A deformation prediction method for the assembly process of a curved conformal antenna

Abstract

The present invention provides a method for predicting deformation during the assembly process of a curved conformal antenna, which relates to the field of intelligent prediction of assembly processes. The method comprises the following steps: establishing a workpiece surface library based on measured surface data and simulated surface data, performing finite element simulation analysis of the assembly process under different process parameters on the workpiece surface library, and obtaining the deformation results after assembly; training and building a GAN-based assembly deformation prediction model using surface manufacturing errors and process parameters as input and the deformation results after assembly as output; importing the surface data detected in real time into the assembly deformation prediction model, predicting the assembly deformation images under different process parameters, and using the result with the minimum error between the assembly deformation image and the ideal surface model as the optimal assembly process parameters that meet the current conditions. The present invention achieves real-time prediction of deformation during the curved surface assembly process and can determine the optimal assembly process parameters based on the actual curved surface manufacturing errors, thereby solving the problem that the curved antenna assembly process is invisible and difficult to assemble and adjust.

Description

Technical Field

[0001] The present invention relates to the field of intelligent prediction of assembly processes, and in particular to a method for predicting deformation during the assembly process of a curved conformal antenna. Background Art

[0002] Curved conformal antennas, which conform to the surface of the carrier platform without disrupting the carrier's external structure and aerodynamic properties, are a research hotspot in the antenna field. They can effectively expand radar apertures, improve detection range, and extend detection range. However, due to their complex structure, curved conformal antennas are difficult to form using traditional manufacturing methods. Therefore, additive manufacturing methods are often used to separately process the multi-layer heterogeneous structure and then assemble it through processes such as gluing and riveting. However, additive manufacturing methods inherently suffer from low manufacturing stiffness and poor precision. Furthermore, the large number and small size of the antenna's electrical interconnect features make deformation during the assembly process a significant factor affecting assembly performance. Existing assembly processes are mostly manual, resulting in high labor costs and low efficiency. Furthermore, since the interlayer structures are invisible, the assembly process relies solely on manual experience and requires repeated attempts, making it difficult to ensure assembly accuracy and quality. Assembly quality often requires post-assembly testing, resulting in low yield and inefficiency.

[0003] Therefore, a method for predicting surface deformation in the assembly process is needed to achieve online control of the assembly process.

[0004] Chinese invention patent publication number CN109918755A proposes a method for predicting the assembly deformation of low-stiffness parts based on point cloud data. This method uses a 3D laser scanner to scan the low-stiffness part to obtain surface information point cloud data and perform point cloud preprocessing. The point cloud coordinates are transformed into a 3D coordinate system, and a 2D mesh is created using x and y plane coordinates. This results in a quadrilateral mesh composed of ordered representative points in the point cloud. This mesh is then input into finite element analysis software for constraint and displacement loading, resulting in the shape change of the low-stiffness part after assembly. This method may take into account actual manufacturing errors in the workpiece and rely solely on finite element simulation analysis for deformation prediction, but it fails to consider real-time control of the assembly process, resulting in low prediction efficiency.

[0005] Chinese invention patent publication number CN116108762A proposes a method for predicting the assembly deformation of large composite components using force sensors. This method uses force sensors and 3D laser scanning equipment to obtain external force and deformation data, and then uses a neural network to predict the deformation of the inner surface of the wall panel after the force is applied during the assembly process. This method uses finite element analysis to build a neural network prediction model to predict the actual state of the workpiece's shape, but does not consider the effects of initial deformation and the coupling of multi-parameter assembly processes. Summary of the Invention

[0006] Purpose of the invention: To propose a deformation prediction method for the assembly process of a curved conformal antenna to solve the above-mentioned problems existing in the prior art.

[0007] The present invention proposes a deformation prediction method for a curved conformal antenna assembly process, comprising the following steps:

[0008] Establishing a workpiece surface library based on measured surface data and simulated surface data, performing finite element simulation analysis of the assembly process with different process parameters on the workpiece surface library to obtain deformation results after assembly;

[0009] Using surface manufacturing errors and process parameters as input and assembly deformation results as output, a GAN-based assembly deformation prediction model is trained and built.

