Aero-engine blade damage deformation model adaptive reconstruction method and system

Abstract

The application belongs to the technical field of engine repair, and particularly discloses a model adaptive reconstruction method and system for aero-engine blade damage deformation, which comprises the following steps: under the deformation conditions of aero-engine blade torsion, bending and the like, the damaged position of the blade is quickly identified; the cutting depth of the damaged position is set, and the part of the damaged blade that is not conducive to the subsequent repair process is removed according to the set depth; the convex hull of the point cloud after the depth setting is obtained, and the removal is performed according to the shape of the convex hull to ensure that the cut position after the removal is smooth; then the boundary of the point cloud after the cutting is extracted, a quadratic surface is fitted according to the boundary, the nearest point corresponding method is used to obtain the filling point cloud of the cross section of the cavity of the blade after the cutting; the cross section filling point cloud and the blade point cloud after the cutting are combined, and the blade machining model is reconstructed. The application is used in the production process of the cutting and machining of the damaged position of the blade, the manual identification and manual polishing process is saved, and the subsequent fusion welding repair process is prepared.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of engine repair, and more particularly relates to a method and system for self-adaptive reconstruction of a model of a damaged and deformed turbine blade. BACKGROUND

[0002] An aero-engine is the heart of an airplane, and its blades account for one-third of the engine. Due to centrifugal load generated by its own operation during service, the geometric shape of the blades is deformed in bending or straightening. Under the action of steady and unsteady airflow, mechanical structure is self-excited and excited by external inhaled materials, so that the blades vibrate and finally fatigue and fail. Automatic repair of damaged blades of an aero-engine blade disc, extension of the service life of the blades, and reduction of the replacement frequency of the entire engine blade disc are of great significance to improve the service performance and economy of the aero-engine.

[0003] At present, the repair methods for aero-engine blades mainly include laser cladding / welding and self-adaptive polishing. The first step of realizing a digital repair system process for a damaged blade of an entire aero-engine is to detect the damaged part and cut off the damaged part of the blade to remove the adverse part generated by the damaged blade, which is not conducive to the subsequent repair process. Because the damage of the blade is various, and the bending and torsional deformation also varies, the detection and cutting of the damaged position of the blade are completed by manual work. Manual work in this link will result in low efficiency of the process, and will increase the time cost and economic cost of the blade repair, which is not conducive to the automation of the entire production line. SUMMARY

[0004] In view of the above defects or improvement needs of the prior art, the present application provides a method and system for self-adaptive reconstruction of a model of a damaged and deformed turbine blade, which aims to realize rapid and accurate identification and repair of the damaged position of the blade, and to save the process of manual identification and manual polishing.

[0005] To achieve the above-mentioned purpose, according to one aspect of the present application, a method for self-adaptive reconstruction of a model of a damaged and deformed turbine blade is provided, which comprises the following steps:

[0006] S1, detecting and identifying the damaged position of the blade; specifically comprising:

[0007] S11, acquiring point cloud data A of the damaged blade, and converting the point cloud data A to the coordinate system of ideal processing model point cloud data B;

[0008] S12, calculating the distance between each point a in the point cloud data A and the corresponding nearest point b in the point cloud data B; i further calculating the geometric error between the points a and b. i i i further calculating the geometric error between the points a and b.​​

[0009] S13. Based on geometric errors, the region growth method is used to determine the location of leaf damage;

[0010] S2. Set the cutting depth at the location of blade damage;

[0011] S3. Based on the cutting depth, set the convex hull of the cutting part to obtain a machining model with a flat cutting boundary;

[0012] S4. Based on the machining model, repair the hollow cross-section after cutting;

[0013] S5. Target model for blade cutting after 3D reconstruction and repair.

