A method for predicting high-temperature oxidation rate of hot-rolled steel

Abstract

The application discloses a kind of high-temperature oxidation rate prediction methods of hot-rolled steel, and the technology belongs to the field of rolling steel.Utilize high-temperature synchronous analyzer TGA, and the oxidation rate of different steel grades at different oxidation temperatures is determined by high-temperature oxidation experiment, and the basic oxidation data set of steel grade is established, which includes the chemical composition of each steel grade, oxidation temperature and oxidation rate.The data set is divided into training set and prediction set by using multiple leave-out method.Using BP neural network, a machine learning model of steel composition-oxidation temperature-oxidation rate is established based on the training data set, and the model accuracy is evaluated using the prediction set.This method greatly shortens the process development time, saves product development cost, and improves new product development efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of steel rolling, in particular to a method for predicting high-temperature oxidation rate of hot-rolled steel. BACKGROUND

[0002] In recent years, under the drive of technology introduction and innovation, the production technology of China's steel industry has developed rapidly, and new breakthroughs have been made in the production technology and performance of steel products. Specifically, the technical equipment of steel enterprises has reached a high level, the product quality has been greatly improved, and the product structure has become more reasonable. However, due to the fact that Chinese steel enterprises have long focused on the performance and shape control of hot-rolled products, they have long neglected the surface quality of hot-rolled products. Therefore, the surface quality of hot-rolled products has gradually become a shortcoming of China's hot-rolled products, and even seriously hinders the improvement of the grade and market competitiveness of hot-rolled products.

[0003] The thickness of the oxide scale directly affects the surface quality of the hot-rolled strip. For example, Chinese patent No. CN201910080963.X discloses a method for controlling the oxide scale of a high-strength steel with a tensile strength of 750 MPa that is free of pickling, which reduces the thickness of the oxide scale and the amount of oxide scale powder by adjusting the rolling process parameters. Chinese patent No. CN201210173143.3 discloses a method for producing a hot-rolled strip that is easy to pickle by thin slab continuous casting and rolling, which reduces the thickness of the oxide scale by controlling the cooling water opening system and laminar cooling system between each rack in the finishing rolling process, thereby realizing the production of a hot-rolled strip that is easy to pickle by thin slab continuous casting and rolling. Chinese patent No. CN201610567648.6 discloses a control method for eliminating the flower spot defect of a medium plate, which reduces the thickness of the oxide scale, improves the flatness of the oxide scale and interface, and effectively suppresses the occurrence rate of the flower spot defect by controlling the heating system, descaling system and temperature system. The above patents all point out that reducing the thickness of the oxide scale by trial and error is an important way to achieve the production of hot-rolled steel with high surface quality.

[0004] Since the thickness of the oxide scale in the hot rolling process is determined by the oxidation rate, predicting the oxidation rate and then predicting the thickness of the oxide scale can reduce the time cost of the trial and error method in the above patents, thereby providing a more efficient and convenient method for controlling the thickness of the oxide scale. SUMMARY

[0005] The purpose of the present application is to provide a method for predicting the high-temperature oxidation rate of hot-rolled steel, which establishes a BP neural network model between the chemical composition of the steel, the oxidation temperature and the oxidation rate, accurately predicts the oxidation rate of a specific steel at different rolling temperatures, and provides guidance for efficiently and conveniently achieving oxide scale thinning and the production of hot-rolled products with high surface quality.

[0006] To achieve the above object, the present application comprises the following steps:

[0007] Step S1. Establishing a high-temperature oxidation database of multiple steel grades at multiple temperatures;

[0008] Step S2. Dividing the data set into a training set and a prediction set according to a division ratio of 8:2;

[0009] Step S3. Establishing a BP neural network model of steel grade chemical composition-oxidation temperature-oxidation rate according to the training set of step S2;

[0010] Step S4. Verifying the calculation accuracy of BP and selecting the best neural network model using the prediction set.

