Tissue prediction model suitable for alloy steel as well as establishment method and application of tissue prediction model

Abstract

The invention provides a structure prediction model suitable for alloy steel and an establishment method and application thereof, and the establishment method comprises the following steps: S1, obtaining component and structure type data of a plurality of steel grades covered by a component range, and calculating the statistical characteristics of each element by taking the mass fraction of the alloy element as a characteristic variable; s2, data processing is conducted, a decision boundary is obtained, and structure types are divided into single-phase austenite, austenite and ferrite or martensite, ferrite and / or martensite; and S3, predicting the model: performing dimensionality reduction on the obtained data by adopting the linear discriminant analysis in the step S2, combining softmax regression, performing linear discriminant analysis dimensionality reduction to obtain two LDA expressions, and obtaining a shaffler-like tissue prediction map according to the decision boundary straight lines of the corresponding three tissue types. Most material organizations can be correctly classified through the method, the accuracy is high, and the method is convenient to popularize and use in new materials and design components.

Description

Technical Field

[0001] This invention relates to the field of microstructure prediction technology for metallic materials, and more specifically, to a microstructure prediction model applicable to alloy steel, its establishment method, and its application. Background Technology

[0002] Alloy steels typically include medium-manganese steel, high-manganese steel, lightweight steel, stainless steel, and nitrogen-containing stainless steel. For example, high-manganese steel and Fe-Mn-Al-C low-density steels have received widespread attention for weight reduction in automotive and shipbuilding structures due to their high strength, high plasticity, and low density. High-nitrogen stainless steel, by substituting nitrogen for nickel, significantly improves its strength and toughness while maintaining its single austenitic structure, and its lower nickel content significantly increases its biocompatibility. Therefore, it is increasingly used and valued in nuclear power and biomedicine. To ensure nitrogen solubility, its alloy composition also contains a relatively high amount of manganese.

[0003] The aforementioned new steel materials all contain a significant mass fraction of manganese. Their microstructure typically comprises ferrite, martensite, austenite, or a mixture thereof, and these microstructures significantly affect their mechanical properties, corrosion resistance, magnetic properties, and other application properties. To obtain the corresponding mechanical and application properties, the composition needs to be designed to achieve the appropriate microstructure type. Generally, the relationship between composition and microstructure is established for Cr-Ni stainless steels using Schaeffler diagrams, WRC-2000 diagrams, and Delong diagrams, with the Schaeffler diagram being the most widely used. The Schaeffler diagram is a common method for predicting the microstructure type of stainless steel based on chemical composition. Compared to equilibrium phase diagrams established using the CALPHAD method, the Schaeffler diagram uses actual experimental data to establish the relationship between microstructure type and chemical composition, resulting in more reliable predictions and the ability to predict non-equilibrium microstructure types. However, the accuracy of the Schaeffler diagram decreases significantly as the alloy composition deviates from that of Cr-Ni stainless steel. Kleuh and Lee et al. established a Scheffler-like diagram applicable to medium- and high-manganese steels and Fe-Mn-Al-C low-density steels by adjusting the weighting coefficients of each component in the Cr and Ni equivalents in the Scheffler diagram and performing operations such as translation and rotation on the decision boundary. However, it is not applicable to high-nitrogen stainless steel. The WRC-2000 diagram and Delong diagram are also not suitable for predicting the microstructure of medium-manganese steel, high-manganese steel, lightweight steel, and nitrogen-containing stainless steel. Summary of the Invention

[0004] In view of this, the present invention aims to propose a microstructure prediction model and its establishment method applicable to alloy steel, so as to solve the problems that existing technologies such as Shaeffler diagrams, WRC-2000 diagrams, and Delong diagrams cannot be uniformly applied to the microstructure prediction of alloy steels such as medium manganese steel, high manganese steel, light steel, and nitrogen-containing stainless steel, as well as the problem of low accuracy in microstructure prediction.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0006] In a first aspect, the present invention proposes a method for establishing a microstructure prediction model applicable to alloy steel, comprising the following steps:

[0007] S1. Obtain composition and microstructure data covering multiple steel grades, and calculate the statistical characteristics of each element using the mass fraction of alloying elements as a characteristic variable.

