Method for predicting curing residual stress by adopting intrinsic strain model considering temperature field
By considering the intrinsic strain model of the temperature field and neural network training, combined with finite element simulation, the problem of low efficiency in curing residual stress prediction of composite thick structural parts is solved, and efficient and accurate residual stress prediction is achieved.
Patent Information
- Application Number
- CN202510261444.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-06
- Publication Date
- 2025-06-27
AI Technical Summary
The prior art is inefficient in predicting the curing residual stress of composite thick structural parts, and the intrinsic strain model cannot accurately predict when considering the temperature field.
The intrinsic strain model considering the temperature field is adopted, and the temperature and curing degree time-varying curves of each layer of the composite material are obtained through thermochemical finite element simulation. Combined with neural network model training and finite element simulation, the residual stress distribution of composite material thick structural parts is predicted.
The efficiency and accuracy of curing residual stress prediction are improved, the requirements for computer hardware are reduced, and the viscoelastic constitutive model can be replaced and the curing process cycle can be optimized.
Smart Images
Figure CN120217757A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of prediction of curing residual stress of composite materials, and particularly relates to a method for predicting curing residual stress by using an eigenstrain model considering a temperature field. Background Art
[0002] Composite material structures are widely used in the fields of aerospace and civil industries due to their excellent properties such as high modulus, high specific strength, corrosion resistance, and fatigue resistance. For thermosetting composite materials, during the curing process, due to the mismatch between the thermal expansion coefficient and the curing shrinkage coefficient of the material, the interaction between the mold and the part, etc., residual stress is generated and accumulated inside the material. If the residual stress is not released, it will lead to a significant reduction in the strength of the part or a shortening of the fatigue life. In order to determine and reduce the residual stress of composite parts and shorten the R & D cycle of composite products, accurately and quickly predicting the residual stress of composite materials is one of the most important issues in the manufacturing process of composite parts.
[0003] Currently, the residual stress in composite materials is mainly predicted by numerical methods, and many reliable finite element models for predicting residual stress have been developed. The viscoelastic constitutive model considering curing shrinkage and stress relaxation is considered to be the most accurate residual stress prediction model, but it is also the most time-consuming model. The eigenstrain model is an effective solution to ensure prediction accuracy and minimize the time used, but it is also for thin structural parts, that is, when the temperature field is not considered. When facing thick structural parts with a temperature gradient, the eigenstrain model cannot accurately predict the residual stress. Since the recommended curing process cycle by the manufacturer is no longer applicable when forming thick composite structural parts, it is necessary to optimize the curing process cycle based on the residual stress distribution state in the thick structural parts. To solve the problem of low efficiency in optimizing the curing process cycle using the viscoelastic constitutive model, using an eigenstrain model considering the temperature field for predicting the residual stress of thick structural parts is an effective solution. Summary of the Invention
[0004] The purpose of the present invention is to provide a method for predicting curing residual stress by using an eigenstrain model considering a temperature field, effectively solving the problem of low efficiency in optimizing the curing process cycle using the viscoelastic constitutive model.
[0005] To solve the above technical problems, the technical solution adopted by the present invention is:
[0006] A method for predicting curing residual stress by using an eigenstrain model considering temperature field, comprising the following steps: S1. Create a composite thick plate according to the thickness and number of layers of a composite thick structural member with a complex geometric structure. Based on the curing process cycle recommended by the manufacturer, obtain the time-varying curves of the temperature and curing degree of each single layer of the composite thick plate through thermochemical finite element simulation; S2. Introduce the time-varying curve data of the temperature and curing degree of each single layer of the composite thick plate into the finite element model of the thin plate, calculate the residual stress and residual strain distributions of each thin plate, and predict the deformation field of each thin plate after demolding; S3. Based on the deformation inversion, calculate the eigenstrain of each single layer of the composite thick plate, construct a neural network mapping relationship between the temperature history and the eigenstrain, divide the training set and the test set, and complete the training of the neural network model; S4. Take the neural network model trained in step S3 as a pre-trained model. For the new curing process cycle designed during the optimization of the curing process cycle, perform thermochemical finite element simulation, obtain the time-varying curves of the temperature and curing degree of each single layer of the composite thick plate, select a part of the time-varying curve data of the temperature and curing degree of some single layers and introduce them into the finite element model of the thin plate, perform mechanical simulation of the thin plate and eigenstrain inversion, and construct a small sample data set; Use the small sample data set to fine-tune the pre-trained model, optimize the network parameters, and use the fine-tuned pre-trained model to predict the eigenstrain of all layers under the new curing process; S5. Establish a finite element model of the composite thick structural member, input the mechanical property parameters of the composite material and the eigenstrain of each single layer predicted in step S4 as pseudo-thermal strain parameters into the finite element model of the composite thick structural member, perform finite element simulation, and predict the residual stress distribution of the composite thick structural member after demolding.
