Forward design method of high-yield-strength stainless steel based on interpretable model

Abstract

The invention relates to the technical field of stainless steel material science, in particular to a high-yield-strength stainless steel forward design method based on an interpretable model, which comprises the following steps: constructing a virtual sample according to the composition and content of stainless steel; the virtual sample is input into a preset stainless steel yield strength prediction model, a yield strength prediction value of the virtual sample is output, the stainless steel yield strength prediction model is obtained based on training of a training set, and the training set comprises yield strength experiment values of a plurality of stainless steel samples; the stainless steel yield strength prediction model is constructed by combining an XGBoost model with an SHAP method; and a virtual sample with preset yield strength is screened through the yield strength predicted value, and forward design of the high-yield-strength stainless steel is completed. The stainless steel with high yield strength is obtained through high-throughput screening, the process is achieved through the computer technology, resource consumption caused by chemical experiments is avoided, and the time cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of stainless steel material science and technology, and in particular to a forward design method for high-yield strength stainless steel based on an interpretable model. Background Art

[0002] Stainless steel, the full name of which is stainless acid-resistant steel, is mainly composed of iron elements. It has excellent properties such as high strength, high toughness, high temperature resistance, ductility, corrosion resistance, and weldability. It is widely used in architectural decoration, oil and gas, automotive industry, transportation and other fields. Yield strength is a key indicator for evaluating the mechanical properties of stainless steel. It is defined as the maximum stress value that the material withstands when it begins to produce plastic deformation. This indicator directly determines the load-bearing capacity and deformation performance of stainless steel under service conditions. In application fields such as petrochemicals and architectural decoration, the demand for high-yield strength stainless steel is particularly urgent. Given the multi-element alloying characteristics of the stainless steel material system, researchers usually optimize or improve the yield strength of stainless steel by regulating the element ratio. However, this traditional trial-and-error R&D model is not only costly, but also has a long development cycle. Therefore, it is of great significance to use machine learning technology to study and design stainless steel with better performance based on experimental data.

[0003] At present, (1) the training of machine learning models requires a large amount of data, but in stainless steel research, the acquisition of experimental data is often time-consuming and costly, resulting in a limited amount of data available for training, which affects the accuracy and generalization ability of the model. In addition, differences in experimental conditions, measurement methods, and other factors may cause noise and bias in the data, which in turn affects the performance of the model. (2) Many machine learning models are trained based on specific data sets and experimental conditions. Their scope of application is often limited to the scenario at the time of training and is difficult to generalize to other types of stainless steel or different processing technologies. When encountering new alloy compositions, process parameters, or service environments, the model may not be able to accurately predict its performance and needs to be retrained or adjusted. (3) Some machine learning models only establish the relationship between input parameters and output performance based on statistical correlation, ignoring the microstructural characteristics and physical metallurgical principles of the material, which may reduce the reliability of the model prediction and make it difficult to conduct in-depth physical mechanism analysis. Due to the lack of support from physical metallurgical principles, the prediction results of the machine learning model are often difficult to interpret and cannot provide clear physical guidance for material design. Therefore, the present invention proposes a forward design method for high yield strength stainless steel based on an interpretable model. Summary of the Invention

[0004] The purpose of the present invention is to provide a forward design method for high yield strength stainless steel based on an interpretable model, obtain stainless steel with high yield strength through high-throughput screening, and use computer technology to implement the process, avoiding the resource consumption caused by chemical experiments and reducing time costs.

[0005] To achieve the above objectives, the present invention provides, on the one hand, a forward design method for high yield strength stainless steel based on an interpretable model, comprising:

[0006] Construct virtual samples based on the composition and content of stainless steel;

[0007] Inputting the virtual sample into a preset stainless steel yield strength prediction model, and outputting a yield strength prediction value of the virtual sample, wherein the stainless steel yield strength prediction model is obtained based on a training set, the training set includes yield strength experimental values ​​of several stainless steel samples, and the stainless steel yield strength prediction model is constructed by combining an XGBoost model with a SHAP method;

[0008] The virtual samples with preset yield strength are screened by the yield strength prediction value to complete the forward design of high yield strength stainless steel.

