A method for predicting deformation of wing docking area driven by measured data

Abstract

The present invention provides a method for predicting deformation of a wing docking area driven by measured data, wherein the docking area shape is scanned by a measuring device, point cloud data is constructed and the point cloud is processed to prepare for model reconstruction; the point cloud obtained by actual measurement is reconstructed into a three-dimensional model, and a digital model reflecting the size error of the real parts is established; for the fuselage docking process, the non-docking area can be regarded as a load module, which only provides support and load, so only the docking area needs to be reconstructed, and the non-docking area uses a theoretical digital model; the docking process is simulated using simulation software to solve the docking gap; batch generation of data is achieved by batch processing of point clouds and finite element simulation, and an agent model is constructed based on a long short-term memory network. The present invention relates to the field of intelligent manufacturing, and is helpful to improve the accuracy, reliability and efficiency of the docking process.

Description

Technical Field

[0001] The present invention relates to the aviation equipment industry and intelligent manufacturing fields, and in particular to a wing docking zone deformation prediction method driven by measured data. Background Art

[0002] In modern aircraft assembly work, real-time monitoring and data transmission have become relatively mature, and a lot of measured data can be used to monitor the current state of aircraft components or to indirectly evaluate and predict the possible completion quality of the next step. For aircraft flexible parts, it is very necessary to predict in advance the assembly gap between them and the structure to be docked due to deformation. After clarifying the possible gap, the assembly quality can be evaluated based on the gap level, and process optimization can be performed after obtaining enough historical data. Furthermore, if it can be proved that the simulation results can well simulate the real physical world, a large amount of data can be generated through the proxy model to assist in optimization.

[0003] At present, the wing-fuselage docking gap of aircraft is limited by the poor spatial openness, and the traditional measurement and observation methods are mostly used in actual operation. In view of these defects, the present invention proposes a simulation model based on measured data, and constructs a proxy model of the wing docking area assembly gap based on it. Summary of the invention

[0004] In view of the above problems in the prior art, the present application proposes a method for predicting deformation of the wing docking area driven by measured data, comprising the following steps:

[0005] S1. Scan the shape of the docking area through the measuring equipment, construct point cloud data and process the point cloud to prepare for model reconstruction;

[0006] S2, by reconstructing the point cloud obtained by actual measurement into a three-dimensional model, a digital model reflecting the actual part size error is established;

[0007] S3. For the fuselage docking process, the non-docking area can be regarded as a load module, which only provides support and load. Therefore, only the docking area needs to be reconstructed, and the non-docking area uses the theoretical numerical model;

[0008] S4, using simulation software to simulate the docking process and solve the docking gap;

[0009] S5. Batch generation of data is achieved through batch processing of point clouds and finite element simulation, and an agent model is constructed based on long short-term memory network.

[0010] Preferably, the scanning work is completed using an optical scanner; the data processing work includes the registration, denoising and downsampling process of the point cloud data, which is completed through reverse engineering software, Python language and open source toolkit.

[0011] Preferably, the point cloud reconstruction method first implements filtering based on a statistical outlier removal algorithm, and then implements point cloud registration and matching based on a new algorithm to obtain a reconstructed model.

[0012] Preferably, during the fuselage docking process, the digital model of the docking area is spliced ​​with the theoretical digital model of the non-docking area in the three-dimensional modeling software to form a new digital model.

[0013] Preferably, the digital model obtained in claim 4 is imported into simulation software to simulate the docking process, and a finite element simulation is performed on the docking gap to simulate the docking gap in this case.

[0014] Preferably, the point cloud data are processed in batches to simulate the initial errors within various tolerance ranges; the models are reconstructed in batches, and finite element parameter modeling is performed on each model in batches to simulate the docking gap under different working conditions and generate a large amount of data; a long-short memory neural network hybrid model based on genetic algorithm optimization is established, and the hyperparameters of the long-short memory neural network hybrid model are optimized by genetic algorithm.

