Method for constructing a prediction model of mechanical properties of porous resin-based composites

Abstract

The application discloses a method for constructing a prediction model of mechanical properties of a resin-based composite material containing pores, relates to the technical field of resin-based composite materials, generates a three-dimensional RVE model of a resin-based fiber-reinforced composite material containing pores of different sizes, develops an RVE simulation script, realizes RVE geometric modeling, material attribute assignment, mesh division, periodic boundary condition application and equivalent mechanical property calculation, generates a certain number of simulation parameter combinations and inputs the simulation parameter combinations into the RVE simulation script to obtain corresponding equivalent mechanical property parameters, combines the simulation parameter combinations and the corresponding equivalent mechanical property parameters to construct a simulation data set, establishes a machine learning model, and trains the model by using the simulation data set to obtain a mechanical property prediction proxy model. The model can be more accurately and quickly used for mechanical property prediction of the resin-based composite material containing pores.

Description

Technical Field

[0001] This invention relates to the field of resin-based composite materials technology, and in particular to a method for constructing a predictive model for the mechanical properties of porous resin-based composite materials. Background Technology

[0002] Resin-based composite materials, primarily composed of a resin matrix and reinforcing phases such as continuous fibers, possess significant advantages such as lightweight, high strength, and high designability, demonstrating immense application potential across various industrial sectors. With the increasing application of resin-based composite materials, accurately understanding their mechanical properties is becoming increasingly crucial to ensuring the safety and reliability of various composite structure designs. During the manufacturing process of resin-based composite materials, microscopic pores inevitably arise within the resin matrix. These pores are primarily caused by factors such as residual air introduced during manufacturing processes (e.g., fiber entanglement), insufficient resin impregnation, and gas release during resin curing. These randomly distributed pores, as manufacturing defects, have a significant adverse impact on the modulus and other mechanical properties of resin-based composite materials. Therefore, accurately predicting the mechanical properties of porous resin-based composite materials is of great importance for improving material manufacturing processes, structural design, and safety assessment.

[0003] To predict the mechanical properties of resin composites, researchers have proposed the Representative Volume Element (RVE) method, which uses microscale simulation models to calculate the equivalent macroscopic mechanical properties of materials. However, to ensure the realism and accuracy of the microstructure, the generation and calculation of a single RVE model often requires a significant amount of time and computational resources. The addition of pores of random size and location further complicates the problem. In practical applications, the simulation model often needs to be adjusted according to the actual microstructure and material parameters, which is time-consuming, labor-intensive, and requires a high level of professional knowledge and skills from the operators, thus limiting its practicality. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, this invention provides a method for constructing a predictive model for the mechanical properties of porous resin-based composite materials. This method combines the advantages of parametric numerical simulation and machine learning. The surrogate model established by this method can be used more accurately and quickly for predicting the mechanical properties of porous resin-based composite materials.

[0005] To achieve the above objectives, the present invention adopts the following technical solution, including:

[0006] A method for constructing a predictive model for the mechanical properties of porous resin-based composite materials includes the following steps:

[0007] S1. Generate a three-dimensional RVE model of resin-based fiber-reinforced composite material with pores of different sizes, develop a corresponding RVE simulation script, and based on the simulation parameters, the script realizes RVE geometric modeling, material property assignment, mesh generation, application of periodic boundary conditions and calculation of equivalent mechanical properties.

[0008] S2, generate a certain number of simulation parameter combinations within a preset range of multiple simulation parameters, and input the simulation parameter combinations into the RVE simulation script respectively, and obtain the corresponding equivalent mechanical performance parameters through calculation;

[0009] S3 combines the simulation parameter combinations and the corresponding equivalent mechanical performance parameters to construct a simulation dataset;

[0010] S4. Establish a machine learning model, train the model using a simulation dataset, and obtain a surrogate model for predicting the mechanical properties of porous resin-based composite materials. The model input is a combination of simulation parameters from a three-dimensional RVE model, and the model output is the equivalent mechanical property parameters of porous resin-based composite materials.

