A composite component forming die compensation method based on manifold neural operator
Through the composite component mold compensation method based on manifold neural operator, a solidified deformation prediction agent model is constructed and combined with an intelligent optimization algorithm, the problem of poor optimization effect of complex deformation modes in the existing technology is solved, and the complex shapes are realized. Efficient control and wide applicability.
Patent Information
- Application Number
- CN202411246296.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-06
AI Technical Summary
The existing composite component mold compensation method is difficult to achieve ideal optimization effects when facing complex deformation modes, and the degree of adjustability is limited, making it difficult to apply to components of complex shapes.
The composite component mold compensation method based on manifold neural operator is adopted. By constructing a curing deformation prediction agent model, and combining intelligent optimization algorithms for massive iterative design, the mold model surface is optimized to reduce the shape difference after curing deformation.
It significantly improves the curing deformation control effect of composite components, has a wide range of application and high efficiency, and can handle mold surface compensation of complex geometric shapes, reducing the time and cost required for design.
Smart Images

Figure CN118761296B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of composite component solidification and forming, and in particular relates to a composite component forming die compensation method based on manifold neural operator. Background Art
[0002] Advanced resin-based composite materials are integrated materials formed by the combination of organic polymer matrix materials and high-performance fiber-reinforced materials. They are lightweight, high in specific strength and modulus, and highly resistant to corrosion, and have been widely used in the aerospace field. The manufacturing of composite components usually includes a shaping process and a curing process. During the shaping process, the prepreg made by combining the matrix and the reinforcement is laid on the molding mold. During the curing process, the material undergoes a series of physical and chemical changes under high temperature and high pressure to cure and form. The state of the material will also change greatly during curing. Due to factors such as the difference in thermal expansion between the matrix and the reinforcement, curing shrinkage, and uneven temperature distribution, the shape and size of the composite component after demoulding deviates from the designed shape.
[0003] In order to control the deviation between the shape of the composite component after demoulding and the required shape within the allowable range, the most direct and effective way is to compensate the mold surface. The mold surface during the curing of the composite component is usually designed as the mold surface of the composite component. The mold surface compensation method trims the mold surface according to the component curing deformation result so that the deformed component shape meets the accuracy requirements. Patent CN114889159A discloses an adjustable mold and compensation method for composite curing molding. The mold can adjust the mold surface through an angle adjustment mechanism and a surface curvature adjustment mechanism to compensate for the angle change and surface warping deformation of the composite C-shaped workpiece. However, the mold has limited freedom to optimize and adjust, and the shape is single, which is difficult to apply to other component shapes. Patent CN 115438394A discloses a composite structural component forming mold compensation method. The composite component is simulated for curing deformation by finite element software to obtain simulation calculation results, and the initial coordinates and deformation of the nodes on the mold surface are extracted. The node deformation is reversely applied to the node initial coordinates by the mirror compensation principle, thereby realizing compensation for the mold surface. However, the mirror compensation method is only suitable for compensating for the springback-type curing deformation of composite components. It is difficult to achieve the ideal optimization effect when the deformation mode of the composite component is more complex. Summary of the invention
[0004] The purpose of the present invention is to provide a composite component forming mold compensation method based on manifold neural operator, which constructs a composite component curing deformation prediction agent model through the manifold neural operator, significantly improves the efficiency of composite component curing deformation prediction, and combines this with an intelligent optimization algorithm as a basis to support massive iterative optimization design of the forming mold surface, thereby significantly improving the composite component curing deformation control effect, and has a wide range of applications and high efficiency.
[0005] To achieve the above object, the present invention provides a composite component forming die compensation method based on manifold neural operator, comprising the following steps:
[0006] Step 1: obtaining a three-dimensional model of the composite component as a target shape, and determining a mastering surface during curing and molding of the composite component according to the three-dimensional model;
[0007] Step 2, parameterize the template surface to determine the design variables and their value ranges;
[0008] Step 3, obtaining a number of master surface design schemes in the design space by a sampling method, obtaining the shape of the composite component corresponding to each master surface design scheme, performing a curing deformation simulation on the shape of the composite component, obtaining the cured deformed shape corresponding to the shape of the composite component, and establishing a curing deformation data set;
[0009] Step 4: Establish a manifold neural operator model with the shape of the composite component as input and the shape after curing deformation as output, and use the curing deformation data set for training to obtain a curing deformation prediction agent model;
[0010] Step 5: Taking minimizing the difference between the shape after curing deformation and the target shape as the optimization goal, the intelligent optimization algorithm is used to optimize the design variables. During the optimization process, the curing deformation prediction agent model is used to calculate the shape after curing deformation, and the shape of the composite component that meets the curing deformation requirements of the composite component is obtained, and then the corresponding mastering surface is used as the molding mold surface.
