Determination Method for Repair Allowance of Aircraft Door and Cover Skin

Abstract

The present invention discloses a method for determining the repair allowance of the skin of aircraft access panels, which includes the following steps: obtaining the point cloud data set Q of the skin manufactured by the factory and the point cloud data set P of the ideal skin; denoising the point cloud data set Q, and uniformly sampling the point cloud data sets Q and P to obtain the source point cloud Q1 and the target point cloud P1; determining the key points in the source point cloud Q1 and the target point cloud P1 and calculating the FPFH histogram feature descriptors of all key points; performing rough registration on the source point cloud Q1 and the target point cloud P1 to obtain the point cloud set Q ’ ; performing fine registration on the point cloud set Q ’ and the target point cloud P1 to obtain a new transformed point cloud set Q''; extracting the boundary lines of the point cloud set Q'' and P1 and calculating the Euclidean distance between the two boundary lines, which is the skin repair amount. The method for determining the repair allowance of the skin of aircraft access panels can be completed by a computer, saving manpower, improving production efficiency, and having a more accurate registration result.

Description

Technical Field

[0001] The present invention relates to the field of aircraft skins, and particularly provides a method for determining the trimming allowance of aircraft hatch skins. Background Art

[0002] The aircraft skin is a key load-bearing component that affects the aerodynamic shape of the aircraft and has a crucial impact on the stability of the aircraft itself. Therefore, the manufacture of the aircraft skin not only has requirements for the shape accuracy and mechanical performance indicators, but also has strict requirements for its installation accuracy. However, the skin belongs to a thin-walled part and is prone to deformation during the processing. Processing according to the theoretical shape will make it difficult for the skin to be installed at the docking connection with the ideal clearance. Therefore, a certain allowance is usually left during the manufacture of the skin, and the allowance is trimmed according to the actual installation size later to achieve the purpose of accurate installation.

[0003] Currently, most domestic aircraft main manufacturers use manual alignment and marking when assembling the skin, and then use manual methods for trimming and cutting to remove the reserved allowance of the skin. However, manually aligning, cutting, and trimming the skin requires a large amount of manpower, high labor costs, and low production efficiency. The processing quality of the skin depends very much on the experience and proficiency of the workers. The errors of the skins processed by different workers may be inconsistent, which will make it difficult to ensure the processing consistency and qualification rate of the skin. When manually marking, a marker pen with a width of 1 mm is generally used, and random errors are easily generated during marking. Manual cutting will also result in low registration accuracy and low processing efficiency of the skin, and it is difficult to ensure the quality of the skin boundary and the contour.

[0004] Therefore, providing a method for determining the trimming allowance of aircraft hatch skins to replace manual alignment and efficiently and accurately obtain the trimming allowance of the skin has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method for determining the trimming allowance of aircraft hatch skins to solve the problems such as low efficiency and low registration accuracy in manual skin registration.

[0006] The technical solution provided by the present invention is: A method for determining the trimming allowance of aircraft hatch skins, comprising the following steps:

[0007] Step 1: Scan the aircraft skin manufactured in the factory by a three-dimensional laser scanner to obtain the point cloud data set Q of the skin manufactured in the factory;

[0008] Step 2: Convert the ideal three-dimensional model of the aircraft skin into point cloud data according to the data density collected by the three-dimensional laser scanner for the aircraft skin manufactured in the factory, and obtain the point cloud data set P of the ideal skin;

[0009] Step 3: Denoise the point cloud dataset Q of the skin manufactured by the factory, and remove the noise points deviating from the outer contour. After that, uniformly sample the point cloud dataset Q of the skin manufactured by the factory and the point cloud dataset P of the ideal skin, and correspondingly obtain the source point cloud Q1 and the target point cloud P1;

[0010] Step 4: Calculate the sum of the normal vectors of each point in the source point cloud Q1 and the target point cloud P1 and the average value of the included angles under different neighborhood radii, and determine the key points in the source point cloud Q1 and the target point cloud P1 according to the sum of the average values of the included angles;

[0011] Step 6: Calculate the FPFH histogram feature descriptors of all key points in the source point cloud Q1 and the target point cloud P1;

[0012] Step 6: Find the points in the source point cloud Q1 with similar FPFH features corresponding to the sampled points in the target point cloud P1 through the K-D tree nearest neighbor search algorithm, and use the random sample consensus algorithm to perform rough registration on the source point cloud Q1 and the target point cloud P1 to obtain the point cloud set Q' corresponding to the target point cloud P1;

[0013] Step 7: Use the improved iterative closest point algorithm for point clouds to perform fine registration on the point cloud set Q' and the target point cloud P1 to obtain a new transformed point cloud set Q";

[0014] Step 8: Adopt the normal estimation extraction boundary algorithm to extract the boundary lines of the new transformed point cloud set Q" and the target point cloud P1 respectively, and calculate the Euclidean distance between the two boundary lines, which is the skin repair amount.

