A method for optimizing curing temperature process of composite components based on manifold neural operator
Through the method based on manifold neural operator, the temperature process parameter field of composite material components is constructed, which solves the problems of low computational efficiency and limited optimization upper limit in traditional optimization solutions, and realizes spatial distribution optimization of composite material components to ensure that Continuity and optimization efficiency of process parameters are achieved.
Patent Information
- Application Number
- CN202411249065.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-09-06
AI Technical Summary
The traditional temperature process optimization scheme has low calculation efficiency, limited optimization limit and high optimization difficulty, making it difficult to meet the temperature process curve optimization requirements in different regions during the curing process of composite material components, and the optimized process parameter field does not guarantee spatial continuity.
Using a method based on manifold neural operator, the process parameter field of the temperature process curve design variable is constructed by obtaining the Laplace operator feature function of composite material components, and combining numerical calculations and intelligent optimization algorithms to construct process parameter field to cure. The prediction model of the state field is optimized to achieve spatial distribution optimization of the temperature process curve.
The number of optimization variables is significantly reduced, the spatial continuity of process parameters is ensured, the efficiency and accuracy of temperature process optimization are improved, and the optimization requirements of multi-region temperature process curves during the curing process of composite material components can be efficiently met.
Smart Images

Figure CN118761298B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of temperature process optimization, and in particular relates to a method for optimizing the curing temperature process of a composite material component based on a manifold neural operator. Background Art
[0002] Composite materials have become important materials in the aerospace field due to their advantages such as light weight, high strength, heat resistance and corrosion resistance. The extensive application of composite materials is conducive to reducing aircraft weight, reducing fuel consumption and thus improving aircraft performance. In order to increase the application of composite materials in aircraft structures, composite components are required to have high geometric quality and stable mechanical properties, which puts higher requirements on the manufacturing process of composite materials.
[0003] Composite material manufacturing mainly includes shaping, curing, and processing steps. Curing refers to the process in which the resin undergoes a cross-linking reaction under high temperature and high pressure, which causes changes in morphology and mechanical properties. It is a key link in composite material molding. Temperature process curves are often used to control the temperature field during the curing process. Composite prepreg manufacturers will provide customers with a guiding temperature process curve, but in actual production and manufacturing, the guiding temperature process curve cannot well guarantee the curing quality of the component due to the influence of the component geometry and thickness. Therefore, iterative optimization of the temperature process curve is of great significance to ensure the curing quality of the parts. However, the traditional temperature process optimization scheme is based on numerical calculation methods, which has low calculation efficiency and is difficult to meet the efficiency requirements of a large number of iterative optimizations. Some studies have proposed using machine learning models to replace numerical calculation methods to improve the efficiency of iterative optimization. However, the above methods all assume that the temperature curing process of all spatial positions of the composite component is the same, so only one temperature process curve needs to be optimized. Some studies have found that by setting different temperature process curves (i.e., process parameter fields) at different positions, the curing molding quality of the composite can be better controlled. At this time, it is required to construct a mapping model from a high-dimensional process parameter field to a curing state field during the optimization process, and the model is suitable for components with complex geometric shapes. The manifold neural operator in deep learning can construct a mapping from field to field on a complex geometric domain, which provides a basis for optimizing the composite curing process parameter field. However, if the process parameter field is directly used as the optimization object, the optimization variable dimension is too high, the optimization is difficult, and the optimized process parameter field may not guarantee spatial continuity, making it difficult to apply in actual engineering production. Specifically, the main defects of the above two traditional optimization schemes are as follows:
[0004] ① For the use of numerical calculation methods or traditional machine learning models, after randomly generating multiple temperature process curves, the finite element method is used for numerical simulation or machine learning model prediction for each temperature process curve to obtain the distribution of curing states during the curing process, and then the intelligent optimization algorithm is used for optimization based on the curing process requirements as constraints to obtain the optimized temperature process curve. The main defects are as follows: The object of the curing temperature process optimization scheme is a temperature process curve, that is, it is assumed that the temperature process curve set when the component is cured is evenly distributed on the component. It is impossible to optimize the scenario of setting different temperature process curves in different areas during the component curing process, and the upper limit of the process optimization of curing molding is limited;
[0005] ② Direct optimization scheme based on manifold neural operator: After discretization of the part space, randomly generate process parameters at each position to obtain the process parameter field, build a prediction model from the process parameter field to the solidification state field based on the manifold neural operator, and then use the solidification process requirements as constraints to optimize using an intelligent optimization algorithm to obtain the optimized process parameter field. The main defects are as follows: the dimension of the optimization variable is too high, the optimization is difficult, and the process parameter field obtained by optimization may not guarantee spatial continuity, making it difficult to apply in actual engineering production. Summary of the invention
[0006] The purpose of the present invention is to provide a method for optimizing the curing temperature process of composite material components based on manifold neural operators, so as to solve the problems existing in the prior art of low computational efficiency, limited optimization upper limit, high optimization difficulty and difficulty in meeting the efficiency requirements of a large number of iterative optimizations.
