Composite material curing deformation simulation model parameter identification method based on measured data

Abstract

The invention discloses a composite material curing deformation simulation model parameter identification method based on measured data, and relates to the field of composite material curing simulation, and the method comprises the specific steps: firstly, obtaining measured curing deformation data of a group of composite material parts under a to-be-identified material system; the method comprises the following steps: establishing a data-driven prediction model from parameters to be identified to curing deformation for each part by utilizing simulation data, and further taking the data-driven models as agent models of a curing deformation simulation model, and determining values of all the parameters to be identified through large-scale search iteration; enabling the difference between the proxy model prediction data and the actually measured solidification deformation data to meet a target set by a user; according to the method for identifying the parameters of the composite material curing deformation simulation model based on the actually measured data, the key parameters in the composite material curing deformation simulation model can be efficiently obtained by directly utilizing the actually measured data, so that the simulation precision of curing deformation of a composite material part is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of composite material curing simulation, and in particular to a composite material curing deformation simulation model parameter identification method based on measured data. Background Art

[0002] Composite materials are increasingly used in key components of high-end equipment due to their high specific modulus and specific strength, good fatigue resistance and corrosion resistance, and rich functional designability. Curing is a key link in the manufacturing process of composite parts. Composite parts will inevitably deform after demolding due to factors such as the thermal expansion and contraction effects of fibers and resins, the flow effect and chemical shrinkage effect of resins, and the interaction between molds and parts during the curing process. Curing deformation will increase the difficulty of subsequent assembly and accumulate assembly stress, thereby affecting the service performance of the structure and increasing the risk of failure. When the curing deformation is too large, the parts will be scrapped due to the inability to assemble.

[0003] Establishing a high-fidelity simulation model to accurately predict the curing deformation of composite parts, and then optimizing the controllable factors related to curing deformation in the design and process links based on the simulation prediction results, is a basic means of controlling curing deformation. The accuracy of the simulation model parameters is one of the main factors affecting the prediction accuracy of the simulation model. However, at this stage, model parameters are mainly obtained through individual measurements, but due to the large number, the measurement cycle is long and the cost is high. Most of the existing measurement methods are aimed at fibers, resins or prepregs alone, and the comprehensive parameters of the materials after prepreg stacking rely on theoretical approximations. In addition, some parameters currently lack accurate measurement methods and are mainly calculated based on assumptions based on experience. Therefore, there is an urgent need for a method that can accurately and efficiently obtain the parameters of the composite curing deformation simulation model. Summary of the invention

[0004] The purpose of the present invention is to provide a composite material curing deformation simulation model parameter identification method based on measured data, so as to solve the problem in the prior art that the composite material curing deformation simulation model parameters are difficult to obtain accurately and efficiently.

[0005] To achieve the above object, the present invention provides a composite material curing deformation simulation model parameter identification method based on measured data, comprising the following steps: Step 1: Obtain the material system to be identified A collection of measured curing deformation data of composite parts ; Step 2: Construct a collection of solidification deformation simulation models of all parts in step 1) , determine all parameters to be identified in the simulation model The value range is selected and sampling is performed within this range. Through the simulation model Calculate the composite part in step 1 In the corresponding A collection of solidification deformation prediction data under a set of parameters to be identified , construct the set of solidification deformation simulation data of all composite parts in step 1 ; Step 3: Using the data set in step 2 Training data-driven predictive models As the simulation model in step 2 Proxy model, construct the simulation model set in step 2 A collection of proxy models ; Step 4: Determine the corresponding values ​​for all parameters to be identified, so that the proxy model set in step 3 A collection of solidification deformation prediction data under its input The combination of the measured curing deformation data in step 1 The gap between them meets the user's pre-set goal.

[0006] Preferably, the curing deformation data in step 1 is index data or deformation field data reflecting the difference between the measured shape and the theoretical shape of the deformed part; wherein the index data includes one or more of the maximum warpage, the maximum deformation, and the springback angle.

