A fully automatic feature extraction and quantification method for three-dimensional point cloud of aircraft engine blades

Abstract

The invention discloses a method for fully automatic feature extraction and quantification of three-dimensional point clouds of aircraft engine blades. After batch point cloud collection of turbine blades using a binocular structured light system, the characteristic cross section of the turbine blade point cloud is firstly oriented, and a non-iterative hierarchical clustering algorithm P-Link is used to establish a connection leading to the clustering center between points, so as to quickly segment the blade point cloud into components and extract the blade body part point cloud. Then, the blade body is oriented using a point cloud bounding box based on principal component analysis and the blade height direction is calculated to obtain the characteristic cross-section normal. Then, the characteristic cross section and the cross-section width are located to slice the blade body, so as to automatically obtain multiple groups of contour slice point clouds of the blade body, avoid manual interception errors, and finally, the fitting based on B-spline is applied to extract the blade body curve features. The method can realize rapid batch feature extraction and quantification of blades.

Description

Technical Field

[0001] The invention belongs to the technical field of three-dimensional point cloud measurement, and more specifically, relates to a method for fully automatic feature extraction and quantification of three-dimensional point clouds of aircraft engine blades. Background Art

[0002] Blades are key components of aircraft engines, and their material properties, processing quality, and geometric accuracy have a significant impact on the aerodynamic performance of the engine. Turbine blades have complex profiles, strong twisting, large differences in curvature between the leading and trailing edges, and numerous design and characterization parameters. In addition, an engine may be designed with hundreds of blades of various shapes and sizes. The processing accuracy and manufacturing level of turbine blades affect engine performance, safety, and life. Therefore, batch and automatic feature extraction and measurement of the three-dimensional morphology of blades is an important part of aviation industry automation. In order to meet the needs of blade detection, detection methods also need to be further developed in the direction of high efficiency, high precision, and high intelligence.

[0003] The three-dimensional measurement technology based on projected structured light has gained wide attention. It has unique advantages by projecting active light stripes with structural information onto the object to be measured and modulating the height structure information of the surface of the object to measure the three-dimensional morphology of the object. In the measurement of spacecraft engine blades, the measurement based on three-dimensional point cloud has broad application potential.

[0004] However, in the batch 3D measurement process of aircraft engine turbine blades, on the one hand, the point cloud data collected by the high-precision binocular structured light point cloud acquisition system is very large, mostly in the millions, which brings higher requirements for batch measurement efficiency. On the other hand, the quality of the blade point cloud data collected each time is inconsistent, the attitude angle of the 3D point cloud data is different, the point cloud is disordered, and there are noise data. In addition, the blade itself is difficult to detect and the surface is complex, which poses a huge challenge to the automated measurement of the blade. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for fully automatic feature extraction and quantification of three-dimensional point clouds of aircraft engine blades, so as to automatically extract and quantify the point cloud features of turbine blades effectively, quickly and with high robustness.

[0006] To achieve the above-mentioned purpose, the method for fully automatic feature extraction and quantification of three-dimensional point cloud of aircraft engine blades of the present invention is characterized by comprising:

[0007] (1) Use the binocular structured light system to batch collect a series of three-dimensional point cloud data of turbine blades, recorded as Q = {PC1,…,PC t ,…,PC T}, where PCt represents the collected t-th turbine blade point cloud, and T represents the total number of point clouds;

[0008] (2) For each PC t ∈Q, perform leaf body point cloud segmentation

[0009] 2.1) Let the t-th turbine blade point cloud PC be t The total number of points is N, for the turbine blade point cloud PC t Every point p in i ∈PC t , find its K nearest neighbor points, expressed as the neighbor point set Ne i ={p i,1 ,…,p i,k ,…,p i,K},p i,k ∈PC t , where p i,k Represents point p i The neighbor point set Ne i The kth point in ;

[0010] 2.2) Using the neighbor point set Ne i All points in , use the least squares method to fit the plane, and obtain the normal vector n of the plane fitting i , a point set in the plane and the fitted residual r i , as point p i The normal vector n i , interior point set and the fitted residual r i ;

[0011] In this way, all PCs are t All points in the plane are fitted;

