WO2024214427A1 - Method for predicting cross-sectional dimension of shape steel, method for manufacturing shape steel, cross-sectional dimension prediction device for shape steel, and method for generating cross-sectional dimension prediction model - Google Patents

Abstract

Provided is a method for predicting a cross-sectional dimension of a shape steel, the method being applicable to shape steels having various cross-sectional dimensions without being limited to a preceding rolled material and succeeding rolled material continuously hot-rolled, as long as the shape steel is manufactured by hot-rolling with the same roll set.　Provided is a cross-sectional dimension prediction method of a shape steel for predicting a cross-sectional dimension for a shape steel manufactured by hot-rolling, wherein the method involves: outputting the difference between cross-sectional dimension deviations by inputting, to a cross-sectional dimension prediction model which uses, as an input, the difference between rolling operation parameters of two shape steels manufactured by hot-rolling with the same roll set and uses, as an output, the difference between cross-sectional dimension deviations of the two shape steels, the difference between rolling operation parameters of a shape steel to be a preceding material manufactured by hot-rolling with the roll set and rolling operation parameters of a shape steel to be manufactured by hot-rolling with the roll set, the cross-sectional dimension deviations being deviations between target dimensions and actual dimensions of the shape steels; and using the output difference between the cross-sectional dimension deviations and the cross-sectional dimension of the shape steel to be the preceding material to predict the cross-sectional dimension of a shape steel to be manufactured.

Description

Method for predicting cross-sectional dimensions of shaped steel, manufacturing method for shaped steel, device for predicting cross-sectional dimensions of shaped steel, and method for generating cross-sectional dimension prediction model

The present invention relates to a method for predicting the cross-sectional dimensions of a steel section having a web and a flange, a method for manufacturing a steel section, a device for predicting the cross-sectional dimensions of a steel section, and a method for generating a cross-sectional dimension prediction model.

　Steel sections with webs and flanges are primarily manufactured by hot rolling, where heated steel slabs are rolled to the target product dimensions using multiple rolling mills. In the rough rolling process, multiple passes are performed using a rough rolling mill with multiple grooves to roughly shape the steel slab to approximate the shape of the product. In the subsequent intermediate rolling process, multiple passes are performed using one or more intermediate universal rolling mills and one or more edging rolling mills to bring the outer dimensions and thickness closer to the target product dimensions. In the final finishing rolling process, a finishing universal rolling mill usually performs one pass of rolling to roll to the target dimensions.

The cross-sectional dimensions of H-beams, a type of shaped steel, have set representative values for dimensions such as web height H, flange width B, web thickness tw, and flange thickness tf, as well as allowable ranges for dimensional tolerances, and all cross-sectional dimensions must be within the dimensional tolerances (within the target dimensions). Therefore, the rolling conditions are adjusted so that the cross-sectional dimensions after hot rolling are within the dimensional tolerances. Specifically, each dimension of the cross-section of the rolled shaped steel is measured, and if any dimension deviates from the dimensional tolerances, the roll spacing of each rolling mill is adjusted so that the dimension is within the dimensional tolerances.

As a method for controlling the dimensions of steel sections, Patent Document 1 discloses a method for controlling the thickness of H-shaped steel sections, in which the dimensions of the hot-rolled steel sections are measured and the amount of roll correction for the horizontal rolls and the vertical rolls is made approximately the same between successive passes, including the final pass of rough universal rolling. According to Patent Document 1, the thickness of the H-shaped steel sections can be controlled by predetermining the correspondence between the web thickness and flange thickness and the amount of roll correction for each roll, measuring the web thickness and flange thickness to determine the deviation from the target value, and adding the amount of roll correction to the setup calculation for the next material. Patent Document 2 describes a method for controlling the thickness of H-shaped steel sections, which specifies a method for adjusting the roll positions to eliminate variations in flange thickness at four points.

Patent Document 3 discloses the following technology for continuously manufacturing steel sections with the same cross-sectional dimensions. According to Patent Document 3, rolling operation parameters are acquired for each of the preceding and succeeding rolled materials of the steel section, and their cross-sectional dimensions are measured. Machine learning is performed using multiple data sets that include the relationship between the difference in rolling operation parameters and the difference in cross-sectional dimensions of the preceding and succeeding rolled materials. Then, a method for generating a prediction model for the amount of change in cross-sectional dimensions of steel sections, which generates a prediction model that predicts the amount of change in cross-sectional dimensions corresponding to the amount of correction of the rolling operation parameters, and a rolling control method using this model are disclosed.

Japanese Patent Application Laid-Open No. 62-151214 Japanese Patent Application Publication No. 10-296311 JP 2021-194701 A

The technologies disclosed in Patent Documents 1 and 2 measure the dimensions of the shaped steel during or after hot rolling, and change the settings of each roll to eliminate the difference from the target dimensions, thereby controlling the dimensions of the shaped steel to some extent. However, there are a great many dimensions of shaped steel, and it would take a lot of effort to predict the dimensions of all of them with high accuracy. In addition, the effects of changes in material temperature during rolling due to differences in the number of passes and material weight, and the effects of steel type and chemical components on the rolling load are not taken into consideration, resulting in problems with the accuracy of dimensional control. In addition, the technology disclosed in Patent Document 3 is effective only when shaped steel with the same cross-sectional dimensions are rolled continuously, and there is a problem that it cannot handle the first cross-sectional change when rolling shaped steel with various cross-sections with the same roll set. In addition, there is also the problem that it takes time to accumulate actual data that serves as learning data for generating a prediction model.

The present invention has been made in consideration of the problems with the conventional technology, and aims to provide a method and device for predicting the cross-sectional dimensions of shaped steel that can be applied to shaped steel of various cross-sectional dimensions, so long as the shaped steel is manufactured by hot rolling with the same roll set. Another aim of the present invention is to provide a method for manufacturing shaped steel using the method for predicting the cross-sectional dimensions of shaped steel, and a method for generating a cross-sectional dimension prediction model to be used in the method for predicting the cross-sectional dimensions of shaped steel.

