A method for processing and axis estimation of a gas film hole point cloud

Abstract

A kind of processing and axis estimation method for gas film hole point cloud, including: first based on the outlier segmentation method of octree adjacent point distance statistics and the point cloud data downsampling based on voxelization grid, complete gas film hole point cloud preprocessing;Then adopt the iterative closest point algorithm with orifice center point position and fitting axis as benchmark, pass through plane fitting and divide out hole wall point cloud and find orifice center coordinate and hole axis, to this as benchmark, point cloud is carried out rough registration, as the basis of iterative closest point algorithm fine registration;Finally, the evaluation of iterative estimation of axis, utilize least square estimation multiple plane ellipse center and fit straight line as the basis of pose transformation, repeatedly iterate until optimal axis is acquired;The present application improves point cloud processing efficiency, improves the accuracy of point cloud registration.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of turbine blade measurement, and particularly relates to a processing and axis estimation method for a film hole point cloud. BACKGROUND

[0002] With the continuous development of current turbine engine technology, in order to improve the engine thrust, higher and higher turbine inlet temperature has to be adopted, which requires high-pressure turbine blades to have higher temperature resistance under complex stress conditions to meet the requirements of high-temperature work. Therefore, specific number and distribution of film holes are often made on modern aero-engine turbine blades, which can form a layer of film on the surface of the blade for cooling, so as to improve the working temperature that the blade can withstand.

[0003] The main parameters of the film hole include hole diameter, hole center coordinate and axis angle, wherein the hole diameter determines the flow of the cooling gas of the film hole, the hole center coordinate determines the action position of the cooling gas flow of the film hole, and the axis angle determines the flow direction of the cooling gas flow, and the three parameters together determine the cooling effect of the film hole on the blade. Therefore, it is necessary to measure the three main parameters of the film hole to ensure that they meet the requirements. However, the film hole has different morphological characteristics, such as cylindrical hole, forward expansion hole, square hole and various special-shaped holes, and the shapes, sizes, axis angles and positions are different, and the surface of the blade is a free curved surface, which brings many inconveniences to the measurement of the film hole. At present, the blade production unit often measures the film hole of the blade by using a contact measurement method, which has many problems such as difficulty in measuring special-shaped holes, difficulty in measuring hole axis angle and low measurement efficiency.

[0004] At present, there has been a scheme of using a machine vision method to perform microscopic measurement on a microstructure. With the gradual development of computer technology, the computer processing efficiency is greatly improved, and the use of a non-contact optical method to measure the film hole (CN201711414874.1, name: a five-axis image measurement device for measuring film hole shape and position parameters) has become the focus of the current film hole measurement field, which has the advantages of fast speed, high precision, strong adaptability and no damage to the hole. Due to the hole structure of the film hole, many challenges are brought to the application of the non-contact optical measurement method. The original point cloud data of the film hole obtained by this method has many problems such as many noise points and difficulty in splicing of multiple measurement directions, and how to effectively process and analyze the film hole point cloud data is a big difficulty. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a processing and axis estimation method for a film hole point cloud, which improves the point cloud processing efficiency and improves the accuracy of point cloud registration.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for processing and estimating the axis of point clouds for air-supported films includes: firstly, using an outlier segmentation method based on octree nearest neighbor distance statistics and point cloud data downsampling based on voxelized raster to preprocess the air-supported film point clouds; then, employing an iterative nearest point algorithm based on the position of the orifice center point and the fitted axis to segment the orifice wall point cloud through plane fitting and find the orifice center coordinates and orifice axis, using these as a reference for coarse registration of the point cloud, serving as the basis for fine registration of the iterative nearest point algorithm; finally, evaluating the iterative estimation of the axis by using least squares to estimate the centers of multiple planar ellipses and fitting straight lines as the basis for pose transformation, iterating repeatedly until the optimal axis is obtained.

[0008] The preprocessing of the air film pore point cloud is as follows: First, outliers in the original air film pore point cloud data are removed. Based on the point cloud data storage scheme of octree data structure, the average distance between each point in the point cloud and its surrounding points is calculated, and outliers are selected according to the standard deviation ratio threshold. Then, the entire point cloud is divided into voxelized grids of unit volume, and the centroid of all points in each voxel is calculated as the position of the point after downsampling.