[0010] The real-time detected surface data is imported into the assembly deformation prediction model to predict the assembly deformation image under different process parameters, and the result with the minimum error between the assembly deformation image and the ideal surface model is used as the optimal assembly process parameter that meets the current conditions.

[0011] In a further embodiment, the measured surface data is obtained by:

[0012] Point cloud data acquisition equipment is used to obtain several measured point cloud data of the assembly surface. The measured point cloud data are filtered and spliced to obtain the overall surface as the measured surface data.

[0013] In a further embodiment, the simulated surface data is obtained by:

[0014] Collect several sets of measured surface data, align them with the ideal surface model, obtain the manufacturing error law of the surface, and construct simulated surface data based on the manufacturing error law.

[0015] In a further embodiment, the ideal surface mathematical model is , for the measured surface data and ideal surface mathematical models ICP registration is performed using the screw hole as a feature, aligning the measured surface data with the ideal surface model and performing coordinate system transformation;

[0016] The measured surface data The measured points in the ideal surface model Projection in the direction of the surface normal and calculation of the projection distance Generate an error field; where Represents measured point cloud data The measured points in Represents an ideal surface model Center and measured points The corresponding point of express Point normal direction vector;

[0017] Ideal surface model Divide along the surface contour into The grid of the specification is used to calculate the mean error of all point clouds in each grid. , maximum value and standard deviation ;

[0018] Repeat the above steps multiple times to obtain multiple sets of manufacturing error data.

[0019] In a further embodiment, constructing simulated surface data based on manufacturing error rules specifically includes:

[0020] The semivariogram is used to calculate and analyze the spatial distribution characteristics of each set of manufacturing error data in the X, Y, and Z directions:

[0021]

[0022]

[0023]

[0024] Where, 、 、 are the X, Y, and Z axis coordinates of the measured point respectively; 、 、 They are the point spacing in the X, Y and Z axis directions respectively; 、 、 The spacing is 、 、 The point logarithm of ;

[0025] Fit the covariance model, calculate the correlation length and variance parameters, and establish the error model;

[0026] A random simulation error field is generated based on Kriging interpolation, and M groups of simulated surface data containing simulated manufacturing errors are generated.

[0027] In a further embodiment, a GAN-based assembly deformation prediction model is trained and constructed with surface manufacturing errors and process parameters as input and deformation results after assembly as output, specifically including:

[0028] Based on the contour corner points and screw hole features, grids with the same number of nodes are arranged for the ideal surface model and the measured surface data respectively. The mapping relationship between the ideal surface model and the measured surface data is established using the nodes, where the node set of the ideal surface model is PI and the node set of the measured surface data is PR.

[0029] Based on the ideal surface model, the predicted surface node coordinates after assembly deformation are D 、Y D , Z D and the actual surface node coordinate X I 、Y I , Z I The coordinates of the contour nodes corresponding to the theoretical surface are X R 、Y R , Z R The coordinate deviations are normalized and mapped to the red, green, and blue values of the RGB color space to prepare the input and output images for training the generative adversarial network.

[0030] The dataset is divided into training set, validation set and test set. The generator inputs the surface manufacturing error image and loading conditions and outputs the predicted surface deformation result. The discriminator inputs the simulated deformation result or the deformation result predicted by the generator and outputs the discrimination result. The training is repeated alternately.

[0031] In a further embodiment, the actual surface node coordinates X are taken as a reference based on the ideal surface model. I 、Y I , Z I Coordinate X of the contour node corresponding to the theoretical surface R 、Y R , Z R The coordinate deviation is normalized to [0,255] and mapped to R, G, and B values respectively to obtain the RGB error image:

[0032]

[0033]

[0034]

[0035] Where, 、 、 are the values of red, green and blue mapped to the i-th point respectively; 、 、 are the coordinate deviations of the i-th point in the X, Y, and Z directions respectively; 、 、 are the minimum values of coordinate deviation in the X, Y, and Z directions respectively; 、 、 They are the maximum values of coordinate deviation in the X, Y, and Z directions respectively.