The means for solving the above problems are as follows.
[1] A method for predicting the cross-sectional dimension of a structural steel to be manufactured by hot rolling, comprising: a cross-sectional dimension prediction model which takes as input the difference in rolling operation parameters for two structural steels manufactured by hot rolling with the same roll set; and outputs the difference in cross-sectional dimension deviation of the two structural steels; the method inputs the difference between the rolling operation parameters of a structural steel to be a predecessor material manufactured by hot rolling with the roll set and the rolling operation parameters of a structural steel to be manufactured by hot rolling with the roll set, and outputs the difference in cross-sectional dimension deviation, the cross-sectional dimension deviation being the deviation between the target dimension and actual dimension of the structural steel; and predicts the cross-sectional dimension of the structural steel to be manufactured using the output difference in cross-sectional dimension deviation and the cross-sectional dimension of the structural steel to be the predecessor material.
[2] The method for predicting cross-sectional dimensions of a structural steel as described in [1], wherein the structural steel is produced by hot rolling a steel slab using a roughing mill, an intermediate rolling mill and a finishing rolling mill, and the rolling operation parameters include the weight of the steel slab, a correction amount from a reference position of the rolling rolls of the intermediate rolling mill, the number of passes through the roughing mill and the intermediate rolling mill, and the rolling time of the roughing mill and the intermediate rolling mill.
[3] The method for predicting cross-sectional dimensions of structural steel according to [1] or [2], wherein the input of the cross-sectional dimension prediction model includes an attribute parameter indicating the steel type classification of the structural steel.
[4] The method for predicting the cross-sectional dimensions of a structural steel according to any one of [1] to [3], wherein the cross-sectional dimensions are at least one of the cross-sectional dimensions of the web thickness of the structural steel, the four flange thicknesses at the top, bottom, left and right, the web height and the flange width.
[5] A method for manufacturing structural steel, comprising: identifying a difference in rolling operation parameters that brings a cross-sectional dimension of a structural steel predicted using the method for predicting a cross-sectional dimension of a structural steel according to any one of [1] to [4] into a range of a target dimension; and manufacturing the structural steel under manufacturing conditions that include rolling operation parameters determined from the identified difference in rolling operation parameters.
[6] A cross-sectional dimension prediction device for structural steel that predicts the cross-sectional dimensions of structural steel to be manufactured by hot rolling, the device having a cross-sectional dimension prediction model that takes as input the difference in rolling operation parameters for two structural steels manufactured by hot rolling with the same roll set and outputs the difference in cross-sectional dimension deviation of the two structural steels, and outputs the difference in cross-sectional dimension deviation by inputting the difference between the rolling operation parameters of a structural steel that is a predecessor material manufactured by hot rolling with the roll set and the rolling operation parameters of a structural steel that is to be manufactured by hot rolling with the roll set, the difference in cross-sectional dimension deviation being the deviation between the target dimension and actual dimension of the structural steel, and the device having a cross-sectional dimension prediction unit that predicts the cross-sectional dimension of the structural steel to be manufactured using the output difference in cross-sectional dimension deviation and the actual value of the cross-sectional dimension of the structural steel that is to be the predecessor material.
[7] A method for generating a cross-sectional dimension prediction model, comprising: training a machine learning model using multiple data sets as training data, each set being a set of actual values of the difference in rolling operation parameters for two structural steels manufactured by hot rolling with the same roll set and actual values of the difference in cross-sectional dimension deviations of the two structural steels; generating a cross-sectional dimension prediction model that uses the difference in rolling operation parameters for the two structural steels as input and the difference in the cross-sectional dimension deviations of the two structural steels as output; and using the deviation between the target dimension and the actual dimension of the structural steel as the cross-sectional dimension deviation.

　According to the present invention, by using the deviation between the target dimension and the actual dimension as the cross-sectional dimension deviation of the steel section, a cross-sectional dimension prediction model can be applied to steel sections of various cross-sectional dimensions, so long as the steel sections are manufactured by hot rolling. This makes it possible to apply the model to any two sets of steel sections, not just the preceding and succeeding rolled materials that are hot rolled in succession, and therefore makes it possible to predict the cross-sectional dimensions of the steel section even if it is the first steel section after a cross-section change. Furthermore, since the actual values of the cross-sectional dimension deviation of any two sets of steel sections hot rolled with the same roll set can be used as learning data, a large amount of learning data for generating a cross-sectional dimension prediction model can be accumulated in a short period of time.

FIG. 1 is a schematic cross-sectional view showing the cross-sectional shape of an H-shaped steel. FIG. 2 is a schematic diagram showing an example of a hot rolling facility including a section dimension prediction device for section steel in which the section dimension prediction method for section steel according to this embodiment can be implemented. FIG. 3 is a functional block diagram of a section steel cross-sectional dimension prediction device. FIG. 4 is a flow diagram showing a flow of a rolling operation parameter specification process by the rolling operation parameter specification unit. FIG. 5 is a graph showing the relationship between the number of learning data and the average value of RMSE for each dimension of H-section steel.

Below, one embodiment of the present invention will be described with reference to the drawings. Note that the embodiment shown below is an example of an apparatus and method for embodying the technical concept of the present invention, and the technical concept of the present invention is not limited to the quality, shape, structure, arrangement, etc. of the components of the embodiment described below.

First, the cross-sectional shape of an H-shaped steel, which is an example of structural steel, will be explained using Figure 1. Figure 1 is a schematic cross-sectional view showing the cross-sectional shape of an H-shaped steel 10. As shown in Figure 1, the relatively thick parts of the H-shaped steel 10 are the flanges 12, and the relatively thin parts that are connected to the pair of flanges 12 are the webs 14. There are many combinations of cross-sectional dimensions of the H-shaped steel 10, such as the thickness tf and width B of the flanges 12, and the thickness tw and height H of the web 14, with dozens for constant inside dimension H-shaped steel (JIS H) and hundreds for constant outside dimension H-shaped steel. Here, the outside dimension is the web height H of the H-shaped steel 10, and the inside dimension is the dimension obtained by subtracting the flange thickness from the outside dimension.

Furthermore, the strength classification of the H-shaped steel 10 is broadly divided into three classes based on tensile strength: 400 N/ ^mm2 , 490 N/mm2, and more. The chemical composition of ^the H-shaped steel 10 is adjusted for each steel type classification (40k steel, 50k steel, and more) according to this strength classification. Such H-shaped steel 10 is mainly manufactured by hot rolling. The H-shaped steel 10 manufactured by hot rolling is called rolled H-shaped steel.

Below, an example of this embodiment will be described using an application to an H-shaped steel, which is an example of structural steel. Figure 2 is a schematic diagram showing an example of a hot rolling facility 100 including a structural steel cross-sectional dimension prediction device capable of implementing the structural steel cross-sectional dimension prediction method according to this embodiment. H-shaped steel is manufactured by rough rolling, intermediate rolling, and finish rolling using multiple rolling mills on a steel slab preheated in a heating furnace 20, and shaping it into structural steel of the target dimensions.

