A method for designing amorphous alloys based on random forest and SHAP analysis

Abstract

The application relates to an amorphous alloy design method based on a random forest and SHAP analysis, which comprises the following steps: obtaining an alloy to be designed, performing SHAP analysis on elements in the alloy to be designed, and screening key elements; the key elements are a plurality of elements that have the greatest influence on the amorphous alloy forming capacity; the key elements are screened by utilizing mutual information analysis, an element combination to be optimized is obtained, the optimal proportion and concentration range of each element in the element combination to be optimized are determined according to a binary phase diagram and a SHAP dependence diagram, and an optimal element combination is obtained; the optimal element combination is input into a random forest model, and the amorphous alloy forming capacity value is predicted; the random forest model is obtained by training a training set; wherein the training set comprises element composition and proportion of the alloy and a target variable. The application effectively reduces the number of experimental verifications, shortens the alloy design time, and provides a scientific and efficient solution for the development of amorphous alloys and multi-component materials.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of material design, and particularly relates to a non-crystalline alloy design method based on random forest and SHAP analysis. BACKGROUND

[0002] Non-crystalline alloys have wide application prospects in aerospace, electronic devices, energy and other fields due to their excellent mechanical properties, corrosion resistance, magnetic properties and other characteristics. However, the design and optimization of non-crystalline alloys often face great challenges. This is because the glass-forming ability (GFA) of non-crystalline alloys is significantly affected by alloy composition, element ratio and complex interaction between elements. Traditional composition optimization process often relies on repeated laboratory tests, which not only has high cost, but also takes a long time, and cannot meet the demand of rapid design.

[0003] With the advancement of the materials genome project, data-driven methods have gradually shown great potential in material design. In particular, machine learning methods can extract rules from large data and efficiently predict material performance, providing new possibilities for material research and development. However, in the process of applying machine learning to non-crystalline alloy design, how to accurately identify and select elements that have a significant impact on GFA, and how to effectively analyze the interaction between different elements, are still difficult problems to be solved:

[0004] 1. Low efficiency of alloy composition screening: The glass-forming ability (GFA) of non-crystalline alloys is affected by alloy composition, element ratio and complex interaction between elements, and traditional experimental methods cannot quickly and efficiently screen the best alloy composition.

[0005] 2. Too many experimental verifications: Current alloy design often relies on a large number of experiments to verify and optimize the composition, which not only requires high cost, but also increases the research and development cycle, and cannot meet the demand of rapid design.

[0006] 3. Complex interaction between elements is difficult to analyze: Traditional methods lack effective analysis methods in identifying and processing complex interactions between elements, resulting in low design efficiency and inaccurate optimization process.

[0007] 4. Difficulty in optimizing element ratio: Even if important elements can be identified, how to reasonably optimize the ratio of these elements, especially in multi-element systems, is still a difficult problem to be solved.

[0008] In order to better understand and optimize the alloy composition, the present application proposes a method for screening and optimizing amorphous alloy composition based on machine learning. This method combines random forest model and SHAP (SHapley Additive exPlanations) analysis, which can effectively identify the elements that have the greatest impact on the target performance (such as GFA), and further screen and optimize through mutual information analysis and binary phase diagram. Specifically, the random forest model provides feature importance evaluation, while the SHAP analysis provides detailed contribution of individual elements and element combinations to the model output. In addition, mutual information analysis is used to identify the dependency between key elements, and binary phase diagram combined with SHAP interaction analysis further optimizes the element ratio, thereby achieving efficient composition design.

[0009] Through this method, the number of experimental verifications and design time can be significantly reduced, providing an effective solution for the design of multi-component alloys such as amorphous alloys. This not only improves the efficiency of alloy development, but also lays a technical foundation for rapid iteration and optimization of new materials. SUMMARY

[0010] In order to solve the above-mentioned problems existing in the prior art, the purpose of the present application is to provide a method for designing amorphous alloys based on random forest and SHAP analysis, which can effectively identify the elements that have the greatest impact on the target performance (such as GFA) by combining random forest model and SHAP analysis, and further screen and optimize through mutual information analysis and binary phase diagram. Specifically, the random forest model provides feature importance evaluation, while the SHAP analysis provides detailed contribution of individual elements and element combinations to the model output. In addition, mutual information analysis is used to identify the dependency between key elements, and binary phase diagram combined with SHAP interaction analysis further optimizes the element ratio, thereby achieving efficient composition design, which significantly reduces the number of experimental verifications and design time, providing an effective solution for the design of multi-component alloys such as amorphous alloys. This not only improves the efficiency of alloy development, but also lays a technical foundation for rapid iteration and optimization of new materials.

