A low-activation high-entropy alloy and its design method based on machine learning

Abstract

The present invention provides a low-activation high-entropy alloy with a molecular formula of Fe a Cr b V c W d Mn e ; wherein, a, b, c, d, and e respectively represent the atomic percentages of the corresponding elements, and satisfy the following conditions: 30 ≤ a ≤ 35, 30 ≤ b ≤ 35, 10 ≤ c ≤ 15, 10 ≤ d ≤ 15, 5 ≤ e ≤ 10, and a + b + c + d + e = 100. The present invention provides a machine learning-based design method for the above alloy. The high-entropy alloy of the present invention has low-activation characteristics, a BCC phase structure, and high hardness; at the same time, the machine learning-based design method constructed by the present invention can synchronously, quickly, and accurately design new multi-principal element alloys according to structural and performance requirements, thereby avoiding the blindness of the traditional experimental trial-and-error method and significantly improving the design and research and development efficiency of new materials.

Description

Technical Field

[0001] The present invention relates to the field of metal materials, and in particular to a low-activated high-entropy alloy and a design method thereof based on machine learning. Background Art

[0002] Low-activated ferrite / martensitic steel (RAFM) has good thermal physical, thermal mechanical, and neutron irradiation resistance properties, and is considered to be an ideal candidate structural material for fusion reactors. The high-temperature strength and thermal creep properties of RAFM steel will drop significantly above 550°C, making the upper limit of the maximum operating temperature of RAFM steel approximately 550°C. In order to further optimize the high-temperature mechanical properties of RAFM steel, researchers have conducted a lot of research on optimizing alloy composition and changing heat treatment parameters. However, this optimization approach is very limited to improving the performance of RAFM steel. For example, at a test temperature of 550°C, its tensile strength is increased by less than 150MPa. It is very difficult to significantly improve the mechanical properties of RAFM to meet the requirements of higher operating temperatures of future fusion reactors.

[0003] In recent years, high entropy alloys have attracted much attention due to their excellent properties such as high strength, toughness, corrosion resistance, oxidation resistance and wear resistance. Among them, the high strength and high temperature softening resistance of BCC structure high entropy alloys are very consistent with the application requirements of advanced fusion reactors. Kumar et al. from Oak Ridge National Laboratory in the United States conducted ion irradiation experiments on FeNiMnCr high entropy alloys and found that they have strong radiation resistance, which also provides a basis for the application of high entropy alloys in fusion reactors. However, the design and research of low-activated BCC high entropy alloys for fusion reactors is very lacking.

[0004] Traditional methods of searching for new high-entropy alloys are often through experiments, theories or calculations. However, the search space for the composition of high-entropy alloys is very huge. It is time-consuming, labor-intensive and difficult to find high-entropy alloys with target structures and properties through these methods. Data-driven machine learning methods do not need to pay too much attention to specific physical details, and can quickly build complex nonlinear relationships between input data and output targets. They are gradually being applied to search problems for different materials. Therefore, data-driven machine learning methods can be used to design low-activated BCC high-entropy alloys for fusion reactors. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a low-activation high-entropy alloy and a design method thereof based on machine learning.

[0006] The present invention adopts the following technical solutions to solve the above technical problems:

[0007] A low-activation high-entropy alloy, wherein the molecular formula of the low-activation high-entropy alloy is Fe aCr b V c W d Mn e ; wherein a, b, c, d and e represent the atomic percentage of the corresponding elements respectively and satisfy the following conditions: 30≤a≤35, 30≤b≤35, 10≤c≤15, 10≤d≤15, 5≤e≤10, a+b+c+d+e=100.

[0008] A design method of the above-mentioned low-activation high-entropy alloy based on machine learning comprises the following steps:

[0009] Step 1: Data Collection

[0010] Collect data and information on existing high entropy alloys;

[0011] Step 2: Feature Construction

[0012] The atomic percentage and element parameters of the high entropy alloy are processed into data to form element characteristics; then, the empirical characteristics corresponding to the target performance are collected and together with the element characteristics, they form characteristic variables;

[0013] Step 3: Feature Screening

[0014] Screening of key characteristic variables;

[0015] Step 4: Machine Learning Algorithm Selection

[0016] Choose different machine learning algorithms according to different models;

[0017] Step 5: Model construction;

[0018] Construct a classification model to identify whether a high entropy alloy is solid solution strengthened, a classification model to identify whether the solid solution strengthened high entropy alloy is body-centered cubic BCC, face-centered cubic FCC, or a mixed structure of BCC and FCC, and a regression model to predict the hardness of high entropy alloys;

[0019] Step 6: Search space settings

[0020] The search space is set to select 4 to 6 low-activation elements, and the atomic percentage of each element ranges from 5 to 35 at.%, with a step size of 1 at.%;

[0021] Step 7: Screening candidate ingredients

[0022] The set search space is screened according to the design requirements of high entropy alloys such as solid solution strengthening, BCC structure, and hardness greater than 700HV, and finally the elements and their corresponding atomic percentages of the low-activation high entropy alloy required for the target are obtained;

[0023] Step 8: Experimental verification.

