A machine learning based prediction method for grain boundary twinning behavior
By combining machine learning with electron backscatter diffraction technology, a grain boundary twin behavior prediction model was constructed, which solved the problem that existing technologies failed to fully capture the nonlinear relationships of multiple influencing factors, and achieved high-precision grain boundary twin behavior prediction, thereby improving the accuracy and reliability of material performance prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING INST OF NEW ENE STOR MATER & EQUIP
- Filing Date
- 2025-11-28
- Publication Date
- 2026-07-21
Smart Images
Figure SMS_1 
Figure HDA0005712260410000011 
Figure HDA0005712260410000012
Abstract
Description
Technical Field
[0001] This invention relates to the interdisciplinary field of materials science and machine learning, specifically to a method for predicting grain boundary twin behavior based on machine learning. Background Technology
[0002] Twinning is a key plastic deformation mechanism in hexagonal close-packed (HCP) metals (such as magnesium alloys), essentially involving uniform shearing of a portion of a crystal along specific crystal planes and directions. This process often originates at grain boundaries, and its microscopic product is the "twin." After nucleation at the grain boundary, the twin extends along a specific crystallographic direction. Based on different morphologies at the grain boundary, twinning behavior can be divided into three categories: (1) no twins, i.e., no twins appear on either side of the grain boundary; (2) twin termination, i.e., the twin stops extending at the grain boundary or only nucleates and grows on one side; (3) twin propagation, i.e., the twin crosses the grain boundary, inducing the formation of new twins in adjacent grains, or the twin nucleates at the grain boundary and forms a continuous twin structure on both sides. Twinning behavior at grain boundaries directly affects the macroscopic mechanical properties and fracture mechanisms of materials, therefore, accurate prediction of it is of great significance.
[0003] Traditional research methods primarily rely on different theoretical frameworks, such as Schmidt's law based on the ease of dislocation slip and geometric compatibility factors based on interfacial strain coordination. However, these theories or analytical equations often focus on single or limited physical factors, leading to inconsistent or even conflicting predictions of the same grain boundary twinning behavior. In practical analysis, researchers must comprehensively weigh the inferences of different theories and rely heavily on their personal theoretical knowledge and experience for subjective decisions, resulting in a non-standardized and low-reproducibility evaluation process. This heavy reliance on researcher experience not only significantly raises the technical threshold but also leads to inconsistent predictions, making it difficult to construct a unified and reliable evaluation system. More importantly, these traditional models based on simplifying assumptions are inherently unable to fully capture the complex nonlinear relationships and interactions among multiple influencing factors, thus often failing to balance the accuracy of macroscopic performance predictions with the consistency of microscopic mechanism explanations. Summary of the Invention
[0004] In view of the above-mentioned shortcomings of the existing technology, the purpose of this invention is to provide a grain boundary twin behavior prediction method based on machine learning, which solves the problem that the existing grain boundary twin behavior prediction model does not fully consider the nonlinear relationship and interaction of various grain boundary twin behavior influencing factors, resulting in inaccurate prediction of twin behavior at grain boundaries.
[0005] The technical solution adopted in this invention is as follows:
[0006] A machine learning-based method for predicting grain boundary twin behavior includes the following steps:
[0007] S1. Plastic deformation of the polycrystalline metal sample and electron backscatter diffraction technique to obtain microstructure data of a certain area of the deformed sample.
[0008] S2. Process the microstructure data, extract the geometric parameters and crystallographic orientation parameters of the grain boundary region, and identify the twinning behavior type at the grain boundary;
[0009] S3. Based on the extracted parameters, calculate multiple feature parameters related to the twinning behavior of each grain boundary. Using a single grain boundary as a sample unit, associate the multiple feature parameters corresponding to each grain boundary with the twinning behavior type label identified in step S2 to construct the original feature dataset.