[0009] The specific steps of the preprocessing of the air film pore point cloud are as follows:

[0010] S101. Fill all the points in the air film pore point cloud into an octree data structure.

[0011] S1011. Find the smallest circumscribed cube of the air film pore point cloud in three-dimensional space, and use it as the cube represented by the root node of the octree.

[0012] S1012. Select all points belonging to the cube and record them under that node;

[0013] S1013. If there are more than 1 child nodes under this node, create 8 child nodes under this node; if there are 1 or less child nodes, stop recursion.

[0014] S1014. Recursively process the 8 child nodes using steps S1012 and S1013. If a child node is found to contain the same number of points as its parent node, stop the recursion.

[0015] S102. Traverse the point cloud to find the n nearest neighbors using the octree, and calculate the n nearest neighbors np. δj Point P δ The mean of the distance dis δ :

[0016]

[0017] Where: dis—the average distance between a point and its neighboring points; n—the number of neighboring points; p δ — Traversing the selected δ-th point; np δj ——p δ The j-th nearest neighbor; δ—the number of the point;

[0018] S103, Calculate all points p in the point cloud δ dis δ mean and standard deviation

[0019]

[0020]

[0021] in: —The mean of the average distances between all points and their neighbors; N —The number of points in the point cloud; dis δ —The average distance between the δth point and its nearest neighbors; —The standard deviation of the average distance between all points and their neighbors;

[0022] S104. Select all that satisfy:

[0023]

[0024] Where: ρ—standard deviation ratio threshold; S dis —The variance of the average distance between all points and their nearest neighbors;

[0025] That is, find dis δ Compare Larger ρ times The point pδ is obtained and removed, and finally the point cloud of the air film pores after removing the discrete points is obtained;

[0026] S105. Calculate the minimum bounding cube of the air film pore point cloud after removing discrete points, and divide the cube into M cube voxel ranges according to the voxel size q.

[0027] S106. Traverse the range of M cube voxels. If there is a point in the range of the Lth voxel, proceed to step S107; otherwise, traverse the next point.

[0028] S107. Calculate the centroid C of the midpoint of the range of the Lth voxel. L :

[0029]

[0030] Where: C L—The centroid of the midpoint of the Lth idiotype range; m —The number of midpoints in the Lth idiotype range;

[0031] S108, the centroid C of the midpoint of the Lth voxel range. L The result of the voxel downsampling is recorded in the downsampled point cloud. After traversing all points in the point cloud, the preprocessed air film pore point cloud is obtained.

[0032] The method for point cloud registration is as follows:

[0033] S201. Approximate the surface of the outer blade as a plane, and segment the preprocessed point cloud results of the air film vent in N directions into two parts: the outer blade surface and the remaining point cloud, using a plane model segmentation based on random sampling consistency; specifically:

[0034] S2011. Randomly select a subset S' of size η from the point cloud G, and use the minimum variance method to calculate the subset model parameters on G', where η is the minimum number of samples required to initialize the model;

[0035] S2012. Obtain the remainder R = G - G' from G. Calculate the deviation of all samples in the remainder under the model. If the deviation is less than the threshold, the sample point is considered to be an in-model sample point; otherwise, it is an out-of-model sample point. Record the number of in-model sample points.

[0036] S2013. Calculate the iteration termination evaluation factor based on the expected error rate set in the model, the total number of samples, the number of in-model sample points of the current best model, and the current iteration number. If the factor reaches a certain threshold, the iteration ends; otherwise, repeat steps S2011 and S2012.

[0037] S2014. After the iteration, the final model parameter estimates are obtained, and the in-model sample points of the final model on the sample set S are used as the plane segmentation results.

[0038] S202. Based on the estimated normal vector of the planar portion of the blade surface outside the hole, obtain the plane equation;

[0039] S203. Perform Euclidean distance-based clustering on the remaining point cloud, and extract the cluster with the most points as the point cloud of the hole's inner wall; specifically:

[0040] S2031. Let the input point cloud dataset be P, set an empty clustering list C, and a point cloud queue Q that needs to be checked.