[0024] As one of the preferred embodiments of the present invention, in step 1, the data information collected is specifically the atomic percentage, cast phase structure and corresponding hardness information of the existing high entropy alloy.

[0025] As one of the preferred embodiments of the present invention, in step 2, the atomic percentage and element parameters of the high entropy alloy are digitized by formulas (1) to (3) to form element characteristics;

[0026]

[0027]

[0028]

[0029] In the formula, δX and DX are the average value, mismatch value and local mismatch value respectively; C i and C j represent the atomic percentage of the i-th element and the j-th element respectively; X i and X j are the element parameters of the i-th and j-th elements respectively.

[0030] As one of the preferred embodiments of the present invention, in step 2, the empirical characteristics are characteristics that are recorded in existing literature and have been confirmed to be related to the target performance.

[0031] As one of the preferred methods of the present invention, in step 3, a three-step method of correlation screening, recursive elimination screening, and exhaustive search screening is used to screen key characteristic variables that affect the structure and performance of high entropy alloys.

[0032] As one of the preferred embodiments of the present invention, in step 4, for the construction of the classification model, the machine learning algorithm includes decision tree classification DTC, gradient boosting classification GBC, radial basis function support vector machine classification SVC, k nearest neighbor classification KNC, random forest classification RFC, and artificial neural network classification ANNC; for the construction of the regression model, the machine learning algorithm includes decision tree regression DTR, gradient boosting regression GBR, radial basis function support vector machine regression SVR, k nearest neighbor regression KNR, random forest regression RFR, and artificial neural network regression ANNR.

[0033] As one of the preferred embodiments of the present invention, in step 6, the low activation elements refer to Fe, Cr, V, Mn, Ti, W, Ta, and Zr elements.

[0034] As one of the preferred methods of the present invention, in step 8, high entropy alloy is smelted according to the screened components, and the cast sample is subjected to structural characterization and hardness test; the results are compared with the predicted results. If the experimental results are consistent with the predicted results, the design is completed; otherwise, the experimental results are added to the training set of the original data, and the design is re-performed until a high entropy alloy that meets the requirements is designed.

[0035] The advantages of the present invention compared to the prior art are:

[0036] (1) The low-activation high-entropy alloy designed by the present invention has the main constituent elements of Fe, Cr, V, W, and Mn, has low activation characteristics, BCC structure, and has the advantages of high strength and high temperature softening resistance;

[0037] (2) The machine learning-based design method for low-activation high-entropy alloys constructed in the present invention can quickly and accurately design new multi-principal component alloys according to structural and performance requirements, thereby avoiding the blindness of traditional experimental trial and error methods and significantly improving the design and development efficiency of new materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a design flow chart of the design method of low-activation high-entropy alloy based on machine learning in Example 4;

[0039] Figure 2 is the XRD pattern of the Fe-Cr-VW-Mn low activation high entropy alloy in Examples 5 and 6. DETAILED DESCRIPTION

[0040] The following is a detailed description of an embodiment of the present invention. This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and a specific operation process are given, but the protection scope of the present invention is not limited to the following embodiment.

[0041] Example 1

[0042] A low-activation high-entropy alloy of this embodiment is designed using the method of the present invention and is smelted strictly according to the composition designed by the present invention.

[0043] The low activation high entropy alloy includes the following chemical components in atomic percentage: 30% Fe, 35% Cr, 15% V, 15% W, and 5% Mn.

[0044] Example 2

[0045] A low-activation high-entropy alloy of this embodiment is designed using the method of the present invention and is smelted strictly according to the composition designed by the present invention.

[0046] The low activation high entropy alloy includes the following chemical components in atomic percentage: 35% Fe, 35% Cr, 10% V, 15% W, and 5% Mn.

[0047] Example 3

[0048] A low-activation high-entropy alloy of this embodiment is designed using the method of the present invention and is smelted strictly according to the composition designed by the present invention.