[0011] Preferably, the simulation parameter combination includes: fiber volume fraction, fiber elastic modulus, fiber shear modulus, fiber Poisson's ratio, resin elastic modulus, resin Poisson's ratio, mean pore diameter, standard deviation of pore diameter, and porosity; the equivalent mechanical property parameters include the elastic modulus, shear modulus, and Poisson's ratio of the composite material.

[0012] Preferably, in the 3D RVE model, the pore shape is cylindrical, and the pore diameter D of pores of different sizes follows a normal distribution. The normal distribution probability density function used for pore diameter sampling is as follows:

[0013] ;

[0014] Where μ is the mean pore diameter and σ is the standard deviation of the pore diameter. For input variables, Let x be the probability density function of the normal distribution;

[0015] The formation process of pore sections of different sizes in the resin matrix cross-section is as follows:

[0016] S121, given the minimum value D of the pore diameter min and maximum value D max Given a step size ΔD, at the minimum value D of the pore diameter min and maximum value D max Within the range, all possible diameter values ​​are generated by discrete step size ΔD;

[0017] S122, randomly select a diameter value as a candidate diameter d, calculate the probability density pdf(d) of the candidate diameter d in the normal distribution, generate a random number rand, and the value of rand is between 0 and pdf(μ); if rand < pdf(d), then accept the candidate diameter d as the pore diameter D, and randomly generate a pore cross section in the resin matrix cross section based on the pore diameter D, and the pore cross section does not interfere with the fiber cross section or the already generated pore cross section; otherwise, reject the candidate diameter d, and reselect a candidate diameter d that satisfies rand < pdf(d), thereby generating a pore cross section;

[0018] S123, Calculate the corresponding porosity based on the total area of ​​all generated pore cross sections. The length and width of the resin matrix cross-section are a and b, respectively; if the porosity Porosity greater than or equal to the set value If the result is positive, the process of generating the pore cross section stops; otherwise, jump to step S122, randomly select a diameter value as a new candidate diameter d, determine whether to accept the new candidate diameter d as the pore diameter D, and continue to generate a new pore cross section.

[0019] Preferably, the specific process of step S1 is as follows:

[0020] S11, the length, width, and height of the 3D RVE model are a, b, and c, respectively; the fiber shape is cylindrical, and the fiber radius is R. f The fiber volume fraction is V f Fiber count N Randomly distributed fiber cross sections are generated within the resin matrix cross section, and there is no interference between two adjacent fiber cross sections; the resin matrix cross section is a×b of the three-dimensional RVE model;

[0021] S12 has cylindrical pores, and the pore diameter D of pores of different sizes satisfies a normal distribution, generating randomly distributed pore cross sections of different sizes in the resin matrix cross section;

[0022] S13, after stretching, yields a cubic resin matrix, cylindrical fibers, and cylindrical pores. Boolean operations are performed on these three components to obtain a three-dimensional RVE model.

[0023] Material properties are assigned to the resin matrix and fibers respectively; wherein the resin matrix is ​​an isotropic material, and the resin material parameters include the elastic modulus E. m Compared to Poisson's ratio v mThe fiber is a transversely isotropic material. A local coordinate system is established, with directions 1, 2, and 3 of the local coordinate system mapped to the z, x, and y directions of the geometric coordinate system. Direction 1 corresponds to the fiber direction and the z direction, while directions 2 and 3 correspond to the x and y directions, respectively. Material properties are assigned to the fiber in the local coordinate system, including the elastic modulus E. f11 E f22 E f33 Shear modulus G f12 G f13 G f23 Poisson's ratio f12 ν f13 ν f23 E f22 =E f33 G f12 =G f13 ,ν f12= ν f13 ;

[0024] S14, Mesh the 3D RVE model;

[0025] S15, invoke the EasyPBC plugin tool to set the periodic boundary conditions of the 3D RVE model, extract stress-strain data, and calculate the equivalent mechanical property parameters of the porous resin-based composite material, including the elastic modulus E. 11 E 22 E 33 Shear modulus G 12 G 13 G 23 Poisson's ratio ν 12 ν 13 ν 23 E 22 =E 33 G 12 =G 13 ,ν 12= ν 13 .

[0026] Preferably, in step S2, a certain number of simulation parameter combinations are generated within a preset range of multiple simulation parameter ranges using the Latin hypercube sampling method.

