Cold-rolled silicon steel same-plate difference prediction and process optimization method based on hot-rolled section characteristics

Abstract

This invention provides a method for predicting and optimizing the thickness difference between hot-rolled and cold-rolled silicon steel strips based on hot-rolled cross-sectional characteristics, belonging to the field of mechanical automation control technology. The method first obtains the process parameters for hot and cold continuous rolling of the strip. Based on the hot continuous rolling process parameters, a mathematical model of the hot-rolled cross-section is constructed to obtain the convexity values ​​of each marker point of the hot-rolled exit strip. Then, a prediction model for the thickness difference between the hot-rolled and cold-rolled cross-sections is trained based on the hot-rolled cross-section and cold-rolling process and cross-sectional data. Finally, the downstream stand roll shifting strategy is optimized based on the established hot-rolled cross-section mathematical model and the cold-rolled BP neural network thickness difference prediction model. This invention predicts the thickness difference of the cold-rolled finished product in a timely manner based on hot-rolling process parameters and strip dimensions, which can guide the timely adjustment of the rolling process parameters of the current batch of silicon steel products to meet the requirements for thickness difference in cold rolling.

Description

Technical Field

[0001] This invention relates to the field of mechanical automation control technology, and in particular to a method for predicting and optimizing the process of cold-rolled silicon steel sheet differences based on the characteristics of hot-rolled cross sections. Background Technology

[0002] The control precision requirements for the thickness variation of cold-rolled silicon steel strip are stringent. However, the configuration of cold rolling equipment varies among different manufacturers, resulting in inconsistent control capabilities for the thickness variation of the strip during the cold rolling process, with a general problem of insufficient strip shape control. Existing research has confirmed a significant correlation between the incoming strip shape of hot-rolled silicon steel and the strip shape of cold-rolled silicon steel. The control of the thickness variation of silicon steel strip needs to be extended from single cold rolling control to coordinated control of the entire hot-rolling and cold-rolling process. Therefore, establishing a full-process silicon steel cross-section prediction model is crucial.

[0003] Currently, mathematical models for hot-rolled sheet shape have achieved high accuracy after extensive research and are widely used in industrial production. However, factors such as friction coefficient and tension significantly affect sheet shape during cold rolling, and cold rolling requires sheet shape control accuracy down to the micrometer level. Traditional mechanism-based cold-rolled sheet shape models are ill-suited to the needs of predicting the same-sheet difference in silicon steel throughout the entire process, necessitating the adoption of data-driven models for accurate prediction. Controlling the same-sheet difference in silicon steel involves multiple processes, resulting in a long process chain and significant fluctuations in process parameters. If each process independently controls its own dimensional targets according to predetermined process regulations, it is easy for individual processes to meet targets while the final finished product's same-sheet difference exceeds the standard. Furthermore, the hot-rolling process has highly variable parameters, and its sheet shape control volatility is higher than that of the cold-rolling process, further exacerbating the difficulty of overall process control.

[0004] Therefore, it is urgent to construct a full-process same-plate difference prediction model that spans hot rolling and cold rolling. Based on hot rolling process parameters and strip dimensions, the same-plate difference of cold-rolled finished products can be predicted in advance, providing guidance for adjusting hot rolling process parameters. At the same time, it can realize the grading treatment of hot-rolled strip steel and the adaptation of subsequent processes, avoid a large number of cold-rolled finished products exceeding the same-plate difference standard due to fluctuations in the previous process, reduce economic losses, and meet the needs of flexible production. Summary of the Invention

[0005] To address the aforementioned technical problems in existing technologies, this invention provides a method for predicting and optimizing the same-plate difference in cold-rolled silicon steel based on hot-rolled cross-sectional characteristics. The technical solution is as follows:

[0006] A method for predicting and optimizing the process differences of cold-rolled silicon steel within the same plate based on hot-rolled cross-sectional characteristics, the method comprising: S1. Obtain the process parameters for hot rolling and cold rolling of strip steel; S2. Based on the hot rolling process parameters obtained in S1, construct a mathematical model of the hot-rolled section and obtain the convexity values ​​of each marker point of the hot-rolled exit strip. S3. Based on the mathematical model of the hot-rolled section of the strip constructed in S2 and the cold-rolling process parameters obtained in S1, a prediction model for the thickness difference between the hot-rolled section and the cold-rolled section is trained, wherein the prediction model adopts a BP neural network model. S4. Based on the mathematical model of the hot-rolled section established in S2 and the prediction model of the thickness difference between the hot-rolled section and the cold-rolled section established in S3, the downstream stand roll shifting strategy is optimized.