[0049] The low activation high entropy alloy includes the following chemical components in atomic percentage: 35% Fe, 35% Cr, 15% V, 10% W, and 5% Mn.

[0050] Example 4

[0051] like Figure 1 As shown, a method for designing low-activated high-entropy alloys in the above-mentioned embodiments 1 to 3 in this embodiment based on machine learning includes the following steps:

[0052] Step 1: Data Collection

[0053] In this embodiment, 611 groups of high entropy alloy compositions and their corresponding cast phase structures are collected; wherein, the elemental composition includes 16 elements including Al, Co, Cr, Cu, Fe, Hf, Mn, Mo, Nb, Ni, Sn, Ta, Ti, V, W and Zr; according to whether the phase structure contains intermetallic compounds and amorphous phases, the 611 groups of samples are divided into solid solution high entropy alloys (329 groups of samples) and non-solid solution high entropy alloys (282 groups of samples); the 329 groups of solid solution high entropy alloy samples are divided into three categories according to the phase structure of BCC, FCC and a mixture of BCC and FCC, wherein there are 159 groups of BCC samples, 99 groups of FCC samples and 71 groups of a mixture of BCC and FCC; a total of 460 groups of high entropy alloy compositions and their corresponding Vickers hardness data are collected.

[0054] Step 2: Feature Construction

[0055] This embodiment collects 64 element parameters corresponding to the 16 elements mentioned in step 1. First, the atomic percentage and element parameters of the high entropy alloy are digitized according to formulas (1) to (3), so there are 192 element characteristics. Next, the empirical characteristics that have been confirmed to be related to the target performance in the literature are collected, and together with the element characteristics, they constitute characteristic variables. This embodiment collects a total of 11 empirical characteristics, including mixing enthalpy, mixing entropy, melting entropy, mixing Gibbs free energy, Ω parameter, Λ parameter, Pauling electronegativity difference, γ parameter, shear modulus mismatch of the strengthening model, energy term of the strengthening model and the sixth power of the work function. Finally, 203 characteristic variables were initially generated. In order to eliminate the influence of dimensional differences on the prediction results during the modeling process, each feature in this embodiment is normalized according to formula (4).

[0056]

[0057]

[0058]

[0059]

[0060] In the formula, δX and DX are the average value, mismatch value and local mismatch value respectively; C i and C j represent the atomic percentage of the i-th element and the j-th element respectively; X i and X j are the element parameters of the i-th and j-th elements respectively; x min and x max are the minimum and maximum values ​​of the feature variable respectively.

[0061] Step 3: Feature Screening

[0062] In this embodiment, three steps of correlation screening, recursive elimination screening, and exhaustive search screening are used to screen the key characteristic variables that affect the structure and performance of high entropy alloys. The first step is correlation screening, the purpose of which is to quickly remove the highly correlated characteristic variables that cause overfitting. Calculate the correlation coefficient (PCC) between any two characteristic variables. When |PCC|>0.9, it means that there is a strong linear correlation between the two characteristic variables. In this case, the remaining features after removing the two characteristic variables are used as input, a prediction model is constructed, the test error is calculated, and the features with large test errors are eliminated. If |PCC|<0.9, these two features are retained. The second step is recursive elimination screening, the purpose of which is to remove characteristic variables that are not related to the research problem. Remove one of the n features in turn, leaving n-1 features, and build a prediction model on this basis to calculate the test error. Keep the n-1 features corresponding to the minimum test error, and continue to perform recursive elimination until only one characteristic variable is left. The third step is exhaustive search screening, that is, traversing all possible options to select a best subset. This embodiment is to select an optimal subset from the characteristic variables corresponding to the minimum test error in the entire recursive elimination process of the second step.

[0063] Step 4: Machine Learning Algorithm Selection

[0064] For the construction of classification models, the candidate machine learning algorithms include decision tree classification (DTC), gradient boosting classification (GBC), support vector machine classification (SVC) with radial basis function, k-nearest neighbor classification (KNC), random forest classification (RFC), and artificial neural network classification (ANNC).

[0065] For the construction of regression models, the candidate machine learning algorithms include decision tree regression (DTR), gradient boosting regression (GBR), support vector machine regression (SVR) with radial basis function, k-nearest neighbor regression (KNR), random forest regression (RFR) and artificial neural network regression (ANNR).

[0066] In this embodiment, the holdout method is used to verify the predictive ability of the models constructed by different machine learning algorithms. In the holdout method, 80% of the normalized data set is used for training, and the remaining 20% ​​is used to calculate the test error. The predictive ability of the classification model is evaluated by the miss rate calculated by formula (5).