[0041] S2032. For each point pi in P, set p i Add to the current queue Q;

[0042] S2033, for each p in Q i Set radius r < d th Search for the nearest neighbor point cloud set P of p in the range ik ;

[0043] S2034, for each nearest neighbor point cloud set P ik , check if the points in it have been processed, if not, add them to Q;

[0044] S2035, when all points in Q have been processed, add Q to the cluster list C, and clear Q, repeat steps S2033, S2034 until all points in the data set P have been processed;

[0045] S2036, select the cluster with the most points in the cluster list C as the output point cloud P';

[0046] S204, estimate the axis direction of the hole wall point cloud using cylindrical fitting, and calculate the intersection of the axis and the plane;

[0047] S205, repeat steps S201, S202, S203, calculate the plane ψ equation and cylindrical part point cloud axis Γ direction of the gas film hole point cloud preprocessing result of all directions, and based on this, take the gas film hole point cloud preprocessing result of the first direction as the reference, correct all point clouds to the same direction, complete the coarse registration of the point cloud;

[0048] S206, use the ICP algorithm to register each direction gas film hole point cloud preprocessing result to the first direction gas film hole point cloud preprocessing result in turn, get the fine registration result of the gas film hole point cloud, that is, the complete gas film hole point cloud.

[0049] The method for estimating the axis of the gas film hole is:

[0050] S301, use multiple sets of parallel planes to obtain multiple two-dimensional sections of the complete gas film hole point cloud, and evaluate the integrity of the section point cloud, which is the basis for the subsequent estimation process; Specifically:

[0051] S3011, use the axis obtained in step S204 as the initial axis to solve the pose transformation matrix T i Transform the initial axis to be perpendicular to the XOY plane and pass through the origin, and use the above pose transformation matrix T i Perform the same pose transformation on the complete gas film hole point cloud as a whole;

[0052] S3012, the transformation result selects multiple planes parallel to the XOY plane as slice planes according to the slice interval h;

[0053] S3013, select a slice threshold hd d

[0054] S3014, numbering the obtained point clouds from bottom to top along the positive direction of Z axis, checking the integrity of the point cloud from the small number, that is, whether it covers at least three quadrants from the Z axis direction; if the integrity is poor, remove the point cloud until the point cloud integrity meets the requirements;

[0055] S302, estimating the center position of each point cloud to estimate the axis position; specifically:

[0056] S3021, taking the average value of the Z coordinates of the points in each point cloud as the Z coordinate of the estimated center point V;

[0057] S3022, the center estimation only uses the X and Y coordinates of each point, which is simplified to the fitting of a two-dimensional graph, and since the shape is approximately elliptical, the general equation of an ellipse is used as a template, the least squares method is used to construct the equation, and the Lagrange multiplier method is used to solve it. The center coordinates of the ellipse are obtained as the X and Y coordinates of the center point V;

[0058] S3023, for each center point V, the least squares fitting is used to obtain a straight line equation, denoted as straight line l;

[0059] S303, according to the obtained straight line l, it is used as the initial axis in S301, and the steps of S301 and S302 are repeated for iteration until the final transformation angle is less than the threshold value, then the iteration is stopped, and the equation of the axis is calculated according to the accumulated transformation matrix T i .

[0060] Compared with the prior art, the beneficial effects of the present application are: the present application combines the topological structure characteristics of the gas film hole, extracts the key structural information of the gas film hole, realizes the point cloud fusion and axis estimation of each measurement direction of the gas film hole. Specifically:

[0061] (1) The present application effectively removes outliers from the point cloud of each measurement direction of the gas film hole, and performs downsampling. Under the premise of retaining the geometric characteristics of the gas film hole and being able to complete the subsequent point cloud fusion and axis estimation, the interference is reduced and the processing efficiency is improved;

[0062] (2) The present application relies on the topological structure of the plane combined with the hole of the gas film hole itself, which provides a relatively close coarse registration result for the point cloud fusion of each direction of the gas film hole, and reduces the time required for fine registration;

[0063] ​​(3) The present application uses an iterative method to approximate the theoretical gas film hole point cloud axis position, which solves the difficulty of estimating the axis due to the inclination angle of the gas film hole axis relative to the machining plane and the standard column that is not perpendicular to the outer surface. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 The flowchart of the present application.