[0015] Preferably, the specific process of docking area shape scanning and data processing mainly includes the following steps:

[0016] Step 101, building a measurement field consisting of an optical dynamic tracking system and a three-dimensional scanner;

[0017] Step 102, scanning the shape of the parts in the docking area;

[0018] Step 103, using a statistical outlier removal algorithm, performing k-neighborhood statistical analysis on each scan point, calculating the average distance, and removing points whose average distance exceeds a given threshold, where the k value represents the average distance between each scan point and its k neighboring points;

[0019] Step 104, searching for neighboring points of a point by using the KD-tree algorithm, calculating the distance between the points, and setting a reasonable threshold to remove neighboring points whose distance is greater than the threshold, iterating the process to remove outliers;

[0020] Step 105, using an equal spacing thinning algorithm to perform overall preliminary simplification on the point cloud data, setting the sampling distance to n in the original point cloud data, and retaining the next point every n points until all points are screened;

[0021] Step 106, further simplify the point cloud using the uniform sampling method, divide the point cloud space into grids with a sphere of radius r, select the point closest to the center of the sphere as the sampling point, and control the sparsity of the sampled point cloud by modifying the radius of the sphere.

[0022] Preferably, the specific process and implementation of the point cloud data-driven docking area reconstruction are as follows:

[0023] Step 201, import the theoretical numerical model into the finite element software to perform meshing operation; the scanning point cloud distribution on the surface of the structural part is uniform, and considering the matching of the mesh nodes and the scanning point cloud, a tetrahedral mesh is selected;

[0024] Step 202, the coordinate data in the point cloud is converted by a matching algorithm. The specific implementation process of the matching algorithm is as follows: read the coordinates of the current grid node; traverse and read the scanning point coordinates in the block file, and calculate the distance between the scanning point coordinates and the grid node coordinates; according to a preset distance threshold, select multiple scanning points with a smaller distance from the scanning points whose distance to the grid node is less than the threshold; calculate the coordinates of the matching point according to the distance value and the coordinates of the scanning point;

[0025] Step 203: Obtain a reconstructed model file based on the point cloud data.

[0026] Preferably, the specific process of splicing the docking area and the non-docking area mainly includes the following steps:

[0027] Step 301, converting the reconstructed point cloud model into an STP format file;

[0028] Step 302, align the reconstructed STP format model with the theoretical model coordinate system, and remove the docking area parts in the theoretical digital model that overlap with the point cloud reconstructed model;

[0029] Step 303, through CATIA assembly quality, the point cloud reconstructed digital model is spliced ​​with the non-docking area theoretical digital model by surface coupling.

[0030] Preferably, the specific process of the finite element simulation of the butt joint gap mainly includes the following steps:

[0031] Step 401: The docking area uses aviation aluminum alloy Al2024-T3 with a density of 2.78×10-6kg / mm 3 ; The Young's modulus and Poisson's ratio of the material are set to 724000MPa and 0.33 respectively;

[0032] Step 402, directly import the stp format model from the 3D modeling software, use hexahedral mesh, the unit type is C3D8R, the mesh size of the upper and lower skins and beams is 15 mm, the mesh size of the ribs is 30 mm, and the mesh size of the remaining structures is 20 mm;

[0033] Step 403, using the power / display analysis step;

[0034] Step 404, the contact between different parts is defined using the contact properties provided by the finite element software, the tangential friction behavior is defined by the penalty function, and the friction coefficient is 0.2; the normal behavior uses "hard" contact; the mutual assembly relationship between the parts is realized by the binding function;

[0035] Step 405, apply gravity load globally, and select gravity acceleration of 9800m / s; apply full constraint at the POGO column support position at the lower end of the structure; apply concentrated load at the node corresponding to the connection between the docking area and the non-docking area;

[0036] Step 406, in the actual docking state, after the docking is completed, fix the simulated part of the fuselage docking area to the position of the actual fuselage docking area, and move the wing to the position of the simulated part; arrange multiple monitoring points at the upper edge and lower edge of the wing docking area respectively, and compare the displacement data of the upper edge and lower edge paths of the finite element simulation.

