A method for predicting mechanical properties of hot-rolled strip integrating physical metallurgy and data-driven

Abstract

The present invention provides a method for predicting the mechanical properties of hot-rolled strip steel that integrates physical metallurgy and data-driven, which belongs to the research field of prediction of mechanical properties of hot-rolled materials, including: obtaining a production data set of hot-rolled strip steel, and calculating the proportion of each organizational component of the hot-rolled strip steel through a physical metallurgical model optimized by a genetic algorithm; constructing a data set required for mechanical property prediction and preprocessing it to obtain a preprocessed data set; dividing the preprocessed data set into a training set and a test set; constructing a random forest model, using the random forest model to calculate the training set, obtaining the optimal parameters, and predicting the mechanical properties of the test set. The present invention adopts the above-mentioned method for predicting the mechanical properties of hot-rolled strip steel that integrates physical metallurgy and data-driven, and uses the component process parameters and the proportion of each phase fraction as input data to predict the mechanical properties of steel, which can effectively improve the accuracy of mechanical property prediction, help to optimize process parameters, and improve product quality.

Description

Technical Field

[0001] The invention relates to the research field of prediction of mechanical properties of hot-rolled materials, and in particular to a method for predicting mechanical properties of hot-rolled strip steel that integrates physical metallurgy and data-driven. Background Art

[0002] With the continuous development of the steel industry, steel with excellent properties such as high strength, corrosion resistance and wear resistance has been sought after by people. At present, high-strength steel has been widely used in many fields such as ships, automobiles, buildings, bridges, high-pressure transmission pipelines, machinery, etc., accounting for a large proportion of social demand. The use of this high-strength steel has achieved quite significant economic and social benefits, and has become a type of steel that various countries are competing to develop and produce.

[0003] The traditional way to obtain the mechanical properties of steel is mainly to conduct mechanical properties experiments, which consumes a lot of time and manpower costs. With the development of machine learning, the method of establishing a neural network model to predict the mechanical properties of steel through machine learning methods has entered everyone's field of vision. At present, there are many patents on the prediction model of mechanical properties of steel, such as: the Chinese patent with publication number CN116825253B discloses a method for establishing a hot-rolled strip mechanical properties prediction model based on feature selection, which predicts the mechanical properties of hot-rolled strip by inputting hot-rolled strip data including C, Si, Mn, P, S, N, Nb, Ti and other component contents and heating time (Heat_Time), solution temperature (SS_Temp), solution time (SS_Time), furnace temperature (DTF), finishing entrance thickness (FEH), finishing entrance temperature (FET) and other process parameters. The Chinese patent with publication number CN112100745B predicts the mechanical properties of automobile beam steel based on the LDA theory. By inputting chemical composition and process parameters (chemical composition includes C, Si, Mn, Ti, V, Nb; process parameters include final rolling temperature FDT, coiling temperature CT, intermediate billet thickness RDT), the yield strength, tensile strength and elongation of automobile beam steel are predicted.

[0004] As we all know, steel composition and process affect the evolution of steel organization, while the microstructure determines the mechanical properties of steel. The above patents predict the mechanical properties of steel by inputting the composition and process parameters of steel, and do not consider the evolution of microstructure in the modeling process. Therefore, taking the influence of organizational evolution behavior on mechanical properties into account in the prediction model and establishing the composition-process-organization-performance relationship of steel is one of the key links to improve the accuracy of mechanical property prediction. Summary of the invention

[0005] The purpose of the present invention is to provide a method for predicting the mechanical properties of hot-rolled strip that integrates physical metallurgy and data-driven, and uses the composition process parameters and the proportion of each phase fraction as input data to predict the mechanical properties of steel, which can effectively improve the accuracy of mechanical property prediction, help to optimize process parameters, and improve product quality.