[0014] As a further preferred embodiment, in step S12, point a i With b i The calculation process for the geometric error between them is as follows:

[0015] Calculate point a i With b i Euclidean distance D i ;

[0016] Determine point a in point cloud data A i The domain point NB(a) i In point cloud data B, determine point b. i The domain point NB(b) i );

[0017] Using the least squares method, fit a i Its neighboring point NB(a i ) plane, to obtain a i The normal vector NC(a) i ); Fit b i Its neighborhood point NB(b) i The plane of ) yields b i The normal vector NC(b) i ); Calculate the normal vector NC(a) i ) and NC(b i The included angle α between ) i ;

[0018] Then calculate point a i With b i Geometric error Ge(a) between i ,b i ) = D i '+kα i '; where D i '、α i 'Let D be the Euclidean distance.' i Angle αi normalized value of the distance between the point a i and the field point NB(a i ), and k is a proportional coefficient.

[0019] As a further preferred, the point a i in the point cloud data A is determined i , and the field point NB(a i ) is specifically:

[0020] In the point cloud data A, a number of neighborhood points closest to the point a i are calculated, and the vector angle between a i and each neighborhood point is calculated, and the neighborhood point corresponding to the vector angle greater than 90° is deleted, and the remaining points are retained.

[0021] The neighborhood point closest to the point a i is re-searched, and the above steps are repeated until the neighborhood point with the vector angle meeting the requirement reaches a predetermined number.

[0022] As a further preferred, in step S13, the region growing method is used to determine the position of the damaged section of the blade, and specifically:

[0023] (1) The geometric error Ge(a i , b i ) corresponding to each a i is calculated, and a i is arranged in descending order according to the size of Ge(a i , b i ), and a data set P is obtained;

[0024] (2) The first point P(1) with a geometric error greater than a threshold value δ1 is found in the data set P in order, which is taken as a seed point and stored in a data set R, and the point P(1) is deleted from the data set P;

[0025] (3) The neighborhood point NB(P(1)) of the seed point P(1) is calculated, and NB(P(1)) is arranged in descending order according to the geometric error, and a data set S is obtained;

[0026] (4) The geometric error of the first point S(1) in the data set S is compared with the threshold value δ1, if greater, S(1) is stored in the data set R, and S(1) is taken as a new seed point, and step (3) is repeated; otherwise, go back to step (2) until there is no point in the data set P with a geometric error greater than the threshold value δ1;

[0027] (5) The points in the data set R are identified as the damaged position of the blade, and the cross-sectional point cloud coordinate data of the damaged position of the blade is extracted and stored in the point cloud data PS.

[0028] As a further preferred, step S2 specifically includes:

[0029] The point cloud data PS is indexed by distance, and the index distance is set as a threshold value δ2, which corresponds to the removal of the cutting depth at the position of the damaged section of the blade; according to the distance index, all points in the point cloud data A with a distance within δ2 from the point cloud data PS are found and stored in the point set C.

[0030] As a further preferred, step S3 specifically comprises:

[0031] The convex hull of the point set C is calculated, the points in the point cloud data A located within the convex hull of the point set C are removed, and the point cloud data A1 with a smooth cutting boundary is obtained; the difference between A1 and A point cloud is calculated, and the actual cutting point cloud data C1, i.e. the machining model, is obtained.

[0032] As a further preferred, step S4 specifically comprises:

[0033] For each point in the point cloud data C1, the distance of the point to all points in the point cloud data A1 is calculated, and if the minimum value of the distance is less than a distance threshold value δ3, the point is extracted; according to all the points extracted from the point cloud data C1, the cutting boundary point cloud Es after the removal of the damaged position of the blade is obtained;

[0034] The cutting boundary point cloud Es is subjected to quadratic surface fitting, and each point c 1i in the point cloud data C1 is projected onto the quadratic surface to find the nearest point e i in the quadratic surface by means of equation-constrained nonlinear optimization, and all e i are stored in the data set E;

[0035] The data set E and the point cloud data A1 are merged and spliced by point cloud, and filtered to obtain the complete cutting blade point cloud data A2, and the repair of the cutting hole section is completed.

