JP7626256B2 - Method for controlling width of rough rolled material and method for manufacturing hot rolled steel sheet - Google Patents

Description

The present invention relates to a method for predicting the width of rough rolled material in a hot rolling line, a method for controlling the width of rough rolled material, a method for manufacturing hot rolled steel plate, and a method for generating a width prediction model for rough rolled material.

In a hot rolling line, first, a slab, which is a steel material, is heated in a heating furnace, the width of the slab is adjusted by a width reduction press (sizing press), and one or more rough rolling mills are used for rough rolling to produce a semi-finished steel plate called a rough bar with a thickness of approximately 30 to 50 mm (hereinafter referred to as rough rolled material). Next, the head and tail ends of the rough rolled material are cut off by a crop shear, and the rough rolled material is finish rolled by a finishing mill consisting of 5 to 7 rolling stands capable of continuous rolling, producing a steel plate with a thickness of approximately 1.0 to 25.0 mm (hereinafter referred to as finish rolled material). Finally, the finish rolled material, which is in a high temperature state, is cooled by a cooling device on the run-out table, and then wound by a coiler (winder) to become a hot-rolled steel plate. In a hot rolling line, the width of a steel sheet varies in a complex manner during the manufacturing process of the hot rolled steel sheet, since the steel sheet is plastically deformed in the thickness direction and width direction in the width reduction press, rough rolling mill, and finish rolling mill. On the other hand, the width accuracy of the hot rolled steel sheet has a direct effect on the product yield. For this reason, in a hot rolling line, the width of the rough rolled material is controlled (rough width control) at the stage after rough rolling and before being charged into the finish rolling mill, and the width of the steel sheet is controlled (finish width control) during the process of passing through the finish rolling mill.

In hot rolling lines, the width of steel sheets changes due to various causes, so various techniques have been proposed to improve the width accuracy of hot-rolled steel sheets. For example, Patent Document 1 describes a method of predicting the amount of width change before and after rolling of a slab using a prediction model constructed based on the measured values of the width and temperature of the slab rolled by a rough rolling mill, and setting the opening of the edger based on the predicted amount of width change. Patent Document 1 also describes a method of improving the setting accuracy of the width reduction by combining the opening setting by an electric motor and the opening adjustment by hydraulic pressure for the opening setting of the edger, and adaptively correcting the parameters of the prediction model online based on the actual value of the slab width.

Patent Document 2 also discloses a method for estimating the width of a finish-rolled material using a width prediction model that shows the relationship between the width of the slab and actual data representing the rolling process and the width of the finish-rolled material, with the aim of controlling the width of the rough-rolled slab with an edger and controlling the width of the finish-rolled material to a target value. Patent Document 2 also describes a method for accumulating the deviation between the estimated value and the actual value of the width of the rolled material after finish rolling for the most recently rolled material, and correcting the target value of the width after finish rolling of the slab to be rolled next time. According to Patent Document 2, this makes it possible to accurately control the width of the rolled material in accordance with the variation in the width of the slab after continuous casting.

Patent Document 3 also describes a method for a rough rolling mill on a hot rolling line, which includes a learning means for learning the relationship between information on the rolled material at the entry side of the edger, roll diameter information for the edger and horizontal rolling mill, actual values of the width after edger rolling and the thickness after horizontal rolling, and the actual value of the width after horizontal rolling, and a prediction means for calculating a predicted value of the width of the rolled material after horizontal rolling in accordance with the knowledge obtained by the learning means, and sets the edger opening so that the difference between the predicted value of the width after horizontal rolling predicted by the prediction means and the target value is zero. Patent Document 3 also describes that the learning means can be configured by a neural network.

Japanese Patent Application Publication No. 7-303909 JP 2010-64103 A Japanese Patent Application Publication No. 9-225513

"Gaussian Processes and Machine Learning", by Daichi Mochihashi and Shigeyuki Ohba, published 2019/03/07 ISBN978-4-06-152926-7, Kodansha

However, the method described in Patent Document 1 uses a physical model as a prediction model to predict the slab width spread behavior caused by the edger and horizontal rolling mill. Therefore, a representative value of the slab width is calculated as the predicted value of the slab width, and it is not possible to predict the variation in the slab width. Furthermore, although Patent Document 1 describes that the slab width varies due to the error between the set edger opening and the actual value, it does not take into account the variation in the slab width due to factors other than those caused by the estimated accuracy of the set edger opening. Therefore, it is inevitable that the width of the rough rolled material will be too small or too large due to the variation in the temperature, etc. of the rough rolled material.