[0037] The above technical features can be combined in various suitable ways or replaced by equivalent technical features as long as the purpose of the present invention can be achieved.

[0038] The present invention provides a method for predicting deformation of a wing docking zone driven by measured data, which has at least the following beneficial effects compared with the prior art:

[0039] First, by accurately scanning and processing the shape of the docking area of ​​the measuring equipment, the accurate construction and processing of the point cloud data of the docking area was achieved. This provides a solid foundation for the subsequent model reconstruction and ensures the accuracy and reliability of the reconstructed model.

[0040] Secondly, the point cloud data-driven docking area reconstruction technology can effectively convert the point cloud data obtained from actual measurements into a 3D model. This conversion not only retains the actual dimensional error of the parts, but also displays it intuitively in the form of digital models, providing strong support for subsequent analysis and optimization.

[0041] In the process of splicing the docking area and the non-docking area, the technology cleverly uses theoretical numerical models to represent the non-docking area, which not only simplifies the operation process but also reduces the time cost. At the same time, this processing method does not affect the overall accuracy, ensuring the accuracy and reliability of the entire docking process.

[0042] In addition, the application of docking gap finite element simulation technology can accurately simulate the docking process and solve the docking gap. This helps to predict and evaluate the docking effect before actual operation, thereby discovering and solving potential problems in advance and improving the docking success rate.

[0043] Finally, by batch processing point cloud data and finite element simulation to achieve batch data generation, and building an agent model based on long short-term memory network, this technology can generate a large amount of data to facilitate process optimization. This not only improves the efficiency of data processing and simulation, but also provides rich data support for subsequent process optimization, which helps to achieve a more refined and efficient process optimization process.

[0044] In summary, this technical process has demonstrated significant technical effects in docking area processing and simulation, which helps to improve the accuracy, reliability and efficiency of the docking process, and provides strong support for technological progress and application promotion in related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The present invention will be described in more detail below based on embodiments and with reference to the accompanying drawings, wherein:

[0046] Figure 1 The specific process of scanning and data processing the docking area shape;

[0047] Figure 2 is a schematic diagram of the measurement field;

[0048] Figure 3 Scan the part shape diagram with a handheld 3D scanner;

[0049] Figure 4 This is the flow chart of the point cloud matching algorithm;

[0050] Figure 5 The specific process of splicing the docking area and the non-docking area;

[0051] Figure 6 The specific process of finite element simulation for butt joint gap;

[0052] Figure 7 The specific process for data generation and proxy model construction. DETAILED DESCRIPTION

[0053] The present invention will be further described below in conjunction with the accompanying drawings.

[0054] The present invention provides a method for predicting deformation of a wing docking area driven by measured data. By establishing a simulation model driven by measured data, and building an agent model based on the simulation model, a large amount of data is generated, and the following contents are mainly achieved:

[0055] (1) Docking area shape scanning and data processing: Scan the docking area shape through measuring equipment, construct point cloud data and process the point cloud to prepare for model reconstruction; (2) Docking area reconstruction driven by point cloud data: Reconstruct the point cloud obtained by actual measurement into a three-dimensional model to establish a digital model that reflects the actual part size error; (3) Docking area and non-docking area splicing: For the fuselage docking process, the non-docking area can be regarded as a load module that only provides support and load. Therefore, only the docking area needs to be reconstructed, and the non-docking area can use the theoretical digital model, which does not affect the accuracy and can reduce the difficulty and time cost of operation; (4) Docking gap finite element simulation: Use Abauqs software to simulate the docking process and solve the docking gap; (5) Data generation and proxy model construction: Batch data generation is achieved through batch processing of point clouds and finite element simulation, and a proxy model is constructed based on long short-term memory network to generate a large amount of data to facilitate process optimization.

[0056] The following is a detailed description of the content:

[0057] The first part is the scanning and data processing of the docking area. The scanning work is completed using an optical scanner; the data processing work includes the registration, denoising and downsampling of point cloud data, which is mainly completed through reverse engineering software, Python language and open source toolkits.