[0065] Figure 2 The octree data structure diagram.

[0066] Figure 3 The gas film hole point cloud preprocessing diagram of the present application.

[0067] Figure 4 The gas film hole each direction point cloud hole axis and hole outer plane estimation diagram of the present application.

[0068] Figure 5 The gas film hole each measurement direction point cloud fusion diagram of the present application.

[0069] Figure 6 The gas film hole point cloud axis direction estimation diagram of the present application. DETAILED DESCRIPTION

[0070] The present application is further illustrated below in combination with the drawings and examples.

[0071] Referring to Figure 1 A processing and axis estimation method for gas film hole point cloud, comprising: a processing and analysis method for gas film hole original point cloud data obtained based on zoom microscopy measurement technology, first removing outliers from the original measurement point cloud of the measured gas film hole in each direction by using an outlier segmentation method based on octree nearest point distance statistics, and the octree data structure is as shown in Figure 2 The method removes noise points and only retains effective data points, and uses a point cloud data downsampling method based on voxel grid to downsample the point cloud, downsampled with the required point cloud precision as the voxel size, while retaining the point cloud topographic features, reducing the number of points in the point cloud, improving the efficiency of the subsequent registration algorithm, and finally obtaining the gas film hole point cloud preprocessing result in each direction;

[0072] Secondly, an iterative closest point algorithm is used, which takes the center point position of the orifice and the fitting axis as the reference. Because the gas film hole itself has obvious characteristics, its topological structure can be summarized as the intersection of a plane and a cylinder. Therefore, by fitting the plane to divide the orifice wall point cloud and finding the orifice center coordinates and the orifice axis, the approximate positional relationship between the point clouds of the gas film holes in each measuring direction can be determined. Then, the point cloud is roughly registered to a position close to the complete registration, which is used as the basis for the iterative closest point algorithm for fine registration, greatly improving the accuracy and efficiency of fine registration. Then, the iterative closest point algorithm is used for fine registration of the point clouds in each direction.

[0073] Finally, the axis is iteratively estimated and evaluated. Multiple plane ellipse centers are estimated using least squares, and a straight line is fitted as the basis for pose transformation. Iterative estimation is repeated until the optimal axis is obtained.

[0074] A method for processing and evaluating gas film hole point clouds, comprising the following steps:

[0075] S1, measuring the gas film hole from k different directions to obtain k different direction gas film hole original point clouds pc1, pc2, pc3……pc k ;

[0076] Taking the point cloud pc i in the i-th direction as an example, find the minimum circumscribed cube of pc i in the three-dimensional space, which is the cube represented by the octree root node; select all points belonging to the cube and record them under the node; if there are more than 1 points under the node, create 8 child nodes under the node, if there are less than or equal to 1 points, stop recursion, recursively 8 child nodes, if it is found that a child node contains the same number of points as its parent node, stop recursion, and finally obtain the point cloud pc i recorded by the octree data structure.

[0077] Referring to Figure 3 , the point cloud is first removed from the outliers, and then down-sampled; traverse the points p i in pc δ , and find the n nearest points np δj to point p δ , and calculate the average distance dis δ between the n nearest points np δ and point p δj :

[0078]

[0079] where: dis is the average distance between a point and its nearest points; n is the number of nearest points; p δThe j-th nearest neighbor; δ—the number of the point;

[0080] Calculate all points p in the point cloud δ dis δ mean and standard deviation

[0081]

[0082]

[0083] in: —The mean of the average distances between all points and their neighbors; N —The number of points in the point cloud; dis δ —The average distance between the δth point and its nearest neighbors; —The standard deviation of the average distance between all points and their neighbors;

[0084] Select PC i All of them satisfy:

[0085]

[0086] Where: ρ—standard deviation ratio threshold; S dis —The variance of the average distance between all points and their nearest neighbors;

[0087] That is, find dis δ Compare Larger ρ times point p δ And remove, finally obtaining pc after removing discrete points. i ′ is the point cloud of the air film pores after removing discrete points in the i-th direction;

[0088] Calculate the point cloud pc of the air film pores after removing discrete points. i Find the smallest bounded cube of ′, and divide the cube into M cube voxel ranges according to the voxel size q;

[0089] Traverse the range of M cube voxels. If there is a point in the range of the Lth voxel, proceed to step S107; otherwise, traverse the next point.

