Method for predicting formation of high-entropy alloy phase structure by adopting machine learning method

Abstract

The invention provides a method for predicting formation of a high-entropy alloy phase structure by adopting a machine learning method, which comprises the following steps of: S1, screening data from a plurality of papers subjected to peer review, and constructing a database with a powerful data set; s2, for the alloy in the database, calculating characteristic parameters by adopting a specific calculation formula; s3, screening out a feature set with the best performance by adopting feature screening; and S4, selecting different machine learning classification models for training. Compared with a traditional experimental method and molecular dynamics simulation, the method for predicting the formation of the high-entropy alloy phase structure by adopting the machine learning method provided by the invention has the advantages that the machine learning technology is more efficient and accurate in predicting the performance of the high-entropy alloy, so that a large amount of experimental time and cost can be saved; and the performance data of the alloy can be more accurately obtained, so that the research and development progress of the material is accelerated.

Description

Technical Field

[0001] This invention relates to the field of high-entropy alloy phase structure analysis, and more particularly to a method for predicting the formation of high-entropy alloy phase structures using machine learning. Background Technology

[0002] High-entropy alloys are a class of alloys composed of five or more main elements in near-equiatomic ratios. Compared to traditional alloys, high-entropy alloys have more complex compositions and structures, thus exhibiting a range of unique properties and application potential. High-entropy alloys have broad application prospects in various fields such as aerospace, energy, and medicine. In particular, high-entropy alloys can achieve an excellent combination of high strength and high ductility, thereby opening up new possibilities for engineering applications.

[0003] Currently, molecular dynamics methods are commonly used to study the microstructure of high-entropy alloys. However, this method has certain limitations, such as low computational efficiency and a high dependence on accurate potential functions. These problems limit the application of molecular dynamics methods in large-scale systems and over long time periods. With the rapid development of machine learning technology, we now have more efficient and accurate tools to address these issues. Machine learning can not only accurately predict various properties of high-entropy alloys, such as hardness, corrosion resistance, and magnetism, but also provide guidance on parameter ranges for synthesizing high-entropy alloys with specific phases.

[0004] The study of modern high-entropy alloys is more complex because their composition, containing five or more metallic elements, can form a large number of different phases and properties. However, this also makes the study of high-entropy compounds more complex, requiring a large investment of manpower, funds and time.

[0005] Therefore, it is necessary to provide a method for predicting the formation of high-entropy alloy phase structures using machine learning to solve the above-mentioned technical problems. Summary of the Invention

[0006] This invention provides a method for predicting the formation of phase structures in high-entropy alloys using machine learning, which solves the problem that the study of high-entropy compounds has become complex and requires a large investment of manpower, funds and time.

[0007] To address the aforementioned technical problems, this invention provides a method for predicting the formation of high-entropy alloy phase structures using machine learning, comprising the following steps:

[0008] S1. Select data from multiple peer-reviewed papers and build a database with a powerful dataset;

[0009] S2. For the alloys in the database, we use specific calculation formulas to calculate the characteristic parameters;

[0010] S3. Use feature filtering to select the feature set that performs best;

[0011] S4. Select different machine learning classification models for training, including K-Nearest Neighbors (KNN), Logistic Regression (LG), Decision Tree (DT), Naive Bayes (NB), Random Forest (RF), Gradient Boosting Tree (GBT), AdaBoost Tree (ADA), and Stacked Generalization Classifier (STA).

[0012] S5. Select the best performing model, perform oversampling, and then perform a second training in step S4.

[0013] Preferably, all papers in S1 are peer-reviewed SCI papers, and the data has a high degree of accuracy.

[0014] Preferably, the machine learning methods used in S4 include simple classification methods such as K-nearest neighbor classification, decision tree classification, and Naive Bayes classification, as well as ensemble classification methods such as random forest and gradient boosting tree.

[0015] Preferably, in step S3, a three-step feature screening method is used to obtain seven most valuable features through PCC, REF, and SFS filters.

[0016] Preferably, the REF is a method used to further remove saliency from irrelevant features. It typically involves calculating a p-value to measure the probability that the correlation coefficient occurred by chance. The PCC is a statistical method used to measure the strength and direction of the linear relationship between two variables. The SFS is an iterative analysis method used to handle the best combination of data. It divides the dataset into different combinations and then performs linear regression analysis within each combination to better obtain the feature set. Preferably, the database in S1 needs to be used with a computer. The computer includes a chassis with a baffle on one side. The baffle has mounting screws threaded inside all four sides. One end of each mounting screw is threaded into the inside of the side of the chassis. The baffle has ventilation holes inside.