[0007] The mathematical model construction process for the hot-rolled section in S2 is as follows: S21. Determine the exit convexity of the frame: The crown genetic model characterizes the relationship between the incoming crown and the standard crown of the mill roll gap on the exit crown, which can be expressed as:

[0008] in, j Rack number; For the first j The exit convexity of the rack, in μm; For the first j The convexity of the rack entrance is obtained at the first level of the communication site, and the unit is μm; For the first j The entrance crown coefficient of the stand is dimensionless and is obtained through rolling experience; The first j The thickness of the rack's exit and entrance is in mm, obtained at the communication site level. For the first j Standard convexity of the frame; k The locations of the marking points on the strip section are 25, 40, 50, and 100. By combining the results, the convexity values ​​at each marked point on the cross-section of the F7 frame outlet strip can be obtained.

[0009] S22. Determine the cross-sectional profile of the strip:

[0010] Where prof represents the strip cross-sectional profile, the convexity value at any point along the strip width can be calculated, and the convexity of the strip exiting the finishing mill is... Discretize the strip cross-section along the width direction into m The convexity at each position is [value missing]. ; S23. Determine the strip cross-sectional model: In continuous rolling, the finishing mill exit target can be represented by the roll gap crown model and the genetic model, with parameters such as rolling force, roll bending force, and roll shape as inputs. The crown value at the strip marker point can be represented by the genetic coefficients of each stand and known parameters.

[0011] in, For the first kn The i-th fitting coefficient of the polynomial corresponding to a strip cross-sectional feature (convexity) is the polynomial coefficient used to quantify the shape of the strip cross-section; for example, K1 represents the K1-th convexity, and i represents the i-th coefficient term of the polynomial; i=1,2,3,...,j, n=1,2,3,...,m.

[0012] After establishing the strip cross-section model, the strip cross-section data from the industrial IMS instrument are extracted to solve for the unknown coefficients at each location, thus obtaining the mathematical model of the hot-rolled cross-section. Based on this model, when certain process parameters change during the finishing rolling process, the convexity values ​​of each marker point on the finishing rolling exit section can be directly obtained.

[0013] In S21, the first j The standard crown of the stand is expressed as a relationship between rolling process parameters, and the formula is:

[0014] in, The first j The rolling force influence coefficient and bending roll force influence coefficient of the stand were obtained by commercial finite element model and corrected by actual production data; The first j The combined crowning influence coefficient of the work roll and support roll of the frame under no-load conditions was obtained from a commercial finite element model and corrected using actual production data. These are the rolling force and bending roll force of the j-th stand, respectively, acquired at the first level of the communication site; The first j The overall convexity of the work rollers and support rollers of the frame under no-load conditions is obtained at the first-level communication site.

[0015] The BP neural network model in S3 is a single-hidden-layer, three-layer feedforward network, including an input layer, hidden layers, and an output layer. The parameter determination and construction process of the BP neural network model is as follows: The parameters of the BP neural network are determined based on the number of nodes in the input layer, the number of nodes in the output layer, the number of network layers, the number of hidden layer nodes, the activation function, the learning rate, the training function, and the range of initial weight values. The range of hidden layer node numbers is calculated using a formula, and the specific number of nodes is determined based on the minimum network error. The optimal training function is selected by comparing the training error and training steps of different training functions. The activation function is determined by combining the nonlinear demand predicted by the difference in cold-rolled sheet thickness. To balance training time and accuracy, determine the learning rate; The initial weights are randomly selected within [0,1] to avoid local optima, resulting in a BP neural network model with complete parameter configuration. This BP neural network model with complete parameter configuration is then used as a prediction model for the thickness difference between the hot-rolled and cold-rolled sections.