[0012] 2.3) PC of turbine blade point cloud t Every point p in i ∈PC t , traverse its interior point set L i Points in in, Represents the lth point in, the size of the internal point set is |L i |, judge Is the plane fitting residual of less than r i If a little satisfy Dominance Set And in the advantage set D i , find and point p i The normal vector n i The point p where the angle deviation is the smallestdm , establish a connection between them:

[0013] p i →{p dm |min <n i ,n dm >,p dm ∈D i}

[0014] If there is no point satisfy Then point p i Relative to the neighbor point set Ne i Any point Both Then further judge r i Whether it meets:

[0015] r i <μ t +σ t

[0016] If yes, then point p i is the local cluster center of the turbine blade point cloud, otherwise, the point is considered a noise point, where μ t It is a turbine blade point cloud PC t The mean of the overall plane fitting residuals for all points in , σ t is the standard deviation of the fitted residual, which is:

[0017]

[0018] 2.4) Using the depth-first search strategy, starting from each local cluster center, search for all directly or indirectly connected points to form a local cluster set C of the turbine blade point cloud = {C1, C2, ..., C G}, where C g (g=1,…,G) represents the gth cluster, G is the total number of clusters;

[0019] 2.5) Using the points in the cluster, in each cluster C g Fit the plane internally and record the cluster internal point set S1, S2, …, S G , where S g (g=1,…,G) represents the point set in the plane of the g-th cluster, and records the normal line of the cluster plane. in represents the plane normal of the g-th cluster;

[0020] 2.6) If there are two points p a 、p b Point sets that are mutually in the neighborhood: p a ∈Lb And p b ∈L a , they are in different cluster internal point sets: p a ∈S p And p b ∈S q When , we examine their corresponding cluster plane normal deviations. If It is considered that the two local clustering features are similar, and the two local clusters C p and C q Merge, C p ,C q ∈C, where θ is the threshold;

[0021] 2.7) Output the merged local cluster set C to complete the fast segmentation of the turbine blade point cloud. Then, find the cluster with the largest number of points in the local cluster set C. This cluster is used as the blade part point cloud of the turbine blade and is recorded as C. body ;

[0022] (3) Use the OBB bounding box to determine the blade part point cloud C body The main direction

[0023] 3.1) Calculate the point cloud C of the blade body body The three-dimensional center of gravity

[0024]

[0025] Among them, N body is the total number of points in the point cloud of the blade body;

[0026] 3.2) Based on three-dimensional center of gravity Construct the covariance matrix:

[0027]

[0028] 3.3) Calculate the eigenvalues ​​λ1,λ2,λ3 and eigenvectors β1,β2,β3 of the covariance matrix Q, and normalize these three vectors to obtain a set of bases. This set of bases is the three direction axes that constitute the directed bounding box. Among them, the eigenvector β1 corresponding to the maximum eigenvalue λ1 is the main direction of the blade point cloud, that is, the direction axis;

[0029] Projecting all points in the point cloud onto the direction axis, we can get the maximum value projected on the direction axis. Convert the point cloud back to the original coordinate axis, and get the vertex of the bounding box by projecting the maximum value onto the direction axis. Let C be the value of the bounding box. body The maximum and minimum values ​​in the three coordinate axes are x max ,y max ,z max and x min,y min ,z min , then the coordinates of the 8 vertices of the bounding box are:

[0030] f1=(x max ,y max ,z max ) T ,f2=(x min ,y max ,z max ) T

[0031] f3=(x max ,y min ,z max ) T ,f4=(x min ,y min ,z max ) T

[0032] f5=(x max ,y max ,z min ) T ,f6=(x min ,y max ,z min ) T

[0033] f7=(x max ,y min ,z min ) T ,f8=(x min ,y min ,z min ) T

[0034] (4) Automatic batch slicing extraction of blade contour features

[0035] 4.1) Calculate the three sides of the bounding box as:

[0036]

[0037] 4.2) Set the total number of blade contour lines to be extracted to H, the section width to ρ, and the main direction of the blade point cloud as the section normal. Determine a point on the low-level section and a point on the high-level section according to the following formula. Cut off the point cloud between the low-level section and the high-level section to obtain the hth (h=0,…,H-1)th contour line point cloud. A point p on the low-level surface of the hth (h=0,…,H-1)th contour line lower for:

[0038]

[0039] Correspondingly, a point p on the high plane of the hth (h=0,…,H-1)th contour line higher for:

[0040]

[0041] The turbine blade point cloud C body By intercepting the point cloud between these two planes with a width of 2ρ, we can get the hth (h=0,…,H-1)th turbine blade contour point cloud of the blade. in is the first point in the hth (h=0,…,H-1)th contour curve, N h is the total number of points of the h-th contour curve;

[0042] (5) Use the segmented cubic B-spline curve fitting method to perform quantitative analysis on the blade contour characteristics

[0043] (5.1), for the hth (h=0,…,H-1) turbine blade contour point cloud Will and As the first and last points of the cubic uniform B-spline curve, the internal contour points As the segment connection points of the cubic uniform B-spline curve, a total of N h +2 control points, both ends of each segmented B-spline curve will pass through the internal points in the turbine blade contour point cloud, and the first N h The control points are:

[0044]

[0045] In this way, we can get N h equations, and use the “catch-up method” to get the control point P i (i=1,…,N h );

[0046] 5.2) Let the curvature of the spline curve at both ends be 0 to obtain two additional control points:

[0047] P0=2P1-P2

[0048]

[0049] 5.3) Based on N h +2 control points, the expression of the i-th segment B-spline curve is obtained as:

[0050]

[0051] Among them, t is the location parameter, t∈[0,1], N j,3 (t) is the jth basis function:

[0052]

[0053] By fitting each turbine blade contour point cloud, the blade contour fitting feature parameters can be obtained, and batch feature extraction and quantitative analysis of blade point clouds can be completed.

[0054] The object of the invention of the present invention is achieved in this way:

[0055] The invention discloses a method for fully automatic feature extraction and quantification of three-dimensional point clouds of aero-engine blades. After batch point cloud collection of turbine blades is performed using a binocular structured light system, the characteristic cross-section of the turbine blade point cloud is firstly oriented, and a non-iterative hierarchical clustering algorithm P-Link is used to establish a connection leading to the cluster center between points, so as to quickly segment the blade point cloud into components and extract the blade body part point cloud. Then, a point cloud bounding box based on principal component analysis is used to orient the blade body and calculate the blade height direction, and obtain the characteristic cross-section normal. Then, the characteristic cross-section and the cross-section width are located to slice the blade body, so as to automatically obtain multiple groups of contour slice point clouds of the blade body, avoid manual interception errors, and finally, B-spline-based fitting is applied to extract blade body curve features. The method can realize rapid batch feature extraction and quantification of blades.

[0056] At the same time, the method for fully automatic feature extraction and quantification of three-dimensional point clouds of aircraft engine blades of the present invention also has the following beneficial effects:

[0057] 1. Based on the demand for batch feature extraction of turbine blades, the present invention automatically separates the blade point cloud from the overall blade point cloud, then identifies the main direction of the blade point cloud, and automatically intercepts the blade point cloud at a certain distance based on the main direction, thereby extracting the blade feature contour point cloud. This solves the problem of long time consumption and low accuracy caused by the traditional blade feature extraction relying on manual selection of feature areas and performing feature fitting, and improves the efficiency of batch blade feature extraction.

[0058] 2. In the process of automatic segmentation of the blade part point cloud, the present invention combines the point cloud segmentation method based on P-Link clustering. By establishing links between points and their local neighborhood points, the cluster center is gradually found to complete the segmentation. The non-iterative hierarchical clustering algorithm improves the efficiency of the blade part segmentation and reduces the time consumption of point cloud segmentation compared to the traditional Euclidean clustering segmentation. At the same time, there is no need to downsample the original point cloud, which ensures the accuracy of blade measurement.

[0059] 3. The present invention uses a bounding box algorithm based on principal component analysis to identify the main direction of the blade point cloud, solving the problem of different posture angles, different point cloud densities, noise, etc. caused by inconsistent experimental environments and conditions during the acquisition of blade point clouds. The contour features of the blade point cloud are extracted at a certain distance through the bounding box plane, realizing batch slicing and quantization of the blade point cloud. The measurement effect and robustness of the batch algorithm for noisy point clouds of different postures and densities are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flowchart of a specific implementation method of the method for fully automatic feature extraction and quantification of three-dimensional point clouds of aircraft engine blades of the present invention;

[0061] Figure 2 It is a specific flow chart of performing point cloud segmentation on a turbine blade point cloud according to the present invention;