[0011] Preferably, in step 2, one of the following methods is used to parameterize the template surface: parameterize the template surface based on Bessel function; parameterize the template surface based on geometric parameters of angle, length, and point coordinates; parameterize the template surface based on free deformation technology.
[0012] Preferably, the sampling method in step 3 is random sampling or Latin hypercube sampling.
[0013] Preferably, the curing deformation simulation of the shape of the composite component in step 3 specifically includes: according to the curing process in the actual molding process, using the finite element method, the finite difference method or commercial simulation software to simulate the curing deformation of the shape of the composite component.
[0014] Preferably, each data in the solidification deformation data set in step 3 consists of the shape of the composite component and the corresponding shape after solidification deformation, specifically, the shape of the composite component and the shape after solidification deformation are discretized into a grid or point cloud, and the coordinate field of the nodes on the shape is used as the data representation of the shape.
[0015] Preferably, the manifold neural operator model in step 4 is composed of a feature dimensionality increase mapping module, one or more Laplace kernel integration modules and a feature dimensionality reduction mapping module, wherein the Laplace kernel integration module includes three modules: frequency domain transformation, linear transformation and nonlinear activation; the frequency domain transformation module includes three modules: encoding, parameterization and decoding; the encoding module uses the Laplace operator characteristic function to convert the function defined on the composite component into coordinates in the frequency domain space; the parameterization module maps the encoded coordinates; the decoding module uses the Laplace operator characteristic function to restore the parameterized frequency domain space coordinates to the function defined on the composite component.
[0016] Preferably, the Laplace operator characteristic function is obtained by defining the Laplace operator in a discrete form according to the data of the shape of the composite component and the shape after solidification deformation, solving the eigenvalue problem of the Laplace operator, and obtaining the Laplace operator characteristic function.
[0017] Preferably, the specific calculation method of the difference between the deformed shape after curing and the target shape in step 5 includes: calculating the maximum deformation between the deformed shape after curing and the target shape; taking multiple points on the deformed shape after curing and the target shape, and calculating the average value, mean square error, root mean square error or relative error of the deformation of the multiple points; calculating the difference between the angle or distance between the corresponding positions on the deformed shape after curing and the target shape.
[0018] Preferably, the intelligent optimization algorithm in step 5 is one of a genetic algorithm, a Bayesian optimization algorithm, a particle swarm optimization algorithm, and an annealing algorithm.
[0019] Therefore, the present invention adopts the above-mentioned composite component forming die compensation method based on manifold neural operator, which has the following beneficial effects:
[0020] (1) The design space of the mastering surface of the present invention is determined by a finite number of parameters of the mastering surface and their parameterization method. A suitable parameterization method will enable the design space of the mastering surface to include any three-dimensional complex geometry, rather than being limited to changes in angles and profile curvature, thereby achieving mold profile compensation for composite components of any complex geometry;
[0021] (2) The present invention constructs a composite component solidification deformation prediction agent model based on the manifold neural operator model, which is more efficient than using simulation prediction. The solidification deformation prediction agent model is optimized and iterated in the intelligent algorithm, and the obtained prototype surface will make the difference between the composite component's solidification deformation shape and the target shape smaller.
[0022] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1It is an overall flow chart of a composite component forming die compensation method based on manifold neural operator of the present invention;
[0024] Figure 2 A structural diagram of a composite component according to an embodiment of the present invention;
[0025] Figure 3 The deformation fields of the post-curing deformed shape and the target shape of the composite component before and after the mold surface compensation of the embodiment of the present invention, wherein (a) is the deformation field of the post-curing deformed shape and the target shape of the composite component before compensation, and (b) is the deformation field of the post-curing deformed shape and the target shape of the composite component after the mold surface compensation. DETAILED DESCRIPTION
[0026] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0027] See also Figure 1-3 , a composite component forming die compensation method based on manifold neural operator, comprising the following steps:
[0028] Step 1: obtaining a three-dimensional model of the composite component as a target shape, and determining a mastering surface during curing and molding of the composite component according to the three-dimensional model;
[0029] Step 2, parameterize the template surface to determine the design variables and their value ranges;
[0030] The following methods are used to parametrically represent the template surface: parametrically represent the template surface based on Bessel function; parametrically represent the template surface based on geometric parameters such as angle, length and point coordinates; parametrically represent the template surface based on free deformation technology.
