High-toughness amorphous alloy component design method and system based on machine learning

Abstract

The invention discloses a high-toughness amorphous alloy component design method and system based on machine learning. The method comprises the following steps: (1) establishing a data set according to historical data; (2) dividing the data into a training set and a test set; (3) carrying out normalization processing on the data; (4) creating various machine learning models; (5) fitting the training set in the model, and adjusting hyper-parameters to obtain an optimal machine learning model; (6) predicting the test set in the model; (7) quantitatively evaluating the performance of various models, and comparing to obtain an optimal model; and (8) inputting component elements to obtain fracture toughness, and performing experimental verification. According to the method, a large number of existing data about the amorphous alloy components and the fracture toughness are excavated, the implicit complex relation between the components and the performance is unlocked through the machine learning technology, and the alloy components are rapidly and accurately designed according to the performance requirement. According to the method, the component design efficiency of the high-toughness amorphous alloy is improved, the trial and error cost is low, and the adaptability is high.

Description

Technical Field

[0001] The invention relates to a high-toughness amorphous alloy composition design method based on machine learning, and belongs to the technical field of metal material composition design. Background Art

[0002] Amorphous alloys have many excellent physical, chemical and mechanical properties, such as high strength, high hardness, high elasticity, low mechanical damping, corrosion resistance, wear resistance and excellent magnetic properties, due to the absence of crystal defects such as dislocations, grain boundaries and phase boundaries. The most prominent mechanical property advantage of amorphous alloys is high strength, and the most prominent mechanical property disadvantage is poor plasticity, which is also the main bottleneck for its application as a structural material.

[0003] Toughness is a comprehensive indicator of strength and plasticity. Some amorphous alloys have good toughness. 59.3 Cu 28.8 Nb 1.5 Al 10.4 The fracture toughness of amorphous materials reaches 114.3 MPam 1 / 2 Some amorphous alloys have very low toughness, such as amorphous Fe 51 Mn 10 Mo 14 Cr 4 C 15 B 6 Fracture toughness 3.5 MPam 1 / 2 How to design amorphous alloys with high toughness has become the focus and difficulty of studying amorphous alloys.

[0004] To date, the design of amorphous alloy compositions has mainly relied on experimental trial and error methods. There is a lack of models that can quantitatively describe the relationship between alloy composition and performance, making it difficult to quickly and accurately design alloy compositions based on given performance requirements, resulting in low efficiency, long cycles, and high costs in the development of new materials. Developing methods to accurately and quickly design alloy compositions based on given performance requirements has become a technical problem that urgently needs to be overcome in the field of metal material design. Based on this, it is necessary to find a new method for theoretical calculation and accurate prediction of fracture toughness.

[0005] As a typical data mining and analysis method, machine learning is an effective means to discover patterns and build models from existing data. In 2016, Raccuglia et al. from Harvard University used machine learning algorithms to analyze "abandoned" experimental data, and achieved a high degree of prediction of the formation conditions of organic template synthesis of inorganic substances, with a success rate of 89%, which is significantly higher than the efficiency of traditional synthesis schemes. At present, researchers have successively predicted the hardness, magnetism, modulus, amorphous formation ability, etc. of amorphous alloys through machine learning models, but toughness has not been studied. Because the fracture toughness of amorphous alloys has high requirements on sample size, only a small number of large-size samples can be prepared for testing. Using different machine learning methods to predict material properties on the large number of material data sets currently generated can effectively improve the accuracy of material performance prediction, save computing costs, and provide guidance for material experiments and applications. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for designing the composition of high-toughness amorphous alloys based on machine learning. By mining the data on amorphous composition and fracture toughness reported in existing literature, machine learning technology is used to unlock the implicit and complex relationship between "composition-performance", so as to achieve the purpose of quickly and accurately designing amorphous composition according to performance requirements and quickly predicting fracture toughness.

[0007] The technical solution implemented by the present invention is as follows: a method for designing high-toughness amorphous alloy composition based on machine learning, the steps are as follows: (1) Data construction: A data set was established based on publicly available literature data, including composition data of 85 amorphous alloys and their corresponding fracture toughness data.

[0008] (2) Data partitioning: Divide the data into a training set and a test set. The training set is used to train the model, and the test set is used to test the model.