[0007] To achieve the above object, the present invention provides a method for optimizing the curing temperature process of a composite material component based on a manifold neural operator, comprising the following steps:
[0008] Step 1: Obtain the Laplace operator characteristic function of the surface geometry of the composite component ;
[0009] Step 2: Before selection Characteristic function , obtained by sampling Dimension coefficient ,Will and Corresponding multiplication, based on the product Constructing the process parameter field of the temperature process curve design variables ;
[0010] Step 3, obtaining the solidification state fields corresponding to the multiple sets of process parameter fields by numerical calculation methods, and constructing a sample data set from the process parameter field to the solidification state field;
[0011] Step 4: construct a prediction model from process parameter field to solidification state field based on manifold neural operator, and train the model on the sample data set;
[0012] Step 5: Use the prediction model trained in step 4 as an efficient solver and use intelligent optimization algorithms to optimize the process according to the process optimization goals. Dimension coefficient Perform iterative solution and finally obtain the optimized process parameter field.
[0013] Preferably, the Laplace operator characteristic function in step 1 is to solve the Laplace operator on the manifold The characteristic function obtained by the characteristic equation ; where the Laplace operator on the manifold is expressed as a second-order differential term: ; Characteristic function To satisfy the characteristic equation The function of is the corresponding eigenvalue, represents a differential operator.
[0014] Preferably, the sampling method in step 2 is random sampling or Latin hypercube sampling within a pre-set range.
[0015] Preferably, the temperature process curve design variables in step 2 include one or more of the following variables: holding temperature, heating rate, holding time, heating time, cooling rate, cooling time.
[0016] Preferably, in step 2, based on the product Constructing the process parameter field of the temperature process curve design variables The process is as follows:
[0017] S21, yes Standardization , the calculation expression is as follows:
[0018] ;
[0019] In the formula, represents the mean, represents variance;
[0020] S22, based on Construct the process parameter field of the corresponding design variables , the calculation expression is as follows:
[0021] ;
[0022] In the formula, is the recommended process parameter value, function is linear or nonlinear.
[0023] Preferably, the numerical calculation method in step 3 is to simulate the curing deformation of the composite material component according to the curing reaction mechanism by using finite element, finite difference and numerical calculation methods of commercial simulation software to obtain the curing state field.
[0024] Preferably, the solidification state field is the state of the composite material during solidification, including temperature, solidification degree, stress, strain and deformation field.