[0009] Optionally, constructing the training set includes:

[0010] Collect chemical formula and yield strength experimental values ​​of several stainless steel samples;

[0011] Using element descriptors to obtain characteristic values ​​corresponding to each stainless steel sample, and performing screening processing on the characteristic values ​​to obtain processed characteristic values;

[0012] The training set is constructed by combining the yield strength experimental value of the stainless steel sample and the corresponding processed characteristic value.

[0013] Optionally, the element descriptor includes a basic descriptor, other class descriptors, and a microstructure feature descriptor.

[0014] Optionally, screening out the characteristic value includes:

[0015] After deleting the constant features in the eigenvalues, redundant features are removed based on feature correlation, wherein the feature correlation adopts a preset Pearson correlation coefficient.

[0016] Optionally, constructing the stainless steel yield strength prediction model by combining the XGBoost model with the SHAP method includes:

[0017] The yield strength experimental value in the training set is used as the target variable, the processed eigenvalue is used as the independent variable, and the XGBoost model is input for training to construct a preliminary stainless steel yield strength prediction model;

[0018] The SHAP value of the processed eigenvalue is calculated using SHAP, and the SHAP value is combined with the preliminary stainless steel yield strength prediction model to perform feature screening to obtain the best feature subset;

[0019] A final stainless steel yield strength prediction model is constructed based on the optimal feature subset and the preliminary stainless steel yield strength prediction model.

[0020] Optionally, the optimal feature subset includes: work function, Zunger atomic radius, vaporization enthalpy, Pietkov dicovalent radius, Slater effective nuclear charge, Pauling electronegativity, first ionization energy, electron cloud density, melting point, boiling point, rigidity modulus, ionic radius, second ionization energy, and Mulliken electronegativity.

[0021] In another aspect, the present invention further provides a forward design system for high yield strength stainless steel based on an interpretable model, comprising:

[0022] A virtual sample construction unit, used for constructing a virtual sample according to the composition and content of stainless steel;

[0023] a yield strength prediction unit, configured to input the virtual sample into a preset stainless steel yield strength prediction model and output a predicted yield strength value of the virtual sample, wherein the stainless steel yield strength prediction model is obtained based on a training set, the training set includes experimental yield strength values ​​of several stainless steel samples, and the stainless steel yield strength prediction model is constructed by combining an XGBoost model with a SHAP method;

[0024] The virtual sample screening unit is used to screen virtual samples with preset yield strength according to the yield strength prediction value to complete the forward design of high yield strength stainless steel.

[0025] On the other hand, the present invention also provides a computer-readable storage medium, which stores a computer program, wherein when the computer program is executed by a processor, the steps of the forward design method for high yield strength stainless steel based on an interpretable model are implemented.

[0026] On the other hand, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the forward design method for high yield strength stainless steel based on an interpretable model are implemented.

[0027] The beneficial effects of the present invention are:

[0028] The present invention can simply and quickly predict the yield strength of stainless steel, generate an optimal feature subset for stainless steel, import the obtained feature data into a prediction model, and obtain a prediction result in seconds.

[0029] The present invention can find stainless steel with high yield strength by setting the element range of stainless steel, designing virtual samples, and predicting through models.

[0030] The method of the present invention is realized entirely by computer, does not involve experiments and chemicals, is low-cost, and complies with the concept of green environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 This is a flow chart of a forward design method for high yield strength stainless steel based on an interpretable model according to an embodiment of the present invention;

[0033] Figure 2 Schematic diagram of preliminary prediction results of four methods: RFR, SVR, KNN, and XGBoost according to an embodiment of the present invention;

[0034] Figure 3 This is a comparison chart of the leave-one-out cross-validation experimental value and the predicted value of the XGBoost model of YS in an embodiment of the present invention;

[0035] Figure 4 This is a comparison chart of the test set experimental value and the predicted value of the XGBoost model of YS in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] On the one hand, this embodiment provides a forward design method for high yield strength stainless steel based on an interpretable model, such as Figure 1 As shown, including:

[0039] Construct virtual samples based on the composition and content of stainless steel;

[0040] Inputting the virtual sample into a preset stainless steel yield strength prediction model, and outputting a yield strength prediction value of the virtual sample, wherein the stainless steel yield strength prediction model is obtained based on a training set, the training set includes yield strength experimental values ​​of several stainless steel samples, and the stainless steel yield strength prediction model is constructed by combining an XGBoost model with a SHAP method;

[0041] The virtual samples with preset yield strength are screened by the yield strength prediction value to complete the forward design of high yield strength stainless steel.

