. The filtered point cloud data is then obtained.

[0020] Voxel downsampling is performed on the filtered point cloud data, and the centroid of the voxel space is calculated as the representative point to obtain the preprocessed blade point cloud.

[0021] The method for reconstructing the three-dimensional surface of complex components of aero-engine blades according to the present invention includes a method for reconstructing the three-dimensional surface of preprocessed blade point clouds based on an optimized leading-edge propulsion algorithm, comprising:

[0022] Local neighborhood smoothing was performed on the preprocessed blade point cloud using second-order Jet polynomial surface fitting.

[0023] ,

[0024] In the formula Let x, y, z be the coordinates of the blade point cloud after preprocessing in three-dimensional space, and a0, a1, ..., a9 be the coefficients of the polynomial to be solved.

[0025] Determine the neighborhood points of the point cloud (x, y, z), and use the least squares method to make the neighborhood points reach the z-axis. Minimize the sum of squared distances and minimize the error function. :

[0026]

[0027] In the formula Let a0, a1, ..., a9 be the number of neighboring points.

[0028] Then p i (x) j y j , z j Substitute In the middle, we obtain a new point p. i ′, using the new point p i ′replace p i ; and then use the new point p i Using ′ as a new neighborhood point, iterative calculations are performed to obtain the fitted surface;

[0029] Based on the fitted surface, normals are estimated, and the normal direction is determined by the minimum spanning tree algorithm according to the angle between the normals of adjacent point clouds.

[0030] Then, a leading-edge propulsion algorithm based on Delaunay triangulation theory is employed. Based on the normal direction, a starting point is selected from the preprocessed blade point cloud. An initial triangular facet is constructed using the two adjacent point clouds of the starting point as indices. The side length of the triangular facet is adaptively adjusted according to the blade point cloud density function ρ(x). Make the side length It is negatively correlated with the leaf point cloud density function ρ(x):

[0031]

[0032] Using the three sides of the triangular face as the initial front edge, the front edge is expanded under the set constraints to obtain two expanded point clouds. Then, the starting point and the two expanded point clouds form a new triangular face. The front edge expansion is repeated until the front edge is empty or cannot be expanded, and the reconstructed three-dimensional surface is obtained.

[0033] The method for reconstructing the three-dimensional surface of complex components of aero-engine blades according to the present invention includes the following method for optimizing the reconstruction of the three-dimensional surface:

[0034] For non-topological holes in the reconstructed 3D surface, determine the smallest cube containing all vertices of the non-topological hole, calculate the size of the smallest cube in each dimension, and determine whether the diameter of the non-topological hole exceeds a set diameter threshold based on the size of the smallest cube in each dimension.

[0035] For non-topological holes that do not exceed the set diameter threshold, Delaunay triangulation is performed to divide the non-topological holes into multiple triangular faces, thereby achieving a rough filling of the non-topological holes to obtain a filled three-dimensional surface.

[0036] After filling, the 3D surface is reduced by midpoint subdivision to reduce surface discontinuities and sharp features, and the mesh is smoothed by Laplacian smoothing.

[0037] ,

[0038] In the formula It refers to the vertex point cloud position in the filled 3D surface. Let λ be the position of the vertex point cloud after smoothing, and λ be the smoothing factor. for The set of adjacent vertex point clouds;

[0039] This results in a complete and continuous three-dimensional blade surface.

[0040] This invention also provides a method for extracting the blade crown profile of an aero-engine, comprising:

[0041] By setting calibration circles on the worktable and the blade respectively, the corresponding relationship of the number of pixels of the calibration circle under different known height differences is obtained; then, given the known number of pixels of the two calibration circles, the height of the blade relative to the worktable is determined, thereby determining the scanning height of the line laser measuring instrument, and scanning to obtain the original point cloud data of the blade;

[0042] Preprocess the original point cloud data of the blade to obtain the preprocessed point cloud of the blade;

[0043] Clustering and segmentation are performed on the preprocessed leaf point cloud to obtain the leaf crown region point cloud. Curvature-based point cloud edge point extraction is performed on the leaf crown region point cloud, followed by edge feature point extraction. The edge feature points are used as segmentation points of the point cloud edge points. The cubic Hermite interpolation method is used to fit curves for each segment of the point cloud edge points to obtain the leaf crown point cloud contour fitting curve.

[0044] According to the method for extracting the blade crown profile of an aero-engine according to the present invention, the method for determining the scanning height of a line laser measuring instrument is as follows:

[0045] Images of the calibration circles on the worktable and blades at different height differences are acquired using a camera; after grayscale transformation and Gaussian filtering of the images, Canny edge detection is performed to obtain the pixel contours of the calibration circles on the worktable and the blades.

[0046] Define the gradient operator Gx0 in the horizontal direction and the gradient operator Gy0 in the vertical direction, and calculate the gradient magnitude threshold M(x0, y0) and the direction θ0(x0, y0):

[0047] ,

[0048] ,

[0049] The x-coordinate of the pixel is The ordinate of the pixel;

[0050] Calculate the gradient magnitude of adjacent pixels in the θ0 (x0, y0) direction of the outline of the calibration circle pixel point on the worktable or the outline of the calibration circle pixel point on the blade. If the gradient magnitude exceeds the gradient magnitude threshold M (x0, y0), the corresponding pixel point is retained as the edge point of the calibration circle; otherwise, it is deleted. The outline of the edge point of the calibration circle on the worktable and the outline of the edge point of the calibration circle on the blade are obtained.

[0051] The least squares method is used to extract the circular contours of the calibration circle edge points on the worktable and the calibration circle edge points on the blade to obtain the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade.

[0052] A BP neural network optimized by random forest is used to establish a mapping model of pixel number-object distance. The mapping model is trained by using the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade under different known height differences. Based on the number of pixels of the two calibration circles obtained under the current height difference, the height of the blade relative to the worktable is determined by the mapping model of pixel number-object distance, and the scanning height of the line laser measuring instrument is determined so that the blade is within the scanning height range of the line laser measuring instrument.

