A contour size rapid analysis method based on curved surface skinning spot hole point cloud

Abstract

The application relates to a contour size rapid analysis method based on a surface skinning spot hole point cloud, which comprises the following steps: S1, performing voxel filtering on the scanned original point cloud P0 according to the installation angle of a laser sensor to obtain a point cloud P0 ＇ , then performing coordinate transformation and compensation of three directions of a laser sensor coordinate system X axis, Y axis and Z axis to obtain a transformed point cloud P1; S2, performing segmentation on the point cloud P1, and segmenting out effective data point cloud P2 and invalid data point cloud P3 by setting a default value; S3, adopting maximum outer circle fitting to extract a bottom hole center c and a bottom hole radius r for the invalid data point cloud P3; and S4, performing dimension reduction processing on the effective data point cloud P2 to obtain a point cloud P4 and the like. According to the structural features of the spot hole, the three-dimensional point cloud is segmented and dimensionally reduced, and the effect of fast, accurate and multi-parameter one-time measurement extraction is achieved, the traditional detection time is shortened from several minutes to 1.5 seconds or below, and the detection precision and stability are greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of aircraft skin manufacturing, and particularly relates to a contour size rapid analysis method based on a curved skin counterbore point cloud. BACKGROUND

[0002] The aircraft skin is characterized by large size, complex shape, and high surface quality requirement, and has a complex profile. In order to reduce the weight of the aircraft, the lightest possible material is used in the aircraft manufacturing according to the local conditions, such as the use of a large amount of composite materials in the most advanced international passenger aircraft to reduce the weight of the aircraft. The special material and thin size characteristics of the aircraft skin make it impossible to use welding process for connection, and the mutual connection through rivets is the best choice at present.

[0003] The rivets play a crucial role in the connection of the aircraft skin, and they are not only simple connectors, but also key factors to ensure the structural integrity and flight safety of the aircraft. The number of rivets on an aircraft is several ten thousand or even up to one million. The precision of the rivet assembly of the aircraft skin will affect the airflow during the flight of the aircraft. In the manufacturing and inspection process of the aircraft skin, the machining error of the rivet hole is usually an important basis for the delivery of the skin inspection, and is an intuitive standard for verifying that the skin meets the requirements of the subsequent assembly process. If the machining precision of the rivet hole does not meet the requirements, it will affect the flatness of the fastener, cause poor assembly, and even have a negative impact on the performance and fuel economy of the aircraft. Military aircraft require higher assembly requirements for fasteners in order to achieve better stealth effect and electrolyte filling in the fastener area. Therefore, whether it is a plane or a curved surface, the rivet hole of the skin needs to be accurately measured to ensure the good assembly of the fastener.

[0004] To accurately measure the contour of the rivet hole on the curved surface, the traditional measurement method is to use special mechanical gauges for measurement, such as using a counter gauge, an internal diameter micrometer, a three-coordinate measuring instrument or a standard part for contact detection. The disadvantage is that these detection methods can only detect one or two parameters at a time, have poor universality, low detection efficiency, and some important parameters such as axial deviation and small diameter cannot be effectively detected. The measurement results are greatly affected by human factors, and cannot meet the application requirements of automatic production lines.

[0005] However, the existing point cloud analysis technology also has significant shortcomings: the circle fitting and detection of three-dimensional point cloud are very complex, the least square method cannot handle high-dimensional cases and is disturbed by abnormal points, the detection iteration number of RANSAC is too large, and the randomness of the initial point selection sometimes has a great influence on the analysis results. The precision, stability and time complexity cannot meet the online detection requirements of automatic production lines. SUMMARY

[0006] In order to solve the above problems, the purpose of the present application is to provide a profile size rapid analysis method based on curved skin spot cloud.

[0007] In order to achieve the above purpose, the technical scheme adopted by the present application is: a profile size rapid analysis method based on curved skin spot cloud, specifically comprising the following steps:

[0008] S1, according to the installation angle of the laser sensor, the original point cloud P0 scanned is voxel filtered to obtain the point cloud P0 ＇ , and then the coordinate transformation and compensation of the X axis, Y axis and Z axis of the laser sensor coordinate system are carried out, and the transformed point cloud P1 is obtained;

[0009] S2, the point cloud P1 is segmented, and the effective data point cloud P2 and the invalid data point cloud P3 are segmented by setting the default value;

[0010] S3, the invalid data point cloud P3 is extracted by adopting the maximum outer circle fitting to extract the bottom hole center c and the bottom hole radius r;

[0011] S4, the effective data point cloud P2 is processed by dimension reduction to obtain the point cloud P4;

[0012] S5, the point cloud P4 is segmented by adopting the point cloud filtering method to obtain the point cloud P5, and the SVD decomposition method is adopted to analyze the plane fitting of the point cloud P5, the three-dimensional plane equation of the large diameter S1 plane is obtained, and the normal vector V of the large diameter S1 plane is extracted;

[0013] S6, taking the large circle center C as the segmentation center, the point cloud P4 is segmented based on the XZ plane of the point cloud coordinate system to obtain two symmetrical point clouds, and one of the two point clouds is taken as P6, the Y axis direction outer contour of the point cloud P6 is extracted to obtain the point cloud P7, and the point cloud P7 is mapped to the YZ plane to obtain the two-dimensional point cloud P8;

