Method for predicting mechanical properties according to metal fracture morphology

Abstract

The invention discloses a method for predicting mechanical properties according to metal fracture morphology, which comprises the following steps: acquiring training data: carrying out tensile mechanical property test on a metal sample, and recording a test result which comprises metal mechanical property parameters and a fracture morphology graph; morphological feature marking: performing feature marking on the fracture morphological graph to form marked features; model construction: constructing a neural network model; model training: training a neural network model by taking the marked features as input features and the mechanical property parameters as output features, so that the neural network model can output corresponding metal mechanical property parameters according to the input marked features; and prediction application: obtaining the fracture morphology graph of the to-be-tested metal, marking the morphology features, inputting the formed marked features into the trained neural network model, and predicting and outputting the mechanical property parameters of the to-be-tested metal. The efficiency and accuracy of predicting the metal mechanical property according to the fracture morphology are improved.

Description

Technical Field

[0001] This invention relates to the field of material property prediction technology, and in particular to a method for predicting mechanical properties based on the fracture morphology of metals. Background Technology

[0002] Traditional mechanical property testing relies on destructive tensile tests, which are time-consuming and costly. Existing research shows that there is a correlation between the fracture morphology of metals and their mechanical properties. However, relying on human experience to visually judge the fracture morphology to determine mechanical properties is inefficient and highly subjective.

[0003] Therefore, the problem of how to improve the efficiency and accuracy of predicting the mechanical properties of metals based on fracture morphology remains to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a method for predicting mechanical properties of metals based on fracture morphology. The technical problem to be solved is how to improve the efficiency and accuracy of predicting the mechanical properties of metals based on fracture morphology.

[0005] To achieve the above objectives, the solution of the present invention is: a method for predicting mechanical properties based on the fracture morphology of metals, comprising the following steps: S1, Obtain training data: Perform tensile mechanical property tests on metal specimens and record the test results, including metal mechanical property parameters and fracture morphology diagrams. S2, Morphological feature marking: The fracture morphology map is marked with features to form marked features. The marked features include tensile temperature, dimple diameter, number of dimples per unit area, area ratio of dimple region, area ratio of cleavage region, number of extension steps per unit length, tear edge spacing, and number of tear edges per unit area. S3, Model Building: Building a neural network model; S4, Training the model: Using the labeled features as input features and the mechanical performance parameters as output features, the neural network model is trained so that it can output the corresponding metal mechanical performance parameters based on the input labeled features. S5, Predictive Application: Obtain the fracture morphology image of the metal to be tested, mark the morphological features, input the formed marked features into the trained neural network model, and predict and output the mechanical property parameters of the metal to be tested.

[0006] Furthermore, the mechanical properties of the metal include tensile strength and yield strength.

[0007] Further, in step S1, a tensile testing machine is used to perform tensile mechanical property tests on multiple standard specimens under multiple different tensile temperature conditions and the same strain rate conditions, respectively, to obtain the corresponding metal mechanical property parameters and fracture morphology diagrams. Then, feature marking is performed on each fracture morphology diagram to form marked features. The mechanical property parameters and marked features are matched and merged into a dataset. The dataset is divided into a training set, a test set, and a validation set according to a predetermined data quantity ratio. The training set is used for learning neural network model parameters, the test set is used for evaluating the generalization ability of the neural network model, and the validation set is used for hyperparameter tuning and model selection.

[0008] Furthermore, in step S4, hyperparameter tuning is performed. The hyperparameters include at least the number of hidden layers, the number of neurons per layer, the learning rate, the batch size, and the number of training iterations. The tuning method adopts Bayesian optimization. The importance of each input feature is analyzed, the influence of different labeled features on the predictive mechanical performance is evaluated, the performance of the final optimized model is evaluated using the test set, and the evaluation index between the predicted value and the true value is calculated, including the coefficient of determination and the Pearson correlation coefficient.

[0009] Furthermore, in steps S1 and S5, a scanning electron microscope is used to photograph the fracture surface to obtain a magnified fracture morphology image.

[0010] Furthermore, in steps S2 and S5, the required feature regions are annotated on the fracture morphology map using ImageJ to form marked features.