[0036] In a further embodiment, the real-time detected surface data is imported into the assembly deformation prediction model to predict the assembly deformation image under different process parameters, specifically including:

[0037] Using a point cloud data acquisition device to acquire the surface data of the current curved conformal antenna in real time, registering it with the ideal surface model to obtain simulated surface data including manufacturing errors, and reducing the dimension of the simulated surface data to a two-dimensional image as one of the inputs of the assembly deformation prediction model;

[0038] Taking minimization of post-assembly deformation error as the optimization goal, the process parameter range is defined, the assembly deformation prediction model is called, the objective function value is calculated, the process parameters are updated, and the iterative optimization is repeated until the convergence conditions are met.

[0039] In a further embodiment, the assembly deformation prediction model includes a generator and a discriminator; the generator is based on The random RGB error image of the resolution and the process parameter vector are input, and the predicted deformation field RGB image is output. The discriminator inputs the real sample or the generated sample and outputs the probability that the sample is true;

[0040] Input real data and generated data, update the discriminator parameters, and maximize the discrimination accuracy; deceive the discriminator by generating data, update the generator parameters, and minimize the discriminator's probability of distinguishing the generated samples;

[0041] The generator is trained once every predetermined number of times for each training of the discriminator; the generator and the discriminator are trained iteratively alternately to input a random manufacturing error image and a process parameter vector and output an assembly deformation image;

[0042] Multi-constraint optimization is performed on the screw tightening sequence and tightening torque to ensure that the gap between the assembly deformation image and the ideal surface model is minimized after assembly.

[0043] In addition, the present invention also proposes an electronic device, which includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the above-mentioned deformation prediction method for the curved conformal antenna assembly process is implemented.

[0044] Compared with the prior art, the present invention has at least the following beneficial effects:

[0045] (1) The present invention realizes the real-time prediction of deformation during the curved surface assembly process and can determine the optimal assembly process parameters based on the actual curved surface manufacturing error, thus solving the problem that the curved antenna assembly process is invisible and difficult to assemble and adjust.

[0046] (2) The present invention adopts a multi-resolution registration method for high-precision point cloud stitching, which effectively improves the speed of point cloud stitching and provides a solid foundation for real-time prediction of assembly deformation.

[0047] (3) The present invention uses images based on GAN to predict surface deformation. Compared with the traditional finite element analysis prediction method, the prediction speed is increased by 20%, which is more suitable for the real-time requirements of actual production.

[0048] (4) The present invention reduces the dimensionality of three-dimensional point cloud data to two-dimensional color image space for processing, and the prediction process is easy to visualize and has significant effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 4 is a flow chart of a method for predicting deformation during the assembly process of a curved conformal antenna in an embodiment.

[0050] Figure 2 Schematic diagram of a multi-resolution point cloud registration method proposed for high-density point cloud stitching in an embodiment.

[0051] Figure 3 3D point cloud to 2D image error mapping model diagram in the embodiment.

[0052] Figure 4 RGB error image example in the embodiment.

[0053] Figure 5 Flowchart of training the assembly deformation prediction model in the embodiment. DETAILED DESCRIPTION

[0054] In the following description, numerous specific details are provided to provide a more thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without one or more of these details. In other instances, certain technical features well known in the art have not been described to avoid confusion with the present invention.

[0055] This embodiment discloses a deformation prediction method for the curved conformal antenna assembly process taking into account manufacturing errors. The flow chart is shown in FIG. Figure 1 , the specific steps are as follows:

[0056] Based on the visual guidance of a large field of view camera, a high-precision laser scanner or a high-precision binocular camera is moved to scan multiple times to obtain T groups of 3D surface point cloud raw data containing manufacturing errors. :

[0057]

[0058] in, ; N is the number of point cloud data points obtained by a single scan or shot.

[0059] 3D surface point cloud raw data Perform outlier removal, based on statistical filtering, remove points that deviate from the mean by k times the standard deviation, and retain those that meet the conditions point ,in Indicates the average z-coordinate value of all point cloud data in a single scan or shot. Indicates the z-coordinate standard deviation of all point cloud data in a single scan or shot.

[0060] Use voxel grid filtering, retain the centroid point in the grid, define the voxel size as v, and use the empirical formula Initialize v, where is the average density of the original point cloud (points / mm²).

[0061] The index of each voxel is calculated from the point coordinates :

[0062] , , ,in is the minimum vertex coordinate of the point cloud bounding box, Indicates rounding down.