[0007] Further, in step S1, use the curing process cycle recommended by the manufacturer to establish a thermochemical finite element model of the composite thick plate, input the thermochemical parameters and curing kinetic parameters of the composite material into the thermochemical finite element model, and obtain the time-varying curves of the temperature and curing degree of each layer of the composite material by solving the nonlinear heat transfer problem with curing heat release as the internal heat source.
[0008] Further, step S2 specifically includes: S21. Take a node from each layer of the composite thick plate along the thickness direction, and extract the time-varying curve data of the temperature and curing degree of each single layer in the simulation result of step S1.
[0009] S22. Establish a finite element model of the thin plate, input the mechanical parameters, thermal expansion coefficient and curing shrinkage coefficient of the composite material into the finite element model of the thin plate. At the same time, input the time-varying curve data of the temperature and curing degree of each single layer of the composite thick plate obtained in step S21 into the finite element model of the thin plate respectively, use the viscoelastic constitutive model to describe the stress-strain relationship of the thin plate, perform simulation analysis, and obtain the residual stress and residual strain distributions of each thin plate.
[0010] S23. Set the demolding boundary conditions for each thin flat plate element model in step S22, turn off geometric nonlinearity, and continue the simulation analysis to obtain the deformation field of each thin flat plate element after demolding.
[0011] Further, step S3 specifically includes: S31. Assume that the eigenstrain of the composite material in the 1 direction is 0, and the eigenstrain in the 3 direction is equal to the eigenstrain in the 2 direction. Based on the deformation field of each thin flat plate element in step S2, utilize the characteristic that the eigenstrain has a linear relationship with the flat plate deformation curvature to inversely obtain the eigenstrains of each layer of the thick flat plate element of the composite material. The 1 direction represents the direction along the composite material fiber, the 2 direction represents the direction perpendicular to the composite material fiber, and the 3 direction represents the composite material thickness direction.
[0012] S32. Build a fully connected neural network model with a five-layer network structure, including an input layer, three hidden layers, and an output layer; the input layer is set to a unified dimension using linear interpolation, and at the same time, the ReLU function is used as the activation function of the hidden layer. Use the temperature time-varying curve data of each layer of the thick flat plate element of the composite material as the input and the corresponding eigenstrain as the output to construct a training data set, train the neural network model, and establish a neural network mapping relationship between the temperature time-varying curve and the eigenstrain.
[0013] Further, in step S3, the data ratio of the training set to the test set is 4:1.
[0014] Compared with the prior art, the beneficial technical effects of the present invention are:
[0015] (1) The present invention proposes a method for considering the temperature field in the eigenstrain model, improves the eigenstrain model architecture, and effectively solves the problem that the existing eigenstrain model cannot accurately predict the curing residual stress of thick structural parts with temperature gradients. The present invention is an alternative model to the reliable viscoelastic constitutive model for residual stress prediction, and has a high consistency with the viscoelastic constitutive model in curing residual stress prediction. In addition, compared with the viscoelastic constitutive model, the present invention has higher efficiency in predicting curing residual stress, shorter time consumption, and lower requirements for computer hardware, effectively solving the problem of low efficiency in using the viscoelastic constitutive model to optimize the curing process cycle.
[0016] (2) The parameters of the eigenstrain model considering the temperature field proposed by the present invention are obtained through neural network training, with high reliability and good feasibility. Description of the Drawings
[0017] Figure 1It is the finite element model of the thermochemical simulation during the curing process and the curing process cycle recommended by the manufacturer. Among them, (a) represents the geometric model of the composite thick flat plate and the Invar flat plate model, (b) represents the model after mesh division, and (c) represents the curing process cycle recommended by the manufacturer.
[0018] Figure 2 It is an example diagram of the linear function between the eigenstrain in the 2 direction and the curvature.
[0019] Figure 3 It is the loss curves of the training set and the test set during the training of the fully connected neural network model built by the present invention.
[0020] Figure 4 It is the prediction result of the training set during the training of the fully connected neural network model built by the present invention.
[0021] Figure 5 It is the prediction result of the test set during the training of the fully connected neural network model built by the present invention.
