Method for predicting tensile strength of carbon fiber composite material based on machine learning

Abstract

The invention discloses a carbon fiber composite material tensile strength prediction method based on machine learning, which comprises the following steps: acquiring component data of a carbon fiber composite material and normal temperature tensile strength data of each original component to form a data set; classifying all original components of the data set; calculating mechanical characteristics of each functional component in the data set to generate corresponding characteristics; feature processing and feature screening are carried out, a feature pool is constructed, and feature importance sorting is carried out; an optimal sub-feature set is screened out from the feature pool through an iterative optimization algorithm; and by taking the optimal sub-feature set as an input independent variable and the normal-temperature tensile strength of the carbon fiber composite material as an output dependent variable, constructing a prediction model by adopting a gradient boosting decision tree regression algorithm suitable for high-dimensional nonlinear data. According to the method, feature extraction and a machine learning algorithm are combined, so that the understanding of the microstructure of the composite material is improved, and the accuracy of predicting the tensile strength of the carbon fiber composite material is further improved.

Description

Technical Field

[0001] The present invention relates to the field of materials science and technology, and particularly to a method for predicting the tensile strength of carbon fiber composites based on machine learning. Background Art

[0002] Due to their excellent mechanical properties and lightweight characteristics, carbon fiber composites are widely used in fields such as aerospace, automotive, and sports equipment. Tensile strength is an important indicator for evaluating the performance of carbon fiber composites, directly affecting the safety and reliability of their structures. In traditional methods, the tensile strength of carbon fiber composites mainly relies on experimental testing to obtain. This method requires the preparation of a large number of standard specimens and repeated testing through a precision tensile testing machine, which is not only time-consuming and laborious but also costly, making it difficult to meet the rapidly developing industrial demands.

[0003] With the development of materials science and computer technology, prediction methods based on machine learning have gradually been applied to the prediction of material properties. Machine learning can quickly process and analyze a large amount of material data without the need to prepare and test materials one by one like traditional experimental methods. By deeply analyzing the existing data of carbon fiber composites, this method explores the internal relationships between various component information such as fibers, resins, and fillers and their tensile strength data under normal temperature conditions, and constructs a prediction model that reflects the complex relationships between these data. In this way, when predicting the tensile strength of carbon fiber composites, only the component characteristic data of the relevant materials need to be input into the trained machine learning model, and the prediction results can be obtained quickly, greatly saving time and labor costs and improving the prediction efficiency, thus meeting the requirements for the rapid evaluation of the performance of carbon fiber composites in industrial production.

[0004] Therefore, the present invention provides a method for predicting the normal temperature tensile strength of carbon fiber composites based on machine learning. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for predicting the tensile strength of carbon fiber composites based on machine learning to solve the problems raised in the above background art. To achieve the above purpose, the present invention provides the following technical solutions:

[0006] A method for predicting the tensile strength of carbon fiber composites based on machine learning, comprising the following steps:

[0007] S1. Collect all the original component data of the carbon fiber composite and the normal temperature tensile strength data of each original component to form a data set, where each carbon fiber composite is composed of two or more components;

[0008] S2. Classify all the original components as functional components according to the functions and roles of the original components in the carbon fiber composite;

[0009] S3. Calculate the mechanical properties of each functional component by numerical simulation methods, and generate corresponding characteristics of each functional component, including the physical properties of the fibers (such as length, diameter, density, volume fraction), the laying angle of the fibers, the elastic modulus of the matrix material, the tensile strength of the matrix material, etc.;

[0010] S4. Using the component ratio as the weight, perform weighted average operations on the corresponding characteristics of each functional component (including fiber components, matrix components, filler components, and other components) to obtain weighted average characteristics, and perform addition and subtraction operations to obtain operation characteristics, and use the weighted average characteristics and the operation characteristics as initial characteristics;