[0016] The formula for calculating the range of hidden layer nodes is:

[0017] In the formula, l This represents the number of hidden layer nodes. n The number of input units, m The number of output units. a It takes a value between 1 and 10; By combining the formula and substituting the number of inputs and outputs, we can obtain that the number of hidden layer neurons is 4 to 13. Then, by using the minimum network error as the criterion, we determine the number of nodes.

[0018] The input variables of the BP neural network are independent of each other, but are related according to the parameters of the same-plate difference prediction model, totaling 12 input variables. The output is only the same-plate difference in cold rolling, and the number of output layers is 1. A three-layer neural network can effectively solve nonlinear problems. A single-hidden-layer error backpropagation neural network can learn and train continuous functions well in any closed interval. Increasing the number of hidden layers can improve the model's accuracy, but it also increases the complexity of the model structure and prolongs the training cycle. Considering the requirements for establishing the same-plate difference prediction model in cold rolling, the number of hidden layers is 1. Commonly used transfer functions for hidden layers are: tan-sigmoid function, log-sigmoid function, and purelin function. For nonlinear problems, the input layer is usually a nonlinear function, and the output layer is a linear function. For highly nonlinear same-plate difference prediction in cold rolling, both the input and hidden layers use the tan-sigmoid function, and the output layer uses the purelin function. During network training, the adjustment of weights and thresholds directly determines the model's training rate and accuracy. To balance training time and accuracy, a learning rate of 0.01 to 0.8 is typically chosen; here, 0.05 is selected. Commonly used backpropagation (BP) algorithms for optimizing training functions include gradient descent, nonlinear least squares (Levenberg-Marquardt) descent, and adaptive Lr (Lagrangian relaxation) gradient descent. In this invention, the optimal training function is determined by the network model's training error and the number of training steps.

[0019] The process of training the prediction model for the thickness difference between hot-rolled and cold-rolled sections in S3 is as follows: n sets of hot-rolled strip cross-sectional data and cold-rolled process parameters are selected (each set contains corresponding hot-rolled strip cross-sectional data and cold-rolled process parameters), where n0 sets are used as the training set and n1 sets are used as the test set, n0+n1=n; the hot-rolled strip cross-sectional data are obtained through a hot-rolled cross-sectional mathematical model. Using hot-rolled strip cross-sectional data and cold-rolled process parameters as inputs and cold-rolled plate differences as outputs, the normalized training set data is input into the Matlab workspace in vector form to train the prediction model.

[0020] The optimization process for the downstream stand roll shifting strategy in S4 is as follows: Input the hot-rolled section mathematical model established in S2 and the prediction model of the thickness difference between the hot-rolled section and the cold-rolled section established in S3 into the hot-rolled section historical process parameters and the current adjustment parameters obtained on site, and predict the impact of the current adjustment parameters on the thickness difference of the cold-rolled section; The optimization of the roll shifting strategy takes roll wear value as the optimization target, while also taking the cold-rolled same-plate difference index as a new optimization reference target. With roll wear value as the optimization target, the optimized roll shifting strategy can reduce the average value and distribution range of cold-rolled sheet difference, and improve the stability of cold-rolled sheet difference.

[0021] The current adjustment parameters include the adjustment parameters of the downstream frame work roll shape. The adjustment of the work roll shape will cause a change in the convexity of the loaded roll gap, which will correspondingly drive a change in the bending roll force if other process and equipment parameters remain unchanged.

[0022] This invention establishes a hot-rolled cross-section calculation model to construct the correspondence between hot-rolling process parameters and the dimensions of the finishing roll exit cross-section. Simultaneously, it builds a cold-rolled strip difference prediction model based on a BP neural network, combining the hot-rolling mechanism model with the cold-rolling data-driven model to form a strip difference prediction model spanning the entire rolling process. This achieves a direct correlation mapping between upstream and downstream process parameters and cold-rolled strip differences. This model provides accurate references for adjusting hot-rolling process parameters and can predict the strip difference level of finished strips in a timely manner based on the shape of the hot-rolled strip, thereby allowing for targeted adjustments to subsequent process procedures, effectively ensuring product quality and meeting the practical needs of flexible production.