[0010] S4. Perform data balancing and standardization on the original feature dataset;
[0011] S5. Using the sample units as units, the processed dataset is divided into a training set and a test set. Based on the training set, the machine learning model is optimized for hyperparameters using an error analysis iterative optimization method to obtain the optimal hyperparameter configuration. Then, the machine learning model with the optimal hyperparameters is trained using the training set to obtain a trained prediction model.
[0012] S6. Input the feature parameters of each grain boundary sample in the test set into the trained prediction model one by one to obtain the prediction results of the twin behavior type corresponding to each grain boundary; then, compare the prediction results of each grain boundary with the corresponding real twin behavior type label in the test set one by one to quantitatively evaluate the prediction performance of the model.
[0013] S7. Obtain the microstructure data of the test area of the metal polycrystalline sample to be predicted before deformation using electron backscatter diffraction technology, and process the microstructure data to extract the geometric parameters and crystallographic orientation parameters of the grain boundary region of the test area; the metal polycrystalline sample to be predicted is the same metal polycrystalline sample as the metal polycrystalline sample in step S1.
[0014] S8. Based on the parameters extracted in step S7, calculate multiple characteristic parameters related to the twinning behavior of each grain boundary. After standardizing the calculated characteristic parameters of each grain boundary, input them into the prediction model after quantitative evaluation. This will allow the prediction of the twinning behavior type at the grain boundary after plastic deformation of the test area with the same strain as in step S1.
[0015] Furthermore, in step S1, the plastic deformation includes at least one of tension, compression, bending, or shear.
[0016] Furthermore, in steps S2 and S7, the method for processing the microstructure data specifically involves: performing noise reduction and grain division processing on the microstructure data.
[0017] Furthermore, in step S1, the certain region is located inside the polycrystalline sample.
[0018] Furthermore, in steps S3 and S8, the multiple characteristic parameters each include: the average value of the minimum active shear stress of grain basal plane slip on both sides of the grain boundary, the absolute value of the difference between the minimum active shear stresses of grain basal plane slip on both sides of the grain boundary, the average value of the minimum active shear stresses of grain tensile twinning on both sides of the grain boundary, the absolute value of the difference between the minimum active shear stresses of grain tensile twinning on both sides of the grain boundary, the maximum deformation compatibility factor between the basal plane slip variants of grains on both sides of the grain boundary, the maximum deformation compatibility factor between the basal plane slip and tensile twinning variants of grains on both sides of the grain boundary, and the maximum deformation compatibility factor between the tensile twinning variants of grains on both sides of the grain boundary. The following parameters are considered: grain boundary length, average grain size on both sides of the grain boundary, absolute value of the difference in grain size on both sides of the grain boundary, angle between the grain boundary trace on the observation surface and the direction of the applied stress, average angle between the shear direction of the basal slip with the maximum Schmitt factor on both sides of the grain boundary and the grain boundary trace, difference in the angle between the shear direction of the basal slip with the maximum Schmitt factor on both sides of the grain boundary and the grain boundary trace, average angle between the shear direction of the tensile twin with the maximum Schmitt factor on both sides of the grain boundary and the grain boundary trace, and difference in the angle between the shear direction of the tensile twin with the maximum Schmitt factor on both sides of the grain boundary and the grain boundary trace.
[0019] Furthermore, in step S4, the data equalization is to process the original feature dataset using oversampling or undersampling methods; in steps S4 and S8, the standardization process is to process the equalized data using Z-score standardization, maximum-minimum scaling, absolute value scaling, or norm-based scaling methods.
[0020] Furthermore, in step S1, the strain of the plastic deformation is 2-8%.
[0021] Furthermore, in step S5, the machine learning model is one of artificial neural networks, support vector machines, decision trees, random forests, Naive Bayes, or gradient boosting decision trees.
[0022] Furthermore, in step S5, the error analysis iterative optimization method is a method based on grid search combined with cross-validation.
[0023] Furthermore, the method is applicable to the prediction of grain boundary twinning behavior in close-packed hexagonal metallic polycrystalline materials.