[0015] Preferably, in Step 4, the method for determining the normal vector of each point in the source point cloud Q1 or the target point cloud P1 is as follows:

[0016] According to the nearest neighbor point search algorithm, find the K nearest neighbor points of any point M in the source point cloud Q1 or the target point cloud P1, and use the K nearest neighbor points to fit the local plane in the sense of the least squares method. This local plane is the tangent plane of the point M, and the normal line of the tangent plane is the line where the normal vector of the point M is located. Specifically: use the principal component analysis method to construct a covariance matrix for the point M and the K nearest neighbor points and perform singular value decomposition, and find its minimum eigenvalue. The eigenvector corresponding to the minimum eigenvalue is the line where the normal vector of the point M is located;

[0017] What is calculated by the principal component analysis method is the line where the normal vector is located. The final direction of the normal vector needs to be further confirmed. Therefore, first set a direction for the normal vector of the point M as Traverse all points. If the direction of the normal vector of the next point to be traversed is And Then Perform flipping. If Then keep unchanged, and thus complete the orientation of the normal vectors of all points in the point cloud set.

[0018] Further preferably, in step 4, the sum of the mean values of the angles between the normal vectors of the point clouds with two different neighborhood radii is obtained by the following formula:

[0019]

[0020] where K1 and K2 are the numbers of points in different neighborhoods, and n i2 , n i3 are the normal vectors with different neighborhood radii.

[0021] Further preferably, in step 4, if the sum of the angles θ between the normal vectors of the point clouds with two different neighborhood radii satisfies ε < θ < 2ε and simultaneously satisfies θ > 45°, then the point M is determined as a key point, where the threshold ε = 1.

[0022] Further preferably, in step 5, according to the normal vectors of the key points in the source point cloud Q1 and the surface curvature within their K neighborhoods, a feature histogram is established to obtain the FPFH histogram feature descriptors of all key points in the source point cloud Q1;

[0023] According to the normal vectors of the key points in the target point cloud P1 and the surface curvature within their K neighborhoods, a feature histogram is established to obtain the FPFH histogram feature descriptors of all key points in the target point cloud P1.

[0024] Further preferably, in step 6, the method for obtaining the point cloud set Q' is as follows:

[0025] Input the key points and the local feature descriptors of the key points obtained in steps 4 and 5, and then randomly sample s points in the target point cloud P1 according to the minimum distance being greater than d min . After that, for each sampled point, a set of points with similar FPFH features is found in the source point cloud Q1 through the K-dtree nearest neighbor search algorithm;

[0026] Next, a rigid transformation matrix is determined for each pair of corresponding points, and the Huber formula is used to perform the registration performance judgment of the transformation parameter "distance error sum ", that is where t1 is the set threshold, and l i is the distance difference of the i-th pair of corresponding points after transformation. The optimal rigid transformation matrix is selected to minimize the point cloud registration distance error, and the random sample consensus algorithm is used to perform rough registration on the point cloud to obtain the transformed point cloud set Q'.

[0027] Further preferably, step 7 specifically includes the following steps: randomly select key points in the target point cloud P1 and find the corresponding points with the closest distance in the point cloud set Q', obtain the corresponding point pairs, and calculate the rotation matrix R and the translation matrix T to minimize the mean square error d k between the above corresponding point pairs. Transform the point cloud set Q' according to the obtained translation and rotation matrices to obtain a new transformed point cloud set Q", and calculate the distance error between Q" and P2. If the error between two iterations is less than the set threshold or the maximum number of iterations is greater than the set number of iterations, the iteration ends. Repeat the iteration process, traverse all points that meet the convergence conditions, and complete the fine registration of the point cloud data using the improved iterative closest point algorithm for point clouds.