[0058] Part II: Docking area reconstruction driven by point cloud data. The embodiment of the present invention proposes a new point cloud reconstruction method, which first implements filtering based on a statistical outlier removal algorithm, then implements point cloud registration and matching based on a new algorithm, and then obtains a reconstructed model.

[0059] The third part is the splicing of the docking area and the non-docking area. The digital model of the docking area is spliced ​​with the theoretical digital model of the non-docking area in the 3D modeling software to form a new digital model. This digital model can accurately reflect the characteristics of the docking area under initial errors such as manufacturing errors, and avoid the cumbersome workload of reconstructing a large number of point clouds in non-critical areas.

[0060] The fourth part is the finite element simulation of the butt joint gap. The digital model obtained in the third part is imported into the finite element simulation software to simulate the butt joint process, and the finite element simulation is performed on the butt joint gap to simulate the butt joint gap in this case; the simulation value is compared with the actual value to ensure the reliability of the simulation result.

[0061] Part V: Data Generation and Proxy Model Construction. Batch modify point cloud data to simulate initial errors within various tolerance ranges; batch reconstruct models, perform batch finite element parameter modeling on each model, simulate the docking gap under different working conditions, and generate a large amount of data; establish a long-short memory neural network hybrid model based on genetic algorithm optimization, and optimize the hyperparameters of the long-short memory neural network hybrid model through genetic algorithm, so as to realize proxy model construction and optimization.

[0062] like Figure 1 As shown, the specific process of the first part of the docking area shape scanning and data processing of the embodiment of the present invention mainly includes the following steps:

[0063] Step 101, build a measurement field consisting of an optical dynamic tracking system and a handheld 3D scanner, such as Figure 2 As shown;

[0064] Step 102, scanning the shape of the parts in the docking area by a handheld 3D scanner, such as Figure 3 As shown, when designing the scanning path, it is necessary to avoid scanning too many placement planes or clamping devices as much as possible to prevent the obtained point cloud data from containing too many outlier noise points;

[0065] Step 103, using a statistical outlier removal algorithm, performing k-neighborhood statistical analysis on each scan point, calculating the average distance, and removing points whose average distance exceeds a given threshold. The k value is set to 6, that is, for each scan point, the average distance between it and the six neighboring points is calculated; the standard deviation threshold multiplier nsigma is set to 1, that is, the average distance threshold is the sum of the average distance and 1 times the standard deviation;

[0066] Step 104: For the point cloud data of other objects such as desktops or fixtures that are inevitably generated during the scanning process, a KD-tree-based clustering algorithm is used to remove such outliers. That is, the KD-tree algorithm is used to search for neighboring points of a point, calculate the distance between the points, and set a reasonable threshold to remove neighboring points whose distance is greater than the threshold. This process is iterated to remove most of the outliers.

[0067] Step 105, using an equal spacing thinning algorithm to perform overall preliminary simplification on the point cloud data, setting the sampling distance to n in the original point cloud data, and retaining the next point every n points until all points are screened;

[0068] Step 106, further simplify the point cloud using the uniform sampling method, divide the point cloud space into grids with a sphere of radius r, select the point closest to the center of the sphere as the sampling point, and control the sparsity of the sampled point cloud by modifying the radius of the sphere.

[0069] Next, the specific process of point cloud data-driven docking area reconstruction in the second part of the embodiment of the present invention is described, and the specific implementation method is as follows:

[0070] Step 201, import the theoretical numerical model into the finite element software and perform meshing operations. The distribution of the scanned point cloud along the surface of the structural part can be regarded as uniform. Considering the matching of the mesh nodes and the scanned point cloud, the mesh size should be as uniform as possible, that is, avoid the appearance of too fine meshes in local areas. Since there are turning points on the surface of the model, tetrahedral meshes are selected to ensure the uniformity of meshing;