[0042] Specifically, this embodiment can simply and quickly predict the yield strength of stainless steel, generate an optimal feature subset for stainless steel, import the obtained feature data into a prediction model, and obtain a prediction result in seconds; by setting the element range of stainless steel, designing a virtual sample, and predicting through the model, stainless steel with high yield strength can be found; this method is implemented entirely by computer, does not involve experiments and chemicals, is low-cost, and conforms to the concept of green environmental protection.

[0043] Furthermore, constructing the training set includes:

[0044] Collect chemical formula and yield strength experimental values ​​of several stainless steel samples;

[0045] Using element descriptors to obtain characteristic values ​​corresponding to each stainless steel sample, and performing screening processing on the characteristic values ​​to obtain processed characteristic values, wherein the element descriptors include basic descriptors, other class descriptors, and microstructure characteristic descriptors;

[0046] The training set is constructed by combining the yield strength experimental value of the stainless steel sample and the corresponding processed characteristic value.

[0047] The step of screening out the characteristic values ​​includes:

[0048] After deleting the constant features in the eigenvalues, redundant features are removed based on feature correlation, wherein the feature correlation adopts a preset Pearson correlation coefficient.

[0049] Furthermore, the stainless steel yield strength prediction model is constructed by combining the XGBoost model with the SHAP method, including:

[0050] The yield strength experimental value in the training set is used as the target variable, the processed eigenvalue is used as the independent variable, and the XGBoost model is input for training to construct a preliminary stainless steel yield strength prediction model;

[0051] The SHAP value of the processed eigenvalue is calculated using SHAP, and the SHAP value is combined with the preliminary stainless steel yield strength prediction model to perform feature screening to obtain the best feature subset;

[0052] A final stainless steel yield strength prediction model is constructed based on the optimal feature subset and the preliminary stainless steel yield strength prediction model.

[0053] The optimal feature subset includes: work function, Zunger atomic radius, vaporization enthalpy, Pietaux dicovalent radius, Slater effective nuclear charge, Pauling electronegativity, first ionization energy, electron cloud density, melting point, boiling point, rigidity modulus, ionic radius, second ionization energy, and Mulliken electronegativity.

[0054] The following is a detailed description of the model construction and application of this embodiment, including the following contents:

[0055] (1) The chemical formula of stainless steel and its experimental values ​​of yield strength (YS) were collected from literature and databases as data set samples.

[0056] Specifically, the chemical formula of stainless steel and 433 yield strength data were collected from literature and databases as shown in Table 1.

[0057] Table 1

[0058]

[0059] (2) The method of using element descriptors to expand the information richness of alloy materials is divided into three categories according to the element descriptor type, including 34 basic descriptors, 66 other descriptors, and finally the microstructural characteristics, a total of 101 descriptors. These include relative atomic mass, atomic number, vaporization enthalpy, ionization energy, etc.

[0060] Specifically, the element descriptors were used to obtain the characteristic values ​​corresponding to each stainless steel sample, some of which and their meanings are shown in Table 2.

[0061] Table 2

[0062] Serial number feature significance 1 atomic number atomic number 2 quantum number quantum ordinal number 3 atomic weight Relative atomic mass 4 group number Atomic family number 5 melting point Melting point 6 boiling point boiling point 7 enthalpy vaporization Enthalpy of vaporization 8 enthalpy melting Melting enthalpy 9 first ionization First ionization energy 10 density density

[0063] (3) The dataset is randomly divided into training set and test set in a ratio of 4:1.

[0064] Specifically, in this embodiment, the data set is randomly divided into a training set and a test set in a ratio of 4:1, and the number of samples in the yield strength training set and the test set are 346 and 87, respectively.

[0065] (4) After deleting the constant features, redundant features are removed based on feature correlation to quickly reduce the dimension of the features.