[0028] The method for determining the repair allowance of the aircraft flap skin provided by the present invention can obtain the repair allowance of the skin by processing the collected point cloud data of the processed aircraft skin and the point cloud data of the ideal aircraft model. This method can greatly save manpower, reduce labor costs, and improve production efficiency. At the same time, due to the absence of human factors, the registration result is more accurate. Workers can repair the processed aircraft skin according to the size of the skin repair allowance, or output the processing program according to the size of the repair allowance to complete the numerical control milling of the aircraft skin, saving the time for aircraft assembly. Specific embodiments

[0029] The present invention will be further explained below in conjunction with specific implementation cases, but the present invention is not limited thereto.

[0030] The present invention provides a method for determining the repair allowance of the aircraft flap skin, including the following steps:

[0031] Step 1: Scan the aircraft skin manufactured in the factory by a three-dimensional laser scanner to obtain the point cloud data set Q of the skin manufactured in the factory;

[0032] Step 2: Convert the three-dimensional model of the ideal aircraft skin into point cloud data according to the data density of the aircraft skin manufactured in the factory collected by the three-dimensional laser scanner to obtain the point cloud data set P of the ideal skin;

[0033] Step 3: Denoise the point cloud data set Q of the skin manufactured in the factory, remove the noise points deviating from the outer contour, and then uniformly sample the point cloud data set Q of the skin manufactured in the factory and the point cloud data set P of the ideal skin to reduce the number of points in the point cloud data set, and correspondingly obtain the source point cloud Q1 and the target point cloud P1;

[0034] Step 4: Calculate the normal vectors of each point in the source point cloud Q1 and the target point cloud P1 and the sum of the mean angles at different neighborhood radii, and determine the key points in the source point cloud Q1 and the target point cloud P1 according to the sum of the mean angles;

[0035] Among them, the method for determining the normal vectors of each point in the source point cloud Q1 or the target point cloud P1 is as follows:

[0036] According to the nearest neighbor search algorithm, find the K nearest neighbors of any point M in the source point cloud Q1 or the target point cloud P1, and use the K nearest neighbors to fit a local plane in the sense of the least squares method. This local plane is the tangent plane of the point M, and the normal line of the tangent plane is the line where the normal vector of the point M is located. Specifically: use the principal component analysis method to construct a covariance matrix for the point M and the K nearest neighbors and perform singular value decomposition, find its minimum eigenvalue, and the eigenvector corresponding to the minimum eigenvalue is the line where the normal vector of the point M is located;

[0037] What is calculated by the principal component analysis method is the line where the normal vector is located. The final direction of the normal vector needs to be further confirmed. Therefore, first set a direction for the normal vector of the point M as Traverse all points. If the direction of the normal vector of the next point to be traversed is and then is flipped. If then keep unchanged. Thus, the orientation of the normal vectors of all points in the point cloud set is completed;

[0038] Use the following formula to calculate the sum of the mean values of the included angles of the normal vectors of the point clouds with two different neighborhood radii:

[0039]

[0040] where, K1 and K2 are the numbers of points in different neighborhoods, n i2 , n i3 are the normal vectors with different neighborhood radii;

[0041] If the sum of the included angles θ of the normal vectors of the point clouds with two different neighborhood radii satisfies ε < θ < 2ε and at the same time satisfies θ > 45°, then determine the point M as a key point, where the threshold ε = 1;

[0042] Step 5: Calculate the FPFH histogram feature descriptors of all key points in the source point cloud Q1 and the target point cloud P1;

[0043] Specifically: establish a feature histogram according to the normal vector of the point cloud of the key points in the source point cloud Q1 and the surface curvature within its K neighborhood to obtain the FPFH histogram feature descriptors of all key points in the source point cloud Q1;

[0044] Establish a feature histogram according to the normal vector of the point cloud of the key points in the target point cloud P1 and the surface curvature within its K neighborhood to obtain the FPFH histogram feature descriptors of all key points in the target point cloud P1;

[0045] Step 6: In the source point cloud Q1, find the points with similar FPFH features corresponding to the sampled points in the target point cloud P1 through the K-D tree nearest neighbor search algorithm, and use the random sample consensus algorithm to perform rough registration on the source point cloud Q1 and the target point cloud P1 to obtain the point cloud set Q' corresponding to the target point cloud P1;