[0062] Figure 3 It is a specific flow chart of establishing the main direction of the blade body based on the bounding box technology of the present invention;

[0063] Figure 4 It is a specific flow chart of the present invention for slicing the blade body point cloud to extract the feature contour line point cloud after obtaining the cross section based on the bounding box technology;

[0064] Figure 5 This is a specific flow chart of the present invention performing B-spline fitting on the blade contour point cloud;

[0065] Figure 6 This is an example of a turbine blade point cloud collected using a binocular structured light system in this example;

[0066] Figure 7 This is the result of fast point cloud segmentation of turbine blade point cloud based on P-Link clustering algorithm in this example;

[0067] Figure 8 This is the bounding box result diagram of the principal component analysis and calculation of the point cloud of the blade body;

[0068] Fig. 9 This is the result of automatically slicing the blade point cloud to extract contour features based on the main direction of the calculated bounding box;

[0069] Fig.10 It is the quantitative analysis result of B-spline fitting based on the contour feature point cloud; DETAILED DESCRIPTION

[0070] The specific implementation of the present invention is described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0071] Figure 1 It is a flow chart of a specific implementation method of the method for fully automatic feature extraction and quantification of three-dimensional point clouds of aircraft engine blades of the present invention.

[0072] In this embodiment, if Figure 1 As shown, the method for fully automatic feature extraction and quantification of three-dimensional point clouds of aircraft engine blades of the present invention comprises the following steps:

[0073] Step S1: Use the binocular structured light system to batch collect a series of three-dimensional point cloud data of turbine blades, and record them as Q = {PC1, ..., PC t ,…,PC T}, where PC t represents the tth turbine blade point cloud collected, and T represents the total number of point clouds.

[0074] Step S2: For each PC t ∈Q, perform leaf body point cloud segmentation

[0075] In this embodiment, if Figure 2 As shown, including:

[0076] Step S2.1: Let the t-th turbine blade point cloud PC t The total number of points is N, for the turbine blade point cloud PC t Every point p in i ∈PC t , find its K nearest neighbor points, expressed as the neighbor point set Ne i ={p i,1 ,…,p i,k ,…,p i,K},p i,k ∈PC t , where p i,k Represents point p i The neighbor point set Ne i The kth point in .

[0077] Step S2.2: Using the neighbor point set Ne i All points in , use the least squares method to fit the plane, and obtain the normal vector n of the plane fitting i , a point set in the plane and the fitted residual r i , as point p i The normal vector n i, interior point set and the fitted residual r i .

[0078] In this way, all PCs are t All points in the plane are fitted.

[0079] Step S2.3: Establish connections and find local cluster centers

[0080] Step S2.3.1: let i=1;

[0081] Step S2.3.2: PC of turbine blade point cloud t Every point p in i ∈PC t , traverse its interior point set L i Points in in, Represents the lth point in, the size of the internal point set is |L i |, judge Is the plane fitting residual of less than r i ,if Dominance Set And in the advantage set D i , find and point p i The normal vector n i The point p where the angle deviation is the smallest dm , establish a connection between them:

[0082] p i →{p dm |min <n i ,n dm >,p dm ∈D i}

[0083] If there is no point satisfy Then point p i Relative to the neighbor point set Ne i Any point Both Then further judge r i Whether it meets:

[0084] r i <μ t +σ t

[0085] If yes, then point p i is the local cluster center of the turbine blade point cloud, otherwise, the point is considered a noise point, where μ t It is a turbine blade point cloud PC tThe mean of the overall plane fitting residuals for all points in , σ t is the standard deviation of the fitted residual, which is:

[0086]

[0087] Step S2.3.3: If i≤N, then i=i+1, go to step S2.3.1, otherwise go to step S2.4.

[0088] Step S2.4: Using the depth-first search strategy, start from each local cluster center to search for all directly or indirectly connected points to form a local cluster set C = {C1, C2, ..., C G}, where C g (g=1,…,G) represents the gth cluster, and G is the total number of clusters;.

[0089] Step S2.5: Use the points in the cluster to find the g Fit the plane internally and record the cluster internal point set S1, S2, …, S G , where S g (g=1,…,G) represents the point set in the plane of the g-th cluster, and records the normal line of the cluster plane. in represents the plane normal of the g-th cluster.