[0031] Step 3, in the design space, a number of master surface design schemes are obtained by random sampling or Latin hypercube sampling, and the shape of the composite component corresponding to each master surface design scheme is obtained, and a curing deformation simulation is performed on the shape of the composite component to obtain a cured deformed shape corresponding to the shape of the composite component, and a curing deformation data set is established; wherein the curing deformation simulation of the shape of the composite component specifically includes: according to the curing process in the actual molding process, the finite element method, the finite difference method or commercial simulation software is used to perform a curing deformation simulation on the shape of the composite component; each data in the curing deformation data set consists of a composite component shape and a corresponding cured deformed shape, specifically, the composite component shape and the cured deformed shape are discretized into a grid or point cloud, and the coordinate field of the nodes on the shape is used as the data representation of the shape.
[0032] Step 4: Establish a manifold neural operator model with the shape of the composite component as input and the shape after solidification deformation as output, and use the solidification deformation data set for training to obtain a solidification deformation prediction agent model; wherein the manifold neural operator model is composed of a feature dimensionality increase mapping module, one or more Laplace kernel integration modules and a feature dimensionality reduction mapping module, wherein the Laplace kernel integration module includes three modules: frequency domain transformation, linear transformation and nonlinear activation, and the frequency domain transformation module includes three modules: encoding, parameterization and decoding. The encoding module uses the Laplace operator characteristic function to convert the function defined on the composite component into coordinates in the frequency domain space; the parameterization module maps the encoded coordinates; the decoding module uses the Laplace operator characteristic function to restore the parameterized frequency domain space coordinates to the function defined on the composite component; the Laplace operator characteristic function is obtained in the following way: define the Laplace operator according to the data discrete form of the shape of the composite component and the shape after solidification deformation, solve the eigenvalue problem of the Laplace operator, and obtain the Laplace operator characteristic function.
[0033] Step 5, taking minimizing the difference between the shape after curing deformation and the target shape as the optimization goal, the intelligent optimization algorithm is used to optimize the design variables, and the curing deformation prediction agent model is used to calculate the shape after curing deformation during the optimization process, so as to obtain the shape of the composite component that meets the curing deformation requirements of the composite component, and then the corresponding mold surface is used as the molding mold surface. The specific calculation method of the difference between the shape after curing deformation and the target shape includes: calculating the maximum deformation between the shape after curing deformation and the target shape; taking multiple points on the shape after curing deformation and the target shape, and calculating the average value, mean square error, root mean square error or relative error of the deformation of multiple points; calculating the difference between the angle or distance between the corresponding position of the shape after curing deformation and the target shape. The intelligent optimization algorithm used is one of the genetic algorithm, Bayesian optimization, particle swarm optimization algorithm, and annealing algorithm.
[0034] Example
[0035] In this embodiment, the forming die compensation method proposed by the present invention is carried out on a three-dimensional U-shaped part, and the component geometry is referred to as shown in the attached figure. Figure 2 , the specific steps of this embodiment are as follows:
[0036] Step 1: obtaining a three-dimensional model of the composite component as a target shape, and determining a mastering surface during curing and forming of the composite component according to the three-dimensional model;
[0037] Get attached Figure 2 The three-dimensional model of the U-shaped composite component is taken as the target shape, and the inner surface of the U-shaped composite component is determined as the mold surface for solidification molding.
[0038] Step 2: parametrically represent the template surface and determine the design variables and their value ranges;
[0039] The mold surface of the composite component is parameterized by five design variables, which respectively represent the length, width, height, chamfer and offset of the mold surface. According to the simulation analysis of the curing deformation of the composite component, in this embodiment, the influence of the length, width and height of the mold surface on the deformation of the component is small and can be ignored, while the chamfer and offset have a greater influence on the curing deformation of the component. Therefore, in this embodiment, the chamfer is set as and bias As optimization variables, the value ranges are , .
