JP7420157B2 - Rolling control device and rolling control method - Google Patents

Description

The present invention relates to a rolling control device and a rolling control method that improve dimensional accuracy such as plate thickness and plate width when manufacturing steel plates by plate rolling.

When rolling a steel plate (rolled material) using a rolling mill, control is performed to bring the plate thickness and plate width after rolling closer to target values and reduce deviations from the target values. For example, control of plate thickness Automatic thickness control (AGC) is used for this purpose. This plate thickness control method mainly involves setting control that predicts the rolling load and sets the initial rolling position, and predicts the plate thickness during rolling using actual measurements or a model to ensure that the deviation from the target value is zero. One example is dynamic control that changes the rolling position so that the The same applies to plate width control.

As a technique for predicting the rolling load in setting control, for example, a rolling load calculation model based on the plate rolling theory described in Non-Patent Document 1 is incorporated into a process computer, and the rolling load of each pass is calculated according to the rolling schedule. Techniques for calculating this are known.

On the other hand, as an example of dynamic control, Patent Document 1 discloses that in a reversible rolling mill, in order to accurately predict the rolling load in a short time and improve the accuracy of plate thickness control, A rolling control technology is disclosed that obtains the actual value of an operating factor, replaces it with the set value of the operating factor included in the target rolling material data, and predicts the rolling load of the next pass based on the calculated influence coefficient. .

Further, Patent Document 2 discloses a thick plate rolling method in which during the rolling pass, actual values for each pass are measured and adaptive control is performed using learning calculations to optimize settings for the next pass.

Japanese Patent Application Publication No. 2015-66569 Japanese Patent Application Publication No. 07-60320

Edited by the Iron and Steel Institute of Japan, Theory and Practice of Plate Rolling (Special Report No. 36), 1982

However, the above-mentioned conventional technology has the following problems that still need to be solved.
As shown in Patent Document 1, in the conventional calculation method, rolling was performed by calculating a predictive calculation model based on rolling theory in a rolling process computer and applying it to the rolling schedule. There was a problem that the accuracy could not be improved due to the influence.

In addition, with the technology described in Patent Document 2, even if calculations are performed using machine learning, the calculations take a long time and cannot be completed within the calculation time allowed between rolling passes. There were problems.

The present invention has been made in view of the above circumstances, and its purpose is to provide a rolling control device and a rolling control method that complete calculations for rolling control within the allowable calculation time between passes during rolling. It's about doing.

The rolling control device according to the present invention, developed in order to solve the above problems and realize the above object, is a device that controls multiple passes of rolling using a reversible rolling mill. The calculation processing section extracts coefficients for model calculation in advance and calculates rolling conditions for reducing dimensional errors after rolling using a rolling model. Based on the rolling model and the predetermined explanatory variables collected and transmitted by the explanatory variable collection section of the rolling process computer, or by reverse analysis, the rolling conditions for the next pass are selected and transmitted to the rolling condition determining section of the rolling process computer. It is characterized by being configured to do so.

Further, the rolling control method according to the present invention is a method of controlling multiple passes of rolling using a reversible rolling mill, and reduces dimensional errors after rolling by using a rolling model created in advance by deep learning. A calculation process for determining rolling conditions is executed by a computer, and in the calculation process, coefficients for model calculation are extracted in advance, and after a predetermined rolling pass is completed, predetermined explanatory variables collected by the rolling process computer and the rolling model are extracted. The present invention is characterized in that the rolling conditions for the next pass are selected based on the following or by reverse analysis, and are transmitted to the rolling condition determining section of the rolling process computer.

As explained above, the rolling control device and the rolling control method according to the present invention include a calculation processing unit that uses a rolling model created in advance by deep learning to determine rolling conditions that reduce dimensional errors after rolling. As a result, calculations for rolling control can now be completed within the allowable calculation time between passes during rolling. Further, it is preferable that the rolling model is created by a computer different from the calculation processing section, and is constructed by deep learning based on training data consisting of actual operation data. Furthermore, the dimension that reduces the error is the plate thickness, the rolling load in the next pass of rolling is predicted and the predicted rolling load is selected as the rolling condition, and the dimension that reduces the error is the plate width. It is preferable to predict the strip width due to rolling passes and select rolling conditions that reduce the difference between the predicted strip width and a target value.