[0090] Step S2.6: If there are two points p a 、p b Point sets that are mutually in the neighborhood: p a ∈L b And p b ∈L a , they are in different cluster internal point sets: p a ∈S p And p b ∈S q When , we examine their corresponding cluster plane normal deviations. If It is considered that the two local clustering features are similar, and the two local clusters C p and C q Merge, C p ,C q ∈C, where θ is the threshold.

[0091] Step S2.7: Output the merged local cluster set C to complete the fast segmentation of the turbine blade point cloud. Then, find the cluster with the largest number of points in the local cluster set C. This cluster is used as the blade part point cloud of the turbine blade and is recorded as C. body .

[0092] Step S3: Use the OBB bounding box to determine the blade part point cloud Cbody The main direction

[0093] In this embodiment, if Figure 3 As shown, including:

[0094] Step S3.1: Calculate the blade part point cloud C body The three-dimensional center of gravity

[0095]

[0096] Among them, N body It is the total number of points in the point cloud of the blade body.

[0097] Step S3.2: Based on 3D centroid Construct the covariance matrix:

[0098]

[0099] S3.3: Calculate the eigenvalues ​​λ1,λ2,λ3 and eigenvectors β1,β2,β3 of the covariance matrix Q, and normalize these three vectors to obtain a set of bases. This set of bases is the three direction axes that constitute the directed bounding box. Among them, the eigenvector β1 corresponding to the maximum eigenvalue λ1 is the main direction of the blade body point cloud, that is, the direction axis;

[0100] Projecting all points in the point cloud onto the direction axis, we can get the maximum value projected on the direction axis. Convert the point cloud back to the original coordinate axis, and get the vertex of the bounding box by projecting the maximum value onto the direction axis. Let C be the value of the bounding box. body The maximum and minimum values ​​in the three coordinate axes are x max ,y max ,z max and x min ,y min ,z min , then the coordinates of the 8 vertices of the bounding box are:

[0101]

[0102] Step S4: Automatic batch slicing extraction of blade contour features

[0103] In this embodiment, if Figure 4 As shown, including:

[0104] Step S4.1: Calculate the three sides of the bounding box as:

[0105]

[0106] Step S4.2: Set the total number of blade contour lines to be extracted to H, the section width to ρ, and the main direction of the blade point cloud as the section normal. Determine a point on the low-level section and a point on the high-level section according to the following formula. Cut off the point cloud between the low-level section and the high-level section to obtain the hth (h=0,…,H-1)th contour line point cloud. A point p on the low-level surface of the hth (h=0,…,H-1)th contour line lower for:

[0107]

[0108] Correspondingly, a point p on the high plane of the hth (h=0,…,H-1)th contour line higher for:

[0109]

[0110] The turbine blade point cloud C body By intercepting the point cloud between these two planes with a width of 2ρ, we can get the hth (h=0,…,H-1)th turbine blade contour point cloud of the blade. in is the first point in the hth (h=0,…,H-1)th contour curve, N h is the total number of points on the hth contour curve.

[0111] Step S5: Use the segmented cubic B-spline curve fitting method to perform quantitative analysis on the blade profile characteristics

[0112] In this embodiment, if Figure 5 As shown, including:

[0113] Step S5.1: For the hth (h=0,…,H-1)th turbine blade contour point cloud Will and As the first and last points of the cubic uniform B-spline curve, the internal contour points As the segment connection points of the cubic uniform B-spline curve, a total of N h +2 control points, both ends of each segmented B-spline curve will pass through the internal points in the turbine blade contour point cloud, and the first N h The control points are:

[0114]

[0115] In this way, we can get N h equations, and use the “catch-up method” to get the control point P i (i=1,…,N h ).

[0116] Step S5.2: Set the curvature of the spline curve at both ends to 0 to obtain two additional control points:

[0117] P0=2P1-P2

[0118]

[0119] Step S5.3: Based on N h +2 control points, the expression of the i-th segment B-spline curve is obtained as:

[0120]

[0121] Among them, t is the location parameter, t∈[0,1], N j,3 (t) is the jth basis function:

[0122]

[0123] By fitting each turbine blade contour point cloud, the blade contour fitting feature parameters can be obtained, and batch feature extraction and quantitative analysis of blade point clouds can be completed.