On the other hand, Patent Document 2 describes estimating the width change of the slab in the rough rolling mill using a physical model that represents the amount of width shrinkage by the edger and the amount of width expansion by the horizontal rolling mill. However, the physical model calculates a representative value of the width of the slab, and does not predict the variation in the width of the slab. Patent Document 2 also describes a method of correcting the target value of the width after rolling of the slab to be rolled next time based on the deviation between the estimated value and the actual value of the width of the rolled material after finish rolling. However, the variation in the width of the rough rolled material is caused not only by the variation in the width of the slab after continuous casting, but also by other factors. For this reason, it is not possible to eliminate the variation in the width of the rough rolled material, and it is inevitable that the width of the rough rolled material will be too small or too large.

Patent Document 3 also describes predicting the width of the rough rolled material using a learning means such as a neural network based on the operational performance data of the edger and horizontal rolling mill. However, the means for predicting the width of the rough rolled material calculates a representative value of the width of the rough rolled material, and does not predict the variation in the width of the rough rolled material. For this reason, it is inevitable that the width of the rough rolled material will be too small or too large due to variations in the temperature of the rough rolled material, etc.

As described above, conventional coarse width control focuses on specific phenomena that cause width variation in the roughly rolled material, and improves the width accuracy of the roughly rolled material by suppressing that variation. However, since width variation in the roughly rolled material also occurs due to, for example, variations in the temperature and deformation resistance of the slab, it is difficult to completely eliminate the width variation in the roughly rolled material. For this reason, the width of the roughly rolled material may become too small or too large, which inevitably leads to poor width accuracy of the roughly rolled material and a decrease in product yield.

The present invention has been made to solve the above problems, and an object of the present invention is to provide a rough rolled material width prediction method capable of predicting statistical information including the width variation of the rough rolled material. Another object of the present invention is to provide a rough rolled material width control method capable of accurately controlling the longitudinal width of the rough rolled material taking into account the width variation of the rough rolled material. Another object of the present invention is to provide a manufacturing method for hot rolled steel sheet capable of improving the product yield of the hot rolled steel sheet. Another object of the present invention is to provide a rough rolled material width prediction model generation method capable of generating a width prediction model that predicts statistical information including the width variation of the rough rolled material.

The method for predicting the width of rough rolled material according to the present invention is a method for predicting the width of rough rolled material in a hot rolling line including a heating furnace for heating slabs, a rough rolling mill for rough rolling the heated slab to produce rough rolled material, and a finishing rolling mill for finish rolling the rough rolled material to produce finished rolled material, and includes a prediction step of predicting statistical information of the width of the rough rolled material using a width prediction model learned by a Gaussian process regression technique, which includes one or more operational parameters selected from the operational parameters of the rough rolling mill as input data and has statistical information of the width of the rough rolled material as output data.

The hot rolling line includes a width reduction press device arranged upstream of the rough rolling mill, which intermittently reduces the width of the slab after it has been heated by the heating furnace, and the width prediction model may include one or more operating parameters selected from the operating parameters of the width reduction press device as input data.

The width prediction model may include one or more operating parameters selected from the operating parameters of the heating furnace as input data.

The width prediction model may include one or more parameters selected from attribute information of the slab as input data.

The rough rolled material width control method of the present invention includes a resetting step of predicting statistical information of the width of the rough rolled material using the rough rolled material width prediction method of the present invention, and resetting one or more operating parameters selected from the operating parameters of the rough rolling mill based on the predicted statistical information so as to reduce the probability that the width of the rough rolled material falls below the target width of the rough rolled material.

The statistical information on the width of the rough rolled material includes an average width _Wm and a standard deviation _Wσ of the width of the rough rolled material, and the resetting step may include a step of setting one or more operation parameters selected from operation parameters of a rough rolling mill so that the target width _Wt of the rough rolled material satisfies the relationship shown in the following formula (1).

The method for manufacturing hot-rolled steel sheet according to the present invention includes a step of manufacturing hot-rolled steel sheet using the width control method for rough rolled material according to the present invention.