[0017] Preferably, a fixing ring is fixedly installed on one side of the baffle, an installation ring is threadedly connected to the outer side of the fixing ring, a retaining ring is fixedly installed at one end of the inner side of the installation ring, the retaining ring is disposed at one end of the fixing ring, and an installation rod is fixedly installed at the other end of the inner side of the installation ring.

[0018] Preferably, an mounting plate is fixedly mounted on the inner side of the mounting ring, and a filter screen is fixedly mounted on the inner side of the mounting plate.

[0019] Preferably, a drive motor is fixedly installed on one side of the mounting rod, a fan blade is fixedly installed on the output end of the drive motor, a rotating shaft is fixedly installed on one side of the fan blade, the rotating shaft is rotatably connected to the inside of the filter screen, a rotating plate is fixedly installed on one end of the rotating shaft, a cleaning brush is fixedly installed on one side of the rotating plate, and the cleaning brush is disposed on one side of the filter screen.

[0020] Preferably, a filter plate is provided on the other side of the baffle, and locking screws are threadedly connected to the inside of the filter plate around its perimeter, with one end of each locking screw threaded into the inside of the baffle.

[0021] Compared with related technologies, the prediction method for the formation of high-entropy alloy phase structures provided by this invention using machine learning has the following advantages:

[0022] This invention provides a method for predicting the formation of phase structures in high-entropy alloys using machine learning. Compared with traditional experimental methods and molecular dynamics simulations, machine learning technology is more efficient and accurate in predicting the performance of high-entropy alloys. This not only saves a lot of experimental time and costs, but also obtains more accurate performance data of the alloys, thereby accelerating the research and development of materials. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the structure of a first embodiment of a method for predicting the formation of high-entropy alloy phase structures using machine learning, provided by the present invention.

[0024] Figure 2 for Figure 1 The detailed flowchart shown below;

[0025] Figure 3 for Figure 1 The first and second feature filtering maps are shown below;

[0026] Figure 4 for Figure 1 The third feature iteration diagram is shown below;

[0027] Figure 5 for Figure 1 The preliminary training results of the machine learning classifier are shown below.

[0028] Figure 6 for Figure 1 The data distribution shown is the result of oversampling.

[0029] Figure 7 for Figure 1 The final training results of the model shown;

[0030] Figure 8 for Figure 1 The calculation formula shown;

[0031] Figure 9 This is a schematic diagram of a second embodiment of a method for predicting the formation of high-entropy alloy phase structures using machine learning, provided by the present invention.

[0032] Figure 10 for Figure 9 The diagram shows a side cross-sectional view of the mounting plate.

[0033] Figure 11 for Figure 10 The enlarged schematic diagram of part A shown below;

[0034] Figure 12 This is a schematic diagram of the third embodiment of a method for predicting the formation of high-entropy alloy phase structures using machine learning, provided by the present invention.

[0035] The following are the labels in the diagram: 1. Chassis, 2. Baffle, 21. Mounting screw, 22. Ventilation hole, 3. Fixing ring, 4. Mounting ring, 41. Retaining ring, 42. Mounting rod, 43. Mounting plate, 44. Filter screen, 5. Drive motor, 51. Fan blade, 52. Shaft, 53. Rotating plate, 54. Cleaning brush, 6. Filter plate, 61. Locking screw. Detailed Implementation

[0036] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0037] First Embodiment

[0038] Please refer to the following: Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 and Figure 8 ,in, Figure 1 This is a schematic diagram of the structure of a first embodiment of a method for predicting the formation of high-entropy alloy phase structures using machine learning, provided by the present invention. Figure 2 for Figure 1 The detailed flowchart shown below; Figure 3 for Figure 1 The first and second feature filtering maps are shown below; Figure 4 for Figure 1 The third feature iteration diagram is shown below; Figure 5 for Figure 1 The preliminary training results of the machine learning classifier are shown below. Figure 6 for Figure 1 The data distribution shown is the result of oversampling. Figure 7 for Figure 1 The final training results of the model shown; Figure 8for Figure 1 The calculation formula is shown. A method for predicting the formation of high-entropy alloy phase structures using machine learning includes the following steps:

[0039] S1. Select data from multiple peer-reviewed papers and build a database with a powerful dataset;

[0040] S2. For the alloys in the database, we use specific calculation formulas to calculate the characteristic parameters;

[0041] S3. Use feature filtering to select the feature set that performs best;

[0042] S4. Select different machine learning classification models for training, including K-Nearest Neighbors (KNN), Logistic Regression (LG), Decision Tree (DT), Naive Bayes (NB), Random Forest (RF), Gradient Boosting Tree (GBT), AdaBoost Tree (ADA), and Stacked Generalization Classifier (STA).