[0124] Examples

[0125] In this example, the high temperature point cloud is a 3D scan point cloud PC of an aircraft engine turbine blade at high temperature. t The point cloud is as follows Figure 6 As shown, the total number of turbine blade point clouds is 1,028,208.

[0126] Then, P-Link is used to establish links between the points in the turbine blades to the cluster center, so as to gradually find the cluster center. Then, the depth-first measurement search is used to form clusters, and clusters with similar features are merged to finally obtain the segmentation result of the turbine blade point cloud. Figure 7 As shown, the point cloud C belonging to the leaf body is found body The total number of points is 642619.

[0127] Then the leaf point cloud C body Extract it, find the main direction of the blade point cloud through principal component analysis, and calculate the bounding box of the blade point cloud, such as Figure 8 As shown. After PCA, the main components of the blade point cloud are λ1=352.632; λ2=128.765; λ1=-0.995126, and the three eigenvectors are β1=[-0.995126,0.0912993,-0.0372705] T , β2=[-0.0943671,-0.991357,0.0911427] T, β3=[-0.0286271,0.0942155,0.99514] T .

[0128] Taking the bottom normal direction of the blade point cloud bounding box as the section direction, the blade point cloud is intercepted at intervals along the orthogonal direction of the central axis to obtain a series of contour line point clouds of the blade point cloud. Set the number of section features to 5 and the section width to 0.02mm. The contour lines of the blade point cloud are intercepted as follows: Fig. 9 shown.

[0129] Finally, the B-spline fitting is performed on the feature point cloud of the blade contour line extracted, and the results are as follows: Fig.10 As shown, the blade point cloud contour line 1 is fitted and the chord length is measured using a B-spline curve. The chord length of the contour line 1 is 35.2957 mm.

[0130] Although the above describes an illustrative specific embodiment of the present invention to facilitate the understanding of the present invention by those skilled in the art, it should be clear that the present invention is not limited to the scope of the specific embodiment. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

