Machine learning method for rapid prediction of hardness of high-entropy alloys and its preparation process

Abstract

The present invention discloses a machine learning method and preparation process for rapidly predicting the hardness of high-entropy alloys, which are used to solve the problem that the traditional "trial and error method" for designing high-entropy alloy compositions is time-consuming, labor-intensive, and produces inaccurate results. The method comprises collecting a high-entropy alloy hardness database as a training set, calculating feature descriptors based on the alloy composition; selecting a suitable model based on the RMSE under different test set partitioning ratios; performing dimensionality reduction processing on the obtained features to obtain the most important feature subset; using the most important feature subset to train and optimize the selected model; using the trained model to predict the hardness of high-entropy alloys of unknown composition; and experimentally verifying the predicted alloy composition. The feature dimensionality reduction method in the present invention comprises correlation analysis, recursive elimination method, and exhaustive method. After dimensionality reduction processing, three most important features are obtained, and the hardness of the alloy can be conveniently and accurately predicted based on these three features.

Description

Technical Field

[0001] The present invention relates to the field of metal material property prediction, and in particular to using a machine learning method to predict the hardness properties of a high entropy alloy, and verifying the prediction results through experiments. Background Art

[0002] High-entropy alloys (HEAs) are a new alloy system that differs from traditional alloy material design concepts. HEAs are composed of five or more primary elements, with each element comprising approximately 5 to 35 at.%. The high configurational entropy in HEAs causes the alloys to tend to form simple random solid solutions rather than intermetallic compounds such as FCC, BCC, or HCP phases. Furthermore, HEAs exhibit four unique effects: high entropy, lattice distortion, slow diffusion, and the cocktail effect. These effects provide HEAs with superior and diverse properties, such as high strength and hardness, high-temperature oxidation resistance, wear resistance, corrosion resistance, and hydrogen embrittlement resistance, offering broad application prospects.

[0003] Traditional high-entropy alloy design often relies on theoretical calculations and experimental verification, with improvements to composition or process often employing a trial-and-error approach. These methods are time-consuming and labor-intensive, and the results are affected by numerous factors. For materials like high-entropy alloys, which have a vast compositional space, finding high-performance high-entropy alloys using traditional methods is particularly difficult. With the widespread adoption of AI technology, it has found widespread applications in areas such as autonomous driving and image recognition. In particular, technological advancements have significantly increased computer computing power, making it possible to apply machine learning methods to identify high-performance materials.

[0004] The steps for predicting material properties using machine learning methods are typically divided into the following: dataset creation, model selection, feature dimensionality reduction, model training and optimization, result prediction, and experimental verification. Feature selection largely determines the model's predictive performance. The number of features often ranges from dozens to hundreds, and considering all of them would significantly increase the computational load. Fortunately, many of these features are redundant or irrelevant to the expected results. Therefore, selecting the appropriate feature combination is crucial in machine learning hardness prediction. Summary of the Invention

[0005] In view of the above problems, the present invention proposes a machine learning method and a preparation process for quickly predicting the hardness of high-entropy alloys to solve the problem that traditional methods are time-consuming and labor-intensive in designing high-performance high-entropy alloys.

[0006] The machine learning method for rapidly predicting the hardness of high entropy alloys comprises the following steps:

[0007] Step 1: Collect and organize the high entropy alloy hardness database, calculate the characteristic descriptors according to the alloy composition, and construct the characteristic data set;

[0008] Step 2: Select the appropriate model based on the RMSE under different test set partition ratios;

[0009] Step 3: Perform dimensionality reduction on the obtained features to obtain the most important feature subset;

[0010] Step 4: Use the most important feature subset obtained to train the selected model and tune the model's hyperparameters;

[0011] Step 5: The trained model can be used to predict the hardness of high-entropy alloys with unknown composition;

[0012] Step 6: Hardness test verification of the predicted composition;

[0013] Furthermore, the high entropy alloy described in step one is an Al-Co-Cr-Cu-Fe-Ni-Mn seven-element alloy system.