[0024] Compared with the prior art, the present invention has the following beneficial effects:
[0025] This invention provides a machine learning-based method for predicting grain boundary twinning behavior. Based on microstructure data obtained through electron backscatter diffraction (EBSD) technology, this method integrates 15 key feature parameters, including grain orientation, grain boundary characteristics, and deformation mechanism parameters, to construct a machine learning prediction model. Compared to traditional methods relying on empirical formulas or simplified assumptions, this invention effectively captures the complex nonlinear relationships and interactions among multiple factors, thereby significantly improving the accuracy of predicting grain boundary twinning behavior and enhancing the generalization ability of the prediction model. This method not only provides strong technical support for the study of the microscopic plastic deformation mechanism of materials but also provides a reliable theoretical basis for the mechanical property control and process optimization of high-performance metallic materials, possessing significant scientific and engineering application value. Attached Figure Description
[0026] Figure 1 These are microscopic data images of polycrystalline metal samples after electron backscatter diffraction characterization in the embodiments of the present invention; wherein, (a) is a microstructure distribution map, and (b) is a grain boundary type distribution map;
[0027] Figure 2 This is a data distribution diagram of each feature parameter in an embodiment of the present invention;
[0028] Figure 3 This is a heatmap of the Pearson correlation coefficients between various feature parameters in this embodiment of the invention.
[0029] Figure 4 This is a comparison chart of the prediction accuracy of different machine learning models in the embodiments of the present invention;
[0030] Figure 5 Here are the confusion matrices of the prediction results of two machine learning models selected in this embodiment of the invention: (a) random forest, (b) artificial neural network;
[0031] Figure 6 The images show the grain patterns of the polycrystalline metal sample before and after deformation; (a) before deformation, and (b) after deformation. Detailed Implementation
[0032] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings and specific examples. These descriptions should not be construed as limiting the scope of protection of the present invention.
[0033] Example
[0034] This embodiment provides a machine learning-based method for predicting grain boundary twinning behavior, and uses the compression deformation of an AZ31 magnesium alloy extrusion bar along the extrusion direction as an example to illustrate the specific implementation process of the method. The entire process includes the following steps:
[0035] S1. Raw Data Acquisition
[0036] (1) Sample preparation
[0037] A compression specimen measuring 7 mm (along the extrusion direction, ED) × 4 mm × 3 mm was cut from a commercial AZ31 magnesium alloy extrusion bar. The compression specimen was then placed in a strain with a strain rate of 10⁻³. s Under conditions of -1, the sample was compressed by 4% along the ED direction. To eliminate the influence of surface effects on grain boundary twinning behavior and ensure data quality, the characterization observation surface was chosen to be inside the sample. The sample was cut in half parallel to a 7mm × 4mm plane using wire electrical discharge machining (EDM) to obtain the internal cross-section as the characterization observation surface. This observation surface was then progressively ground and electropolished using commercially available magnesium alloy ACII polishing slurry to prepare a high-quality sample for electron backscatter diffraction (EBSD) analysis.
[0038] (2) EBSD characterization
[0039] EBSD characterization was performed on the central region of the sample's observation surface using a TESCAN MIRA3 scanning electron microscope equipped with an EBSD probe to obtain microstructural information. The characterization parameters were: voltage 20 kV, step size 1 μm, and characterization area 300 μm × 300 μm.
[0040] (3) Data processing
[0041] This step utilizes commercial crystallography analysis software (such as the MTEX open-source library based on MATLAB) to process the microstructure information in step S1-(2), mainly including:
[0042] ① Data preprocessing: The data undergoes noise reduction and grain segmentation to obtain structured data containing clear grain and grain boundary information. For example... Figure 1 As shown in (a), it can be observed that the grain orientation in the deformed sample mainly exhibits two distribution characteristics: one type is grains whose c-axis is roughly perpendicular to ED (blue or green), representing the original grains of the sample; the other type is strip-shaped grains whose c-axis is parallel to ED (red), representing twins generated by deformation.