[0011] Preferably, the steel grade in step S1 is one of IF steel, low-carbon microalloyed steel, C-Mn steel, high-carbon steel, Fe-Cr alloy and Fe-Si alloy, and the chemical composition of each steel grade is 0.002% to 0.810% of C, 0.01% to 2.2% of Si, 0.08% to 2.21% of Mn, 0 to 0.04% of Al, 0 to 1.5% of Cr, and the rest is Fe and impurities during smelting.

[0012] Preferably, the high-temperature oxidation database in step S1 comprises the following steps:

[0013] Step 1.1, obtain multiple strength levels of steel grades by changing the proportion of each element, and then smelt according to the element ratio to obtain multiple component ingots;

[0014] Step 1.2, heat the ingot to 1200℃ and keep for 1 hour, then hot roll, and sample from the surface of the rolled piece after cooling to room temperature;

[0015] Step 1.3, use a high-temperature simultaneous thermal analyzer to obtain the oxidation weight gain-time curve at multiple temperatures through high-temperature oxidation experiments, the isothermal oxidation temperature is set to 500-1300℃, the temperature interval is 50℃, and the isothermal time is set to 120min;

[0016] Step 1.4, according to the oxidation weight gain-time curve, use the oxidation kinetics model shown in formula (1) to fit the oxidation rate k P (T) at oxidation temperature T, formula (1) is as follows:

[0017] W 2 =k P (T)t(1)

[0018] In the formula, W is the oxidation weight gain, t is the oxidation time, k P(T) is the oxidation rate at oxidation temperature T, R is the gas constant;

[0019] Step 1.5, for a plurality of component systems of steel grades, repeating step 1.1 and step 1.4, obtaining the oxidation rate of a plurality of component systems of steel grades in a plurality of temperature ranges;

[0020] Step 1.6, collating the experimental results obtained in step 1.4 to establish a high-temperature oxidation database.

[0021] Preferably, the BP neural network model in the step S3 specifically comprises the following steps:

[0022] Step 3.1, normalizing the input features and output features to reduce the difference in the order of magnitude between the input features;

[0023] Step 3.2, the BP neural network comprises an input layer, a hidden layer and an output layer, and the basic unit is a neuron. The relationship between the input x and the output y of each neuron is:

[0024] y = f(w*x+b) (2)

[0025] Wherein, f is an activation function, w is a weight, and b is a threshold value;

[0026] Step 3.3, the loss function J between the oxidation rate calculated by the BP network and the measured oxidation rate is defined as,

[0027]

[0028] In the formula, n is the size of the training set, E i is the measured value of the oxidation rate of the i th data, P i is the oxidation rate prediction of the i th data;

[0029] Step 3.4, the number of input layer neurons of the BP neural network is set to 6, the number of output layer neurons is set to 1, and the number of intermediate layer neurons is set to 12;

[0030] Step 3.5, using Levenberg-Marquardt algorithm to optimize the weight w and the threshold value b, obtaining the best weight and threshold value group. In the neural network training process, the learning rate is set to 0.001, the Dropout is set to 0.2, and the early stopping strategy is used for training.

[0031] Preferably, the oxidation temperature range in the step S3 is 500-1300℃.

[0032] Preferably, the oxidation rate range in the step S3 is 0.003-287mg / cm 2 / min.

[0033] Preferably, the activation function between the input layer and the intermediate layer is selected as a Sigmoid function, and the activation function between the intermediate layer and the output layer is selected as a Relu function.

[0034] Therefore, the method for predicting the high-temperature oxidation rate of hot-rolled steel provided by the application can quickly calculate the oxidation rate at different oxidation temperatures, and then calculate the oxide scale thickness in the rolling process. In addition, the oxide scale structures under different process conditions can be compared through the calculation results, so as to obtain the optimal process, obtain the same results as the "experience-trial and error" method, and compared with the same, the method greatly shortens the process development time, saves the product development cost, and improves the new product development efficiency.