[0014] At the same time, taking the large circle center C as the segmentation center, the point cloud P4 is segmented based on the YZ plane of the point cloud coordinate system to obtain two symmetrical point clouds, and one of the two point clouds is taken as P6 ＇ , the X direction outer contour of the point cloud P6 ＇ is extracted to obtain the point cloud P7 ＇ , and the point cloud P7 is mapped to the XZ plane to obtain the two-dimensional point cloud P8 ＇ ;

[0015] S7, the least square method is adopted to analyze the straight line fitting of the two-dimensional point cloud P8 to obtain four straight lines, two of which are the generatrix L1 and L2 of the conical surface, and the other two are the generatrix L3 and L4 of the bottom hole cylindrical surface;

[0016] Similarly, the two-dimensional point cloud P8 ＇The least squares method was used for line fitting analysis, resulting in four lines, two of which are generatrices L1 of the conical surface. ＇ and L2 ＇ Two are the generatrices L3 of the bottom hole cylindrical surface. ＇ and L4 ＇ ;

[0017] S8, to obtain the depth H of the notch and the diameter of the bottom hole, i.e., the minor diameter S2;

[0018] S9, calculate the axis L of the conical surface cavity using the generatrices L1 and L2 of the conical surface. w The hole axis L is calculated using the generatrices L3 and L4 of the bottom hole cylindrical surface. k And according to L w With L k The coaxiality deviation A of the hole axis is obtained by calculating the included angle δ and the depth H of the hole.

[0019] Furthermore, in step S4, the set of points on a straight line parallel to the Z-axis passing through the center C of the large circle is used as key points. A point cloud dimensionality reduction method based on gradient descent with key points as the center is used to reduce the dimensionality of the effective data point cloud P2 to obtain point cloud P4.

[0020] Furthermore, the point cloud dimensionality reduction method in step S4 specifically includes the following steps:

[0021] S401. Define the objective function:

[0022] a) Assume that the valid data point cloud P2 has n 3D point cloud data points ,as well as m Key points ; The point after dimensional reduction At the key point Gathering around; Represents three-dimensional Euclidean space;

[0023] b) Use the Gaussian kernel function to measure the 3D point cloud data points. With key points Similarity, Gaussian kernel function for:

[0024] ;

[0025] Where exp(·) is the natural exponential function, and σ is a hyperparameter that controls the width of the Gaussian kernel function; the Gaussian kernel function The larger the value, the better. and The more similar they are, the closer they are to each other;

[0026] Define the objective function for:

[0027] ;

[0028] Objective function The meaning is: for each three-dimensional point cloud data point , the similarity between it and all key points is calculated by the Gaussian kernel function , and the similarity is used as a weight to measure the distance square between the reduced point of the three-dimensional point cloud data point and the reduced point of each key point ;

[0029] S402, calculate the gradient:

[0030] In order to minimize the objective function using the gradient descent method , it is necessary to calculate the gradient of :

[0031] c) Set the weight , that is, the Gaussian kernel function as the weight, then:

[0032] ;

[0033] For the vector , w is the dimension of the vector, respectively Calculate the partial derivative of the l component of ;

[0034] The partial derivative of with respect to is obtained by the following formula, where represents the component of the i point of k , represents the component of the j point of k :

[0035] ;

[0036] Therefore:

[0037] ;

[0038] Then, the gradient is:

[0039] ;

[0040] in, for about The gradient, i.e., only focusing on When changing The rate of change; Represents the variable Find the gradient; and For vectors, ,represent Pointed l Quantity , ,represent Pointed l Quantity ; w Let be the dimension of the vector;

[0041] S403, Gradient Descent Update:

[0042] e) Gradient descent minimizes the objective function by continuously updating the parameters along the negative direction of the gradient;

[0043] f) In each iteration number t In the middle, the dimensionality-reduced point vectors are updated based on the gradient. The learning rate α controls the step size for each update, and the update formula is:

[0044] ;

[0045] Specifically, for The updated for:

[0046] ;

[0047] Repeat this update process until the objective function converges.

[0048] Further, in step S5, when dividing point cloud P4 to obtain point cloud P5, the radius R and center C of the large circle at the top of the conical hole are detected along the point cloud scanning direction with the center c of the bottom hole as the center. With the center C of the large circle as the center, the upper surface thickness Z of the effective data point cloud P2 as the radius and the upper surface thickness Z of the effective data point cloud P2 as the upper surface thickness difference, the point cloud P4 is divided by the point cloud filtering method to obtain point cloud P5.

[0049] Furthermore, in step S6, the method for obtaining the two-dimensional point cloud P8 is specifically as follows:

[0050] S601, Assume the point cloud in point cloud P4 is... in is the i-th point in the point cloud P4; k

[0051] S602, taking the center of the circle as the segmentation center, respectively, the coordinate values of the X-axis, Y-axis and Z-axis directions of the center of the great circle C in the point cloud coordinate system, and the XZ plane is segmented: For the XZ plane, its equation is

[0052] = 0; with the plane passing through the center of the great circle C and parallel to the XZ plane y y y c Segmentation is performed; after segmentation, two symmetrical point clouds are obtained, and one of them is taken as P6, assuming that the part of y > y c is taken, denoted as the point cloud ;

[0053] S603, extracting the Y-axis direction outer contour of the point cloud P 1 to obtain a new point cloud P7:

[0054] Let P 2 be the new point cloud P7, for each three-dimensional point in the point cloud P7, its spatial coordinates can be represented as x , y , z ), x , y , z respectively corresponding to the coordinate values of three orthogonal directions in the three-dimensional space; wherein x , z represent the x component and the z component of the point coordinate, that is, the position parameters of the point in the X-axis direction and the Z-axis direction in the point cloud coordinate system; for each group of x , z value, find the point with the maximum P value in y 1; that is, for given x and z , if there are multiple points satisfying , select the point with the maximum value y from these points to join P 2;

[0055] S604, mapping P 2 to the YZ plane to obtain a two-dimensional point cloud P8:

[0056] Since in the YZ plane, the X-axis coordinate x is 0;

[0057] ​​​So for each point in P2 P 2 , it is mapped to ;

[0058] Let the mapped 2D point cloud P8 be P8 P 3

[0059] .