In the rough rolling process using the roughing mill 22, approximately 5 to 30 passes of reverse rolling are performed using grooved rolls. The intermediate rolling mill 24 is often a combination of one or more intermediate universal rolling mills 26 and edger rolling mills 28. In the intermediate rolling process, approximately 5 to 30 passes of reverse rolling are performed, and the material is rolled to a state that is roughly close to the cross-sectional shape of the final product. In the finish rolling process using the finish rolling mill 30, typically one pass of rolling is performed, and the material is formed into the target thickness and cross-sectional shape.

The intermediate universal rolling mill 26, which is one of the intermediate rolling mills 24, is a rolling mill that has a total of four rolls: upper and lower horizontal rolls that are driven to rotate about a horizontal axis, and left and right vertical rolls that rotate freely about a vertical axis.

The diameter of the upper and lower horizontal rolls is approximately 1000 to 1500 mm, and the width of the upper and lower horizontal rolls is appropriately adjusted according to the web height H of the H-shaped steel to be rolled, and is installed in the rolling mill. The diameter of the left and right vertical rolls is approximately 600 to 1000 mm. These upper and lower horizontal rolls and left and right vertical rolls are each designed so that their positions can be adjusted with a reduction device, and the spacing between the upper and lower horizontal rolls and the spacing between the sides of the horizontal rolls and the vertical rolls can be set as desired. The intermediate universal rolling mill 26 has the function of simultaneously reducing the web thickness tw and the flange thickness tf, and is characterized by rolling with the flange 12 tilted outward by a maximum of approximately 10°.

The edger rolling machine 28, which is one of the intermediate rolling machines 24, has two horizontal rolls (hereinafter referred to as E1 rolls) with grooves, one above the other, and adjusts the flange width B by pressing down the tip of the flange 12 from above and below with the E1 rolls. The diameter of the E1 rolls is about 800 to 1200 mm, and both E1 rolls are driven.

A finishing universal rolling mill is used for the finishing rolling mill 30. In the finishing rolling mill 30, the intermediate rolled material is finished into the product cross-sectional shape in one pass. The dimensions and configuration of the rolls of the finishing universal rolling mill are the same as those of the intermediate universal rolling mill 26, but in the finishing universal rolling mill, the flange 12 is rolled so that it is perpendicular to the web 14.

The cross-sectional dimensions of the H-shaped steel formed and manufactured by the above-mentioned hot rolling are measured in the hot state by a hot dimension gauge 32 installed downstream of the finishing rolling mill 30. The hot dimension gauge 32 measures the web height H of the H-shaped steel, the left and right flange widths B, the web thickness tw, and the flange thicknesses tf at four points on the top, bottom, left and right. It is preferable that the cross-sectional dimensions are measured by the hot dimension gauge 32 at as small intervals as possible over the entire longitudinal length. In addition, the measured cross-sectional dimensions are measured in the hot state, and when cooled to room temperature, the dimensions change due to thermal contraction. For this reason, it is preferable to measure the temperature of the H-shaped steel in or near the hot dimension gauge 32, predict the thermal contraction due to cooling using the temperature and linear expansion coefficient, and convert it into the product dimensions at room temperature. It is preferable that the actual dimensions used for predicting and controlling the cross-sectional dimensions are, for example, the average values of the cross-sectional dimensions measured over the entire length. Furthermore, the actual dimensions may be the average cross-sectional dimensions excluding a few meters from the end, which is the unsteady part at the tip and tail, or the average cross-sectional dimensions at one or more specific positions, or even the cross-sectional dimension at one specific position, such as the center of the length.

The process computer 36 can be a general-purpose computer such as a workstation or a personal computer. The process computer 36 is connected to the heating furnace 20, the roughing mill 22, the intermediate mill 24, the finishing mill 30, and the hot dimension gauge 32 by wire or wirelessly, and controls the manufacturing process of the H-beam by the hot rolling equipment 100. The process computer 36 also acquires attribute parameters of the H-beam to be manufactured from a higher-level computer. The attribute parameters include information on the target dimensions of the H-beam (web height H, left and right flange widths B, web thickness tw, and flange thicknesses tf at four locations (top, bottom, left, and right)), steel type classification (40k steel, 50k steel, and three types higher than this), chemical composition (contents of C, Si, Mn, Cr, Mo, V, etc.), and target values of mechanical properties (yield stress, tensile strength, elongation, toughness, hardness, etc.).

The process computer 36 sets rolling operation parameters such as the number of passes and roll spacing for each rolling mill according to the attribute information of the H-shaped steel. The set values of normal rolling operation parameters are based on past rolling performance and are set as table values associated with the attribute parameters of the H-shaped steel.

The process computer 36 sets, as rolling operation parameters for the rolling process, a correction value from the standard conditions for the horizontal roll gap in each pass in the roughing mill 22 and a correction value for the axial relative position of the horizontal roll. The process computer 36 also sets, as rolling operation parameters for the rolling process, the correction amount from the standard conditions for the horizontal roll gap in the intermediate universal mill 26, the correction amount for the thrust position of the horizontal roll, the correction amount for the horizontal roll height, the correction amount from the standard conditions for the vertical roll gap, and the center correction amount for the vertical rolls, as well as the correction amount from the standard conditions for the E1 roll gap in the edger mill 28, the leveling correction amount for the E1 roll, and the correction amount for the thrust relative position of the E1 roll. Furthermore, the process computer 36 sets the amount of correction from the standard conditions for the horizontal roll gap in the finishing mill 30, the amount of correction for the thrust position of the horizontal roll, the amount of correction for the height of the horizontal roll, the amount of correction from the standard conditions for the vertical roll gap, and the amount of center correction for the vertical roll as rolling operation parameters for the rolling process.

The process computer 36 outputs the set rolling operation parameters to the roughing mill 22, the intermediate universal mill 26, the edger mill 28, and the finishing mill 30. Each rolling mill controls the roll spacing, roll position, etc. of each rolling mill in accordance with the acquired rolling operation parameters.

The process computer 36 also collects the actual dimensions of the H-beam measured by the hot dimension gauge 32. The process computer 36 stores in the database of the process computer 36 the target dimensions of the manufactured H-beam, the steel type classification, the chemical composition, the target characteristics, the cross-sectional dimensions of the steel piece, the weight of the steel piece, the rolling operation parameters, the actual dimensions, and an identification number identifying the roll set used, in association with the manufacturing lot number of the H-beam.

The section steel cross-sectional dimension prediction device 38 acquires from the process computer 36 the rolling operation parameters of the H-shaped steel, which is the precursor material that has been hot rolled with the same roll set as the H-shaped steel to be manufactured. The difference between the rolling operation parameters of the H-shaped steel to be manufactured and the rolling operation parameters of the precursor material is then input into the cross-sectional dimension prediction model, and the difference in the cross-sectional dimension deviation of the H-shaped steel is output to predict the cross-sectional dimensions of the H-shaped steel to be manufactured.