[0025] Preferably, the prediction model constructed based on the manifold neural operator in step 4 is specifically: composed of one or more Laplace kernel integration modules based on the Laplace characteristic function Construct, the expression is as follows:
[0026] ;
[0027] in, is the input function, and is the feature mapping layer, represents the manifold neural operator, Indicates Layer Laplacian kernel integration module;
[0028] The Laplace kernel integral module is constructed by combining three submodules: frequency domain transformation, linear transformation or nonlinear activation. The combination method is: only including the frequency domain transformation submodule, including the frequency domain transformation and linear transformation submodule, including the frequency domain transformation and nonlinear activation submodule, including the frequency domain transformation, linear transformation and nonlinear activation submodule;
[0029] Get the frequency domain transform submodule ( ) The method is: including encoding , parameterization ,decoding The three steps are as follows: the encoding is performed using the obtained Laplacian operator characteristic function The module's input function Mapped to the frequency domain, is the number of nodes, is the solidified state dimension, and the coordinates of the input under the characteristic function are obtained ,in is the Laplace eigenfunction matrix The pseudo-inverse of , is the number of characteristic functions; parameterization is to use the parameterization matrix to linearly transform the obtained coordinates, expressed as: ,in represents the parameterized matrix; decoding is based on the Laplacian operator characteristic function The parameterized coordinates are restored to the original function space, expressed as .
[0030] Preferably, the intelligent optimization algorithm is one of a genetic algorithm, a Bayesian optimization algorithm, a particle swarm optimization algorithm, an annealing algorithm, and an expectation-maximization algorithm.
[0031] Preferably, the process optimization target is a state index of the composite material during the curing process, including one or more of the following: minimizing the curing cycle, maximizing the degree of curing, minimizing the thermal peak, minimizing the thermal hysteresis, minimizing the maximum curing deformation, and minimizing the average curing deformation.
[0032] Therefore, the present invention adopts the above-mentioned composite material component curing temperature process optimization method based on manifold neural operator, constructs the composite material curing temperature process parameter field based on the Laplace operator characteristic function, and uses The group characteristic function characterizes the process parameter field, which can realize the optimization of the temperature process curve considering the spatial distribution, and significantly reduces the number of optimization variables while ensuring the continuity of the process parameters in space; combined with the manifold neural operator, an operator mapping from the process parameter field to the solidification state field is constructed, so that the optimization process can be carried out efficiently and accurately.
[0033] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 It is an overall flow chart of a composite material component curing temperature process optimization method based on manifold neural operator of the present invention;
[0035] Figure 2 A simulation model of the flat plate after meshing analyzed in the embodiment of the present invention;
[0036] Figure 3 The temperature process curve diagram used in the embodiment of the present invention is obtained;
[0037] Figure 4 It is a schematic diagram of the geometric Laplace operator characteristic function of the flat plate surface calculated according to an embodiment of the present invention;
[0038] Figure 5 is a flow chart of the optimization process of an embodiment of the present invention;
[0039] Figure 6 It is a comparison diagram of the curing deformation of the flat plate before and after optimization in the simulation software according to the embodiment of the present invention. DETAILED DESCRIPTION
[0040] 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.
[0041] See also Figure 1-6 , a method for optimizing the curing temperature process of composite material components based on manifold neural operator, comprising the following steps:
[0042] Step 1: Obtain the Laplace operator characteristic function of the surface geometry of the composite component ; The Laplace operator characteristic function is to solve the Laplace operator on the manifold The characteristic function obtained by the characteristic equation ; where the Laplace operator on the manifold is expressed as a second-order differential term: ; Characteristic function To satisfy the characteristic equation The function of is the corresponding eigenvalue, represents a differential operator.
[0043] Step 2: Before selection Characteristic function , obtained by random sampling or Latin hypercube sampling Dimension coefficient ,Will and Corresponding multiplication, based on the product Constructing the process parameter field of the temperature process curve design variables ; Wherein the temperature process curve design variables include one or more of the following variables: holding temperature, heating rate, holding time, heating time, cooling rate, cooling time;
[0044] Based on the product Constructing the process parameter field of the temperature process curve design variables The process is as follows:
[0045] S21, yes Standardization , the calculation expression is as follows:
[0046] ;
[0047] In the formula, represents the mean, represents variance;
[0048] S22, based on Construct the process parameter field of the corresponding design variables , the calculation expression is as follows:
[0049] ;
[0050] In the formula, is the recommended process parameter value, function is linear or nonlinear.