[0009] (3) Data processing: normalize the data.

[0010] (4) Model creation: creating various machine learning models.

[0011] (5) Model training: Fit the training set in the machine learning model, obtain the optimal hyperparameters of the machine learning model through cross-validation and grid search, and thus obtain the machine learning model under the optimal hyperparameters.

[0012] (6) Model prediction: making predictions on the test set in the machine learning model.

[0013] (7) Model evaluation, using the regression model evaluation index R 2 The performance of various machine learning models was quantitatively evaluated with RMSE, and the optimal model was obtained by comparison.

[0014] (8) Model verification: input the element content of the designed component into the optimal machine learning model, predict the fracture toughness corresponding to this component, and conduct experimental verification.

[0015] The normalization process uses a standardization method to process the data, which is expressed as: ; in is the element content of the amorphous alloy in the data set, is the standard value, is the average value, is the standard deviation.

[0016] The model training adopts Anaconda3 software and uses machine learning algorithm to perform model training.

[0017] The creation of various machine learning models includes: ridge regression model, support vector machine model, k nearest neighbor model, decision tree model, extreme forest model, random forest, integrated regression model, gradient boosting model and elastic network model.

[0018] When evaluating the model, the indicator of the machine learning model is the goodness of fit R 2 and mean absolute percentage error MAE :

[0019] in, n is the sample size; y i is the actual value of the sample; is the predicted value of the model; is the mean value of the sample; R 2 The value is [0,1]. The closer R2 is to 1, the better the model fit is, and the closer it is to 0, the worse the model fit is. The closer MAE is to 0, the smaller the difference between the true value and the predicted value or the average value is.

[0020] The model verification is to conduct experimental verification on the predicted amorphous alloy composition and its neighboring amorphous alloy compositions.

[0021] The method is applicable to all amorphous alloys, including palladium-based, platinum-based, zirconium-based, copper-based, magnesium-based, nickel-based, iron-based, titanium-based, cerium-based, calcium-based, and lanthanum-based.

[0022] A high-toughness amorphous alloy composition design system based on machine learning, the system comprising: (1) Data construction module, in which the input data is the element content of the amorphous alloy composition, and the element distribution is obtained; the output data is the corresponding fracture toughness, and the fracture toughness distribution is obtained; (2) Data partitioning module: divides the data into training set and test set, of which 80% is set as training set for training the model and 20% is set as test set for testing the model; (3) Data processing module, which performs standard normalization on the data, uses the transformation fitting process fit_transform on the training set, and then uses the transformation process transform on the test set, and obtains the correlation between elements through the Pearson correlation coefficient; (4) Model creation module, which creates various machine learning models in the database; (5) Model training module: fit the data in the training set to the machine learning model, and use cross-validation and grid search hyperparameters to obtain the optimal machine learning model; (6) Model prediction module, which uses the data in the test set to predict the training model; (7) Model evaluation module, using R 2 and MAE to quantitatively evaluate the performance of various models and obtain R 2 Machine learning models with the highest and lowest MAE; (8) Model verification module: input the set content of amorphous alloy components into the model, obtain the fracture toughness, and conduct experimental verification.

[0023] The beneficial effect of the present invention is that the present invention uses a machine learning model to unlock the implicit complex relationship between "composition-performance" by mining a large amount of existing data on amorphous alloy composition and fracture toughness, and realizes the rapid and accurate design of amorphous alloy composition according to performance requirements. The present invention improves the efficiency of high-toughness amorphous alloy composition design, and through data mining and analysis methods, the trial and error cost is low, the cycle is short, and the adaptability is strong. The technical solution provided by the present invention is applicable to all amorphous alloy composition designs. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A flowchart of a method for designing the composition of a high-toughness amorphous alloy based on machine learning; Figure 2 is the distribution map of amorphous elements in the data set; Figure 3 This is the distribution diagram of amorphous fracture toughness in the data set; Figure 4 is the distribution diagram of Pearson correlation coefficient between amorphous elements in the data set; Figure 5 R for 9 machine learning models 2 Comparison chart; Figure 6 This is a comparison chart of MAE for various machine learning models; Figure 7Predicted and experimental plots for the training and test sets of the extreme forest model; Figure 8 This is a tree flow chart of the extreme forest model. DETAILED DESCRIPTION