In hot rolling of H-shaped steel, multiple H-shaped steels with approximately the same web height H or web inner width and flange width B among the target cross-sectional dimensions are rolled together continuously with the same roll set. For this reason, there are many cases where H-shaped steels with the same web height H and flange width B but with changed web thickness tw and flange thickness tf are manufactured with the same roll set, and when changing such cross-sectional dimensions, the dimensional tolerance may be removed. Of course, when the dimensional tolerance is removed, but even if a dimension within the dimensional tolerance is obtained, if the dimension becomes close to the upper or lower limit of the dimensional tolerance, the rolling operation parameters are changed. In this way, the cross-sectional dimension prediction device 38 of the shape steel according to this embodiment is used to predict the cross-sectional dimensions of the H-shaped steel when the rolling operation parameters are changed while rolling with the same roll set. Note that "hot rolling with the same roll set" means that all rolls used in hot rolling (rough rolling, intermediate rolling, and finishing rolling) are rolled without rearrangement or replacement.

In the section dimension prediction device 38 of this embodiment, the deviation between the target dimension and the actual dimension in the cross section of the H-shaped steel is used as the predicted cross-sectional dimension. As a result, this cross-sectional dimension prediction model is a model that can predict the effect of changes in rolling operation parameters on the deviation from the target dimension of the H-shaped steel, so that it is a cross-sectional dimension prediction model that can be applied to various types of hot-rolled steel with different cross-sectional dimensions.

In this way, the cross-sectional dimension prediction model used in the cross-sectional dimension prediction device 38 for shaped steel in this embodiment can be applied to various shaped steels with different cross-sectional dimensions. This makes it possible to apply it to any two H-shaped steels hot rolled with the same roll set, not just to the preceding and succeeding rolled materials that are hot rolled in succession. As a result, it becomes possible to predict the cross-sectional dimensions of even the first H-shaped steel produced by hot rolling after changing the cross-sectional dimensions. Furthermore, since the actual values of the difference between the rolling operation parameters of any two H-shaped steels and the actual values of the difference between the cross-sectional dimension deviations can be used as learning data for the cross-sectional dimension prediction model, a large amount of learning data for generating the cross-sectional dimension prediction model can be accumulated in a short period of time.

For example, when generating a cross-sectional dimension prediction model using operational performance data of 100 H-shaped steels manufactured by hot rolling with the same roll set, the maximum number of sets of learning data that can be obtained from two H-shaped steels that are continuously rolled as disclosed in Patent Document 3 is 99 sets. In contrast, when any two H-shaped steels are selected from the operational performance data of 100 H-shaped steels, a maximum of 4950 sets of learning data can be obtained, and the number of learning data obtained from the operational performance data increases significantly. When generating a cross-sectional dimension prediction model using operational performance data of 100 H-shaped steels manufactured by hot rolling with two different roll sets, the difference in the operational performance data and the difference in the cross-sectional dimension deviation are limited to data from the same roll set. However, these can be used as learning data for the same cross-sectional dimension prediction model, so the number of learning data is 4950 sets x 2.

Next, the section dimension prediction device 38 for predicting the section dimensions of section steel will be described. FIG. 3 is a functional block diagram of the section dimension prediction device 38 for section steel. The section dimension prediction device 38 for section steel can be a general-purpose computer such as a workstation or a personal computer. The section dimension prediction device 38 for section steel has a control unit 40, an input unit 42, an output unit 44, and a storage unit 46. The control unit 40 is, for example, a CPU, and functions as a data acquisition unit 50, a difference calculation unit 52, a section dimension prediction unit 54, a rolling operation parameter identification unit 56, and a section dimension prediction model generation unit 58 by executing a program stored in the storage unit 46.

The input unit 42 is, for example, a keyboard, a touch panel that is integrated with a display, or the like. The output unit 44 is, for example, an LCD or CRT display, or the like. The storage unit 46 is, for example, an information recording medium such as an updatable flash memory, a hard disk built-in or connected via a data communication terminal, a memory card, or the like, and a read/write device for the information recording medium. The storage unit 46 stores programs and data for implementing each function of the section dimension prediction device 38 for section steel. The storage unit 46 further stores a database 60 and a section dimension prediction model 62. The database 60 stores 10,000 or more sets of learning data sets, each set consisting of a difference in rolling operation parameters of H-beam steel and a difference in section dimension deviation for each roll set of the hot rolling equipment 100. The number of learning data sets stored in the database 60 is preferably 20,000 or more sets, and more preferably 50,000 or more sets.

The cross-sectional dimension prediction model 62 is a trained machine learning model that has been machine-learned using the learning data stored in the database 60 as training data. The cross-sectional dimension prediction model 62 according to this embodiment is a trained machine learning model that takes the difference in the rolling operation parameters as input and outputs the difference in the cross-sectional dimension deviation of the H-beam 10, and uses the deviation between the target dimension and the actual dimension as the cross-sectional dimension deviation. The difference in the cross-sectional dimension deviation that is the output of the cross-sectional dimension prediction model 62 is the difference in at least one cross-sectional dimension deviation selected from seven cross-sectional dimensions of the H-beam web height H, flange width B, web thickness tw, and flange thickness tf at four locations above, below, left, and right.

Next, the processing performed by the data acquisition unit 50, difference calculation unit 52, and cross-sectional dimension prediction unit 54 when predicting the cross-sectional dimensions of the H-shaped steel 10 to be manufactured will be described. The data acquisition unit 50 acquires from the process computer 36 the rolling operation parameters of the H-shaped steel to be manufactured, the target dimensions of the cross-sectional dimensions of the H-shaped steel to be manufactured, and the rolling operation parameters, target dimensions, and actual dimensions of a preceding steel beam manufactured in the past with the same roll set. The data acquisition unit 50 outputs the acquired rolling operation parameters to the difference calculation unit 52, and outputs the acquired target dimensions and actual dimensions to the cross-sectional dimension prediction unit.

The difference calculation unit 52 calculates the difference between the rolling operation parameters of the H-shaped steel to be manufactured and the rolling operation parameters of the H-shaped steel that will be the precursor material manufactured in the past with the same roll set. The difference calculation unit 52 outputs the calculated difference in the rolling operation parameters to the cross-sectional dimension prediction unit 54. It is preferable to select a steel section that is as close in time as possible to the H-shaped steel section to be manufactured as the precursor material. This makes it possible to reduce the influence of parameters that change over time, such as roll wear, and that are not included in the input.