[0036] As a further preferred, step S5 specifically comprises:

[0037] The point cloud data A2 is filtered and subjected to average voxel sparse downsampling; according to the sampled point cloud data, a greedy projection triangulation algorithm is used: the point cloud normal is calculated, and the point cloud is projected onto a certain two-dimensional plane; then the projected point cloud is triangulated in the plane to obtain the topological connection relationship of each point, and the topological connection of the original three-dimensional point cloud is determined according to the topological connection relationship in the plane, and finally the triangular mesh reconstruction model, i.e. the blade cutting machining target model, is obtained.

[0038] As a further preferred, in the process of planar triangulation, a spatial region growing algorithm based on Delaunay triangulation is used to determine the topological connection relationship of each point.

[0039] According to another aspect of the present application, a turbine blade damage deformation model adaptive reconstruction system is provided, comprising a processor configured to perform the above-mentioned turbine blade damage deformation model adaptive reconstruction method.

[0040] Overall, compared with the prior art, the above technical solutions conceived by the present application mainly have the following technical advantages:

[0041] 1. The present application can quickly identify the different damage positions of the blade damage under the conditions of torsion, bending and other deformations of the aero-engine blade. Then, by setting the cutting depth and shape of the part to be cut, the damaged area is set to obtain a smooth blade cutting model, and the cross-sectional shape of the cutting is optimized. Further, the target model of the three-dimensional reconstruction of the blade cutting processing is repaired and cut. The present application is used in the production process of blade damage cutting processing, which saves the process of manual identification and manual polishing, and prepares for the subsequent melting and welding repair process.

[0042] 2. The present application detects and identifies the damaged cross section of the blade with any damage and different deformation conditions by using the region growing algorithm, which overcomes the problem that it is difficult to find a comparative model for automatic detection and identification due to the complex deformation and damage of the blade.

[0043] 3. The present application obtains the point cloud of the missing part of the model cross section after cutting by using the quadratic surface fitting projection method, which is more suitable for this processing scene. Compared with the traditional hole filling algorithm, this algorithm has the characteristics of faster speed and improves the time efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The flowchart of the turbine blade damage deformation model adaptive reconstruction method of the embodiment of the present application;

[0045] Figure 2 The flowchart of the region growing method of the embodiment of the present application;

[0046] Figure 3 The CAD ideal point cloud diagram of a certain type of blade registered to the same coordinate system of the embodiment of the present application;

[0047] Figure 4 The point cloud diagram of a certain type of damaged blade registered to the same coordinate system of the embodiment of the present application;

[0048] Figure 5 Figures (a) and (b) are the damaged cross section diagrams of the blade extracted in step S1 of the embodiment of the present application;

[0049] Figure 6 The point cloud data diagram of the given cutting depth of step S2 of the embodiment of the present application;

[0050] Figure 7 A point cloud convex hull graph planned according to a given cutting depth for step S3 of the embodiment of the present application;

[0051] Figure 8 A blade point cloud data graph after cutting of the service damaged blade for step S3 of the embodiment of the present application;

[0052] Figure 9 A filled cross section point cloud data graph obtained according to the fitting quadratic surface projection method for step S4 of the embodiment of the present application;

[0053] Figure 10 A final point cloud data graph obtained after blade cutting of the point cloud after merging for step S4 of the embodiment of the present application;

[0054] Figure 11 A geometric patch model graph after three-dimensional reconstruction of the final blade point cloud of the embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the objectives, technical solutions and advantages of the present application clearer and more comprehensible, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0056] The method for self-adaptive reconstruction of the model of the damaged and deformed aero-engine blade provided by the embodiment of the present application, as shown in Figure 1 includes the following steps:

[0057] S1, detecting and identifying the damaged position and damaged cross section of the blade;

[0058] S2, setting the cutting depth at the identified damaged position of the blade;

[0059] S3, setting the convex hull of the cutting part to obtain a machining model with smooth cutting boundary;

[0060] S4, repairing the hollow cross section after cutting;

[0061] S5, reverse modeling, three-dimensional reconstruction of the target model of the blade cutting machining after cutting and repairing.