[0053] According to the method for extracting the blade crown contour of an aero-engine according to the present invention, the method for preprocessing the original point cloud data of the blade includes:

[0054] The original point cloud data of the blade is transformed from the point spacing on the laser line to the point spacing distributed along the scanning direction, and then subjected to pass-through filtering and statistical filtering. Statistical filtering uses the K-nearest neighbor algorithm to select the n nearest neighbors of the center point, calculates the average distance, and calculates the standard deviation of the distance. A distance threshold is set based on the average distance and the standard deviation. The average distance between the center point and the n nearest neighbors is calculated, and center points with an average distance greater than the set distance threshold are considered outliers and removed. The filtered point cloud data is then obtained.

[0055] Voxel downsampling is performed on the filtered point cloud data, and the centroid of the voxel space is calculated as the representative point to obtain the preprocessed blade point cloud.

[0056] The method for extracting the blade crown contour of an aero-engine according to the present invention provides the following method for obtaining the point cloud of the blade crown region:

[0057] The DBSCAN density point cloud clustering algorithm is adopted. By traversing the preprocessed leaf point cloud, the number of point cloud points in the neighborhood radius eps of each point cloud is calculated. If the number of point cloud points is not less than the set minimum number of points MinPts, the current point cloud is marked as the current core point; otherwise, it is marked as a non-core point. Then, the core points and non-core points in the neighborhood of the current core point are recursively classified into the canopy region cluster, thereby obtaining the canopy region point cloud from the preprocessed leaf point cloud.

[0058] The method for extracting edge points in a point cloud is as follows:

[0059] A curvature-based point cloud edge point extraction method is adopted, and a neighborhood search is performed using a Kd tree to obtain a set of neighboring points in the leaf canopy region point cloud to form a neighboring point set. The covariance matrix C of the neighboring point set is then calculated.

[0060]

[0061] In the formula Let be the set of neighboring points, and q be the neighboring points in the set of neighboring points. Let be the centroid of the nearest neighbor set;

[0062] Find the eigenvalues ​​λ1, λ2, and λ3 of the covariance matrix C, such that λ1 ≥ λ2 ≥ λ3;

[0063] Calculate the curvature κ of the nearest point q:

[0064]

[0065] If the curvature κ exceeds the set curvature threshold, then the neighboring point q will be taken as the edge point of the point cloud.

[0066] The edge feature point extraction method is as follows:

[0067] The SUSAN algorithm is used to extract edge feature points from the point cloud edge points. Within a set radius r, a neighborhood search is performed using a Kd tree to obtain the neighborhood point set. The Euclidean distance d(P,Q) between the center point cloud edge point P in the neighborhood point set and other points Q in the neighborhood point set is calculated.

[0068] ,

[0069] In the formula Let x be the three-dimensional coordinates of point P at the edge of the center point cloud. Q y Q z Q Let Q be the three-dimensional coordinates of point Q;

[0070] Calculate the normal angle θ between point P and point Q at the edge of the center point cloud:

[0071]

[0072] In the formula Let P be the normal vector of the edge point P of the center point cloud. Let be the normal vector of point Q;

[0073] Calculate the intensity difference I(P, Q) between point P and point Q at the edge of the center point cloud:

[0074] ,

[0075] In the formula The local intensity of point P at the edge of the center point cloud. Let Q be the local intensity.

[0076] The Euclidean distance d(P,Q), normal vector angle θ, and intensity difference I(P,Q) are compared with the corresponding set thresholds. Points Q whose comparison results are all less than the corresponding set thresholds are retained. If the number of retained points Q is less than the set threshold, the corresponding points Q are taken as edge feature points.

[0077] The method for extracting the blade crown profile of an aero-engine according to the present invention includes the following steps for obtaining the blade crown point cloud profile fitting curve:

[0078] Using edge feature points as segmentation points and other point cloud edge points as base points, the leaf crown edge curve is segmented and fitted. Each pair of adjacent segmentation points serves as the two ends of a fitted curve segment.

[0079] For each segment of the fitted curve, the cubic Hermite interpolation method was used for curve fitting:

[0080] Each fitted curve segment includes U point cloud edge points. Calculate the tangent vector between two adjacent point cloud edge points:

[0081] ,

[0082] ,

[0083] In the formula The number of base points in each fitted curve segment. For the first Tangent vectors at each fundamental point For the first +1 tangent vector at the base point For the first The coordinates of the base points;

[0084] Define weight function , , and :

[0085] ,

[0086] Where t∈[0,1] is the uniform sampling parameter, defined In the formula For the number of interpolation points, For interpolation points, v = 0, 1, ..., N-1;

[0087] Calculate the interpolation point function value H(t):

[0088] ,

[0089] Combining the interpolation point function value H(t) and the number of interpolation points The system uses smoothing and sorting auxiliary functions to obtain the fitting result of the current segment's fitted curve, and then obtains the leaf crown point cloud contour fitting curve.

[0090] The beneficial effects of this invention are: This invention is used for image mapping adaptive surface reconstruction optimization and contour extraction of complex components of aero-engine blades.

[0091] The method of this invention features high data acquisition accuracy and strong adaptability. By adaptively adjusting the scanning height through a neural network mapping model of pixel count versus object distance, it avoids the sparse or distorted point cloud caused by fixed-height scanning, and can adapt to various complex components of aero-engine blades.

[0092] The method of this invention achieves a high degree of completeness in the surface reconstruction. By adaptively adjusting the triangulation side length and using a reconstruction optimization method, non-topological holes are filled and smoothed, thus realizing the complete reconstruction of the three-dimensional surface of the blade.

[0093] The method of this invention achieves high accuracy in contour curve fitting. It uses feature points as segmentation points and edge points as base points for piecewise curve fitting, leveraging the high-order continuity of cubic Hermite interpolation to generate a smooth curve that closely matches the actual contour, thus avoiding the sharp edges or distortion problems of traditional fitting methods.