[0006] To achieve the above object, the present invention provides a method for predicting the mechanical properties of hot-rolled strip steel by integrating physical metallurgy and data-driven, comprising the following steps:

[0007] S1. Obtain a production data set of hot-rolled strip steel, and calculate the proportion of each organizational component of the hot-rolled strip steel through a physical metallurgical model optimized by a genetic algorithm;

[0008] S2, constructing a data set required for mechanical property prediction and preprocessing it to obtain a preprocessed data set;

[0009] S3, dividing the preprocessed data set into a training set and a test set;

[0010] S4. Construct a random forest model, use the random forest model to calculate the training set, obtain the optimal parameters, and predict the mechanical properties of the test set.

[0011] Preferably, in step S1, the production data set includes component content data and process data.

[0012] Preferably, the component content data includes C content, Si content, Mn content, S content, P content, N content, Nb content, V content, Ti content, Als content, Cu content, Cr content, Ni content, Co content, Mo content and B content;

[0013] Preferably, the process data include soaking temperature, heating temperature, time in the furnace, rolling speed, rolling force, strip running speed in the annealing section, exit thickness, finishing rolling entrance temperature, final rolling temperature and coiling temperature.

[0014] Preferably, in step S1, each structure includes: ferrite, pearlite and bainite formed after phase transformation.

[0015] Preferably, in step S1, calculating the proportion of each organizational component of the hot-rolled strip by a physical metallurgical model optimized by a genetic algorithm comprises the following steps:

[0016] S11. Using the component content data and process data in the production data set, the temperature field is calculated using the finite difference method to obtain the optimal heat transfer coefficient;

[0017] The best heat transfer coefficient is obtained as follows:

[0018]

[0019] in, are the calculated values ​​of rough rolling outlet temperature, final rolling temperature and coiling temperature of the i-th group of data respectively; are the measured values ​​of rough rolling outlet temperature, final rolling temperature and coiling temperature of the i-th group of data respectively; FUN temp is the fitness function of the temperature field; n is the total number of data sets obtained; when FUN temp The best heat transfer coefficient is obtained when the value of is the smallest;

[0020] S12, after obtaining the best heat transfer coefficient, the physical metallurgical model is optimized by using a genetic algorithm, and the optimal physical metallurgical model parameters are obtained by optimizing the genetic algorithm parameters, and finally the predicted values ​​of each phase fraction are obtained;

[0021] The objective function to be optimized in the physical metallurgical model is as follows:

[0022]

[0023] in, are the calculated values ​​of yield strength, tensile strength and elongation of the i-th group of data respectively; are the measured values ​​of yield strength, tensile strength and elongation of the i-th group of data respectively; FUN phy is the fitness function of the physical metallurgical model; when FUN phy When the value of is the smallest, the optimal parameters of the population number, evolutionary generations, scaling coefficient and crossover factor in the genetic algorithm are obtained.

[0024] Preferably, in step S11, the heat transfer coefficient includes an air cooling heat transfer coefficient, a descaling heat transfer coefficient, a rolling pass interval heat transfer coefficient, a rolling heat transfer coefficient and a laminar cooling heat transfer coefficient.

[0025] Preferably, in step S2, the data set required for mechanical property prediction includes component content data, process data and the proportion of each tissue component.

[0026] Therefore, the present invention adopts the above-mentioned method for predicting the mechanical properties of hot-rolled strip that integrates physical metallurgy and data-driven, and the technical effects are as follows: using the composition process parameters and the proportion of each phase fraction as input data at the same time to realize the prediction of the mechanical properties of steel, can effectively improve the accuracy of mechanical property prediction, help to optimize process parameters, and improve product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of a method for predicting mechanical properties of hot-rolled strip steel that integrates physical metallurgy and data-driven technology in the present invention;

[0028] Figure 2 This is a comparison chart of model prediction results; Figure 2 (a) is a comparison chart of the predicted and actual yield strength values ​​of the traditional mechanism model and the composition-process-structure relationship model; Figure 2 (b) is a comparison chart of the predicted and actual values ​​of tensile strength using the traditional mechanism model and the composition-process-structure relationship model; Figure 2 (c) is a comparison chart of the predicted and actual values ​​of elongation using the traditional mechanism model and the composition-process-structure relationship model;