[0011] Furthermore, the metal mechanical property parameters and fracture morphology diagrams from step S1 were augmented using the Gaussian noise generation method.

[0012] Furthermore, in the aforementioned marking features: in step S2, the dimple region and the cleavage region are distinguished by the grayscale threshold method, and the area ratios are statistically analyzed to obtain the area ratios of the dimple region and the cleavage region.

[0013] The beneficial effects of the present invention after adopting the above scheme are as follows: a neural network model is trained by using the mechanical property parameters of metals and the marked features of the corresponding fracture morphology diagrams. The marked features include tensile temperature, dimple diameter, number of dimples per unit area, area ratio of dimple region, area ratio of cleavage region, number of extension steps per unit length, tear ridge spacing, and number of tear ridges per unit area. The marked features are used as input features, and the mechanical property parameters are used as output features to train the neural network model, so that the neural network model can output the corresponding mechanical property parameters of metals according to the input marked features, thereby improving the efficiency and accuracy of predicting the mechanical properties of metals based on fracture morphology. Attached Figure Description

[0014] Figure 1 This is a flowchart of the present invention; Figure 2 A heatmap showing the correlation between input features and mechanical properties; Figure 3 For feature weight map; Figure 4 This is a comparison chart of the predicted results. Detailed Implementation

[0015] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] Embodiments of the invention will now be described in full with reference to the accompanying drawings. It should be noted that the invention may be implemented in various forms and is not limited to the embodiments set forth herein. These embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0017] This invention provides a method for predicting mechanical properties based on the fracture morphology of metals, such as... Figure 1 As shown, it includes the following steps: S1, Obtain training data: Perform tensile mechanical property tests on metal specimens and record the test results. The test results include metal mechanical property parameters and fracture morphology diagrams. The mechanical property parameters of the metal include tensile strength and yield strength. Using a tensile testing machine, perform tensile mechanical property tests on multiple standard specimens under multiple different tensile temperature conditions and the same strain rate conditions, respectively, and obtain the corresponding metal mechanical property parameters and fracture morphology diagrams. Then, mark the features of each fracture morphology diagram to form marked features. Merge the mechanical property parameters and marked features into a dataset. Since experimental data is usually limited, in a preferred embodiment, the metal mechanical property parameters and fracture morphology diagrams in step S1 are augmented using a Gaussian noise generation method. This method expands the sample size by adding small Gaussian noise (the noise intensity is controlled to be 5% of the standard deviation of each feature) to the original data while retaining the distribution characteristics of the original data: First, calculate the mean and standard deviation of each feature, and then randomly select the original samples and add a Gaussian noise following the N(0, New samples are generated using Gaussian noise with a σ² distribution, where σ is 5% of the standard deviation of the corresponding feature. This method has three major advantages over traditional oversampling techniques: 1. It preserves the statistical characteristics and physical meaning of the original data; 2. It avoids introducing synthetic samples that deviate significantly from the true data distribution; 3. It has high computational efficiency and the expansion scale can be controlled. The sample distribution before and after data expansion is verified by a correlation heatmap to ensure that the correlation structure between key features is preserved. Finally, the dataset is divided into training, testing, and validation sets according to a predetermined data quantity ratio. The training set is used for learning neural network model parameters, the testing set is used for evaluating the generalization ability of the neural network model, and the validation set is used for hyperparameter tuning and model selection. S2, Morphological feature marking: The fracture morphology map is marked with features to form marked features. The marked features include tensile temperature, dimple diameter, number of dimples per unit area, area ratio of dimple region, area ratio of cleavage region, number of extension steps per unit length, tear edge spacing, and number of tear edges per unit area. S3, Model Building: Building a neural network model; S4, Model Training: Using labeled features as input features and mechanical performance parameters as output features, the neural network model is trained so that it can output the corresponding metal mechanical performance parameters based on the input labeled features. Specifically, hyperparameter tuning is performed, including at least: the number of hidden layers, the number of neurons per layer, the learning rate, the batch size, and the number of training iterations. Bayesian optimization is used for tuning. The importance of each input feature is analyzed, the influence of different labeled features on the predicted mechanical performance is evaluated, the performance of the final optimized model is evaluated using the test set, and evaluation metrics between predicted and true values ​​are calculated, including the coefficient of determination and the Pearson correlation coefficient. S5, Predictive Application: Obtain the fracture morphology image of the metal to be tested, mark the morphological features, input the formed marked features into the trained neural network model, and predict and output the mechanical property parameters of the metal to be tested.