[0094] The method of this invention features high process efficiency and a high degree of automation. It achieves full-process algorithm-driven operation by adaptively adjusting the scanning height through a neural network, combined with point cloud preprocessing, optimized surface reconstruction, and automatic contour extraction. Attached Figure Description

[0095] Figure 1 This is a flowchart of the three-dimensional surface reconstruction method for complex components of aero-engine blades described in this invention;

[0096] Figure 2 This is a flowchart of the method for extracting the blade crown contour of an aero-engine as described in this invention;

[0097] Figure 3 This is a schematic diagram of the blade point cloud after pass-through filtering of the original blade point cloud data;

[0098] Figure 4 Yes Figure 3 A schematic diagram of the point cloud of a leaf after statistical filtering and voxel downsampling;

[0099] Figure 5 This is a schematic diagram of the reconstructed 3D surface of the blade.

[0100] Figure 6This is a schematic diagram of the local blade surface before reconstruction and optimization;

[0101] Figure 7 This is a schematic diagram of the local blade surface after reconstruction and optimization using the method of this invention;

[0102] Figure 8 This is a schematic diagram of the leaf canopy region point cloud obtained by clustering and segmenting the preprocessed leaf point cloud;

[0103] Figure 9 This is a schematic diagram of the leaf crown point cloud edge points after point cloud edge point extraction based on curvature;

[0104] Figure 10 This is a schematic diagram of the results of extracting feature points at the leaf crown edge;

[0105] Figure 11 This is a schematic diagram of the leaf crown point cloud contour fitting curve obtained by the method of the present invention. Detailed Implementation

[0106] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0107] Specific Implementation Method 1: Combination Figure 1 , Figures 3 to 5 As shown, this invention provides a method for reconstructing the three-dimensional curved surface of complex components of an aero-engine blade, comprising:

[0108] By setting calibration circles on the worktable and the blade respectively, the corresponding relationship of the number of pixels of the calibration circle under different known height differences is obtained; then, given the known number of pixels of the two calibration circles, the height of the blade relative to the worktable is determined, thereby determining the scanning height of the line laser measuring instrument, and scanning to obtain the original point cloud data of the blade;

[0109] Preprocess the original point cloud data of the blade to obtain the preprocessed point cloud of the blade;

[0110] An optimized frontier propulsion algorithm is used to reconstruct the three-dimensional surface of the preprocessed blade point cloud, and the reconstruction is optimized to obtain a complete and continuous three-dimensional surface of the blade.

[0111] In this embodiment, calibration circles are placed on the worktable and the blade, respectively. Images are captured by a camera, and the captured images are preprocessed to extract the number of pixels in the outline area of ​​the calibration circle in the image. A mapping model between the number of image pixels and the object distance is established based on a neural network to obtain the height relationship between the worktable and the blade, thereby adaptively determining the scanning height. Adaptive control is achieved through an algorithm. Based on the point cloud preprocessing algorithm, three-dimensional surface reconstruction and optimization are performed to obtain a complete and continuous three-dimensional surface of the blade, which allows for a good and intuitive understanding of the overall shape of the blade.

[0112] Furthermore, the method for determining the scanning height of the line laser measuring instrument is as follows:

[0113] To achieve adaptive acquisition of blade point clouds, the scanning height of the line laser measuring instrument needs to be determined. A camera at a specific mounting height is used to acquire images of the calibration circles on the worktable and blade at different height differences. After grayscale transformation and Gaussian filtering of the images, Canny edge detection is performed to obtain the pixel outlines of the calibration circles on the worktable and blade.

[0114] Define the gradient operator Gx0 in the horizontal direction and the gradient operator Gy0 in the vertical direction, and calculate the gradient magnitude threshold M(x0, y0) and the direction θ0(x0, y0):

[0115] ,

[0116] ,

[0117] In the formula The x-coordinate of the pixel is The ordinate of the pixel;

[0118] Calculate the gradient magnitude of adjacent pixels in the θ0 (x0, y0) direction of the outline of the calibration circle pixel point on the worktable or the outline of the calibration circle pixel point on the blade. If the gradient magnitude exceeds the gradient magnitude threshold M (x0, y0), the corresponding pixel point is retained as the edge point of the calibration circle; otherwise, it is deleted. The outline of the edge point of the calibration circle on the worktable and the outline of the edge point of the calibration circle on the blade are obtained.

[0119] The least squares method is used to extract the circular contours of the calibration circle edge points on the worktable and the blade to obtain the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade; the blade height is changed multiple times to obtain the number of pixels of the corresponding blade marking circle image.

[0120] A BP neural network optimized by random forest is used to establish a mapping model of pixel number-object distance. The mapping model is trained by using the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade under different known height differences. Based on the number of pixels of the two calibration circles obtained under the current height difference, the height of the blade relative to the worktable is determined by the mapping model of pixel number-object distance, and the scanning height of the line laser measuring instrument is determined so that the blade is within the scanning height range of the line laser measuring instrument.

[0121] Methods for preprocessing raw point cloud data of blades include:

[0122] The original point cloud data of the blade is transformed from the point spacing on the laser line to the point spacing distributed along the scanning direction, and then subjected to pass-through filtering and statistical filtering. Statistical filtering uses the K-nearest neighbor algorithm to select the n nearest neighbors of the center point, calculates the average distance, and calculates the standard deviation of the distance. A distance threshold is set based on the average distance and the standard deviation. The average distance between the center point and the n nearest neighbors is calculated, and center points with an average distance greater than the set distance threshold are considered outliers and removed. The filtered point cloud data is then obtained.

[0123] Voxel downsampling is performed on the filtered point cloud data, and the centroid of the voxel space is calculated as the representative point to obtain the preprocessed blade point cloud.

[0124] The point spacing on the laser line of a line laser measuring instrument is a fixed value, and the point spacing distributed along the scanning direction is given by the following formula:

[0125] The dot spacing v varies with the scanning speed, where... The scanning frequency of the line laser.

[0126] Pass-through filtering is used to remove outlier / invalid points. By retaining all points in the z-axis direction, the blade point cloud is extracted from the original point cloud data.

[0127] Statistical filtering is used to remove outliers. The specific process is as follows:

[0128] The K-nearest neighbor algorithm is used to select the n nearest neighbors of the center point, and the average distance μ is calculated using the following formula:

[0129] ,

[0130] Among them, D i The distance between any nearest neighbor and the center point;

[0131] Furthermore, the standard deviation σ of the distance is calculated using the following formula:

[0132] ,

[0133] Furthermore, a distance threshold D is set. max The mean μ and the standard deviation coefficient V σ The sum of the products of the standard deviation σ and the standard deviation σ is given by the formula:

[0134] ;

[0135] If the average distance between the center point and its n nearest neighbors is greater than the distance threshold, then this point is considered an outlier and should be removed.