[0043] S5. Select the best performing model, perform oversampling, and then perform a second training in step S4.

[0044] All papers in S1 are peer-reviewed SCI papers, and the data has a high degree of accuracy.

[0045] The machine learning methods used in S4 include simple classification methods such as K-nearest neighbor classification, decision tree classification, and Naive Bayes classification, as well as ensemble classification methods such as random forest and gradient boosting tree.

[0046] The S3 method employs a three-step feature selection process, using PCC, REF, and SFS filters to obtain seven of the most valuable features.

[0047] The REF is a method used to determine whether the PCC or other correlation coefficients are statistically significant. It typically involves calculating a p-value to measure the probability that the correlation coefficient occurred by chance. The PCC is a statistical method used to measure the strength and direction of the linear relationship between two variables. The SFS is a regression analysis method used to handle situations where there are non-linear relationships in the data. It divides the dataset into different segments and then performs linear regression analysis within each segment to better fit the data.

[0048] Data was selected from multiple peer-reviewed papers, ultimately yielding 2085 valuable data points. The specific data distribution is shown below. Figure 1For the above data, we used a specific calculation formula to calculate the feature parameters to construct a dataset with 64 feature parameters. The specific calculation formula can be found in Table S1. We used a three-step feature selection method, employing PCC, REF, and SFS filters to obtain the seven most valuable features. The results of the first step are shown in [Table S1]. Figure 3 (a) The results of the second screening are shown in Figure 3 (b) The final screening process is shown in Figure 4 The dataset is divided into a 70% training set and a 30% test set for training and validation. The best model is then oversampled and retrained. For convenience, the distribution of the oversampled data is displayed on a parallel coordinate graph. Figure 6 ;

[0049] KNN: K-Nearest Neighbors, a simple machine learning algorithm;

[0050] LG: Logistic Regression, a simple machine learning algorithm;

[0051] DT: Decision Tree, a simple machine learning algorithm;

[0052] NB: Naive Bayes, a simple machine learning algorithm;

[0053] RF: Random Forest, an ensemble machine learning algorithm;

[0054] GBT: Gradient Boosting Tree, an integrated machine learning algorithm;

[0055] ADA: AdaBoost Tree, an integrated machine learning algorithm;

[0056] STA: Stacked Generalized Classifier, an integrated machine learning algorithm;

[0057] This invention utilizes machine learning technology to predict and optimize the properties and phases of high-entropy alloys, providing materials researchers with an efficient and accurate tool. The computing platforms used in this invention include Python programs and Origin data processing software. This invention is applicable to both Windows and Linux systems, offering a flexible computing environment and saving experimental time and costs.

[0058] The working principle of the prediction method for the formation of high-entropy alloy phase structure using machine learning provided by this invention is as follows:

[0059] We compared the performance of simple models (such as linear regression) and ensemble models (such as random forest). Simple models are suitable for small data and simple features, but their expressive power is limited; while ensemble models perform better on complex data. We used four basic models and four ensemble methods, and selected seven optimal features from 64 features through a three-stage feature selection process. Model performance was evaluated through 8-fold cross-validation and comparison with peer studies. Finally, the model was further optimized using oversampling techniques. Feature selection is the process of removing irrelevant or redundant features from the main feature space, thereby reducing the dimensionality of the feature set. This process can accelerate model training, simplify model structure, enhance model interpretability, reduce the risk of overfitting, and improve model prediction accuracy. In this study, we used a combination of feature selection techniques, including recursive feature elimination (REF), Pearson correlation coefficient (PCC) feature selector, and sequential forward selection (SFS) method, forming a three-stage feature selection process. The selection process is as follows: Figure 3 and Figure 4 As shown; Figure 5After training the models, we observed that basic classification models like LG performed relatively poorly in this study, scoring below 0.6 on the training data. In contrast, ensemble classification methods showed significant advantages in this classification task. RF, in particular, not only achieved the highest score of 0.779 on the training set but also reached 72% accuracy on the test set, performing best overall. This model achieved over 80% accuracy for AM and IM classification. This is likely due to the clear boundaries between these two phases and other phases. For FCC and BCC, although there was some transition to other phases, the test accuracy also reached over 75% due to the support of the large amount of data. However, the accuracy of the other phases was below 70%. This may be partly because these phases transition from other simpler phases, and their relatively low proportion in the dataset leads to insufficient training on them. Specifically, for the IMS phase, the prediction accuracy is only 47%. Considering the presence of complex phases like BCC+IM and others in IMS, and the inclusion of features like BCC+FCC in SS, the model struggles to distinguish them accurately and is easily confused with phases like IM+BCC+FCC. In the original data, a significant problem is sample imbalance, which causes the model to perform poorly when predicting minority class samples. To address this, we decided to use oversampling to enhance the model's generalization ability to minority class samples. Oversampling aims to address class imbalance in the dataset, where some classes have significantly fewer samples than others. Using this technique, we can increase the number of minority class samples, making the samples more balanced across all classes, which helps improve the overall performance and accuracy of the model. After applying oversampling, we maintained the original dataset's test set partitioning while oversampling samples from other parts. Subsequently, we trained four ensemble models. Detailed processing and training procedures are as follows: Figure 5 As shown.