[0011] To achieve the above-mentioned purpose, the present application provides the following scheme:

[0012] A method for designing amorphous alloys based on random forest and SHAP analysis, comprising:

[0013] Obtaining an alloy to be designed, performing SHAP analysis on the elements in the alloy to be designed, and screening out key elements; the key elements are a number of elements that have the greatest impact on the amorphous alloy forming ability;

[0014] screening the key elements by using mutual information analysis, obtaining an element combination to be optimized, determining an optimal proportion and a concentration range of each element in the element combination to be optimized according to a binary phase diagram and a SHAP dependence diagram, and obtaining an optimal element combination;

[0015] inputting the optimal element combination into a random forest model to predict an amorphous alloy forming ability value; the random forest model is obtained by training a training set; wherein the training set includes element composition and proportion of an alloy and a target variable.

[0016] Optionally, the screening of the key elements comprises:

[0017] performing SHAP analysis on the element composition and proportion to quantify the contribution degree of each element to the amorphous alloy forming ability under different conditions, i.e., a Shapley value;

[0018] Based on the Shapley value, the elements in the designed alloy are sorted, a screening threshold is set, and the key elements are further screened out.

[0019] Optionally, obtaining the element combination to be optimized comprises:

[0020] The mutual dependence between each element in the key elements is evaluated by using mutual information analysis, and element redundancy of the key elements is removed according to the mutual dependence, to obtain an element combination to be optimized; the element combination to be optimized is an element combination that has the greatest impact on the amorphous alloy forming ability.

[0021] Optionally, determining the optimal proportion of each element in the element combination to be optimized comprises:

[0022] According to the eutectic point of the binary phase diagram, the solubility, phase transition temperature and behavior of liquid alloy of each element in the element combination to be optimized are obtained, and based on the solubility, phase transition temperature and behavior of liquid alloy of each element, the optimal proportion of each element in the element combination to be optimized is determined.

[0023] Optionally, determining the concentration range of each element in the element combination to be optimized comprises:

[0024] According to the SHAP dependence diagram, the sensitivity of the concentration of each element in the element combination to be optimized to the amorphous alloy forming ability is obtained, and the target relationship between the element concentration and the amorphous alloy forming ability is obtained in combination with the eutectic point;

[0025] Based on the target relationship between the element concentration and the amorphous alloy forming ability, the concentration range of each element in the element combination to be optimized is determined.

[0026] Optionally, obtaining the random forest model comprises:

[0027] obtaining the element composition and proportion of the alloy and a target variable; the target variable is a target parameter of amorphous alloy forming ability;

[0028] Taking the element composition and proportion of the alloy and the target variable as input data, preprocessing the input data, inputting the preprocessed input data into an original random forest model, optimizing the hyperparameters of the model, and balancing multiple performance indicators of the model by using an NSGA-II algorithm to obtain the random forest model.

[0029] Optionally, preprocessing the input data comprises:

[0030] The element content in the input data is scaled to a target interval by using a data cleaning and standardization method.

[0031] Optionally, the hyperparameters of the model comprise: the number of trees, the maximum depth, and the minimum sample size.

[0032] The present application has the following beneficial effects:

[0033] Efficient screening of alloy components: the present application uses a random forest regression model to extract key factors affecting GFA from a large amount of experimental data. Random forest can automatically screen out the elements that have the greatest impact on glass forming ability through feature importance evaluation, realize efficient and accurate alloy component screening, and significantly improve the efficiency of alloy design.

[0034] Reducing the number of experimental verifications: the present application can directly predict the influence of alloy components on GFA without a large number of experiments through the trained random forest model, reducing the limitations of traditional methods that rely on experiments. Through the data-driven prediction model, the alloy design cycle is greatly shortened, the experimental cost is reduced, and the problems of high cost and long cycle in the traditional design process are solved.

[0035] Precise analysis of the interaction between elements: the present application can deeply mine and explain the contribution of each element to GFA by introducing SHAP analysis, not only considering the effect of single element, but also revealing the complex interaction between elements. SHAP analysis provides detailed contribution of each element and element combination to the model output, ensuring the accuracy and reliability of the design process.