[0022] Figure 6 It is the new curing process cycle used by the present invention.
[0023] Figure 7 It is the loss curves of the training set and the test set during the fine-tuning of the pre-trained model proposed by the present invention.
[0024] Figure 8 It is the prediction result of the training set during the fine-tuning of the pre-trained model proposed by the present invention.
[0025] Figure 9 It is the prediction result of the test set during the fine-tuning of the pre-trained model proposed by the present invention.
[0026] Figure 10 It is the schematic diagram of the finite element model of the Bouligand spiral structure.
[0027] Figure 11 It is the simulation results of the viscoelastic constitutive model, the eigenstrain model without considering the temperature field, and the eigenstrain model considering the temperature field proposed by the present invention. Among them, (a) represents the through-thickness residual stress / MPa of the viscoelastic constitutive model considering the temperature field, (b) represents the through-thickness residual stress / MPa of the eigenstrain model without considering the temperature field, (c) represents the through-thickness residual stress / MPa of the eigenstrain model considering the temperature field, (d) represents the error between (a) and (b) / %, and (e) represents the error between (a) and (c) / %. Detailed implementation manners
[0028] Example 1: In this example, the CCF800H / AC531 material system is selected, and an example of a unidirectional composite thick flat plate is used to elaborate in detail on the method of predicting curing residual stress using the eigenstrain model considering the temperature field in the present invention.
[0029] It includes the following steps: (1) Thermochemical simulation during the curing process.
[0030] Assume that the composite thick structural part is a 170-layer bionic Bouligand spiral structure flat plate, with an interlayer angle of 2° and a maximum thickness of 25 mm. Then, use abaqus software to establish a geometric model of a composite thick flat plate with a length and width of 10 mm each and a thickness of 25 mm (the single-layer thickness of the CCF800H / AC531 unidirectional prepreg is 0.144 mm, a total of 170 layers, and the fiber direction is the 1 direction), as well as an Invar flat plate model with a length and width of 10 mm each and a thickness of 15 mm, and apply time-temperature boundary conditions on the model surface using the curing process cycle recommended by the manufacturer, as shown in Figure 1 (a) and Figure 1 (c).
[0031] Perform mesh division on the geometric model. In the thickness direction, divide the mesh according to the number of composite material layers, as shown in Figure 1 (b). Input the thermochemical properties and curing kinetics properties of the composite material, and the thermal properties of Invar into the finite element software. By solving the nonlinear heat transfer problem with curing heat release as the internal heat source, obtain the time-varying curves of the temperature and degree of cure of each layer of the composite material.
[0032] (2) Extraction of single-layer temperature and degree of cure data.
[0033] Take one node for each layer of the composite thick flat plate in the thickness direction, and extract the time-varying curve data of the temperature and degree of cure of 170 layers in the simulation results of step (1).
[0034] (3) Analysis of residual stress and residual strain of the thin flat plate.
[0035] Establish a finite element model of the thin flat plate: the length and width are 200 mm, and the thickness is 0.576 mm (4 layers, ply layup is [902,02]). Input the mechanical parameters, thermal expansion coefficient, and curing shrinkage coefficient of the composite material into the finite element model of the thin flat plate. At the same time, use the time-varying curve data of the temperature and degree of cure of each layer of the composite material obtained in step (2) as the temperature boundary conditions for the entire model. Use the viscoelastic constitutive model to describe the stress-strain relationship of the thin flat plate, and perform simulation analysis to obtain the residual stress and residual strain distributions of each thin flat plate.
[0036] (4) Prediction of the demolding deformation field of the thin flat plate.
[0037] Set the demolding boundary conditions for the finite element models of each thin flat plate in step (3), turn off geometric nonlinearity, and continue the simulation analysis to obtain the deformation field of each thin flat plate after demolding.
[0038] (5) Inverse calculation of eigenstrain model parameters.
[0039] Assume that the eigenstrain of the composite material in the 1 direction is 0, and the eigenstrain in the 3 direction is equal to the eigenstrain in the 2 direction, where the 1 direction represents the direction along the composite material fiber, the 2 direction represents the direction perpendicular to the composite material fiber, and the 3 direction represents the composite material thickness direction.
[0040] Based on the deformation fields of each thin flat plate in step (4), according to the characteristic that the eigenstrain and curvature are linearly related, the eigenstrains in the 2 direction are taken as 0.004, 0.005, and 0.006 respectively to simulate and obtain the curvatures of the thin flat plates, and fit the linear function between the eigenstrain in the 2 direction and the curvature. As Figure 2 shown, inverse calculate the eigenstrains of each layer of the composite material thick flat plate.
