A machine learning method for mechanical properties of hot-rolled steel guided by physical metallurgy

Abstract

The present invention discloses a machine learning method for mechanical properties of hot-rolled steel guided by physical metallurgy, which belongs to the intersection of steel plate production and data statistical modeling, including: collecting hot-rolled steel plate data and preprocessing the data; calculating physical metallurgical parameters based on steel plate data and physical metallurgy theory; optimizing mechanical properties calculation model parameters using particle swarm optimization algorithm, and constructing a mechanical properties prediction model. The present invention adopts the above-mentioned machine learning method for mechanical properties of hot-rolled steel guided by physical metallurgy, builds a high-quality data set, selects the optimal algorithm modeling to predict mechanical properties, and guides steel plate production.

Description

Technical Field

[0001] The invention relates to the cross-technical field of steel plate production and data statistical modeling, and in particular to a machine learning method for mechanical properties of hot-rolled steel guided by physical metallurgy. Background Art

[0002] Accurately predicting mechanical properties based on the chemical composition of steel, production process parameters, etc. is an important research topic in the steel industry. It has high application value in actual production and is used to predict mechanical properties, reduce sampling and improve detection efficiency, optimize production processes, and develop new steel grades.

[0003] At present, the prediction of mechanical properties mainly adopts physical metallurgical models and data-driven models. Although the physical metallurgical model has good prediction regularity, the prediction accuracy is low and the field application effect is not good; although the data-driven machine learning model has high prediction accuracy, the quality of field data and the complexity of physical mechanisms have a significant impact on the generalization and stability of the model, which hinders its promotion in many practical applications. In addition, compared with strip steel, the temperature field and process flow in the steel plate production process are more complicated, and further exploration is needed in the high-precision mechanical property prediction modeling. Summary of the invention

[0004] The purpose of the present invention is to provide a machine learning method for the mechanical properties of hot-rolled steel guided by physical metallurgy, adopt data processing methods such as hierarchical clustering and outlier removal to further improve data quality, build a high-quality data set, and optimize the physical model parameters by combining machine learning algorithms and physical metallurgy theory, and finally establish a mechanical properties prediction model with high precision and good generalization.

[0005] To achieve the above object, the present invention provides a machine learning method for mechanical properties of hot-rolled steel guided by physical metallurgy, the steps comprising:

[0006] S1, collecting hot-rolled steel plate data and preprocessing the data;

[0007] S2, calculate physical metallurgical parameters based on the data in S1 and physical metallurgical theory;

[0008] S3. Use particle swarm optimization algorithm to optimize the parameters of the mechanical properties calculation model and construct a mechanical properties prediction model.

[0009] Preferably, the data collected in step S1 include steel plate composition specification parameters, hot rolling process parameters and mechanical property data, and the data preprocessing includes:

[0010] S11. Perform hierarchical clustering on the multidimensional data according to the parameter range, select the number of clusters, divide the data into different clusters, and classify the data, specifically:

[0011] Each sample in the data is taken as an initial cluster, and then the two closest clusters are found and merged at each step of the algorithm. When the key components and process parameters in the same cluster are the same or similar, they can be considered as data produced by the same rolling procedure. The selection of the hierarchical clustering algorithm is based on the merging criterion of the Mahalanobis distance, that is, if the distance between cluster 1 and cluster 2 is the shortest Mahalanobis distance between all different clusters, cluster 1 and cluster 2 are merged. Mahalanobis distance d M It can be found by the following formula:

[0012]

[0013] In the formula, x is the difference matrix of two objects; is the mean vector.

[0014] S12. According to the Grubbs criterion and the 3σ criterion, the values ​​with large detection errors in the classification data are judged and the abnormal values ​​are eliminated. Specifically, on the basis of hierarchical clustering, different criteria are selected according to the number of steel plates in the same cluster. That is, when the number of samples in the cluster is greater than 30, the 3σ criterion is used, otherwise, the Grubbs criterion is used. Assume that the production data of a group is:

[0015]

[0016] In the formula, n is the number of samples; p is the number of variables. j 1 (j = 1, 2, ..., n) is the mechanical property data, then the mean and variance of the performance data can be calculated by the following formula:

[0017]

[0018]

[0019] According to the 3σ criterion, if x~N(u,σ), then the probability P(|x-μ|>3σ)≤0.003. Therefore, when the residual error m of the yield strength of a steel plate is j =x j When 1–μ satisfies the following formula, it can be considered as an outlier and should be removed.