[0062] Further, step S1 includes:

[0063] S11, obtaining initial point cloud data;

[0064] A developer was uniformly sprayed onto the surface of the blade parts, and calibration points were set on the bottom turntable. By rotating the chassis holding the blade, a blue light 3D scanner was used to record the 3D point cloud data of each surface of the blade. The point clouds were merged using a point cloud stitching algorithm to obtain the final point cloud data of the service-damaged blade model. The data was then filtered and denoised to obtain relatively clean experimental point cloud data.

[0065] The point cloud data of the original ideal CAD blade model is digitized, and the point cloud data of the damaged blade and the ideal blade converted from CAD are uniformly sampled to obtain the damaged point cloud data A and the ideal CAD processing model point cloud data B with the same average point spacing.

[0066] Through rigid ICP iterative registration, the damaged point cloud data A is transformed into the coordinate system of the ideal machining model point cloud data B, where m and n are the number of points in the damaged point cloud data A and the CAD ideal model point cloud data B, respectively.

[0067] In this embodiment, the final point cloud data A and B each contain approximately 150,000 points. The point cloud diagrams corresponding to the ideal CAD point cloud data B and the damaged point cloud data A are shown below. Figure 3 , Figure 4 As shown.

[0068] S12, Calculate geometric error;

[0069] Calculate each point a in the damaged point cloud data A i The nearest point b in point cloud data B i The Euclidean distance D between the two is obtained. i .

[0070] D i =min(dist(a i ,b j (j=1,2,...,n)

[0071] Where n is the number of point clouds in the ideal CAD model point cloud data B.

[0072] Using the neighborhood point search method, point a is first calculated in point cloud data A. i The 10 nearest neighbors; to ensure that the neighbors are in a i The uniformity of the surrounding distribution, calculate a. i The angle between vectors and neighboring points is determined, ensuring that the angle between adjacent vectors is no greater than 90°. If the angle is greater than 90°, the nearest points corresponding to the adjacent vectors with angles greater than 90° are deleted, and the search for the nearest point is restarted until the requirement is met, thus obtaining point a. i All neighboring points NB(a i ).

[0073] According to the uniformity neighborhood point algorithm, the least square method is used to fit a i The plane of a i and its neighborhood points NB(a i ), and the normal vector NC(a i ) of a i is obtained. The normal vector NC(b i ) of the corresponding nearest point b i is obtained in the same way; the included angle α i between NC(a i ) and NC(b i ) is calculated.

[0074] Specifically, taking a i as an example, the normal vector of the plane obtained by the least square plane fitting is:

[0075] z=k0x+k1y+k2

[0076] For all neighborhood points (x i , y i , z i ), i=0, 1,..., n-1, there is a penalty function:

[0077]

[0078] Minimizing the penalty function, we have:

[0079]

[0080]

[0081] where n is the number of neighborhood points, and the parameters k j , j=0, 1, 2 obtained by the final solution can obtain the fitted plane equation, and the plane normal vector NC(a i )=(-k0, -k1, 1) can be obtained.

[0082] The Euclidean distance D i between each point a i in the normalized damaged blade point cloud A and the corresponding point b i in the CAD ideal model point cloud B, and the included angle α i between their normal vectors, are obtained to obtain the normalized distance D i ' and α i ', and the weighted sum is calculated to obtain the geometric error Ge(a i , b i ) between a i and b i :

[0083]

[0084]

[0085] Ge(a i ,b i ) = D i '+kα i '

[0086] Among them, D i '、α i 'For point a i With b i Normalized values ​​of the distance and angle between them, D i and α i Point a i With point b i The values ​​of the Euclidean distance and the angle between the normal vectors are given, where m is the number of points in the damaged point cloud data A, and k is a set scaling factor used to adjust the weight of the influence of distance and vector angle on the geometric difference, which affects the accuracy of subsequent identification and extraction of the damaged blade cross-section.