[0090] Calculate the centroid C of the midpoint of the range of the Lth voxel. L :

[0091]

[0092] Where: C L —The centroid of the midpoint of the Lth idiotype range; m —The number of midpoints in the Lth idiotype range;

[0093] The center of gravity C of the midpoint of the Lth voxel range L As the result of the voxel downsampling, record the down-sampled point cloud pc" in the ith direction i In the middle, after traversing all points in the point cloud, the pre-processed gas film hole point cloud pc" in the ith direction is obtained i ;

[0094] S2, since the actual size of the gas film hole is small, it can be approximated that the above pre-processed gas film hole point cloud can be divided into two parts of a plane with a hole and a hole wall structure, therefore, as shown in Figure 4 , using a plane model segmentation based on random sampling consistency, the pre-processed gas film hole point cloud pc1", pc2", pc3" … pc k " is segmented into a plane part, and its plane equation is solved, and then another part of the point cloud is clustered and segmented based on Euclidean distance, and the cluster with the most points is extracted as the hole wall point cloud to fit a cylindrical surface, and the axis equation is solved;

[0095] Taking the point cloud pc1" and the point cloud pc ω " as an example, suppose the segmented plane is E1 and E ω , and the segmented cylindrical axis is Γ1 and Γ ω , and their equations are:

[0096] Ψ1:a1x+b1y+c1z=d1

[0097] Ψ i :a ω x+b ω y+c ω z=d ω

[0098] Γ1:

[0099] Γ i :

[0100] Where: Ψ ω - the plane after segmentation of the point cloud pc ω "; Γ ω - the cylindrical axis after segmentation of the point cloud pc ω "; a ω - the first parameter of the plane equation; b ω - the second parameter of the plane equation, same as the cylindrical axis; c ω - the third parameter of the plane equation; d ω - the fourth parameter of the plane equation; e ω - Ψ ω and Γ ωThe x-coordinate of the intersection point; f ω ——Ψ ω With Γ ω The ordinate of the intersection point; gω——Ψ ω With Γ ω The perpendicular coordinates of the intersection point; x—horizontal coordinate; y—vertical coordinate; z—vertical coordinate; ω—point cloud and plane number;

[0101] Where point C1:(e1,f1,g1) and point C ω :(e ω ,f ω ,g ω ) are the intersection points of Ψ1 and Γ1, and Ψ ω With Γ ω The intersection point;

[0102] Then from point cloud PC ω The rotation transformation matrix for "coarse registration to pc1" is:

[0103]

[0104] Where: R ω1 —From point cloud PC ω Rotation transformation matrix for "coarse registration to pc1"; a1, a ω —The first parameter of the plane equation; b1, b ω —The second parameter of the plane equation is the same as the axis of the cylinder; c1, c ω —The third parameter of the plane equation;

[0105] The translation transformation vector is:

[0106]

[0107] Where: M ω1 —Point Cloud PC ω Translation vector from coarse registration to pc1;

[0108] Then from point cloud PC ω The homogeneous transformation matrix from coarse registration to pc1 is:

[0109]

[0110] Wherein: T ω1 —From point cloud PC ω The homogeneous transformation matrix from coarse registration to pc1; R ω1 —From point cloud PC ω The rotation transformation matrix for "coarse registration to pc1";

[0111] The point cloud pc is obtained using the method described above. ωThe coarse registration transformation of "coarse registration to pc1", using which the fine registration is performed using the IPC algorithm to obtain the point cloud PC;