[0036] Optimizing element ratio: the present application identifies the mutual dependence between elements through mutual information analysis, so as to further screen out element combinations with high synergy. Combined with binary phase diagram and SHAP dependence diagram analysis, the composition ratio of the alloy can be accurately optimized, and the effect of ratio optimization is improved.

[0037] Improving alloy design efficiency: The invention effectively reduces the number of experiments and significantly improves the efficiency of alloy composition screening and optimization by using machine learning techniques, especially random forests and SHAP analysis. Through data-driven methods, the alloy design process is optimized, greatly shortening the research and development cycle.

[0038] Reducing experimental costs and time: The invention can avoid tedious experimental processes through model prediction, saving a large amount of experimental resources and reducing overall research and development costs. At the same time, the alloy design cycle is greatly shortened, adapting to the demand for rapid research and development.

[0039] Accurate identification of key elements and interactions: The invention uses SHAP analysis to provide detailed explanations for the contribution of each element, clearly showing the interactions between elements, optimizing alloy composition ratios, and improving design accuracy. This not only improves the quality of amorphous alloys but also provides a reference for the design of other alloys.

[0040] Efficient multi-element system optimization: The invention can handle complex multi-element systems, improving the design efficiency and reliability of multi-component alloys through precise ratio optimization, reducing errors and trial-and-error costs in experiments.

[0041] Improving alloy design efficiency: The invention effectively reduces the number of experiments and significantly improves the efficiency of alloy composition screening and optimization by using machine learning techniques, especially random forests and SHAP analysis. Further through data-driven methods, the alloy design process is optimized, greatly shortening the research and development cycle.

[0042] Reducing experimental costs and time: The invention can avoid tedious experimental processes through model prediction, saving a large amount of experimental resources and reducing overall research and development costs. At the same time, the alloy design cycle is greatly shortened, adapting to the demand for rapid research and development.

[0043] Efficient multi-element system optimization: The invention can handle complex multi-element systems, improving the design efficiency and reliability of multi-component alloys through precise ratio optimization, reducing errors and trial-and-error costs in experiments.

[0044] Promoting the progress of the Materials Genome Project: The invention combines machine learning and data analysis methods, conforming to the development trend of the Materials Genome Project, providing new technical support for rapid material research and development, and promoting the digital transformation of material design.

[0045] Expanding the industrial application of amorphous alloys: The invention reduces experimental verification and design cycle, reducing the research and development costs of amorphous alloys and improving their industrialization feasibility. This method not only has important application prospects in the fields of aerospace, electronic devices, etc., but also provides new ideas for the design of other high-end materials.

[0046] Provide solutions for other material systems: Although the present invention focuses on the design of amorphous alloys, its method has wide generality. Whether it is other metal alloys, ceramic materials, or multi-component composite materials, it can be used for efficient screening and optimization.

[0047] Enhance the intelligent and automated level of alloy design: The present invention introduces intelligent algorithms and automated optimization processes, reduces manual intervention, improves the automation level of the alloy design process, and further improves design efficiency and accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative labor.

[0049] Figure 1 A non-crystalline alloy design method flowchart based on random forest and SHAP analysis for an embodiment of the present invention;

[0050] Figure 2 R 2 and mean square error (MSE) index schematic diagram for the model in the training set and the test set of the embodiment of the present invention;

[0051] Figure 3 Performance diagram of the model in the embodiment of the present invention for predicting the formation energy (Dmax) of amorphous alloys;

[0052] Figure 4 Element feature importance diagram based on SHAP analysis for the embodiment of the present invention;

[0053] Figure 5 Weighted mutual information value schematic diagram of different alloy systems for the embodiment of the present invention;

[0054] Figure 6 Column chart of the interaction intensity between element pairs for the embodiment of the present invention;

[0055] Figure 7 Binary phase diagram of each element system for the embodiment of the present invention;

[0056] Figure 8 Distribution situation schematic diagram of the contribution degree of different elements in the amorphous alloy composition design with the concentration change based on SHAP analysis for the embodiment of the present invention;

[0057] Figure 9 Parallel coordinate diagram for the embodiment of the present invention;

[0058] Figure 10 Al for different thicknesses of the embodiment of the application 39 Co 17 Ce 22 La 22 X-ray diffraction (XRD) spectrum of the material. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the application will be apparently and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the application.