[0029] Figure 3 This is a comparison chart of R value, MAE, and RMSE results; among them, Figure 3 (a) is a comparison chart of yield strength R value, MAE, and RMSE results between the traditional mechanism model and the composition-process-structure relationship model; Figure 3 (b) is a comparison chart of tensile strength R value, MAE, and RMSE results between the traditional mechanism model and the composition-process-structure relationship model; Figure 3 (c) is a comparison chart of the elongation R value, MAE, and RMSE results of the traditional mechanism model and the composition-process-structure relationship model. DETAILED DESCRIPTION

[0030] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0031] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0032] Embodiment 1

[0033] like Figure 1 As shown, it is a flow chart of a method for predicting mechanical properties of hot-rolled strip steel that integrates physical metallurgy and data-driven according to the present invention, which specifically includes the following steps:

[0034] Compared with other high-strength steels, S500MC steel not only has good mechanical properties, but also has the characteristics of low cost and good environmental protection. Therefore, it has received widespread attention in the industry and has high research value. Reliable performance prediction can not only save test costs and time, but also help to optimize the steel composition and production process, thereby improving product quality. Therefore, it is very necessary to predict the mechanical properties of S500MC steel in advance.

[0035] This embodiment takes S500MC steel as an example.

[0036] S1. Organizational Information Acquisition

[0037] The production data of several S500MC steels were selected to establish the basic data for predicting the proportion of the microstructure. The proportion of each phase in the steel was predicted through the physical metallurgical model. The microstructure included ferrite, pearlite and bainite formed after phase transformation.

[0038] The specific steps for obtaining organizational information are as follows:

[0039] S11. Using the component content data and process data in the production data set, the temperature field is calculated using the finite difference method to obtain the optimal heat transfer coefficient;

[0040] The best heat transfer coefficient is obtained as follows:

[0041]

[0042] in, are the calculated values ​​of rough rolling outlet temperature, final rolling temperature and coiling temperature of the i-th group of data respectively; are the measured values ​​of rough rolling outlet temperature, final rolling temperature and coiling temperature of the i-th group of data respectively; FUN temp is the fitness function of the temperature field; n is the total number of data sets obtained; when FUN temp The best heat transfer coefficient is obtained when the value of is the smallest;

[0043] S12. After obtaining the best heat transfer coefficient, the physical metallurgical model is optimized by using a genetic algorithm. The optimal model parameters are obtained by optimizing the genetic algorithm parameters, and finally the predicted values ​​of the fractions of each phase with an error range of more than 77% and the actual value within ±5% are obtained;

[0044] The objective function to be optimized in the physical metallurgical model is as follows:

[0045]

[0046] in, are the calculated values ​​of yield strength, tensile strength and elongation of the i-th group of data respectively; are the measured values ​​of yield strength, tensile strength and elongation of the i-th group of data respectively; FUN phy is the fitness function of the physical metallurgical model; when FUN phy When the value of is the smallest, the optimal parameters of the population number, evolutionary generations, scaling coefficient and crossover factor in the genetic algorithm are obtained.

[0047] S2. Selection of data samples

[0048] The production data of S500MC steel with obtained organizational information was selected, and samples with large differences in component and process data such as the main components of steel and slab thickness were collected. Finally, the data set required for mechanical property prediction (component content data, process data and the proportion of each organizational component) was obtained.

[0049] The specific model parameters are shown in Table 1.

[0050] Table 1 Model parameters

[0051]

[0052]

[0053] S3. Outlier removal

[0054] The program was used to find missing data. If the data contained missing components or main process parameters, such data were deleted as 0. The components, process parameters and organizational information were eliminated according to the Laida criterion for outliers, with minimal noise interference, and finally a total of 960 groups of complete data were obtained.

[0055] S4. Data partitioning

[0056] The data set after removing outliers is randomly divided into two parts: training data and prediction data, so that the ratio of training data to prediction data remains at about 3:1, and finally 720 groups of training data and 240 groups of prediction data are obtained.

[0057] S5. Constructing a component-process-organization correspondence model

[0058] Determine the number of decision trees in the random forest. Here, 20 decision trees are used, and there is no limit on the maximum number of leaf nodes. Based on this, random forest modeling is performed on the training data, and parameters with good fitting effects are obtained. Finally, the optimal structure is obtained and the mechanical properties of the test set are tested.