[0067]

[0068] Where T and F represent the number of correctly classified samples and the number of incorrectly classified samples, respectively. When the calculated Miss Rate is 0, it indicates a perfect fit.

[0069] The predictive ability of the regression model is calculated by the root mean square error (RMSE) calculated by formula (6) and the coefficient of determination (R 2 ) to evaluate the predictive ability of the model.

[0070]

[0071]

[0072] Where n is the number of samples; y i and are the actual value and predicted value of the i-th sample (i=1,2,…,n); is the mean of the actual values. For an ideal model, RMSE is 0 and R 2 is equal to 1.

[0073] Step 5: Model construction

[0074] The constructed models include a classification model for identifying whether a high entropy alloy is solid solution strengthened (named SS-Model), a classification model for identifying whether a solid solution strengthened high entropy alloy is a body-centered cubic (BCC), face-centered cubic (FCC) or a mixed structure of BCC and FCC (named BCC-Model), and a regression model for predicting the hardness of high entropy alloys (named HV-Model). Table 1 lists the optimal machine learning algorithms and their corresponding key feature variables selected to construct the above three models after steps 3 and 4.

[0075] Table 1 Algorithms and key characteristic variables for constructing SS-Model, BCC-Model and HV-Model

[0076]

[0077] Step 6: Search space settings

[0078] The search space is set to select 4 to 6 low activation elements, and the atomic percentage range of each element is 5-35 at.%, with a step size of 1 at.%. In this embodiment, the low activation elements selected are Fe, Cr, V, Mn, W, Ti, Ta and Zr.

[0079] Step 7: Screening candidate ingredients

[0080] The set search space was screened according to the design requirements of high entropy alloys with solid solution strengthening, BCC structure, and hardness greater than 700HV. After 100 repeated designs, it was found that the low-activated high-entropy alloys that met the design requirements were all Fe a Cr b V c W d Mn e , where a, b, c, d and e represent the atomic percentage of the corresponding elements respectively, and satisfy the following conditions: 30≤a≤35, 30≤b≤35, 10≤c≤15, 10≤d≤15, 5≤e≤10, a+b+c+d+e=100.

[0081] Step 8: Experimental verification

[0082] High entropy alloys are melted according to the screened components, and the structure and hardness of the cast samples are characterized and tested, which are compared with the predicted results. If the experimental results are consistent with the predicted results, the design is completed. Otherwise, the experimental results are added to the training set of the original data, and the design is repeated until a high entropy alloy that meets the requirements is designed.

[0083] Example 5

[0084] This embodiment is a specific experimental verification of the low-activation high-entropy alloy in the above-mentioned embodiments 1 to 3.

[0085] From the results designed in this embodiment, a low-activated high-entropy alloy is selected for experimental verification. On the one hand, this can verify the reliability of the design method based on machine learning, and on the other hand, it can achieve the purpose of designing a low-activated high-entropy alloy.

[0086] According to the designed low activation high entropy alloy Fe 30 Cr 35 V 15 W 15 Mn 5(Example 1) A 200 g ingot was prepared by vacuum arc melting. To ensure the uniformity of the alloy composition, the alloy was turned over and melted repeatedly for at least 5 times during the preparation process.

[0087] The XRD characteristics of the cast alloy are as follows: Figure 2 As shown in the figure, the results show that the alloy consists of two types of BCC phase structures. This indicates that the high entropy alloy is solid solution strengthened and has a BCC structure, which is consistent with the predicted results and fully demonstrates the effectiveness of the SS-Model and BCC-Model models.

[0088] The experimentally measured Vickers hardness of the high entropy alloy was 699.3±27.6HV, which is consistent with the predicted result of 703.4±11.4HV, indicating that the HV-Model is very effective in predicting Vickers hardness.

[0089] The above experimental results show that the design method based on machine learning of the present invention can provide a fast and effective method for the design of low-activation high-entropy alloys.

[0090] Example 6

[0091] This embodiment is a specific experimental verification of the low-activation high-entropy alloy in the above-mentioned embodiments 1 to 3.

[0092] From the results designed in this embodiment, a low-activated high-entropy alloy is selected for experimental verification. On the one hand, this can verify the reliability of the design method based on machine learning, and on the other hand, it can achieve the purpose of designing a low-activated high-entropy alloy.