[0027] Preferably, in step S3, the simulation dataset is standardized and preprocessed to transform the simulation parameters into a distribution with a mean of 0 and a standard deviation of 1.

[0028] Preferably, in step S4, an XGBoost model is built using the Python platform, and the XGBoost model is trained using a simulation dataset, with the objective function being the mean squared error.

[0029] Preferably, an independent XGBoost model is trained for each equivalent mechanical performance parameter.

[0030] The present invention also provides a readable storage medium having a computer program stored thereon, which, when executed, implements the method for constructing a predictive model of the mechanical properties of porous resin-based composite materials.

[0031] The present invention also provides an electronic device, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for constructing a predictive model of the mechanical properties of porous resin-based composite materials.

[0032] The advantages of this invention are:

[0033] (1) This invention proposes a method for constructing a prediction model for the mechanical properties of porous resin-based composite materials. It combines the advantages of parametric numerical simulation and machine learning methods. The surrogate model established by this method can be used more accurately and quickly for predicting the mechanical properties of porous resin-based composite materials.

[0034] (2) Based on random sequential expansion and accept-reject sampling, this invention realizes efficient random generation of RVE models of resin-based composite materials with different porosities. The pore size distribution is described by probability density function, which is closer to the microstructure of real composite materials. The relevant simulation parameters can be adjusted, the modeling freedom is high, and the scalability is strong.

[0035] (3) The present invention realizes the full parameterization of the modeling and calculation process of the RVE model through Python script, and generates a certain number of uniformly distributed sample points in the multidimensional parameter space through the Latin hypercube sampling method, thereby automatically and efficiently obtaining the calculation results under different simulation parameter conditions, and quickly constructing the equivalent mechanical properties dataset of porous resin-based composite materials at a low cost.

[0036] (4) This invention constructs an XGBoost model and optimizes hyperparameters through step-by-step adjustment and cross-validation. Finally, the predictive performance of the surrogate model is evaluated from multiple aspects and interpretability analysis is performed. This surrogate model can efficiently and accurately predict multiple mechanical properties of porous composite materials, such as modulus and Poisson's ratio, and can explain the contribution of each simulation parameter, which is convenient for practical application and can provide important support for material performance prediction and evaluation and process optimization. Attached Figure Description

[0037] Figure 1 This is a flowchart of the method for constructing a predictive model for the mechanical properties of porous resin-based composite materials according to the present invention.

[0038] Figure 2V is the fiber volume fraction. f =0.55, porosity The RVE model when =0.02.

[0039] Figure 3 For the proxy model, on the test set, for label E 22 A comparison chart of predicted and actual values.

[0040] Figure 4 Predicting E using surrogate model with 12 independent simulation parameters 22 Contribution ranking chart. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Depend on Figure 1 As shown, this invention proposes a method for constructing a predictive model for the mechanical properties of porous resin-based composite materials, comprising the following steps:

[0043] S1 uses random sequential expansion (RSE) and accept-reject sampling methods to generate three-dimensional RVE models of resin-based fiber-reinforced composite materials with pores of different sizes. A corresponding Python script (i.e., RVE simulation script) is developed to parametrically (automatically) realize functions such as RVE geometric modeling, material property assignment, mesh generation, application of periodic boundary conditions, and calculation of equivalent mechanical properties.

[0044] The specific process of step S1 is as follows:

[0045] S11, the length, width, and height of the 3D RVE model are a, b, and c, respectively; the fiber shape is cylindrical, and the fiber radius is R. f The fiber volume fraction is V f The formula for calculating the number of fibers N is: The randomized sequential expansion (RSE) algorithm is used to generate randomly distributed fiber cross-sections within the resin matrix. Specifically, the resin matrix cross-section is an a×b section of the three-dimensional RVE model. The center coordinates of each fiber cross-section are randomly generated within the resin matrix cross-section. Each newly generated center coordinate must be located within the a×b rectangular region, and the center distance between two adjacent fiber cross-sections must be within L. min +2R f With L max +2R fBetween them, to ensure that there is no interference between two adjacent fibers. Among them, L min L max The minimum and maximum values ​​for the set edge distance.