[0059] The pure mechanism model established with the empirical formula was used to predict the mechanical properties of steel, and the correlation coefficient, mean absolute error, and root mean square error were compared between the two to obtain the influence of the composition-process-organization correspondence on the accuracy of mechanical property prediction.

[0060] The empirical formula is as follows:

[0061] Yield strength σ s Empirical formula:

[0062] σ s =σ C +σ GD +σ S +σ P

[0063] Among them, σ C is a constant, generally taken as 108.69MPa; GD is the contribution of the interaction between grain boundary strengthening and dislocation strengthening to the yield strength; σ s is the contribution of solid solution strengthening to yield strength; σ P is the contribution of precipitation strengthening to yield strength.

[0064] Tensile strength b Empirical formula:

[0065]

[0066] in, and As constants, take 1.6MPa and 19.8N / mm respectively. 3 / 2 , 228MPa; X F , X P and X B are the volume fractions of ferrite, pearlite and bainite respectively; H F , H P and H B These are the microhardness of ferrite, pearlite and bainite respectively.

[0067] Elongation EL empirical formula:

[0068]

[0069] Among them, d F is the equivalent average grain diameter of ferrite, and h is the thickness of the strip.

[0070] Prediction results:

[0071] Through this embodiment, the following prediction results can be obtained:

[0072] By establishing a composition-process-organization correspondence model, the prediction accuracy of the yield strength of S500MC steel can reach 96.8% within the relative error of ±6% between the predicted value and the actual value. Figure 2 As shown in (a) in the figure, the relative error between the predicted value and the measured value of tensile strength is within ±6%, and the prediction accuracy can reach 98.9%. Figure 2 As shown in (b) in the figure, the prediction accuracy can reach 100% when the absolute error between the predicted value and the measured value of the elongation is within ±4%. Figure 2 As shown in (c) in .

[0073] Through the pure mechanism model established by empirical formula, the prediction accuracy of the yield strength of S500MC steel can reach 79.8% within the relative error of ±6% between the predicted value and the actual value. Figure 2As shown in (a) in the figure, the relative error between the predicted value and the measured value of tensile strength is within ±6%, and the prediction accuracy can reach 85.4%. Figure 2 As shown in (b) in the figure, the prediction accuracy can reach 83.6% when the absolute error between the predicted value and the measured value of the elongation is within ±4%. Figure 2 As shown in (c) in .

[0074] pass Figure 3 (a) Figure 3 (b) Figure 3 The following conclusions can be drawn from the comparison of the steel yield strength, tensile strength, elongation R value, MAE, and RMSE results corresponding to (c):

[0075] By establishing the composition-process-organization correspondence model, the correlation coefficient of the yield strength of S500MC steel can reach 0.65, the correlation coefficient of the tensile strength of S500MC steel can reach 0.71, and the correlation coefficient of the elongation of S500MC steel can reach 0.75.

[0076] Through the pure mechanism model established with empirical formula, the correlation coefficient of yield strength of S500MC steel can reach 0.27, the correlation coefficient of tensile strength of S500MC steel can reach 0.28, and the correlation coefficient of elongation of S500MC steel can reach 0.29.

[0077] By establishing a composition-process-organization correspondence model, the average absolute errors of the mechanical properties (yield strength, tensile strength, and elongation) of S500MC steel are obtained to be 21.2, 22.3, and 1.6, respectively. By using a pure mechanism model established with empirical formulas, the average absolute errors of the mechanical properties (yield strength, tensile strength, and elongation) of S500MC steel are obtained to be 28.7, 22.2, and 4.2, respectively.

[0078] By establishing a composition-process-organization correspondence model, the root mean square errors of the mechanical properties (yield strength, tensile strength, and elongation) of S500MC steel are obtained to be 26.1, 11.3, and 2.1, respectively. By using a pure mechanism model established with empirical formulas, the root mean square errors of the mechanical properties (yield strength, tensile strength, and elongation) of S500MC steel are obtained to be 34.5, 27.9, and 5.9, respectively.