[0014] Furthermore, the feature descriptors in step 1 include 20 features, namely, average atomic radius, atomic radius difference, work function, average melting point, mixing enthalpy, mixing entropy, Ω parameter, valence electron concentration, average electronegativity, electronegativity difference, Λ parameter, number of mobile electrons, electron affinity, density, first ionization energy, average Young's modulus, lattice distortion energy, cohesive energy, local size mismatch, and average shear modulus.

[0015] Furthermore, the machine learning models described in step 2 include linear regression, random forest regression, support vector machine-radial basis kernel function, support vector machine-sigmoid kernel function, support vector machine-polynomial kernel, K-nearest neighbor and ridge regression.

[0016] Furthermore, the test set partitions described in step 2 account for 15%, 20%, 25%, 30%, 35%, and 40% of the total data set, respectively.

[0017] Furthermore, the appropriate machine learning model in step 2 is a random forest regression algorithm model.

[0018] Furthermore, in step three, the dimensionality reduction process of the feature data set includes first using the Pearson correlation coefficient to remove features with high correlation, retaining only one feature in the feature subset with high correlation, and then using the recursive elimination method to obtain the feature subset with the best model fitting effect. Finally, the exhaustive method is used to further reduce the dimensionality of the remaining features by using permutations and combinations until the most important feature subset is obtained.

[0019] Furthermore, the Pearson correlation coefficient measures the highly correlated nature of two features based on a Pearson correlation coefficient value greater than or equal to 0.95.

[0020] Furthermore, the basis for removing one of the two highly correlated features is the importance to the model, and the feature with high importance is retained.

[0021] Furthermore, the optimal characteristic combination is three characteristics: work function, valence electron concentration, and Young's modulus.

[0022] Furthermore, the random forest regression algorithm in step 4 uses a grid search method to perform hyperparameter tuning.

[0023] Furthermore, the hardness prediction of the unknown alloy in step five is achieved by calculating the three important features in step three.

[0024] A method for preparing the Al-Co-Cr-Cu-Fe-Ni-Mn high entropy alloy as described above comprises the following steps:

[0025] Step 1, ultrasonic cleaning and material weighing: high-purity metal particles Al, Co, Cr, Cu, Fe, Ni, and Mn of the Al-Co-Cr-Cu-Fe-Ni-Mn high-entropy alloy are placed in a container respectively, and cleaned in an ultrasonic cleaning device with acetone and anhydrous ethanol for 10-20 minutes in sequence, and finally the materials are blown dry with cold air for standby use; the Al-Co-Cr-Cu-Fe-Ni-Mn high-entropy alloy is converted into corresponding weight according to atomic percentage and weighed; the weighed materials are placed in a copper crucible of a WK-II vacuum arc furnace in order of increasing melting point, and a layer of the surface of the pure titanium ingot is ground off with 400-mesh sandpaper. After rinsing with alcohol and blowing dry, the ingot is placed in the middle crucible position of the vacuum furnace, and then the furnace door is closed and locked;

[0026] Step 2: Vacuuming: After step 1, use mechanical and molecular pumps to vacuum to 1-5×10 -4 Pa, and then slowly fill in argon gas to make the vacuum degree reach 1~5×10 4 Pa, and then the vacuum degree of the arc furnace is pumped to 1~5×10 -4 Pa; repeat this process three times to ensure that there is no oxygen in the vacuum chamber; finally, slowly introduce argon to -0.02~-0.05Mpa.

[0027] Step 3: Melting: First, melt the pure titanium ingot 3 to 5 times, with the current controlled at 100-300A and the time being 30-150s. After each melting, turn the pure titanium ingot over and carry out the next melting; then melt the high entropy alloy. At this time, the current should be increased slowly to avoid the temperature rising too fast to generate large thermal stress and cause the sample to crack. The arc current is increased to a maximum of 300A and the duration is 50-200s; during the melting process, slightly rotate the arc rod to heat the material evenly; when turning off the current, also slowly reduce the current to avoid cracking caused by rapid cooling; use magnetic stirring with an external magnetic field during melting, and set the current to 1-3mA to fully ensure the uniformity of the composition;

[0028] Step 4: Repeat the smelting 4 to 7 times according to the requirements of step 2 to ensure uniform composition, and cool for 10 to 30 minutes to obtain a high-hardness Al-Co-Cr-Cu-Fe-Ni-Mn high-entropy alloy.