[0112] Referring to Figure 5 The point clouds of each part are registered together to obtain the overall point cloud of the gas film hole; in actual implementation, since the point clouds of adjacent visual angles have more overlapping parts and higher registration accuracy, PC is obtained by registering pc2" and pc1", and then the point clouds are registered to PC one by one starting from pc3", and finally the overall point cloud PC of the gas film hole is obtained;

[0113] S3, since the axis of the gas film hole has an inclination angle relative to the machining plane and is not a regular cylinder, an iterative method is used to approximate the theoretical axis position; referring to Figure 6 , first, the initial axis is estimated, and the pose transformation matrix T i The point cloud is aligned, that is, the initial axis after transformation passes through the origin and is perpendicular to the XOY plane, and then a plurality of slice planes are selected with h as the slice interval:

[0114] S τ :I τ x+J τ y+K τ (z-h*τ)=μ τ (τ=1,2,3,……)

[0115] Wherein: S - plane equation; I - first parameter of plane equation; J - second parameter of plane equation; K - third parameter of plane equation; μ - fourth parameter of plane equation; x - horizontal coordinate; y - vertical coordinate; z - vertical coordinate; h - slice interval; τ - plane number;

[0116] From the origin along the positive direction of the Z axis, for each plane, take the plane as the reference, and extend along the Z axis in two directions to form a space with a space of h d / 2, wherein the points contained in the space are the τth slice point cloud U τ ;

[0117] Starting from τ=0, for each point cloud U τ , verify whether the X and Y coordinates of all points cover at least three quadrants, if not, remove the point cloud, continue to verify the U1 point cloud, until a point cloud that meets the integrity requirement is found;

[0118] For all points in each of the above point clouds, the mean value of the Z coordinates of the points is calculated as the Z coordinate of the estimated center point V, that is: Where ε is the number of points contained in the point cloud;

[0119] For the X, Y coordinates of the center point V, the least square method is used to construct an equation to fit the ellipse in the following way:

[0120] The general equation of the ellipse is: Ax 2 +Bxy+Cy 2 +Dx+Ey+F=0

[0121] Let W=[A,B,C,D,E,F] T ,X=[x 2 ,xy,y 2 ,x,y,1] T ,

[0122] The optimization objective is: min‖W T X‖ 2 =W T XX T W, where due to the parameter constraint of the ellipse 4AC-B 2 >0

[0123] Therefore, W T HW>0, where

[0124] A - the first parameter of the ellipse equation; B - the second parameter of the ellipse equation; C - the third parameter of the ellipse equation; D - the fourth parameter of the ellipse equation; E - the fifth parameter of the ellipse equation; F - the sixth parameter of the ellipse equation; x - the horizontal coordinate; y - the vertical coordinate; W, X, H - equation parameter matrix;

[0125] Since ‖W T X‖ 2 =0, W can have a scaling factor α, i.e. W′=αW also satisfies the above condition, so here W T HW=1, so the problem is simplified to min‖W T X‖ 2 =W T XX T W,(W T HW=1);

[0126] The Lagrange function L(W, λ)=W T XX T W-λ(W T HW-1)

[0127] Take the partial derivative of W and let it equal to 0, i.e.

[0128] XX T W=λHW

[0129] λ - the Lagrange multiplier;​

[0130] i.e. transformed into the feature vector ;

[0131] After obtaining a plurality of center points, a straight line formed by fitting these points using the least square principle is used as the obtained axis line to replace the initial axis line above and the above process is repeated to calculate and update the pose transformation matrix T i Calculate the rotation Euler angle:

[0132] Take the rotation matrix The quaternion Q can be calculated as

[0133] Where: Q - the rotation matrix obtained from the pose transformation matrix T i ; r - the rotation parameter;

[0134] - the quaternion; q0, q1, q2, q3 - four parameters of the quaternion equation; a - the first imaginary part of the quaternion; β - the second imaginary part of the quaternion; γ - the third imaginary part of the quaternion;

[0135]

[0136] Then

[0137]

[0138] Where: yaw - the yaw angle; pitch - the pitch angle; roll - the roll angle;

[0139] If the three Euler angles yaw, pitch, roll are all less than the set threshold, the iteration is terminated;

[0140] The final axis pose is multiplied by the inverse of the pose transformation matrix T i (i from back to front) multiple times, respectively, i.e. the pose of the axis in the initial state is obtained.