[0060] In order to make the above objectives, characteristics and advantages of the application more apparent, comprehensible and easier to understand, the application will be further described in detail below with reference to the drawings and specific embodiments.

[0061] As Figure 1 shown, the embodiment discloses a non-crystalline alloy design method based on random forest and SHAP analysis, including: obtaining an alloy to be designed, performing SHAP analysis on elements in the alloy to be designed, and screening out key elements; the key elements are a number of elements that have the greatest impact on the non-crystalline alloy forming ability; the key elements are screened by mutual information analysis, and a combination of elements to be optimized is obtained; according to a binary phase diagram and a SHAP dependence diagram, the best proportion and concentration range of each element in the combination of elements to be optimized are determined, and an optimal element combination is obtained; the optimal element combination is input into a random forest model to predict a non-crystalline alloy forming ability value; the random forest model is obtained by training a training set; wherein the training set includes: element composition and proportion of the alloy and a target variable.

[0062] Specifically:

[0063] The embodiment discloses a non-crystalline alloy design method based on random forest and SHAP analysis, including:

[0064] Data acquisition and preprocessing: collect alloy composition and corresponding glass forming ability (GFA) parameters from experimental data, databases or literature, clean the data, remove outliers, and use standardization techniques to adjust the content of each element to the [0, 1] interval;

[0065] Model training and evaluation: use a random forest regression model to train the preprocessed data to predict the maximum forming ability (Dmax) of the alloy, and evaluate the performance of the model by R 2 and mean square error (MSE);

[0066] Element screening: Determine the contribution of each element in the designed alloy to Dmax through SHAP analysis, and select key elements that have a significant impact on GFA;

[0067] Combination screening: Perform mutual information analysis on the selected key elements to evaluate the synergy between element combinations, and select element combinations with higher synergy;

[0068] Component ratio optimization: Based on the high-synergy element combination, combined with binary phase diagram and SHAP dependency diagram analysis, determine the optimal concentration range of each element in the alloy to improve the amorphous alloy forming ability;

[0069] Experimental verification: According to the optimized element combination, perform experimental verification, synthesize the alloy and verify its amorphous structure through X-ray diffraction (XRD) analysis.

[0070] Further, obtaining the random forest model comprises: obtaining the element composition and proportion of the alloy and the target variable; the target variable is the target parameter of the amorphous alloy forming ability; taking the element composition and proportion of the alloy and the target variable as input data, preprocessing the input data, inputting the preprocessed input data into the original random forest model, optimizing the hyperparameters of the model, and balancing multiple performance indicators of the model using the NSGA-II algorithm to obtain the random forest model.

[0071] Further, preprocessing the input data comprises: using data cleaning and standardization methods to scale the content of each element in the input data to the target interval.

[0072] Specifically:

[0073] Data collection and preprocessing: Collect alloy composition and GFA parameters (maximum forming ability Dmax) from experimental data, databases or related literature. Use data cleaning and standardization methods (MinMaxScaler) to scale the content of each element to the [0,1] interval to ensure the consistency of feature weights during model training. The data set is divided into training set and test set in the ratio of 8:2 to ensure the training and verification of the model.

[0074] Random forest model training and performance evaluation: Use the random forest regression model (Random Forest Regressor) to learn the training set to predict the Dmax of the alloy and obtain the feature importance. Evaluate the model performance through the determination coefficient (R 2 ) and mean square error (MSE) indicators to ensure that the prediction ability of the model on the training set and test set has good generalization.

[0075] The random forest model training includes: input data: the elemental composition and proportion of the alloy, such as the content and proportion of elements such as Ni, Cu, Fe, etc. in the alloy. The characteristics of each alloy sample are the percentages of these elements. Output data: the glass forming ability (GFA) of the alloy, i.e. the predicted Dmax value. Through the random forest regression model (Random Forest Regressor), the present application learns the training set. The goal of training is to establish the relationship between the alloy composition and its glass forming ability, so that the model can predict Dmax (GFA).

[0076] The random forest regression model is optimized, mainly in the following two aspects:

[0077] Hyperparameter optimization: the hyperparameters of the model (such as the number of trees, maximum depth, minimum sample size, etc.) are automatically optimized by Optuna, which significantly improves the prediction accuracy and stability of the model. In this way, the present application realizes more efficient alloy composition design and reduces the risk of model overfitting.