[0020] |m j |=|x j1 -μ|>3σ

[0021] When using the Grubbs criterion, first calculate the Grubbs critical value G of each coil of steel. If G>G(n,α), the yield strength of the corresponding steel plate is considered an abnormal value and is eliminated. The calculation formula of the Grubbs critical value G is as follows:

[0022]

[0023] In the formula, M is the median of n groups of data, and α is the significance level.

[0024] Preferably, step S2 comprises:

[0025] S21. Analyze the microstructure evolution behavior and the influencing factors of the microstructure characteristics on the mechanical properties during hot rolling based on the physical metallurgical theory;

[0026] S22. Based on the data collected in S1, the physical metallurgical model is used to calculate the physical metallurgical parameters that affect the final microstructure characteristics and mechanical properties model.

[0027] Preferably, step S3 includes:

[0028] S31, based on the calculated microstructure characteristic information and mechanical properties calculation model parameters, the mechanical properties calculation model parameters are optimized using a particle swarm optimization algorithm;

[0029] S32, taking component specifications and hot rolling process parameters as input and mechanical property model optimization parameters as output, dividing the training set and the test set, and using a random forest network to construct a mechanical property prediction model;

[0030] S33. Predict the test set based on the mechanical property prediction model to obtain the predicted values ​​of the mechanical property parameters of the steel to be predicted. At the same time, calculate the mechanical properties using the mechanical property calculation model, and compare the calculation results with the measured mechanical properties.

[0031] Preferably, step S21 specifically includes: analyzing, according to the mechanical property calculation model, the influencing factors related to the austenite grain growth during heating, the recrystallization, grain growth and precipitation during rolling, and the phase transformation behavior during cooling, wherein the influencing factors include the precipitation fraction and size, dislocation density, ferrite fraction and size, pearlite fraction and bainite fraction.

[0032] Preferably, the mechanical property calculation model includes a yield strength model, a tensile strength model and an elongation model, specifically:

[0033] Yield Strength Model

[0034] YS=σ C +σ Grain +σ dis +σ pre

[0035] σ C =77+60Si+80Cu+45Ni+60Cr+80Mn+11Mo+5000N+750P

[0036] σ Grain =K g d F-0.5

[0037]

[0038]

[0039] Tensile Strength Model

[0040]

[0041] H F =458-0.357T mF +50Si

[0042] H B =669-0.588T mB +50Si

[0043] Elongation

[0044]

[0045] Where, YS is the yield strength (MPa), σ C is the solid solution strengthening strength (MPa), σ Grain is the fine grain strengthening strength (MPa), σ dis is the dislocation strengthening strength (MPa), Si, Cu, Ni, Cr, Mn, Mo, N and P are the mass fractions of the elements (wt%), d F is the final ferrite grain size (μm), ρ dis is the dislocation density (m2), d pre is the average size of the precipitated phase particles (nm), f pre is the fraction of precipitated phase, v is the Poisson's ratio of 0.3, and b is the Burgers vector of 2.48×10 -10 m, μ is the shear modulus of 8×10 10 Pa. TS is the tensile strength (MPa), X F , X P and X B are the volume fractions of ferrite, pearlite and bainite, respectively, and H F , H P and H B are the microhardness (HV) of ferrite, pearlite and bainite, T mF and T mB are the phase equilibrium temperatures of ferrite and bainite, EL is the elongation after fracture (%), h is the thickness of the steel product (mm), K g , K d , K p 、f 1 、f 2 、f 3,EL 0 , k F , k d , k h Both are adjustment coefficients.

[0046] Preferably, in step S2, the physical metallurgical parameters include precipitation fraction, ferrite fraction and size, pearlite fraction and bainite fraction;

[0047] During the heating process, the austenite grains grow, and the austenite grain size depends on the heating temperature and the holding time; the accumulated strain ε during hot rolling a Exceeding the critical strain ε required for dynamic recrystallization DRX C , new DRX nuclei are generated along the elongated grain boundaries and lead to metadynamic recrystallization MDRX in the subsequent pass interval; otherwise, ε C ≥ε a Static recrystallization SRX occurs in the interval between passes, and the deformed grains are completely or partially replaced by new recrystallized grains; at the same time, the alloy elements dissolved in the heating process undergo strain induced precipitation during the hot rolling process, and the austenite grains decompose into ferrite, pearlite and bainite during the cooling process.