[0008] S2. Data processing to obtain decision boundaries and classify microstructure types into single-phase austenite, austenite + ferrite or martensite, ferrite and / or martensite.

[0009] S3. Prediction Model:

[0010] The obtained data are subjected to dimensionality reduction using linear discriminant analysis in step S2 and combined with softmax regression. After dimensionality reduction by linear discriminant analysis, two LDA expressions are obtained. Based on the decision boundary lines of the corresponding three organizational types, the Shaeffler-like organizational prediction map is obtained.

[0011] Furthermore, step S2 specifically includes the following:

[0012] S21. Standardize all samples containing d components to standardize the d-dimensional data;

[0013] S22. Use the LDA algorithm to obtain the transformation matrix W from the standardized data. d×2 ;

[0014] S23. Multiply the standardized standard deviation by the transformation matrix W. d×2 This is used to reduce d-dimensional data to two dimensions;

[0015] S24. Use the multivariate softmax classification method for classification and obtain the decision boundary.

[0016] Furthermore, we first calculate the score s of sample x belonging to category k. k (x), and then the softmax function is used to calculate the probability that sample x belongs to a certain class. Then, cross-entropy is used as the loss function, and the final parameter matrix is ​​solved using gradient descent. .

[0017] Furthermore, the score s is calculated using equation (1). k (x), calculate the probability using equation (2). The parameter matrix The form is shown in equation (3).

[0018] .

[0019] Furthermore, in step S3, the two LDA expressions obtained after the dimensionality reduction by the linear discriminant analysis are shown in equations (4) and (5).

[0020] .

[0021] Furthermore, in step S3, the decision boundary lines for the three corresponding organizational types are shown in equations (6) to (8).

[0022] .

[0023] Furthermore, in step S1, the composition and microstructure data of many steel types, including medium manganese steel, high manganese steel, light steel, stainless steel, and nitrogen-containing stainless steel, are obtained as samples to ensure that the corresponding microstructures cover fully austenitic, austenitic + ferrite / martensite, ferrite, martensite, and ferrite + martensite microstructures.

[0024] Secondly, this invention proposes a microstructure prediction model applicable to alloy steel, which is obtained using the aforementioned method for establishing a microstructure prediction model.

[0025] Thirdly, the present invention also proposes an application of a microstructure prediction model suitable for alloy steel in microstructure prediction, using the above-mentioned method for establishing the microstructure prediction model and / or the microstructure prediction model.

[0026] Furthermore, the main components of C, Si, Mn, Cr, Ni, Mo, Al, and Cu in the new material or design composition are marked with corresponding points in a Shaeffler-like microstructure prediction map according to two LDA expressions. The microstructure type can be intuitively determined based on the region of the new component in the Shaeffler-like microstructure prediction map.

[0027] Compared with existing technologies, the microstructure prediction model for alloy steel, its establishment method, and its application described in this invention have the following advantages:

[0028] (1) The microstructure of most materials can be correctly classified by the microstructure prediction model and its establishment method of the present invention, and the accuracy is significantly improved compared with the traditional Cr / Ni equivalent method and the improved Cr / Ni equivalent method, which is convenient for promotion and use in new materials and design composition.

[0029] (2) This invention is applicable to the microstructure prediction of alloy steels such as medium manganese steel, high manganese steel, light steel, and nitrogen-containing stainless steel.

[0030] (3) By simply marking the position of the corresponding point in the Shaeffler-like tissue prediction map, the tissue type can be intuitively determined based on the region of the new component in the Shaeffler-like tissue prediction map. Attached Figure Description

[0031] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0032] Figure 1 This is a flowchart illustrating the data dimensionality reduction and classification using the LDA+softmax method of the present invention.

[0033] Figure 2 This invention establishes a Shaeffler-like tissue prediction map.