[0023] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In the above scheme, the constructed hot-rolled strip cross-section model and the cold-rolled BP neural network same-plate difference prediction model can integrate historical hot-rolling process parameters with current adjustment parameters to accurately predict the impact of parameter changes on cold-rolled same-plate difference, providing a scientific and direct guidance for hot-rolling process adjustments. This breaks through the limitations of previous methods that only used roll wear as a single optimization target, incorporating the cold-rolled same-plate difference index into the optimization system, providing a new core reference target, and achieving a multi-dimensional balance in process optimization. Validation based on historical process data from low-grade silicon steel plans shows that the model can clearly predict changes in cold-rolled same-plate difference before and after process adjustments, significantly improving the overall trend of same-plate difference, narrowing the mean and distribution range, and greatly enhancing the control accuracy and stability of cold-rolled same-plate difference, thus providing a solid guarantee for product quality. Meanwhile, this technical solution can flexibly adapt to production needs based on the prediction results. It can guide the optimization of hot rolling process parameters for this batch of products through prediction, and provide a basis for adjusting the process system of subsequent processes. It can effectively avoid the problem of excessive differences in the same plate of finished products caused by the fluctuation of the previous process, reduce economic losses, and take into account both product quality and production flexibility. It has extremely strong industrial application value. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a method for predicting and optimizing the same-plate difference of cold-rolled silicon steel based on the characteristics of hot-rolled cross sections, provided by an embodiment of the present invention. Figure 2 This is a comparison chart of the predicted and actual values ​​of the cold-rolled silicon steel strip with the same plate difference model in an embodiment of the present invention; Figure 3 This is a distribution diagram of the prediction error of the same plate difference model for cold-rolled silicon steel strip in an embodiment of the present invention; Figure 4 This is a comparison of the work roll curves before and after process adjustment in an embodiment of the present invention; Figure 5 This is a prediction of the difference between the same board before and after process adjustment in this embodiment of the invention. Detailed Implementation

[0026] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0027] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0028] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0029] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0030] This invention provides a method for predicting and optimizing the process of cold-rolled silicon steel sheet defects based on hot-rolled cross-sectional characteristics. For example... Figure 1 The flowchart shown illustrates a method for predicting and optimizing the same-plate difference in cold-rolled silicon steel based on hot-rolled cross-sectional characteristics. This method may include the following steps:

[0031] S1. Obtain the process parameters for hot rolling and cold rolling of strip steel; S2. Based on the hot rolling process parameters obtained in S1, construct a mathematical model of the hot-rolled section and obtain the convexity values ​​of each marker point of the hot-rolled exit strip. S3. Based on the mathematical model of the hot-rolled section of the strip constructed in S2 and the cold-rolling process parameters obtained in S1, a prediction model for the thickness difference between the hot-rolled section and the cold-rolled section is trained, wherein the prediction model for the thickness difference of the same plate in cold rolling adopts a BP neural network model. S4. Based on the mathematical model of the hot-rolled section established in S2 and the prediction model of the thickness difference between the hot-rolled section and the cold-rolled section established in S3, the downstream stand roll shifting strategy is optimized.

[0032] The following description, in conjunction with specific embodiments, illustrates this point.

[0033] In the specific case, the selected data is 50SW1300 low-grade non-oriented silicon steel, with a finished product width of 1200mm and a finished product thickness of 0.5mm. The hot-rolled incoming material width is 1250mm and the hot-rolled incoming material thickness is 2.5mm.

[0034] The following procedure is used to predict board-to-board differences and optimize the process.