[0043] ② Twin Behavior Recognition: Based on the preprocessed data, the twinning behavior type (no twins, twin termination, twin propagation) at each grain boundary is determined through manual identification or automatic discrimination based on grain orientation characteristics, and each grain boundary is labeled with a type label. The results are as follows: Figure 1 (b) shows the distribution of grain boundary types. Figure 1 (b) The sample shows that all three types of grain boundaries have a significant number distribution, which is representative and extensive for constructing a machine learning training dataset.
[0044] ③ Parameter extraction: Based on the preprocessed structured data, the geometric parameters and crystallographic orientation parameters of the grain boundary region are extracted. These parameters will serve as the basis for the next step of calculating the characteristic parameters related to grain boundary twinning behavior.
[0045] In addition, considering that the information on grains and grain boundaries (such as grain size and grain boundary length) at the edge of the EBSD scan area may be incomplete, the edge grain boundaries are marked separately and removed in subsequent analysis.
[0046] S2, Dataset Construction
[0047] (1) Through literature review and theoretical analysis, this invention identifies 15 characteristic parameters that are closely related to grain boundary twinning behavior, as shown in Table 1.
[0048] Table 1 describes the feature parameters in the dataset.
[0049]
[0050] In constructing the characteristic parameters, to more comprehensively characterize the physical mechanisms of grain boundary regions, the same physical concept can be quantified using different mathematical forms. For example, to characterize the influence of grain orientation on twinning behavior on both sides of the grain boundary, this invention not only calculates the Schmidt factor, a geometric parameter, but also further calculates the related shear stress, a mechanical parameter, thus providing complementary information from both geometric advantages and mechanical driving forces. Based on this design principle, this invention ultimately selected 15 key characteristic parameters as shown in Table 1 to comprehensively capture various factors affecting grain boundary twinning behavior.
[0051] (2) Based on the grain boundary-related parameters extracted in steps S1-(3) and the identified grain boundary type information, a total of 1012 grain boundary data instances were extracted. For each grain boundary, the 15 feature parameters listed in Table 1 were calculated. Then, using a single grain boundary as a sample unit, multiple feature parameters corresponding to each grain boundary and their corresponding identified twin behavior type labels were associated to construct the original feature dataset. The numerical distribution of each feature parameter is as follows: Figure 2 As shown. Figure 2 The data shows that these 15 features are generally distributed in the samples, indicating that they have good discriminative power and representativeness, and are suitable as input variables for machine learning models.
[0052] To further prevent model overfitting, improve generalization ability, and increase training efficiency, a heatmap of Pearson correlation coefficients between features was created, such as... Figure 3As shown in the figure, most features exhibit low correlation coefficients, indicating high independence among them and demonstrating the rationality and effectiveness of feature selection. Notably, MBB and MTT show a high correlation, which is based on a clear theoretical foundation to simultaneously consider the potential competition and synergy between different deformation mechanisms (basal slip and tensile twinning) in terms of geometric compatibility.
[0053] S3, Machine Learning Model Construction and Result Display
[0054] (1) Sample equalization processing
[0055] To address the imbalance in the number of grain boundary samples among the three types in the dataset constructed in step S2-(2), a random oversampling method is used to enhance the grain boundary types with fewer samples, ensuring that the number of samples among the three types of grain boundary is balanced, thereby avoiding the degradation of classification performance due to sample bias during model training. After oversampling, the dataset contains a total of 1515 grain boundaries.
[0056] (2) Data standardization processing
[0057] The balanced dataset is then standardized to eliminate the influence of differences in the units of measurement of each feature, thereby improving the stability and convergence speed of model training. Specifically, Z-score standardization, maximum-minimum scaling, absolute value scaling, or norm-based scaling methods can be used to standardize the data.
[0058] (3) Dataset partitioning
[0059] To ensure the trained machine learning model has good generalization ability and avoid overfitting due to insufficient data, the dataset used for model training should be of a certain size. Specifically, it is recommended that the total number of data samples in the training set be no less than 500. Based on this, in this embodiment, the standardized dataset is randomly divided into a training set and a test set at a ratio of 9:1, using individual grain boundary sample units as the unit. The training set is used for model training, and the test set is used for model performance evaluation.