When the cross-sectional dimension prediction unit 54 acquires the difference in the rolling operation parameters, it reads out the cross-sectional dimension prediction model 62 from the storage unit 46. The cross-sectional dimension prediction unit 54 inputs the difference in the rolling operation parameters into the cross-sectional dimension prediction model 62, and outputs the difference in the cross-sectional dimension deviation between the H-shaped steel to be manufactured and the H-shaped steel that will be the preceding material manufactured in the past with the same roll set. In the cross-sectional dimension prediction device 38 for structural steel in this embodiment, the deviation between the target dimension and the actual dimension is used as the cross-sectional dimension. Therefore, the difference in the cross-sectional dimension deviation that is output is the sum of the difference in the target dimension between the H-shaped steel that will be the preceding material manufactured in the past and the H-shaped steel to be manufactured, and the difference in the actual dimension between the H-shaped steel to be manufactured and the H-shaped steel that will be the preceding material manufactured in the past. Of the differences in the target dimensions and the differences in the actual dimensions, the target dimension, the actual dimension, and the target dimension of the H-shaped steel to be manufactured of the preceding material manufactured in the past are acquired from the data acquisition unit 50. Therefore, the cross-sectional dimension prediction unit 54 can predict the cross-sectional dimensions of the H-shaped steel to be manufactured by using the difference between the predicted cross-sectional dimensions and the above-mentioned known dimensions. The cross-sectional dimension prediction unit 54 may output the predicted cross-sectional dimensions of the H-shaped steel to be manufactured to the output unit 44, and display the predicted cross-sectional dimensions of the steel on the output unit 44. This allows the operator to check the predicted cross-sectional dimensions of the H-shaped steel by visually checking the output unit 44.

The data acquisition unit 50 preferably acquires from the process computer 36 the rolling operation parameters to be input to the cross-sectional dimension prediction model 62, such as the weight of the steel billet, the amount of correction from the standard conditions of the rolling rolls of the intermediate universal rolling mill 26 and the edger rolling mill 28 that constitute the intermediate rolling mill, and the number of passes and rolling time in the roughing mill 22 and the intermediate rolling mill 24.

Here, the correction amount from the standard conditions of the rolling rolls of the intermediate universal rolling mill 26 refers to the correction amount from the standard conditions of the horizontal roll opening of the intermediate universal rolling mill 26, the correction amount of the thrust position of the horizontal roll, the correction amount of the horizontal roll height, the correction amount from the standard conditions of the vertical roll opening, and the difference in the opening of the left and right vertical rolls. Also, the correction amount from the standard conditions of the rolling rolls of the edger rolling mill 28 refers to the correction amount from the standard conditions of the E1 roll opening in the edger rolling mill 28.

In intermediate rolling, the roughly rolled material is rolled in multiple passes until it roughly assumes the cross-sectional shape of the product. For this reason, the amount of correction for the roll opening and position in intermediate rolling significantly affects the cross-sectional dimensions of the H-beam. For this reason, it is preferable to include a correction amount from the standard conditions of the rolling rolls of the intermediate universal rolling mill 26 and edger rolling mill 28 that make up the intermediate rolling mill in the rolling operation parameters that are input to the cross-sectional dimension prediction model 62.

Furthermore, since the number of passes and rolling time at the roughing mill 22 and intermediate rolling mill 24 also have a significant effect on the cross-sectional dimensions of the H-shaped steel, it is preferable to include the number of passes and rolling time at the roughing mill 22 and intermediate rolling mill 24 in the rolling operation parameters that are input to the cross-sectional dimension prediction model 62.

Furthermore, since the weight of the steel billet, which is the raw material, also has a large effect on the cross-sectional dimensions of the H-beam, it is preferable to include the weight of the billet in the rolling operation parameters that are input to the cross-sectional dimension prediction model 62. If the weight of the billet is heavy, the rolling length and rolling time will be longer, which will result in different cross-sectional temperature distributions and deformation resistance. For this reason, if the weight of the billet is different, the reduction rate will be different even if the billet is rolled with the same roll spacing, and as a result, the cross-sectional dimensions of the H-beam will change. Note that instead of the weight of the billet, the length of the billet may be included in the rolling operation parameters.

In addition, the data acquisition unit 50 preferably acquires from the process computer 36 at least one of the following rolling operation parameters to be input to the cross-sectional dimension prediction model 62: the cross-sectional dimension of the slab, the rolling time of the finishing rolling process, the transport time from the heating furnace to the roughing mill, the transport time from the roughing mill to the intermediate mill, the transport time from the intermediate mill to the finishing mill, the ambient temperature of the heating furnace 20, the time in the furnace, the gas flow rate in the furnace, the extraction temperature, the correction amount from the reference position of the horizontal roll opening of the finishing mill 30, the correction amount from the reference position of the vertical roll opening, and the correction amount of the thrust position of the horizontal roll. Since these parameters also directly or indirectly affect the cross-sectional dimension of the H-beam, by including these parameters in the rolling operation parameters to be input to the cross-sectional dimension prediction model 62, it becomes possible to predict the cross-sectional dimension of the H-beam with higher accuracy.

Next, the processing of the rolling operation parameter identification unit 56 will be described. The rolling operation parameter identification unit 56 identifies the rolling operation parameters that will bring the cross-sectional dimensions predicted by the cross-sectional dimension prediction unit 54 within the range of the target dimensions (within the tolerance range), and outputs the rolling operation parameters to the process computer 36 to reflect them in the manufacturing conditions of the H-beam.

FIG. 4 is a flow diagram showing the flow of the rolling operation parameter identification process by the rolling operation parameter identification unit 56. The flow shown in FIG. 4 is started, for example, by receiving an input from an operator to start the process.

First, the data acquisition unit 50 acquires actual data of H-shaped steel rolled with the same roll set from the process computer 36 (step S101). The rolling operation parameter identification unit 56 sets the rolling operation parameters of the H-shaped steel to be manufactured (step S102). The rolling operation parameters may be set based on the target dimensions of the H-shaped steel to be manufactured and attribute parameters such as the steel type classification. The data acquisition unit 50 and the rolling operation parameter identification unit 56 output these rolling operation parameters to the difference calculation unit 52.