[0035] The technical solutions of the application will be further described in detail below with reference to the drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 is a high-temperature oxidation kinetics curve of IF steel selected by the method for predicting the high-temperature oxidation rate of hot-rolled steel;

[0037] Figure 2 is the fitting results of the oxidation rate of IF steel selected by the method for predicting the high-temperature oxidation rate of hot-rolled steel at 700 DEG C and 800 DEG C;

[0038] Figure 3 is the fitting results of the oxidation rate of IF steel selected by the method for predicting the high-temperature oxidation rate of hot-rolled steel at 900 DEG C and 1000 DEG C;

[0039] Figure 4 is the fitting results of the oxidation rate of IF steel selected by the method for predicting the high-temperature oxidation rate of hot-rolled steel at 1100 DEG C and 1200 DEG C;

[0040] Figure 5 is the relationship curve between the oxidation rate and the temperature of IF steel selected by the method for predicting the high-temperature oxidation rate of hot-rolled steel;

[0041] Figure 6 is the prediction accuracy of the BP neural network training set of the method for predicting the high-temperature oxidation rate of hot-rolled steel;

[0042] Figure 7 is the prediction accuracy of the BP neural network prediction set of the method for predicting the high-temperature oxidation rate of hot-rolled steel. DETAILED DESCRIPTION

[0043] The method for predicting the high-temperature oxidation rate of hot-rolled steel comprises the following steps:

[0044] Step S1. Establish a high-temperature oxidation database for multiple steel grades at multiple temperatures;

[0045] Step 1.1. Obtain multiple steel grades with different strength levels by changing the proportions of each element, and then melt the ingots according to the element ratio to obtain multiple component ingots;

[0046] Step 1.2. Heat the ingots to 1200℃ and keep for 1 hour, then hot rolling, and sample from the surface of the rolled piece after cooling to room temperature;

[0047] Step 1.3. Obtain the oxidation weight gain-time curve at multiple temperatures by high-temperature oxidation experiment using a high-temperature simultaneous thermal analyzer, with isothermal oxidation temperature set to 500-1300℃ and temperature interval of 50℃, and isothermal time set to 120min;

[0048] Step 1.4. According to the oxidation weight gain-time curve, use the oxidation kinetics model shown in formula (1) to fit the oxidation rate k P (T) at oxidation temperature T, formula (1) is as follows:

[0049] W 2 =k P (T)t(1)

[0050] In the formula, W is the oxidation weight gain, t is the oxidation time, k P (T) is the oxidation rate at oxidation temperature T, and R is the gas constant;

[0051] Step 1.5. Repeat steps 1.1 and 1.4 for multiple component systems to obtain the oxidation rate of multiple component systems at multiple temperature ranges;

[0052] Step 1.6. Organize the experimental results obtained in step 1.4 to establish a high-temperature oxidation database.

[0053] Step S2. Divide the data set into training set and prediction set according to the division ratio of 8:2;

[0054] Step S3. According to the training set of step S2, establish a BP neural network model of steel chemical composition-oxidation temperature-oxidation rate;

[0055] Step 3.1. Normalize the input features and output features to reduce the difference in the number of orders between the input features;

[0056] Step 3.2. The BP neural network includes input layer, hidden layer and output layer, and its basic unit is neuron. The relationship between the input x and the output y of each neuron is:

[0057] y=f(w*x+b)(2)

[0058] wherein f is an activation function, w is a weight, and b is a threshold value;

[0059] Step 3.3, the loss function J between the oxidation rate calculated by the BP network and the oxidation rate actually measured in the experiment is defined as,

[0060]

[0061] wherein n is the size of the training set, E i is the actual measured value of the oxidation rate of the ith data, P i is the prediction of the oxidation rate of the ith data;

[0062] Step 3.4, the number of input layer neurons of the BP neural network is set to 6, the number of output layer neurons is set to 1, and the number of intermediate layer neurons is set to 12.

[0063] Step 3.5, the Levenberg-Marquardt algorithm is used to optimize the weight w and the threshold value b to obtain the best weight and threshold value group. In the process of neural network training, the learning rate is set to 0.001, the Dropout is set to 0.2, and the early stopping strategy is used for training.

[0064] Step S4. The prediction accuracy of BP is verified by using the prediction set, and the best neural network model is selected.