[0025] like Figure 1 As shown, this embodiment provides a method for designing a composition of a high-toughness amorphous alloy based on machine learning, comprising the following steps: (1) Data construction: By collecting and organizing publicly available literature, 85 sets of amorphous composition data and fracture toughness data were obtained to establish a basic data set; amorphous alloys include palladium-based, platinum-based, zirconium-based, copper-based, magnesium-based, nickel-based, iron-based, titanium-based, cerium-based, calcium-based, lanthanum-based amorphous alloys, and the element distribution is as follows: Figure 2 As shown; the fracture toughness data are all notch fracture toughness obtained by single-edge notch three-point bending K Q , the fracture toughness distribution is as follows Figure 3 shown.

[0026] (2) Data partitioning: The data is divided into a training set and a test set, with 80% set as the training set for training the model and 20% set as the test set for testing the model.

[0027] (3) Data processing: The data is normalized and the correlation between features is obtained through the Pearson correlation coefficient. The distribution of the Pearson correlation coefficient between amorphous elements in the data set is as follows: Figure 4 shown.

[0028] (4) Model creation: Establish various machine learning models in the database.

[0029] (5) Model training: Fit the data in the training set to the machine learning model and obtain the optimal hyperparameters through cross-validation and grid search.

[0030] (6) Model prediction: Use the data in the test set to predict the training model.

[0031] (7) Model evaluation: using R 2 and MAE to quantitatively evaluate various models, such as Figure 5 and 6 As shown, R 2 The extreme forest model with the highest and lowest MAE. The experiments and predictions of the training set and test set in the extreme forest model are as follows Figure 7 As shown, the tree flow chart of the extreme forest model is as follows Figure 8 shown.

[0032] (8) Model verification: The set amorphous component element content is input into the model to obtain the fracture toughness, and then experimental verification is performed.

[0033] Example 1 This embodiment designs a ternary Cu x Zr y Ti z For amorphous components, the boundary conditions are set as 50£x<65, 30£y<45, z=100-xy. The amorphous components and fracture toughness values ​​are predicted by the extreme forest model. The fracture toughness values ​​of the two components have been compared in the open literature (Scripta Mater. 2009 (61)137). As shown in Table 1, the fracture toughness values ​​predicted by machine learning are K Q Verified by test K Q The value is within the error range, indicating that machine learning is feasible to predict the fracture toughness of amorphous materials. x Zr y Ti z The highest fracture toughness value in the series is 62MPa√m, and the amorphous Cu 50 Zr 30 Ti 20 .

[0034]

[0035] Example 2 Design of quaternary Pd x Ni y Cu z P 20 For amorphous components, the boundary conditions were set to 31£x<63, 5£y<17, z=80-xy. The amorphous components and fracture toughness values ​​were predicted by the extreme forest model. The fracture toughness values ​​of the six components that have been experimentally tested in the open literature (Materialia2020 (12)100828 and Acta Mater. 2018 (145)477) were compared, as shown in Table 2. It was found that the fracture toughness values ​​predicted by machine learning were K Q Verified by test K Q The value is within the error range, indicating that machine learning is feasible to predict the fracture toughness of amorphous materials. x Ni y Cu z P 20 The amorphous Pd with the highest fracture toughness value of 68MPa√m 31 Ni 5 Cu 44 P 20 .

[0036]

[0037] Example 3 Design of hexavalent Fe x Cr 15 Mo 14 C y B z Er m For amorphous components, the boundary conditions are set to 40£x<50, 10£y<20, 1£z<10, and m=71-xyz. The amorphous components and fracture toughness values ​​are predicted by the extreme forest model. The fracture toughness values ​​of the three components have been compared with those of the experiments in the open literature (Appl. Phys. Lett. 2008 (92)091918). As shown in Table 3, the fracture toughness values ​​predicted by machine learning are K Q Verified by test K Q The value is within the error range, indicating that machine learning is feasible to predict the fracture toughness of amorphous glass. 58 Ti 6 Cu 19 Ni 7 Al 10 The highest fracture toughness value in the series is 42MPa√m, and the amorphous Fe 49 Cr 15 Mo 14 C 19 B 3 Er 0 .