[0041] (6) Construction and training of the neural network model.
[0042] Build a fully connected neural network model with a five-layer network structure, including an input layer, an output layer, and three hidden layers. The input layer is set to a unified dimension of 512 using linear interpolation, and the ReLU function is used as the activation function of the hidden layer.
[0043] Use the temperature time-varying curve data of each layer of the composite material thick flat plate as the input and the corresponding eigenstrain as the output to construct a training dataset (where the data ratio of the training set to the test set is 4:1; that is, 136 groups of data are used as the training set and 34 groups of data are used as the test set). Train the neural network model to establish the mapping relationship between the temperature time-varying curve and the eigenstrain. The loss curves of the training set and the test set, the prediction results of the training set, and the prediction results of the test set are respectively as Figure 3 、 Figure 4 and Figure 5 shown.
[0044] (7) Transfer learning and small sample prediction.
[0045] Use the neural network model trained in step (6) as the pre-trained model. In actual production, it is necessary to adjust and optimize the curing process cycle when forming composite material thick structural parts. Therefore, for the new curing process cycle designed during the optimization of the curing process cycle (such as Figure 6As shown in the figure, a thermochemical finite element simulation is carried out on the thick composite plate to obtain the time-varying curve data of the temperature of each single layer of the thick composite plate. 17 groups of data are selected from them for the mechanical simulation of the thin plate and the inversion of the eigenstrain. The 17 groups of temperature time-varying curves and eigenstrains are constructed into a small sample data set.
[0046] The pre-trained model is fine-tuned using the small sample data set to optimize the network parameters. The pre-trained model after fine-tuning is used to predict the eigenstrain of all layers under the new curing process. The loss curves of the training set and the test set, the prediction results of the training set and the prediction results of the test set are respectively as Figure 7 , Figure 8 and Figure 9 shown.
[0047] (8) Prediction of residual stress in thick structural components.
[0048] A finite element model of a 170-layer bionic Bouligand spiral structure of the composite material is established, and the interlayer angle is set to 2°, as Figure 10 shown. Since the mechanical property parameters and eigenstrain of the material are required when using the eigenstrain model to describe the stress-strain relationship of the material, the mechanical property parameters of the composite material and the eigenstrain of each single layer predicted in step (7) are used as pseudo-thermal strain parameters and input into the model for finite element simulation to predict the residual stress distribution after the thick composite structural component is demolded.
[0049] Furthermore, the thick structural component in step (8) is simulated using a viscoelastic constitutive model, an eigenstrain model without considering the temperature field, and an eigenstrain model considering the temperature field proposed in this embodiment. The new curing process cycle shown in Figure 6 is used, and the simulation results are as Figure 11 shown. Compared with the eigenstrain model without considering the temperature field, the eigenstrain model considering the temperature field improves the prediction consistency of the residual stress distribution between the eigenstrain model and the viscoelastic constitutive model by 50%. This proves that the eigenstrain model considering the temperature field proposed in this embodiment is an alternative model to the reliable viscoelastic constitutive model for residual stress prediction and has a high consistency with the viscoelastic constitutive model in curing residual stress prediction.
[0050] Table 1 shows the comparison of the time used for predicting the curing residual stress by the eigenstrain model considering the temperature field proposed in this embodiment and the time used for prediction by the viscoelastic constitutive model. The results show that the time used for prediction by the eigenstrain model considering the temperature field proposed in this embodiment is only 26% of the time used for prediction by the viscoelastic constitutive model. This proves that compared with the viscoelastic constitutive model, the eigenstrain model considering the temperature field proposed in this embodiment has higher efficiency in predicting the curing residual stress, shorter time used, and lower requirements for computer hardware.
[0051] Table 1 Time taken for the present embodiment and the viscoelastic constitutive model to predict curing residual stress
[0052] Viscoelastic constitutive model This embodiment Prediction time 540s 140s Proportion 100％ 26％
[0053] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions, or substitutions made by those skilled in the art within the scope of the essence of the present invention should also fall within the protection scope of the present invention.