[0007] Preferably, the parameters to be identified in step 2 are selected from parameters that are difficult to accurately measure in the curing deformation simulation model and are more sensitive to curing deformation, and the value range is reasonably determined based on the physical meaning of the parameters, preliminary measurement results or parameter values ​​of similar material systems.

[0008] Preferably, the type of data-driven prediction model in step 3 depends on the type of solidified deformation data in step 1. When the solidified deformation is indicator data, the data-driven model is a fully connected neural network, a decision tree model, a support vector machine, or a Gaussian process regression model; when the solidified deformation is deformation field data, the data-driven model includes various types of reduced-order models, convolutional neural networks, recurrent neural networks, graph neural networks, Transformer models, and neural operator models suitable for high-dimensional field data prediction.

[0009] Preferably, in step 4, corresponding values ​​are determined for all parameters to be identified by using a non-iterative method such as Monte Carlo search or enumeration or an iterative optimization method such as gradient descent or genetic algorithm.

[0010] Preferably, the gap between the set of predicted solidification deformation data of the proxy model set under its input and the set of measured solidification deformation data in step 4 is obtained by adopting a multi-objective combination method including average percentage absolute error and maximum percentage absolute error; wherein, when the deformation field is used to characterize the deformation, it is also necessary to introduce an indicator characterizing the consistency of the distribution of the predicted deformation field data and the distribution of the measured deformation field data into the optimization objective, which is obtained by adopting KL divergence, Wasserstein distance, and mutual information.

[0011] Therefore, the present invention adopts the above-mentioned composite material curing deformation simulation model parameter identification method based on measured data, directly introduces the measured curing deformation data as the basis for parameter identification, and performs large-scale search iterations on the parameters to be identified, and only uses the curing deformation data of a small number of parts to significantly improve the prediction accuracy of the simulation model; in addition, the present invention establishes a data-driven prediction model from the parameters to be identified to the curing deformation as a simulation proxy model for parameter identification, and the computational efficiency of a single iteration can be improved by several orders of magnitude, thereby making a large number of search iterations possible and ensuring the final identification effect.

[0012] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is an overall flow chart of a composite material curing deformation simulation model parameter identification method based on measured data of the present invention; Figure 2 : C / U / V-shaped part simulation model diagrams of embodiments of the present invention, wherein (a) is a C-shaped part simulation model diagram, (b) is a U-shaped part simulation model diagram, and (c) is a V-shaped part simulation model diagram; Figure 3 1 is a C / U / V-shaped part parameter-curing deformation sensitivity analysis result diagram of an embodiment of the present invention, wherein (a) is the sensitivity analysis result of the C-shaped part under zero-degree ply, (b) is the sensitivity analysis result of the U-shaped part under zero-degree ply, (c) is the sensitivity analysis result of the V-shaped part under zero-degree ply, (d) is the sensitivity analysis result of the C-shaped part under 45-degree orthogonal ply, (e) is the sensitivity analysis result of the U-shaped part under 45-degree ply, and (f) is the sensitivity analysis result of the V-shaped part under 45-degree orthogonal ply; Figure 4 A schematic diagram of a neural network proxy model according to an embodiment of the present invention; Figure 5 This is a diagram showing the curing deformation simulation test results of the Z-shaped part according to an embodiment of the present invention. DETAILED DESCRIPTION

[0014] 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.

[0015] See also Figure 1-Figure 5 A composite material curing deformation simulation model parameter identification method based on measured data comprises the following steps: Step 1: Obtain the material system to be identified A collection of measured curing deformation data of composite parts ; Wherein, the solidification deformation data is index data or deformation field data reflecting the difference between the measured shape of the deformed part and the theoretical shape; wherein the index data includes one or more of the maximum warpage, the maximum deformation, and the springback angle; Step 2: Construct a collection of solidification deformation simulation models of all parts in step 1) , determine all parameters to be identified in the simulation model The value range is selected and sampling is performed within this range. Through the simulation model Calculate the composite part in step 1 In the corresponding A collection of solidification deformation prediction data under a set of parameters to be identified , construct the set of solidification deformation simulation data of all composite parts in step 1 Among them, the parameters to be identified are those that are difficult to measure accurately and are more sensitive to the solidification deformation in the solidification deformation simulation model, and the value range is reasonably determined according to the physical meaning of the parameters, preliminary measurement results or parameter values ​​of similar material systems.