[0060] Further, the step S8 specifically includes the following steps:

[0061] Suppose L1 and L3 are left busbars, and L2 and L4 are right busbars, find the intersection point D1 of the straight line L1 and the straight line L3, and the intersection point D2 of the straight line L2 and the straight line L4; find the first depth value H1 through the point D1, the point D2 and the center C of the circle, and the calculation formula is:

[0062] H1=Z C -(D1 z +D2 z ) / 2;

[0063] Wherein, Z C represents the coordinate value of the center C of the circle in the Z-axis direction of the point cloud coordinate system, D1 z represents the coordinate value of the intersection point D1 in the Z-axis direction of the point cloud coordinate system, and D2 z represents the coordinate value of the intersection point D2 in the Z-axis direction of the point cloud coordinate system.

[0064] Find the first minor diameter value S2 ＇ through the straight line L3 and the straight line L4, S2 ＇ is the shortest horizontal distance between the straight line L3 and the straight line L4 from the intersection point D1 and the intersection point D2 to the bottom surface of the bottom hole;

[0065] Suppose L1 ＇ and L3 ＇ are left busbars, and L2 ＇ and L4 ＇ are right busbars, find the intersection point D1 ＇ of the straight line L1 ＇ and the straight line L3 ＇ , and the intersection point D2 ＇ of the straight line L2 ＇ and the straight line L4 ＇ , find the second depth value H2 through the point D1 ＇ , the point D2 ＇ and the center C of the circle, and the calculation formula is:

[0066] H2=Z C -(D1 z +D2 ＇ ) / 2;z ＇ ) / 2;

[0067] Wherein, Z C represents the coordinate value of the great circle center C point in the Z axis direction of the point cloud coordinate system, D1 z ＇ represents the intersection D1 ＇ The coordinate value of the Z axis in the point cloud coordinate system, D2 z ＇ represents the intersection D2 ＇ The coordinate value of the Z axis in the point cloud coordinate system;

[0068] H2 is compared with H1, when the difference is within the tolerance range of the countersinking structure of the aircraft skin surface, the further calibration is obtained, the depth H= (H1+H2) / 2;

[0069] At the same time, the second small diameter value S2 ＇ is obtained by the straight line L3 ＇ and the straight line L4 ＂ , S2 ＂ is compared with S2 ＇ , when the difference is within the tolerance range, the small diameter S2 is further calibrated, and the calculation formula is:

[0070] S2= (S2 ＇ +S2 ＂ ) / 2.

[0071] Further, in the step S9, the conical surface hole axis L w is calculated by the straight line L1 and L2, and the vector equation is:

[0072] L w = (L1+L2) / 2;

[0073] The hole axis L k is calculated by the straight line L3 and L4, and the vector equation is:

[0074] L k = (L3+L4) / 2.

[0075] Further, in the step S9, the calculation formula of the hole axis coaxial degree deviation A is:

[0076] A=H×tan (δ);

[0077] In the formula, δ is the included angle of L w and L k .

[0078] Compared with the prior art, the present application has the following beneficial effects:

[0079] 1. The application effectively processes the data point cloud P2 through the point cloud dimension reduction processing of the gradient descent method to obtain the point cloud P4, which not only realizes the point cloud dimension reduction purpose, but also maximally retains the key feature data, and ensures the rapidity and accuracy of subsequent analysis.

[0080] 2. On the basis of the point cloud P4, the point cloud filtering method is used for segmentation to obtain the point cloud P5, and the decomposition method is used for plane fitting analysis on the point cloud P5 to extract the large diameter S1 and the normal vector V of the counterbore.

[0081] 3. The application segments the point cloud P4 through the XZ plane and the YZ plane to obtain the point clouds P7 and P7 ＇ , and after mapping, two-dimensional point clouds P8 and P8 ＇ are obtained, key contour features are accurately extracted, and the calculation complexity is greatly reduced; and the two-dimensional point clouds P8 and P8 ＇ are subjected to straight line fitting analysis respectively to obtain the corresponding conical surface generatrix and bottom hole cylindrical surface generatrix, and on this basis, the counterbore depth H, the small diameter S2 and the hole axis coaxiality deviation A are calculated.

