Prediction method for hardenability of medium-low carbon alloy steel

Abstract

The invention discloses a method for predicting the hardenability of medium-low carbon alloy steel, and the method comprises the steps: collecting historical production data from an industrial production MES system, and constructing a high-dimensional industrial data set containing steel grade components, smelting parameters, rolling parameters and hardenability indexes; performing preliminary screening on the high-dimensional industrial data set based on prior knowledge of a smelting physical mechanism, and selecting features significantly related to a hardenability target value through a Pearson correlation coefficient analysis method; taking the selected characteristics obviously related to the hardenability target value as an input variable, taking the Jominy distance hardness as an output variable, and establishing an explicit hardenability mathematical model by adopting a symbol regression method; and inputting components and process parameters of the to-be-predicted medium-low carbon alloy steel into the hardenability mathematical model to realize prediction of the hardenability of the to-be-predicted medium-low carbon alloy steel. The method has guiding significance for accurately predicting the hardenability value of the medium-low carbon alloy steel.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of metal material heat treatment performance prediction, and in particular to a method for predicting hardenability of medium and low carbon alloy steel. BACKGROUND

[0002] Hardenability is one of the key performance indicators of medium and low carbon alloy steel in the heat treatment process, which directly affects the strength, toughness and service life of the material. Accurate prediction of hardenability has important engineering significance for optimizing alloy composition design, developing reasonable heat treatment process and improving product quality stability.

[0003] At present, the methods for predicting hardenability mainly include empirical formula method, numerical simulation method based on physics and machine learning method driven by data. Among them, the empirical formula method such as Grossmann method and Jominy curve empirical formula usually depends on the statistical relationship between the composition and hardenability in a specific alloy system, which is simple in form but limited in scope, and difficult to reflect the influence of complex smelting process and multi-element interaction. The numerical simulation method based on physics such as phase transition dynamics model and finite element thermal-mechanical coupling simulation has clear mechanism, but the model is complex, the calculation cost is high, and it depends on a large number of material parameters which are difficult to obtain accurately, so it is difficult to popularize and apply in actual production.

[0004] In recent years, with the development of industrial big data technology, the method for predicting hardenability based on machine learning has gradually attracted attention. This kind of method can learn the nonlinear relationship from historical data and has high prediction accuracy, but its model is usually a "black box" and lacks physical interpretability, which makes it difficult to guide the actual process optimization and composition adjustment. Especially in the actual production of medium and low carbon alloy steel, the hardenability is influenced by the interaction of multiple factors, and the existing methods often fail to fully integrate multi-source process data, resulting in limitations of the prediction results in actual application.

[0005] Therefore, how to accurately predict the hardenability of medium and low carbon alloy steel is a technical problem to be solved in the field. SUMMARY

[0006] In order to solve the above technical problems, the present application provides the following technical solutions: The embodiment of the present application provides a method for predicting hardenability of medium and low carbon alloy steel, comprising: Collecting historical production data from an industrial production MES system to construct a high-dimensional industrial data set containing steel composition, smelting parameters, rolling parameters and hardenability index; Selecting features significantly related to the target value of hardenability by Pearson correlation coefficient analysis method after preliminary screening of the high-dimensional industrial data set based on smelting physical mechanism prior knowledge; The selected features that are significantly related to the hardenability target value are used as input variables, and the Jominy distance hardness is used as the output variable. An explicit mathematical model of hardenability is established using the symbolic regression method. The composition and process parameters of the medium-low carbon alloy steel to be predicted are input into the hardenability mathematical model to achieve the prediction of the hardenability of the medium-low carbon alloy steel to be predicted.

[0007] In one possible implementation, the step of collecting historical production data from the industrial production MES system and constructing a high-dimensional industrial dataset containing steel composition, smelting parameters, rolling parameters, and hardenability indicators includes: The steel composition, smelting parameters, rolling parameters, and hardenability index of medium and low carbon alloy steel are collected from the industrial production MES system. The steel composition includes C, Si, Mn, Cr, Ni, Mo, V, Ti, and B. The smelting parameters include refining temperature, deoxidation time, N content, and O content. The rolling parameters include final rolling temperature, initial cooling temperature, and final cooling temperature. The collected steel composition, smelting parameters, rolling parameters, and hardenability indicators are fused to generate a high-dimensional industrial dataset.