[0060] (4) Preliminary model comparison and selection
[0061] This study selected six typical machine learning algorithms for performance comparison, including Decision Tree, Random Forest, Gradient Boosting Decision Tree (GBDT), Artificial Neural Network (ANN), Support Vector Machine (SVM), and Naive Bayes. Without adjusting hyperparameters, each model was trained and tested using the Sklearn toolkit in Python. During training, the training set was used to train each model. During testing, the feature parameters of each grain boundary sample in the test set were input into the trained prediction model to obtain the predicted twin behavior type for each grain boundary. Subsequently, the predicted result for each grain boundary was compared with its corresponding real twin behavior type label in the test set to quantitatively evaluate the model's predictive performance. The results are as follows: Figure 4 As shown.
[0062] from Figure 4 As can be seen, the prediction accuracies of decision trees, random forests, gradient boosting decision trees, and neural networks on the test set are all between 0.868 and 0.895, showing good performance. However, the accuracies of support vector machines and Naive Bayes models are only 0.632 and 0.500, respectively, significantly lower than the aforementioned four models, indicating their poor applicability in predicting grain boundary twinning behavior. Given that random forests and gradient boosting decision trees are both ensemble learning methods using decision trees as basic units, and considering both performance, random forests and artificial neural networks were ultimately selected as the core models for this study.
[0063] (5) Hyperparameter tuning
[0064] To further improve the model's prediction accuracy, and based on the aforementioned model error analysis results, we employ a grid search combined with cross-validation method to optimize and iterate the hyperparameters of the random forest and artificial neural network models. This process systematically traverses different hyperparameter combinations and evaluates the performance (i.e., error) of each parameter set on the cross-validation set, thereby finding the configuration that optimizes model performance. The final optimal hyperparameter combination is as follows:
[0065] Random Forest: Maximum tree depth is 15, and the number of trees is 500;
[0066] Artificial Neural Network: The activation function is the hyperbolic tangent function (tanh), the L2 regularization coefficient α is 0.01, and the maximum number of iterations is 2000.
[0067] (6) Model training and performance evaluation
[0068] Based on the optimal hyperparameter combination obtained in step S3-(5), the random forest and artificial neural network models are trained using the training set to obtain the prediction model. The feature parameters of each grain boundary sample in the 152 test sets are input into the trained prediction model one by one to obtain the prediction results of the twin behavior type corresponding to each grain boundary. Then, the prediction results of each grain boundary are compared with the real twin behavior type label corresponding to it in the test set to quantitatively evaluate the prediction performance of the model.
[0069] Figure 5 The prediction confusion matrices of the two models are shown. As can be seen from the figure, both models achieve a prediction accuracy exceeding 0.80 for the three types of grain boundary twinning behavior, indicating that the machine learning method proposed in this invention has high accuracy in the identification and classification of grain boundary twinning behavior. Statistical analysis and calculation of the prediction accuracies for the three types of grain boundary twinning behavior show that the overall prediction accuracies of the random forest and artificial neural network models reach 0.969 and 0.923, respectively, demonstrating excellent classification performance.
[0070] (7) Prediction of twinning behavior type at grain boundaries of undeformed samples
[0071] ① The microstructure data of a certain area of the AZ31 magnesium alloy extrusion bar before deformation was obtained by electron backscatter diffraction technology, and the microstructure data was processed to extract the geometric parameters and crystallographic orientation parameters of the grain boundary region.
[0072] ②Based on the parameters extracted in step S3-(7)-①, calculate multiple characteristic parameters related to the twinning behavior of each grain boundary, and after standardizing the calculated characteristic parameters of each grain boundary, input them into the prediction model after quantitative evaluation, so as to predict the twinning behavior type at the grain boundary after the test area undergoes plastic deformation of 4% along the ED direction.