[0029] Beneficial technical effects of the present invention:

[0030] The present invention first determines the number of features in the feature combination, then uses the Pearson correlation coefficient to screen features with high correlation, and then uses a genetic algorithm to screen to obtain the optimal feature combination. Compared with the existing exhaustive method that requires arranging all feature combinations to find the optimal feature, when applying a machine learning algorithm to predict the hardness of high-entropy alloys, the feature screening method in the present invention is not only more accurate in predicting the performance of high-entropy alloys, but also saves computing resources and time.

[0031] The Al predicted by the present invention 51 Co 22 Cr 19 The hardness of Ni4Mn4 high entropy alloy is 805.3±1.9HV.

[0032] The Al obtained by the test of the present invention 51 Co 22 Cr 19 The hardness of Ni4Mn4 high entropy alloy is 785.9±9.9HV.

[0033] The Al predicted by the present invention 51 Co 22 Cr 19 The as-cast structure of Ni4Mn4 high entropy alloy is a typical dendrite, mainly containing a large amount of B2 and BCC phases and a trace amount of FCC phase.

[0034] The Al obtained by the experiment of the present invention 51 Co 22 Cr 19The dendritic morphology of a Ni4Mn4 high-entropy alloy shows that the dendrites are rich in Al, Co, and Ni, while the interdendritic regions are rich in Al, Cr, and Mn. This is because Al easily combines with other elements. The Cr-rich regions are primarily BCC phases, while the Ni- and Co-rich regions are primarily B2 phases. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 RMSE of the seven different models of the present invention under different test set partitions;

[0036] Figure 2 The heat map of the Pearson correlation coefficient distribution among the 20 features of the present invention;

[0037] Figure 3 This is a flow chart of feature selection based on the Pearson correlation coefficient of the present invention;

[0038] Figure 4 Result graph of the recursive elimination method of the present invention;

[0039] Figure 5 This is a result diagram of feature elimination using the exhaustive method of the present invention;

[0040] Figure 6 The fitting performance of the model established by the present invention on the training set and the test set respectively;

[0041] Figure 7 The model established by the present invention predicts Al 51 Co 22 Cr 19 XRD pattern of Ni4Mn4 high entropy alloy;

[0042] Figure 8 Al 51 Co 22 Cr 19 The microstructure of Ni4Mn4 high entropy alloy and the surface distribution of Al, Co, Cr, Mn and Ni elements. DETAILED DESCRIPTION

[0043] The following is a further description of the machine learning method and preparation process for rapidly predicting the hardness of high-entropy alloys according to the present invention in conjunction with the examples and drawings. In order to clearly and concisely describe the embodiments, not all features are described in the specification. The drawings only show the device structure and / or processing steps closely related to the solution according to the present invention, while omitting other details that are not closely related to the present invention. The scope of protection of the present invention is not limited to the contents of the examples.

[0044] A machine learning method for rapidly predicting the hardness of high entropy alloys and a preparation process thereof include the following steps:

[0045] Step 1: Collect and organize the high entropy alloy hardness database, calculate the characteristic descriptors based on the alloy composition, and construct a characteristic data set.

[0046] According to an embodiment of the present invention, the data set used for the seven-element high entropy alloy system Al-Co-Cr-Cu-Fe-Ni-Mn in the present invention contains 178 components, and the composition of the alloy is expressed as Al a Co b Cr c Cu d Fe e Ni f Mn g , where 0≤a≤46.2, 0≤b≤42.9, 0≤c≤36.8, 0≤d≤29, 0≤e≤50, 0≤f≤50, and 10≤g≤28 (atomic percentage). The mean hardness in the data set is 402.9 HV, and the standard deviation is 177.0 HV.