[0078] Multi-objective optimization: in order to balance the multiple performance indicators (such as R 2 and mean square error MSE) of the model, the present application uses NSGA-II algorithm (non-dominated sorting genetic algorithm II) for multi-objective optimization, so that the model can find the optimal balance between prediction accuracy and error, further enhancing the generalization ability of the model.

[0079] Random forest model training and performance evaluation: in the present application, the random forest regression model (Random Forest Regressor) is used to learn the training set to predict the glass forming ability (GFA, referred to as Dmax) of the alloy. Specifically, the input data of the training set is the elemental composition and proportion of the alloy, which is obtained by analyzing the chemical composition of the alloy, and the output data is the GFA value (i.e. Dmax) of the alloy. The goal of the random forest model is to predict the glass forming ability of the alloy based on its elemental composition.

[0080] Further, the key elements are screened out, including: SHAP analysis of the elemental composition and proportion, quantifying the contribution of each element to the amorphous alloy forming ability under different conditions, i.e. Shapley value; based on the Shapley value, the elements in the designed alloy are sorted, and a screening threshold is set to further screen out the key elements.

[0081] Specifically:

[0082] To evaluate the impact of input features (i.e., element composition and proportion) on the amorphous alloy forming ability, the present invention employs the SHAP analysis (SHapley Additive exPlanations) method for feature importance analysis. SHAP is an explanation method based on cooperative game theory, which can provide a Shapley value for each feature, measuring the specific contribution of each input feature to the model output.

[0083] SHAP provides the precise contribution of each feature to the model output by considering all possible arrangements of feature values, and can quantify the interaction effects between features. Specifically, SHAP analysis calculates the marginal contribution of each feature under all possible feature combinations, thereby evaluating its impact on the prediction of the target variable (Dmax).

[0084] SHAP analysis for element screening: Introduce the elements of the alloy to be designed into the SHAP value analysis, quantify the contribution of each element to the amorphous alloy forming ability under different conditions. The application of SHAP value makes the present invention not only able to evaluate the importance of elements, but also to explain the specific influence of each element and its combination on the amorphous alloy forming ability, providing higher interpretability and scientific basis.

[0085] To accurately quantify the contribution of each element in the prediction of the glass forming ability (GFA, Dmax) of the alloy, the present invention introduces SHAP analysis (SHapley Additive exPlanations). SHAP analysis is based on the idea of game theory, which evaluates the influence of each element (feature) on the glass forming ability by calculating the Shapley value. This process not only reveals the marginal contribution of each element to the glass forming ability, but also identifies elements with important influence, thereby guiding alloy design.

[0086] The specific analysis process is as follows:

[0087] Calculate SHAP value:

[0088] Use SHAP interpreter (shap.TreeExplainer(model)) to calculate the contribution of each element of the alloy to be designed to Dmax.

[0089] Shapley value is based on the idea of cooperative game theory, measuring the marginal contribution of each element under all possible feature arrangements. For each alloy sample, SHAP value provides a specific numerical value for each element, indicating the positive or negative impact of the element on the glass forming ability.

[0090] Shapley value calculation: By considering all possible orders of features, the SHAP explainer calculates the marginal contribution of each element to the glass-forming ability. This process takes into account the interactions between features, ensuring that the evaluation of feature contributions is both comprehensive and accurate.

[0091] Quantifying feature contributions: Global feature importance: By calculating the Shapley value of each element (feature), the invention obtains a global ranking of feature importance. The importance of a feature is quantified by its average Shapley value, with a larger Shapley value indicating a greater contribution to the amorphous alloy-forming ability. The invention ranks the average contribution of each element, thereby screening out the elements that have the greatest impact on the amorphous alloy-forming ability.

[0092] Element screening process: Based on the SHAP value ranking, the invention can screen out the elements that play the most important role in alloy design. Specifically, elements with greater contributions will be given priority, as they have the greatest impact on the amorphous alloy-forming ability in the model. Through the SHAP summary plot, the invention can visually display the average contribution of all elements to the amorphous alloy-forming ability. This allows designers to quickly identify key elements and optimize alloy composition based on the impact of these elements.