[0011] S5. Calculate the Pearson correlation coefficient between any two characteristics in the initial characteristics. If the correlation coefficient is greater than 0.9, it is considered that the two characteristics are highly correlated. Remove redundant characteristics and only retain one of them. Construct a feature pool containing relatively independent and important characteristics, and rank the importance of each characteristic in the feature pool;

[0012] S6. Screen out the optimal sub-feature set from the feature pool through an iterative optimization algorithm: taking the accuracy of the gradient boosting decision tree regression algorithm in machine learning as the objective function, perform iterative screening of sub-features, and screen out the sub-feature set corresponding to the highest accuracy of the machine learning gradient boosting decision tree regression model from the feature pool;

[0013] S7. Using the optimal sub-feature set as the input independent variable and the room temperature tensile strength of the carbon fiber composite as the output dependent variable, adopt the gradient boosting decision tree (GBDT) regression algorithm applicable to high-dimensional non-linear data to construct a prediction model, and correct the prediction residuals by gradient descent to minimize the root mean square error between the predicted value of the room temperature tensile strength output by the model and the measured value.

[0014] Preferably, all the original components included in the carbon fiber composite material include carbon fiber, epoxy resin, polyimide resin, calcium carbonate filler, silane coupling agent, and other modification additives that affect the tensile strength, and the sum of the mass percentages of each component is 100%.

[0015] Preferably, the functional components include: fiber components, matrix components, filler components, and other components; among them, the fiber components include different types of carbon fibers, such as the type, orientation, and length of the fibers; the matrix components include different types of resins, such as epoxy resin, polyester resin, etc.; the filler components include inorganic fillers such as silicate, calcium carbonate, etc., and organic fillers such as lignin, polymer particles, etc.; the other components include additives and modifiers that may affect the performance of the composite material, such as toughening agents, flame retardants, etc.

[0016] Preferably, the corresponding characteristics of each functional component include: elastic modulus and elongation at break of the fiber component; glass transition temperature and curing shrinkage of the matrix component; particle size distribution and aspect ratio of the filler component.

[0017] Preferably, after performing weighted average calculation and addition and subtraction operations on the corresponding characteristics of each functional component, the weighted average characteristics and the operation characteristics are normalized to ensure that different characteristics are compared under the same dimension.

[0018] Preferably, the normalization method includes but is not limited to minimum-maximum normalization and Z-score normalization, wherein the specific calculation formula of the Z-score normalization method is:

[0019]

[0020] Among them, X new is the normalized feature, x is the initial feature before normalization, μ is the mean of the feature, and σ is the standard deviation of the feature.

[0021] Preferably, the calculation formula of the Pearson correlation coefficient between any two features in the initial features is:

[0022]

[0023] Among them, r X,Y is the correlation between features X and Y; X i 、Y i are the sample values of features X and Y respectively; are the means of features X and Y respectively.

[0024] Preferably, the use of the gradient boosting decision tree regression algorithm to construct a prediction model includes hyperparameter optimization: the decision tree depth range is 3-8 layers, the learning rate range is 0.01-0.3, the number of iterations is 50-200 times, and the optimal parameter combination is determined by grid search so that the correlation coefficient between the predicted value and the measured value of the test set reaches above 0.95.

[0025] Preferably, the hyperparameter optimization specifically includes: when the base learner is a decision tree with a depth of 5, the optimal learning rate is 0.1; when the number of iterations reaches 150, the RMSE of the validation set decreases and tends to converge.

[0026] Preferably, the construction of the prediction model using the gradient boosting decision tree regression algorithm further includes model training: the ten-fold cross-validation method is used to evaluate the training set. During each training process, the root mean square error and the correlation coefficient are calculated, and based on this, the hyperparameters of the gradient boosting decision tree regression model are adjusted. The hyperparameters are continuously optimized. When the root mean square error is minimized and the correlation coefficient is maximized, the prediction effect of the model reaches the optimal, and finally the optimal machine learning model for predicting the room temperature tensile strength of carbon fiber composites is obtained.