[0034] Figure 3 This is a comparative diagram of the model described in this invention, the Cr / Ni equivalent method, and the improved Cr / Ni equivalent method (Lee);

[0035] Figure 4 The tissue type predicted in Embodiment 1 of the present invention;

[0036] Figure 5 This is a metallographic image of the tissue in Embodiment 1 of the present invention. Detailed Implementation

[0037] The present invention will be further described below with reference to specific embodiments. First, it should be noted that the data in the following experimental examples were obtained by the inventors through numerous experiments. Due to space limitations, only a portion of these data is shown in the specification, and those skilled in the art can understand and implement the present invention based on this data. These embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the contents of this invention, those skilled in the art can make various modifications or alterations to the invention, and these modifications or alterations also fall within the scope of protection of this application.

[0038] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0039] The present invention provides a method for establishing a microstructure prediction model applicable to alloy steel, comprising the following steps:

[0040] S1. Obtain composition and microstructure data covering multiple steel grades, and calculate the statistical characteristics of each element using the mass fraction of alloying elements as a characteristic variable.

[0041] Specifically, in step S1, based on publicly available literature and experimental results, data on the composition and microstructure types of numerous steel grades, including medium-manganese steel, high-manganese steel, light steel, stainless steel, and nitrogen-containing stainless steel, are obtained as samples. This ensures that the corresponding microstructures cover fully austenitic, austenitic + ferrite / martensite, ferrite, martensite, and ferrite + martensite microstructures. The statistical characteristics of each element include its maximum, minimum, and average values.

[0042] The data sources for this invention total 219, including 52 data points for conventional Cr-Ni-Mo stainless steel, 117 data points for nitrogen-containing stainless steel, and 50 data points for Fe-Mn-C-(Al) series medium and high manganese steel. The statistical characteristics of each element are shown in Table 1.

[0043] Table 1

[0044]

[0045] S2. Data processing to obtain decision boundaries and classify the microstructure into single-phase austenite (denoted as A), austenite + ferrite or martensite (denoted as A+F / M), and ferrite and / or martensite (denoted as F / M).

[0046] Specifically, such as Figure 1 As shown, step S2 includes the following process:

[0047] S21. Standardize all samples containing d components to standardize the d-dimensional data;

[0048] S22. Use the LDA algorithm to obtain the transformation matrix W from the standardized data. d×2 ;

[0049] S23. Multiply the standardized standard deviation by the transformation matrix W. d×2 This is used to reduce d-dimensional data to two dimensions;

[0050] S24. Use the multivariate softmax classification method for classification and obtain the decision boundary.

[0051] In step S21, the d components refer to each element in Table 1. The standardization process refers to processing the data in Table 1 according to the following formula: x j '=(x j -μ j ) / σ j , where x j' refers to the standardized value of the content of the j-th element, x j This refers to the content of the j-th element, μ. j It refers to the average content of the j-th element, σ j It refers to the variance of the content of the j-th element.

[0052] Fisher linear discriminant analysis (LDA) is commonly used to reduce the dimensionality and classify high-dimensional data containing class information, and it has been widely applied in biomedicine, statistics, and computer science. LDA and principal component analysis (PCA) are similar in their construction methods, the main difference being that LDA is a supervised learning method that includes class information, while PCA is an unsupervised learning method for data dimensionality reduction. LDA assumes that each class of data follows a multivariate normal distribution, that the covariance matrices of each class are identical, and that each sample is independent. This method aims to minimize the within-group variance of samples within the same class and maximize the between-group variance of samples from different classes, thereby achieving the goal of data dimensionality reduction. Therefore, this invention employs the LDA algorithm.

[0053] In step S22, given that the transformation matrix W is obtained using the LDA algorithm... d×2 This is a standard technical method and will not be elaborated upon here. Transformation matrix W d×2 The values ​​correspond to the coefficients in the two LDAs. This matrix can be solved using standard machine learning algorithms, directly obtained by calling the Linear Discriminant Analysis API.

[0054] In step S24, softmax regression is a multi-class generalization of logistic regression, and its decision boundary is a linear boundary, which can calculate the probability value of each sample belonging to a certain class. First, the score s of sample x belonging to class k is calculated by equation (1). k (x), and then the probability that sample x belongs to a certain class is calculated using the softmax function through equation (2). Then, cross-entropy is used as the loss function, and the final parameter matrix is ​​solved using gradient descent. The parameter matrix The form is shown in equation (3):

[0055]

[0056] Since this application classifies organization types into 3 categories, k=3.