[0071] Step 202, since the nodes divided by the finite element mesh do not coincide with the point positions in the point cloud, it is necessary to use a matching algorithm to convert the coordinate data in the point cloud as the new finite element mesh node coordinate values. The specific implementation process of the matching algorithm is: read the coordinates of the current mesh node; traverse and read the scanning point coordinates (in the block file), and calculate the distance between the scanning point coordinates and the mesh node coordinates; according to the preset distance threshold, select the 3 scanning points with the smallest distance from the scanning points whose distance to the mesh node is less than the threshold; calculate the coordinates of the matching point based on the 3 distance values ​​and the coordinates of the 3 scanning points. The algorithm flow is as follows: Figure 4 The above three distance values ​​are d1, d2, and d3, respectively, which are expressed by formula (1).

[0072]

[0073] Assume that the coordinates of the three scanning points are (x1, y1, z1), (x2, y2, z2) and (x3, y3, z3), respectively. The coordinate calculation method of the matching points is expressed by equations (2), (3) and (4).

[0074] x p =w1x1+w2x2+w3x3 (2)

[0075] y p =w1y1+w2y2+w3y3 (3)

[0076] z p =w1z1+w2z2+w3z3 (4)

[0077] In the above formula, the distance weights w1, w2 and w3 must satisfy the relationship shown in formulas (5) and (6).

[0078] w1+w2+w3=1 (5)

[0079] d1w1=d2w2=d3w3 (6)

[0080] After solving equations (5) and (6) simultaneously, we can get the expressions of w1, w2 and w3 as follows:

[0081]

[0082]

[0083]

[0084] Substituting equations (7), (8) and (9) into equations (2), (3) and (4) respectively, the coordinates of the matching points can be obtained.

[0085] In the above matching algorithm, it takes a long time to traverse and read the scanning point coordinates. In order to improve the efficiency of the algorithm, the scanning point coordinate file is divided into n blocks along the x direction of the part, so that the number of coordinates read each time is reduced to 1 / n of the original, and the calculation time is also shortened accordingly. Assume that the length of the part along the x direction is L, and the scanning point coordinate range is (x min ,x max ), then the coordinate range of each block is given by formula (10).

[0086]

[0087] In the embodiment of the present invention, n=50, and each block contains about 8000 coordinate data. After the point cloud data is divided into blocks, when reading the point cloud coordinates, the x coordinate of the current grid node is used to determine in which block the neighboring point to be searched is located;

[0088] Step 203: Obtain an inp file of a reconstructed model based on point cloud data.

[0089] like Figure 5 As shown, the specific process of splicing the docking area and the non-docking area in the third part of the embodiment of the present invention mainly includes the following steps:

[0090] Step 301, converting the reconstructed point cloud model into an STP format file;

[0091] Step 302, align the reconstructed STP format model with the theoretical model coordinate system, and remove the docking area parts in the theoretical digital model that overlap with the point cloud reconstructed model.

[0092] Step 303, through CATIA assembly quality, the point cloud reconstructed digital model is spliced ​​with the non-docking area theoretical digital model by surface coupling.

[0093] like Figure 6 As shown in the figure, the specific process of the finite element simulation of the docking gap in the fourth part of the embodiment of the present invention mainly includes the following steps:

[0094] Step 401, the docking area uses aviation aluminum alloy Al2024-T3, with a density of 2.78×10-6kg / mm3. Since the structural deformation is small during the docking process, it is generally believed that plastic deformation is not allowed, so only the Young's modulus and Poisson's ratio of the material are set to 724000MPa and 0.33 respectively;

[0095] Step 402, directly import the stp format model from the 3D modeling software, use hexahedral mesh, the unit type is C3D8R, the mesh size of the upper and lower skins and beams is 15 mm, the mesh size of the ribs is 30 mm, and the mesh size of the remaining structures is 20 mm;

[0096] Step 403, using the power / display analysis step;

[0097] Step 404, the contact between different parts is defined using the contact properties provided by Abaqus, the tangential friction behavior is defined by the penalty function, and the friction coefficient is 0.2; the normal behavior uses "hard" contact; the mutual assembly relationship between the parts is realized by the binding function;

