A method for predicting curing temperature field of composite components based on manifold neural operator
Through the method based on manifold neural operator, the temperature field mapping model is constructed using spatial basis functions, which solves the problem of poor temperature field prediction caused by model architecture limitation in the prior art, and realizes the efficient temperature field prediction and the Process optimization.
Patent Information
- Application Number
- CN202411198806.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-12-24
- Estimated Expiration
- 2044-08-29
AI Technical Summary
In the prediction of composite curing temperature field, the prior art requires simplification of the input and output due to model architecture limitations, resulting in unsatisfactory prediction results and cannot effectively meet the efficiency requirements of iterative optimization.
Using a method based on manifold neural operator, the manifold neural operator model is constructed by obtaining the spatial basis function of the solidified temperature process data and the temperature field data, so as to realize the mapping from the temperature process to the temperature field, avoiding the simplified processing of input and output. , reduce the difficulty of modeling and improve the prediction effect.
Accurate prediction of the curing temperature field of composite material components is achieved, prediction efficiency and effect are improved, the curing process can be quickly optimized, and the occurrence of curing defects is reduced.
Smart Images
Figure CN118709577B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of composite material curing, and in particular to a method for predicting the curing temperature field of a composite material component based on a manifold neural operator. Background Art
[0002] Composite materials have the advantages of light weight and high strength, and have become the preferred material for weight reduction and efficiency improvement of high-end equipment in the fields of aerospace, automobiles, ships, etc. Composite material curing molding refers to the process of causing physical and chemical changes in the resin under a certain curing process to form composite parts with both geometric shape and load-bearing performance. Unreasonable curing process will lead to the formation of uneven temperature field, curing degree field and large curing residual stress inside the composite parts, thereby causing curing defects such as ablation, delamination, and deformation, greatly reducing the manufacturing qualification rate and molding quality of composite parts. Therefore, the curing process design process of composite parts often requires a large number of iterative optimizations based on the prediction results of the curing state such as temperature field, curing degree field, and stress field. The numerical simulation method represented by finite element has low computational efficiency and cannot meet the efficiency requirements of a large number of iterative optimizations. The data-driven method directly learns the mapping relationship between the curing temperature process and the temperature curing state from the curing data sample. Once the model is trained, it can quickly predict the curing state, which provides a feasible idea for iterative optimization of the curing temperature process curve of composite components.
[0003] The curing temperature field describes the evolution of temperature in time and space on composite components. The input curing temperature process of the curing temperature field prediction scenario is a time function, and the output curing temperature field is a time-space function. It is a typical operator mapping and belongs to a non-equivalent domain mapping. Existing technologies have simplified the input and output of the composite material curing temperature field prediction scenario to varying degrees. One method is to simplify the curing temperature process curve into a small number of process parameters (heating rate, cooling rate, insulation temperature, insulation time, etc.), and simplify the temperature field into a small number of statistical indicators (maximum temperature, maximum temperature difference, etc.), and then use neural network and other models for fitting; another method is to process the input and output into equal domain (same domain) functions, that is, the input and output functions are either time functions or space functions. The prior art can refer to the patent with announcement number CN114065578A, which discloses a composite material curing process modeling method, which only predicts the temperature evolution of a certain point on the component, thereby learning the operator mapping whose input and output are both time functions; for example, a composite material part curing state prediction method based on Laplace spectrum with announcement number CN116258072A only predicts the temperature distribution at a certain moment, and simplifies the input into spatially distributed curing process parameters, thereby learning the operator mapping whose input and output are both spatial functions.
[0004] The above technology does not design a new model architecture for the characteristics of the curing temperature field prediction task belonging to non-equivalent domain mapping, and can only simplify the input or output to adapt to the existing model architecture. In practical applications, it is necessary to build several parallel prediction models (such as the patent with announcement number CN114065578A needs to build a prediction model for each spatial position point; the patent with announcement number CN116258072A needs to build a prediction model for each moment), resulting in unsatisfactory prediction of the curing temperature field. In view of the above reasons, it is necessary for the present application to provide a method for predicting the curing temperature field of composite components based on manifold neural operators. Summary of the invention
[0005] The purpose of the present invention is to provide a method for predicting the curing temperature field of composite components based on manifold neural operators, so as to solve the problem that the existing data-driven methods in the above-mentioned background technology are limited by the existing model architecture and require simplified processing of the input and output, resulting in unsatisfactory prediction effect of the curing temperature field.