[0048] Preferably, in step S31, a particle swarm optimization algorithm is used to optimize the mechanical property model parameters under different components and process conditions.

[0049] Preferably, in step S33, the calculated results are compared with the measured mechanical properties, and the root mean square error (RMSE) and the coefficient of determination (R 2 ), the formula is:

[0050]

[0051]

[0052] In the formula, n is the number of samples, i is the sample number, is the predicted value, is the measured value, It is the average of the measured values, and the performance prediction value and the performance measured value are judged as pass / fail respectively.

[0053] Therefore, the present invention adopts the above-mentioned physical metallurgy-guided hot-rolled steel mechanical properties machine learning method, which has the following beneficial effects:

[0054] (1) Data classification is achieved through data processing methods such as hierarchical clustering and outlier removal, which simplifies redundant data, removes abnormal data with large errors, and makes the physical regularity of the data more significant;

[0055] (2) Under the guidance of the optimization algorithm, the model parameters are optimized based on the physical model of the entire hot rolling process, and the complex physical metallurgical phenomena of each process in the steel plate production process are fully considered;

[0056] (3) Based on key components and process parameters, a machine learning model for physical model parameter prediction was established. This model not only takes into account the good physical regularity of the physical model, but also includes the high precision of the data-driven model, thereby improving the detection efficiency, optimizing the composition and process parameters, and guiding the development of new steel grades.

[0057] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a flow chart of the present invention;

[0059] Figure 2 Comparison of predicted values ​​and measured values ​​of different machine learning algorithms of the present invention;

[0060] Figure 3 Comparison of the predicted and measured values ​​of this patented model with the data-driven machine learning model. DETAILED DESCRIPTION

[0061] Example

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0063] Reference Figure 1 The present invention discloses a machine learning method for mechanical properties of hot-rolled steel guided by physical metallurgy, the steps comprising:

[0064] S1. Collect the composition specification parameters and hot rolling process parameter data of hot-rolled AH36 steel and pre-process the data. The composition specification data include the mass fraction of C, Si, Mn, S, P, N, Nb, V, Ti, Al, Cu, Cr, Ni, Mo, B, the thickness and width of slabs and steel plates; hot rolling process parameters: heating furnace soaking temperature and holding time, rough rolling start temperature, finishing rolling start temperature, finishing rolling final temperature, rough rolling and finishing rolling each pass steel throwing temperature, rolling speed, rolling force, start cooling temperature, final cooling temperature and cooling rate; mechanical properties: yield strength, tensile strength and elongation.

[0065] Preprocessing includes:

[0066] S11. Perform hierarchical clustering on the multidimensional data according to the parameter range, select the number of clusters, divide the data into different clusters, and classify the data, specifically:

[0067] Each sample in the data is taken as an initial cluster, and then the two closest clusters are found and merged at each step of the algorithm. When the key components and process parameters in the same cluster are the same or similar, they can be considered as data produced by the same rolling procedure. The selection of the hierarchical clustering algorithm is based on the merging criterion of the Mahalanobis distance, that is, if the distance between cluster 1 and cluster 2 is the shortest Mahalanobis distance between all different clusters, cluster 1 and cluster 2 are merged. Mahalanobis distance d M It can be found by the following formula:

[0068]

[0069] Where x is the difference matrix of two objects; x is the mean vector.

[0070] S12. According to the Grubbs criterion and the 3σ criterion, the values ​​with large detection errors in the classification data are judged and the abnormal values ​​are eliminated. Specifically, on the basis of hierarchical clustering, different criteria are selected according to the number of steel plates in the same cluster. That is, when the number of samples in the cluster is greater than 30, the 3σ criterion is used, otherwise, the Grubbs criterion is used. Assume that the production data of a group is:

[0071]

[0072] In the formula, n is the number of samples; p is the number of variables. j 1 (j = 1, 2, ..., n) is the mechanical property data, then the mean and variance of the performance data can be calculated by the following formula:

[0073]

[0074]

[0075] According to the 3σ criterion, if x~N(u,σ), then the probability P(|x-μ|>3σ)≤0.003. Therefore, when the residual error m of the yield strength of a steel plate is j =x j When 1–μ satisfies the following formula, it can be considered as an outlier and should be removed.