[0040] Step 3: Obtain a certain number of master surface design schemes in the design space through sampling methods, obtain the shape of the composite component corresponding to each master surface design scheme, perform curing deformation simulation on the shape of the composite component, obtain the cured deformed shape corresponding to the shape of the composite component, and establish a curing deformation data set;
[0041] Firstly, 200 sets of template surface design schemes are obtained by random sampling in the design space, and the shape of the composite component corresponding to each template surface design scheme is obtained;
[0042] Secondly, COMSOL software was used to establish a composite component curing deformation simulation model for simulation. The curing deformation simulation model geometry was set to the U-shaped composite component of this embodiment, the physical field was set to the curing heat transfer and linear elastic deformation during the curing process of the composite component, and the mesh was divided into quadrilateral meshes with a mesh node number of . Write a MATLAB subroutine to call the curing deformation simulation model to simulate the shape of each composite component, and obtain the corresponding curing deformation shape of each composite component.
[0043] Finally, the coordinates of the quadrilateral grid are used as the data discrete form of the composite component shape and the shape after solidification deformation to derive the shape of each composite component. And the corresponding shape after curing deformation , construct a deformation data set with the shape of the composite component and the shape after curing deformation as data pairs .
[0044] Step 4: Establish a manifold neural operator model with the shape of the composite component as input and the shape after curing deformation as output, and use the curing deformation data set for training to obtain a curing deformation prediction agent model;
[0045] First, a manifold neural operator model is established. A fully connected neural network is used to construct a feature dimensionality mapping module, and the input composite component shape is upgraded from 3 channels to 64 channels. The Laplace operator is defined according to the discrete form of the composite component shape and the shape after curing deformation. The eigenvalue problem of the Laplace operator is solved to obtain the Laplace operator characteristic function. The encoding module is constructed using the Laplace operator characteristic function. The fully connected layer of the CNN constructs a parameterized module, the Laplace operator function is used to construct the decoding module, and the convolutional layer for linear transformation and the CNN for nonlinear activation are combined. Activation function, constructing Laplace kernel integration module. This embodiment uses 4 Laplace kernel integration modules. Use the fully connected layer to construct the feature dimension reduction mapping module, and restore the data passing through the Laplace kernel integration module from 64 channels to 3 channels.
[0046] Secondly, the solidification deformation dataset is used to train the manifold neural operator model to obtain the solidification deformation prediction agent model. The 200 sets of data in the solidification deformation dataset are divided into training sets and test sets. The training set contains 190 sets of data for training model parameters, and the test set contains 10 sets of data for testing model performance. The Adam optimizer is used to train the model, the learning rate is set to 0.01, the number of iterations is 1000, and the loss function is defined as the relative value of the manifold neural operator model's prediction value of the sample's solidification deformation shape and the sample label value. Error, by solving the empirical risk minimization problem of the loss function, training the manifold neural operator model, and obtaining the solidification deformation prediction agent model.
[0047] Step 5: Taking minimizing the difference between the shape after curing deformation and the target shape as the optimization goal, the intelligent optimization algorithm is used to optimize the design variables. During the optimization process, the curing deformation prediction agent model is used to calculate the shape after curing deformation, and the shape of the composite component that meets the curing deformation requirements of the composite component is obtained, and then the corresponding master surface is used as the molding mold surface.
[0048] In this embodiment, the difference between the shape after solidification deformation and the target shape is reflected by the maximum deformation between the shape after solidification deformation and the target shape, and the optimization goal is to minimize the maximum deformation between the shape after solidification deformation and the target shape. The design variables chamfer and offset of the U-shaped part backing surface are used as optimization variables. For the backing surface design scheme in the design space, the shape of the composite component of the backing surface design scheme is obtained, and the shape after solidification deformation is calculated using the solidification deformation prediction agent model, and the maximum deformation between the shape after solidification deformation and the target shape is calculated. Genetic algorithm is used for optimization, and the number of iterations is set to 20, the number of backing surface design schemes in each round of iteration is 20, and the mutation rate of each round of iteration is randomly generated in the range of 0-0.3.
[0049] The parameters of the template surface before and after optimization and the maximum deformation results of the shape after solidification deformation and the target shape are shown in Table 1. The results show that the maximum deformation after optimization is reduced by 44.16% compared with that before optimization, which proves the effectiveness of the method of the present invention.
[0050] Table 1 Parameters of the template surface and the maximum deformation results before and after optimization
[0051] .
[0052] Therefore, the present invention adopts the above-mentioned composite component forming mold compensation method based on manifold neural operator, establishes a prediction model of the shape of the composite component to the shape after curing deformation based on the manifold neural operator, and combines it with the intelligent optimization algorithm, with minimizing the difference between the shape after curing deformation and the target shape as the optimization goal, iteratively searches in the design space to obtain the shape of the composite component that meets the curing deformation requirements of the composite component, and then uses its corresponding mastering surface as the forming mold surface, thereby improving the mold compensation method for the curing deformation control effect when the shape of the composite component is complex and reducing the time cost required for mold design.