[0087] S13. Use the regional growth method to determine the location of leaf damage;

[0088] like Figure 2 As shown, it specifically includes:

[0089] S131. Find each a i Geometric error Ge(a) i ,b i ), and according to Ge(a i ,b i ) for a i Sort the dataset in descending order to obtain a new dataset P;

[0090] S132. In the dataset P, find the first point P(1) with a geometric error greater than the threshold δ1 as the seed point and store P(1) in the dataset R.

[0091] S133. Next, find the neighboring points NB(P(1)) of the seed point P(1), and sort NB(P(1)) in descending order according to the geometric error Ge to obtain a new dataset S;

[0092] S134. Compare the geometric error of the first point S(1) in dataset S with the threshold δ1. If it is greater, store S(1) in dataset R and use S(1) as the new seed point, repeating step S133; otherwise, return to step S132 until there are no points in dataset P with a geometric error greater than the threshold δ1. It should be noted that when calculating the neighborhood points of each new seed point, the previous seed points need to be removed from the total set A to avoid the infinite loop situation where each point is a neighbor of another point.

[0093] S135, the points in the data set R are extracted and stored in the data set PS, which is the position of the damaged section identified by the algorithm; the section point cloud coordinate data of the damaged part is extracted and stored in the data set PS.

[0094] In this embodiment, the threshold value δ1 is 0.4, and the size of the threshold value needs to be set differently according to different types of blades, which is an empirical value. The damaged point cloud section data graph is shown in FIG. 4. Figure 5

[0095] Further, step S2 includes:

[0096] The point cloud data PS of the extracted damaged section is indexed by distance, and the index distance is set as the threshold value δ2, which corresponds to the cutting depth of the damaged part of the blade. According to the distance index, all points in the damaged point cloud A with a distance within δ2 from the section point cloud PS are found and stored in the point set C.

[0097] The threshold value δ2 is set according to the actual depth of the damaged micro-cracks of the blade, and in this embodiment, the depth is selected as 6. The final cutting depth point cloud is shown in FIG. 5. Figure 6

[0098] Further, step S3 includes:

[0099] The convex hull of the obtained point set C is calculated, and the points in the damaged point cloud A located within the convex hull of C are removed, thereby obtaining the point cloud model A1 with a smooth cutting boundary, and the difference between A1 and A is calculated, that is, the different point clouds of the coordinates are extracted and stored in the data set C1, thereby obtaining the actual cutting point cloud data set C1. The convex hull is shown in FIG. 6, and the point cloud data graph after removing the section of the damaged blade is shown in FIG. 7. Figure 7 Figure 8

[0100] Further, step S4 includes:

[0101] A distance threshold value δ3 is set, and if the distance between the cutting part point cloud data set C1 and the blade point cloud data set A1 after cutting is less than the distance threshold value δ3, the part is extracted. By setting the value of δ3, the cutting boundary Es after cutting the damaged part of the blade is obtained.

[0102] After removing the C1 data set from the point cloud data A, the data set A1 obtained at the damaged section is a hollow without point cloud, and if there is a hollow, the cutting model for machining cannot be reconstructed by three-dimensional reconstruction. The extracted cutting boundary point cloud Es is fitted by a quadratic surface, and the removed point cloud data set C1 is projected onto the quadratic surface, and each point c 1i in C1 finds the nearest point e in the quadratic surface by an equality constrained nonlinear optimization problem​​​​i and all e i are stored in dataset E.

[0103] The obtained dataset E is merged with dataset A1 by point cloud splicing, and then filtered, and finally the complete cutting blade point cloud data A2 is obtained.