[0063] All points are assigned to corresponding voxels according to their coordinates. The point set included is: ;

[0064] For each non-empty voxel, calculate the centroid of all points inside it as the representative point: , where n is the number of points within the voxel, further weighted calculation: , preserving the normal vector features.

[0065] Output the filtered point cloud, which is the set of all voxel representative points: , , where M is the number of retained points after filtering.

[0066] Perform coarse registration based on the position of the target point of a high-precision scanner or binocular camera under the field of view of a large-field camera;

[0067] The point cloud sets after filtering of T groups are Construct a Gaussian pyramid to generate L layers of point cloud with different resolutions and l layers of voxel size. , resolution ,in is the initial voxel size, initialized according to the empirical formula , the set of voxel representative points in the l-layer is , .

[0068] Clustering of adjacent points and The first layer of point cloud and Use FPFH descriptors to extract features and calculate feature correspondences to obtain matching scores , the retention score is above the threshold Matching pairs, randomly sample matching pairs, estimate rigid body transformation, and calculate the initial transformation matrix of each layer .

[0069] Starting from the coarsest layer l=0, optimize the transformation parameters layer by layer, with the number of iterations decreasing for each layer. Let the transformation after optimization of the lth layer be , which is used as the initial value of the l+1th layer. At the lth layer, the source point cloud and target point cloud The registration error is: ,in , are corresponding point pairs, is the weight, ,in , is the outlier threshold.

[0070] Introducing regularization terms Limiting inter-layer transformation mutations, ,in is the balance factor, is the Frobenius norm, and an optimization algorithm such as SVD or Levenberg-Marquardt is used to solve the optimal transformation: .

[0071] The registration results of each layer are integrated and the motion consistency constraint of adjacent layers is introduced to avoid registration mutations caused by resolution changes. . Construct a global error function :

[0072]

[0073] in Indicates mapping the SE(3) transformation matrix to the Lie algebra space se(3), represents the weighted norm of the Lie algebra vector, Represents the weight of the ICP error term of each layer, which increases with the improvement of resolution. Represents the weight of the smoothness term, balancing the registration accuracy and transformation continuity;

[0074] Transform each layer As the node of graph optimization, the difference between layers is used as the edge constraint, and the sparse solver is used to solve the optimal adjacent point cloud slices. .

[0075] Repeat the above steps to obtain the complete point cloud data of the curved antenna including manufacturing errors. :

[0076]

[0077] in, ; N is the number of point cloud data points obtained by a single scan or shot.

[0078] See the above process Figure 2 shown.

[0079] Assume that the ideal surface mathematical model is , for the measured point cloud data and ideal surface mathematical models ICP registration is performed using the screw holes as features, aligning the point cloud data with the ideal surface model and performing coordinate system transformation;

[0080] The measured point cloud data The measured points in the ideal surface model Projection in the direction of the surface normal and calculation of the projection distance Generate error field;

[0081] Ideal surface model Divide along the surface contour into The grid of the specification is used to calculate the mean error of all point clouds in each grid. , maximum value and standard deviation ;

[0082] Repeat the above steps n times to obtain multiple sets of measured surface error data.

[0083] The semivariogram is used to calculate and analyze the spatial distribution characteristics of each set of errors in the X, Y, and Z directions:

[0084]

[0085]

[0086]

[0087] in, 、 、 are the X, Y, and Z axis coordinates of the measured point respectively; 、 、 They are the point spacing in the X, Y and Z axis directions respectively; 、 、 The spacing is 、 、 The number of point pairs.

[0088] Fit the covariance model, estimate the correlation length and variance parameters, and establish the error model;

[0089] A random simulation error field is generated based on Kriging interpolation, and M groups of surface models containing simulated manufacturing errors are generated.

[0090] Import the surface model including manufacturing errors into Abaqus, define material parameters including density, Young's modulus, Poisson's ratio, etc., and set the material direction separately for composite materials;

[0091] Assemble component instances to ensure effective contact between models;

[0092] Use translator equivalent to replace the screw connection and create an analysis step for each screw tightening process;

[0093] Apply loads and boundary conditions, create meshes, and submit the job;

[0094] Using the Python interface provided by Abaqus for parametric modeling, the screw tightening sequence and related boundary conditions were modified, and the solution was automatically looped to obtain multiple sets of simulated deformation data of surfaces with different manufacturing errors under different process parameters, forming a simulation database.