[0006] To achieve the above object, the present invention provides a method for predicting the curing temperature field of a composite material component based on a manifold neural operator, comprising the following steps:
[0007] S1. For a given composite component, obtain a set of curing temperature process data , curing temperature field data and a set of spatial basis functions for the curing temperature field data ,in, The curing temperature process data data, is the first data, Indicates the index of the data. represents the index of the spatial basis function, Indicates the number of data. represents the number of basis functions;
[0008] S2, using the spatial basis function obtained in step S1 and manifold neural operator models Construct a mapping model from curing temperature process to curing temperature field;
[0009] S3, using the curing temperature process data, curing temperature field data and spatial basis functions obtained in step S1 to train the mapping model constructed in step S2, to obtain a prediction model for the curing temperature field of the composite material component;
[0010] S4. For a new curing temperature process, the curing temperature field prediction model for the composite material component trained in step S3 is used to predict its corresponding curing temperature field.
[0011] Preferably, in step S1 The representation of is a time discrete vector of time function, which is ; The representation of is the space-time discrete matrix of the space-time function, which is ; The representation of is a spatial discrete vector of spatial function, which is ,in, and Discrete quantities representing the time dimension and space dimension respectively.
[0012] Preferably, the spatial basis functions in step S1 are basis functions related to the training data obtained based on a modal decomposition method, such as an intrinsic orthogonal decomposition method, or basis functions inherent in the spatial domain and independent of the training data, such as a Laplace Beltrami operator basis function.
[0013] Preferably, the architecture of the mapping model in step S2 is as follows: Input Manifold Neural Operator Model , and obtain a set of time weight functions , the time weight function The spatial basis function obtained in step S1 Multiply to get the output curing temperature field .
[0014] Preferably, the manifold neural operator model in step S2 Includes process for building curing temperatures To the time weight function Neural operator models that map between temporal functions, such as Fourier neural operators and wavelet neural operators.
[0015] Therefore, the present invention adopts the above-mentioned method for predicting the curing temperature field of composite components based on manifold neural operators, and uses a set of spatial basis functions and a manifold neural operator model that can be used to construct mappings between time functions to design a mapping model architecture for the prediction scenario of the curing temperature field of composite components. Compared with the existing technology, this architecture does not require simplified processing of the input or output; by introducing spatial basis functions, non-equivalent domain mappings are converted into equivalent domain mappings, so that the manifold neural operator model only needs to construct mappings between time functions, which reduces the difficulty of modeling and thereby improves the prediction effect.
[0016] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 is a flow chart of the present invention;
[0018] Figure 2 It is the curing temperature process curve diagram of the present invention;
[0019] Figure 3 is a schematic diagram of a composite wing leading edge component in an embodiment of the present invention;
[0020] Figure 4 Schematic diagram of a solidification temperature field prediction model in an embodiment of the present invention;
[0021] Figure 5 It is a prediction error distribution diagram of all nodes on all samples in the test set in the embodiment of the present invention. DETAILED DESCRIPTION
[0022] Example
[0023] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.
[0024] like Figure 1 As shown, a method for predicting the curing temperature field of a composite material component based on a manifold neural operator is used to realize the mapping of the curing temperature process to the spatiotemporal curing temperature field. The prediction method can be used for rapid prediction of the information of the curing process temperature, curing process optimization, and curing process temperature monitoring, etc., and includes the following steps:
[0025] S1. For a given composite component, obtain a set of curing temperature process data , curing temperature field data and a set of spatial basis functions for the curing temperature field data ,in, The curing temperature process data data, is the first data, Indicates the index of the data. represents the index of the spatial basis function, Indicates the number of data. represents the number of basis functions;
[0026] like Figure 2-Figure 3As shown, this embodiment uses a composite wing leading edge component and adopts zoned heating curing, that is, the inner and outer surfaces of the component are divided into 6 areas, and different curing temperature processes are applied to each area. In this embodiment, Python code is first used to randomly generate 500*6 groups of double insulation temperature process curves to obtain curing temperature process data. Then, the commercial multi-physics simulation software COMSOL Multiphysics is used to calculate the temperature field of the composite material component during the curing process corresponding to the curing temperature process, that is, to obtain the curing temperature field data. . The discrete number of time dimensions Take 151, the discrete number of spatial dimensions Take 2743.
[0027] In this embodiment, based on 500 sets of curing temperature field data, the intrinsic orthogonal decomposition method is used to obtain the first 32 modes in the spatial domain and use them as a set of spatial basis functions, that is, .