[0136] By employing a voxel downsampling method, the centroid of the voxel space is calculated and retained as a representative point, thereby significantly reducing the number of points and reducing the computational load of subsequent processing.

[0137] Furthermore, methods for reconstructing 3D surfaces from preprocessed blade point clouds based on optimized front-end propulsion algorithms include:

[0138] Local neighborhood smoothing was performed on the preprocessed blade point cloud using second-order Jet polynomial surface fitting.

[0139] ,

[0140] In the formula Let x, y, z be the coordinates of the blade point cloud after preprocessing in three-dimensional space, and a0, a1, ..., a9 be the coefficients of the polynomial to be solved.

[0141] Determine the neighborhood points of the point cloud (x, y, z), and use the least squares method to make the neighborhood points reach the z-axis. Minimize the sum of squared distances and minimize the error function. :

[0142]

[0143] In the formula Let a0, a1, ..., a9 be the number of neighboring points.

[0144] Then p i (x) j y j , z j Substitute In the middle, we obtain a new point p. i ′, using the new point p i ′replace p i ; and then use the new point p i Using ′ as a new neighborhood point, iterative calculations are performed to obtain the fitted surface;

[0145] Normals are estimated based on the fitted surface, and the normal direction is determined by the minimum spanning tree (MST) algorithm based on the angle between the normals of adjacent point clouds.

[0146] Then, a leading-edge propulsion algorithm based on Delaunay triangulation theory is employed. According to the normal direction, a suitable starting point is selected from the preprocessed blade point cloud. An initial triangular facet is constructed using the two adjacent point clouds of the starting point as indices. The side length of the triangular facet is adaptively adjusted according to the blade point cloud density function ρ(x). Make the side length It is negatively correlated with the leaf point cloud density function ρ(x):

[0147]

[0148] Using the three sides of the triangular face as the initial front edge, the front edge is expanded under the set constraints to obtain two expanded point clouds. Then, the starting point and the two expanded point clouds form a new triangular face. The front edge expansion is repeated until the front edge is empty or cannot be expanded, and the reconstructed three-dimensional surface is obtained.

[0149] The method for reconstructing and optimizing the 3D surface is as follows:

[0150] For non-topological holes in the reconstructed 3D surface, determine the smallest cube containing all vertices of the non-topological hole, i.e., the bounding box, calculate the size of the smallest cube in each dimension, and determine whether the diameter of the non-topological hole exceeds a set diameter threshold based on the size of the smallest cube in each dimension.

[0151] For non-topological holes that do not exceed the set diameter threshold, Delaunay triangulation is performed using an algorithm to divide the non-topological holes into multiple well-shaped triangular faces, thereby achieving a rough filling of the non-topological holes and obtaining a filled three-dimensional surface.

[0152] After filling, the 3D surface is reduced by midpoint subdivision to reduce surface discontinuities and sharp features, and the mesh is smoothed by Laplacian smoothing.

[0153] ,

[0154] In the formula It refers to the vertex point cloud position in the filled 3D surface. Let λ be the position of the vertex point cloud after smoothing, and λ be the smoothing factor. for The set of adjacent vertex point clouds;

[0155] This results in a complete and continuous three-dimensional blade surface.

[0156] Specific Implementation Method Two: Combination Figure 2 , Figures 6 to 11 As shown, the present invention also provides a method for extracting the blade crown profile of an aero-engine, comprising:

[0157] By setting calibration circles on the worktable and the blade respectively, the corresponding relationship of the number of pixels of the calibration circle under different known height differences is obtained; then, given the known number of pixels of the two calibration circles, the height of the blade relative to the worktable is determined, thereby determining the scanning height of the line laser measuring instrument, and scanning to obtain the original point cloud data of the blade;

[0158] Preprocess the original point cloud data of the blade to obtain the preprocessed point cloud of the blade;

[0159] Clustering and segmentation are performed on the preprocessed leaf point cloud to obtain the leaf crown region point cloud. Curvature-based point cloud edge point extraction is performed on the leaf crown region point cloud, followed by edge feature point extraction. The edge feature points are used as segmentation points of the point cloud edge points. The cubic Hermite interpolation method is used to fit curves for each segment of the point cloud edge points to obtain the leaf crown point cloud contour fitting curve.

[0160] In this embodiment, calibration circles are placed on the worktable and the blade respectively. Images are captured by a camera, and the captured images are preprocessed to extract the number of pixels in the outline region of the calibration circle in the image. A mapping model between the number of image pixels and the object distance is established based on a neural network to obtain the height relationship between the worktable and the blade, thereby adaptively determining the scanning height. Adaptive control is achieved through an algorithm. A piecewise fitting method for the leaf crown profile curve based on cubic Hermite interpolation is adopted, with feature points as segmentation points and edge points as base points. The leaf crown profile curve is successfully obtained with high accuracy.

[0161] Furthermore, the method for determining the scanning height of the line laser measuring instrument is as follows:

[0162] To achieve adaptive acquisition of blade point clouds, the scanning height of the line laser measuring instrument needs to be determined. A camera at a specific mounting height is used to acquire images of the calibration circles on the worktable and blade at different height differences. After grayscale transformation and Gaussian filtering of the images, Canny edge detection is performed to obtain the pixel outlines of the calibration circles on the worktable and blade.

[0163] Define the gradient operator Gx0 in the horizontal direction and the gradient operator Gy0 in the vertical direction, and calculate the gradient magnitude threshold M(x0, y0) and the direction θ0(x0, y0):

[0164] ,

[0165] ,

[0166] The x-coordinate of the pixel is The ordinate of the pixel;

[0167] Calculate the gradient magnitude of adjacent pixels in the θ0 (x0, y0) direction of the outline of the calibration circle pixel point on the worktable or the outline of the calibration circle pixel point on the blade. If the gradient magnitude exceeds the gradient magnitude threshold M (x0, y0), the corresponding pixel point is retained as the edge point of the calibration circle; otherwise, it is deleted. The outline of the edge point of the calibration circle on the worktable and the outline of the edge point of the calibration circle on the blade are obtained.