[0066] Specifically, after deleting the constant features, this embodiment uses the Pearson correlation coefficient to remove redundant features to reduce the feature dimension. The Pearson correlation coefficient is set at 0.95. In the yield strength dataset, 50 features remain.

[0067] (5) The yield strength experimental values ​​of stainless steel collected in step (1) are used as target variables, and the features screened in step (4) are used as independent variables. Based on the training set divided in step (3), a preliminary prediction model of the yield strength of stainless steel is established using RFR, SVR, KNN, and XGBoost.

[0068] Specifically, the yield strength experimental value of stainless steel collected in step (1) is used as the target variable, the features in step (4) are used as the independent variable, and based on the training set divided in step (3), SVR, XGBoost, KNN, and RFR are used to establish a preliminary prediction model for the yield strength performance of stainless steel. The results of the preliminary prediction model are as follows: Figure 2 shown.

[0069] (6) Based on the prediction model of stainless steel established in step (5), SHAP value of each feature is calculated using SHAP and the preliminary prediction model with the best result is packaged for feature screening to obtain the best feature subset.

[0070] Specifically, according to the prediction model of the yield strength performance of stainless steel established in step (5), XGBoost with the best preliminary prediction model result is selected, and the best feature subset is obtained by combining the SHAP method. The best features are shown in Table 3, totaling 14 features.

[0071] Table 3

[0072] Serial number Yield Strength Optimum Characteristics significance 1 work function Work function 2 radii pseudo zunger Zunger atomic radius 3 enthalpy vaporization Enthalpy of vaporization 4 covalent radius pyykko double Piéko dicovalent radius 5 nulear charge effective slater Slater effective nuclear charge 6 en pauling Pauling electronegativity 7 first ionization First ionization energy 8 <![CDATA[n ws 1 / 3 miedema]]> Electron cloud density 9 melting point Melting point 10 boiling point boiling point 11 modulus rigidity Modulus of rigidity 12 ionic radius Ionic radius 13 second ionization Second ionization energy 14 en mulliken Mulliken electronegativity

[0073] (7) Based on the best feature subset selected in step (6) and the best XGBoost prediction model, quickly predict the performance of the test set samples in step (3).

[0074] Specifically, based on the optimal feature subset and the optimal model XGBoost selected in step (6), a stainless steel yield strength performance prediction model is established to quickly predict the yield strength of the test set samples in step (3).

[0075] (8) Generate virtual samples, use the model to predict performance, and find stainless steel with excellent target performance.

[0076] Specifically, the XGBoost model is applied to the generated virtual samples, and the model is used to make predictions and screen stainless steel with high yield strength.

[0077] Example 1:

[0078] In this example, the leave-one-out cross-validation results of the stainless steel yield strength model established with XGBoost using SHAP are as follows: Figure 3As shown in the figure, the coefficient of determination between the experimental value and the predicted value is 0.81, and the mean absolute error is 57.74.

[0079] Example 2:

[0080] In this embodiment, the stainless steel yield strength prediction model established by combining SHAP's XGBoost is used to predict the test sample. Figure 4 As shown in the figure, the coefficient of determination between the experimental value and the predicted value is 0.85, and the mean absolute error is 45.03.

[0081] Example 3:

[0082] In this example, based on the composition and content of stainless steel in the collected dataset, high-throughput virtual samples were generated using WAE. These samples contained 24 elements: Fe, Mn, Cr, Ni, Si, C, S, P, N, Mo, Se, Nb, V, Cu, W, Co, Ti, Al, Ta, Ce, B, La, and Sn. The XGBoost model was used to predict the performance of the virtual samples, and the prediction results for stainless steel samples with excellent performance are shown in Table 4.

[0083] Table 4

[0084]

[0085] On the other hand, this embodiment also provides a forward design system for high yield strength stainless steel based on an interpretable model, including:

[0086] A virtual sample construction unit, used for constructing a virtual sample according to the composition and content of stainless steel;

[0087] a yield strength prediction unit, configured to input the virtual sample into a preset stainless steel yield strength prediction model and output a predicted yield strength value of the virtual sample, wherein the stainless steel yield strength prediction model is obtained based on a training set, the training set includes experimental yield strength values ​​of several stainless steel samples, and the stainless steel yield strength prediction model is constructed by combining an XGBoost model with a SHAP method;

[0088] The virtual sample screening unit is used to screen virtual samples with preset yield strength according to the yield strength prediction value to complete the forward design of high yield strength stainless steel.