[0018] In a preferred embodiment, in steps S1 and S5, a scanning electron microscope is used to photograph the fracture surface to obtain a magnified fracture morphology image.

[0019] In a preferred embodiment, in steps S2 and S5, the required feature regions are marked on the fracture morphology map using ImageJ to form marked features.

[0020] In a preferred embodiment, in step S2, the dimple region and the cleavage region are distinguished by the grayscale threshold method, and the area ratios are statistically analyzed to obtain the area ratios of the dimple region and the cleavage region.

[0021] The following is a more specific example: In this embodiment, the metal sample is made of rare earth magnesium alloy, with the mass fraction expressed as % as: Gd 9.55, Y 3.28, Zn 1.77, Zr 0.34, Mg balance. A total of 15 samples were used. The method for predicting mechanical properties based on the fracture morphology of the metal includes the following steps: S1. Tensile mechanical properties were tested on each metal sample, and the results were recorded. The results included the metal mechanical property parameters and fracture morphology images. The metal mechanical property parameters were tensile strength (UTS) and yield strength (YS). For the fracture morphology images, scanning electron microscopy (SEM) was used to obtain representative microstructural feature data. During observation, the scanning images of the fracture ends of the metal samples were set to 200x magnification. Starting from the center region of the fracture, one image was acquired every 0.2 mm along the radial direction, for a total of 5 images for each sample. This was to avoid errors caused by the inhomogeneity of microscopic features in different regions of the sample. S2, feature marking is performed on the fracture morphology image to form marked features, including: tensile temperature T, dimple diameter DD, number of dimples per unit area DDA, area ratio of dimple region DAF, area ratio of cleavage region CAF, number of extended steps per unit length SDL, tear ridge spacing TRS, and number of tear ridges per unit area TRDA. The images acquired in step S1 are calibrated using ImageJ software and quantitatively labeled. The marked feature data from 5 images acquired for each sample are averaged, and the result is used as the marked features for subsequent input of that sample. More specifically, for fracture surfaces with a porous network dimple structure, the dimple diameter is the maximum distance between the pit edges. Ten typical dimples are selected, their equivalent diameters are measured, and the average value is taken. The formula for calculating the number of dimples per unit area is: DDA = total number of dimples / field area. The dimple region and cleavage region are distinguished by grayscale thresholding, and their area ratios are statistically analyzed. The number of extended steps per unit length is the number of "river pattern branches" on the cleavage surface at 1... The number of tear ridges within a length of mm is determined by the fact that the cleavage zone accounts for a relatively small proportion, consists mostly of local small planes, and has extremely weak step density. Typical areas of the cleavage zone are selected for measurement. Tear ridges are "ridge-like protrusions" between dimples, and their spacing is strongly correlated with the dimple diameter; therefore, TRS ≈ dimple diameter. Tear ridges are distributed in a network within the field of view. Based on the "ridge continuity" judgment, the calculation formula is: TRDA = field area / total length of tear ridges. Through the above process, the obtained quantitative data provides a quantitative basis for subsequent material property analysis. The training data is shown in Table 2. Table 2 Training Dataset

[0022] By adding small Gaussian noise (noise intensity controlled at 5% of the standard deviation of each feature) to the original data, the sample size was expanded to 60 groups while preserving the original data distribution characteristics. The original input features included seven dimensions: T, DD, DDA, DAF, CAF, SDL, and TRDA, with UTS and YS as the target variables. Figure 2 The data matrix shown is from the correlation heatmap, with values ​​ranging from [ Between 1 and 1, positive numbers indicate positive correlation, and negative numbers indicate negative correlation. The larger the absolute value, the stronger the correlation. It can be seen that YS, UTS, SDL, DDA, and TRDA are mainly strongly positively correlated, while DD is strongly negatively correlated with DDA, SDL, UTS, and YS. Considering the differences in the scale of different features (e.g., DDA values ​​can reach 4800, while CAF values ​​are as low as 0), in order to ensure that each feature has the same order of magnitude and to improve the convergence speed and prediction stability of the subsequent model, the Z-score normalization method is used to preprocess the feature data. The remaining steps are the same as described above and will not be repeated here.