[0008] In one possible implementation, after preliminary screening of the high-dimensional industrial dataset based on prior knowledge of smelting physical mechanisms, features significantly correlated with the hardenability target value are selected using Pearson correlation coefficient analysis, including: Based on prior knowledge of the physical mechanism of smelting, redundant variables that have no physical connection with hardenability are eliminated. Calculate the Pearson correlation coefficients between the remaining feature variables in the high-dimensional industrial dataset and the hardenability target value; The absolute value of the calculated correlation coefficient is compared with a set threshold, and the feature variable whose absolute value of the correlation coefficient is greater than the set threshold is selected.

[0009] In one possible implementation, the step of using selected features significantly correlated with the hardenability target value as input variables and Jominy distance hardness as the output variable, and establishing an explicit mathematical model of hardenability using a symbolic regression method, includes: The dataset containing significantly relevant features and hardenability target values ​​is divided into training set and test set according to a preset ratio. The HeuristicLab algorithm, based on a genetic algorithm, is used to automatically search for a mathematical expression that fits the relationship between input features and output hardness values ​​in the training set. During the search process, a Pareto front optimization strategy is adopted to optimize the prediction accuracy and complexity until a model that meets the preset accuracy requirements is obtained. The model is validated using the test set, and the final hardenability mathematical model is output.

[0010] In one possible implementation, the hardenability mathematical model includes a first sub-model for predicting Cr content, a second sub-model for predicting J9 hardness, and a third sub-model for predicting J15 hardness.

[0011] In one possible implementation, the calculation formula for the first sub-model is: in, This represents the mass fraction of carbon (C) in the steel. This represents the mass fraction of Mn in the steel. - These are the empirical parameters to be fitted.

[0012] In one possible implementation, the calculation formula for the second sub-model is: in, This represents the mass fraction of carbon (C) in the steel. This represents the mass fraction of Cr in the steel. This represents the mass fraction of Ti in the steel. The mass fraction of silicon in the steel is denoted by , and FDT is the finishing rolling temperature. - These are the empirical parameters to be fitted.

[0013] In one possible implementation, the calculation formula for the third sub-model is: in, This represents the mass fraction of carbon (C) in the steel. This represents the mass fraction of Cr in the steel. This represents the mass fraction of Ti in the steel. The value represents the mass fraction of Si in the steel, SFT is the initial cooling temperature, and A represents the Class A sulfide inclusion level. - These are the empirical parameters to be fitted.

[0014] Compared with the prior art, the beneficial effects of this application are as follows: This application uses key factors affecting the hardenability of low-carbon alloy steel, such as chemical composition and heat treatment process, summarized from big data from the production site as the basis for building a model. By minimizing the prediction error, a method to automatically discover mathematical expressions can automatically find the best mathematical model from a large amount of data without having to pre-define a specific function form. The hardenability prediction method in this application has the characteristics of low cost, short cycle, and high accuracy.

[0015] This application uses industrial big data mining based on empirical knowledge and Pearson correlation for feature selection, and combines a hardenability modeling strategy based on symbolic regression, which has guiding significance for accurately predicting the hardenability value of low-carbon alloy steel. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a method for predicting the hardenability of medium and low carbon alloy steel provided in this application embodiment; Figure 2 The Pearson correlation analysis results provided for the embodiments of this application; Figure 3 This is a schematic diagram comparing the model predictions and measured values ​​provided in the embodiments of this application. Detailed Implementation

[0017] The present solution will now be described in conjunction with the accompanying drawings and specific embodiments.

[0018] Figure 1 A flowchart illustrating a method for predicting the hardenability of medium- and low-carbon alloy steel provided in this application embodiment is shown below. Figure 1 This application provides a method for predicting the hardenability of medium and low carbon alloy steel, comprising: S101 collects historical production data from the industrial production MES system and constructs a high-dimensional industrial dataset containing steel composition, smelting parameters, rolling parameters, and hardenability indicators.

[0019] In this embodiment, the steel composition, smelting parameters, rolling parameters, and hardenability index of medium and low carbon alloy steel are collected from the industrial production MES system. The steel composition includes C, Si, Mn, Cr, Ni, Mo, V, Ti, and B. The smelting parameters include refining temperature, deoxidation time, N content, and O content. The rolling parameters include final rolling temperature, initial cooling temperature, and final cooling temperature. The collected steel composition, smelting parameters, rolling parameters, and hardenability index are fused to generate a high-dimensional industrial dataset.