[0065] wherein the steel type in step S1 is one of IF steel, low-carbon microalloyed steel, C-Mn steel, high-carbon steel, Fe-Cr alloy and Fe-Si alloy, and the chemical composition of each steel type is 0.002% to 0.810% of C, 0.01% to 2.2% of Si, 0.08% to 2.21% of Mn, 0 to 0.04% of Al, 0 to 1.5% of Cr, and the rest is Fe and impurities during smelting; the oxidation temperature range in step S3 is 500 to 1300℃; the oxidation rate range in step S3 is 0.003 to 287 mg / cm 2 / min; the activation function between the input layer and the intermediate layer is selected as a Sigmoid function, and the activation function between the intermediate layer and the output layer is selected as a Relu function.

[0066] Embodiment

[0067] In this example, IF steel is used, and the chemical composition is shown in Table 1,

[0068] Table 1 Chemical composition of IF steel (mass fraction, wt. %)

[0069]

[0070] The oxidation rate prediction method based on the BP neural network model comprises the following steps:

[0071] Step 1. Sample processing and treatment.

[0072] Step 1.1, according to the component proportion shown in Table 1, each element is added in the smelting furnace for smelting. After the smelting is completed, the ingot is placed in a heating furnace and heated to 1200℃ and kept for 1 hour, and then hot rolling is carried out, and the thickness of the slab after rolling is set to 2mm.

[0073] Step 1.2, using wire cutting, a sample with a size of 2mmx8mmx10mm is taken on the slab, and the sample surface is polished to smooth.

[0074] Step 2. Establish the high-temperature oxidation kinetics model of IF steel.

[0075] Step 2.1, using a high-temperature simultaneous thermal analyzer, the oxidation weight gain curve of IF steel at 700-1200℃ for 60min is measured by high-temperature oxidation experiment, as shown in Figure 1 , the horizontal axis is the oxidation time, and the vertical axis is the oxidation rate (i.e. oxidation weight gain).

[0076] Step 2.2, as shown in Figures 2-4 , the high-temperature oxidation rate of IF steel at different temperatures is obtained by formula (1) regression, the horizontal coordinate is time, and the vertical coordinate is oxidation weight square, W2. The oxidation rate at each temperature is shown in Table 2. Figure 5 , the relationship between lnKp and oxidation temperature is shown, the horizontal coordinate is the logarithm of oxidation rate, lnKp, and the vertical coordinate is the inverse of temperature, 1000 / T.

[0077] Table 2 High-temperature oxidation rate of IF steel at different temperatures

[0078]

[0079]

[0080] Step 3. For different component systems of steel, repeat step 1 to obtain the oxidation rate of different component systems of steel at different temperatures, and summarize the data and establish a high-temperature oxidation basic database at different temperatures of different steel grades.

[0081] Step 4. The data set is randomly divided into a training set and a prediction set according to a division ratio of 8:2.

[0082] Step 5. According to the training set of step 3, a BP neural network model of steel grade chemical composition-oxidation temperature-oxidation rate is established. The output of the model is the mass fraction of steel grade chemical composition (C, Si, Mn, Al and Cr) and oxidation temperature, and the output is the logarithm lnKp of oxidation rate Kp. The activation function between the input layer and the intermediate layer of the BP network is selected as Sigmoid function, and the activation function between the intermediate layer and the output layer is selected as Relu function. The Levenberg-Marquardt algorithm is used to optimize the weights and thresholds of the BP network, and the best weight and threshold set is obtained, so that the loss function J is minimized.

[0083] Step 6. The calculation accuracy of the BP neural network is verified by using the prediction set, and the best neural network model is selected. Figure 6 and Figure 7 The prediction results of the selected optimal BP neural network model for the oxidation rate of the training set and the test set are shown in the figure, in which the horizontal axis is the measured value of the oxidation rate (i.e. the experimental value of lnKp), and the vertical axis is the predicted value of the oxidation rate (i.e. the calculated value of lnKp). It can be seen that in the two data sets, the fitting degree between the measured value and the predicted value of the oxidation rate is very high, which indicates that the optimal neural network model has high calculation accuracy and good generalization ability.