[0057] LDA is used to reduce the original nine components of each material to two dimensions. Therefore, in equation (1), x is a vector with two elements. x is two LDAs transformed by the LDA algorithm, distributed as LDA1 and LDA2. = (k0, k1, k2), where k0 is a constant term, k1 is the coefficient of one element after dimensionality reduction, and k2 is the coefficient of the other element after dimensionality reduction. The parameter matrix... The parameter matrix is ​​composed of the k vectors in the solved equation (2).

[0058] For example, x = (x1, x2) T So S k (x) = K0 + x1*K1 + x2*K2.

[0059] S3. Prediction Model:

[0060] The obtained data were subjected to dimensionality reduction using linear discriminant analysis in step S2, combined with softmax regression. The two LDA expressions obtained after dimensionality reduction by linear discriminant analysis are shown in equations (4) and (5). The decision boundary lines for the three organizational types are shown in equations (6) to (8). The resulting Shaeffler-like organizational prediction map is shown in [reference needed]. Figure 2 .

[0061] Combination Figure 2 It can be seen that linear discriminant analysis can reduce the dimensionality of various data points and aggregate them, resulting in a significant classification effect.

[0062]

[0063] Wherein, equation (6) corresponds to Figure 2 The straight line boundary at the lower left corner, corresponding to equation (7). Figure 2 The straight line boundary in the upper right corner, corresponding to equation (8) Figure 2 The straight line boundary in the lower right corner.

[0064] As a specific example of this application, the score for each class can be obtained by multiplying the parameter matrix calculated by the softmax regression algorithm with the corresponding LDA and bias term, as detailed in equation (9), where the matrix on the left is the parameter matrix containing the bias term. The right side represents the two LDAs after dimensionality reduction.

[0065] ,

[0066] Among them, S A (x) represents the score for belonging to austenite, S A (x) represents the score for belonging to austenite + ferrite or martensite, S A(x) represents the score for belonging to ferrite and / or martensite.

[0067] The microstructure prediction model of the present invention applicable to alloy steel is established using the above-described method.

[0068] This invention also proposes an application of a microstructure prediction model suitable for alloy steel in microstructure prediction, employing the aforementioned method for establishing and / or modeling the microstructure prediction system. Specifically, for microstructure prediction of new materials or designed compositions, the main components C, Si, Mn, Cr, Ni, Mo, Al, and Cu in the new material or designed composition are expressed using two LDA expressions. Figure 2 The location of corresponding points is marked in the Shaeffler-like tissue prediction map, and the tissue type can be intuitively determined based on the region of the new component in the Shaeffler-like tissue prediction map.

[0069] This invention obtains data from relevant literature and experiments. Based on this data, a linear discriminant method is used to compress the multivariate components to a binary component, reducing the d-dimensional data to two dimensions. Combined with softmax regression analysis, a highly reliable microstructure prediction model is established that can unify the microstructure prediction of medium manganese steel, high manganese steel, stainless steel, and nitrogen-containing stainless steel. The two LDA expressions after dimensionality reduction are shown in equations (4) and (5), and the decision boundary lines for the three microstructure types are shown in equations (6) to (8). The classification effect of the model is shown in equation (6). Figure 2 As can be seen, the tissue structure of most materials can be correctly classified using the tissue prediction model and its establishment method of this invention, and this has been verified, for example... Figure 3 As shown, the accuracy reaches 93%, which is a significant improvement over both the traditional Cr / Ni equivalent method and the improved Cr / Ni equivalent method, making it convenient for widespread use in the design of new materials and compositions. This invention is applicable to the microstructure prediction of alloy steels such as medium manganese steel, high manganese steel, lightweight steel, and nitrogen-containing stainless steel.

[0070] Example 1

[0071] Objective: To obtain the chemical composition of materials for which tissue type needs to be predicted.

[0072] The composition of a newly developed FeMnAl-C type lightweight steel ingot was determined as follows: C: 0.96%, Si: 0.137%, Mn: 24.16%, Al: 9.93%.