[0027] The beneficial effects of the present invention are as follows: by organically combining the feature extraction technology with the machine learning algorithm, the present invention deepens the understanding of the microstructure of the composite material, thereby improving the accuracy and reliability of the prediction of the tensile strength of carbon fiber composites. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0029] Figure 1 It is a flowchart of the steps of the method for predicting the tensile strength of carbon fiber composites based on machine learning of the present invention.

[0030] Figure 2 It is a flowchart of the method for predicting the tensile strength of carbon fiber composites based on machine learning of the present invention.

[0031] It should be noted that the drawings are not necessarily drawn to scale, but are only shown in a schematic manner that does not affect the understanding of the reader. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0033] It should be further understood that the term " / and" used in the specification and claims of the present invention refers to any combination and all possible combinations of one or more of the related listed items, and includes these combinations.

[0034] Such as Figure 1 and Figure 2As shown, the method for predicting the tensile strength of carbon fiber composites based on machine learning disclosed in the embodiments of the present application includes the following steps:

[0035] I. Step 1: Data collection

[0036] S101. Construct a data set containing 120 groups of carbon fiber composite samples through the following method:

[0037] Material composition:

[0038] ① Fiber component: T800 high-strength carbon fiber (tensile modulus 294 GPa) and M40J high-modulus carbon fiber (modulus 377 GPa);

[0039] ② Matrix component: Bisphenol A epoxy resin (E-51, curing agent 4,4'-diaminodiphenyl sulfone) and thermoplastic polyimide resin (PI-3);

[0040] ③ Filler component: Calcium carbonate (average particle size 75 μm, purity ≥ 99%) and nano-montmorillonite (particle size 50 - 100 nm);

[0041] ④ Interface modification component: γ-aminopropyltriethoxysilane (KH550) and titanate coupling agent (NDZ-201);

[0042] The total mass percentage of each component is strictly controlled within an error range of 100% ± 0.2%.

[0043] S102. Tensile strength test:

[0044] ① Specimen preparation:

[0045] Prepare standard specimens using a compression molding process, with dimensions of 250 mm × 25 mm × 2 mm (length × width × thickness), and the fiber orientation parallel to the tensile direction.

[0046] ② Test conditions:

[0047] According to ASTM D3039 standard, test using an INSTRON 5969 universal testing machine in an environment of 23 ± 0.5 °C and humidity 50 ± 3% RH (compliant with ISO 291 standard).

[0048] ③ Loading parameters:

[0049] Apply a tensile load at a constant rate of 1 mm / min until the specimen fractures, and record the stress-strain curve and the maximum load value F max .

[0050] ④ Strength calculation:

[0051] Calculate the tensile strength σ according to the following formula t :

[0052]

[0053] (σ t Unit: MPa, F max Unit: N, A0 unit: mm 2 ), where the initial cross-sectional area A0 of the specimen = 50 mm 2 .

[0054] S103. Data recording specification:

[0055] Table 1 Typical sample data

[0056]

[0057] Note: The test results are the average of 3 parallel tests, and the relative standard deviation ≤ 1.5%

[0058] S104. Dataset construction:

[0059] ① Data division: Divide 120 groups of data into a training set (96 groups) and a validation set (24 groups) according to a 4:1 ratio;

[0060] ② Distribution characteristics:

[0061] Gradient of carbon fiber mass fraction: 30%, 40%, 50%, 60%;

[0062] Fixed proportion of coupling agent 5%;

[0063] The resin / filler ratio varies between 1:3 and 3:1;

[0064] Outlier handling: Use the 3σ criterion to eliminate measurement data with deviations exceeding ±3 standard deviations.