[0028] S2, using the spatial basis function obtained in step S1 and manifold neural operator models Construct a mapping model from curing temperature process to curing temperature field;
[0029] The manifold neural operator model used in this implementation is a Fourier neural operator, which includes 4 Laplace kernel integration modules. Each Laplace kernel integration module includes a frequency domain transformation submodule, a linear transformation submodule and a nonlinear activation submodule, wherein the implementation steps of the frequency domain transformation submodule are as follows: (1) using fast Fourier transform FFT to reduce the dimension of the submodule input function to obtain a coefficient vector with a dimension of 16; (2) performing a linear transformation on the 16-dimensional coefficient vector; (3) using inverse fast Fourier transform IFFT to increase the dimension of the coefficient vector after linear transformation to obtain the submodule output function. The nonlinear activation submodule uses the relu activation function. In this embodiment, in order to enhance the manifold neural operator model In order to improve the learning ability of the neural network, a fully connected neural network with 32 neurons is set before the four Laplace kernel integration modules, and three fully connected neural networks with 128, 128 and 32 neurons are set after the four kernel integration modules. The fully connected layers use the relu activation function. The output of is the time weight function.
[0030] The forward architecture of the mapping model is: Input Manifold Neural Operator Model , and get the time weight function , the weight function With spatial basis functions Multiply to get the predicted spatiotemporal curing temperature field .
[0031] S3, using the curing temperature process data, curing temperature field data and spatial basis function obtained in step S1 to train the mapping model constructed in step S2, and obtain a prediction model for the curing temperature field of the composite material component, such as Figure 4 As shown;
[0032] In this embodiment, 500 sets of curing temperature process and curing temperature field data pairs are used and the classic neural network training method is adopted. The curing temperature field prediction model is obtained by iterative training 500 times, where the batch size is 50; the initial learning rate is set to 0.01 (decayed by 50% every 100 iterations); the optimizer is Adam, and the loss function is defined as follows:
[0033] ;
[0034] in, is the number of training data, Indicates the curing temperature field label value, Represents the predicted value of the curing temperature field.
[0035] S4. For a new curing temperature process, the corresponding curing temperature field is predicted using the composite material component curing temperature field prediction model trained in step S3;
[0036] In this embodiment, 100 sets of curing temperature process and curing temperature field data pairs are generated as test data in the same manner as step S1. The error is only 0.257%. Figure 5 The prediction error distribution diagram of all nodes on all samples in the test set is given, and it can be seen that the prediction error (absolute value) of 94.17% of the nodes is less than 2 K. It can be seen that the method has the potential to accurately predict the curing temperature field of composite components.
[0037] Therefore, the present invention adopts the above-mentioned method for predicting the curing temperature field of composite components based on manifold neural operators, and designs a mapping model architecture for the prediction scenario of the curing temperature field of composite components by utilizing a set of spatial basis functions and a manifold neural operator model that can be used to construct mappings between time functions. The architecture is simple and direct, does not require simplified processing of the input or output, and can improve the prediction effect of the curing temperature field.
[0038] 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 predicting the curing temperature field of composite material components based on manifold neural operator, characterized in that: The following steps are involved: S1. For a given composite component, obtain a set of curing temperature process data , curing temperature field data and a set of spatial basis functions for the curing temperature field data ,in, and They are The curing temperature process and its corresponding A curing temperature field, Indicates the index of the data. represents the index of the spatial basis function, Indicates the number of data. represents the number of basis functions; S2, using the spatial basis function obtained in step S1 and manifold neural operator models Construct a mapping model from curing temperature process to curing temperature field; S3, using the curing temperature process data, curing temperature field data and spatial basis functions obtained in step S1 to train the mapping model constructed in step S2, to obtain a prediction model for the curing temperature field of the composite material component; S4. For a new curing temperature process, the corresponding curing temperature field is predicted using the composite material component curing temperature field prediction model trained in step S3; In step S1 The representation of is a time discrete vector of time function, which is ; The representation of is the space-time discrete matrix of the space-time function, which is ; The representation of is a spatial discrete vector of spatial function, which is ,in, and The discrete quantities representing the time dimension and the space dimension respectively; The spatial basis function in step S1 is a basis function related to the training data obtained based on the modal decomposition method or a basis function inherent in the spatial domain and unrelated to the training data; The architecture of the mapping model in step S2 is as follows: Input Manifold Neural Operator Model , and obtain a set of time weight functions , the time weight function The spatial basis function obtained in step S1 Multiply to get the output curing temperature field ; Manifold neural operator model in step S2 Includes process for building curing temperature To the time weight function A neural operator model for mapping between temporal functions.