[0093] Further, obtaining the combination of elements to be optimized includes: using mutual information analysis to evaluate the interdependence between each element in the key elements, according to the interdependence, eliminating the element redundancy of the key elements, and obtaining the combination of elements to be optimized; the combination of elements to be optimized is the combination of elements that has the greatest impact on the amorphous alloy-forming ability.

[0094] Specifically:

[0095] Element combination screening and mutual information analysis: Use mutual information analysis to screen the element combinations screened by SHAP, evaluate their correlation and synergy.

[0096] SHAP analysis for element screening and mutual information analysis for alloy system screening:

[0097] After quantifying the contribution of each element to the amorphous alloy-forming ability (GFA) (i.e. Dmax) through SHAP analysis and screening out the key elements, the next step is to further screen the combinations of these screened elements based on mutual information analysis to determine the optimal alloy system.

[0098] Mutual information analysis to screen element combinations: After identifying the key elements, the next step is to further screen the combinations of key elements based on mutual information analysis to identify the optimal element combination. Mutual information analysis is mainly used to evaluate the mutual dependence between two or more variables, helping to identify which element combination can most effectively affect the glass forming ability (GFA) of amorphous alloys.

[0099] Objective of mutual information analysis: The main goal of mutual information analysis is to identify element combinations that have the greatest information gain and can effectively synergize. By evaluating the mutual information value of each pair or each multi-element combination, the invention can find which element combination has the greatest impact on the glass forming ability of amorphous alloys and can reduce the redundant effects of other elements.

[0100] Mutual information analysis can quantify the dependence between elements to identify element combinations with the best synergistic effect. This process ensures that the screened element combinations have the potential to improve the glass forming ability (GFA) of amorphous alloys.

[0101] Specific process of mutual information analysis:

[0102] Define the input features of mutual information analysis, select the key elements screened from SHAP analysis as the input features of mutual information analysis.

[0103] For each pair of elements, use mutual information regression analysis (mutual_info_regression) to calculate their mutual information value with Dmax, i.e. the contribution of these element combinations to the glass forming ability of amorphous alloys. The higher the mutual information value, the greater the impact of these element combinations on the glass forming ability of amorphous alloys, and the higher the predictive power.

[0104] Synergistic effect of screened combinations: By calculating the mutual information value of each element combination, the invention ranks the element combinations. Select element combinations with high mutual information values as priority combinations. High mutual information value element combinations have strong synergistic effects. Element combinations with high information gain: According to the ranking results, screen element combinations that can maximize information gain. These combinations help reduce the impact of redundant features and may have strong synergistic effects in alloys.

[0105] Screening high mutual information combinations: From mutual information analysis, screen element combinations with high mutual information, which are usually elements with strong synergistic effects with each other and can effectively improve the glass forming ability (Dmax) of amorphous alloys. By selecting element combinations with high mutual information values, designers can effectively optimize alloy composition to ensure that element combinations in the alloy system can play the greatest role and improve overall performance.

[0106] Further, the determining the optimal proportion of each element in the element combination to be optimized comprises: obtaining the solubility, phase transition temperature and behavior of liquid alloy of each element in the element combination to be optimized according to the eutectic point of the binary phase diagram; and determining the optimal proportion of each element in the element combination to be optimized based on the solubility, phase transition temperature and behavior of liquid alloy of each element.

[0107] Further, the determining the concentration range of each element in the element combination to be optimized comprises: obtaining the sensitivity of the concentration of each element in the element combination to be optimized to the amorphous alloy forming ability according to the SHAP dependence diagram, and obtaining the target relationship between the element concentration and the amorphous alloy forming ability in combination with the eutectic point; and determining the concentration range of each element in the element combination to be optimized based on the target relationship between the element concentration and the amorphous alloy forming ability.

[0108] Specifically:

[0109] Combining the proportion optimization of the binary phase diagram: on the basis of the determined element combination, the physical interaction between elements is studied by binary phase diagram analysis, especially the eutectic point and the amorphous forming region. In combination with the SHAP dependence diagram, the influence of a single element and its combination on the GFA at different concentrations is analyzed, and the specific composition ratio of the element is optimized to improve the amorphous forming ability of the alloy.