[0076] |m j |=|x j1 -μ|>3σ

[0077] When the Grubbs criterion is used, the Grubbs critical value G of each coil is first calculated. When the significance level α is known, G(n,α) is found by looking up Table 1. If G>G(n,α), the yield strength of the corresponding steel plate is considered an abnormal value and is eliminated. The calculation formula of the Grubbs critical value G is as follows:

[0078]

[0079] Where M is the median of n groups of data.

[0080] Table 1 Grubbs critical value table

[0081]

[0082] In this embodiment, due to the limitation of actual production of steel plates, the collected data are all discrete and uneven, which may lead to incoherence between prediction rules when training the model. For the marginal data with less data volume, due to its low frequency of occurrence, the information learned during training is weak, which will also affect the accuracy of the model. Therefore, by statistically analyzing the frequency distribution of mechanical property values ​​in each interval, the mechanical property interval with the highest frequency can be selected. Based on this frequency, other data groups with lower frequencies can be copied through the program to increase the number of data, so as to ensure that all data can be evenly learned during model training, and finally obtain a high-quality data set.

[0083] S2, calculate physical metallurgical parameters based on the data in S1 and physical metallurgical theory;

[0084] S21. According to the mechanical properties calculation model, the influencing factors related to austenite grain growth during heating, recrystallization, grain growth and precipitation during rolling, and phase transformation behavior during cooling are analyzed. The mechanical properties calculation model is shown in Table 2.

[0085] Table 2 Mechanical properties calculation model and parameters

[0086]

[0087]

[0088] Where, YS is the yield strength (MPa), σ C is the solid solution strengthening strength (MPa), σ Grain is the fine grain strengthening strength (MPa), σ dis is the dislocation strengthening strength (MPa), Si, Cu, Ni, Cr, Mn, Mo, N and P are the mass fractions of the elements (wt%), d F is the final ferrite grain size (μm), ρ dis is the dislocation density (m2), d preis the average size of the precipitated phase particles (nm), f pre is the fraction of precipitated phase, v is the Poisson's ratio of 0.3, and b is the Burgers vector of 2.48×10 -10 m, μ is the shear modulus of 8×10 10 Pa. TS is the tensile strength (MPa), X F , X P and X B are the volume fractions of ferrite, pearlite and bainite, respectively, and H F , H P and H B are the microhardness (HV) of ferrite, pearlite and bainite, T mF and T mB are the phase equilibrium temperatures of ferrite and bainite respectively, EL is the elongation after fracture (%), and h is the thickness of the steel product (mm).

[0089] S22. Based on the data collected in S1, the physical metallurgical model is used to calculate the physical metallurgical parameters that affect the final microstructure characteristics and mechanical properties model, including precipitation fraction, ferrite fraction and size, pearlite fraction and bainite fraction, etc.

[0090] During the heating process, the initial austenite grains continue to grow at high temperatures, and the austenite grain size depends on the heating temperature and holding time. During the hot rolling process, if the accumulated strain ε a Exceeding the critical strain ε required for dynamic recrystallization (DRX) C , new DRX nuclei will be generated along the elongated grain boundaries and lead to metadynamic recrystallization (MDRX) in the subsequent pass interval. Otherwise, if ε C ≥ε a , static recrystallization (SRX) will occur in the interval between passes. Eventually, the deformed grains are completely or partially replaced by newly recrystallized grains. At the same time, alloying elements such as Nb in the steel undergo solid solution during reheating and strain-induced precipitation occurs during the subsequent hot rolling process. During cooling, austenite grains decompose into ferrite, pearlite, and bainite. The physical metallurgical model is shown in Table 3.

[0091] Table 3 Physical metallurgical model and parameters in hot rolling process

[0092]

[0093]

[0094] Where D is the austenite grain size during heating (μm), R is the gas constant 8.314 J / (mol·K), T heat is the average temperature during heating (℃), v heatis the heating rate during heating (℃ / s), ε C is the critical strain, D γ is the average size of austenite grains (μm), T def is the deformation temperature (℃), ε is the deformation rate (s-1), X RX is the dynamic recrystallization fraction X DRX Or static recrystallization fraction X SRX , w Mn 、w C and w Si etc. are the mass fraction of the element (wt%), ε * is the equivalent strain, [Nb] is the molar fraction of the alloying element, t 0.05 and t 0.95 is the precipitation start and end time (s), f P is the precipitation fraction, X F , X P and X B is the phase transformation fraction of ferrite, pearlite and bainite, d F is the ferrite grain size (μm), t is the time (s), CR is the cooling rate (℃ / s), C eq is the carbon equivalent; the parameters involved below the parameters in Table 3 are all adjustment coefficients.