[0104] Specifically, the least square surface fitting projection related calculation method is as follows:

[0105] x 2 +k1y 2 +k2z 2 +k3x+k4y+k5z+k6=0

[0106] For all boundary points (x i ,y i ,z i ), i = 0, 1,..., n-1, there is a penalty function:

[0107]

[0108] Minimize the penalty function, and obtain:

[0109]

[0110]

[0111] k = [k1 k2 k3 k4 k5 k6] T

[0112]

[0113] Pk = -Q

[0114] Where N is the number of boundary points, and the finally obtained parameter k j , j = 1, 2, 3, 4, 5, 6, the fitted quadratic surface function is obtained, and the nonlinear equality constraint optimization problem is solved:

[0115] f(x,y,z) = (x ci -x) 2 +(y ci -y) 2 +(z ci -z) 2

[0116]

[0117] Where (x ci ,y ci ,z ci), i=0, 1, …, n-1 is the point cloud data coordinate of the removed part, each point can be obtained by optimizing the distance function f(x, y, z) through the nonlinear equation constraint, and the projection point in the quadric surface is obtained, and then the filling point cloud of the interface cavity is obtained.

[0118] In this embodiment, the distance threshold δ3 for extracting the boundary is set to 1.5, which can be adjusted according to different point cloud density, and the finally obtained filling cross-section point cloud data graph is as shown in Figure 9

[0119] Further, the step S5 comprises:

[0120] The obtained cutting blade point cloud A2 is filtered and average voxel sparse downsampling is performed to ensure that the point cloud data used for three-dimensional reconstruction is relatively uniform.

[0121] The point cloud grid is reconstructed using the greedy triangulation algorithm, the point cloud normal is calculated, the point cloud is projected onto a certain two-dimensional plane, then the projected point cloud is triangulated in the plane, and then the topological connection relationship of each point is obtained, the spatial region growing algorithm based on Delaunay triangulation is used, and finally the topological connection of the original three-dimensional point cloud is determined according to the topological connection relationship in the plane to obtain the triangular mesh reconstruction model. The finally obtained complete point cloud data of the cutting blade is as shown in Figure 10 , and the three-dimensional model after three-dimensional reconstruction is as shown in Figure 11 .

[0122] In summary, the three-dimensional scanning technology is used to detect and identify the damaged cross-section of the blade in any damage and different deformation conditions by using the proposed region growing algorithm, which overcomes the problem that it is difficult to find a comparative model for automatic detection and identification due to the complex deformation and damage of the blade; the proposed depth setting and shape setting algorithm is used to set the removed part of the damaged area to obtain a smooth blade cutting model of the removed cross-section, which optimizes the shape of the removed cross-section; the proposed quadric surface fitting projection method is used to obtain the point cloud of the missing part of the model cross-section after cutting, which is more suitable for this processing scene, and compared with the traditional hole filling algorithm, the algorithm has the characteristics of faster speed and improves the time efficiency.

[0123] Those skilled in the art will readily understand that the above description is only a preferred embodiment of the present application, and is not intended to limit the present application, and any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.​