[0046] Specifically: Input the key points and the local feature descriptors of the key points obtained in Steps 4 and 5, and then according to the minimum distance being greater than d min Randomly sample s points in the target point cloud P1. After that, for each sampled point, find a set of points with similar FPFH features in the source point cloud Q1 through the K-dtree nearest neighbor search algorithm. Preferably, d min = 0.005;

[0047] Next, determine a rigid transformation matrix for each set of corresponding point pairs, and use the Huber formula to perform "distance error sum " registration performance judgment, that is where t1 is the set threshold, l i is the distance difference of the i-th set of corresponding points after transformation. Select the optimal rigid transformation matrix to minimize the point cloud registration distance error, complete the rough registration of the point cloud by the random sample consensus algorithm, and obtain the transformed point cloud set Q';

[0048] Step 7: Use the improved iterative closest point algorithm for point cloud (i.e., GICP algorithm) to perform fine registration on the point cloud set Q' and the target point cloud P1 to obtain a new transformed point cloud set Q"; compared with the traditional ICP algorithm, the GICP algorithm has higher robustness;

[0049] Specifically: Arbitrarily select key points in the target point cloud P1 and find the closest corresponding points in the point cloud set Q' to obtain corresponding point pairs, and obtain the rotation matrix R and the translation matrix T to minimize the mean square error d k between the above corresponding point pairs. Transform the point cloud set Q' according to the obtained translation and rotation matrices to obtain a new transformed point cloud set Q", and calculate the distance error between Q" and P2. If the error between two iterations is less than the set threshold or the maximum number of iterations is greater than the set number of iterations, the iteration ends. Repeat the iteration process, traverse all points that meet the convergence conditions, and complete the fine registration of the point cloud data using the improved iterative closest point algorithm;

[0050] Step 8: Use the normal estimation extraction boundary algorithm to extract the boundary lines of the new transformed point cloud set Q" and the target point cloud P1 respectively, and calculate the Euclidean distance between the two boundary lines, which is the skin repair amount.

[0051] The method for determining the repair allowance of the aircraft skin of the access panel type can obtain the repair allowance of the skin through data processing of the collected point cloud data of the processed aircraft skin and the point cloud data of the ideal aircraft model. This method can greatly save manpower, reduce labor costs, and improve production efficiency. At the same time, due to the absence of the influence of human factors, the registration result is more accurate. Workers can repair the processed aircraft skin according to the size of the skin repair allowance, or output the processing program according to the size of the repair allowance to complete the numerical control milling of the aircraft skin, saving the time for aircraft assembly.

[0052] The above has made a detailed description of the embodiments of the present invention. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the gist of the present invention within the scope of knowledge possessed by those of ordinary skill in the art.

Claims

1. A method for determining the repair allowance of the skin of aircraft access panels, characterized in that, It includes the following steps: Step 1: Scan the aircraft skin manufactured in the factory using a 3D laser scanner to obtain the point cloud data set Q of the skin manufactured in the factory; Step 2: According to the data density of the aircraft skin manufactured in the factory collected by the 3D laser scanner, convert the ideal 3D model of the aircraft skin into point cloud data to obtain the point cloud data set P of the ideal skin; Step 3: Denoise the point cloud data set Q of the skin manufactured in the factory, and remove the noise points deviating from the outer contour. Then, uniformly sample the point cloud data set Q of the skin manufactured in the factory and the point cloud data set P of the ideal skin to correspondingly obtain the source point cloud Q1 and the target point cloud P1; Step 4: Calculate the normal vectors of each point in the source point cloud Q1 and the target point cloud P1 and the sum of the mean angles at different neighborhood radii, and determine the key points in the source point cloud Q1 and the target point cloud P1 according to the sum of the mean angles; Step 5: Calculate the FPFH histogram feature descriptors of all key points in the source point cloud Q1 and the target point cloud P1; Step 6: Find points in the source point cloud Q1 that are similar to the FPFH features corresponding to the sampled points in the target point cloud P1 through the K-D tree nearest neighbor search algorithm, and use the random sample consensus algorithm to perform rough registration on the source point cloud Q1 and the target point cloud P1 to obtain the point cloud set Q corresponding to the target point cloud P1 ’ ; Step 7: Use the improved point cloud iterative closest point algorithm to perform fine registration on the point cloud set Q ’ and the target point cloud P1 to obtain a new transformed point cloud set Q ” ; Step 8: Use the boundary extraction algorithm based on normal estimation to extract the new transformed point cloud set Q ” and the boundary lines of the target point cloud P1, and calculate the Euclidean distance between the two boundary lines, which is the skin fitting amount; Among them, step 7 specifically includes the following steps: Arbitrarily select key points in the target point cloud P1 and find the corresponding points with the shortest distance in the point cloud set Q ’ to obtain corresponding point pairs, and calculate the rotation matrix R and the translation matrix T to minimize the mean square error d k between the above corresponding point pairs. Transform the point cloud set Q ’ according to the obtained translation and rotation matrices to obtain a new transformed point cloud set Q ” , and calculate the distance error between Q ” and P2. If the error between two iterations is less than the set threshold or the maximum number of iterations is greater than the set number of iterations, the iteration ends. Repeat the iteration process, traverse all points that meet the convergence conditions, and complete the fine registration of the point cloud data using the improved iterative closest point algorithm for point clouds.