[0168] The least squares method is used to extract the circular contours of the calibration circle edge points on the worktable and the blade to obtain the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade; the blade height is changed multiple times to obtain the number of pixels of the corresponding blade marking circle image.

[0169] A BP neural network optimized by random forest is used to establish a mapping model of pixel number-object distance. The mapping model is trained by using the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade under different known height differences. Based on the number of pixels of the two calibration circles obtained under the current height difference, the height of the blade relative to the worktable is determined by the mapping model of pixel number-object distance, and the scanning height of the line laser measuring instrument is determined so that the blade is within the scanning height range of the line laser measuring instrument.

[0170] Methods for preprocessing raw point cloud data of blades include:

[0171] The original point cloud data of the blade is transformed from the point spacing on the laser line to the point spacing distributed along the scanning direction, and then subjected to pass-through filtering and statistical filtering. Statistical filtering uses the K-nearest neighbor algorithm to select the n nearest neighbors of the center point, calculates the average distance, and calculates the standard deviation of the distance. A distance threshold is set based on the average distance and the standard deviation. The average distance between the center point and the n nearest neighbors is calculated, and center points with an average distance greater than the set distance threshold are considered outliers and removed. The filtered point cloud data is then obtained.

[0172] Voxel downsampling is performed on the filtered point cloud data, and the centroid of the voxel space is calculated as the representative point to obtain the preprocessed blade point cloud.

[0173] The point spacing on the laser line of a line laser measuring instrument is a fixed value, and the point spacing distributed along the scanning direction is given by the following formula:

[0174] The dot spacing v varies with the scanning speed, where... The scanning frequency of the line laser.

[0175] Pass-through filtering is used to remove outlier / invalid points. By retaining all points in the z-axis direction, the blade point cloud is extracted from the original point cloud data.

[0176] Statistical filtering is used to remove outliers. The specific process is as follows:

[0177] The K-nearest neighbor algorithm is used to select the n nearest neighbors of the center point, and the average distance μ is calculated using the following formula:

[0178] ,

[0179] Among them, D i The distance between any nearest neighbor and the center point;

[0180] Furthermore, the standard deviation σ of the distance is calculated using the following formula:

[0181] ,

[0182] Furthermore, a distance threshold D is set. max The mean μ and the standard deviation coefficient V σ The sum of the products of the standard deviation σ and the standard deviation σ is given by the formula:

[0183] ;

[0184] If the average distance between the center point and its n nearest neighbors is greater than the distance threshold, then this point is considered an outlier and should be removed.

[0185] By employing a voxel downsampling method, the centroid of the voxel space is calculated and retained as a representative point, thereby significantly reducing the number of points and reducing the computational load of subsequent processing.

[0186] Furthermore, the method for obtaining the point cloud of the leaf canopy region is as follows:

[0187] The DBSCAN density point cloud clustering algorithm is adopted. By traversing the preprocessed leaf point cloud, the number of point cloud points in the neighborhood radius eps of each point cloud is calculated. If the number of point cloud points is not less than the set minimum number of points MinPts, the current point cloud is marked as the current core point; otherwise, it is marked as a non-core point. Then, the core points and non-core points in the neighborhood of the current core point are recursively classified into the canopy region cluster, thereby obtaining the canopy region point cloud from the preprocessed leaf point cloud.

[0188] The method for extracting edge points in a point cloud is as follows:

[0189] A curvature-based point cloud edge point extraction method is adopted, and a neighborhood search is performed using a Kd tree to obtain a set of neighboring points in the leaf canopy region point cloud to form a neighboring point set. The covariance matrix C of the neighboring point set is then calculated.

[0190]

[0191] In the formula Let be the set of neighboring points, and q be the neighboring points in the set of neighboring points. Let be the centroid of the nearest neighbor set;

[0192] Find the eigenvalues ​​λ1, λ2, and λ3 of the covariance matrix C, such that λ1 ≥ λ2 ≥ λ3;

[0193] Calculate the curvature κ of the nearest point q:

[0194]

[0195] If the curvature κ exceeds the set curvature threshold, then the neighboring point q will be retained as the edge point of the point cloud.

[0196] The edge feature point extraction method is as follows:

[0197] The SUSAN algorithm is used to extract edge feature points from the point cloud edge points. Within a set radius r, a neighborhood search is performed using a Kd tree to obtain the neighborhood point set. The Euclidean distance d(P,Q) between the center point cloud edge point P in the neighborhood point set and other points Q in the neighborhood point set is calculated.

[0198] ,

[0199] In the formula Let x be the three-dimensional coordinates of point P at the edge of the center point cloud. Q y Q z Q Let Q be the three-dimensional coordinates of point Q;

[0200] Calculate the normal angle θ between point P and point Q at the edge of the center point cloud:

[0201]

[0202] In the formula Let P be the normal vector of the edge point P of the center point cloud. Let be the normal vector of point Q;

[0203] Calculate the intensity difference I(P, Q) between point P and point Q at the edge of the center point cloud:

[0204] ,

[0205] In the formula The local intensity of point P at the edge of the center point cloud. Let Q be the local intensity.

[0206] The Euclidean distance d(P,Q), normal vector angle θ, and intensity difference I(P,Q) are compared with the corresponding set thresholds. Points Q whose comparison results are all less than the corresponding set thresholds are retained. If the number of retained points Q is less than the set threshold, the corresponding points Q are taken as edge feature points.

[0207] Methods for obtaining the fitted curve of the leaf canopy point cloud profile include:

[0208] Using edge feature points as segmentation points and other point cloud edge points as base points, the leaf crown edge curve is segmented and fitted. Feature points are marked in the edge points, and each pair of adjacent segmentation points is used as the two ends of a fitted curve segment.