[0023] Feature importance is a metric used in the Random Forest algorithm to measure the contribution of each feature to the model's decision. A higher value indicates a more important feature. Figure 3 The results show that T, DD, and SDL make the largest contributions to the prediction of UTS and YS, with feature importance weights of 0.3106, 0.2753, and 0.2653, respectively.

[0024] Specifically, to achieve high-precision prediction of UTS and YS, this embodiment constructs a prediction system comprising three models: ridge regression, random forest, and neural network. The ridge regression model serves as the linear baseline model, employing L2 regularization to address potential multicollinearity issues between features. The regularization objective function is:

[0025] Where ω is the feature weight vector, α is the constraint term, X is the feature matrix, and y is the target value vector.

[0026] The random forest model employs an ensemble of 100 decision trees, controlling the complexity of the tree structure through a maximum depth limit (max_depth=5) to avoid overfitting. The model achieves ensemble learning through bootstrap sampling and random feature selection, demonstrating a strong ability to capture the non-linear relationship between UTS and YS.

[0027] The neural network model employs a three-layer feedforward neural network. The model structure is as follows: the input layer receives seven features, which are then passed through a fully connected layer using the ReLU activation function, and finally through another fully connected layer to output the result. The hidden layer size is optimized between 16 and 128 as a hyperparameter, and the learning rate is 1×10⁻⁶. -4~1 ×10 -2 Optimization within the specified range. The Adam optimizer is used, with the loss function defaulting to mean squared error (MSE), and the number of training epochs is fixed at 200. Hyperparameter optimization employs the BayesSearchCV method.

[0028] Five-fold cross-validation (K-Fold) was used to evaluate the generalization ability of each model to ensure the reliability of the evaluation in small sample scenarios. The evaluation indicators included the coefficient of determination (R²) and the Pearson correlation coefficient (R), and the formulas are as follows. The performance comparison of the three models is summarized in Table 3.

[0029] ; Where yi represents the true value of the i-th sample. R represents the predicted value of the i-th sample. 2 The closer the model is to 1, the better the fit.

[0030] ; in, R represents the mean of the variables, R=1 indicates perfect positive correlation, R= 1 indicates a completely negative correlation, and R=0 indicates no linear relationship.

[0031] Table 3 Summary of Model Performance Comparison

[0032] The tensile strength (UTS) and yield strength (YS) of the material were predicted using three machine learning models (ridge regression, random forest, and neural network), and the model performance was evaluated by scatter plots of the predicted and actual values.

[0033] Figure 4 The scatter plot compares the true and predicted values. The closer each data point is to the diagonal (y=x), the higher the model's prediction accuracy. From the overall distribution, the neural network model shows the best fit for both the UTS and YS target variables. Its data points are highly concentrated near the diagonal, indicating that it can effectively capture the complex nonlinear relationship between input features and mechanical properties.

[0034] In contrast, while random forest models also possess some nonlinear modeling capabilities, they still exhibit slight biases in high-value regions, and their prediction stability is slightly inferior to that of neural networks.

[0035] As a linear model, Ridge regression, limited by its assumptions, exhibits significant dispersion in both tasks, particularly in high-intensity regions, reflecting its limited ability to fit nonlinear relationships.

[0036] In conclusion, neural networks exhibited the strongest predictive ability under the experimental conditions and are suitable for such material property prediction tasks.