[0046] In this embodiment, a=65μm, b=65μm, c=5μm, R f =3.5μm, L min =0.4μm, L max =1.5μm.

[0047] S12, the pore shape is cylindrical, and the pore diameter D of different sizes is assumed to follow a normal distribution. The normal distribution probability density function used for pore diameter sampling is as follows:

[0048] ;

[0049] Where μ is the mean pore diameter and σ is the standard deviation of the pore diameter. For input variables, Let x be the probability density function of x in a normal distribution.

[0050] The formation process of pore sections of different sizes in the resin matrix cross-section is as follows:

[0051] S121, given the minimum value D of the pore diameter min =1μm and maximum value D max =5μm, given a step size ΔD=0.1μm, generate all possible diameter values ​​by discretizing within the range of minimum to maximum pore diameter by step size;

[0052] S122, randomly select one of the diameter values ​​as the candidate diameter d, calculate the probability density pdf(d) of the candidate diameter d in the normal distribution, and generate a random number rand, where the value of rand is between 0 and pdf(μ). If rand < pdf(d), then accept the candidate diameter d as the pore diameter D. Then, randomly generate the center coordinates of the pore cross-section. Based on the pore diameter D and the center coordinates, generate the corresponding pore cross-section within the resin matrix cross-section, and ensure that the pore cross-section does not interfere with the fiber cross-section or the already generated pore cross-section. The interference judgment condition is that the center distance between the pore cross-section and the fiber cross-section or the already generated pore cross-section is less than the sum of their radii plus Δ. l Δ l A value of 0.5 μm can be used to ensure the quality of the local mesh; if rand ≥ pdf(d), the candidate diameter d is rejected and a new candidate diameter d is selected until rand < pdf(d) is satisfied, thereby generating the corresponding pore section.

[0053] S123, Calculate the corresponding porosity based on the total area of ​​all generated pore cross sections. If porosity Reaching or exceeding the set porosity If the result is positive, the process of generating the pore cross section stops; otherwise, jump to step S122, randomly select a diameter value as a new candidate diameter d, determine whether to accept the new candidate diameter d as the pore diameter D, and continue to generate a new pore cross section.

[0054] S13, after stretching, yields a cubic resin matrix (a×b×c), cylindrical fibers, and cylindrical pores. Boolean operations are performed on these three components to obtain the geometry of the three-dimensional RVE model.

[0055] Material properties are assigned to both the resin matrix and the fibers. The resin matrix is ​​an isotropic material, and its material parameters include the elastic modulus E. m Compared to Poisson's ratio v m The fiber is a transversely isotropic material. A local coordinate system is established, with directions 1, 2, and 3 of the local coordinate system mapped one-to-one with the z, x, and y directions of the geometric coordinate system. Direction 1 corresponds to the fiber direction and the z direction, while directions 2 and 3 correspond to the x and y directions, respectively. Material properties are assigned to the fiber in the local coordinate system, including the elastic modulus E. f11 E f22 E f33 Shear modulus G f12 G f13 G f23 Poisson's ratio ν f12 ν f13 ν f23 E f22 =E f33 G f12 =G f13 ,ν f12= ν f13 .

[0056] S14: Mesh the 3D RVE model. The mesh type is mainly hexahedral, with a mesh cell size of 0.8μm. A sweep technique is used to generate a periodic mesh. Figure 2 It shows the fiber volume fraction V f =0.55, porosity The 3D RVE model when =0.02.

[0057] S15, using the open-source EasyPBC plugin, sets the periodic boundary conditions for the 3D RVE model. This automatically identifies the boundary nodes of the RVE and generates constraint equations and displacement boundary conditions. Then, in the post-processing stage, stress and strain data are extracted, and the equivalent mechanical properties of the composite material, including the elastic modulus E, are calculated. 11 E 22 E 33Shear modulus G 12 G 13 G 23 Poisson's ratio ν 12 ν 13 ν 23 E 22 =E 33 G 12 =G 13 ,ν 12= ν 13 .

[0058] At this point, the modeling and calculation of the three-dimensional RVE model of the porous resin-based composite material is completed, and six independent equivalent mechanical property parameters of the porous resin-based composite material are obtained. The relevant calculation files are saved.