1 is a schematic block diagram illustrating a rolling control device according to an embodiment of the present invention. It is a conceptual diagram explaining the rolling control method concerning the above-mentioned embodiment. It is a conceptual diagram explaining other rolling control methods concerning the above-mentioned embodiment. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a schematic conceptual diagram illustrating a plate rolling process using a reversible rolling mill to which the present invention is applied.

Embodiments of the present invention will be described below with reference to the drawings. Note that each drawing is schematic and may differ from the actual drawing. Furthermore, the following embodiments are intended to exemplify devices and methods for embodying the technical idea of the present invention, and the configuration is not limited to the following. That is, the technical idea of the present invention can be modified in various ways within the technical scope described in the claims.

FIG. 4 is a schematic conceptual diagram illustrating a plate rolling process using a reversible rolling mill to which the present invention is applied. The steel slab 100 heated to a predetermined temperature in a heating furnace is subjected to forming rolling 110 using a reversible rolling mill so that it has a desired shape that can be rolled to a target width and thickness in subsequent rolling. The material to be rolled 101, which has a desired shape, undergoes a 90° turn 140 within a horizontal plane. The rolled material 101 whose rolling direction has been rotated by 90 degrees is subjected to tentering rolling 120 using the same reversible rolling mill so as to stretch it in the width direction of the product. The rolled material 101 that has been rolled to the desired target width W is again rotated 140 by 90 degrees within the horizontal plane. The rolled material 101 whose rolling direction has been rotated by 90 degrees is subjected to thickening rolling 130 in the same reversible rolling mill so as to be stretched in the longitudinal direction of the product, and rolled to the target thickness of the product 102. In the tentering rolling 120, control is required to improve the prediction accuracy of the strip width, that is, to reduce the dimensional error between the target strip width W and the actual strip width. In the thickening rolling 130, it is required to improve the prediction accuracy of the rolling load so that the target product thickness is achieved, that is, control to reduce the difference between the target rolling load and the actual rolling load. Accordingly, the dimensional error between the target plate thickness and the actual product plate thickness can be reduced.

FIG. 1 shows a schematic block diagram of a rolling control device 1 according to an embodiment of the present invention. In this embodiment, a learning unit 2 that creates a rolling model in advance using training data related to plate rolling through deep learning, and a calculation processing unit 3 that performs calculations by giving explanatory variables to the created rolling model are operated on different computers. It is arranged in The rolling model created by the learning section is, for example, a deep learning model trained using a deep learning algorithm.

The rolling model uses the rolling conditions for plate rolling and the properties of the rolled material after rolling as a set of training data. For example, when the purpose is to improve the accuracy of rolling loads in order to reduce dimensional errors in product plate thickness, it is possible to Learning is performed using training data with one set of actual loads. In addition, when the purpose is to improve the accuracy of plate width, extracting the plate thickness at the end of tentering during tentering rolling and the heating furnace for past thick steel plate products with similar material properties and similar dimensions. Learning is performed using training data that includes one set of rolling conditions including temperature and width dimension after rolling.

The calculation processing unit 3 extracts coefficients necessary for rolling model calculation and stores them together with the model. The rolling process computer 4 includes an explanatory variable collection section 41, a rolling condition determining section 42, and a rolling execution section. The explanatory variable collection unit 41 collects explanatory variables necessary for rolling model calculation after completing a predetermined pass of plate rolling consisting of a plurality of passes using a reversible rolling mill, and transmits the explanatory variables to the calculation processing unit 3. The calculation processing unit 3 selects rolling conditions for the next pass of the predetermined pass based on the transmitted explanatory variables and rolling model or by inverse analysis. The calculation processing unit 3 transmits the selected rolling conditions for the next pass to the rolling condition determining unit 42. Here, the calculation processing section 3 may be constructed in the rolling process computer 4, or may be constructed in a separately prepared computer and communicated with the rolling process computer.

The rolling condition determining unit 42 of the rolling process computer 4 determines whether or not the rolling conditions selected and transmitted by the calculation processing unit 3 can be adopted, determines the rolling conditions, and transmits the rolling conditions to the rolling execution unit 43. The rolling execution unit 43 performs the next pass of rolling based on the determined rolling conditions. Specifically, the rolling condition determining unit 42 compares the result of the model calculation by the calculation processing unit 3 with a reference value, and determines whether or not to adopt the model. If it is determined that the method can be adopted, the model calculation results are reflected in the rolling conditions for the next pass. It is preferable that the information regarding whether or not the application is accepted is added to the teacher data of the learning section 2.