[0093] According to the designed low activation high entropy alloy Fe 35 Cr 35 V 10 W 15 Mn 5 (Example 2) A vacuum arc melting method was used to prepare an ingot of about 200 g. In order to ensure the uniformity of the alloy composition, the alloy was turned over and repeatedly melted at least 5 times during the preparation process.

[0094] The XRD characteristics of the cast alloy are as follows: Figure 2 As shown in the figure, the results show that the alloy consists of two types of BCC phase structures. This indicates that the high entropy alloy is solid solution strengthened and has a BCC structure, which is consistent with the predicted results and fully demonstrates the effectiveness of the SS-Model and BCC-Model models.

[0095] The experimentally measured Vickers hardness of the high entropy alloy was 696.2±34.4HV, which was consistent with the predicted result of 701.4±30.3HV, indicating that the HV-Model is very effective in predicting Vickers hardness.

[0096] The above experimental results show that the design method based on machine learning of the present invention can provide a fast and effective method for the design of low-activation high-entropy alloys.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A machine learning-based design method for low-activation high-entropy alloys, It is characterized in that The molecular formula of the low activation high entropy alloy is Fe a Cr b V c W d Mn e ; wherein a, b, c, d and e represent the atomic percentage of the corresponding elements respectively and satisfy the following conditions: 30≤a≤35, 30≤b≤35, 10≤c≤15, 10≤d≤15, 5≤e≤10, a+b+c+d+e=100; the design method comprises the following steps: Step 1: Data Collection Collect the atomic percentage, cast phase structure and corresponding hardness information of existing high entropy alloys; Step 2: Feature Construction The atomic percentage and element parameters of the high entropy alloy are processed digitally to form element characteristics; then, empirical characteristics corresponding to the target performance are collected and combined with the element characteristics to form characteristic variables; wherein the atomic percentage and element parameters of the high entropy alloy are processed digitally through formulas (1) to (3) to form element characteristics; In the formula, δX and DX are the average value, mismatch value and local mismatch value respectively; C i and C j Respectively represent the atomic percentage of the i-th element and the j-th element; X i and X j are the element parameters of the i-th and j-th elements respectively; Step 3: Feature Screening A three-step method of correlation screening, recursive elimination screening, and exhaustive search screening is used to screen the key characteristic variables that affect the structure and performance of high entropy alloys; Step 4: Machine Learning Algorithm Selection Choose different machine learning algorithms according to different models; Step 5: Model construction; Construct a classification model to identify whether a high entropy alloy is solid solution strengthened, a classification model to identify whether the solid solution strengthened high entropy alloy is body-centered cubic BCC, face-centered cubic FCC, or a mixed structure of BCC and FCC, and a regression model to predict the hardness of high entropy alloys; Step 6: Search space settings The search space is set to select 4 to 6 low-activation elements, and the atomic percentage of each element ranges from 5 to 35 at.%, with a step size of 1 at.%; Step 7: Screening candidate ingredients The set search space is screened according to the design requirements of high entropy alloys such as solid solution strengthening, BCC structure, and hardness greater than 700HV, and finally the elements and their corresponding atomic percentages of the low-activation high entropy alloy required for the target are obtained; Step 8: Experimental verification.

2. The machine learning-based design method for low-activation high-entropy alloys according to claim 1, It is characterized in that In step 2, the empirical features are features that are recorded in existing literature and have been proven to be related to the target performance.

3. The design method of low-activation high-entropy alloy based on machine learning according to claim 1, It is characterized in that In step 4, for the construction of the classification model, the machine learning algorithms include decision tree classification DTC, gradient boosting classification GBC, radial basis function support vector machine classification SVC, k nearest neighbor classification KNC, random forest classification RFC, and artificial neural network classification ANNC; for the construction of the regression model, the machine learning algorithms include decision tree regression DTR, gradient boosting regression GBR, radial basis function support vector machine regression SVR, k nearest neighbor regression KNR, random forest regression RFR, and artificial neural network regression ANNR.

4. The machine learning-based design method for low-activation high-entropy alloys according to claim 1, It is characterized in that In step 6, the low activation elements refer to Fe, Cr, V, Mn, Ti, W, Ta, and Zr elements.

5. The machine learning-based design method for low-activation high-entropy alloys according to claim 1, It is characterized in that In step 8, high entropy alloy is smelted according to the screened components, and the cast sample is subjected to structural characterization and hardness test; the results are compared with the predicted results. If the experimental results are consistent with the predicted results, the design is completed; otherwise, the experimental results are added to the training set of the original data, and the design is repeated until a high entropy alloy that meets the requirements is designed.

Images