[0209] For each segment of the fitted curve, the cubic Hermite interpolation method was used for curve fitting:

[0210] Each fitted curve segment includes U point cloud edge points. Calculate the tangent vector between two adjacent point cloud edge points:

[0211] ,

[0212] ,

[0213] In the formula The number of base points in each fitted curve segment. For the first Tangent vectors at each fundamental point For the first +1 tangent vector at the base point For the first The coordinates of the base points;

[0214] Define weight function , , and :

[0215] ,

[0216] Where t∈[0,1] is the uniform sampling parameter, defined In the formula For the number of interpolation points, For interpolation points, v = 0, 1, ..., N-1;

[0217] Calculate the interpolation point function value H(t):

[0218] ,

[0219] Combining the interpolation point function value H(t) and the number of interpolation points The system uses smoothing and sorting auxiliary functions to obtain the fitting result of the current segment's fitted curve, and then obtains the leaf crown point cloud contour fitting curve. Specific Implementation

[0220] Combination Figures 1 to 11 Explanation:

[0221] First, collect the raw point cloud data of the blade; the specific processing flow is as follows: Figure 1 and Figure 2 As shown.

[0222] During the scanning process of a line laser measuring instrument, various factors such as the scanning environment and scanning strategy may lead to the presence of abnormal / invalid points and outliers in the scanning results. These points can adversely affect subsequent steps. Furthermore, excessively dense point cloud data can lead to reduced data processing efficiency and increased noise interference. Therefore, point cloud preprocessing is necessary to extract blade point clouds from the raw point cloud data that are easy to process in subsequent steps.

[0223] The leaf point cloud after pass-through filtering of the original leaf point cloud data is as follows: Figure 3 As shown, the leaf point cloud after statistical filtering and point cloud downsampling is as follows: Figure 4 As shown.

[0224] Point cloud data, as a discrete data type, cannot accurately describe the geometric structure of objects to some extent. However, 3D surface reconstruction can fit these discrete data points into a continuous surface model, resulting in a more compact data representation with a clear geometric structure and topological relationships. Furthermore, it is more intuitive and natural, making it easier to visually grasp the shape and structure of objects. Therefore, 3D surface reconstruction is performed on the point cloud of a blade. During the reconstruction process, due to factors such as uneven point cloud distribution and algorithm parameter limitations, holes appear on the surface. Filling these holes in the 3D surface optimizes the reconstruction. The reconstructed blade surface is shown below. Figure 5 As shown.

[0225] To achieve path planning for the welded repair area of ​​the blade crown, it is necessary to obtain the contour of the blade crown area. The crown area point cloud is extracted from the blade point cloud using a segmentation and clustering algorithm. Then, the edge points of the crown area point cloud are obtained. Since the obtained edge points are difficult to fit directly, an edge feature point extraction method is used to extract the feature points that best describe the crown shape. These edge feature points are then used as segmentation points, prioritizing edge points during the fitting process. Piecewise curve fitting based on cubic Hermite interpolation is used to fit the crown area contour, making the crown area contour curve as close as possible to the real data and improving fitting accuracy.

[0226] By setting the number of interpolation points in the interpolation function and combining auxiliary functions such as smoothing and sorting, a high-precision point cloud contour fitting curve for the leaf canopy region is obtained. The piecewise fitting result of the leaf canopy region contour curve is shown below. Figure 11 As shown.

[0227] In this embodiment, the number of pixels on the worktable and blade marker circles is obtained through image processing algorithms. This information is used to establish a pixel count-object distance neural network mapping model to determine the scanning height. Adaptive adjustment is implemented through code, with the height adjustment error within 5%. Raw point cloud data is acquired using a line laser measuring instrument. The raw point cloud data is processed by passing through filtering to extract the blade point cloud, followed by statistical filtering to remove outliers. Voxel downsampling is then used to simplify the blade point cloud. The final point cloud preprocessing effect is as follows: Figure 4 As shown.

[0228] After preprocessing the blade point cloud, to better achieve 3D surface reconstruction and thus understand the shape and characteristics of the blade, a 3D surface reconstruction and optimization method is adopted. Multiple second-order Jet polynomial surface fittings are used to smooth the local neighborhood of the point cloud, and the surface normals are calculated. A leading-edge propulsion algorithm based on Delaunay triangulation theory is used to establish the 3D surface of the blade, and a hole-filling algorithm is used to reconstruct and optimize the surface, obtaining a complete and continuous 3D surface of the blade. The final 3D surface effect of the blade is shown in the figure. Figure 7 As shown.

[0229] To obtain the blade region contour and lay the foundation for welding path planning, the DBSCAN density point cloud clustering algorithm was used to extract the crown region point cloud from the blade point cloud. A curvature-based point cloud edge point extraction method was employed to calculate the local curvature of the point cloud and compare it with a manually set curvature threshold to extract the edge points of the crown region point cloud. The SUSAN algorithm was then used to extract feature points from these edge points. Finally, using these feature points as segmentation points and edge points as base points, a segmented fitting of the crown region contour curve based on cubic Hermite interpolation was performed. The final crown region contour curve is shown below. Figure 11 As shown, the size error between the profile curve of the leaf crown region and the actual leaf is within 0.1 mm.

[0230] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A method for reconstructing the three-dimensional curved surface of complex components of an aero-engine blade, characterized in that, include: By setting calibration circles on the worktable and the blade respectively, the corresponding relationship of the number of pixels of the calibration circle under different known height differences is obtained; then, given the known number of pixels of the two calibration circles, the height of the blade relative to the worktable is determined, thereby determining the scanning height of the line laser measuring instrument, and scanning to obtain the original point cloud data of the blade; Preprocess the original point cloud data of the blade to obtain the preprocessed point cloud of the blade; An optimized frontier propulsion algorithm is used to reconstruct the three-dimensional surface of the preprocessed blade point cloud, and the reconstruction is optimized to obtain a complete and continuous three-dimensional surface of the blade.