[0073] It should be noted that the reliability of the prediction method of this invention is based on the following: when the strain of plastic deformation is small (2-8%), the changes in grain morphology (changes in the geometric parameters and crystallographic orientation parameters of grain boundaries) are negligible. That is, before and after deformation of the same polycrystalline metal sample, the values of the 15 characteristic parameters related to grain boundary twinning behavior calculated from the geometric parameters and crystallographic orientation of the grain boundaries remain unchanged. Therefore, by inputting the characteristic parameters of the sample's test area before deformation into the prediction model, the type of grain boundary twinning behavior can be accurately predicted.
[0074] Figure 6The grain patterns of AZ31 magnesium alloy extruded bars before and after deformation are shown. The figures reveal that the plastic deformation of this magnesium alloy is primarily caused by twinning and slip. Tensile twinning leads to an orientation rotation of approximately 86.3° in the crystal lattice, as shown in the red box area—a red twin region with a specific orientation relationship is generated within the original blue grain. Notably, the crystallographic orientation of the matrix region outside the twins remains largely unchanged, thus allowing for accurate identification and calculation of the crystallographic parameters before deformation. Figure 6 It was verified that the crystallographic orientation parameters of the polycrystalline sample remained essentially unchanged before and after deformation, and could be ignored.
[0075] On the other hand, although slip also causes lattice rotation, the rotation angle is usually very small (generally less than 2°), which is often negligible within the accuracy range of practical EBSD analysis. Therefore, from the perspective of crystallographic orientation, the orientation information that can be used to calculate characteristic parameters remains basically the same before and after deformation.
[0076] Regarding geometric parameters, since the strain used in this experiment was relatively small (4%), it did not cause significant grain coarsening or reconstruction, and the grain boundary morphology and grain size did not change significantly. Therefore, when constructing feature parameters, it can be assumed that geometric parameters such as grain boundary length and grain size remain consistent before and after deformation, and their minor changes have negligible impact on machine learning modeling.
[0077] In summary, this invention provides a simple, efficient, and highly accurate method for predicting grain boundary twinning behavior. When predicting the grain boundary twinning behavior type of a pre-deformation polycrystalline metallic sample after a certain amount of plastic deformation, it is only necessary to obtain the microstructure data of a specific region of the sample after plastic deformation under the same strain. After data processing, a prediction model can be constructed, which can then be used to predict the grain boundary twinning behavior type of other regions of the same polycrystalline metallic sample after plastic deformation under the same strain, greatly reducing the research and testing time. Furthermore, this method does not rely on traditional analytical equations, significantly lowering the technical threshold for research. Simultaneously, it systematically incorporates multiple key factors affecting grain boundary twinning behavior and their interactions, significantly improving the overall accuracy and adaptability of the prediction model. This provides a reliable technical path and theoretical support for in-depth research on the microscopic plastic deformation mechanism of materials and the design optimization of high-performance materials.
[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit the technical solutions. Those skilled in the art should understand that any modifications or equivalent substitutions to the technical solutions of the present invention without departing from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.