[0051] Step 3, obtain the curing state fields corresponding to multiple groups of process parameter fields through numerical calculation methods, and construct a sample data set from the process parameter field to the curing state field; the numerical calculation method is specifically based on the curing reaction mechanism, through finite element, finite difference, and numerical simulation methods of commercial simulation software to perform curing simulation on the composite material component to obtain the curing state field; the curing state field is the state of the composite material during the curing process, including temperature, curing degree, stress, strain and deformation field.
[0052] Step 4: Construct a prediction model from process parameter field to solidification state field based on manifold neural operator, and train the model on the sample data set; the constructed prediction model is specifically composed of one or more Laplace kernel integration modules based on Laplace characteristic function Construct, the expression is as follows:
[0053] ;
[0054] in, is the input function, and is the feature mapping layer, represents the manifold neural operator, Indicates Layer Laplace kernel integration module;
[0055] The Laplace kernel integral module is constructed by combining three submodules: frequency domain transformation, linear transformation or nonlinear activation. The combination method is: only including the frequency domain transformation submodule, including the frequency domain transformation and linear transformation submodule, including the frequency domain transformation and nonlinear activation submodule, including the frequency domain transformation, linear transformation and nonlinear activation submodule;
[0056] Get the frequency domain transform submodule ( ) The method is: including encoding , parameterization ,decoding The three steps are as follows: the encoding is performed using the obtained Laplacian operator characteristic function The module's input function Mapped to the frequency domain, is the number of nodes, is the solidified state dimension, and the coordinates of the input under the characteristic function are obtained ,in is the Laplace eigenfunction matrix The pseudo-inverse of , is the number of characteristic functions; parameterization is to use the parameterization matrix to linearly transform the obtained coordinates, expressed as: ,in represents the parameterized matrix; decoding is based on the Laplacian operator characteristic function The parameterized coordinates are restored to the original function space, expressed as .
[0057] Step 5: Use the prediction model trained in step 4 as an efficient solver and use intelligent optimization algorithms to optimize the process according to the process optimization goals. Dimension coefficient Iterate the solution and finally obtain the optimized process parameter field. The intelligent optimization algorithm is one of genetic algorithm, Bayesian optimization, particle swarm optimization algorithm, annealing algorithm and expected maximum algorithm. The process optimization goal is the state quantity of the composite material during the curing process, including one or more of the shortest curing cycle, maximizing the degree of curing, minimizing the thermal peak, minimizing the thermal hysteresis, minimizing the maximum curing deformation and minimizing the average curing deformation.
[0058] Example
[0059] This embodiment focuses on the Figure 2 The curing temperature process optimization of the flat plate shown in the figure, in this embodiment, the temperature process parameter fields are set for the upper and lower surfaces of the flat plate respectively, and then the temperature process parameter fields of the upper and lower surfaces are optimized based on the manifold neural operator to obtain the temperature process parameter field that meets the requirements, which specifically includes the following steps:
[0060] Step 1: First, obtain the Laplace operator characteristic function of the surface geometry of the composite component .
[0061] In this embodiment, firstly Figure 2 The flat plate shown in the figure is meshed. After meshing, the part has 5202 nodes and 2500 units. Then the .stl file containing the mesh node and unit information is exported. The Laplace operator of the surface geometry is constructed to solve the characteristic equation to obtain the Laplace operator characteristic function. ,like Figure 4 shown.
[0062] Step 2: Before selecting Characteristic function , obtained by sampling Dimension coefficient ,Will and Corresponding multiplication, based on the product Constructing the process parameter field of the temperature process curve design variables .