[0059] S2 uses the Latin hypercube sampling method to generate a certain number of samples composed of simulation parameters within multiple preset simulation parameter ranges, and inputs the samples into the RVE simulation script to obtain the corresponding result files through calculation.

[0060] The specific process of step S2 is as follows:

[0061] First, several simulation parameters for RVE simulation of porous resin-based composite materials were determined. In this example, 12 independent simulation parameters were selected, namely fiber volume fraction V... f Fiber mechanical property parameters E f11 E f22 G f12 G f23 ν f12 ν f23 Resin mechanical property parameters E m v m and pore parameters The values ​​of μ and σ, and their respective discretization step sizes are shown in Table 1 below.

[0062] Table 1. 12-dimensional simulation parameters for RVE simulation

[0063]

[0064] In order to efficiently generate a set of uniformly distributed sample points in the multidimensional parameter space to cover the entire parameter space, Latin hypercube sampling is used to generate 900 simulation parameter combinations, each combination representing a sample point, i.e. a three-dimensional RVE model.

[0065] Then, the 900 simulation parameter combinations, each containing 12 independent simulation parameters, are sequentially input into the RVE simulation script to automatically complete batch simulation calculations and obtain result files for 900 different 3D RVE models.

[0066] S3 reads all result files, extracts mechanical performance parameters such as modulus and Poisson's ratio in different directions, combines and merges them with the corresponding simulation parameters to jointly construct the simulation dataset, and completes preprocessing operations such as standardization.

[0067] The specific process of step S3 is as follows:

[0068] The result files of 900 three-dimensional RVE models obtained from batch simulation calculations were read sequentially, and six independent equivalent mechanical property parameters of the porous resin-based composite material output from each result file were extracted, namely the elastic modulus E. 11 E 22 Shear modulus G 12 G 23 and Poisson's ratio ν 12 ν 23 The dataset consists of 900 data points, corresponding to the simulation parameter combinations used to generate 900 3D RVE models. The former (equivalent mechanical performance parameters) are the label values, and the latter (simulation parameter combinations, i.e., 12 independent simulation parameters) are the model input values. The two are combined to form the simulation dataset, which contains a total of 900 data points.

[0069] Check for missing values ​​in the data and delete the entire row containing the missing values.

[0070] To eliminate the influence of dimensions and improve model performance, simulation parameters are standardized by transforming them into a distribution with a mean of 0 and a standard deviation of 1. The formula is as follows:

[0071] ;

[0072] Where x and x′ are the simulation parameters before and after standardization, respectively, and μ and σ are the mean and standard deviation of the simulation parameters, respectively.

[0073] The simulation dataset is divided into a training set and a test set, with the training set accounting for 80% and the test set accounting for 20%.

[0074] S4. Establish a machine learning model, train the model using a simulation dataset, and obtain a surrogate model for predicting the mechanical properties of porous resin-based composite materials. The model input is a combination of simulation parameters from a three-dimensional RVE model, and the model output is the equivalent mechanical property parameters of porous resin-based composite materials.

[0075] The specific process of step S4 is as follows:

[0076] An XGBoost model is defined using the Python platform. This model can effectively capture the nonlinear relationships behind large amounts of data. The objective function for the regression task is the mean squared error (MSE).

[0077] ;

[0078] in, It is a tag value. is the predicted value, and n is the number of samples.

[0079] Furthermore, this invention trains an independent XGBoost model for each label (each equivalent mechanical performance parameter), and adopts a synchronous modeling and training and unified hyperparameter optimization strategy to ensure the consistency of each label proxy model.

[0080] The model's hyperparameters are coarsely tuned and finely tuned using a grid search method. Hyperparameters include the number of trees, learning rate, and maximum tree depth. The coarse tuning stage searches for optimal hyperparameter values ​​within a relatively wide range, quickly finding approximate optimal parameter ranges, such as a learning rate search range of [0.01, 0.05, 0.1, 0.2, 0.3]. The fine tuning stage further refines the parameters within the approximate range found in the coarse tuning, adjusting the parameters with smaller step sizes to find more precise optimal values, such as a learning rate search range of [0.08, 0.09, 0.1, 0.11, 0.12]. This step-by-step search method improves the efficiency of parameter search, finding hyperparameter combinations closer to the global optimum by comprehensively searching the parameter space. To avoid overfitting, cross-validation is used in the hyperparameter optimization process on the training set. Five-fold cross-validation is chosen, further dividing the training set into five subsets: four for training and one for validation. This evaluates the model's performance on different subsets and selects the optimal hyperparameter combination. The XGBoost model is retrained on the entire training set using the optimal combination of hyperparameters. The saved model is the predictive surrogate model for the mechanical properties of porous resin-based composite materials.