2. The method for reconstructing the three-dimensional curved surface of complex components of aero-engine blades according to claim 1, characterized in that, The method for determining the scanning height of a line laser measuring instrument is as follows: Images of the calibration circles on the worktable and blades at different height differences are acquired using a camera; after grayscale transformation and Gaussian filtering of the images, Canny edge detection is performed to obtain the pixel contours of the calibration circles on the worktable and the blades. Define the gradient operator Gx0 in the horizontal direction and the gradient operator Gy0 in the vertical direction, and calculate the gradient magnitude threshold M(x0, y0) and direction θ0(x0, y0): ， ， In the formula The x-coordinate of the pixel is The ordinate of the pixel; Calculate the gradient magnitude of adjacent pixels in the θ0 (x0, y0) direction of the outline of the calibration circle pixel point on the worktable or the outline of the calibration circle pixel point on the blade. If the gradient magnitude exceeds the gradient magnitude threshold M (x0, y0), the corresponding pixel point is retained as the edge point of the calibration circle; otherwise, it is deleted. The outline of the edge point of the calibration circle on the worktable and the outline of the edge point of the calibration circle on the blade are obtained. The least squares method is used to extract the circular contours of the calibration circle edge points on the worktable and the calibration circle edge points on the blade to obtain the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade. A BP neural network optimized by random forest is used to establish a mapping model of pixel number-object distance. The mapping model is trained by using the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade under different known height differences. Based on the number of pixels of the two calibration circles obtained under the current height difference, the height of the blade relative to the worktable is determined by the mapping model of pixel number-object distance, and the scanning height of the line laser measuring instrument is determined so that the blade is within the scanning height range of the line laser measuring instrument.

3. The method for reconstructing the three-dimensional curved surface of complex components of aero-engine blades according to claim 2, characterized in that, Methods for preprocessing raw point cloud data of blades include: The original point cloud data of the blade is transformed from the point spacing on the laser line to the point spacing distributed along the scanning direction, and then subjected to pass-through filtering and statistical filtering. Statistical filtering uses the K-nearest neighbor algorithm to select the n nearest neighbors of the center point, calculates the average distance, and calculates the standard deviation of the distance. A distance threshold is set based on the average distance and the standard deviation. The average distance between the center point and the n nearest neighbors is calculated, and center points with an average distance greater than the set distance threshold are considered outliers and removed. The filtered point cloud data is then obtained. Voxel downsampling is performed on the filtered point cloud data, and the centroid of the voxel space is calculated as the representative point to obtain the preprocessed blade point cloud.

4. The method for reconstructing the three-dimensional curved surface of complex components of aero-engine blades according to claim 3, characterized in that, Methods for reconstructing 3D surfaces from preprocessed blade point clouds based on optimized front-end propulsion algorithms include: Local neighborhood smoothing was performed on the preprocessed blade point cloud using second-order Jet polynomial surface fitting. ， In the formula Let x, y, z be the coordinates of the blade point cloud after preprocessing in three-dimensional space, and a0, a1, ..., a9 be the coefficients of the polynomial to be solved. Determine the neighborhood points of the point cloud (x, y, z), and use the least squares method to make the neighborhood points reach the z-axis. Minimize the sum of squared distances and minimize the error function. : In the formula Let a0, a1, ..., a9 be the number of neighboring points. Then p i (x) j y j , z j Substitute In the middle, we obtain a new point p. i ′, using the new point p i ′replace p i ; and then use the new point p i Using ′ as a new neighborhood point, iterative calculations are performed to obtain the fitted surface; Based on the fitted surface, normals are estimated, and the normal direction is determined by the minimum spanning tree algorithm according to the angle between the normals of adjacent point clouds. Then, a leading-edge propulsion algorithm based on Delaunay triangulation theory is employed. Based on the normal direction, a starting point is selected from the preprocessed blade point cloud. An initial triangular facet is constructed using the two adjacent point clouds of the starting point as indices. The side length of the triangular facet is adaptively adjusted according to the blade point cloud density function ρ(x). Make the side length It is negatively correlated with the leaf point cloud density function ρ(x): Using the three sides of the triangular face as the initial front edge, the front edge is expanded under the set constraints to obtain two expanded point clouds. Then, the starting point and the two expanded point clouds form a new triangular face. The front edge expansion is repeated until the front edge is empty or cannot be expanded, and the reconstructed three-dimensional surface is obtained.

5. The method for reconstructing the three-dimensional curved surface of complex components of aero-engine blades according to claim 4, characterized in that, The method for reconstructing and optimizing the 3D surface is as follows: For non-topological holes in the reconstructed 3D surface, determine the smallest cube containing all vertices of the non-topological hole, calculate the size of the smallest cube in each dimension, and determine whether the diameter of the non-topological hole exceeds a set diameter threshold based on the size of the smallest cube in each dimension. For non-topological holes that do not exceed the set diameter threshold, Delaunay triangulation is performed to divide the non-topological holes into multiple triangular faces, thereby achieving a rough filling of the non-topological holes to obtain a filled three-dimensional surface. After filling, the 3D surface is reduced by midpoint subdivision to reduce surface discontinuities and sharp features, and the mesh is smoothed by Laplacian smoothing. ， In the formula It refers to the vertex point cloud positions in the filled 3D surface. Let λ represent the position of the vertex point cloud after smoothing, and λ be the smoothing factor. for The set of adjacent vertex point clouds; This results in a complete and continuous three-dimensional blade surface.

6. A method for extracting the blade crown contour of an aero-engine, characterized in that, include: By setting calibration circles on the worktable and the blade respectively, the corresponding relationship of the number of pixels of the calibration circle under different known height differences is obtained; then, given the known number of pixels of the two calibration circles, the height of the blade relative to the worktable is determined, thereby determining the scanning height of the line laser measuring instrument, and scanning to obtain the original point cloud data of the blade; Preprocess the original point cloud data of the blade to obtain the preprocessed point cloud of the blade; Clustering and segmentation are performed on the preprocessed leaf point cloud to obtain the point cloud of the leaf crown region; Curvature-based point cloud edge point extraction is performed on the point cloud of the leaf canopy region, followed by edge feature point extraction. The edge feature points are used as segmentation points of the point cloud edge points. The cubic Hermite interpolation method is used to fit curves for each segment of the point cloud edge points to obtain the leaf canopy point cloud contour fitting curve.