[0082] In summary, the contour size rapid analysis method based on the counterbore point cloud of the curved skin of the application, according to the structural characteristics of the counterbore, through the segmentation and dimension reduction operation of the three-dimensional point cloud, the effect of rapid, accurate and multi-parameter (parameters such as the counterbore normal vector V, the hole axis coaxiality deviation A and the small diameter S2 can be detected) one-time measurement extraction is achieved, the traditional detection time is shortened from several minutes to less than 1.5 seconds, and the detection accuracy and stability are greatly improved. The method not only can replace the traditional manual measurement method based on mechanical gauge measurement, but also is more suitable for integration into the skin manufacturing automation production line to realize online rapid measurement, real-time information feedback, improve the integration degree, size measurement accuracy and measurement efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0083] Figure 1 is a flow chart of a contour size rapid analysis method based on the counterbore point cloud of the curved skin of the application;

[0084] Figure 2 is a schematic view of the original point cloud P0 (not containing the invalid data point cloud P3);

[0085] Figure 3 is a schematic view of the maximum circle fitting point cloud of the invalid data point cloud P3;

[0086] Figure 4 is a top view of the point cloud P5;

[0087] Figure 5 is a schematic view of the two-dimensional point cloud P8;

[0088] Figure 6This is a schematic diagram of the countersink structure on the aircraft skin surface, showing the conical generatrices L1 and L2, the cylindrical generatrices L3 and L4 of the bottom hole, the intersection point D1, the intersection point D2, and the hole axis L. w Hole axis L k ;

[0089] In the diagram, 1 is the conical hole; 2 is the pilot hole; S1 is the major diameter; S2 is the minor diameter; T1 is the aircraft skin surface; T2 is the curved transition surface; H is the recess depth; L is the depth of the recess. w , socket axis; L k A. Hole axis; B. Hole axis coaxiality deviation; C. Center of the large circle; D. Center of the bottom hole; R. Radius of the bottom hole. Detailed Implementation

[0090] The following examples are used to illustrate the present invention, but are not intended to limit the scope of the invention.

[0091] Aircraft curved skin countersink structure such as Figure 6 As shown, the countersink includes a conical hole 1 and a bottom hole 2. The conical hole 1 has a conical surface inside, and the bottom hole 2 has a cylindrical surface inside. The diameter of the major circle at the top of the conical hole 1 is the major diameter S1, the center of the major circle is C, the radius is R, and the depth of the notch is H. The diameter of the bottom hole 2 is the minor diameter S2, and the major diameter S1 is larger than the minor diameter S2. The center of the bottom hole 2 is c, and the axis of the notch is L. w The hole axis is L k L-axis of the nest w and hole axis L k The included angle between them is δ, the coaxiality deviation of the hole axis is A, and the surface type of the aircraft skin is T1; the conical surface of the conical hole 1 and the cylindrical surface of the bottom hole 2 are connected by an arc-shaped transition surface T2.

[0092] Example

[0093] like Figure 1 As shown, a rapid analysis method for the contour dimensions of countersunk hole point clouds based on curved skin is specifically included in the following steps:

[0094] S1, based on the installation angle of the laser sensor (laser sensors are generally installed at an angle with a fixed angle during measurement), voxel filtering is performed on the original scanned point cloud P0 to obtain point cloud P0. ＇ Then, transformations and compensations are performed in the X, Y, and Z axes of the laser sensor coordinate system to obtain the transformed point cloud P1; the original point cloud P0 (which does not contain invalid data point cloud P3) is as follows. Figure 2 As shown;

[0095] Wherein, the X-axis direction is the direction of the laser dot matrix line of the laser sensor, the Y-axis direction is the scanning direction of the laser sensor, and the Z-axis direction is the laser emission direction (existing technology).

[0096] Coordinate transformation and compensation method is a known technology, such as Journal of Metrology, Vol. 41, No. 12, December 2020, Error compensation method of workpiece size measurement system based on line laser sensor;

[0097] S2, segmenting the point cloud P1, and segmenting the effective data point cloud P2 and the invalid data point cloud P3 by setting a default value; the effective point cloud P2 is a relatively complete counterbore structure, and the invalid data point cloud P3 is an invalid point cloud formed by the bottom hole 2 of the counterbore;

[0098] The laser sensor (in the present application, a line laser sensor) is measured by triangulation, including a transmitter and a receiver. When a counterbore structure similar to the intermediate through hole is scanned by a line laser, the intermediate through hole part cannot reflect the laser back, causing the data of this part to be invalid, so a default value is assigned. The default value is generally set to be greater than the maximum range of the Z direction of the laser sensor and less than the invalid data of the laser sensor. For example, the measurement range of a certain laser sensor is -12.5~12.5, and the default value of the invalid data is 65535, which is much larger than the measurement range of the laser sensor. This feature can quickly realize the segmentation of the effective point cloud and the invalid point cloud;

[0099] S3, analyzing the invalid data point cloud P3, and extracting the bottom hole center c and the bottom hole radius r by maximum circle fitting; the maximum circle fitting point cloud of the invalid data point cloud P3 is as shown in Figure 3 ;

[0100] S4, performing dimension reduction processing on the effective data point cloud P2 to obtain the point cloud P4: taking a point on the straight line of the bottom hole center c as a key point, performing dimension reduction processing on the point cloud P2, and adopting a point cloud dimension reduction method based on gradient descent method with the key point as the center to obtain the point cloud P4; both the point cloud dimension reduction purpose and the key feature data are maximally retained, and the rapidity and accuracy of subsequent analysis are ensured;

[0101] The point cloud dimension reduction method specifically includes the following steps:

[0102] S401, defining a target function:

[0103] a) assuming that the effective data point cloud P2 has n three-dimensional point cloud data points , and m key points ; it is hoped that the points after dimension reduction gather around the key points ; representing a three-dimensional Euclidean space;

[0104] b) using a Gaussian kernel function to measure the three-dimensional point cloud data points and the key points similarity of the points, the Gaussian kernel function is:

[0105] ;

[0106] where exp(·) is the natural exponential function, and σ is a hyperparameter that controls the width of the Gaussian kernel function; the larger the value of the Gaussian kernel function indicates that the points are more similar and closer in distance;