[0065] S105. Data preprocessing:

[0066] Perform normalization before model training:

[0067] II. Step 2: Component classification

[0068] S201 Functional component classification:

[0069] Classify the components of carbon fiber composites according to the following rules:

[0070] ① Fiber components:

[0071] Contain T800 high-strength carbon fiber (mass fraction 50 ± 0.2%) and M40J high-modulus carbon fiber (mass fraction 60 ± 0.3%), as the main load-bearing phase, and their orientation distribution directly affects the anisotropy of the composite material;

[0072] ②Matrix component:

[0073] It contains E-51 epoxy resin (mass ratio: 30±0.1%) and PI-3 polyimide resin (mass ratio: 35±0.2%), bonds fibers and fillers, transfers stress and protects the reinforcing phase;

[0074] ③Filler component:

[0075] It contains calcium carbonate (mass ratio: 15±0.1%) and nano-montmorillonite (mass ratio: 10±0.1%), adjusts interface properties and inhibits crack propagation;

[0076] ④Interface modification component:

[0077] It contains KH550 silane coupling agent (mass ratio: 5±0.05%) and NDZ-201 titanate (mass ratio: 5±0.05%), improves the interfacial bonding strength between fiber and matrix (the bonding force is increased by 38% measured by atomic force microscope).

[0078] S202 Mechanical characteristic calculation:

[0079] Obtain characteristic parameters through finite element simulation and experimental testing:

[0080] ①Fiber component characteristics:

[0081] Elastic modulus: 294 GPa (T800), 377 GPa (M40J) (tested according to ASTM D4018 standard);

[0082] Elongation at break: 1.8±0.1% (T800), 1.5±0.1% (M40J) (obtained through INSTRON5969 tensile test);

[0083] ②Matrix component characteristics:

[0084] Glass transition temperature: 120±2 °C (E-51), 250±5 °C (PI-3) (tested by DSC, heating rate 10 °C / min);

[0085] Curing shrinkage rate: 3.2±0.3% (E-51), 1.5±0.2% (PI-3) (measured by laser displacement sensor, accuracy 0.1 μm);

[0086] ③Filler component characteristics:

[0087] Calcium carbonate particle size distribution: D10 = 45 μm, D50 = 75 μm, D90 = 110 μm (tested by Malvern Mastersizer 3000);

[0088] Aspect ratio of nano-montmorillonite: 82 ± 5 (analyzed by transmission electron microscopy images, according to JIS Z8901 standard).

[0089] S203 Structured data construction:

[0090] Table 2 Functional component feature database

[0091]

[0092] S204 Feature weighting processing:

[0093] Fiber feature weighting:

[0094]

[0095] Matrix feature operation:

[0096]

[0097] III. Step 3: Feature calculation

[0098] S301 Multi-scale modeling:

[0099] ① Microscopic representative volume element (RVE) modeling:

[0100] Use Digimat-FE 2023 software to construct a three-phase RVE model (size 50μm × 50μm × 50μm);

[0101] Fiber phase: T800 carbon fiber (diameter 7μm, randomly distributed, volume fraction 60%);

[0102] Matrix phase: E-51 epoxy resin (coating the fiber surface, thickness 2μm);

[0103] Filler phase: calcium carbonate particles (particle size 75μm, volume fraction 15%);

[0104] Mesh generation: hexahedral elements (C3D8R, size 1μm) are used in the fiber region, and tetrahedral elements (C3D4, size 3μm) are used in the matrix / filler region;

[0105] ② Macroscopic laminate modeling:

[0106] Establish a [0° / 90°] symmetric ply structure (single ply thickness 0.125mm, total thickness 2mm) in ABAQUS2022;

[0107] Material properties:

[0108] Fiber layer: Elastic modulus E 11 =294GPa, E 22 =13.7GPa, G 12= 5.2 GPa;

[0109] Matrix layer: Elastic modulus E m = 3.2 GPa, Poisson's ratio ν = 0.35;

[0110] S302 fiber characteristic calculation:

[0111] ① Elastic modulus:

[0112] Apply x-direction displacement boundary condition (U x = 0.1% Lx) to the RVE model and calculate the reaction force:

[0113]

[0114] The deviation from the measured value of ASTM D4018 (294 ± 5 GPa) is < 0.3%;

[0115] ② Elongation at break:

[0116] Define Cohesive contact at the fiber-matrix interface (critical stress σ c = 80 MPa, fracture energy G c = 0.5 N / mm),

[0117] When the damage factor D = 0.95, record the fiber axial strain as 1.8 ± 0.05%;

[0118] S303 matrix characteristic calculation:

[0119] ① Glass transition temperature:

[0120] Fit the time-temperature equivalence curve (WLF equation) through DSC test data:

[0121]

[0122] When the storage modulus drops to 10 g then determine T g = 120 ± 1 °C;

[0123] ② Curing shrinkage rate:

[0124] Simulate the curing reaction in the thermo-mechanical coupling field (curing degree α from 0 to 1):

[0125] ΔV / V0 = β(1 - α) (β = 0.032),

[0126] The final volume shrinkage rate is 3.2 ± 0.1%, and the deviation from the measured value by laser interferometry is < 0.5%;

[0127] S304 filler characteristic calculation:

[0128] ① Particle size distribution:

[0129] Perform Avizo Fire image processing on the micro-CT scan data (resolution 1 μm 3 ):

[0130] D10 = 45 ± 2 μm, D50 = 75 ± 3 μm, D90 = 110 ± 5 μm

[0131] The correlation coefficient R with the test results of Malvern Mastersizer 3000 2 = 0.98R;

[0132] ②Aspect ratio:

[0133] Measure 100 nano-montmorillonite particles through TEM images (JEM-2100F, 200 kV):

[0134]

[0135] S305 feature verification:

[0136] Input the obtained feature parameters into the GBDT model and test on the validation set (24 groups of data):

[0137]

[0138] IV. Step 4: Feature processing

[0139] S401 Component weight assignment:

[0140] Set the typical ratio according to the mass percentage limit:

[0141] Fiber component: T800 carbon fiber (mass 50% ± 0.2%)

[0142] Matrix component: E-51 epoxy resin (mass 30% ± 0.1%)

[0143] Filler component: calcium carbonate (mass 15% ± 0.1%)

[0144] Interface modification component: KH550 silane (mass 5% ± 0.05%)

[0145] Overall verification: 50% + 30% + 15% + 5% = 100% (error ≤ 0.25%)

[0146] S402 Weighted feature calculation:

[0147] Perform weighting according to the defined features:

[0148] ①Fiber features:

[0149]

[0150] ②Matrix characteristics:

[0151]

[0152] ③Filler characteristics:

[0153]

[0154] AR ω = 82 × 0.15 = 12.3

[0155] S403 Operation characteristic generation:

[0156] Addition and subtraction operations between characteristics:

[0157] ①Modulus difference characteristic:

[0158] ΔE f-m = 294 PGPa - 3.2 GPa = 290.8 GPa

[0159] ②Temperature - shrinkage correlation characteristic:

[0160] T g ×S m = 120 °C × 3.2% = 384 °C·%

[0161] ③Aspect ratio - particle size ratio:

[0162]

[0163] S404 Normalization processing:

[0164] Perform Z - score standardization:

[0165] ①Calculate the training set statistics (96 groups of data):

[0166]

[0167] ②Standardization calculation example:

[0168]

[0169] S405 Feature pool construction:

[0170] Generate a structured data set containing the following features:

[0171]

[0172] V. Step 5: Feature screening

[0173] S501 Pearson correlation coefficient calculation:

[0174] Calculate the Pearson correlation coefficient between any two features, X and Y. The formula is as follows:

[0175]

[0176] Key feature pair detection:

[0177]

[0178] S502 Feature importance evaluation:

[0179] ① Evaluation method:<s

[0180] According to the GBDT model, calculate the feature contribution:

[0181] # Parameter settings: max_depth = 5, learning_rate = 0.1, n_estimators = 150

[0182] gbdt = GradientBoostingRegressor()

[0183] gbdt.fit(X_train, y_train)

[0184] importance = gbdt.feature_importances_

[0185] ② Sorting results:

[0186] Feature Importance Score Rank Fiber Elastic Modulus Weighted Value 0.62 1 Curing Shrinkage Rate Calculation Feature 0.18 2 Aspect Ratio - Particle Size Ratio 0.12 3

[0187] S503 Iterative feature selection:

[0188] Optimize using ten-fold cross-validation:

[0189] ① Parameter settings:

[0190] Stepwise forward selection: Add 1 feature each time

[0191] Evaluation metrics: RMSE (preferred) and R 2 (Auxiliary)

[0192] ② Performance comparison:

[0193] Characteristic Number RMSE (MPa) <![CDATA[R 2 > Selected Parameters (Depth / Learning Rate / Number of Iterations) 3 15.2 0.958 5 / 0.1 / 150 5 12.7 0.976 5 / 0.1 / 150

[0194] S504 Optimal subset determination:

[0195] Optimal performance is achieved when the number of features increases to 5: weighted value of fiber elastic modulus; operation feature of glass transition temperature; aspect ratio - particle size ratio; weighted value of curing shrinkage rate; modulus difference feature.[[ID=**66]]

[0196] S505 Model Verification:

[0197] Index Full Feature Set Optimal Subset Improvement Magnitude RMSE (MPa) 14.9 12.7 14.8％ <![CDATA[Coefficient of correlation R 2 > 0.963 0.976 1.3％

[0198] VI. Step 6: Model Construction and Optimization

[0199] S601 Model Initialization:

[0200] Initialize the GBDT model according to the defined parameters:

[0201] gbdt = GradientBoostingRegressor(

[0202] max_depth = 5, # Decision tree depth

[0203] learning_rate = 0.1, # Learning rate

[0204] n_estimators = 150 # Number of iterations )

[0206] S602 Residual Correction Training:

[0207] Execute the gradient descent strategy:

[0208] ① Initial predicted value calculation:

[0209]

[0210] ② Iterative update (m = 1 to 150):

[0211] r im = y i - F m-1 (x i )(Residual of the m-th round)

[0212] (Decision tree for fitting the residual)

[0213] F m (x) = F m-1 (x)+0.1·h m (x)(Learning rate 0.1)

[0214] S603 Cross-Validation Optimization:

[0215] Execute ten-fold cross-validation:

[0216] Fold Number Training Set RMSE (MPa) Validation Set RMSE (MPa) 1 13.2 14.5 ... ... ... 10 12.9 13.8

[0217] Average performance: RMSE of the validation set = 13.1 ± 0.7 MPa, R 2 = 0.972

[0218] S604 Early Stopping Mechanism:

[0219] When the decrease in the RMSE of the validation set is less than 1% for 5 consecutive iterations, terminate the training and retain the optimal model:

[0220] Iteration Interval RMSE Change Rate Whether to Terminate 120-125 0.8％ No 125-130 0.6％ No 130-135 0.4％ Yes

[0221] S605 Final Model Deployment:

[0222] Lock the model parameters according to the performance criteria:

[0223] Parameter Item Optimal Value Decision Tree Depth 5 Learning Rate 0.1 Number of Iterations 135