[0047] Although machine learning based on alloy composition can be used to predict alloy properties, it is only phenomenal and cannot reveal the physical and chemical characteristics of the elements, the interactions and reactions between the elements, the types and proportions of the constituent phases, and other physical and metallurgical mechanisms that affect alloy properties. This method is very limited in its application in screening new alloys. Calculating the physical and chemical characteristics of the material based on the alloy composition can address this problem to a certain extent. The feature descriptor includes 20 features: average atomic radius (r), atomic radius difference (δr), work function (W), average melting point (T), mixing enthalpy (ΔH), mixing entropy (ΔS), Ω parameter (Ω), valence electron concentration (VEC), average electronegativity (χ), electronegativity difference (χ), Λ parameter (Λ), mobile electron number (e / a), electron affinity (EAE), density (ρ), first ionization energy (FIE), Young's modulus (E), lattice distortion energy (μ), cohesive energy (Ec), local size mismatch (Dr), and shear modulus (G), to construct a feature dataset.

[0048] Before training a model, it is important to understand the generally accepted fact that variations in the magnitude and range of input vectors can affect the performance of ML algorithms. This is because features with higher magnitudes often have greater comparable weights than features with lower magnitudes, which can mislead machine learning models. To avoid this, it is necessary to ensure that all features are normalized using the following formula:

[0049]

[0050] where X norm represents the normalized value of the i-th feature, X iRepresents the value of the i-th feature, μ represents the mean value of the feature, σ represents the standard deviation, and the processed feature value will be between [-1, 1].

[0051] Step 2: Select the appropriate model based on the RMSE under different test set partition ratios;

[0052] According to the no free lunch theorem, to ensure the accuracy of the model, at least one machine learning model is required. Therefore, the models selected in the present invention include linear regression (LR), random forest regression (RFR), support vector machine-radial basis kernel function (SVR-r), support vector machine-sigmoid kernel function (SVR-s), support vector machine-polynomial kernel (SVR-p), K-nearest neighbor (KNN) and ridge regression (Ridge).

[0053] In order to explore the impact of different test set partitions on model performance, the dataset was randomly divided, and the test set partition ratios accounted for 15%, 20%, 25%, 30%, 35%, and 40% of the total dataset, respectively.

[0054] The prediction accuracy of the model is calculated by the root mean square error, which is calculated as follows:

[0055]

[0056] Where RMSE represents the root mean square error; n represents the total number of samples; y i Represents the true value of the i-th sample; Represents the predicted value of the i-th sample.

[0057] Figure 1 The RMSE of the 7 different models selected for the present invention under different test set divisions can be seen from the figure. Under different test set ratios, the RMSE of the random forest regression (RFR) model is the smallest, so the random forest regression model is selected as the prediction model in the present invention.

[0058] Step 3: Perform dimensionality reduction on the obtained features to obtain the most important feature subset;

[0059] The feature dataset includes a variety of features related to predicted performance, but not all features affect the predicted target performance. Features can be categorized as useful, useless, or redundant for predicting material physical properties. Therefore, dimensionality reduction of the original feature dataset can reduce model complexity while maintaining accuracy, improving the model's predictive efficiency and enhancing its generalization capabilities for predicting high-entropy alloys of unknown composition.

[0060] First, for all feature data sets, use the Pearson correlation coefficient to remove features with high correlation. The Pearson correlation coefficient (PCC) is used to measure the correlation between two quantities. Its value ranges from -1 to +1, with -1 representing a perfect negative correlation and +1 representing a perfect positive correlation. The closer the absolute value is to 1, the greater the correlation between the two features. In this case, one feature can be used to replace the other. The replaced feature is called a redundant feature. The calculation formula of the Pearson correlation coefficient is as follows:

[0061]

[0062] where r xy Represents the Pearson correlation coefficient between features x and y, x i ,y i Represents two numerical values, x m 、y m Represents the average of two features.

[0063] Figure 2 A heat map of the Pearson correlation coefficients between different features is shown. Brighter colors indicate a more negative correlation between the two features, while darker colors indicate a more positive correlation. Correlations greater than 0.95 indicate highly correlated features. Analysis of the Pearson correlation heat map shows a high correlation between characteristic atomic radius, atomic radius difference, local size mismatch, lattice distortion energy, and density; a high correlation between characteristic valence electron concentration and mobile electron number; and a high correlation between melting point, Young's modulus, and shear modulus.