[0098] Step 405, apply gravity load globally, select gravity acceleration of 9800m / s; apply full constraint at the POGO column support position at the lower end of the structure; apply concentrated load at the corresponding node at the connection between the docking area and the non-docking area. It should be noted that the concentrated load here is not the gravity of the docking area, because there is also POGO column support in the non-docking area. Therefore, the concentrated load here is the difference between the support force of the non-docking area and the gravity;

[0099] Step 406, in the actual docking state, after the docking is completed, the docking area will be completely surrounded by the fuselage and the wing. This makes it impossible for the measuring equipment to measure the docking gap. Therefore, a simulation part of the fuselage docking area is processed, the simulation part is fixed to the position of the actual fuselage docking area, and the wing is moved to the position of the simulation part. The problem of the inability to measure the gap caused by poor spatial openness can be solved. 10 and 11 monitoring points are arranged at the upper and lower edges of the wing docking area respectively, and the displacement data of the upper and lower edge paths of the finite element simulation are compared. If the absolute error is less than or equal to 0.01mm, the finite element simulation is considered to be valid.

[0100] like Figure 7 As shown in FIG. 5 , the specific process of data generation and agent model construction in the fifth part of the embodiment of the present invention mainly includes the following steps:

[0101] Step 501, batch modifying parameters such as point cloud density and coordinates representing initial errors such as manufacturing errors in point cloud data, but the modification range shall not exceed the allowable tolerance of part manufacturing and the allowable tolerance of assembly;

[0102] Step 502, reconstructing the model for each set of modified point cloud data, further performing finite element parametric modeling on different models based on Python language, and performing batch simulation;

[0103] Step 503, extracting deformation data of nodes corresponding to the measured positions of the docking area as labels, taking key process information such as docking movement path and load as input, and constructing a data set;

[0104] Step 504, normalize the data set and divide the training set and the data set into a ratio of 8:2;

[0105] Step 505, taking the learning rate, maximum number of iterations, number of neurons in the hidden layer, dropout rate, and batch size as optimization targets, establishing a long short-term memory network based on genetic algorithm optimization, and taking the root mean square error of the prediction result as the optimization target;

[0106] Step 506, return the optimal hyperparameters, compare the test set results and labels, evaluate the network quality, and complete the proxy model construction.

[0107] The above disclosure is only a preferred embodiment of the present invention, but the present invention is not limited thereto. Any non-creative changes that can be thought of by a person skilled in the art, as well as several improvements and modifications made without departing from the principle of the present invention, should be within the scope of protection of the present invention.