[0095] S3. Use particle swarm optimization algorithm to optimize the parameters of the mechanical properties calculation model and construct a mechanical properties prediction model.

[0096] S31. Based on the calculated microstructure characteristic information and mechanical property calculation model parameters, the particle swarm optimization algorithm is used to optimize the mechanical property model parameters under different compositions and process conditions. The optimized parameters are shown in Table 2.

[0097] S32. Taking the component specifications and key hot rolling process parameters as input, the main input parameters selected are the mass fractions of chemical components such as C, Si, Mn, Nb, Ti, and production process parameters such as rolling thickness, steel-out temperature, intermediate billet thickness, finishing rolling temperature, cooling temperature and red-returning temperature, as shown in Table 4. The optimization parameters of the mechanical properties model are taken as output, and the training set and test set are divided. The random forest network (RF) is used to construct the mechanical properties prediction model.

[0098] Table 4 RF machine learning model parameters

[0099]

[0100] S33, predict the test set based on the mechanical property prediction model to obtain the predicted values ​​of the mechanical property parameters of the steel to be predicted, and use the mechanical property calculation model to calculate the mechanical properties, compare the calculated results with the measured mechanical properties, and use the root mean square error (RMSE) and determination coefficient (R 2 ), the formula is:

[0101]

[0102]

[0103] In the formula, n is the number of samples, i is the sample number, is the predicted value, is the measured value, It is the average of the measured values, and the performance prediction value and the performance measured value are judged as pass / fail respectively.

[0104] The prediction method of the present invention is adopted, AH36 steel is selected, and the data is divided into training data and test data at a ratio of 4:1, with 400 sets of training data and 100 sets of prediction data. Random forest (RF), support vector machine (SVM) and artificial neural network (ANN) are compared.

[0105] Comparison of predicted and measured values ​​of yield strength, tensile strength and elongation of AH36 steel based on RF, SVM and ANN algorithms Figure 2 As shown in the figure, the predicted values ​​of the model based on the random forest algorithm have the best consistency with the measured values, and the prediction accuracy is higher than that based on the SVM and ANN models.

[0106] The root mean square error of the machine learning model based on RF, SVM and ANN algorithms is as follows: Figure 2 As shown in the middle bar graph, the random forest algorithm has the lowest root mean square error in the three mechanical properties of service strength, tensile strength and elongation, indicating that the difference between the predicted value and the actual value of the random forest algorithm is small and the prediction effect is good.

[0107] A traditional data-driven machine learning model was established for comparison with the model of the present invention. The specific modeling method of the traditional data-driven machine learning model is to use a random forest model based on the physical metallurgical mechanism and correlation analysis results, select key components and process parameters that are strongly correlated with mechanical properties and have strong physical metallurgical regularity as input, and use mechanical properties as output to establish data-driven machine learning models for predicting yield strength, tensile strength and elongation. Finally, based on the same data set, the test results of the model developed by the present invention and the traditional data-driven machine learning model are compared.

[0108] Test results such as Figure 3 As shown, Figure 3a compares the RMSE of the predicted value and the measured value, and b compares the R 2 It can be seen that the RMSE of yield strength prediction has decreased from 18.81MPa to 14.91MPa, the RMSE of tensile strength has increased from 21.2MPa to 16.47MPa, and the RMSE of elongation has increased from 4.2% to 3.0%.

[0109] Therefore, the present invention adopts the above-mentioned physical metallurgy-guided hot-rolled steel mechanical properties machine learning method, which can establish a high-quality data set, and select the optimal algorithm modeling to predict the mechanical properties, thereby guiding steel plate production.