[0016] Step 3: Using the data set in step 2 Training data-driven predictive models As the simulation model in step 2 Proxy model, construct the simulation model set in step 2 A collection of proxy models ; Among them, the type of data-driven prediction model depends on the type of solidified deformation data in step 1. When the solidified deformation is indicator data, the data-driven model is a fully connected neural network, a decision tree model, a support vector machine, and a Gaussian process regression model; when the solidified deformation is deformation field data, the data-driven model includes various types of reduced-order models, convolutional neural networks, recurrent neural networks, graph neural networks, Transformer models, and neural operator models suitable for high-dimensional field data prediction.

[0017] Step 4: Determine the corresponding values ​​for all parameters to be identified, so that the proxy model set in step 3 A collection of solidification deformation prediction data under its input The combination of the measured curing deformation data in step 1 The gap between them meets the goal set in advance by the user. Among them, the corresponding values ​​of all parameters to be identified are determined by using the non-iterative method of Monte Carlo search and enumeration method or the iterative optimization method of gradient descent and genetic algorithm; in addition, the gap between the set of solidification deformation prediction data under the input of the proxy model set and the set of measured solidification deformation data is obtained by using a multi-objective combination method including the average percentage absolute error and the maximum percentage absolute error; among them, when the deformation field is used to characterize the deformation, it is also necessary to introduce an indicator that characterizes the consistency of the distribution of the predicted deformation field data and the distribution of the measured deformation field data in the optimization target, which is obtained by using KL divergence, Wasserstein distance, and mutual information.

[0018] Example This implementation case takes a certain material system as the research object and conducts a 0-degree ply ( ) and 45 degree cross ply ( ) were used to perform curing tests on typical C / U / V composite specimens with a thickness of 3 mm and measure their maximum springback deformation; a curing deformation simulation model of the above composite parts group was established, such as Figure 2 As shown, (a) is the simulation model diagram of C-type parts, (b) is the simulation model diagram of U-type parts, and (c) is the simulation model diagram of V-type parts; global sensitivity analysis of model parameters is performed to determine the identification parameter group; a neural network proxy model from the identification parameter group to the curing deformation is established; a multi-objective genetic algorithm is used to identify parameters using the neural network proxy model and the actual curing deformation data of C / U / V-type composite parts under two layups. Three curing experiments were carried out on C / U / V-type parts under two layups, and the maximum rebound deformation of the mold surface was measured. Finally, the average of the three measurement results was taken, as shown in Table 1. Among them, 0 represents zero-degree layup and 1 represents 45-degree orthogonal layup.

[0019] Table 1 Measured data of curing deformation of C / U / V-shaped parts under two types of layups ;

[0020] The 12 parameters in the curing deformation simulation model were preliminarily screened and the Sobol method was used to perform a global sensitivity analysis of the curing deformation. The 12 parameters are resin rubber modulus Er0, resin glass modulus Er00, gel point curing degree alphac1, glass transition point curing degree alphac2, resin Poisson's ratio vr, ratio of thermal expansion coefficients of rubber resin and glass resin a, thermal expansion coefficient of glass resin AAmg, volume shrinkage rate of viscous flow resin curing chemr, ratio of volume shrinkage rate of glass resin and rubber resin curing b, fiber axial thermal expansion coefficient AA1f, fiber radial thermal expansion coefficient AA2f and mold-part contact friction coefficient Friction. The results of the sensitivity analysis of C / U / V-shaped parts parameters under two layups are shown in the figure. Figure 3 As shown, (a) is the sensitivity analysis result of C-shaped parts under zero-degree ply, (b) is the sensitivity analysis result of U-shaped parts under zero-degree ply, (c) is the sensitivity analysis result of V-shaped parts under zero-degree ply, (d) is the sensitivity analysis result of C-shaped parts under 45-degree orthogonal ply, (e) is the sensitivity analysis result of U-shaped parts under 45-degree orthogonal ply, and (f) is the sensitivity analysis result of V-shaped parts under 45-degree orthogonal ply. Figure 3 From (a)-(e), we can see that there are 8 parameters that have a great influence on the solidification deformation, namely Er0, Er00, alphac2, AAmg, chemr, b, AA2f, and Friction. Therefore, the above 8 parameters are determined as the parameters to be identified. The above parameters of similar material systems are taken as the intermediate values, and the range of ±40% is set as its value range on this basis, and the Latin hypercube algorithm is used for sampling.