[0110] Combining the eutectic point of the binary phase diagram and the SHAP dependence diagram to optimize the composition ratio of the amorphous alloy: after determining the key elements and screening out the elements with high contribution, the next step is to optimize the composition ratio of the alloy through the eutectic point criterion of the binary phase diagram and SHAP analysis to improve the glass forming ability (GFA) (i.e. Dmax). Specifically, by combining the eutectic point analysis and the SHAP dependence diagram, the influence of a single element and its combination on the GFA at different concentrations can be analyzed, so as to obtain the specific composition ratio of the amorphous alloy formation.

[0111] The eutectic point criterion is a very important theoretical basis in the binary phase diagram, which describes the phase transition behavior between the liquid phase and the solid phase of two elements or alloys at a certain temperature and element ratio. In alloy design, the use of eutectic point can help the present application to understand the solubility, phase transition temperature and behavior of liquid alloy of different elements. In the design of alloy composition, the eutectic point criterion helps to determine the optimal proportion of elements in the alloy, so that it can effectively avoid the formation of crystals and promote the generation of amorphous phase during the cooling process. In a binary alloy system, the eutectic point is often the boundary of the amorphous forming region, so the element ratio in this region has a high glass forming ability GFA.

[0112] SHAP analysis is used to quantify the contribution of each element to the glass forming ability Dmax. After screening the key elements through SHAP analysis, the invention can further analyze the specific impact of individual elements and element combinations on Dmax at different concentrations using SHAP dependence plots. By analyzing SHAP dependence plots, designers can identify the sensitivity of each element concentration to the amorphous alloy forming ability, and which concentration interval can maximize Dmax. Using the eutectic point criterion and SHAP dependence plots, the composition of the alloy can be further optimized. Through SHAP dependence plots, it can be clearly pointed out that near the eutectic point, the concentration change of certain elements has a significant impact on GFA. By combining this information with the amorphous forming region in the phase diagram, designers can determine the optimal element ratio and concentration range to improve the glass forming ability of the alloy.

[0113] Through SHAP interaction analysis and physical interaction research, the invention can analyze the impact of element combinations on Dmax (glass forming ability). Certain element combinations may have strong synergistic effects near the eutectic point, which can optimize the performance of amorphous alloys. Combined with SHAP analysis, designers can identify element combinations with the greatest synergistic effect and adjust their concentrations to optimize the Dmax of the alloy. Concentration changes affect GFA: using SHAP dependence plots, the contribution of each element to GFA at different concentrations can be analyzed, and the concentration distribution of these elements is ensured to be in the eutectic point region, i.e. the optimal concentration interval between the liquid and solid phases of the alloy, to enhance the amorphous alloy forming ability.

[0114] Combined with SHAP analysis, phase diagram analysis and eutectic point criteria, the optimal concentration interval of key elements in the alloy can be determined. By optimizing the concentration of these elements, they are ensured to be in the amorphous forming region, i.e. the eutectic point region or its vicinity, to promote the formation of amorphous phase in the alloy during rapid cooling.

[0115] The trained random forest model is used to predict the optimal element combination designed, to evaluate its forming ability (Dmax). Finally, experimental verification is carried out to ensure the feasibility of the design. The verification results should be consistent with the model prediction, confirming the reliability of the optimization method in practical application.

[0116] As shown in Figure 2 , the R 2 and mean squared error (MSE) indicators of the model on the training set and test set are used to measure the fitting performance and prediction ability of the model.

[0117] As shown in Figure 3Figure 2 shows the model's performance in predicting the formation energy (Dmax) of amorphous alloys. The horizontal axis represents the measured Dmax, and the vertical axis represents the model's predicted Dmax. The blue asterisks in the figure represent the training set data, the red circles represent the test set data, and the black dashed line represents the ideal "perfect fit," where the predicted values ​​are completely consistent with the measured values.

[0118] like Figure 4 The element feature importance diagram based on SHAP analysis shows the weight of the influence of different elements on the model output. This diagram is used to illustrate the importance ranking of each element in the design of amorphous alloys.

[0119] like Figure 5 As shown in Figure 3, the weighted mutual information values ​​of different alloy systems can be used to identify the correlation between elements and screen out the element combination with the largest information gain through mutual information analysis.

[0120] like Figure 6 The histogram of the interaction strength between element pairs shows the mutual influence of different element combinations in the design of amorphous alloy compositions. This graph is used to evaluate the synergistic effect of element combinations. The higher the interaction strength, the more significant the interaction between these element pairs, which may have a positive impact on the target properties of the alloy.

[0121] like Figure 7 As shown, the binary phase diagrams of each element system show the eutectic points of different alloy systems.