[0141] It is apparent that the described embodiments are part of the embodiments of the present application, but not all. The components of the embodiments of the present application described and shown herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

Claims

1. A method for processing and axis estimation of a gas film hole point cloud, characterized in that, The method comprises the following steps of: S201, approximating the blade surface outside the hole as a plane, and dividing the N-direction gas film hole point cloud preprocessing results into the blade surface outside the hole and the remaining point cloud by a plane model segmentation based on a random sample consensus (RANSAC); S202, obtaining a plane equation according to an estimated normal vector of the blade surface outside the hole; S203, performing clustering segmentation on the remaining point cloud based on a Euclidean distance, and extracting a cluster with the largest number of points as the hole inner wall point cloud; S204, estimating the axis direction of the hole inner wall point cloud by using a cylindrical fitting, and calculating the intersection point of the axis and the plane; S205, repeating steps S201, S202 and S203 to calculate the plane equation and the cylindrical part point cloud axis direction of the gas film hole point cloud preprocessing results in all directions, and correcting all the point clouds to the same direction based on the gas film hole point cloud preprocessing results in the first direction to complete the coarse registration of the point clouds; S206, using an iterative closest point (ICP) algorithm to sequentially register the gas film hole point cloud preprocessing results in each direction to the gas film hole point cloud preprocessing results in the first direction to obtain the fine registration result of the gas film hole point cloud, that is, the complete gas film hole point cloud; The method for estimating the gas film hole axis comprises the following steps of: S301, obtaining a plurality of two-dimensional sections by using a plurality of parallel planes, and evaluating the completeness of the section point cloud to lay a foundation for the subsequent estimation process; step S301 specifically comprises the following steps of: S3012, selecting a plurality of planes parallel to the XOY plane as slice planes according to the slice interval h; S3014, numbering the obtained point cloud from bottom to top along the positive direction of the Z axis, and checking the completeness of the point cloud from the point cloud with the smallest number, that is, observing whether the point cloud covers at least three quadrants from the Z axis direction; if the completeness is poor, the point cloud is removed until the completeness of the point cloud meets the requirements; S3011, using the axis obtained in step S204 as an initial axis, solving a pose transformation matrix T i transforming the initial axis to be perpendicular to the XOY plane and passing through the origin, using the pose transformation matrix T i performing the same pose transformation on the complete gas film hole point cloud as a whole; S302, estimating the center position of each point cloud to estimate the axis position; step S302 specifically comprises the following steps of: S3013, selecting a slice threshold h d , using the slice plane in step S3012 to slice the complete gas film hole point cloud to obtain a plurality of point clouds with a height range of h d ; S3021, taking the average value of the Z coordinates of the points in each point cloud as the Z coordinate of the estimated center point V; S3022, the center estimation only uses the X and Y coordinates of each point, which is simplified to a two-dimensional graph fitting, an elliptic general equation is used as a template, a least square method is used to construct an equation, and a Lagrange multiplier method is used to solve the equation to obtain the center coordinates of the ellipse as the X and Y coordinates of the center point V; S3023, for each center point V, a straight line equation is obtained by using a least square fitting, which is denoted as a straight line l; ​ ​ S303、According to the obtained straight line l, repeat the steps of S301 and S302 as the initial axis in S301 until the final transformation angle is less than the threshold value, then stop iteration, and the accumulated transformation matrix T is obtained according to the transformation matrix T i The equation of the axis is calculated.

2. The method for processing and axis estimation of a gas film hole point cloud according to claim 1, characterized in that, The gas film hole point cloud preprocessing specifically comprises: firstly, removing outliers of original gas film hole point cloud data, based on an octree data structure point cloud data storage scheme, counting the average distance of each point on the point cloud and surrounding points, and screening out outliers according to a standard deviation multiple threshold; and then, dividing the entire point cloud into a voxelized grid of unit volume, calculating the center of gravity of all points in each voxel as the position of the point after downsampling.