[0035] S1. Obtain the process parameters for hot and cold strip rolling processes; Specifically, in this embodiment, the production data of the hot strip rolling process includes: the entry and exit thicknesses (in mm) of each of the seven stands (F1-F7), obtained through on-site primary communication; the rolling force and bending force (in kN) of each stand, obtained through on-site primary communication; the comprehensive crown of the work rolls and support rolls of each stand under no-load conditions (in μm), obtained through on-site primary communication; the roll shape parameters of the work rolls (including roll shape curve characteristics, taper control section parameters, etc.), collected through equipment testing and process records; the measured data of the cross-sectional profile of the intermediate billet after rough rolling, obtained through manual sampling measurement; and the strip cross-sectional data at the exit of the finishing mill, collected by the industrial site IMS multi-functional integrated measurement instrument, including the thickness coordinates of discrete points in the strip width direction and the crown values ​​of each marker point (C25, C40, C50, C100, i.e., the crown at 25mm, 40mm, 50mm, and 100mm from the edge of the strip, in μm). S2. Based on the hot rolling process parameters obtained in S1, construct a mathematical model of the hot-rolled section and obtain the convexity values ​​of each marker point of the hot-rolled exit strip. The process of constructing the mathematical model of the hot-rolled section is as follows: S21. Determine the exit convexity of the frame: The crown genetic model characterizes the relationship between the incoming crown and the standard crown of the mill roll gap on the exit crown, which can be expressed as:

[0036] in, j Rack number; For the first j The exit convexity of the rack, in μm; For the first j The convexity of the rack entrance is obtained at the first level of the communication site, and the unit is μm; For the first j The entrance crown coefficient of the stand is dimensionless and is obtained through rolling experience; The first j The thickness of the rack's exit and entrance is in mm, obtained at the communication site level. For the first j The standard crown of the stand is expressed as a formula relating the rolling process parameters:

[0037] in, For the first j The rolling force influence coefficient and bending roll force influence coefficient of the stand were obtained by commercial finite element model and corrected by actual production data; For the first jThe combined crowning influence coefficient of the work roll and support roll of the frame under no-load conditions was obtained from a commercial finite element model and corrected using actual production data. The rolling force and bending force of the j-th stand are obtained at the first-level communication site. For the first j The overall convexity of the work rollers and support rollers of the frame under no-load conditions is obtained at the first-level communication site.

[0038] By combining the equations, the convexity values ​​at each marked point on the strip cross-section at the F7 frame exit can be obtained. The convexity value at any point along the strip width can be calculated.

[0039] S22. Determine the cross-sectional profile of the strip:

[0040] Where, prof represents the strip cross-sectional profile, and the crown of the strip exiting the finishing mill is... Discretize the strip cross-section along the width direction into m The convexity at each position is [value missing]. ; S23. Determine the strip cross-sectional model: In continuous rolling, the finishing mill exit target can be represented by the roll gap crown model and the genetic model, with parameters such as rolling force, roll bending force, and roll shape as inputs. The crown value at the strip marker point can be represented by the genetic coefficients of each stand and known parameters.

[0041] in, For the first kn The i-th fitting coefficient of the polynomial corresponding to each convexity, i=1,2,3,...,j, n=1,2,3,...,m.

[0042] After establishing the strip cross-section model, the strip cross-section data from the industrial IMS instrument are extracted to solve for the unknown coefficients at each location, thus obtaining the mathematical model of the hot-rolled cross-section. Based on this model, when certain process parameters change during the finishing rolling process, the convexity values ​​of each marker point on the finishing rolling exit section can be directly obtained.