[0021] The data set of the data-driven prediction model is constructed by batch simulation of C / U / V-shaped parts under two types of plies. The training set is 80 sets of samples generated by 8 parameters in the parameter space, the test set is 30 sets of samples generated by 8 parameters in the parameter space, and the validation set is 10 sets of samples generated by 8 parameters in the parameter space. Figure 4 The fully connected neural network shown in Figure 1 is used as a data-driven proxy prediction model. The input of the network model is the identification parameter group, and the output is the maximum rebound deformation. The established fully connected neural network model contains five hidden layers with dimensions of 25, 50, 100, 50, and 25, respectively. The prediction model was validated and tested in 10 groups of samples in the validation set, and the test results are shown in Table 2.

[0022] Table 2 Prediction error of the neural network model on the validation set ;

[0023] As shown in Table 2, the average error of the established data-driven prediction model is less than 0.3%, and the maximum error is less than 1.2%. This accuracy meets the needs of subsequent parameter identification. In addition, it takes 5 minutes to complete a simulation using multi-core acceleration, while it only takes 0.002 seconds to predict the maximum springback deformation using the data-driven prediction model, which greatly improves the prediction efficiency.

[0024] An iterative optimization model is constructed based on the non-dominated sorting genetic algorithm with elite strategy (NSGA-II). The identifiable parameter combinations determined after parameter analysis are encoded as chromosomes, and the two fitness functions are defined as the average percentage absolute error and the maximum percentage error between the predicted value and the experimental data value. The above optimization problem can be expressed as: ; ; in, For the Prediction value of curing deformation of composite parts, For the The measured value of the curing deformation of the composite parts, To identify the parameter vector, Represents the number of parts. The forward prediction model from the identification parameter group to the solidification deformation value adopts the established neural network proxy model. The Pareto frontier of the identification parameter group is obtained through the above iterative optimization problem, and the weights of the average percentage absolute error and the maximum percentage error are finally considered to select the final optimization result. The optimization results are shown in Table 3.

[0025] Table 3 Parameter identification results of a material system ;

[0026] The above iteratively optimized identification parameter group and the remaining unidentified parameters are used as simulation model parameters to simulate the C / U / V-shaped parts under the two types of layups. The test results are shown in Table 4.

[0027] Table 4 Simulation test results of C / U / V-shaped parts under two types of layups ;

[0028] As shown in Table 4, the maximum error of the solidification deformation prediction is 4.69%, and the average error is 2.98%, which achieves a relatively ideal simulation prediction accuracy. The above iteratively optimized identification parameter group and the remaining unidentified parameters are further used as simulation model parameters to simulate the Z-type parts under zero-degree ply. The prediction results are as follows: Figure 5 As shown. Figure 5It can be seen that the maximum rebound deformation value of the 60-degree angle area obtained by simulation is 1.768mm, and the maximum rebound deformation value of the 90-degree angle area is 0.8648mm. The curing experiment was carried out on the Z-shaped part under the zero-degree ply and the mold surface was measured and analyzed. The results showed that the maximum rebound deformation value of the 60-degree angle area was 1.653mm, and the maximum rebound deformation value of the 90-degree angle area was 0.824mm. Therefore, the prediction errors of the curing deformation simulation model are +6.96% and +4.95%, respectively, both of which are within 10%. With the introduction of more measured curing deformation data of composite parts in the parameter identification process, the practical application effect of the method of the present invention will be further improved.