The difference calculation unit 52 calculates the difference of the acquired rolling operation parameters (step S103). The difference calculation unit 52 outputs the difference of the rolling operation parameters to the cross-sectional dimension prediction unit 54. The cross-sectional dimension prediction unit 54 reads the cross-sectional dimension prediction model 62 from the storage unit 46. The cross-sectional dimension prediction unit 54 inputs the difference of the acquired rolling operation parameters to the cross-sectional dimension prediction model 62 and outputs the difference of the cross-sectional dimension deviation of the H-shaped steel, thereby predicting the cross-sectional dimension of the H-shaped steel to be manufactured (step S104). The cross-sectional dimension prediction unit 54 obtains a predicted value of the cross-sectional dimension of the H-shaped steel to be manufactured using the difference of the cross-sectional dimension deviation output from the cross-sectional dimension prediction model 62, the target dimension of the H-shaped steel rolled by the same roll, the actual dimension, and the target dimension of the H-shaped steel to be manufactured. The cross-sectional dimension prediction unit 54 outputs the cross-sectional dimension of the H-shaped steel to the rolling operation parameter identification unit 56.

The rolling operation parameter identification unit 56 determines whether the predicted cross-sectional dimensions are within the range of the target dimensions (within the range of the dimensional tolerance) (step S105). The range of the target dimensions may be input by the operator from the input unit 42, or may be acquired from the process computer 36 via the data acquisition unit 50.

If the predicted cross-sectional dimensions are not within the range of the target dimensions (step S105: No), the rolling operation parameter identification unit 56 changes the rolling operation parameters (step S106). The rolling operation parameter identification unit 56 returns the process to step S103, and repeats the processes of steps S103 to S106 until the cross-sectional dimensions predicted in step S105 are within the range of the target dimensions.

On the other hand, if the predicted cross-sectional dimension is within the range of the target dimension (step S105: Yes), the rolling operation parameter identification unit 56 identifies the difference in the rolling operation parameters used to predict the cross-sectional dimension as the difference in the rolling operation parameters that can bring the cross-sectional dimension within the range of the target dimension (step S107). The rolling operation parameter identification unit 56 calculates the rolling operation parameters to be manufactured using the rolling operation parameters of the H-section steel rolled by the same roll set obtained in step S101 and the identified difference value (step S108). By executing this process, the flow of the rolling operation parameter identification process shown in Figure 4 is completed. The rolling operation parameter identification unit 56 outputs the calculated rolling operation parameters to the process computer 36, thereby reflecting the rolling operation parameters in the manufacturing conditions of the hot rolling equipment 100.

In this way, it is possible to calculate the rolling operation parameters that will bring the cross-sectional dimensions of the H-shaped steel within the target range. Then, by manufacturing the H-shaped steel under manufacturing conditions that include these rolling operation parameters, it becomes possible to stably manufacture H-shaped steel whose cross-sectional dimensions fall within the target range.

Next, a method for generating the cross-sectional dimension prediction model 62 used to predict the cross-sectional dimensions of H-shaped steel will be described. The data acquisition unit 50 acquires the actual values of the rolling operation parameters and the cross-sectional dimensions (deviation between the target dimensions and the actual dimensions) of two H-shaped steels manufactured in the past with the same roll set from the process computer 36. The data acquisition unit 50 outputs the acquired actual values of the rolling operation parameters and the cross-sectional dimension deviation of the two H-shaped steels to the difference calculation unit 52. The difference calculation unit 52 calculates the difference between the two rolling operation parameters and the difference between the cross-sectional dimension deviation, and stores a data set consisting of these differences as one set in the database 60 of the storage unit 46. The number of data sets stored in the database 60 is preferably 10,000 sets or more, more preferably 20,000 sets or more, and even more preferably 50,000 sets or more. The difference between the rolling operation parameters and the difference between the cross-sectional dimension deviation are calculated using the actual values of the rolling operation parameters and the cross-sectional dimensions of the steel hot rolled with the same roll set. However, the data set that takes the difference as one set is not limited to H-shaped steel hot rolled with the same roll set, and a data set of H-shaped steel hot rolled with different roll sets can be used.

The cross-sectional dimension prediction model generating unit 58 reads out the machine learning model stored in advance in the storage unit 46, and trains the machine learning model using the dataset stored in the database 60 as training data to generate a trained machine learning model. This trained machine learning model becomes the cross-sectional dimension prediction model. The cross-sectional dimension prediction model generating unit 58 stores the generated cross-sectional dimension prediction model 62 in the storage unit 46. The machine learning model used in this embodiment may be any of the commonly used neural networks, decision tree learning, random forests, and support vector regression.

Furthermore, the cross-sectional dimension prediction model may be updated to a new cross-sectional dimension prediction model, for example, by retraining it every month or year. The data acquisition unit 50 acquires actual data each time an H-shaped steel is manufactured and stores it in the database 60. As actual data of newly manufactured H-shaped steel is stored in the database 60, the amount of actual data stored increases. As the amount of actual data and the amount of training data increase, the prediction accuracy of the cross-sectional dimension prediction model improves, making it possible to predict cross-sectional dimensions with higher accuracy. Furthermore, by using new actual data, the latest state of the hot rolling equipment 100 is reflected in the cross-sectional dimension prediction model, so that by periodically training the cross-sectional dimension prediction model through machine learning, it becomes possible to predict cross-sectional dimensions with even higher accuracy.

As described above, in the cross-sectional dimension prediction model used in the cross-sectional dimension prediction device 38 of the present embodiment, the deviation between the target dimension and the actual dimension in the cross section of the H-shaped steel is used as the predicted cross-sectional dimension. As a result, the cross-sectional dimension prediction model 62 becomes a model that can predict the influence on the deviation from the target dimension, and in the case of hot-rolled steel, it becomes a cross-sectional dimension prediction model that can be applied to various steels having different cross-sectional dimensions that are the target dimensions. By using this cross-sectional dimension prediction model 62, it becomes possible to predict the cross-sectional dimension of the first steel that is manufactured by hot rolling after changing the cross-sectional dimension. Furthermore, it becomes possible to use the actual values of the difference in the rolling operation parameters and the actual values of the difference in the cross-sectional dimension deviation of any two steels hot-rolled with the same roll set as learning data. As a result, it becomes possible to accumulate a large amount of learning data for generating a cross-sectional dimension prediction model of steel in a short time, and it becomes possible to predict the cross-sectional dimension with high accuracy.

In this embodiment, an example has been given in which the difference in rolling operation parameters is used as the input to the cross-sectional dimension prediction model 62 in the cross-sectional dimension prediction device 38 for structural steel, but this is not limiting. It is preferable to include the steel type classification, which is an attribute parameter of structural steel, in addition to the rolling operation parameters as the input to the cross-sectional dimension prediction model. Since the steel type classification also affects the cross-sectional dimensions of structural steel, by including the steel type classification as the input to the cross-sectional dimension prediction model 62, it becomes possible to predict the cross-sectional dimensions with high accuracy even for structural steels with different steel type classifications.