2. The method for determining the repair allowance of the skin of the aircraft hatch cover according to claim 1, characterized in that: In Step 4, the method for determining the normal vector of each point in the source point cloud Q1 or the target point cloud P1 is as follows: According to the nearest neighbor search algorithm, find the K nearest neighbor points of any point M in the source point cloud Q1 or the target point cloud P1, and use the K nearest neighbor points to fit a local plane in the sense of the least squares method. This local plane is the tangent plane of the point M, and the normal line of the tangent plane is the line where the normal vector of the point M is located. Specifically: use the principal component analysis method to construct a covariance matrix for the point M and the K nearest neighbor points and perform singular value decomposition, and find its minimum eigenvalue. The eigenvector corresponding to the minimum eigenvalue is the line where the normal vector of the point M is located; What is calculated by the principal component analysis method is the straight line where the normal vector lies. The final direction of the normal vector needs to be further confirmed. Therefore, first set a direction for the normal vector of point M as Traverse all points. If the direction of the normal vector of the next point to be traversed is and then is flipped. If then keep unchanged. Thus, the orientation of the normal vectors of all points in the point cloud is completed.

3. The method for determining the repair allowance of the aircraft hatch skin according to claim 1, wherein: In Step 4, use the following formula to calculate the sum of the mean angles of the normal vectors of the point clouds with two different neighborhood radii: where K1 and K2 are the numbers of points in different neighborhoods, and n i2 , n i3 are the normal vectors of different neighborhood radii.

4. The method for determining the repair allowance of the aircraft hatch skin according to claim 2, wherein: In Step 4, if the sum of the angles θ of the normal vectors of the point clouds with two different neighborhood radii satisfies ε < θ < 2ε and at the same time satisfies θ > 45°, then determine the point M as a key point, where the threshold ε = 1; 5. The method for determining the repair allowance of the skin of the aircraft hatch cover according to claim 1, wherein: In Step 5, establish a feature histogram according to the normal vector of the key point cloud in the source point cloud Q1 and the surface curvature within its K neighborhood to obtain the FPFH histogram feature descriptors of all key points in the source point cloud Q1; Establish a feature histogram according to the normal vector of the key point cloud in the target point cloud P1 and the surface curvature within its K neighborhood to obtain the FPFH histogram feature descriptors of all key points in the target point cloud P1.

6. The method for determining the repair allowance of the aircraft hatch skin according to claim 1, characterized in that: In step 6, the point cloud set Q ’ is obtained as follows: Input the key points and the local feature descriptors of the key points obtained in Steps 4 and 5, and then according to the minimum distance being greater than d min Randomly sample s points in the target point cloud P1. After that, for each sampled point, find a set of points with similar FPFH features in the source point cloud Q1 through the K-dtree nearest neighbor search algorithm; Next, determine a rigid transformation matrix for each set of corresponding point pairs, and use the Huber formula to perform the "distance error sum" of the transformation parameters for registration performance judgment, that is ", where t1 is the set threshold, and l is the distance difference of the i-th set of corresponding points after transformation. Select the optimal rigid transformation matrix to minimize the point cloud registration distance error, complete the rough registration of the point cloud by the random sample consensus algorithm, and obtain the transformed point cloud set Q i . ’ .

Images