[0064] Basis for removing highly correlated features Figure 3 To determine the features to be retained, the importance of a feature to the model was evaluated and the importance of each feature was ranked, as shown in Table 1. Finally, six highly relevant and low-importance features, including atomic radius (r), local size mismatch (Dr), lattice distortion energy (μ), mobile electron number (e / a), melting point (T), and shear modulus (G), were removed, leaving 14 features.

[0065] Table 1

[0066]

[0067] To further screen for the best-performing features, features were recursively eliminated. We randomly selected one feature in turn (14 features after excluding highly correlated features), and the remaining n-1 features were used as input vectors to build the RFR model. In this process, the alloy features corresponding to the least important were then eliminated, leaving n-1 alloy features, which were then used to build the model again. We recursively performed elimination until no features remained, and subsequently obtained several feature subsets with different numbers of features. We utilized a cross-validation method to test the score of the model using various feature subsets, during which the model itself remained unchanged.

[0068] from Figure 4 It can be seen from Figure 2 that when the number of features is less than 7, the cross-validation score increases with the number of selected features within 7 and reaches the maximum value at 7. We obtain a new feature subset containing 7 key features, namely δr, W, VEC, EAE, ρ, FIE and E, which closely affect the hardness of HEAs.

[0069] The final step in feature dimensionality reduction is exhaustive feature screening. We developed a series of RFR models using exhaustive combinations of all seven remaining alloy features as input vectors. By comparing the RMSE values of the model predictions for different feature combinations, we ultimately identified the three key features that most significantly impacted alloy hardness.

[0070] The features are further filtered by considering all possible combinations of these features to identify the subset with the lowest model error. Figure 5 The RMSE values of the RFR model based on different subsets of features ranging from 1 to 7 are shown. As can be seen from the figure, the minimum RMSE (marked with "★" in the figure) corresponds to 3 features, including W, VEC, and E.

[0071] Based on the above feature screening process, we built the RFR model using 80% of the dataset and 3 additional features, and the remaining 20% of the dataset was used to test the model. The prediction performance of our model is shown in Figure 2. Figure 6 As shown in the figure, the scatter points corresponding to the experimental values and predicted values of the alloys in the training set and the test set are close to the diagonal y=x, indicating that the model is appropriate.

[0072] Step 4: Use the most important feature subset obtained to train the selected model and tune the model's hyperparameters;

[0073] Using 10-fold cross-validation, we plotted learning curves to adjust the hyperparameter n_estimators, which has the greatest impact on the random forest model. A value of 31 yielded the best predictive performance, so we set n_estimators to 31. We then adjusted the parameter to reduce model complexity, thereby minimizing generalization error. We used max_depth, which has the second-highest impact and reduces model complexity. A value of 14 yielded the best model performance.

[0074] Step 5: The trained model can be used to predict the hardness of high-entropy alloys of unknown composition. The three important features obtained in step 3 of the unknown composition alloy are calculated and input into the trained model as the prediction set to obtain the corresponding hardness value. The high-entropy alloy with the highest predicted hardness is: Al 51 Co 22 Cr 19 Ni4Mn4, its predicted hardness is 805.3±1.9HV.

[0075] Step 6: Verify the alloy hardness of the predicted composition through testing.

[0076] Preparation of Al 51 Co 22 Cr 19 Ni4Mn4 high entropy alloy, wherein the atomic percentage of each element is as follows: Al 50-52%, Co 21-23%, Cr 18-20%, Ni 3-5%, and Mn 3-5%.

[0077] Step 1: Al 51 Co 22 Cr 19 The materials for preparing Ni4Mn4 high entropy alloy (Al material, Co material, Cr material, Ni material, Mn material) were placed in glass beakers respectively, and cleaned in ultrasonic cleaning device with acetone and anhydrous ethanol for 10-20 minutes, and finally dried with cold air for later use. 51 Co 22 Cr 19 The Ni4Mn4 high-entropy alloy material was converted to its corresponding weight based on atomic percentage, and each metal component was weighed. The weighed materials were placed into the copper crucible of a WK-II vacuum arc furnace in order of increasing melting point. Simultaneously, a layer of the surface of the pure titanium ingot was abraded using 400-grit sandpaper. After rinsing with alcohol and drying, the ingot was placed in the center crucible position of the vacuum furnace, and the furnace door was closed and locked.