[0110] 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 machine learning method for mechanical properties of hot-rolled steel guided by physical metallurgy, characterized in that: Includes steps: S1, collecting hot-rolled steel plate data and preprocessing the data; S2. Calculate the physical metallurgical parameters based on the data in S1 and the physical metallurgical theory. The specific steps include: S21. According to the mechanical property calculation model, the influencing factors related to austenite grain growth during heating, recrystallization, grain growth and precipitation during rolling, and phase transformation behavior during cooling are analyzed. The mechanical property calculation model includes a yield strength model, a tensile strength model, and an elongation model, specifically: Yield Strength Model: YS=σ C +s Grain +s dis +s pre ; σ C =77+60Si+80Cu+45Ni+60Cr+80Mn+11Mo+5000N+750P; s Grain =K g d F -0.5 ; Tensile Strength Model: H F =458-0.357T mF +50Si; H B =669-0.588T mB +50Si; Elongation: Where, YS is the yield strength (MPa), σ C is the solid solution strengthening strength (MPa), σ Grain is the fine grain strengthening strength (MPa), σ dis is the dislocation strengthening strength (MPa), Si, Cu, Ni, Cr, Mn, Mo, N and P are the mass fractions of the elements (wt%), d F is the final ferrite grain size (μm), ρ dis is the dislocation density (m2), d pre is the average size of the precipitated phase particles (nm), f pre is the fraction of precipitated phase, v is the Poisson's ratio of 0.3, and b is the Burgers vector of 2.48×10 -10 m, μ is the shear modulus of 8×10 10 Pa. TS is the tensile strength (MPa), X F , X P and X B are the volume fractions of ferrite, pearlite and bainite, respectively, and H F , H P and H B are the microhardness (HV) of ferrite, pearlite and bainite, T mF and T mB are the phase equilibrium temperatures of ferrite and bainite, EL is the elongation after fracture (%), h is the thickness of the steel product (mm), k g , k d , k p ,f1,f2,f3,EL0,k F , k d , k h All are adjustment coefficients; S22, based on the data collected in S1, the physical metallurgical model is used to calculate the physical metallurgical parameters that affect the final microstructure characteristics and mechanical properties model; S3. Use particle swarm optimization algorithm to optimize the parameters of mechanical properties calculation model and construct mechanical properties prediction model. The specific steps include: S31, based on the calculated microstructure characteristic information and mechanical properties calculation model parameters, the mechanical properties calculation model parameters are optimized using a particle swarm optimization algorithm; S32, taking component specifications and hot rolling process parameters as input and mechanical property model optimization parameters as output, dividing the training set and the test set, and using a random forest network to construct a mechanical property prediction model; S33. Predict the test set based on the mechanical property prediction model to obtain the predicted values ​​of the mechanical property parameters of the steel to be predicted. At the same time, calculate the mechanical properties using the mechanical property calculation model, and compare the calculation results with the measured mechanical properties.

2. The method for machine learning of mechanical properties of hot-rolled steel guided by physical metallurgy according to claim 1, characterized in that: The data collected in step S1 include steel plate composition specification parameters, hot rolling process parameters and mechanical property data, and the data preprocessing includes: S11, hierarchically clustering the multidimensional data according to the parameter range, selecting the number of clusters, dividing the data into different clusters, and classifying the data; S12. According to the Grubbs criterion and the 3σ criterion, determine the values ​​with large detection errors in the classification data and eliminate outliers.

3. The method for machine learning of mechanical properties of hot-rolled steel guided by physical metallurgy according to claim 1, characterized in that: In step S22, the physical metallurgical parameters include precipitation fraction, ferrite fraction and size, pearlite fraction and bainite fraction; During the heating process, the austenite grains grow, and the austenite grain size depends on the heating temperature and the holding time; the accumulated strain ε during hot rolling a Exceeding the critical strain ε required for dynamic recrystallization DRX C , new DRX nuclei are generated along the elongated grain boundaries and lead to metadynamic recrystallization MDRX in the subsequent pass interval; otherwise, ε C ≥ε a Static recrystallization SRX occurs in the interval between passes, and the deformed grains are completely or partially replaced by new recrystallized grains; at the same time, the alloy elements dissolved in the heating process undergo strain induced precipitation during the hot rolling process, and the austenite grains decompose into ferrite, pearlite and bainite during the cooling process.

4. The method for machine learning of mechanical properties of hot-rolled steel guided by physical metallurgy according to claim 1, characterized in that: In step S31, a particle swarm optimization algorithm is used to optimize the mechanical properties model parameters under different compositions and process conditions.

5. The method for machine learning of mechanical properties of hot-rolled steel guided by physical metallurgy according to claim 1, characterized in that: In step S33, the calculated results are compared with the measured mechanical properties, and the accuracy is evaluated using the root mean square error RMSE and the determination coefficient R 2 The formula is: In the formula, n is the number of samples, i is the sample number, is the predicted value, is the measured value, It is the average of the measured values, and the performance prediction value and the performance measured value are judged as pass / fail respectively.

Images