[0063] In this embodiment, the temperature process curve is set as follows: Figure 3 The single holding temperature process curve shown in the figure, that is, the curing temperature process only includes three stages in sequence: heating-holding-cooling. The design variable of the temperature process curve is the holding temperature. , other parameters such as heating time , insulation time , Cooling time The temperature process parameters set on the upper and lower surfaces are set to the same difference at the corresponding nodes. , is a randomly selected integer in the interval [-20, 20]. The first 128 characteristic functions are selected and 600 sets of 128-dimensional coefficients randomly selected in the interval [-5, 5] Corresponding multiplication, product Standardization . Then construct 600 sets of insulation temperature parameter fields ,function Use linear function ,in , , .
[0064] Step 3: Obtain the solidification state field corresponding to the process parameter field through numerical calculation methods, and build a sample database from the process parameter field to the solidification state field.
[0065] In this embodiment, the curing deformation field corresponding to the process parameter field is obtained based on the numerical simulation finite element software ABAQUS combined with secondary development batch simulation, and 600 sets of sample data sets from process parameter field to curing deformation field are constructed for subsequent prediction model training.
[0066] Step 4: Build a prediction model from process parameter field to solidification state field based on manifold neural operator, and train the model on the sample data set.
[0067] In this embodiment, a four-layer iterative kernel integration module is constructed based on the Laplace operator characteristic function obtained in step 1, and each layer of the kernel integration module includes a frequency domain transformation, a linear transformation and a nonlinear activation module.
[0068] In the frequency domain transformation, the feature function is truncated at a high frequency of 128 to remove the high-frequency errors in the data. The 4-layer iterative kernel integration module combines the feature mapping layer at the input and output ends to complete the construction of the manifold neural operator and establish an operator mapping model from the process parameter field to the solidification deformation field. 500 groups are randomly selected from the sample database obtained in step 3 as the model training set, and the model is trained using the classic neural network training method. After 2000 iterations, the test error on the 100 test sets is 1.35% (relative to error), thereby achieving rapid forward inference from process parameter field to solidification deformation field.
[0069] Step 5: Use the prediction model constructed in step 4 as an efficient solver and use intelligent optimization algorithms to optimize the process according to the process optimization goals. Dimension coefficient Perform efficient iterative solution and finally obtain the process parameter field that meets the process requirements.
[0070] In this embodiment, the intelligent optimization algorithm used is a genetic algorithm, which iteratively optimizes the coefficients with the optimization goal of minimizing the average curing deformation. In this process, the optimization variable is a 128-dimensional coefficient vector And the corresponding temperature difference The 129-dimensional parameter vector The initial population of the genetic algorithm is 100, and the optimization is terminated after 100 generations of iteration to obtain the optimized parameter vector , and the corresponding curing temperature parameter field is obtained. The optimization process is as follows Figure 5 The results before and after optimization are shown in Figure 6 As shown in the figure, the maximum curing deformation of the flat plate cured with the recommended temperature process curve before optimization is 1.176 The average curing deformation is 0.402 After optimization, the same flat plate is cured using the optimized temperature process parameters, and the maximum curing deformation is 0.451 , which is 62% lower than before optimization, and the average curing deformation is 0.146 , which is only 0.36 times of that before optimization.
[0071] Therefore, the present invention adopts the above-mentioned composite material component curing temperature process optimization method based on manifold neural operator, constructs the composite material curing temperature process parameter field based on the Laplace operator characteristic function, and uses The group characteristic function characterizes the process parameter field, which can realize the optimization of the temperature process curve considering the spatial distribution, and significantly reduces the number of optimization variables while ensuring the continuity of the process parameters in space. At the same time, the manifold neural operator is combined to construct an operator mapping from the parameter field to the solidification state field, so that the optimization process can be carried out efficiently and accurately.