[0060] Compared with related technologies, the prediction method for the formation of high-entropy alloy phase structures provided by this invention using machine learning has the following advantages:

[0061] Compared with traditional experimental methods and molecular dynamics simulations, machine learning technology is more efficient and accurate in predicting the performance of high-entropy alloys. This not only saves a lot of experimental time and costs, but also provides more accurate performance data of alloys, thereby accelerating the research and development of materials.

[0062] Second Embodiment

[0063] Please refer to the following: Figure 9 , Figure 10 and Figure 11Based on the first embodiment of this application, which provides a method for predicting the formation of high-entropy alloy phase structures using machine learning, the second embodiment of this application proposes another method for predicting the formation of high-entropy alloy phase structures using machine learning. The second embodiment is merely a preferred embodiment of the first embodiment, and its implementation will not affect the independent implementation of the first embodiment.

[0064] Specifically, the second embodiment of this application provides a method for predicting the formation of high-entropy alloy phase structures using machine learning, which differs in that the database in S1 needs to be used in conjunction with a computer. The computer includes a chassis 1, and a baffle 2 is provided on one side of the chassis 1. The interior of the baffle 2 is threaded with mounting screws 21 on all four sides. One end of the mounting screws 21 is threaded to the interior of one side of the chassis 1. The interior of the baffle 2 has ventilation holes 22.

[0065] A fixing ring 3 is fixedly installed on one side of the baffle 2. A mounting ring 4 is threadedly connected to the outer side of the fixing ring 3. A retaining ring 41 is fixedly installed at one end of the inner side of the mounting ring 4. The retaining ring 41 is located at one end of the fixing ring 3. A mounting rod 42 is fixedly installed at the other end of the inner side of the mounting ring 4.

[0066] An installation plate 43 is fixedly installed on the inner side of the installation ring 4, and a filter screen 44 is fixedly installed on the inner side of the installation plate 43.

[0067] A drive motor 5 is fixedly installed on one side of the mounting rod 42. A fan blade 51 is fixedly installed at the output end of the drive motor 5. A rotating shaft 52 is fixedly installed on one side of the fan blade 51. The rotating shaft 52 is rotatably connected to the inside of the filter screen 44. A rotating plate 53 is fixedly installed at one end of the rotating shaft 52. A cleaning brush 54 is fixedly installed on one side of the rotating plate 53. The cleaning brush 54 is located on one side of the filter screen 44.

[0068] The working principle of the prediction method for the formation of high-entropy alloy phase structure using machine learning provided by this invention is as follows:

[0069] When in use, the user starts the drive motor 5, which drives the fan blades 51 to rotate, thereby causing the fan blades 51 to rotate the shaft 52. At this time, the fan blades 51 draw outside air into the inside of the casing 1, thereby dissipating heat from the components inside the casing 1. Simultaneously, the rotation of the shaft 52 drives the rotating plate 53 and the cleaning brush 54 to rotate mechanically, thereby cleaning the impurities adsorbed on the filter screen 44. Afterwards, the user can rotate the mounting ring 4 to make the mounting ring 4 threadedly connected to the outer side of the fixing ring 3, thus removing the mounting ring 4. At this time, the user can clean the impurities set between the filter screen 44 and the retaining ring 41.

[0070] Compared with related technologies, the prediction method for the formation of high-entropy alloy phase structures provided by this invention using machine learning has the following advantages:

[0071] The components, including baffle 2, fixing ring 3, mounting ring 4, mounting plate 43, filter screen 44, mounting rod 42, drive motor 5, fan blade 51, rotating shaft 52, rotating plate 53, and cleaning brush 54, work together to not only dissipate heat from the internal components of the casing 1, but also clean the impurities adsorbed on the filter screen 44.