[0224] For the embodiments of the present invention, it should also be noted that, without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0225] As described above, it is only the preferred embodiments of the present invention, and it does not impose any formal limitations on the present invention. The protection scope of the present invention should be subject to the protection scope of the claims. Although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content to make equivalent embodiments of equivalent changes, but as long as it does not depart from the technical content of the present invention, any simple modifications, equivalent changes and modifications made to the above embodiments according to the technical essence of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for predicting the tensile strength of carbon fiber composite materials based on machine learning, characterized in that It includes the following steps: Collect all the original component data and the room-temperature tensile strength data of each original component contained in the carbon fiber composite material to form a data set; Classify all the original components as functional components according to the functions and roles of each original component in the carbon fiber composite material; Calculate the mechanical properties of each functional component by numerical simulation methods to generate the corresponding characteristics of each functional component; Taking the component ratio as the weight, perform weighted average operations on the corresponding characteristics of each functional component to obtain weighted average characteristics, and perform addition and subtraction operations to obtain operation characteristics. Use the weighted average characteristics and the operation characteristics as initial characteristics; Calculate the Pearson correlation coefficient between any two of the initial characteristics. If the correlation coefficient is greater than 0.9, remove the redundant characteristics and only retain one of them. Construct a feature pool and sort the importance of each feature in the feature pool; Select the optimal sub-feature set from the feature pool through an iterative optimization algorithm; Taking the optimal sub-feature set as the input independent variable and the room-temperature tensile strength of the carbon fiber composite material as the output dependent variable, use the gradient boosting decision tree regression algorithm to construct a prediction model, and correct the prediction residuals by gradient descent to minimize the root mean square error between the predicted value and the measured value of the room-temperature tensile strength output by the model.

2. The method according to claim 1, wherein All the original components contained in the carbon fiber composite material include carbon fiber, epoxy resin, polyimide resin, calcium carbonate filler, silane coupling agent and other modified additives that affect the tensile strength, and the sum of the mass percentages of each component is 100%.

3. The method according to claim 1, wherein The functional components include: fiber components, matrix components, filler components and other components; among them, the fiber components include different types of carbon fibers, the matrix components include different types of resins, the filler components include inorganic fillers and organic fillers, and the other components include additives and modifiers that may affect the properties of the composite material.

4. The method according to claim 1, wherein The corresponding characteristics of each functional component include: the elastic modulus and elongation at break of the fiber component; the glass transition temperature and curing shrinkage rate of the matrix component; the particle size distribution and aspect ratio of the filler component.

5. The method according to claim 1, wherein, After performing weighted average operations and addition and subtraction operations on the corresponding characteristics of each functional component, perform normalization processing on the weighted average characteristics and the operation characteristics.

6. The method according to claim 5, characterized in that, The methods of the normalization processing include but are not limited to min-max normalization and Z-score standardization.

7. The method according to claim 1, wherein The calculation formula for the Pearson correlation coefficient between any two of the initial characteristics is: ， wherein, is the correlation between feature X and Y; , are the sample values of feature X and Y respectively; , are the means of feature X and Y respectively.

8. The method according to claim 1, wherein The construction of the prediction model using the gradient boosting decision tree regression algorithm includes hyperparameter optimization: the decision tree depth range is 3-8 layers, the learning rate range is 0.01-0.3, and the number of iterations is 50-200 times. Determine the optimal parameter combination through grid search to make the correlation coefficient between the predicted value and the measured value of the test set reach more than 0.

95.

9. The method according to claim 8, characterized in that, The specific hyperparameter optimization includes: When the base learner is a decision tree with a depth of 5, the optimal learning rate is 0.1; when the number of iterations reaches 150 times, the RMSE of the validation set decreases and tends to converge.

10. The method according to claim 9, characterized in that, The construction of the prediction model using the gradient boosting decision tree regression algorithm also includes model training: the ten-fold cross-validation method is used for training set evaluation. During each training process, the root mean square error and the correlation coefficient are calculated, and based on this, the hyperparameters of the gradient boosting decision tree regression model are adjusted. The hyperparameters are continuously optimized. When the root mean square error is minimized and the correlation coefficient is maximized, the prediction effect of the model reaches the optimal state, and finally the optimal machine learning model for predicting the room temperature tensile strength of carbon fiber composites is obtained.

Images