[0081] S5 evaluates the performance of the surrogate model on the test set and performs interpretability analysis on the prediction process of the mechanical properties of porous resin-based composites based on SHAP values.

[0082] The specific process of step S5 is as follows:

[0083] The performance of the surrogate model was evaluated on the test set, and the evaluation metrics included mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). 2 The performance of the proxy model on the test set is shown in Table 2. The three errors for all six labels are relatively small, and R0 is [value missing]. 2 The values ​​are all no less than 0.9639, indicating that the surrogate model is relatively accurate in predicting the mechanical properties of porous resin-based composite materials. Figure 3 This demonstrates the proxy model's performance on label E on the test set. 22 The comparison chart of predicted and actual values ​​shows that the model's prediction performance is good.

[0084] To further analyze the impact of each feature on label prediction, based on the XGBoost model corresponding to each label, a SHAP interpreter is initialized for each label, and the SHAP value on the training data is calculated, which represents the prediction contribution of each feature to each sample. Figure 4 The simulation parameters of 12 independent simulation parameters are shown to affect the surrogate model's prediction of E. 22 The contribution ranking shows that V f E m v m as well as The transverse elastic modulus E of porous resin-based composites 22 The effect is significant, while the size distribution characteristics of the pores (μ, σ) are almost unaffected.

[0085] Table 2 Evaluation results of the XGBoost model on the test set

[0086]

[0087] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing a predictive model for the mechanical properties of porous resin-based composite materials, characterized in that, Includes the following steps: S1. Generate a three-dimensional RVE model of resin-based fiber-reinforced composite material with pores of different sizes, develop a corresponding RVE simulation script, and based on the simulation parameters, the script realizes RVE geometric modeling, material property assignment, mesh generation, application of periodic boundary conditions and calculation of equivalent mechanical properties. S2, generate a certain number of simulation parameter combinations within a preset range of multiple simulation parameters, and input the simulation parameter combinations into the RVE simulation script respectively, and obtain the corresponding equivalent mechanical performance parameters through calculation; S3 combines the simulation parameter combinations and the corresponding equivalent mechanical performance parameters to construct a simulation dataset; S4. Establish a machine learning model, train the model using a simulation dataset, and obtain a surrogate model for predicting the mechanical properties of porous resin-based composite materials. The model input is the combination of simulation parameters of the three-dimensional RVE model, and the model output is the equivalent mechanical property parameters of the porous resin-based composite materials. In the 3D RVE model, the pores are cylindrical, and the pore diameter D of different sizes follows a normal distribution. The normal distribution probability density function used for pore diameter sampling is as follows: Where μ is the mean pore diameter and σ is the standard deviation of the pore diameter. For input variables, Let x be the probability density function of the normal distribution; The formation process of pore sections of different sizes in the resin matrix cross-section is as follows: S121, given the minimum value D of the pore diameter min and maximum value D max Given a step size ΔD, at the minimum value D of the pore diameter min and maximum value D max Within the range, all possible diameter values ​​are generated by discrete step size ΔD; S122, randomly select a diameter value as a candidate diameter d, calculate the probability density pdf(d) of the candidate diameter d in the normal distribution, generate a random number rand, and the value of rand is between 0 and pdf(μ); if rand < pdf(d), then accept the candidate diameter d as the pore diameter D, and randomly generate a pore cross section in the resin matrix cross section based on the pore diameter D, and the pore cross section does not interfere with the fiber cross section or the already generated pore cross section; otherwise, reject the candidate diameter d, and reselect a candidate diameter d that satisfies rand < pdf(d), thereby generating a pore cross section; S123, Calculate the corresponding porosity based on the total area of ​​all generated pore cross sections. The length and width of the resin matrix cross-section are a and b, respectively; if the porosity Porosity greater than or equal to the set value If the result is positive, the process of generating the pore cross section stops; otherwise, jump to step S122, randomly select a diameter value as a new candidate diameter d, determine whether to accept the new candidate diameter d as the pore diameter D, and continue to generate a new pore cross section.