Claims

1. A method for fully automatic feature extraction and quantification of three-dimensional point clouds of aircraft engine blades, characterized in that: include: (1) Use the binocular structured light system to batch collect a series of three-dimensional point cloud data of turbine blades, recorded as Q = {PC1,…,PC t ,…,PC T }, where PC t represents the collected t-th turbine blade point cloud, and T represents the total number of point clouds; (2) For each PC t ∈Q, perform leaf body point cloud segmentation 2.1) Let the t-th turbine blade point cloud PC t The total number of points is N, for the turbine blade point cloud PC t Every point p in i ∈PC t , find its K nearest neighbor points, expressed as the neighbor point set Ne i ={p i,1 ,…,p i,k ,…,p i,K },p i,k ∈PC t ,in, p i,k Represents point p i The neighbor point set Ne i The kth point in ; 2.2) Using the neighbor point set Ne i All points in , use the least squares method to fit the plane, and obtain the normal vector n of the plane fitting i , the point set L in the plane i , and the fitted residual r i , as point p i The normal vector n i , interior point set L i , and the fitted residual r i ; Based on the neighbor point set, all PC t All points in the plane are fitted; 2.3) PC of turbine blade point cloud t Every point p in i ∈PC t , traverse its interior point set L i Points in in, Represents the lth point in, the size of the internal point set is |L i |, judge Is the plane fitting residual of less than r i If a little satisfy Then put it into the advantage set D i , And in the advantage set D i , find and point p i The normal vector n i The point p where the angle deviation is the smallest dm , establish a connection between them: on i →{p dm |my <n i ,n dm >,p dm ∈D i } If there is no point satisfy Then point p i Relative to the neighbor point set Ne i Any point Both Then further judge r i Whether it meets: r i ≤μ t +s t If yes, then point p i is the local cluster center of the turbine blade point cloud, otherwise, the point is considered a noise point, where μ t It is a turbine blade point cloud PC t The mean of the overall plane fitting residuals for all points in t is the standard deviation of the fitted residual, which is: 2.4) Using the depth-first search strategy, starting from each local cluster center, search for all directly or indirectly connected points to form a local cluster set C = {C1, C2, ..., C g ,…,C G }, where C g (g=1,…,G) represents the gth cluster, G is the total number of clusters; 2.5) Using the points in the cluster, in each cluster C g Fit the plane internally and record the cluster internal point set S1, S2, …, S g ,…,S G , where S g (g=1,…,G) represents the point set in the plane of the g-th cluster, recording the normal line of the cluster plane in represents the plane normal of the g-th cluster; 2.6) If there are two points p a 、p b Point sets that are mutually in the neighborhood: p a ∈L b And p b ∈L a , they are in different cluster internal point sets: p a ∈S p And p b ∈S q When , we examine their corresponding cluster plane normal deviations. If It is considered that the two local clustering features are similar, and the two local clusters C p and C q Merge, C p ,C q ∈C, where θ is the threshold; 2.7) Output the merged local cluster set C to complete the fast segmentation of the turbine blade point cloud. Then, find the cluster with the largest number of points in the local cluster set C. This cluster is used as the blade part point cloud of the turbine blade and is recorded as C. body ; (3) Use the OBB bounding box to determine the blade part point cloud C body The main direction 3.1) Calculate the point cloud C of the blade body body The three-dimensional center of gravity in, N body is the total number of points in the point cloud of the blade body; 3.2) Based on three-dimensional center of gravity Construct the covariance matrix: 3.3) Calculate the eigenvalues ​​λ1,λ2,λ3 and eigenvectors β1,β2,β3 of the covariance matrix Q, and normalize these three vectors to obtain a set of bases. This set of bases is the three direction axes that constitute the directed bounding box. Among them, the eigenvector β1 corresponding to the maximum eigenvalue λ1 is the main direction of the blade point cloud, that is, the direction axis; Projecting all points in the point cloud onto the direction axis, we can get the maximum value projected on the direction axis. Convert the point cloud back to the original coordinate axis, and get the vertex of the bounding box by projecting the maximum value onto the direction axis. Let C be the value of the bounding box. body The maximum and minimum values ​​in the three coordinate axes are x max ,y max ,z max and x min ,y min ,z min , then the coordinates of the 8 vertices of the bounding box are: f1=(x max ,y max ,z max ) T ,f2=(x min ,y max ,z max ) T f3=(x max ,y min ,z max ) T ,f4=(x min ,y min ,z max ) T f5=(x max ,y max ,z min ) T ,f6=(x min ,y max ,z min ) T f7=(x max ,y min ,z min ) T ,f8=(x min ,y min ,z min ) T (4) Automatic batch slicing extraction of blade contour features 4.1) Calculate the three sides of the bounding box as: 4.2) Set the total number of blade contour lines to be extracted to H, the section width to ρ, and the main direction of the blade point cloud as the section normal. Determine a point on the low-level section and a point on the high-level section according to the following formula. Cut off the point cloud between the low-level section and the high-level section to obtain the hth (h=0,…,H-1)th contour line point cloud. The hth (h=0,…,H-1)th contour line point p on the low-level surface is lower for: Correspondingly, a point p on the high plane of the hth (h=0,…,H-1)th contour line higher for: The turbine blade point cloud C body By intercepting the point cloud between these two planes with a width of 2ρ, we can get the hth (h=0,…,H-1)th turbine blade contour point cloud of the blade. in is the first point in the hth (h=0,…,H-1)th contour curve, N h is the total number of points of the h-th contour curve; (5) Use the segmented cubic B-spline curve fitting method to perform quantitative analysis on the blade contour characteristics 5.1) For the hth (h=0,…,H-1)th turbine blade contour point cloud Will and As the first and last points of the cubic uniform B-spline curve, the internal contour points As the segment connection points of the cubic uniform B-spline curve, a total of N h +2 control points, both ends of each segmented B-spline curve will pass through the internal points in the turbine blade contour point cloud, and the first N h The control points are: In this way, we can get N h equations, and use the "catch-up method" to get the control point P i (i=1,…,N h ); 5.2) Let the curvature of the spline curve at both ends be 0 to obtain two additional control points: P0=2P1-P2 5.3) Based on N h +2 control points, the expression of the i-th segment B-spline curve is obtained as: in, t is the location parameter, t∈[0,1], N j,3 (t) is the jth basis function: By fitting each turbine blade contour point cloud, the blade contour fitting feature parameters can be obtained, and batch feature extraction and quantitative analysis of blade point clouds can be completed.

Images