[0107] The objective function is defined as:

[0108] ;

[0109] The objective function has the following meaning: for each three-dimensional point cloud data point , the similarity between it and all key points is calculated by the Gaussian kernel function , and this similarity is used as a weight to measure the distance squared between the reduced point of the three-dimensional point cloud data point and the reduced point of each key point ;

[0110] By double-summing the weighted distances of all three-dimensional point cloud data points and all key points , the overall optimization objective is formed; when minimizing this objective function, each will be closer to the reduced point corresponding to the key point similar to it in the original space (the anchor point with a larger kernel function value has a stronger influence), and ultimately the reduced points will gather around the corresponding key points, thereby preserving the local structure relationship in the original space;

[0111] S402, calculate the gradient:

[0112] In order to minimize the objective function using gradient descent, the gradient of with respect to needs to be calculated:

[0113] c) Set the weight , i.e. use the Gaussian kernel function as the weight, then:

[0114] ;

[0115] For the vector ,​​w Let be the dimension of the vector, and calculate respectively. about of l Quantity partial derivatives ;

[0116] Using the following formula Seeking information about The partial derivatives of, where, express The i Pointed k Quantity, express The j Pointed k Quantity:

[0117] ;

[0118] so:

[0119] ;

[0120] Then, gradient for:

[0121] ;

[0122] in, for about The gradient, i.e., only focusing on When changing The rate of change; Represents the variable Find the gradient; and For vectors, ,represent Pointed l Quantity , ,represent Pointed l Quantity ; w Let be the dimension of the vector;

[0123] S403, Gradient Descent Update:

[0124] e) Gradient descent minimizes the objective function by continuously updating the parameters along the negative direction of the gradient;

[0125] f) In each iteration number t In the middle, the dimensionality-reduced point vectors are updated based on the gradient. The learning rate α controls the step size for each update, and the update formula is:

[0126] ;

[0127] Specifically, for The updated for:

[0128] ;

[0129] Repeat this update process until the objective function converges.

[0130] Repeat this update process until the objective function converges; for example, when the norm of the gradient is less than a certain threshold, or when the maximum number of iterations is reached. In this way, the points after dimensionality reduction are gradually adjusted. The position of makes the objective function Minimize, thereby maximizing the density of points around key points after dimensionality reduction.

[0131] S5, with the center c of the bottom hole as the center, a one-dimensional straight line is used to detect the radius R and center C of the large circle at the top of the conical hole along the point cloud scanning direction. With the center C of the large circle as the center, and with a radius of (1.5~2)×R and the upper surface thickness Z of the effective data point cloud P2, the point cloud P4 is segmented using the point cloud filtering method to obtain point cloud P5. Point cloud P5 is as follows: Figure 4 As shown, the point cloud P5 was subjected to plane fitting analysis using the SVD decomposition method to obtain the three-dimensional plane equation of the plane containing the large diameter S1. The normal vector V of the plane containing the large diameter S1 of the counterboring hole was extracted. The center C of the large circle, the radius R of the large circle at the top of the conical hole 1, the upper surface thickness Z of the effective data point cloud P2, and the normal vector V are shown in the figure. Figure 2 ;

[0132] Among them, the point cloud filtering method and SVD decomposition method are known techniques for performing plane fitting analysis to obtain the three-dimensional plane equation of the plane containing the large diameter S1, and will not be elaborated here.

[0133] S6. Using the center C of the great circle as the dividing center, the point cloud P4 is divided into two symmetrical point clouds using the XZ plane based on the point cloud coordinate system. One of these point clouds is arbitrarily selected as P6. The outer contour of point cloud P6 along the Y-axis is extracted to obtain point cloud P7. Point cloud P7 is then mapped onto the YZ plane to obtain the two-dimensional point cloud P8. Point cloud P8 is shown below. Figure 5 As shown;

[0134] The method for obtaining the 2D point cloud P8 specifically includes the following steps:

[0135] S601, Assume the point cloud in point cloud P4 is... ,in It is the first point cloud in P4. k One point;

[0136] S602, take the center of the circle C as the center of segmentation, , respectively, the coordinate values of the X-axis, Y-axis and Z-axis directions of the center of the great circle C in the point cloud coordinate system, and segment in the XZ plane:

[0137] For the XZ plane, its equation is y = 0; take the plane passing through the center of the great circle C and parallel to the XZ plane y = y c Segment; after segmentation, 2 symmetrical point clouds are obtained, and one of them is taken as P6, assuming that the part of y > y c is taken, denoted as point cloud ;

[0138] S603, extract the new point cloud P7 from the Y-axis direction outer contour of the point cloud P 1:

[0139] To find the Y-axis direction outer contour, the boundary points in the Y-axis direction need to be found; for the point cloud P 1, in the Y-axis direction, these points are the points with the maximum value when the given x and z values are taken (because the part of y y > y c is taken before), then:

[0140] Let P 2 be the new point cloud P7, for each three-dimensional point in the point cloud P7, its spatial coordinates can be represented as ( x , y , z ), x , y , z corresponding to the coordinate values of the three orthogonal directions in the three-dimensional space; among them, ( x , z ) represents the x component and the z component of the point coordinate, that is, the position parameters of the point in the X-axis direction and the Z-axis direction in the point cloud coordinate system; for each group of ( x , z ) values, find the point with the maximum P value in y 1; that is, for the given x and z , if there are multiple points satisfying , select the point with the maximum value y from these points to join P 2;

[0141] S604,​P 2 mapping to YZ plane, get two-dimensional point cloud P8:

[0142] Since in YZ plane, X-axis coordinate x 0;

[0143] So for P 2 each point , it is mapped to ;

[0144] Let the mapped two-dimensional point cloud P8 be P 3, then:

[0145] .