[0043] S3. Based on the mathematical model of the hot-rolled section of the strip constructed in S2 and the cold-rolling process parameters obtained in S1, a prediction model for the thickness difference between the hot-rolled section and the cold-rolled section is trained, wherein the prediction model for the thickness difference of the same plate in cold rolling adopts a BP neural network model. The BP neural network model is a single-hidden-layer, three-layer feedforward network, including an input layer, hidden layers, and an output layer. The parameter determination and construction process of the BP neural network model is as follows: The parameters are determined based on the number of nodes in the input layer, the number of nodes in the output layer, the number of network layers, the number of nodes in the hidden layer, the activation function, the learning rate, the training function, and the range of initial weight values. The network has 3 layers (1 input layer, 1 hidden layer, and 1 output layer). The single hidden layer structure can effectively solve nonlinear problems, balance model accuracy and training efficiency, and avoid excessive number of hidden layers, which would lead to complex structure and prolonged training cycle. The number of input layer nodes is set to 12, and the corresponding input parameters are the convexity values ​​of the four marker points of the hot-rolled section (C25, C40, C50, C100), trimming amount, roll insertion amount, cone height and rolling pass reduction rate. Each input variable is independent of each other and strongly correlated with the same plate difference prediction model. The number of output layer nodes is set to 1, and only the core indicator of cold-rolled same-plate difference is output.

[0044] The number of hidden layer nodes ranges from 4 to 13. Using the minimum network training error as the criterion, the optimal number of hidden layer nodes is determined to be 11 by comparing the training errors corresponding to different numbers of nodes.

[0045] In light of the highly nonlinear requirements for predicting the difference between cold-rolled sheets, the tan-sigmoid (Tansig) function is used as the activation function for both the input layer and the hidden layer, while the purelin linear function is used for the output layer, to adapt to the requirements of nonlinear mapping and final linear output. By comparing the training error and training steps of different training functions, the Levenberg-Marquardt (LM) method was selected as the training function (Trainlm), with a training error of only 0.00094537 and a training step count of 1673.

[0046] The training structures of the BP network under different training functions are shown in Table 1.

[0047] Table 1

[0048] The learning rate is set to 0.05, balancing model training speed and prediction accuracy, and is suitable for the common value range of 0.01 to 0.8. The initial weights are randomly selected within [0,1] to avoid the model getting trapped in local optima due to excessively large initial values, thus ensuring that the model error is small after training. The maximum number of iterations was set to 2500, the expected error was 1e-3, the Zscore function was used for data normalization, and the LM algorithm was used for the learning rule.

[0049] Nine hundred sets of hot-rolled cross-sections and cold-rolling process and cross-section data of strip silicon steel were selected as training samples. The data objects were low-grade non-oriented silicon steel of grade 50SW1300, corresponding to a finished product width of 1200mm and a finished product thickness of 0.5mm, and a hot-rolled incoming material width of 1250mm and a hot-rolled incoming material thickness of 2.5mm. In order to eliminate the interference of hot-rolled wedges on the prediction results, data with hot-rolled wedges smaller than 3μm were selected for training and prediction, of which 700 sets were used as the training set and 200 sets were used as the test set.

[0050] Using the cross-sectional parameters of hot-rolled strip steel and key process parameters of cold rolling (including roll shape parameters, trimming amount, reduction rate, etc.) as inputs and the difference between the same plate and the cold-rolled strip steel as outputs, the normalized training set data is input into the Matlab workspace in vector form. The weights and thresholds of the hidden layers are adjusted through iterative calculations until the expected error requirements are met, thus completing the training of the BP neural network model.

[0051] The model accuracy was evaluated using the reliable interval method, with an allowable error of 2 μm. On 200 sets of test data, the model was used to predict the thickness difference between hot-rolled and cold-rolled sections. The model accuracy reached 75.5%, and the offline training accuracy meets the requirements of industrial applications. It can effectively achieve accurate prediction of the thickness difference between hot-rolled and cold-rolled sections. The prediction results and error distribution are shown in the figure below. Figure 2 , Figure 3 As shown.

[0052] S4. Based on the mathematical model of the hot-rolled section established in S2 and the prediction model of the thickness difference between the hot-rolled section and the cold-rolled section established in S3, the downstream stand roll shifting strategy is optimized.

[0053] The process of optimizing the downstream stand roll shifting strategy is as follows: Input historical hot rolling process parameters and current adjustment parameters into the hot-rolled section mathematical model established in S2 and the prediction model of the thickness difference between the hot-rolled section and the cold-rolled section established in S3, and predict the impact of the current adjustment parameters on the thickness difference of the cold-rolled section. Break through the limitation of the traditional roll shifting strategy which only uses roll wear value as the single optimization target, incorporate the thickness difference index of the cold-rolled section into the optimization system as a new core reference target, realize the multi-dimensional balance of process optimization, and avoid the problem of excessive thickness difference of the final product caused by single-objective optimization.