[0084] Finally, it should be pointed out that: the above examples are only used to illustrate the technical solutions of the present application but not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that: the technical solutions of the present application can still be modified or replaced by equivalent, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. A method of predicting the high temperature oxidation rate of hot-rolled steel, characterized in that , comprising the following steps: Step S1. Establishing a high-temperature oxidation database of multiple steel grades at multiple temperatures, including the following steps: The steel grade in the step S1 is one of IF steel, low-carbon micro-alloyed steel, C-Mn steel, high-carbon steel, Fe-Cr alloy and Fe-Si alloy, and the chemical composition of each steel grade is 0.002% to 0.810% of C, 0.01% to 2.2% of Si, 0.08% to 2.21% of Mn, 0 to 0.04% of Al, 0 to 1.5% of Cr, and the rest is Fe and impurities during smelting; The high-temperature oxidation database in the step S1 includes the following steps: Step 1.1, obtaining multiple strength levels of steel grades by changing the proportion of each element, and then smelting according to the element ratio to obtain multiple component ingots; Step 1.2, heating the ingot to 1200℃ and keeping for 1 hour, then hot rolling, and sampling from the surface of the rolled piece after cooling to room temperature; Step 1.3, using a high-temperature simultaneous thermal analyzer, obtaining multiple temperature oxidation weight gain-time curves through high-temperature oxidation experiments, setting the isothermal oxidation temperature to 500-1300℃, and the temperature interval to 50℃, and setting the isothermal time to 120min; Step 1.

4. The oxidation rate k at oxidation temperature T is fitted according to the weight gain-time curve using the oxidation kinetic model shown in equation (1) P (T) is fitted according to the weight gain-time curve using the oxidation kinetic model shown in equation (1) k = k0exp(-Ea / RT) (1) W 2 = k P (T)t(1) where W is the weight gain, t is the oxidation time, k P (T) is the oxidation rate at temperature T, R is the gas constant; Step 1.5, repeating steps 1.1 and 1.4 for multiple component systems of steel grades to obtain the oxidation rate of multiple component systems of steel grades in multiple temperature ranges; Step 1.6, collating the experimental results obtained in step 1.4 to establish a high-temperature oxidation database; Step S2. Dividing the data set into a training set and a prediction set according to a division ratio of 8:2; Step S3. Establishing a BP neural network model of steel chemical composition-oxidation temperature-oxidation rate according to the training set of step S2; The BP neural network model in the step S3 specifically includes the following steps: Step 3.1, normalizing the input features and output features to reduce the difference in the number of orders between the input features; Step 3.2, the BP neural network includes an input layer, a hidden layer and an output layer, and the basic unit is a neuron, the relationship between the input x and the output y of each neuron is: y=f(w*x+b)(2) Where f is the activation function, w is the weight, and b is the threshold value; Step 3.3, the loss function J between the oxidation rate calculated by the BP network and the measured oxidation rate is defined as, where n is the size of the training set, E i is the measured oxidation rate of the i-th data, P i is the predicted oxidation rate of the i-th data; Step 3.4, the number of input layer neurons of the BP neural network is set to 6, the number of output layer neurons is set to 1, and the number of intermediate layer neurons is set to 12; Step 3.5, using Levenberg-Marquardt algorithm to optimize the weight w and the threshold value b to obtain the best weight and threshold value group, and setting the learning rate to 0.001, setting the Dropout to 0.2, and using the early stopping strategy for training during the neural network training process; Step S4. Using the prediction set to verify the calculation accuracy of BP and selecting the best neural network model.

2. The method of predicting the high temperature oxidation rate of a hot-rolled steel product according to claim 1, characterized in that: The oxidation temperature range in the step S3 is 500-1300℃.

3. The method of claim 1, wherein the method is characterized by: The oxidation rate in the step S3 ranges from 0.003 to 287 mg / cm 2 / min.

4. The method of claim 1, wherein the method is characterized by: The activation function between the input layer and the intermediate layer is a Sigmoid function, and the activation function between the intermediate layer and the output layer is a Relu function. The activation function between the input layer and the intermediate layer is a Sigmoid function, and the activation function between the intermediate layer and the output layer is a Relu function.

Images