[0078] Step 2: Vacuuming: After step 1, use mechanical and molecular pumps to vacuum to 1-5×10 -4 Pa, and then slowly fill in argon gas to make the vacuum degree reach 1~5×104 Pa, and then the vacuum degree of the arc furnace is pumped to 1~5×10 -4 Pa; repeat this process three times to ensure that there is no oxygen in the vacuum chamber; finally, slowly introduce argon to -0.02~-0.05Mpa.

[0079] Step 3: Melting: First, melt the pure titanium ingot 3 to 5 times, with the current controlled at 100-300A and the time being 30-150s. After each melting, turn the pure titanium ingot over and carry out the next melting; then melt the high entropy alloy. At this time, the current should be increased slowly to avoid the temperature rising too fast to generate large thermal stress and cause the sample to crack. The arc current is increased to a maximum of 300A and the duration is 50-200s; during the melting process, slightly rotate the arc rod to heat the material evenly; when turning off the current, also slowly reduce the current to avoid cracking caused by rapid cooling; use magnetic stirring with an external magnetic field during melting, and set the current to 1-3mA to fully ensure the uniformity of the composition;

[0080] Step 4: Repeat the smelting 4 to 7 times according to the requirements of step 2 to ensure uniform composition, and cool for 10-30 minutes to obtain high hardness Al 51 Co 22 Cr 19 Ni4Mn4 high entropy alloy.

[0081] The prepared Al 51 Co 22 Cr 19 Ni4Mn4 high entropy alloy phase structure analysis, such as Figure 7 Al 51 Co 22 Cr 19 The XRD pattern of Ni4Mn4 high entropy alloy shows that Al 51 Co 22 Cr 19 Ni4Mn4 high entropy alloy is mainly composed of B2 and BCC phases, with a small amount of FCC phase.

[0082] The Al prepared in Example 1 51 Co 22 Cr 19 The hardness of Ni4Mn4 high entropy alloy was tested according to Figure 3 It can be seen that Al 51 Co 22 Cr 19 Ni4Mn4 has good hardness, with a hardness value of 785.9±9.9HV.

[0083] The Al prepared in Example 1 51 Co 22 Cr 19Ni4Mn4 high entropy alloy was used to measure its microstructure and composition. Figure 8 Al 51 Co 22 Cr 19 The microstructure of Ni4Mn4 high entropy alloy and the surface distribution of Al, Co, Cr, Mn and Ni elements.

[0084] from Figure 8 The surface distribution diagram shows a distinct dendritic morphology. Dendrites are rich in Al, Co, and Ni, while interdendritic regions are rich in Al, Cr, and Mn. This is because Al easily combines with other elements. The Cr-rich region primarily represents the BCC phase, while the Ni- and Co-rich regions primarily represent the B2 phase.

[0085] From the above examples, it can be seen that the present invention provides a machine learning method and preparation process for quickly predicting the hardness of high entropy alloys. By mining a large amount of existing composition-performance data, the machine learning method can conveniently and accurately predict the hardness of the alloy, which can to some extent solve the problem that the traditional "trial and error method" of designing high entropy alloy composition is time-consuming, labor-intensive, and has inaccurate results.

[0086] Although the present invention has been described with respect to a limited number of embodiments, it will be apparent to those skilled in the art, having benefit of the foregoing description, that other embodiments are contemplated within the scope of the invention thus described. Furthermore, it should be noted that the language used in this specification has been selected primarily for readability and didactic purposes, rather than for the purpose of explaining or limiting the subject matter of the present invention. Consequently, many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the appended claims. The disclosure of the present invention is intended to be illustrative rather than restrictive of the scope of the invention, which is defined by the appended claims.