[0072] Third Embodiment

[0073] Please refer to the following: Figure 12 Based on the first embodiment of this application, which provides a method for predicting the formation of high-entropy alloy phase structures using machine learning, the third embodiment of this application proposes another method for predicting the formation of high-entropy alloy phase structures using machine learning. The third embodiment is merely a preferred embodiment of the first embodiment, and its implementation will not affect the separate implementation of the first embodiment.

[0074] Specifically, the third embodiment of this application provides a method for predicting the formation of high-entropy alloy phase structure using machine learning, which differs in that, in this method, a filter plate 6 is provided on the other side of the baffle 2, and locking screws 61 are threadedly connected to the interior of the filter plate 6 around its perimeter, with one end of the locking screws 61 threadedly connected to the interior of the baffle 2.

[0075] The working principle of the prediction method for the formation of high-entropy alloy phase structure using machine learning provided by this invention is as follows:

[0076] When in use, the user places the filter plate 6 on the other side of the baffle 2, and then rotates the locking screw 61 so that one end of the locking screw 61 is threaded inside the other side of the baffle 2, thereby installing the filter plate 6.

[0077] Compared with related technologies, the prediction method for the formation of high-entropy alloy phase structures provided by this invention using machine learning has the following advantages:

[0078] The filter plate 6 and locking screw 61 work together to enhance the filtration of the gas entering the casing 1 by placing the filter plate 6 on one side of the ventilation hole 22 during use.

[0079] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for predicting phase structure formation of high-entropy alloys using a machine learning method, characterized by, It comprises the following steps: S1, data is screened from a plurality of papers that have undergone peer review to construct a database with a strong data set; S2, for the alloy of the database, a specific calculation formula is used to calculate the characteristic parameters; S3, the best feature set is screened out by using feature screening; S4, different machine learning classification models are selected for training, including K-nearest neighbor (KNN), logistic regression (LG), decision tree (DT) and naive Bayes (NB), random forest (RF), gradient boosting tree (GBT), AdaBoost tree (ADA) and stacked generalization classifier (STA); S5, the best model is selected, and oversampling technology is used for the second training of S4. 2.The method of claim 1, wherein the method is characterized by, All papers in S1 are peer-reviewed SCI papers, and the data has high accuracy. 3.The method of claim 1, wherein the method is characterized by: The machine learning methods used in S4 include K-nearest neighbor classification, decision tree classification, naive Bayes classification, and other simple classification methods, as well as random forest, gradient boosting tree, and other ensemble classification methods. 4.The method of claim 1, wherein the method is characterized by: In S3, three-step feature screening method is used, and seven most valuable features are obtained through PCC, REF, and SFS filters.

5. The method of claim 4, wherein the method is characterized by: REF is a significant method for further removing irrelevant features, which usually involves calculating a p-value to measure the likelihood of the correlation coefficient occurring by chance. PCC is a statistical method for measuring the strength and direction of the linear relationship between two variables. SFS is an iterative analysis method for handling the best combination of data, which divides the data set into different combinations and then performs linear regression analysis within each combination to better obtain the feature set.

6. The method of claim 1, wherein the method is characterized by: The database in S1 needs to be used with a computer, and the computer comprises a case, one side of the case is provided with a baffle, the inside of the baffle is screw connected with a mounting screw around, one end of the mounting screw is screw connected with the inside of the case, the inside of the baffle is provided with a ventilation hole.

7. The method of claim 6, wherein the method is characterized by: One side of the baffle is fixedly installed with a fixed ring, the outer side of the fixed ring is screw connected with a mounting ring, one end of the inner side of the mounting ring is fixedly installed with a blocking ring, the blocking ring is arranged at one end of the fixed ring, the other end of the inner side of the mounting ring is fixedly installed with a mounting rod. 8.The method of claim 7, wherein the method is characterized by, The inner side of the mounting ring is fixedly installed with a mounting plate, and the inner side of the mounting plate is fixedly installed with a filter screen. 9.The method of claim 8, wherein the method is characterized by, One side of the mounting rod is fixedly installed with a driving motor, the output end of the driving motor is fixedly installed with a fan blade, one side of the fan blade is fixedly installed with a rotating shaft, the rotating shaft is rotatably connected to the inside of the filter screen, one end of the rotating shaft is fixedly installed with a rotating plate, one side of the rotating plate is fixedly installed with a cleaning brush, and the cleaning brush is arranged on one side of the filter screen. 10.The method of claim 9, wherein the method is characterized by: The other side of the baffle is provided with a filter plate, the inside of the filter plate around is screw connected with a locking screw, one end of the locking screw is screw connected with the inside of the baffle.

Images