[0089] On the other hand, this embodiment also provides a computer-readable storage medium, which stores a computer program, wherein when the computer program is executed by a processor, the steps of the forward design method for high yield strength stainless steel based on an interpretable model are implemented.

[0090] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the forward design method for high yield strength stainless steel based on an interpretable model are implemented.

[0091] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A forward design method for high yield strength stainless steel based on an interpretable model, characterized in that: include: Construct virtual samples based on the composition and content of stainless steel; Inputting the virtual sample into a preset stainless steel yield strength prediction model, and outputting a yield strength prediction value of the virtual sample, wherein the stainless steel yield strength prediction model is obtained based on a training set, the training set includes yield strength experimental values ​​of several stainless steel samples, and the stainless steel yield strength prediction model is constructed by combining an XGBoost model with a SHAP method; The virtual samples with preset yield strength are screened by the yield strength prediction value to complete the forward design of high yield strength stainless steel.

2. The forward design method for high yield strength stainless steel based on an interpretable model according to claim 1, characterized in that: Constructing the training set includes: Collect chemical formula and yield strength experimental values ​​of several stainless steel samples; Using element descriptors to obtain characteristic values ​​corresponding to each stainless steel sample, and performing screening processing on the characteristic values ​​to obtain processed characteristic values; The training set is constructed by combining the yield strength experimental value of the stainless steel sample and the corresponding processed characteristic value.

3. The forward design method for high yield strength stainless steel based on an interpretable model according to claim 2, characterized in that: The element descriptors include basic descriptors, other class descriptors, and microstructure feature descriptors.

4. The forward design method for high yield strength stainless steel based on an interpretable model according to claim 2, characterized in that: Screening out the characteristic value includes: After deleting the constant features in the eigenvalues, redundant features are removed based on feature correlation, wherein the feature correlation adopts a preset Pearson correlation coefficient.

5. The forward design method for high yield strength stainless steel based on an interpretable model according to claim 1, characterized in that: The stainless steel yield strength prediction model constructed by combining the XGBoost model with the SHAP method includes: The yield strength experimental value in the training set is used as the target variable, the processed eigenvalue is used as the independent variable, and the XGBoost model is input for training to construct a preliminary stainless steel yield strength prediction model; The SHAP value of the processed eigenvalue is calculated using SHAP, and the SHAP value is combined with the preliminary stainless steel yield strength prediction model to perform feature screening to obtain the best feature subset; A final stainless steel yield strength prediction model is constructed based on the optimal feature subset and the preliminary stainless steel yield strength prediction model.

6. The forward design method for high yield strength stainless steel based on an interpretable model according to claim 5, characterized in that: The optimal feature subset includes: work function, Zunger atomic radius, vaporization enthalpy, Pietkov dicovalent radius, Slater effective nuclear charge, Pauling electronegativity, first ionization energy, electron cloud density, melting point, boiling point, rigidity modulus, ionic radius, second ionization energy, and Mulliken electronegativity.

7. A forward design system for high yield strength stainless steel based on an interpretable model, characterized in that: include: A virtual sample construction unit, used for constructing a virtual sample according to the composition and content of stainless steel; a yield strength prediction unit, configured to input the virtual sample into a preset stainless steel yield strength prediction model and output a predicted yield strength value of the virtual sample, wherein the stainless steel yield strength prediction model is obtained based on a training set, the training set includes experimental yield strength values ​​of several stainless steel samples, and the stainless steel yield strength prediction model is constructed by combining an XGBoost model with a SHAP method; The virtual sample screening unit is used to screen virtual samples with preset yield strength according to the yield strength prediction value to complete the forward design of high yield strength stainless steel.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the forward design method for high yield strength stainless steel based on an interpretable model according to any one of claims 1 to 6 are implemented.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the forward design method for high yield strength stainless steel based on an interpretable model according to any one of claims 1 to 6 are implemented.

Images