[0054] The current adjustment parameters include the downstream frame work roll shape adjustment parameters. This work roll shape adjustment causes a change in the convexity of the loaded roll gap, which, with other process and equipment parameters remaining unchanged, correspondingly leads to a change in the bending roll force. The specific optimization and verification process is as follows:

[0055] Taking the downstream F6 stand of hot rolling mill as the optimization target, such as Figure 4As shown, the work roll shape was optimized from the original curve A to curve B. The transmission and changes in the crown of each stand after the roll shape adjustment were captured using a hot-rolled strip cross-section model. Combined with a cold-rolled BP neural network model for predicting the same-plate difference, and inputting the hot-rolling process parameters before and after optimization, as well as historical process data, the trend of the same-plate difference in the cold-rolled finished product was predicted. Historical process data from a low-grade 50SW800 silicon steel production plan was selected as a verification sample to fully reproduce the parameter iteration process from process A (original roll shape) to process B (optimized roll shape).

[0056] The prediction model outputs data on the difference in cold-rolled sheet thickness before and after process adjustment, as shown in the following figures. Figure 5 As shown, the overall trend of the cold-rolled sheet difference is significantly improved after optimization, with both the mean and distribution range narrowing considerably, and the stability of the sheet difference is greatly enhanced. This verifies the positive effect of the optimized roll shifting strategy on improving the cold-rolled sheet difference. Furthermore, this optimization logic can be extended to single-parameter or multi-parameter adjustment scenarios such as downstream stand work roll shape adjustment and multi-stand load distribution adjustment, all of which can predict changes in the cold-rolled finished sheet difference through the model.

[0057] This optimization scheme allows for the prediction of the same-plate difference in cold-rolled finished products based on process parameters and strip dimensions during the hot rolling process. This not only provides timely guidance for adjusting the hot rolling process parameters of the current batch of silicon steel products to ensure that the same-plate difference requirements for cold rolling are met, but also enables the grading of already produced hot-rolled silicon steel strips, allowing for targeted adjustments to subsequent process procedures and product customer matching. This avoids the problem of excessive same-plate difference in cold-rolled finished products caused by fluctuations in the previous process, reducing economic losses and balancing product quality with flexible production needs.

[0058] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting and optimizing the process of cold-rolled silicon steel sheet defects based on hot-rolled cross-sectional characteristics, characterized in that, The method includes: S1. Obtain the process parameters for hot rolling and cold rolling of strip steel; S2. Based on the hot rolling process parameters obtained in S1, construct a mathematical model of the hot-rolled section and obtain the convexity values ​​of each marker point of the hot-rolled exit strip. S3. Based on the mathematical model of the hot-rolled section of the strip constructed in S2 and the cold-rolling process parameters obtained in S1, a prediction model for the thickness difference between the hot-rolled section and the cold-rolled section is trained, wherein the prediction model adopts a BP neural network model. S4. Based on the mathematical model of the hot-rolled section established in S2 and the prediction model of the thickness difference between the hot-rolled section and the cold-rolled section established in S3, the downstream stand roll shifting strategy is optimized.

2. The method for predicting and optimizing the same-plate difference of cold-rolled silicon steel based on hot-rolled cross-sectional characteristics according to claim 1, characterized in that, The mathematical model construction process for the hot-rolled section in S2 is as follows: S21. Determine the exit convexity of the frame: ； in, j Rack number; For the first j The exit convexity of the rack, in μm; For the first j The convexity of the rack entrance is obtained at the first level of the communication site, and the unit is μm; For the first j The entrance crown coefficient of the stand is dimensionless and is obtained through rolling experience; The first j The thickness of the rack's exit and entrance is in mm, obtained at the communication site level. For the first j Standard convexity of the frame; k The locations of the marking points on the strip section are 25, 40, 50, and 100. S22. Determine the cross-sectional profile of the strip: ； Where, prof represents the strip cross-sectional profile, and the crown of the strip exiting the finishing mill is... Discretize the strip cross-section along the width direction into m The convexity at each position is [value missing]. ; S23. Determine the strip cross-sectional model: ； in, For the first kn The i-th fitting coefficient of the polynomial corresponding to each convexity, i=1,2,3,...,j, n=1,2,3,...,m.