Claims
1. A method for predicting grain boundary twin behavior based on machine learning, characterized in that, Includes the following steps: S1. Plastic deformation of the polycrystalline metal sample and electron backscatter diffraction technique to obtain microstructure data of a certain area of the deformed sample. S2. Process the microstructure data, extract the geometric parameters and crystallographic orientation parameters of the grain boundary region, and identify the twinning behavior type at the grain boundary; the twinning behavior type includes no twinning, twin termination, and twin propagation; S3. Based on the extracted parameters, calculate multiple feature parameters related to the twinning behavior of each grain boundary. Using a single grain boundary as a sample unit, associate the multiple feature parameters corresponding to each grain boundary with the twinning behavior type label identified in step S2 to construct the original feature dataset. S4. Perform data balancing and standardization on the original feature dataset; S5. Using the sample units as units, the processed dataset is divided into a training set and a test set. Based on the training set, the machine learning model is optimized for hyperparameters using an error analysis iterative optimization method to obtain the optimal hyperparameter configuration. Then, the machine learning model with the optimal hyperparameters is trained using the training set to obtain a trained prediction model. S6. Input the feature parameters of each grain boundary sample in the test set into the trained prediction model one by one to obtain the prediction results of the twin behavior type corresponding to each grain boundary; then, compare the prediction results of each grain boundary with the corresponding real twin behavior type label in the test set one by one to quantitatively evaluate the prediction performance of the model. S7. Obtain the microstructure data of the test area of the metal polycrystalline sample to be predicted before deformation using electron backscatter diffraction technology, and process the microstructure data to extract the geometric parameters and crystallographic orientation parameters of the grain boundary region of the test area; the metal polycrystalline sample to be predicted is the same metal polycrystalline sample as the metal polycrystalline sample in step S1. S8. Based on the parameters extracted in step S7, calculate multiple characteristic parameters related to the twinning behavior of each grain boundary. After standardizing the calculated characteristic parameters of each grain boundary, input them into the prediction model after quantitative evaluation. This will allow the prediction of the twinning behavior type at the grain boundary after plastic deformation of the same strain as in step S1 in the area to be tested. In steps S3 and S8, the multiple characteristic parameters include: the average value of the minimum active shear stress of grain basal plane slip on both sides of the grain boundary, the absolute value of the difference between the minimum active shear stresses of grain basal plane slip on both sides of the grain boundary, the average value of the minimum active shear stresses of grain tensile twinning on both sides of the grain boundary, the absolute value of the difference between the minimum active shear stresses of grain tensile twinning on both sides of the grain boundary, the maximum deformation compatibility factor between the basal plane slip variants of grains on both sides of the grain boundary, the maximum deformation compatibility factor between the basal plane slip and tensile twin variants of grains on both sides of the grain boundary, the maximum deformation compatibility factor between the tensile twin variants of grains on both sides of the grain boundary, and the grain... The average value of the grain size on both sides of the grain boundary, the absolute value of the difference in grain size on both sides of the grain boundary, the angle between the grain boundary trace on the observation surface and the direction of the externally applied stress, the average value of the angle between the shear direction of the basal slip with the maximum Schmitt factor on both sides of the grain boundary and the grain boundary trace, the difference in the angle between the shear direction of the basal slip with the maximum Schmitt factor on both sides of the grain boundary and the grain boundary trace, the average value of the angle between the shear direction of the tensile twin with the maximum Schmitt factor on both sides of the grain boundary and the grain boundary trace, and the difference in the angle between the shear direction of the tensile twin with the maximum Schmitt factor on both sides of the grain boundary and the grain boundary trace; In step S1, the strain of the plastic deformation is 2-8%.
2. The method for predicting grain boundary twin behavior based on machine learning according to claim 1, characterized in that, In step S1, the plastic deformation includes at least one of tension, compression, bending, or shear.
3. The method according to claim 1, characterized in that, In steps S2 and S7, the method for processing the microstructure data specifically involves: performing noise reduction and grain division processing on the microstructure data.
4. The method for predicting grain boundary twin behavior based on machine learning according to claim 1, characterized in that, In step S1, the certain region is located inside the polycrystalline sample.
5. The method for predicting grain boundary twin behavior based on machine learning according to claim 1, characterized in that, In step S4, the data equalization is to process the original feature dataset using oversampling or undersampling methods; in steps S4 and S8, the standardization process is to process the equalized data using Z-score standardization, maximum-minimum scaling, absolute value scaling, or norm-based scaling methods.
6. The method for predicting grain boundary twin behavior based on machine learning according to claim 1, characterized in that, In step S5, the machine learning model is one of artificial neural networks, support vector machines, decision trees, random forests, Naive Bayes, or gradient boosting decision trees.
7. The method for predicting grain boundary twin behavior based on machine learning according to claim 1, characterized in that, In step S5, the error analysis iterative optimization method is a method based on grid search combined with cross-validation.
8. The method for predicting grain boundary twin behavior based on machine learning according to any one of claims 1 to 7, characterized in that, The method is applicable to the prediction of grain boundary twinning behavior in close-packed hexagonal metallic polycrystalline materials.