Claims
1. A method for predicting curing residual stress using an intrinsic strain model taking into account a temperature field, characterized in that: The following steps are involved: S1. Create a composite thick flat plate according to the thickness and number of layers of a composite thick structural part with a complex geometric structure. Based on the curing process cycle recommended by the manufacturer, obtain the time-varying curves of the temperature and curing degree of each single layer of the composite thick flat plate through thermochemical finite element simulation; S2, introducing the temperature and curing degree time-varying curve data of each single layer of the composite thick flat plate into the finite element model of the thin flat plate, calculating the residual stress and residual strain distribution of each thin flat plate, and predicting the deformation field of each thin flat plate after demoulding; S3. Calculate the intrinsic strain of each single layer of the thick flat plate of the composite material based on deformation inversion, construct a neural network mapping relationship between temperature history and intrinsic strain, divide the training set and the test set, and complete the training of the neural network model; S4, using the neural network model trained in step S3 as a pre-trained model, and performing thermochemical finite element simulation on a new curing process cycle designed in the process of curing process cycle optimization, to obtain the temperature and curing degree time-varying curves of each single layer of the thick flat plate of the composite material, and selecting a portion of the temperature and curing degree time-varying curve data of the single layer to introduce into the finite element model of the thin flat plate, and performing mechanical simulation and intrinsic strain inversion of the thin flat plate, and constructing a small sample data set; The pre-trained model is fine-tuned using a small sample data set to optimize network parameters, and the fine-tuned pre-trained model is used to predict the intrinsic strain of all layers under the new curing process; S5. Establish a finite element model of a thick composite material structure, input the mechanical property parameters of the composite material and the intrinsic strain of each single layer predicted in step S4 as pseudo-thermal strain parameters into the finite element model of the thick composite material structure, perform finite element simulation, and predict the residual stress distribution of the thick composite material structure after demolding.
2. The method for predicting curing residual stress using an intrinsic strain model taking temperature field into consideration according to claim 1, characterized in that: In step S1, a thermochemical finite element model of a thick flat plate of a composite material is established using the curing process cycle recommended by the manufacturer, the thermochemical parameters and curing kinetic parameters of the composite material are input into the thermochemical finite element model, and the time-varying curves of the temperature and curing degree of each layer of the composite material are obtained by solving the nonlinear heat transfer problem with curing exotherm as the internal heat source.
3. The method for predicting curing residual stress using an intrinsic strain model taking temperature field into consideration according to claim 2, characterized in that: Step S2 specifically includes: S21, taking a node from each layer of the composite material thick flat plate along the thickness direction, and extracting the time-varying curve data of the temperature and curing degree of each single layer in the simulation result of step S1; S22, establishing a finite element model of a thin flat plate, inputting the mechanical parameters, thermal expansion coefficient and curing shrinkage coefficient of the composite material into the finite element model of the thin flat plate, and inputting the temperature and curing degree time-varying curve data of each single layer of the thick flat plate of the composite material obtained in step S21 into the finite element model of the thin flat plate, using a viscoelastic constitutive model to describe the stress-strain relationship of the thin flat plate, performing simulation analysis, and obtaining the residual stress and residual strain distribution of each thin flat plate; S23, setting demoulding boundary conditions for the finite element models of each thin flat plate component in step S22, turning off geometric nonlinearity, and continuing simulation analysis to obtain the deformation field of each thin flat plate component after demoulding.
4. The method for predicting curing residual stress using an intrinsic strain model taking temperature field into consideration according to claim 3, characterized in that: Step S3 specifically includes: S31, assuming that the intrinsic strain of the composite material in direction 1 is 0, and the intrinsic strain in direction 3 is equal to the intrinsic strain in direction 2, based on the deformation field of each thin flat plate member in step S2, using the characteristic that the intrinsic strain is linearly related to the plate deformation curvature, inversion is performed to obtain the intrinsic strain of each layer of the thick flat plate member of the composite material; Direction 1 represents the direction along the composite fiber, direction 2 represents the direction perpendicular to the composite fiber, and direction 3 represents the thickness direction of the composite. S32. Build a five-layer fully connected neural network model, including an input layer, three hidden layers and an output layer; The input layer is set to a unified dimension using linear interpolation, and the ReLU function is used as the activation function of the hidden layer. The temperature time-varying curve data of each layer of the thick flat plate of the composite material is taken as input, and the corresponding intrinsic strain is used as output. A training data set is constructed, and a neural network model is trained to establish a neural network mapping relationship between the temperature time-varying curve and the intrinsic strain.
5. The method for predicting curing residual stress using an intrinsic strain model taking temperature field into consideration according to claim 4, characterized in that: In step S3, the data ratio of the training set and the test set is 4:1.