Claims

1. A machine learning method for rapidly predicting the hardness of high entropy alloys, characterized in that: The following steps are included: Step 1: Collect and organize the high entropy alloy hardness database, which is the Al-Co-Cr-Cu-Fe-Ni-Mn seven-element alloy system; calculate the characteristic descriptors based on the alloy composition, which include 20 features, namely average atomic radius, atomic radius difference, work function, average melting point, mixing enthalpy, mixing entropy, Ω parameter, valence electron concentration, average electronegativity, electronegativity difference, Λ parameter, number of mobile electrons, electron affinity, density, first ionization energy, Young's modulus, lattice distortion energy, cohesive energy, local size mismatch, and average shear modulus; construct a characteristic data set; Step 2: Select a suitable model based on the RMSE under different test set partition ratios. The suitable model is the random forest regression model; Step 3: Perform dimensionality reduction on the obtained features. Using the Pearson correlation coefficient, remove the six highly correlated and low-importance features, including atomic radius, local size mismatch, lattice distortion energy, number of mobile electrons, melting point, and shear modulus. The remaining 14 features are recursively eliminated by the random forest regression model to obtain a new feature subset containing 7 key features, namely atomic radius difference, work function, valence electron concentration, electron affinity, density, first ionization energy, and Young's modulus. By performing exhaustive feature screening on the random forest regression model, the three key features that have the greatest impact on the hardness of the alloy are determined, including work function, valence electron concentration, and Young's modulus. These three features are used to obtain the most important feature subset. Step 4: Use the most important feature subset obtained to train the selected random forest regression model and tune the hyperparameters of the random forest regression model; Step 5: The trained model can be used to predict the hardness of high-entropy alloys with unknown composition; Step 6: Verify the alloy hardness of the predicted composition through testing.

2. The machine learning method for rapidly predicting the hardness of high entropy alloys according to claim 1, characterized in that: The different test sets described in step 2 account for 15%, 20%, 25%, 30%, 35%, and 40% of the total dataset respectively.

3. The machine learning method for rapidly predicting the hardness of high-entropy alloys according to claim 1, characterized in that: The Pearson correlation coefficient measures the highly correlated nature of two features based on a Pearson correlation coefficient value greater than or equal to 0.

95. The basis for removing one of the two highly correlated features is the importance of the feature to the model, and the feature with high importance is retained.

4. The machine learning method for rapidly predicting the hardness of high entropy alloys according to claim 1, characterized in that The preparation method of Al-Co-Cr-Cu-Fe-Ni-Mn high entropy alloy comprises the following steps: Step 1, ultrasonic cleaning and material weighing: high-purity metal particles Al, Co, Cr, Cu, Fe, Ni, and Mn of the high-entropy alloy are placed in a container respectively, and cleaned in an ultrasonic cleaning device with acetone and anhydrous ethanol for 10-20 minutes in sequence, and finally the materials are blown dry with cold air for use; the high-entropy alloy is converted into corresponding weight according to atomic percentage and weighed; the weighed materials are placed in a copper crucible of a WK-II vacuum arc furnace in order of increasing melting point, and a layer of the surface of the pure titanium ingot is ground off with 400-grit sandpaper. After rinsing with alcohol and drying, the ingot is placed in the middle crucible position of the vacuum furnace, closed, and the furnace door is tightened and locked; Step 2: Vacuuming: After step 1, use mechanical and molecular pumps to vacuum to 1~5×10 -4 Pa, and then slowly fill in argon gas to make the vacuum degree reach 1~5×10 4 Pa, and then the vacuum degree of the arc furnace is pumped to 1~5×10 -4 Pa; repeat this process 3 times to ensure that there is no oxygen in the vacuum chamber; finally, slowly introduce argon gas to -0.02~-0.05Mpa; Step 3: Melting: First, melt the pure titanium ingot 3 to 5 times, with the current controlled at 100-300A and the time being 30-150 seconds. After each melting, turn the pure titanium ingot over and carry out the next melting. Then, melt the high-entropy alloy, with the arc current increased to a maximum of 300A and the duration being 50-200 seconds. During the melting process, slightly rotate the arc rod to ensure uniform heating of the material. When turning off the current, slowly lower the current to avoid cracking caused by rapid cooling. Use magnetic stirring with an external magnetic field during melting, and set the current to 1-3 mA to fully ensure the uniformity of the composition. Step 4: Repeat the smelting 4 to 7 times according to the requirements of step 2 to ensure uniform composition, and cool for 10 to 30 minutes to obtain a high-hardness Al-Co-Cr-Cu-Fe-Ni-Mn high-entropy alloy.

Images