[0029] Therefore, the present invention adopts the above-mentioned composite material curing deformation simulation model parameter identification method based on measured data. First, the measured curing deformation data of a group of composite parts under the material system to be identified are obtained, and the simulation data is used to establish a data-driven prediction model from the parameters to be identified to the curing deformation for each part, and then these data-driven models are used as proxy models of the curing deformation simulation model. The values ​​of all parameters to be identified are determined through large-scale search iterations, so that the gap between the proxy model prediction data and the measured curing deformation data meets the target set by the user; through this method, the measured data can be directly used to efficiently obtain the key parameters in the composite material curing deformation simulation model, thereby significantly improving the simulation accuracy of the curing deformation of the composite parts.

[0030] 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 material curing deformation simulation model parameter identification method based on measured data, characterized in that: The following steps are involved: Step 1: Obtain the material system to be identified A collection of measured curing deformation data of composite parts ; Step 2: Construct a collection of solidification deformation simulation models of all parts in step 1) , determine all parameters to be identified in the simulation model The value range is selected and sampling is performed within this range. Through the simulation model Calculate the composite part in step 1 In the corresponding A collection of solidification deformation prediction data under a set of parameters to be identified , construct the set of solidification deformation simulation data of all composite parts in step 1 ; Step 3: Using the data set in step 2 Training data-driven predictive models As the simulation model in step 2 Proxy model, construct the simulation model set in step 2 A collection of proxy models ; Step 4: Determine the corresponding values ​​for all parameters to be identified, so that the proxy model set in step 3 A collection of solidification deformation prediction data under its input The combination of the measured curing deformation data in step 1 The gap between them meets the user's pre-set goals.

2. The method for identifying parameters of a composite material curing deformation simulation model based on measured data according to claim 1, characterized in that: The curing deformation data in step 1 is index data or deformation field data reflecting the difference between the measured shape of the deformed part and the theoretical shape; wherein the index data includes one or more of the maximum warpage, the maximum deformation, and the springback angle.

3. The method for identifying parameters of a composite material curing deformation simulation model based on measured data according to claim 2, characterized in that: The parameters to be identified in step 2 are selected from the parameters that are difficult to measure accurately and are more sensitive to the curing deformation in the curing deformation simulation model, and the value range is reasonably determined according to the physical meaning of the parameters, preliminary measurement results or parameter values ​​of similar material systems.

4. The method for identifying parameters of a composite material curing deformation simulation model based on measured data according to claim 3, characterized in that: The type of data-driven prediction model in step 3 depends on the type of solidified deformation data in step 1. When the solidified deformation is indicator data, the data-driven model is a fully connected neural network, a decision tree model, a support vector machine, and a Gaussian process regression model; when the solidified deformation is deformation field data, the data-driven model includes various types of reduced-order models, convolutional neural networks, recurrent neural networks, graph neural networks, Transformer models, and neural operator models suitable for high-dimensional field data prediction.

5. The method for identifying parameters of a composite material curing deformation simulation model based on measured data according to claim 4, characterized in that: In step 4, the corresponding values ​​of all the parameters to be identified are determined by using a non-iterative method such as Monte Carlo search or enumeration or an iterative optimization method such as gradient descent or genetic algorithm.

6. The method for identifying parameters of a composite material curing deformation simulation model based on measured data according to claim 5, characterized in that: In step 4, the gap between the set of predicted solidification deformation data of the proxy model set under its input and the set of measured solidification deformation data is obtained by adopting a multi-objective combination method including the average percentage absolute error and the maximum percentage absolute error; among them, when the deformation field is used to characterize the deformation, it is also necessary to introduce an indicator characterizing the consistency between the distribution of the predicted deformation field data and the distribution of the measured deformation field data into the optimization objective, which is obtained by adopting KL divergence, Wasserstein distance, and mutual information.

Images