[0079] By comparing and analyzing the prediction model of composition-process-organization correspondence and the traditional empirical formula calculation model, it can be seen that the prediction model of composition-process-organization correspondence is better than the traditional empirical formula calculation model in predicting the mechanical properties of S500MC steel.

[0080] It can be seen from this example that the prediction model of the composition-process-organization correspondence is more accurate in predicting the mechanical properties of S500MC steel than the traditional empirical formula calculation model. It has guiding significance for the optimization of product composition and production process, and helps to improve product quality.

[0081] Therefore, the present invention adopts the above-mentioned hot-rolled strip mechanical properties prediction method that integrates physical metallurgy and data-driven, and uses the composition process parameters and the proportion of each phase fraction as input data to realize the prediction of the mechanical properties of steel, which can effectively improve the accuracy of mechanical properties prediction, help to optimize process parameters, and improve product quality.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for predicting mechanical properties of hot-rolled strip steel that integrates physical metallurgy and data-driven, characterized in that: The following steps are involved: S1. Obtain a production data set of hot-rolled strip steel, and calculate the proportion of each organizational component of the hot-rolled strip steel through a physical metallurgical model optimized by a genetic algorithm; S2, constructing a data set required for mechanical property prediction and preprocessing it to obtain a preprocessed data set; S3, dividing the preprocessed data set into a training set and a test set; S4. Construct a random forest model, use the random forest model to calculate the training set, obtain the optimal parameters, and predict the mechanical properties of the test set; In step S1, the production data set includes component content data and process data; Among them, the component content data include C content, Si content, Mn content, S content, P content, N content, Nb content, V content, Ti content, Als content, Cu content, Cr content, Ni content, Co content, Mo content and B content; Process data include soaking temperature, heating temperature, furnace time, rolling speed, rolling force, strip running speed in annealing section, exit thickness, finishing rolling entrance temperature, final rolling temperature and coiling temperature; The various structures include: ferrite, pearlite and bainite formed after phase transformation; In step S1, the proportion of each organizational component of the hot-rolled strip is calculated by a physical metallurgical model optimized by a genetic algorithm, including the following steps: S11. Using the component content data and process data in the production data set, the temperature field is calculated using the finite difference method to obtain the optimal heat transfer coefficient; The best heat transfer coefficient is obtained as follows: in, are the calculated values ​​of rough rolling outlet temperature, final rolling temperature and coiling temperature of the i-th group of data respectively; are the measured values ​​of rough rolling outlet temperature, final rolling temperature and coiling temperature of the i-th group of data respectively; FUN temp is the fitness function of the temperature field; n is the total number of data sets obtained; when FUN temp The best heat transfer coefficient is obtained when the value of is the smallest; S12, after obtaining the best heat transfer coefficient, the physical metallurgical model is optimized by using a genetic algorithm, and the optimal physical metallurgical model parameters are obtained by optimizing the genetic algorithm parameters, and finally the predicted values ​​of each phase fraction are obtained; The objective function to be optimized in the physical metallurgical model is as follows: in, are the calculated values ​​of yield strength, tensile strength and elongation of the i-th group of data respectively; are the measured values ​​of yield strength, tensile strength and elongation of the i-th group of data respectively; FUN phy is the fitness function of the physical metallurgical model; when FUN phy When the value of is the smallest, the optimal parameters of the population number, evolutionary generations, scaling coefficient and crossover factor in the genetic algorithm are obtained.

2. The method for predicting mechanical properties of hot-rolled steel strip integrating physical metallurgy and data-driven according to claim 1, characterized in that: In step S11, the heat transfer coefficient includes the air cooling heat transfer coefficient, the descaling heat transfer coefficient, the rolling pass interval heat transfer coefficient, the rolling heat transfer coefficient and the laminar cooling heat transfer coefficient.

3. The method for predicting mechanical properties of hot-rolled steel strip integrating physical metallurgy and data-driven according to claim 2, characterized in that: In step S2, the data set required for mechanical property prediction includes component content data, process data and the proportion of each tissue component.

Images