[0095] Based on the features such as contour corner points and screw holes, the surface after assembly deformation is arranged with the same number of nodes as the ideal surface and the measured surface. Mesh, referring to the manufacturing error processing method of measured surface, generates the assembly deformation error field;

[0096] Establish a node mapping relationship from deformation error to image space, normalize the deviation values of the measured surface, the assembly deformation surface and the ideal surface in the x, y, and z axes to [0, 255], and map the x, y, and z axis deviations to R, G, and B values respectively. The conversion process is shown in Figure 3 As shown, the converted RGB error image is as follows Figure 4 shown.

[0097]

[0098]

[0099]

[0100] The process of training the assembly deformation prediction model is shown in Figure 5 The generator is The random RGB error image of the resolution and the process parameter vector are input, and the predicted deformation field RGB image is output. The discriminator inputs the real sample or the generated sample and outputs the probability that the sample is true;

[0101] Input real data and generated data, update the discriminator parameters, and maximize the discrimination accuracy; deceive the discriminator by generating data, update the generator parameters, and minimize the discriminator's probability of distinguishing the generated samples;

[0102] Each time the discriminator is trained 5 times, the generator is trained once, and the generator and discriminator are trained alternately and iteratively to input random manufacturing error images and process parameter vectors and output assembly deformation images;

[0103] Multi-constraint optimization is performed on process parameters such as screw tightening sequence and tightening torque to ensure that the deformation after assembly is minimal compared to the ideal surface.

[0104] The technical process of the deformation prediction method for the curved conformal antenna assembly process disclosed in the above embodiment can be implemented in whole or in part through software, hardware, firmware or any other combination.

[0105] When implemented using hardware, the aforementioned embodiments can be run on an electronic device by compiling all or part of the operating logic and computing processes into software. The electronic device includes a processor, a memory, a communication interface, and a communication bus. The processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which enables the processor to execute the technical process disclosed in the aforementioned embodiments.

[0106] When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. If the above methods are implemented in the form of software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present application, or the part that contributes to the relevant technology, can be embodied in the form of a software product. The software product is stored in a storage medium and includes several instructions for enabling an electronic device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of the present application. The aforementioned storage media include various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), magnetic disks, or optical disks. Thus, the embodiments of the present application are not limited to any specific hardware, software, or firmware, or any combination of hardware, software, and firmware.

[0107] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A method for predicting deformation during the assembly process of a curved conformal antenna, characterized in that: The steps include: Establishing a workpiece surface library based on measured surface data and simulated surface data, performing finite element simulation analysis of the assembly process with different process parameters on the workpiece surface library to obtain deformation results after assembly; Using surface manufacturing errors and process parameters as input and assembly deformation results as output, we trained and built a GAN-based assembly deformation prediction model, specifically including: Based on the contour corner points and screw hole features, grids with the same number of nodes are arranged for the ideal surface model and the measured surface data respectively. The mapping relationship between the ideal surface model and the measured surface data is established using the nodes, where the node set of the ideal surface model is PI and the node set of the measured surface data is PR. Based on the ideal surface model, the predicted surface node coordinates after assembly deformation are D 、Y D , Z D and the actual surface node coordinate X I 、Y I , Z I The coordinates of the contour nodes corresponding to the theoretical surface are X R 、Y R , Z R The coordinate deviations are normalized and mapped to the red, green, and blue values of the RGB color space to prepare the input and output images for training the generative adversarial network. The dataset is divided into training set, validation set and test set. The generator inputs the surface manufacturing error image and loading conditions and outputs the predicted surface deformation result. The discriminator inputs the simulated deformation result or the deformation result predicted by the generator and outputs the discrimination result. The training is repeated alternately. Importing the real-time detected surface data into the assembly deformation prediction model, predicting the assembly deformation image under different process parameters, and taking the result with the minimum error between the assembly deformation image and the ideal surface model as the optimal assembly process parameter that meets the current conditions, specifically including: Using a point cloud data acquisition device to acquire the surface data of the current curved conformal antenna in real time, registering it with the ideal surface model to obtain simulated surface data including manufacturing errors, and reducing the dimension of the simulated surface data to a two-dimensional image as one of the inputs of the assembly deformation prediction model; Taking minimization of post-assembly deformation error as the optimization goal, the process parameter range is defined, the assembly deformation prediction model is called, the objective function value is calculated, the process parameters are updated, and the iterative optimization is repeated until the convergence conditions are met.