(Example 1)
FIG. 2 shows a conceptual diagram of a rolling control method when the above embodiment is applied to tentering rolling 120 in rolling a thick steel plate. In tentering rolling 120, the strip width after rolling corresponds to the product width, so it is necessary to improve the accuracy of strip width prediction. Here, a rolling model for deep learning calculation processing was created on the model creation personal computer 11 using training data consisting of actual operation data. A calculation processing unit 3 that provides explanatory variables to the created rolling model and performs calculations was installed in the rolling process computer 4, and parts necessary for model construction were installed in the model storage personal computer 12. Here, the training data is data that is thought to affect the tenting accuracy among the actual operation data that the thick plate rolling process computer 4 stores in a separate storage device (server), which is past rolling results. was used. For example, component composition, material properties, slab dimensions, shape and dimensions after forming rolling 110, shape and dimensions after tentering rolling 120, dimensions after thickening rolling 130, the number of times of each rolling, and the number of times the rolls of the rolling mill are used. It is selected from among factors such as the degree of wear, degree of wear, time between rolling, heating furnace extraction temperature, and calculated temperature of the rolled material before and after rolling, and has a high degree of contribution to affecting the dimensional error (width dimension) after the next rolling pass. selected in order. Further, as the explanatory variables, among the data currently being rolled that is similarly collected by the plate rolling process computer 4, the same items as those used for the teacher data were used.

First, the explanatory variables necessary for the model calculation at the completion of the forming rolling 110 of the tentering rolling 120, such as the target thickness at the end of tentering, the extraction temperature from the heating furnace, and the product width, are set in the rolling process computer 4. and transmits it to the installed calculation processing unit 3. The calculation processing unit 3 uses deep learning calculation processing to determine rolling conditions that reduce the difference between the predicted plate width and the target plate width in the next pass of tentering rolling.

Optimization by deep learning calculation processing uses inverse analysis to change the value of one variable among the explanatory variables based on a predetermined criterion. In the example of FIG. 2, the "target thickness at the end of tentering" is extracted from the explanatory variables, and the estimated width value based on the rolling model calculation is calculated for a plurality of target thicknesses at the end of tentering. The calculated width estimated value is compared with the target target width, and the closest width estimated value x, that is, the one with the smallest error from the target board width is determined. As a reverse analysis, the target thickness at the end of tentering from which the estimated width value x was calculated is selected as the rolling condition for tentering rolling. The rolling for the next pass and subsequent passes is repeated in the same manner.

(Example 2)
FIG. 3 shows a conceptual diagram of a rolling control method when the above embodiment is applied to thickening rolling in rolling of a thick steel plate. In thickening rolling, the plate thickness after rolling corresponds to the product plate thickness, so it is necessary to improve the accuracy of predicting the rolling load. Here, a rolling model for deep learning calculation processing was created on the model creation personal computer 11 using training data consisting of actual operation data. A calculation processing unit 3 that provides explanatory variables to the created rolling model and performs calculations was installed in the rolling process computer 4, and parts necessary for model construction were installed in the model storage personal computer 12. Here, as the training data, among the actual operation data that the thick plate process computer has stored separately in the storage device (server), which is the past rolling results, we use the one that is considered to affect the rolling load accuracy. there was. For example, component composition, material properties, pass information, rolling dimensions, heating furnace information, slab heating information, calculated temperatures of the rolled material before and after rolling, rolling mill roll information including roll changes and roll history, number of rolling rolls, pass schedule. information, previous pass performance information, etc., and selected in descending order of contribution to influencing the rolling load error of the next pass. Further, as the explanatory variables, among the data currently being rolled that is similarly collected by the plate rolling process computer 4, the same items as those used for the teacher data were used. In the example of FIG. 3, coefficients necessary for model calculation are given to the calculation processing section 3 in the rolling process computer 4, and when rolling is started, the calculation processing section 3 calculates the "rolling load" from the explanatory variables for each pass. is extracted and returned to the load calculation process. Then, in the load calculation process, after determining whether the "rolling load" is appropriate, load application processing is performed, and rolling is performed for that pass. The rolling for the next pass and subsequent passes is repeated in the same manner.