3. The method for processing and axis estimation of a gas film hole point cloud according to claim 1, characterized in that, The specific steps of the gas film hole point cloud preprocessing are as follows: S101, filling all points in the gas film hole point cloud into an octree data structure; S102, Traversing point p in the point cloud δ And find the n nearest neighbors using the octree, and calculate np of the n nearest neighbors. δj Point P δ The mean of the distance dis δ : Wherein: dis - the average distance between a certain point and its adjacent points; n - the number of adjacent points; p δ - Traverse the selected δ point; np δj - Point p δ jthadjacent point of point p; δ - the number of points; S103、Calculate all points p in the point cloud δ dis δ mean and standard deviation where: — the mean of the average distances of all points to their neighbors; N — the number of points in the point cloud; dis δ — the average distance between the δth point and its neighbors; — the standard deviation of the average distances of all points to their neighbors; S104, selecting all points satisfying: where: p - standard deviation ratio threshold; S dis - variance of the average distance of all points from their nearest neighbors; i.e. find dis δ compared to large p times point p δ and remove, finally obtaining the gas film hole point cloud after removing the discrete points; S105, calculating the minimum circumscribed cube of the gas film hole point cloud after removing discrete points, and dividing the cube into M square voxel ranges according to the voxel size q; S106, traversing the M square voxel ranges, if there is a point in the Lth voxel range, then performing step S107, otherwise traversing the next point; S107, calculate the center of gravity C of the point of the Lth voxel range L : where: C L - the barycenter of the points of the Lth voxel range; m - the number of points of the Lth voxel range; S108, the center of gravity C of the Lth voxel range L As the result of the voxel downsampling, the preprocessed gas film hole point cloud is obtained after traversing all the points in the point cloud is recorded in the downsampled point cloud.

4. The method for processing and axis estimation of a gas film hole point cloud according to claim 3, characterized in that, The step S101 specifically comprises: S1011, finding the minimum circumscribed cube of the gas film hole point cloud in the three-dimensional space as the cube represented by the root node of the octree; S1012, selecting all points belonging to the cube and recording them under the node; S1013, if there are more than 1 points under the node, creating 8 child nodes under the node, if there are less than or equal to 1 points, stopping recursion; S1014, recursively using steps S1012 and S1013 for 8 child nodes, if it is found that the number of points contained in a child node is equal to the number of points contained in its parent node, then stopping recursion.

5. The method for processing and axis estimation of a gas film hole point cloud according to claim 1, wherein, The step 201) specifically comprises: S2011, randomly extracting a subset S' of size η from the point cloud G, and using the least squares method to calculate the subset model parameters on G', wherein η is the minimum sample size required for initializing the model; S2012, obtaining the residual set R=G-G' of G' from G, calculating the deviation of all samples in the residual set under the model, if the deviation is less than a threshold, considering the sample point as an in-model sample point, otherwise as an out-of-model sample point, and recording the number of in-model sample points; S2013, calculating an iteration end evaluation factor according to the expected error rate of the model, the total number of samples, the number of in-model sample points of the current best model, and the current iteration number, if the factor reaches a certain threshold, ending the iteration, otherwise repeating steps S2011 and S2012; S2014, obtaining the final model parameter estimate after iteration, and taking the in-model sample points of the final model on the sample set S as the plane segmentation result.

6. The method for processing and axis estimation of a gas film hole point cloud of claim 1, wherein, The step 203) specifically comprises: S2031, setting the input point cloud data set as P, setting an empty clustering list C, and setting a point cloud queue Q to be checked; S2032, for each point p in P i add p to the current queue Q; i add p to the current queue Q; S2033, p in each Q i Set radius r < d th Search p in this range ii Nearest neighbor point cloud set P ik ; S2034, for each nearest neighbor point cloud set P ik , check whether the points in it have been processed, if not, add them to Q; S2035, when all points in Q are processed, adding Q to the clustering list C, and emptying Q, repeating steps S2033 and S2034 until all points in the data set P are processed; S2036, selecting a cluster with the most points in the clustering list C as the output point cloud P'.

Images