3. The method for predicting and optimizing the same-plate difference of cold-rolled silicon steel based on hot-rolled cross-sectional characteristics according to claim 2, characterized in that, In S21, the first j The standard convexity formula for the frame is: ； in, The first j The rolling force influence coefficient and the bending roll force influence coefficient of the stand; The first j The combined convexity influence coefficient of the work rolls and support rolls of the frame under no-load conditions; These are the rolling force and bending roll force of the j-th stand, respectively; The first j The overall convexity of the work rolls and support rolls of the frame under no-load conditions.

4. The method for predicting and optimizing the same-plate difference of cold-rolled silicon steel based on hot-rolled cross-sectional characteristics according to claim 1, characterized in that, The BP neural network model in S3 is a single-hidden-layer, three-layer feedforward network, including an input layer, hidden layers, and an output layer. The parameter determination and construction process of the BP neural network model is as follows: The parameters are determined based on the number of nodes in the input layer, the number of nodes in the output layer, the number of network layers, the number of nodes in the hidden layer, the activation function, the learning rate, the training function, and the range of initial weight values. The range of hidden layer node numbers is calculated using a formula, and the specific number of nodes is determined based on the minimum network error. The optimal training function is selected by comparing the training error and training steps of different training functions. The activation function is determined by combining the nonlinear demand predicted by the difference in cold-rolled sheet thickness. To balance training time and accuracy, determine the learning rate; The initial weights are randomly selected within [0,1] to avoid local optima, resulting in a BP neural network model with complete parameter configuration. This BP neural network model with complete parameter configuration is then used as a prediction model for the thickness difference between the hot-rolled and cold-rolled sections.

5. The method for predicting and optimizing the same-plate difference of cold-rolled silicon steel based on hot-rolled cross-sectional characteristics according to claim 1, characterized in that, The process of training the prediction model for the thickness difference between hot-rolled and cold-rolled sections in S3 is as follows: n sets of hot-rolled strip cross-sectional data and cold-rolling process parameters are selected, where n0 sets are used as the training set and n1 sets are used as the test set, and n0+n1=n; the hot-rolled strip cross-sectional data are obtained through a hot-rolled cross-sectional mathematical model. Using hot-rolled strip cross-sectional data and cold-rolled process parameters as inputs and cold-rolled plate differences as outputs, the normalized training set data is input into the Matlab workspace in vector form to train the prediction model.

6. The method for predicting and optimizing the same-plate difference of cold-rolled silicon steel based on hot-rolled cross-sectional characteristics according to claim 1, characterized in that, The optimization process for the downstream stand roll shifting strategy in S4 is as follows: Input the hot-rolled section mathematical model established in S2 and the prediction model of the thickness difference between the hot-rolled section and the cold-rolled section established in S3 into the hot-rolled section historical process parameters and the current adjustment parameters obtained on site, and predict the impact of the current adjustment parameters on the thickness difference of the cold-rolled section; With roll wear value as the optimization target, the optimized roll shifting strategy can reduce the average value and distribution range of cold-rolled sheet difference, and improve the stability of cold-rolled sheet difference.

7. The method for predicting and optimizing the same-plate difference of cold-rolled silicon steel based on hot-rolled cross-sectional characteristics according to claim 6, characterized in that, The current adjustment parameters include the adjustment parameters of the downstream frame work roll shape. The adjustment of the work roll shape will cause a change in the convexity of the loaded roll gap, which will correspondingly drive a change in the bending roll force if other process and equipment parameters remain unchanged.

Images