Thickening rolling was performed for general purpose thick steel plates and thick steel plates requiring temperature adjustment using rolling load prediction control applying the conventional method (approximately 5000 pieces each) and the present embodiment (approximately 5000 pieces each). In the conventional method, the ratio of the actual tip load to the predicted tip load had an average error of 0.6 to 0.7%, and the standard deviation σ was about 0.07 to 0.11. By applying this embodiment, the error in the average value was reduced to 0.1 to 0.2%, and the variation was also reduced to standard deviation σ=0.06 to 0.07. By applying this embodiment, improvement in rolling load prediction accuracy is recognized.

1 Rolling control device 2 Learning section 3 Calculation processing section 4 Rolling process computer 41 Explanatory variable collection section 42 Rolling condition determination section 43 Rolling execution section 11 Model creation personal computer 12 Model storage personal computer 100 Slab 101 Material to be rolled 102 Product 110 Forming rolling 120 Tenter rolling 130 Thickening rolling 140 90° rotation W Target width (target plate width)

Claims (6)

An apparatus for controlling multiple passes of rolling in the order of forming rolling, tentering rolling, and thickening rolling using a reversible rolling mill,
It has a calculation processing unit that uses a rolling model created in advance through deep learning to find rolling conditions that reduce dimensional errors after rolling.
The calculation processing section extracts coefficients for model calculation in advance, and after the completion of a predetermined rolling pass, based on the predetermined explanatory variables collected and transmitted by the explanatory variable collection section of the rolling process computer and the rolling model, Alternatively, it is configured to select the rolling conditions for the next pass through reverse analysis and transmit it to the rolling condition determining section of the rolling process computer.
In the tentering rolling, the dimension for reducing errors in the calculation processing section is the plate width,
The explanatory variables include the target thickness at the end of tentering, the extraction temperature from the heating furnace, and the product width,
The calculation processing unit predicts the sheet width by rolling the next pass for past thick steel plate products with similar material properties and similar dimensions, and calculates the predicted sheet width and target value for a plurality of tentering end target thicknesses. A rolling control device characterized in that it is configured to select rolling conditions that reduce the difference between . 2. The rolling model according to claim 1, wherein the rolling model is created by a computer different from the calculation processing unit, and is constructed by deep learning based on training data consisting of actual operation data. The rolling control device described. The dimension that reduces the error in the calculation processing unit is the plate thickness,
The rolling control device according to claim 1 or 2, wherein the calculation processing unit is configured to predict a rolling load in the next pass of rolling, and select the predicted rolling load as a rolling condition. . A method of controlling multiple passes of rolling in the order of forming rolling, tentering rolling, and thickening rolling using a reversible rolling mill, the method comprising:
Using a rolling model created in advance using deep learning, a computer executes calculation processing to determine rolling conditions that reduce dimensional errors after rolling.
In the calculation process, coefficients for model calculation are extracted in advance, and after the completion of a predetermined rolling pass, the coefficients for the next pass are calculated based on predetermined explanatory variables collected by the rolling process computer and the rolling model, or by inverse analysis. When selecting rolling conditions and transmitting them to the rolling condition determination section of the rolling process computer,
In the tentering rolling, the dimension that reduces errors in the calculation process is the plate width,
The explanatory variables include the target thickness at the end of tentering, the extraction temperature from the heating furnace, and the product width,
In the calculation process, the width of the next pass of rolling is predicted for past thick steel plate products with similar material properties and similar dimensions for multiple target thicknesses at the end of tentering, and the predicted plate width and target value are calculated. A rolling control method characterized in that rolling conditions that reduce the difference in are selected by inverse analysis . 5. The rolling control method according to claim 4 , wherein the rolling model is created by a computer different from the computer that performs the calculation processing, and deep learning is performed based on training data consisting of actual operation data. The dimension that reduces the error in the calculation process is the plate thickness,
6. The rolling control method according to claim 4 , wherein in the calculation process, a rolling load in the next pass of rolling is predicted, and the predicted rolling load is selected as the rolling condition.

Images