[0020] S102. Based on prior knowledge of smelting physical mechanisms, a high-dimensional industrial dataset was initially screened, and then features significantly correlated with the hardenability target value were selected using Pearson correlation coefficient analysis.

[0021] See Figure 2 In this embodiment, based on prior knowledge of the physical mechanism of smelting, redundant variables that have no physical relationship with hardenability are excluded, and the Pearson correlation coefficient between each remaining feature variable in the high-dimensional industrial dataset and the target value of hardenability is calculated. The absolute value of the calculated correlation coefficient is compared with a set threshold, and feature variables with an absolute value of correlation coefficient greater than the set threshold are selected.

[0022] In this embodiment, the features significantly related to the hardenability target value include: , , , , The specific information regarding the nitrogen content, oxygen content, inclusions, final rolling temperature, final cooling temperature, and cooling bed temperature of the finished product is shown in Table 1. S103 uses the selected features that are significantly related to the hardenability target value as input variables and the Jominy distance hardness as the output variable, and establishes an explicit mathematical model of hardenability using the symbolic regression method.

[0023] Figure 3 This is a comparison chart of the model predictions and measured values ​​provided in this embodiment. In this embodiment, the dataset containing significantly correlated features and hardenability target values ​​is divided into a training set and a test set in a 4:1 ratio. HeuristicLab, based on a genetic algorithm, is used to automatically search for a mathematical expression that can fit the relationship between the input features and the output hardness values ​​in the training set. During the search process, a Pareto front optimization strategy is used to optimize the prediction accuracy index and complexity until a model that meets the accuracy requirement of MAE less than or equal to 2 HRC is obtained. The test set is used to verify the model that meets the requirements, and the final hardenability mathematical model is output.

[0024] The symbolic regression used in this embodiment differs from black-box models such as machine learning. Symbolic regression aims to discover explicit mathematical expressions that describe the relationship between inputs (composition and process parameters) and outputs (hardenability indicators). The generated model is interpretable and helps to understand how chemical composition, processes, etc., quantitatively affect hardenability, potentially guiding alloy design and process optimization.

[0025] In this embodiment, the hardenability mathematical model includes a first sub-model for predicting Cr content, a second sub-model for predicting J9 hardness, and a third sub-model for predicting J15 hardness. The fitted values ​​of the parameters in the model are shown in Table 2. The calculation formula for the first sub-model is as follows: in, This represents the mass fraction of carbon (C) in the steel. This represents the mass fraction of Mn in the steel. - These are the empirical parameters to be fitted.

[0026] The calculation formula for the second sub-model is: in, This represents the mass fraction of carbon (C) in the steel. This represents the mass fraction of Cr in the steel. This represents the mass fraction of Ti in the steel. The mass fraction of silicon in the steel is denoted by , and FDT is the finishing rolling temperature. - These are the empirical parameters to be fitted.

[0027] The calculation formula for the third sub-model is: in, This represents the mass fraction of carbon (C) in the steel. This represents the mass fraction of Cr in the steel. This represents the mass fraction of Ti in the steel. The value represents the mass fraction of Si in the steel, SFT is the initial cooling temperature, and A represents the Class A sulfide inclusion level. - These are the empirical parameters to be fitted.

[0028] S104: Input the composition and process parameters of the medium-low carbon alloy steel to be predicted into the hardenability mathematical model to achieve the prediction of the hardenability of the medium-low carbon alloy steel to be predicted.

[0029] In this embodiment, 20CrMnTiH steel is used as the medium-low carbon alloy steel to be predicted. Its composition is: C: 0.17~0.23%, Si: 0.17~0.37%, Mn: 0.80-1.20%, Cr: 1.00~1.45%, Ti: 0.04~0.10%. Three different chemical compositions are selected for prediction. The specific chemical composition ratios are shown in Table 3, the specific production process is shown in Table 4, and the output model and predicted hardenability results are shown in Table 5. The composition data and process parameters of 20CrMnTiH steel are input into the hardenability mathematical model for prediction, and the final prediction results are obtained. The prediction results for the first group are J9=44.2HRC and J15=37.6HRC. The prediction results for the second group are J9=44.1HRC and J15=37.3HRC; and the prediction results for the third group are J9=42.5HRC and J15=35.4HRC.