2. The method for predicting deformation during the assembly process of a curved conformal antenna according to claim 1, wherein: The measured surface data is obtained by: Point cloud data acquisition equipment is used to obtain several measured point cloud data of the assembly surface. The measured point cloud data are filtered and spliced to obtain the overall surface as the measured surface data.

3. The method for predicting deformation during the assembly process of a curved conformal antenna according to claim 1, wherein: The simulated surface data is obtained in the following manner: Collect several sets of measured surface data, align them with the ideal surface model, obtain the manufacturing error law of the surface, and construct simulated surface data based on the manufacturing error law.

4. The method for predicting deformation during the assembly process of a curved conformal antenna according to claim 3, wherein: Assume that the ideal surface mathematical model is , for the measured surface data and ideal surface mathematical models ICP registration is performed using the screw hole as a feature, aligning the measured surface data with the ideal surface model and performing coordinate system transformation; The measured surface data The measured points in the ideal surface model Projection in the direction of the surface normal and calculation of the projection distance Generate an error field; where Represents measured point cloud data The measured points in Represents an ideal surface model Center and measured points The corresponding point of express Point normal direction vector; Ideal surface model Divide along the surface contour into The grid of the specification is used to calculate the mean error of all point clouds in each grid. , maximum value and standard deviation ; Repeat the above steps multiple times to obtain multiple sets of manufacturing error data.

5. The method for predicting deformation during the assembly process of a curved conformal antenna according to claim 4, wherein: Constructing simulated surface data based on manufacturing error rules, specifically including: The semivariogram is used to calculate and analyze the spatial distribution characteristics of each set of manufacturing error data in the X, Y, and Z directions: ； ； ； Where, 、 、 are the X, Y, and Z axis coordinates of the measured point respectively; 、 、 They are the point spacing in the X, Y and Z axis directions respectively; 、 、 The spacing is 、 、 The point logarithm of ; Fit the covariance model, calculate the correlation length and variance parameters, and establish the error model; A random simulation error field is generated based on Kriging interpolation, and M groups of simulated surface data containing simulated manufacturing errors are generated.

6. The method for predicting deformation during the assembly process of a curved conformal antenna according to claim 1, wherein: Taking the ideal surface model as the benchmark, the actual surface node coordinates X I 、Y I , Z I Coordinate X of the contour node corresponding to the theoretical surface R 、Y R , Z R The coordinate deviation is normalized to [0,255] and mapped to R, G, and B values respectively to obtain the RGB error image: ； ； ； Where, 、 、 are the values of red, green and blue mapped to the i-th point respectively; 、 、 are the coordinate deviations of the i-th point in the X, Y, and Z directions respectively; 、 、 are the minimum values of coordinate deviation in the X, Y, and Z directions respectively; 、 、 They are the maximum values of coordinate deviation in the X, Y, and Z directions respectively.

7. The method for predicting deformation during the assembly process of a curved conformal antenna according to claim 6, wherein: The assembly deformation prediction model includes a generator and a discriminator; the generator is based on The random RGB error image of the resolution and the process parameter vector are input, and the predicted deformation field RGB image is output. The discriminator inputs the real sample or the generated sample and outputs the probability that the sample is true; Input real data and generated data, update the discriminator parameters, and maximize the discrimination accuracy; deceive the discriminator by generating data, update the generator parameters, and minimize the discriminator's probability of distinguishing the generated samples; The generator is trained once every predetermined number of times for each training of the discriminator; the generator and the discriminator are trained iteratively alternately to input a random manufacturing error image and a process parameter vector and output an assembly deformation image; Multi-constraint optimization is performed on the screw tightening sequence and tightening torque to ensure that the gap between the assembly deformation image and the ideal surface model is minimized after assembly.

8. An electronic device, characterized in that: The electronic device comprises: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the method for predicting deformation during the assembly process of a curved conformal antenna according to any one of claims 1 to 7 is implemented.

Images