Claims

1. A method for predicting deformation of a wing docking zone driven by measured data, characterized in that: The following steps are involved: S1. Scan the shape of the docking area through the measuring equipment, construct point cloud data and process the point cloud to prepare for model reconstruction; S2. Reconstruct the point cloud obtained by actual measurement into a three-dimensional model to establish a digital model that reflects the actual part size error. The specific process of point cloud data-driven docking area reconstruction is as follows: Step 201, import the theoretical numerical model into the finite element software to perform meshing operation; the scanning point cloud distribution on the surface of the structural part is uniform, and considering the matching of the mesh nodes and the scanning point cloud, a tetrahedral mesh is selected; Step 202, the coordinate data in the point cloud is converted by a matching algorithm. The specific implementation process of the matching algorithm is as follows: read the coordinates of the current grid node; traverse and read the scanning point coordinates in the block file, and calculate the distance between the scanning point coordinates and the grid node coordinates; according to a preset distance threshold, select multiple scanning points from the scanning points whose distances to the grid nodes are less than the threshold; calculate the coordinates of the matching points according to the distance value and the coordinates of the scanning points; Step 203, obtaining a reconstructed model file based on the point cloud data; S3. For the fuselage docking process, only the docking area is reconstructed, and the non-docking area uses the theoretical digital model; the specific process of splicing the docking area and the non-docking area includes the following steps: Step 301, converting the reconstructed point cloud model into an STP format file; Step 302, the reconstructed stp format model and the theoretical model coordinate system 1, the theoretical digital model and the part of the docking area that overlaps with the point cloud reconstructed model are removed; Step 303, using CATIA assembly quality, the point cloud reconstruction digital model is spliced ​​with the non-docking area theoretical digital model by means of surface coupling; S4. Use simulation software to simulate the docking process and solve the docking gap. The specific process of finite element simulation of the docking gap mainly includes the following steps: Step 401, using aluminum alloy in the docking area; setting the Young's modulus and Poisson's ratio of the material; Step 402, directly import the model from the 3D modeling software, use a hexahedral mesh, and the unit type is C3D8R; Step 403, using the power / display analysis step; Step 404, the contact between different parts is defined using the contact properties provided by the finite element software, the tangential friction behavior is defined by the penalty function; the normal behavior uses "hard" contact; the mutual assembly relationship between the parts is realized by the binding function; Step 405, applying gravity load globally; applying full constraint at the POGO column support position at the lower end of the structure; applying concentrated load at the node corresponding to the connection between the docking area and the non-docking area; Step 406, in the actual docking state, after the docking is completed, the simulated part of the fuselage docking area is fixed to the position of the actual fuselage docking area, and the wing is moved to the position of the simulated part; multiple monitoring points are arranged at the upper edge and the lower edge of the wing docking area respectively, and the displacement data of the upper edge and the lower edge path of the finite element simulation are compared; S5. Batch modify point cloud data to simulate initial errors within various tolerance ranges; batch reconstruct models, perform batch finite element parameter modeling on each model, simulate the docking gap under different working conditions, and generate a large amount of data; establish a long-short memory neural network hybrid model based on genetic algorithm optimization, and optimize the hyperparameters of the long-short memory neural network hybrid model through genetic algorithm, so as to realize the construction and optimization of the proxy model.

2. The method for predicting wing docking zone deformation driven by measured data according to claim 1, characterized in that: The scanning work is completed using an optical scanner; the data processing work includes the alignment, denoising and downsampling of point cloud data, which is completed through reverse engineering software, Python language and open source toolkits.

3. The method for predicting wing docking zone deformation driven by measured data according to claim 1, characterized in that: The point cloud reconstruction method first implements filtering based on a statistical outlier removal algorithm, then implements point cloud registration and matching, and then obtains a reconstructed model.

4. The method for predicting wing docking zone deformation driven by measured data according to claim 1, characterized in that: During the fuselage docking process, the digital model of the docking area is spliced ​​with the theoretical digital model of the non-docking area in the 3D modeling software to form a new digital model.

5. The method for predicting wing docking zone deformation driven by measured data according to claim 4, characterized in that: The digital model obtained in claim 4 is imported into the simulation software to simulate the docking process and perform finite element simulation on the docking gap.

6. The method for predicting wing docking zone deformation driven by measured data according to claim 1, characterized in that: The specific process of docking area shape scanning and data processing includes the following steps: Step 101, building a measurement field consisting of an optical dynamic tracking system and a three-dimensional scanner; Step 102, scanning the shape of the parts in the docking area; Step 103, using a statistical outlier removal algorithm, performing k-neighborhood statistical analysis on each scan point, calculating the average distance, and removing points whose average distance exceeds a given threshold, where the k value represents the average distance between each scan point and its k neighboring points; Step 104, searching for neighboring points of a point by using the KD-tree algorithm, calculating the distance between the points, and setting a reasonable threshold to remove neighboring points whose distance is greater than the threshold, iterating the process to remove outliers; Step 105, use the equal spacing thinning algorithm to perform overall preliminary simplification on the point cloud data, and set the sampling distance in the original point cloud data to n , each interval n Keep the next point until all points are screened; Step 106, further simplify the point cloud using the uniform sampling method, divide the point cloud space into grids with a sphere of radius r, select the point closest to the center of the sphere as the sampling point, and control the sparsity of the sampled point cloud by modifying the radius of the sphere.

Images