[0122] like Figure 8 Figure 1 shows the distribution of the contribution of different elements in the amorphous alloy composition design based on SHAP analysis as their concentrations change. These figures are used to illustrate the contribution pattern of each high-priority element combination in different concentration ranges, helping to optimize the alloy composition design.

[0123] like Figure 9 As shown in Figure 1, the parallel coordinate plot shows the effect of the concentrations of four elements, Al, Co, Ce, and La, on the amorphous alloy forming energy (Dmax). A larger Dmax value indicates a stronger amorphous alloy forming ability. The plot uses a color mapping path to show the contribution of different component combinations to Dmax, with the color gradually changing from blue (indicating high Dmax) to red (indicating low Dmax).

[0124] like Figure 10 As shown, different thicknesses of Al 39 Co 17 Ce 22 La 22 X-ray diffraction (XRD) patterns of the material are used to analyze the crystal structure characteristics of the material at different casting diameters (4mm, 5mm, and 6mm).

[0125] The above described embodiments are only to illustrate the preferred modes of the present application, and are not intended to limit the scope of the present application. Any modification and improvement made by those skilled in the art to the technical solutions of the present application without departing from the design spirit of the present application shall fall within the protection scope of the present application as defined by the claims.

Claims

1. A non-crystalline alloy design method based on random forest and SHAP analysis, characterized by, The method comprises the following steps: obtaining an alloy to be designed, performing SHAP analysis on elements in the alloy to be designed, and screening out key elements; the key elements are a plurality of elements that have the greatest influence on the amorphous alloy forming ability; screening the key elements by using mutual information analysis, obtaining an element combination to be optimized, determining the optimal proportion and concentration range of each element in the element combination to be optimized according to a binary phase diagram and a SHAP dependence diagram, and obtaining an optimal element combination; determining the optimal proportion of each element in the element combination to be optimized comprises: obtaining the solubility, phase transition temperature and behavior of liquid alloy of each element in the element combination to be optimized according to the eutectic point of the binary phase diagram, and determining the optimal proportion of each element in the element combination to be optimized based on the solubility, phase transition temperature and behavior of liquid alloy of each element; determining the concentration range of each element in the element combination to be optimized comprises: obtaining the sensitivity of the concentration of each element in the element combination to be optimized to the amorphous alloy forming ability according to the SHAP dependence diagram, and obtaining the target relationship between the element concentration and the amorphous alloy forming ability in combination with the eutectic point; determining the concentration range of each element in the element combination to be optimized based on the target relationship between the element concentration and the amorphous alloy forming ability; inputting the optimal element combination into a random forest model to predict the amorphous alloy forming ability value; the random forest model is obtained by training a training set; wherein the training set comprises the element composition and proportion of the alloy and a target variable; obtaining the random forest model comprises: obtaining the element composition and proportion of the alloy and the target variable; the target variable is a target parameter of the amorphous alloy forming ability; taking the element composition and proportion of the alloy and the target variable as input data, preprocessing the input data, inputting the preprocessed input data into an original random forest model, optimizing the hyperparameters of the model, and balancing multiple performance indicators of the model by using an NSGA-II algorithm to obtain the random forest model.

2. The random forest and SHAP analysis based amorphous alloy design method of claim 1, wherein, screening the key elements comprises: performing SHAP analysis on the element composition and proportion, quantifying the contribution of each element to the amorphous alloy forming ability under different conditions, i.e. Shapley value; based on the Shapley value, sorting the elements in the alloy to be designed, setting a screening threshold, and further screening out the key elements.

3. The random forest and SHAP analysis based amorphous alloy design method of claim 1, wherein, obtaining the element combination to be optimized comprises: evaluating the mutual dependence between each element in the key elements by using mutual information analysis, eliminating element redundancy of the key elements according to the mutual dependence, and obtaining an element combination to be optimized; the element combination to be optimized is an element combination that has the greatest influence on the amorphous alloy forming ability.

4. The random forest and SHAP analysis based amorphous alloy design method of claim 1, wherein, the preprocessing of the input data comprises: scaling the content of each element in the input data to a target interval by using data cleaning and standardization methods.

5. The random forest and SHAP analysis based amorphous alloy design method of claim 1, wherein, the hyperparameters of the model include the number of trees, the maximum depth, and the minimum number of samples.

Images