[0038] .

Claims

1. A method for designing high-toughness amorphous alloy composition based on machine learning, characterized in that: The method comprises the following steps: (1) Data construction: A data set was established based on publicly available literature data, including composition data of 85 amorphous alloys and corresponding fracture toughness data; (2) Data partitioning: Divide the data into a training set and a test set. The training set is used to train the model, and the test set is used to test the model. (3) Data processing: normalizing the data; (4) Model creation: creating various machine learning models; (5) Model training: Fit the training set in the machine learning model, obtain the optimal hyperparameters of the machine learning model through cross-validation and grid search, and thus obtain the machine learning model under the optimal hyperparameters; (6) Model prediction: predicting the test set in the machine learning model; (7) Model evaluation, using the regression model evaluation index R 2 Quantitatively evaluate the performance of various machine learning models using RMSE and compare to obtain the optimal model; (8) Model verification: input the element content of the designed component into the optimal machine learning model, predict the fracture toughness corresponding to this component, and conduct experimental verification.

2. The method for designing high-toughness amorphous alloy composition based on machine learning according to claim 1, characterized in that: The normalization process uses a standardization method to process the data, which is expressed as: ; in is the element content of the amorphous alloy in the data set, Standard value, is the average value, is the standard deviation.

3. The method for designing high-toughness amorphous alloy composition based on machine learning according to claim 1, characterized in that: The model training adopts Anaconda3 software and uses machine learning algorithm to perform model training.

4. The method for designing high-toughness amorphous alloy composition based on machine learning according to claim 1, characterized in that: The creation of various machine learning models includes: ridge regression model, support vector machine model, k nearest neighbor model, decision tree model, extreme forest model, random forest, integrated regression model, gradient boosting model and elastic network model.

5. The method for designing high-toughness amorphous alloy composition based on machine learning according to claim 1, characterized in that: When evaluating the model, the indicator of the machine learning model is the goodness of fit R 2 and mean absolute percentage error MAE : ; in, n is the sample size; y i is the actual value of the sample; is the predicted value of the model; is the mean value of the sample; R 2 The value is [0,1]. The closer R2 is to 1, the better the model fit is, and the closer it is to 0, the worse the model fit is. The closer MAE is to 0, the smaller the difference between the true value and the predicted value or the average value is.

6. The method for designing high-toughness amorphous alloy composition based on machine learning according to claim 1, characterized in that: The model verification is to conduct experimental verification on the predicted amorphous alloy composition and its neighboring amorphous alloy compositions.

7. The method for designing high-toughness amorphous alloy composition based on machine learning according to claim 1, characterized in that: The method is applicable to all amorphous alloys, including palladium-based, platinum-based, zirconium-based, copper-based, magnesium-based, nickel-based, iron-based, titanium-based, cerium-based, calcium-based, and lanthanum-based.

8. A high-toughness amorphous alloy composition design system implementing a high-toughness amorphous alloy composition design method based on machine learning as described in any one of claims 1 to 9, characterized in that: The system comprises: (1) Data construction module, in which the input data is the element content of the amorphous alloy composition, and the element distribution is obtained; the output data is the corresponding fracture toughness, and the fracture toughness distribution is obtained; (2) Data partitioning module: divides the data into training set and test set, of which 80% is set as training set for training the model and 20% is set as test set for testing the model; (3) Data processing module, which performs standard normalization on the data, uses the transformation fitting process fit_transform on the training set, and then uses the transformation process transform on the test set, and obtains the correlation between elements through the Pearson correlation coefficient; (4) Model creation module, which creates various machine learning models in the database; (5) Model training module: fit the data in the training set to the machine learning model, and use cross-validation and grid search hyperparameters to obtain the optimal machine learning model; (6) Model prediction module, which uses the data in the test set to predict the training model; (7) Model evaluation module, using R 2 and MAE to quantitatively evaluate the performance of various models and obtain R 2 Machine learning models with the highest and lowest MAE; (8) Model verification module: input the set content of amorphous alloy components into the model, obtain the fracture toughness, and conduct experimental verification.

Images