[0030] In this embodiment, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0031] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0032] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A method of predicting hardenability of a medium-low carbon alloy steel, characterized by, The application relates to a method for predicting hardenability of low-carbon and medium-carbon alloy steel. The method comprises the following steps: Collecting historical production data from an industrial production MES system, and constructing a high-dimensional industrial data set containing steel composition, smelting parameters, rolling parameters and hardenability indexes; Preliminarily screening the high-dimensional industrial data set based on prior knowledge of smelting physical mechanism, and selecting features significantly related to the hardenability target value through a Pearson correlation coefficient analysis method; Taking the selected features significantly related to the hardenability target value as input variables and Jominy distance hardness as output variables, and establishing an explicit hardenability mathematical model by using a symbolic regression method; 2. The method of predicting hardenability of medium-low carbon alloy steel according to claim 1, characterized in that, Inputting the composition and process parameters of the low-carbon and medium-carbon alloy steel to be predicted into the hardenability mathematical model to realize prediction of the hardenability of the low-carbon and medium-carbon alloy steel to be predicted. The method for collecting historical production data from an industrial production MES system, and constructing a high-dimensional industrial data set containing steel composition, smelting parameters, rolling parameters and hardenability indexes comprises the following steps: Collecting steel composition, smelting parameters, rolling parameters and hardenability indexes of low-carbon and medium-carbon alloy steel from an industrial production MES system, wherein the steel composition comprises C, Si, Mn, Cr, Ni, Mo, V, Ti and B, the smelting parameters comprise refining temperature, deoxidation time, N content and O content, and the rolling parameters comprise finishing temperature, open cooling temperature and final cooling temperature; 3. The method of predicting hardenability of medium-low carbon alloy steel according to claim 1, characterized in that, Fusing the collected steel composition, smelting parameters, rolling parameters and hardenability indexes to generate a high-dimensional industrial data set. The method for preliminarily screening the high-dimensional industrial data set based on prior knowledge of smelting physical mechanism, and selecting features significantly related to the hardenability target value comprises the following steps: Excluding redundant variables irrelevant to the hardenability based on prior knowledge of smelting physical mechanism; Calculating Pearson correlation coefficients between the remaining feature variables in the high-dimensional industrial data set and the hardenability target value; 4. The method of predicting hardenability of medium-low carbon alloy steel according to claim 1, characterized in that, Comparing the absolute values of the calculated correlation coefficients with a set threshold value, and selecting feature variables with absolute values greater than the set threshold value. The method for taking the selected features significantly related to the hardenability target value as input variables, Jominy distance hardness as output variables, and establishing an explicit hardenability mathematical model by using a symbolic regression method comprises the following steps: Dividing a data set containing significantly related features and the hardenability target value into a training set and a test set according to a preset ratio; Using a HeuristicLab based on a genetic algorithm to automatically search for a mathematical expression capable of fitting the relationship between the input features and the output hardness value in the training set; In the searching process, a Pareto front optimization strategy is used to optimize the prediction accuracy index and the complexity until a model meeting the preset accuracy requirement is obtained; 5. The method of predicting hardenability of medium-low carbon alloy steel according to claim 4, characterized in that, Using the test set to verify the model, and outputting a final hardenability mathematical model.

6. The method of predicting hardenability of medium-low carbon alloy steel according to claim 5, characterized in that, The hardenability mathematical model comprises a first sub-model for predicting Cr content, a second sub-model for predicting J9 hardness and a third sub-model for predicting J15 hardness. wherein, is the mass fraction of C element in the steel, is the mass fraction of Mn element in the steel, - is an empirical parameter to be fitted.

7. The method of predicting hardenability of medium-low carbon alloy steel according to claim 5, characterized in that, The calculation formula of the first sub-model is as follows: wherein, is the mass fraction of C element in the steel, is the mass fraction of Cr element in the steel, is the mass fraction of Ti element in the steel, is the mass fraction of Si element in the steel, FDT is the finishing temperature of the finishing rolling, - is an empirical parameter to be fitted.

8. The method of predicting hardenability of medium-low carbon alloy steel according to claim 5, characterized in that, The calculation formula of the second sub-model is as follows: The calculation formula of the third sub-model is as follows: wherein, is the mass fraction of C element in the steel, is the mass fraction of Cr element in the steel, is the mass fraction of Ti element in the steel, is the mass fraction of Si element in the steel, SFT is the open cooling temperature, A is the grade of class A sulfide inclusions, - is an empirical parameter to be fitted.

Images