[0146] At the same time, referring to the above method of obtaining two-dimensional point cloud P8 (step S601-step S604), taking the great circle center C as the segmentation center, the point cloud P4 is segmented based on the YZ plane of the point cloud coordinate system, and 2 symmetrical point clouds are obtained, and one of them is taken as P6 ＇ , the X outer contour of the point cloud P6 ＇ is extracted to obtain the point cloud P7 ＇ , and the point cloud P7 ＇ is mapped to the XZ plane to obtain the two-dimensional point cloud P8 ＇ ;

[0147] S7, the least square method is used for straight line fitting analysis on two-dimensional point cloud P8, and 4 straight lines are obtained, 2 of which are conical surface generatrix L1 and L2, and 2 of which are bottom hole 2 cylindrical surface generatrix L3 and L4. The position diagram of L1, L2, L3 and L4 is shown in Figure 6 ;

[0148] Similarly, two-dimensional point cloud P8 ＇ is analyzed, and the least square method is used for straight line fitting analysis, and 4 straight lines are obtained, 2 of which are conical surface generatrix L1 ＇ and L2 ＇ , and 2 of which are bottom hole cylindrical surface generatrix L3 ＇ and L4 ＇ ;

[0149] S8, the depth H and the diameter of the bottom hole 2, i.e. the small diameter S2, are obtained; the specific method is:

[0150] Assuming that L1 and L3 are left generatrix, and L2 and L4 are right generatrix, the intersection point D1 of straight line L1 and straight line L3 and the intersection point D2 of straight line L2 and straight line L4 are obtained, and the position diagram of D1 and D2 is shown in Figure 6 ; the first depth value H1 is obtained through the intersection point D1, the intersection point D2 and the great circle center C, and the calculation formula is:

[0151] H1=ZC -(D1 z +D2 z ) / 2;

[0152] Among them, Z C D1 represents the coordinates of the center C of the great circle along the Z-axis in the point cloud coordinate system. z D1 represents the coordinate value of the intersection point D1 in the Z-axis direction of the point cloud coordinate system, and D2 represents the intersection point D2. z This represents the coordinate value of intersection point D2 in the Z-axis direction of the point cloud coordinate system;

[0153] At the same time, the first minor diameter value S2 is obtained through lines L3 and L4. ＇ S2 ＇ It is the shortest horizontal distance between straight lines L3 and L4, extending downwards from points D1 and D2 to the bottom of hole 2.

[0154] Assume L1 ＇ and L3 ＇ For the left busbar, L2 ＇ and L4 ＇ Given the right generatrix, find line L1. ＇ and line L3 ＇ Intersection point D1 ＇ and line L2 ＇ and line L4 ＇ Intersection point D2 ＇ Through intersection point D1 ＇ Intersection point D2 ＇ The second fossa depth H2 is obtained by finding the center C of the larger circle. The calculation formula is as follows:

[0155] H2=Z C -(D1 z ＇ +D2 z ＇ ) / 2;

[0156] Among them, Z C D1 represents the coordinates of the center C of the great circle along the Z-axis in the point cloud coordinate system. z ＇ Indicates the intersection point D1 ＇ The coordinate value of D2 in the Z-axis direction of the point cloud coordinate system. z ＇ Indicates the intersection point D2 ＇ The coordinate value along the Z-axis in the point cloud coordinate system;

[0157] Compare H2 with H1. When the difference is within the tolerance range of the countersink structure on the aircraft skin surface, further calibration is performed to obtain the pit depth H = (H1 + H2) / 2.

[0158] Simultaneously passing through straight line L3＇ and line L4 ＇ The second minor diameter value S2 is obtained. ＂ S2 ＂ It is line L3 ＇ and line L4 ＇ From intersection point D1 ＇ and intersection point D2 ＇ Starting from the bottom of hole 2, extend the shortest horizontal distance within this section, and then S2 ＂ With S2 ＇ By comparison, when the difference is within the tolerance range, further calibration is performed to obtain the minor diameter S2, calculated using the following formula:

[0159] S2 = (S2) ＇ +S2 ＂ ) / 2;

[0160] S9, calculate the axis L of the conical surface cavity using lines L1 and L2. w The hole axis L is calculated using straight lines L3 and L4. k And according to L w and L k The included angle δ is used to calculate the coaxiality deviation A, L of the hole axis. w L k Refer to the schematic diagram of the positions of δ and A. Figure 6 The specific method is as follows:

[0161] Calculate the axis L of the conical notch using lines L1 and L2. w The vector equation is:

[0162] L w = (L1 + L2) / 2;

[0163] Calculate the hole axis L using lines L3 and L4. k The vector equation is:

[0164] L k = (L3 + L4) / 2;

[0165] The formula for calculating the coaxiality deviation A of the hole axis is:

[0166] A = H × tan (δ);

[0167] In the formula, δ is L w With L k The included angle.

[0168] In this invention, the coaxiality deviation A of the hole axis is determined by L. w and L k The included angle δ forms a horizontal distance with the plane containing the major diameter S1.