2. The method for constructing a predictive model for the mechanical properties of porous resin-based composite materials according to claim 1, characterized in that, The simulation parameter combination includes: fiber volume fraction, fiber elastic modulus, fiber shear modulus, fiber Poisson's ratio, resin elastic modulus, resin Poisson's ratio, mean pore diameter, standard deviation of pore diameter, and porosity; the equivalent mechanical property parameters include the elastic modulus, shear modulus, and Poisson's ratio of the composite material.

3. The method for constructing a predictive model for the mechanical properties of porous resin-based composite materials according to claim 1, characterized in that, The specific process of step S1 is as follows: S11, the length, width, and height of the 3D RVE model are a, b, and c, respectively; the fiber shape is cylindrical, and the fiber radius is R. f The fiber volume fraction is V f Fiber count N Randomly distributed fiber cross sections are generated within the cross section of the resin matrix, and there is no interference between two adjacent fiber cross sections; The cross section of the resin matrix is ​​the a×b section of the three-dimensional RVE model; S12 has cylindrical pores, and the pore diameter D of pores of different sizes satisfies a normal distribution, generating randomly distributed pore cross sections of different sizes in the resin matrix cross section; S13, after stretching, yields a cubic resin matrix, cylindrical fibers, and cylindrical pores. Boolean operations are performed on these three components to obtain a three-dimensional RVE model. Material properties are assigned to the resin matrix and fibers respectively; wherein the resin matrix is ​​an isotropic material, and the resin material parameters include the elastic modulus E. m Compared to Poisson's ratio v m The fiber is a transversely isotropic material. A local coordinate system is established, with directions 1, 2, and 3 of the local coordinate system mapped to the z, x, and y directions of the geometric coordinate system. Direction 1 corresponds to the fiber direction and the z direction, while directions 2 and 3 correspond to the x and y directions, respectively. Material properties are assigned to the fiber in the local coordinate system, including the elastic modulus E. f11 E f22 E f33 Shear modulus G f12 G f13 G f23 Poisson's ratio f12 ν f13 ν f23 E f22 =E f33 G f12 =G f13 ,ν f12= ν f13 ; S14, mesh the 3D RVE model; S15, invoke the EasyPBC plugin tool to set the periodic boundary conditions of the 3D RVE model, extract stress-strain data, and calculate the equivalent mechanical property parameters of the porous resin-based composite material, including the elastic modulus E. 11 E 22 E 33 Shear modulus G 12 G 13 G 23 Poisson's ratio ν 12 ν 13 ν 23 E 22 =E 33 G 12 =G 13 ,ν 12= ν 13 .

4. The method for constructing a predictive model for the mechanical properties of porous resin-based composite materials according to claim 1, characterized in that, In step S2, a certain number of simulation parameter combinations are generated within a preset range of multiple simulation parameter ranges using the Latin hypercube sampling method.

5. The method for constructing a predictive model for the mechanical properties of porous resin-based composite materials according to claim 1, characterized in that, In step S3, the simulation dataset is standardized and preprocessed to transform the simulation parameters into a distribution with a mean of 0 and a standard deviation of 1.

6. The method for constructing a predictive model for the mechanical properties of porous resin-based composite materials according to claim 1, characterized in that, In step S4, an XGBoost model is built using the Python platform, and the XGBoost model is trained using a simulation dataset. The objective function is the mean squared error.

7. The method for constructing a predictive model for the mechanical properties of porous resin-based composite materials according to claim 6, characterized in that, Train an independent XGBoost model for each equivalent mechanical performance parameter.

8. A readable storage medium, characterized in that, It stores a computer program, which, when executed, implements the method for constructing a predictive model of the mechanical properties of porous resin-based composite materials as described in any one of claims 1 to 6.

9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for constructing a predictive model of the mechanical properties of porous resin-based composite materials as described in any one of claims 1 to 6.

Images