7. The method for extracting the blade crown profile of an aero-engine according to claim 6, characterized in that, The method for determining the scanning height of a line laser measuring instrument is as follows: Images of the calibration circles on the worktable and blades at different height differences are acquired using a camera; after grayscale transformation and Gaussian filtering of the images, Canny edge detection is performed to obtain the pixel contours of the calibration circles on the worktable and the blades. Define the gradient operator Gx0 in the horizontal direction and the gradient operator Gy0 in the vertical direction, and calculate the gradient magnitude threshold M(x0, y0) and direction θ0(x0, y0): ， ， The x-coordinate of the pixel is The ordinate of the pixel; Calculate the gradient magnitude of adjacent pixels in the θ0 (x0, y0) direction of the outline of the calibration circle pixel point on the worktable or the outline of the calibration circle pixel point on the blade. If the gradient magnitude exceeds the gradient magnitude threshold M (x0, y0), the corresponding pixel point is retained as the edge point of the calibration circle; otherwise, it is deleted. The outline of the edge point of the calibration circle on the worktable and the outline of the edge point of the calibration circle on the blade are obtained. The least squares method is used to extract the circular contours of the calibration circle edge points on the worktable and the calibration circle edge points on the blade to obtain the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade. A BP neural network optimized by random forest is used to establish a mapping model of pixel number-object distance. The mapping model is trained by using the number of pixels of the calibration circle on the worktable and the number of pixels of the calibration circle on the blade under different known height differences. Based on the number of pixels of the two calibration circles obtained under the current height difference, the height of the blade relative to the worktable is determined by the mapping model of pixel number-object distance, and the scanning height of the line laser measuring instrument is determined so that the blade is within the scanning height range of the line laser measuring instrument.

8. The method for extracting the blade crown profile of an aero-engine according to claim 7, characterized in that, Methods for preprocessing raw point cloud data of blades include: The original point cloud data of the blade is transformed from the point spacing on the laser line to the point spacing distributed along the scanning direction, and then subjected to pass-through filtering and statistical filtering. Statistical filtering uses the K-nearest neighbor algorithm to select the n nearest neighbors of the center point, calculates the average distance, and calculates the standard deviation of the distance. A distance threshold is set based on the average distance and the standard deviation. The average distance between the center point and the n nearest neighbors is calculated, and center points with an average distance greater than the set distance threshold are considered outliers and removed. The filtered point cloud data is then obtained. Voxel downsampling is performed on the filtered point cloud data, and the centroid of the voxel space is calculated as the representative point to obtain the preprocessed blade point cloud.

9. The method for extracting the blade crown profile of an aero-engine according to claim 8, characterized in that, The method for obtaining the point cloud of the leaf canopy region is as follows: The DBSCAN density point cloud clustering algorithm is adopted. By traversing the preprocessed leaf point cloud, the number of point cloud points in the neighborhood radius eps of each point cloud is calculated. If the number of point cloud points is not less than the set minimum number of points MinPts, the current point cloud is marked as the current core point; otherwise, it is marked as a non-core point. Then, the core points and non-core points in the neighborhood of the current core point are recursively classified into the canopy region cluster, thereby obtaining the canopy region point cloud from the preprocessed leaf point cloud. The method for extracting edge points in a point cloud is as follows: A curvature-based point cloud edge point extraction method is adopted, and a neighborhood search is performed using a Kd tree to obtain a set of neighboring points in the leaf canopy region point cloud to form a neighboring point set. The covariance matrix C of the neighboring point set is then calculated. In the formula Let be the set of neighboring points, and q be the neighboring points in the set of neighboring points. Let be the centroid of the nearest neighbor set; Find the eigenvalues ​​λ1, λ2, and λ3 of the covariance matrix C, such that λ1 ≥ λ2 ≥ λ3; Calculate the curvature κ of the nearest point q: If the curvature κ exceeds the set curvature threshold, then the neighboring point q will be taken as the edge point of the point cloud. The edge feature point extraction method is as follows: The SUSAN algorithm is used to extract edge feature points from the point cloud edge points. Within a set radius r, a neighborhood search is performed using a Kd tree to obtain the neighborhood point set. The Euclidean distance d(P,Q) between the center point cloud edge point P in the neighborhood point set and other points Q in the neighborhood point set is calculated. ， In the formula Let x be the three-dimensional coordinates of point P at the edge of the center point cloud. Q y Q z Q Let Q be the three-dimensional coordinates of point Q; Calculate the normal angle θ between point P and point Q at the edge of the center point cloud: In the formula Let P be the normal vector of the edge point P of the center point cloud. Let be the normal vector of point Q; Calculate the intensity difference I(P, Q) between point P and point Q at the edge of the center point cloud: ， In the formula The local intensity of point P at the edge of the center point cloud. Let Q be the local intensity. The Euclidean distance d(P,Q), normal vector angle θ, and intensity difference I(P,Q) are compared with the corresponding set thresholds. Points Q whose comparison results are all less than the corresponding set thresholds are retained. If the number of retained points Q is less than the set threshold, the corresponding points Q are taken as edge feature points.

10. The method for extracting the blade crown profile of an aero-engine according to claim 9, characterized in that, Methods for obtaining the fitted curve of the leaf canopy point cloud profile include: Using edge feature points as segmentation points and other point cloud edge points as base points, the leaf crown edge curve is segmented and fitted. Each pair of adjacent segmentation points serves as the two ends of a fitted curve segment. For each segment of the fitted curve, the cubic Hermite interpolation method was used for curve fitting: Each fitted curve segment includes U point cloud edge points. Calculate the tangent vector between two adjacent point cloud edge points: ， ， In the formula The number of base points in each fitted curve segment. For the first Tangent vectors at each fundamental point For the first +1 tangent vector at the base point For the first The coordinates of the base points; Define weight function , , and : ， Where t∈[0,1] is the uniform sampling parameter, defined In the formula For the number of interpolation points, For interpolation points, v = 0, 1, ..., N-1; Calculate the interpolation point function value H(t): ， Combining the interpolation point function value H(t) and the number of interpolation points The system uses smoothing and sorting auxiliary functions to obtain the fitting result of the current segment's fitted curve, and then obtains the leaf crown point cloud contour fitting curve.

Images