[0169] Through the above steps, the profile size of the surface skin counterbore point cloud can be quickly obtained, including the normal vector V of the plane where the major diameter S1 of the conical hole 1 of the counterbore is located, the counterbore depth H, the minor diameter S2 and the hole axis coaxiality deviation A. The traditional detection time can be shortened from several minutes to 1.5 seconds or less, and the detection accuracy and stability are greatly improved.

[0170] It is to be understood that while the embodiments of the application have been illustrated and described, it is to be understood that changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for rapid contour dimension analysis based on point clouds of countersunk holes in curved skin, characterized in that, Specifically, the following steps are included: S1, Based on the installation angle of the laser sensor, voxel filtering is performed on the original scanned point cloud P0 to obtain point cloud P0. ＇ Then, coordinate transformation and compensation are performed in the three directions of the laser sensor coordinate system (X-axis, Y-axis, and Z-axis) to obtain the transformed point cloud P1. S2, segment the point cloud P1, and use the default value to segment into valid data point cloud P2 and invalid data point cloud P3; S3, extract the center c and radius r of the bottom hole by using maximum outer circle fitting for invalid data point cloud P3; S4, perform dimensionality reduction on the effective data point cloud P2 to obtain point cloud P4; S5, point cloud P4 is segmented using point cloud filtering to obtain point cloud P5, and point cloud P5 is subjected to plane fitting analysis using SVD decomposition method to obtain the three-dimensional plane equation of the plane where the large diameter S1 is located, and the normal vector V of the plane where the large diameter S1 of the countersink is located is extracted. S6. Using the center C of the large circle as the dividing center, the point cloud P4 is divided into two symmetrical point clouds by the XZ plane based on the point cloud coordinate system. One of the point clouds is randomly selected as P6. The outer contour of the point cloud P6 in the Y-axis direction is extracted to obtain the point cloud P7. The point cloud P7 is mapped to the YZ plane to obtain the two-dimensional point cloud P8. Simultaneously, using the center C of the large circle as the dividing center, the point cloud P4 is divided into two symmetrical point clouds using the YZ plane based on the point cloud coordinate system. One of these point clouds is then arbitrarily selected as P6. ＇ Extracting point cloud P6 ＇ The point cloud P7 is obtained from the X-direction outer contour. ＇ Point cloud P7 ＇ Mapping to the XZ plane yields a two-dimensional point cloud P8. ＇ ; S7. The least squares method is used to perform linear fitting analysis on the two-dimensional point cloud P8, and four straight lines are obtained, two of which are the generatrices of the conical surface L1 and L2, and two are the generatrices of the cylindrical surface of the bottom hole L3 and L4. Similar to the 2D point cloud P8, for the 2D point cloud P8 ＇ The least squares method was used for line fitting analysis, resulting in four lines, two of which are generatrices L1 of the conical surface. ＇ and L2 ＇ Two are the generatrices L3 of the bottom hole cylindrical surface. ＇ and L4 ＇ ; S8, to obtain the depth H of the notch and the diameter of the bottom hole, i.e., the minor diameter S2; S9, calculate the axis L of the conical surface cavity using the generatrices L1 and L2 of the conical surface. w The hole axis L is calculated using the generatrices L3 and L4 of the bottom hole cylindrical surface. k And according to L w With L k The coaxiality deviation A of the hole axis is obtained by calculating the included angle δ and the depth H of the hole.

2. The method for rapid contour dimension analysis based on point cloud of countersunk holes in curved skin as described in claim 1, characterized in that, Step S4 uses the set of points on a straight line parallel to the Z-axis passing through the center C of the large circle as key points. A point cloud dimensionality reduction method based on gradient descent with key points as the center is used to reduce the dimensionality of the effective data point cloud P2 to obtain point cloud P4.

3. The method for rapid contour dimension analysis based on point cloud of countersunk holes in curved skin as described in claim 1, characterized in that, The point cloud dimensionality reduction method in step S4 specifically includes the following steps: S401. Define the objective function: a) Assume that the valid data point cloud P2 has n 3D point cloud data points ,as well as m Key points ; The point after dimensional reduction At the key point Gathering around; Represents three-dimensional Euclidean space; b) Use the Gaussian kernel function to measure the 3D point cloud data points. With key points Similarity, Gaussian kernel function for: ； Where exp(·) is the natural exponential function, and σ is a hyperparameter that controls the width of the Gaussian kernel function; the Gaussian kernel function The larger the value, the better. and The more similar they are, the closer they are to each other; Define the objective function for: ； objective function The significance is: for each 3D point cloud data point Through Gaussian kernel function Calculate its relationship with all key points The similarity is used as a weight to measure the similarity of the points in the 3D point cloud data after dimensionality reduction. Dimensionally reduced points of each key point Square distance between ; S402, Calculate the gradient: To minimize the objective function using gradient descent. , needs to be calculated about gradient : c) Assume weights Gaussian kernel function As a weight, then: ； For vectors , w Let be the dimension of the vector, and calculate respectively. about of l Quantity partial derivatives ; Using the following formula Seeking information about The partial derivatives of, where, express The i Pointed k Quantity, express The j Pointed k Quantity: ； so: ； Then, gradient for: ； in, for about The gradient, i.e., only focusing on When changing The rate of change; Represents the variable Find the gradient; and For vectors, ,represent Pointed l Quantity , ,represent Pointed l Quantity ; w Let be the dimension of the vector; S403, Gradient Descent Update: e) Gradient descent minimizes the objective function by continuously updating the parameters along the negative direction of the gradient; f) In each iteration number t In the middle, the dimensionality-reduced point vectors are updated based on the gradient. The learning rate α controls the step size for each update, and the update formula is: ； Specifically, for The updated for: ； Repeat this update process until the objective function converges.