Claims

1. A method for adaptive reconstruction of a post-damage deformation model of a turbine blade, characterized in that, Comprise the following steps: S1, detecting the damaged position of the blade; specifically comprising: S11, obtaining the point cloud data A of the damaged blade, and converting the point cloud data A to the coordinate system of the ideal machining model point cloud data B through rigid ICP iterative registration; S12. Calculate the points in point cloud data A. The nearest point in point cloud data B Then calculate the point and Geometric errors between them; including: computing points with euclidean distance ; determining a point of the domain in the point cloud data A determining a point of the domain in the point cloud data B ; Using the least squares method, the fit is... Its neighboring points The plane, to obtain normal vector ; Fitting Its domain point The plane, to obtain normal vector ; Calculate the normal vector and The angle between ; further calculate the geometric error between the points and ; wherein, , are normalized values of the Euclidean distance , the included angle , respectively, is a proportional coefficient;​ S13, based on the geometric error, the damaged position of the blade is determined by using region growing method; the cross-section point cloud coordinate data of the damaged position of the blade is extracted and stored in the point cloud data PS; S2, setting the cutting depth at the damaged position of the blade; comprising: The point cloud data PS is indexed by distance, and the index distance is set as a threshold value The threshold value The cutting depth at the position of the damaged section of the blade is removed accordingly; according to the distance index, all points in the point cloud data A with a distance in The point set C is obtained by storing all points in the point cloud data A with a distance in S3, setting the convex hull of the cutting part based on the cutting depth, and then obtaining the machining model with smooth cutting boundary; comprising: The convex hull of the point set C is calculated, and points in the point cloud data A located within the convex hull of the point set C are removed to obtain point cloud data with a smooth cutting boundary ; and The actual cutting point cloud data is obtained by subtracting the point cloud A , that is, the machining model S4, based on the machining model, repairing the hollow cross-section after cutting; comprising: Point cloud data For each point in the graph, calculate the point cloud data from that point. The distance to all points, if the minimum value of that distance is less than the distance threshold. If so, then extract that point; based on the point cloud data All points extracted are used to obtain the cut boundary point cloud after the leaf damage location is removed. ; Fitting a quadric surface to the cutting boundary point cloud Es, projecting the point cloud data onto the quadric surface, finding the nearest point in the quadric surface for each of the points by equality constrained non-linear optimization, and storing all of in a dataset E; The dataset E is combined with the point cloud data The point cloud is combined and spliced, and filtered to obtain complete cutting blade point cloud data The repair of the hollow section after cutting is completed; S5, three-dimensional reconstruction of the repaired blade cutting machining target model.

2. The method of claim 1, wherein the method further comprises: determining a point in the point cloud data A of the domain point , in particular: Calculate the relationship between points in point cloud data A. Calculate the nearest neighboring points of each. The vector angle between the vector and each neighboring point is used to delete the neighboring points when the vector angle is greater than 90°, and the remaining points are retained. Re-searching with the point The nearest neighbor point, and repeat the above steps until the vector angle meets the requirements of the neighborhood point to achieve a predetermined number.

3. The method of claim 1, wherein the method further comprises: In step S13, the region growing method is used to determine the damaged cross-section position of the blade, specifically: (1) Find each Corresponding geometric error and according to Size pair Sort the dataset P in descending order; (2) Find the first point in the dataset P with a geometric error greater than the threshold value in order and store it in the dataset R as a seed point while removing the point from the dataset P. ​ (3) Find the seed point neighboring points and will Sort the dataset S in descending order of geometric error; (4) Take the first point in dataset S Geometric error and threshold Compare; if it is greater, then... Store it in dataset R, and use As a new seed point, repeat step (3); otherwise, return to step (2) until no geometric error in dataset P is greater than the threshold. point; (5) The points in the data set R are the damaged positions of the blade, and the cross-section point cloud coordinate data of the damaged position of the blade is extracted and stored in the point cloud data PS.

4. The method of claim 1, wherein the method further comprises: Step S5, specifically comprising: To point cloud data Filtering, and average voxel sparse downsampling; according to the sampled point cloud data, using greedy projection triangulation algorithm: calculating point cloud normal, and the point cloud is projected onto a two-dimensional plane; then the point cloud obtained by projection is triangulated in the plane, and the topological connection relationship of each point is obtained, and the topological connection of the original three-dimensional point cloud is determined according to the topological connection relationship in the plane, and finally the triangular mesh reconstruction model, i.e. the blade cutting processing target model is obtained.

5. The method of claim 4, wherein the method further comprises: In the process of planar triangulation, the spatial region growing algorithm based on Delaunay triangulation is used to determine the topological connection relationship of each point.

6. An aero-engine blade damage deformation post-model adaptive reconstruction system, characterized in that, Comprise a processor, the processor is used for executing the aero-engine blade damage deformation model adaptive reconstruction method as claimed in any one of claims 1-5.

Images