It is preferable to further include the chemical composition of the structural steel as an attribute parameter of the structural steel used as input to the cross-sectional dimension prediction model 62. Since the chemical composition of the structural steel also affects the cross-sectional dimensions, by including the chemical composition as input to the cross-sectional dimension prediction model 62, it becomes possible to predict the cross-sectional dimensions with high accuracy even for structural steels with different chemical compositions.

In the description of this embodiment, the hot rolling equipment 100 shown in FIG. 2 has a process computer 36 and a section steel cross-sectional dimension prediction device 38, but this is not limited to the above. For example, the process computer 36 may have the function of the section steel cross-sectional dimension prediction device 38, and these may be configured as a single device.

In the description of this embodiment, in the section dimension prediction device 38 of the section steel shown in FIG. 3, an example was shown in which the control unit 40 has the data acquisition unit 50, the difference calculation unit 52, the section dimension prediction unit 54, the rolling operation parameter identification unit 56, and the section dimension prediction model generation unit 58, but this is not limited to this. If the section dimension prediction device 38 of the section steel predicts the section dimension, the control unit 40 does not need to have the rolling operation parameter identification unit 56. In addition, if the section dimension prediction model 62 is generated externally and the generated section dimension prediction model 62 is stored in the storage unit 46 via the data acquisition unit 50, the section dimension prediction device 38 of the section steel does not need to have the section dimension prediction model generation unit 58. Furthermore, if difference data such as the difference in the rolling operation parameters is stored in the database of the process computer 36 and the data acquisition unit 50 acquires the difference data, the control unit 40 does not need to have the difference calculation unit 52.

In the explanation of this embodiment, an example of predicting the cross-sectional dimensions of H-shaped steel using the section dimension prediction device 38 has been described, but the section steel for which the cross-sectional shape is predicted is not limited to H-shaped steel. The section dimension prediction device 38 can also predict the cross-sectional shapes of other sections manufactured by rolling with an intermediate universal rolling mill, such as channel steel, I-shaped steel, steel sheet piles, or rails.

In order to verify the effects of the present invention, an example of manufacturing H-shaped steel using the hot manufacturing equipment 100 shown in Figure 2 will be described. In this example, H-shaped steel with a web height H of 600 mm and a flange width B of 300 mm was manufactured by hot rolling using the same roll set. There are two types of web thickness, 14 mm and 16 mm, and five types of flange thickness, 19 mm, 22 mm, 25 mm, 28 mm, and 32 mm, and there are 10 types of cross-sectional dimensions as combinations of these. 100 pieces of each of these 10 types of H-shaped steel, for a total of 1,000 H-shaped steel pieces, were manufactured by changing the rolling operation parameters.

The cross-sectional dimensions of the H-beam were measured hot using a hot dimension gauge installed downstream of the finishing universal rolling mill, measuring the web height H, left and right flange width B, web thickness tw, and top, bottom, left and right flange thicknesses tf. The measured cross-sectional dimensions were converted to dimensions at room temperature using the linear expansion coefficient.

The cross-sectional dimensions used to generate the cross-sectional dimension prediction model are the room temperature converted values of the web height H, left and right flange width B, web thickness tw, and top, bottom, left and right flange thickness tf of the H-beam measured with a hot dimension gauge, and in the example of the invention, the deviation between the target dimensions and the actual dimensions was used. The rolling operation parameters used to generate the cross-sectional dimension prediction model are the values shown in (1) to (7) below.

(1) Correction amount of horizontal roll gap of the intermediate universal rolling mill from the standard conditions (2) Correction amount of horizontal roll height of the intermediate universal rolling mill (3) Correction amount of vertical roll gap of the intermediate universal rolling mill from the standard conditions (4) Center correction amount of vertical rolls of the intermediate universal rolling mill (5) Correction amount of E1 roll gap of the edger rolling mill from the standard conditions (6) Weight of steel billet (7) Rolling time at the intermediate universal rolling mill

In the invention example, 10 sets of H-shaped steel made with the same roll set of 100 pieces were prepared, and all combinations of two pieces extracted from the steel pieces made with each roll set were selected. Since there were 4,950 sets of combinations of two pieces extracted from 100 pieces selected from the same roll set, a total of 49,500 sets of manufacturing performance data were used as learning data to generate a cross-sectional dimension prediction model.

First, we will explain the results of checking the number of training data and the prediction accuracy of the cross-sectional dimension prediction model. Figure 5 is a graph showing the relationship between the number of training data and the average RMSE when predicting each dimension of 100 H-shaped steel pieces that were not used for training. In Figure 5, the horizontal axis is the number of training data (sets), and the vertical axis is the average RMSE (root mean square error: mm) of each predicted dimension of the H-shaped steel (web height H, left and right flange width B, web thickness tw, and four flange thicknesses tf at top, bottom, left and right).

As shown in Figure 5, the average RMSE value for each predicted dimension of H-beam decreases as the number of training data used to train the cross-sectional dimension prediction model increases. Considering the dimensional tolerances of H-beam, it is preferable that the average RMSE value be 0.1 mm or less. In order to make the average RMSE value 0.1 mm or less, it is preferable to use 10,000 or more sets of training data to generate the cross-sectional dimension prediction model.

As described above, the method for predicting the cross-sectional dimensions of structural steel according to this embodiment allows learning data to be accumulated in a short time, making it possible to obtain more than 10,000 sets of learning data in a shorter time than before. It has been confirmed that by using a cross-sectional dimension prediction model trained using a large amount of learning data acquired in this way, it is possible to predict the dimensions of H-shaped steel with high accuracy.

Next, we will explain the results of confirming the manufacturing method of section steel using the section steel cross-sectional dimension prediction method. Using the cross-sectional dimension prediction model generated as described above, we first predicted the cross-sectional dimensions of web height, left and right flange widths, web thickness, and top, bottom, left and right flange thicknesses. Next, we identified the rolling operation parameters that would result in the predicted cross-sectional dimensions falling within the dimensional tolerances of these cross-sectional dimensions. The tolerances for each dimension are as follows:

Web height: ±2.0mm
Left and right flange width: ±2.0mm
Web thickness: ±0.7mm
Top, bottom, left and right flange thickness: -0.7 to +2.3 mm

99 H-beams were manufactured by setting the rolling operation parameters as manufacturing conditions so that the predicted cross-sectional dimensions would fall within the above tolerances. The operation parameters for the second rolled piece were determined using the actual cross-sectional dimensions of the first rolled piece, the operation parameters for the first and second rolled pieces, and the target cross-sectional dimensions of the second rolled piece. The operation parameters for the third rolled piece were determined using the actual cross-sectional dimensions of the second rolled piece, the operation parameters for the second and third rolled pieces, and the target cross-sectional dimensions of the third rolled piece. Similarly, for subsequent H-beams, the operation parameters for the next H-beam to be rolled were set using the operation parameters of the preceding material and the H-beam to be next rolled. As a result, H-beams with good cross-sectional dimensions, all of which were within the above tolerances, were manufactured.