[0037] This invention predicts the UTS (tensile strength) and YS (yield strength) of rare-earth magnesium alloys by comprehensively applying data augmentation techniques and multi-model machine learning algorithms, particularly an integrated method based on random forests and optimized neural networks. Technically, addressing the limited original sample size (15 groups), this invention innovatively employs a Gaussian noise data augmentation method, increasing the sample size to 60 groups while preserving the original data distribution characteristics. Combined with feature importance analysis (revealing T, DD, and SDL as key influencing factors), it significantly improves the accuracy of mechanical property prediction (R² reaches 0.98), enabling the prediction model to adapt to the complex nonlinear relationship between fracture characteristic parameters and mechanical properties. Economically, the small-sample augmentation technique reduces reliance on large amounts of experimental data, lowers material preparation and mechanical testing costs, and shortens the development cycle of new materials. The application of this method can effectively achieve rapid evaluation of the mechanical properties of metallic materials, providing a theoretical basis for the optimized design of material microstructures, and has broad application prospects and significant practical implications in aerospace, automotive manufacturing, and other fields.

[0038] The above description is only a preferred embodiment of the present invention and is not intended to limit the design of this case. All equivalent changes made based on the key design features of this case shall fall within the protection scope of this case.

Claims

1. A method for predicting mechanical properties based on the fracture morphology of a metal, characterized in that, Includes the following steps: S1, Obtain training data: Perform tensile mechanical property tests on metal specimens and record the test results, including metal mechanical property parameters and fracture morphology diagrams. S2, Morphological feature marking: The fracture morphology map is marked with features to form marked features. The marked features include tensile temperature, dimple diameter, number of dimples per unit area, area ratio of dimple region, area ratio of cleavage region, number of extension steps per unit length, tear edge spacing, and number of tear edges per unit area. S3, Model Building: Building a neural network model; S4, Training the model: Using the labeled features as input features and the mechanical performance parameters as output features, the neural network model is trained so that it can output the corresponding metal mechanical performance parameters based on the input labeled features. S5, Predictive Application: Obtain the fracture morphology image of the metal to be tested, mark the morphological features, input the formed marked features into the trained neural network model, and predict and output the mechanical property parameters of the metal to be tested.

2. The method for predicting mechanical properties based on the fracture morphology of a metal as described in claim 1, characterized in that: The mechanical properties of the metal include tensile strength and yield strength.

3. The method for predicting mechanical properties based on the fracture morphology of a metal as described in claim 1, characterized in that: In step S1, a tensile testing machine is used to perform tensile mechanical property tests on multiple standard specimens under multiple different tensile temperature conditions and the same strain rate conditions, respectively, to obtain the corresponding metal mechanical property parameters and fracture morphology diagrams. Then, feature marking is performed on each fracture morphology diagram to form marked features. The mechanical property parameters and marked features are matched and merged into a dataset. The dataset is divided into a training set, a test set, and a validation set according to a predetermined data quantity ratio. The training set is used for learning neural network model parameters, the test set is used for evaluating the generalization ability of the neural network model, and the validation set is used for hyperparameter tuning and model selection.

4. The method for predicting mechanical properties based on the fracture morphology of a metal as described in claim 1, characterized in that: In step S4, hyperparameter tuning is also performed. The hyperparameters include at least the number of hidden layers, the number of neurons per layer, the learning rate, the batch size, and the number of training iterations. The tuning method is Bayesian optimization. The importance of each input feature is analyzed, the influence of different labeled features on the predictive mechanical performance is evaluated, the performance of the final optimized model is evaluated using the test set, and the evaluation index between the predicted value and the true value is calculated, including the coefficient of determination and the Pearson correlation coefficient.

5. The method for predicting mechanical properties based on the fracture morphology of a metal as described in claim 1, characterized in that: In steps S1 and S5, a scanning electron microscope is used to photograph the fracture surface to obtain a magnified image of the fracture morphology.

6. The method for predicting mechanical properties based on the fracture morphology of a metal as described in claim 1, characterized in that: In steps S2 and S5, the required feature regions are marked on the fracture morphology map using ImageJ to form marked features.

7. The method for predicting mechanical properties based on the fracture morphology of a metal as described in claim 1, characterized in that: The mechanical property parameters and fracture morphology diagrams of the metal in step S1 were augmented using the Gaussian noise generation method.

8. The method for predicting mechanical properties based on the fracture morphology of a metal as described in claim 1, characterized in that: In the aforementioned marking features: in step S2, the dimple region and the cleavage region are distinguished by the gray-scale threshold method, and the area ratios are statistically analyzed to obtain the area ratios of the dimple region and the cleavage region.

Images