[0072] 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 method for optimizing the curing temperature process of composite material components based on manifold neural operator, characterized in that: The following steps are involved: Step 1: Obtain the Laplace operator characteristic function of the surface geometry of the composite component ; Step 2: Before selection Characteristic function , obtained by sampling Dimension coefficient ,Will and Corresponding multiplication, based on the product Constructing the process parameter field of the temperature process curve design variables ; Step 3, obtaining the solidification state fields corresponding to the multiple sets of process parameter fields by numerical calculation methods, and constructing a sample data set from the process parameter field to the solidification state field; Step 4: construct a prediction model from process parameter field to solidification state field based on manifold neural operator, and train the model on the sample data set; Step 5: Use the prediction model trained in step 4 as an efficient solver and use intelligent optimization algorithms to optimize the process according to the process optimization goals. Dimension coefficient Perform iterative solution and finally obtain the optimized process parameter field; In step 2, based on the product Constructing the process parameter field of the temperature process curve design variables The process is as follows: S21, yes Standardization , the calculation expression is as follows: ; In the formula, represents the mean, represents variance; S22, based on Construct the process parameter field of the corresponding design variables , the calculation expression is as follows: ; In the formula, is the recommended process parameter value, function is linear or nonlinear; The prediction model constructed based on the manifold neural operator in step 4 is specifically composed of one or more Laplace kernel integration modules based on the Laplace characteristic function. Construct, the expression is as follows: ; in, is the input function, is the feature mapping layer, represents the manifold neural operator, Indicates Layer Laplace kernel integration module; The Laplace kernel integral module is constructed by combining three submodules: frequency domain transformation, linear transformation or nonlinear activation. The combination method is: only including the frequency domain transformation submodule, including the frequency domain transformation and linear transformation submodule, including the frequency domain transformation and nonlinear activation submodule, including the frequency domain transformation, linear transformation and nonlinear activation submodule; Get the frequency domain transform submodule ( ) The method is: including encoding , parameterization ,decoding The three steps are as follows: the encoding is performed using the obtained Laplacian operator characteristic function The module's input function Mapped to the frequency domain, is the number of nodes, is the solidified state dimension, and the coordinates of the input under the characteristic function are obtained ,in is the Laplace eigenfunction matrix The pseudo-inverse of , is the number of characteristic functions; parameterization is to use the parameterization matrix to linearly transform the obtained coordinates, expressed as: ,in represents the parameterized matrix; decoding is based on the Laplacian operator characteristic function The parameterized coordinates are restored to the original function space, expressed as ; The intelligent optimization algorithm is one of a genetic algorithm, a Bayesian optimization algorithm, a particle swarm optimization algorithm, an annealing algorithm, and an expectation maximum algorithm; The process optimization goal is the state index of the composite material during the curing process, including one or more of the following: shortening the curing cycle, maximizing the degree of curing, minimizing the thermal peak, minimizing the thermal hysteresis, minimizing the maximum curing deformation, and minimizing the average curing deformation.
2. According to claim 1, a method for optimizing the curing temperature process of composite material components based on manifold neural operators is characterized in that: The characteristic function of the Laplace operator in step 1 is to solve the Laplace operator on the manifold The characteristic function obtained by the characteristic equation ; where the Laplace operator on the manifold is expressed as a second-order differential term: ; Characteristic function To satisfy the characteristic equation The function of is the corresponding eigenvalue, represents a differential operator.
3. The method for optimizing the curing temperature process of composite material components based on manifold neural operator according to claim 2 is characterized in that: The sampling method in step 2 is random sampling or Latin hypercube sampling within a pre-set range.
4. The method for optimizing the curing temperature process of composite material components based on manifold neural operator according to claim 3 is characterized in that: The temperature process curve design variables in step 2 include one or more of the following variables: holding temperature, heating rate, holding time, heating time, cooling rate, cooling time.
5. The method for optimizing the curing temperature process of composite material components based on manifold neural operator according to claim 4 is characterized in that: The numerical calculation method in step 3 is specifically to perform curing simulation on the composite material component according to the curing reaction mechanism through finite element, finite difference, and numerical calculation methods of commercial simulation software to obtain the curing state field.
6. The method for optimizing the curing temperature process of composite material components based on manifold neural operator according to claim 5 is characterized in that: The curing state refers to the state of the composite material during the curing process, including temperature, degree of curing, stress, strain and deformation field.