In the comparative example, on the other hand, a cross-sectional dimension prediction model was generated using the method disclosed in Patent Document 3. In the method disclosed in Patent Document 3, the difference between two sections of steel that were hot rolled in succession (the preceding rolled material and the following rolled material) is used. However, since the data between two pieces with different target cross-sectional dimensions does not become learning data, 90 sets of data are extracted from 100 pieces selected from the same roll set. Therefore, in the comparative example, a total of 900 sets of manufacturing performance data were used as training data to generate a cross-sectional dimension prediction model.

This cross-sectional dimension prediction model was used to predict the cross-sectional dimensions of the web height, flange width, web thickness, and the four flange thicknesses at the top, bottom, left, and right, and the rolling operation parameters were identified that would result in the predicted cross-sectional dimensions falling within the dimensional tolerances of these cross-sectional dimensions. The tolerances for each dimension are as shown above.

99 H-beams were manufactured by setting the rolling operation parameters as manufacturing conditions so that the predicted cross-sectional dimensions would fall within the above tolerances. The rolling operation parameters for the next H-beam to be rolled were determined using the operation parameters for the preceding material and the next H-beam to be rolled. The rolling operation parameters for the H-beam to be rolled after the cross-section was changed were determined to be the standard conditions for that cross-section without using this model. As a result, some dimensions of 15 H-beams from the second H-beam onwards were outside the tolerances, and these H-beams required additional dimensional corrections and remanufacturing, and it was not possible to manufacture H-beams with good cross-sectional dimensions.

REFERENCE SIGNS LIST 10 H-section steel 12 Flange 14 Web 20 Heating furnace 22 Roughing mill 24 Intermediate rolling mill 26 Intermediate universal rolling mill 28 Edger rolling mill 30 Finishing rolling mill 32 Hot dimension gauge 36 Process computer 38 Section dimension prediction device for section steel 40 Control unit 42 Input unit 44 Output unit 46 Storage unit 50 Data acquisition unit 52 Difference calculation unit 54 Section dimension prediction unit 56 Rolling operation parameter identification unit 58 Section dimension prediction model generation unit 60 Database 62 Section dimension prediction model 100 Hot rolling equipment

Claims (9)

A method for predicting a cross-sectional dimension of a shaped steel to be manufactured by hot rolling, comprising:
a cross-sectional dimension prediction model which takes as input the difference in rolling operation parameters for two shaped steels produced by hot rolling with the same roll set and outputs the difference in cross-sectional dimension deviations of the two shaped steels, by inputting the difference between the rolling operation parameters of a shaped steel to be a precursor material produced by hot rolling with the roll set and the rolling operation parameters of a shaped steel to be produced by hot rolling with the roll set, and outputting the difference in cross-sectional dimension deviations;
The cross-sectional dimension deviation is a deviation between a target dimension and an actual dimension of a shaped steel,
A method for predicting the cross-sectional dimension of structural steel, which predicts the cross-sectional dimension of the structural steel to be manufactured using the output difference in cross-sectional dimension deviation and the cross-sectional dimension of the preceding structural steel material. The method for predicting cross-sectional dimensions of a steel section as described in claim 1, wherein the steel section is manufactured by hot rolling a steel slab using a roughing mill, an intermediate rolling mill, and a finishing rolling mill, and the rolling operation parameters include the weight of the steel slab, a correction amount from a reference position of the rolling rolls of the intermediate rolling mill, the number of passes through the roughing mill and the intermediate rolling mill, and the rolling time of the roughing mill and the intermediate rolling mill. The method for predicting the cross-sectional dimensions of steel sections according to claim 1, wherein the input of the cross-sectional dimension prediction model includes attribute parameters indicating the steel type classification of the steel section. The method for predicting the cross-sectional dimensions of steel sections according to claim 2, wherein the input of the cross-sectional dimension prediction model includes attribute parameters indicating the steel type classification of the steel section. The method for predicting the cross-sectional dimensions of a steel section according to any one of claims 1 to 4, wherein the cross-sectional dimensions are at least one of the cross-sectional dimensions of the web thickness of the steel section, the thicknesses of the four flanges at the top, bottom, left and right, the web height and the flange width. A method for manufacturing shaped steel, comprising: identifying differences in rolling operation parameters that bring the cross-sectional dimensions of the shaped steel predicted using the method for predicting the cross-sectional dimensions of the shaped steel described in any one of claims 1 to 4 into the range of the target dimensions; and manufacturing the shaped steel under manufacturing conditions that include rolling operation parameters that are determined from the identified differences in the rolling operation parameters. A method for manufacturing shaped steel, which identifies differences in rolling operation parameters that will bring the cross-sectional dimensions of the shaped steel predicted using the method for predicting the cross-sectional dimensions of shaped steel described in claim 5 into the range of target dimensions, and manufactures the shaped steel under manufacturing conditions that include rolling operation parameters determined from the identified differences in rolling operation parameters. A section dimension prediction device for predicting a section dimension of a section steel manufactured by hot rolling, comprising:
a cross-sectional dimension prediction model which takes as input the difference in rolling operation parameters for two shaped steels produced by hot rolling with the same roll set and outputs the difference in cross-sectional dimension deviations of the two shaped steels, by inputting the difference between the rolling operation parameters of a shaped steel to be a precursor material produced by hot rolling with the roll set and the rolling operation parameters of a shaped steel to be produced by hot rolling with the roll set, and outputting the difference in cross-sectional dimension deviations;
The cross-sectional dimension deviation is a deviation between a target dimension and an actual dimension of a shaped steel,
A cross-sectional dimension prediction device for structural steel having a cross-sectional dimension prediction unit that predicts the cross-sectional dimension of the structural steel to be manufactured using the difference in the output cross-sectional dimension deviations and the actual value of the cross-sectional dimension of the preceding structural steel material. A machine learning model is trained using a plurality of data sets as training data, each set being a combination of actual values of differences in rolling operation parameters for two shaped steels produced by hot rolling with the same roll set and actual values of differences in cross-sectional dimensional deviations for the two shaped steels;
A cross-sectional dimension prediction model is generated, in which the difference between the rolling operation parameters of the two shaped steels is input and the difference between the cross-sectional dimension deviations of the two shaped steels is output;
A method for generating a cross-sectional dimension prediction model, in which the deviation between a target dimension and an actual dimension of a steel section is used as the cross-sectional dimension deviation.

Images