[0073] Calculate the values ​​of its two LDA values ​​according to equations (4) and (5) respectively.

[0074]

[0075] Tissue type prediction based on LDA values: The coordinate values ​​formed by the calculated LDA1=-1.667 and LDA2=-2.618 are plotted in... Figure 2 The location is marked, and the marked point is... Figure 4 The yellow pentagram dots indicate that the material in Example 1 has an A+F / M structure, which is a dual-phase structure of austenite + ferrite / martensite.

[0076] After polishing the metal samples of the newly developed FeMnAl-C type lightweight steel ingots, the sample surface was etched with a 5% nitric acid aqueous solution. The actual microstructure was observed using a metallographic microscope. Figure 5 As shown, the actual microstructure consists of a dendritic proeutectoid ferrite and austenite matrix, consistent with the prediction results obtained using the prediction model of this invention.

[0077] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can make various alterations and modifications without departing from the spirit and scope of the invention; therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A method for establishing a microstructure prediction model applicable to alloy steel, characterized in that, Includes the following steps: S1. Obtain composition and microstructure data covering multiple steel grades, and calculate the statistical characteristics of each element using the mass fraction of alloying elements as a characteristic variable. S2. Data processing to obtain decision boundaries and classify microstructure types into single-phase austenite, austenite + ferrite or martensite, ferrite and / or martensite. S3. Prediction Model: The obtained data are subjected to dimensionality reduction using linear discriminant analysis in step S2 and combined with softmax regression. After dimensionality reduction by linear discriminant analysis, two LDA expressions are obtained. Based on the decision boundary lines of the corresponding three organizational types, the Shaeffler-like organizational prediction map is obtained.

2. The method for establishing an organizational prediction model according to claim 1, characterized in that, Step S2 specifically includes the following: S21. Standardize all samples containing d components to standardize the d-dimensional data; S22. Use the LDA algorithm to obtain the transformation matrix W from the standardized data. d×2 ; S23. Multiply the standardized standard deviation by the transformation matrix W. d×2 This is used to reduce d-dimensional data to two dimensions; S24. Use the multivariate softmax classification method for classification and obtain the decision boundary.

3. The method for establishing an organization prediction model according to claim 2, characterized in that, First, calculate the score s of sample x belonging to category k. k (x), and then the softmax function is used to calculate the probability that sample x belongs to a certain class. Then, cross-entropy is used as the loss function, and the final parameter matrix is ​​solved using gradient descent. .

4. The method for establishing an organizational prediction model according to claim 1, characterized in that, The score s is calculated using equation (1). k (x), calculate the probability using equation (2). The parameter matrix The form is shown in equation (3). 。 5. The method for establishing an organizational prediction model according to claim 1, characterized in that, In step S3, the two LDA expressions obtained after dimensionality reduction by the linear discriminant analysis are shown in equations (4) and (5). 。 6. The method for establishing an organizational prediction model according to claim 1, characterized in that, In step S3, the decision boundary lines for the three corresponding organizational types are shown in equations (6) to (8). 。 7. The method for establishing an organizational prediction model according to claim 1, characterized in that, In step S1, the composition and microstructure data of many steel types, including medium manganese steel, high manganese steel, light steel, stainless steel, and nitrogen-containing stainless steel, are obtained as samples to ensure that the corresponding microstructures cover fully austenitic, austenitic + ferrite / martensite, ferrite, martensite, and ferrite + martensite microstructures.

8. A microstructure prediction model applicable to alloy steel, characterized in that, The method for establishing the organization prediction model according to any one of claims 1 to 7 is used to obtain the model.

9. An application of a microstructure prediction model suitable for alloy steel in microstructure prediction, characterized in that, The method for establishing the organizational prediction model according to any one of claims 1 to 7 and / or the organizational prediction model according to claim 8 are adopted.

10. The application according to claim 9, characterized in that, The main components of C, Si, Mn, Cr, Ni, Mo, Al, and Cu in the new material or design composition are marked with corresponding points in a Shaeffler-like microstructure prediction map according to two LDA expressions. The microstructure type can be intuitively determined based on the region of the new component in the Shaeffler-like microstructure prediction map.

Images