[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A composite component forming die compensation method based on manifold neural operator, characterized in that: The following steps are involved: Step 1: obtaining a three-dimensional model of the composite component as a target shape, and determining a mastering surface during curing and molding of the composite component according to the three-dimensional model; Step 2, parameterize the template surface to determine the design variables and their value ranges; Step 3, obtaining a number of master surface design schemes in the design space by a sampling method, obtaining the shape of the composite component corresponding to each master surface design scheme, performing a curing deformation simulation on the shape of the composite component, obtaining the cured deformed shape corresponding to the shape of the composite component, and establishing a curing deformation data set; Step 4: Establish a manifold neural operator model with the shape of the composite component as input and the shape after curing deformation as output, and use the curing deformation data set for training to obtain a curing deformation prediction agent model; Step 5: Taking minimizing the difference between the shape after curing deformation and the target shape as the optimization goal, the design variables are optimized using an intelligent optimization algorithm. During the optimization process, a curing deformation prediction agent model is used to calculate the shape after curing deformation, and the shape of the composite component that meets the curing deformation requirements of the composite component is obtained, and the corresponding mastering surface is used as the molding mold surface; The manifold neural operator model in step 4 is composed of a feature dimensionality increase mapping module, one or more Laplace kernel integration modules and a feature dimensionality reduction mapping module, wherein the Laplace kernel integration module includes three modules: frequency domain transformation, linear transformation and nonlinear activation; the frequency domain transformation module includes three modules: encoding, parameterization and decoding; the encoding module uses the Laplace operator characteristic function to convert the function defined on the composite component into coordinates in the frequency domain space; the parameterization module maps the encoded coordinates; the decoding module uses the Laplace operator characteristic function to restore the parameterized frequency domain space coordinates to the function defined on the composite component; The Laplace operator characteristic function is obtained by defining the Laplace operator in a discrete form according to the data of the shape of the composite component and the shape after solidification deformation, solving the eigenvalue problem of the Laplace operator, and obtaining the Laplace operator characteristic function.
2. The method for compensating a composite component forming die based on a manifold neural operator according to claim 1, characterized in that: In step 2, one of the following methods is used to parameterize the template surface: parameterize the template surface based on Bessel function; parameterize the template surface based on geometric parameters such as angle, length, and point coordinates; parameterize the template surface based on free deformation technology.
3. The method for compensating a composite component forming die based on a manifold neural operator according to claim 2, characterized in that: The sampling method in step 3 is random sampling or Latin hypercube sampling.
4. The method for compensating a composite component forming die based on a manifold neural operator according to claim 3, characterized in that: The curing deformation simulation of the shape of the composite component in step 3 specifically includes: according to the curing process in the actual molding process, using the finite element method, the finite difference method or commercial simulation software to simulate the curing deformation of the shape of the composite component.
5. A composite component forming die repair method based on manifold neural operator according to claim 4 The compensation method is characterized by: Each data in the solidification deformation data set in step 3 consists of the shape of the composite component and the corresponding shape after solidification deformation. Specifically, the shape of the composite component and the shape after solidification deformation are discretized into a grid or point cloud, and the coordinate field of the nodes on the shape is used as the data representation of the shape.
6. The method for compensating a composite component forming die based on a manifold neural operator according to claim 5, characterized in that: The specific calculation method of the difference between the solidified deformed shape and the target shape in step 5 includes: calculating the maximum deformation between the solidified deformed shape and the target shape; taking multiple points on the solidified deformed shape and the target shape, and calculating the average value, mean square error, root mean square error or relative error of the deformation of the multiple points; calculating the difference between the angle or distance between the corresponding positions on the solidified deformed shape and the target shape.
7. The method for compensating a composite component forming die based on a manifold neural operator according to claim 6, characterized in that: The intelligent optimization algorithm in step 5 is one of a genetic algorithm, a Bayesian optimization algorithm, a particle swarm optimization algorithm, and an annealing algorithm.
Citation Information
Patent Citations
Adjustable mold for curing molding of composite material and compensation method
CN114889159A
Compensation method and system for forming die of composite material structural part
CN115438394A
Neural operator construction method for complex geometrical shape
CN116187386A
Composite material part curing state prediction method based on Laplace spectrum
CN116258072A
Underwater cavity noise control method and device and medium
CN117094219A