4. The method for rapid contour dimension analysis based on point cloud of countersunk holes in curved skin as described in claim 1, characterized in that, In step S5, when dividing point cloud P4 to obtain point cloud P5, the radius R and center C of the large circle at the top of the conical hole are detected along the point cloud scanning direction with the center c of the bottom hole as the center. With the center C of the large circle as the center, the upper surface thickness Z of the effective data point cloud P2 as the radius and the upper surface thickness Z of the effective data point cloud P2 as the upper surface thickness difference, the point cloud P4 is divided by the point cloud filtering method to obtain point cloud P5.

5. The method for rapid contour dimension analysis based on point cloud of countersunk holes in curved skin as described in claim 1, characterized in that, In step S6, the method for obtaining the two-dimensional point cloud P8 is as follows: S601, Assume the point cloud in point cloud P4 is... in It is the first point cloud in P4. k One point; S602, with the center of the circle As the dividing center, These are the coordinates of the center C of the great circle in the X, Y, and Z axes of the point cloud coordinate system, divided by the XZ plane: For the XZ plane, its equation is: y =0; a plane passing through the center C of the great circle and parallel to the XZ plane. y = y c Perform segmentation; after segmentation, two symmetrical point clouds are obtained. Arbitrarily select one of these point clouds as P6. Let's assume we take... y≥y c This part is denoted as point cloud. ; S603, Extracting Point Clouds P The outer contour of 1 along the Y-axis yields a new point cloud P7: set up P 2 represents the new point cloud P7. For each 3D point in point cloud P7, its spatial coordinates can be represented as ( x , y , z ), x , y , z These correspond to the coordinate values ​​of three orthogonal directions in three-dimensional space, respectively. in,( x , z ) represents the coordinates of that point x Quantity and z Components, i.e., the position parameters of a point in the X-axis and Z-axis directions in the point cloud coordinate system; for each set ( x , z ) value, in P 1. Find the corresponding maximum y The point of value; that is, for a given x and z If there are multiple points satisfy Then select values ​​from these points. y Add the largest point P 2; S604, will P 2. Mapping to the YZ plane yields a two-dimensional point cloud P8: Because on the YZ plane, the X-axis coordinate x =0; Therefore for P Each point in 2 Map it to ; Let the mapped 2D point cloud P8 be P 3 items: 。 6. The method for rapid contour dimension analysis based on point cloud of countersunk holes in curved skin as described in claim 1, characterized in that, Step S8 specifically includes the following steps: Assuming L1 and L3 are the left generatrixes and L2 and L4 are the right generatrixes, find the intersection point D1 of lines L1 and L3, and the intersection point D2 of lines L2 and L4; obtain the first dent depth H1 using points D1, D2, and the center C of the great circle, calculated using the following formula: H1=Z C -(D1) z +D2 z (2) Among them, Z C D1 represents the coordinates of the center C of the great circle along the Z-axis in the point cloud coordinate system. z D1 represents the coordinate value of the intersection point D1 in the Z-axis direction of the point cloud coordinate system, and D2 represents the intersection point D2. z This represents the coordinate value of intersection point D2 in the Z-axis direction of the point cloud coordinate system; The first minor diameter value S2 is obtained by using lines L3 and L4. ＇ S2 ＇ It is the shortest horizontal distance between straight lines L3 and L4, which extend downwards from the intersection points D1 and D2 to the bottom of the hole; Assume L1 ＇ and L3 ＇ For the left busbar, L2 ＇ and L4 ＇ Given the right generatrix, find line L1. ＇ and line L3 ＇ Intersection point D1 ＇ and line L2 ＇ and line L4 ＇ Intersection point D2 ＇ Through point D1 ＇ Point D2 ＇ The second fossa depth H2 is obtained by finding the center C of the larger circle. The calculation formula is as follows: H2=Z C -(D1) z ＇ +D2 z ＇ (2) Among them, Z C D1 represents the coordinates of the center C of the great circle along the Z-axis in the point cloud coordinate system. z ＇ Indicates the intersection point D1 ＇ The coordinate value of D2 in the Z-axis direction of the point cloud coordinate system. z ＇ Indicates the intersection point D2 ＇ The coordinate value along the Z-axis in the point cloud coordinate system; Compare H2 with H1. When the difference is within the tolerance range of the countersink structure on the aircraft skin surface, further calibration is performed to obtain the pit depth H = (H1 + H2) / 2. Simultaneously passing through straight line L3 ＇ and line L4 ＇ The second minor diameter value S2 is obtained. ＂ , will S2 ＂ With S2 ＇ By comparison, when the difference is within the tolerance range, further calibration is performed to obtain the minor diameter S2, calculated using the following formula: S2 = (S2 ＇ +S2 ＂ ) / 2.

7. The method for rapid contour dimension analysis based on point cloud of countersunk holes in curved skin as described in claim 1, characterized in that, In step S9, the axis L of the conical cavity is calculated using lines L1 and L2. w The vector equation is: L w =(L1+L2) / 2; Calculate the hole axis L using lines L3 and L4. k The vector equation is: L k =(L3+L4) / 2.

8. The method for rapid contour dimension analysis based on point cloud of countersunk holes in curved skin as described in claim 1, characterized in that, In step S9, the formula for calculating the coaxiality deviation A of the hole axis is: A = H × tan (δ); In the formula, δ is L w With L k The included angle.

Images