The method for generating a width prediction model of rough rolled material according to the present invention is a method for generating a width prediction model for predicting the width of the rough rolled material in a hot rolling line including a heating furnace for heating slabs, a rough rolling mill for rough rolling the heated slab to produce a rough rolled material, and a finishing rolling mill for finish rolling the rough rolled material to produce a finish rolled material, and includes a learning data acquisition step for acquiring multiple learning data including one or more operational performance data selected from operational performance data of the rough rolling mill and performance data on the width of the rough rolled material, and a step for generating the width prediction model using a Gaussian process regression technique using the multiple learning data acquired in the learning data acquisition step, the input performance data including one or more operational performance data selected from the operational performance data of the rough rolling mill, and output data including statistical information on the width of the rough rolled material.

According to the rough rolled material width prediction method of the present invention, it is possible to predict statistical information including the width variation of the rough rolled material. Furthermore, according to the rough rolled material width control method of the present invention, it is possible to accurately control the longitudinal width of the rough rolled material taking into account the width variation of the rough rolled material. Furthermore, according to the hot rolled steel sheet manufacturing method of the present invention, it is possible to improve the product yield of the hot rolled steel sheet. Furthermore, according to the rough rolled material width prediction model generation method of the present invention, it is possible to generate a width prediction model that predicts statistical information including the width variation of the rough rolled material.

FIG. 1 is a schematic diagram showing an example of the configuration of a hot rolling line to which the present invention is applied. FIG. 2 is a schematic diagram showing an example of the configuration of the heating furnace shown in FIG. FIG. 3 is a schematic diagram of the inside of the heating furnace shown in FIG. 1 as viewed from above. FIG. 4 is a schematic diagram showing a configuration example of the width reduction press device shown in FIG. FIG. 5 is a diagram showing the configuration of a rolling stand constituting the roughing mill shown in FIG. FIG. 6 is a schematic diagram for explaining an optical width measuring method. FIG. 7 is a flowchart showing the flow of the model generation steps according to one embodiment of the present invention. FIG. 8 is a flowchart showing the flow of the prediction step according to one embodiment of the present invention. FIG. 9 is a block diagram showing the configuration of a width prediction model generating unit according to an embodiment of the present invention. FIG. 10 is a diagram for explaining a method of setting a control target width in a conventional width control method. FIG. 11 is a block diagram showing the configuration of a width prediction unit according to an embodiment of the present invention. FIG. 12 is a diagram showing the relationship between the probability density distribution of the rough output side width predicted by the width prediction model and the target width. FIG. 13 is a diagram for explaining a method for setting a control target value for coarse width control in this embodiment. FIG. 14 is a diagram showing the relationship between the probability density distribution of the rough output side width predicted by the width prediction model and the target width. FIG. 15 is a diagram showing the relationship between the deviation of the rough output side width from the target width and the amount of the steel sheet to be cut off.

Below, with reference to the drawings, a method for predicting the width of rough rolled material, a method for controlling the width of rough rolled material, a method for manufacturing hot-rolled steel plate, and a method for generating a width prediction model for rough rolled material, which are one embodiment of the present invention, are described in detail.

[Hot rolling line]
First, the configuration of a hot rolling line to which the present invention is applied will be described with reference to Figs.

Figure 1 is a schematic diagram showing an example of the configuration of a hot rolling line to which the present invention is applied. As shown in Figure 1, the hot rolling line 1 to which the present invention is applied is equipped with a heating furnace 2, a descaling device 3, a width reduction press device 4, a rough rolling mill 5, a finishing rolling mill 6, a cooling device 7, and a coiler (winder) 8. A cast slab (not shown) is charged into the heating furnace 2, heated to a predetermined set temperature, and extracted from the heating furnace 2 as a hot slab. The hot slab extracted from the heating furnace 2 has the primary scale formed on the surface removed by the descaling device 3, and then the width is reduced to a predetermined set width by the width reduction press device 4. The width-reduced slab is then rolled to a predetermined thickness in the rough rolling mill 5 to become a rough rolled material, and is transported to the finishing rolling mill 6. In the finishing rolling mill 6, the rough rolled material is rolled to a product thickness by a continuous rolling mill consisting of 5 to 7 rolling stands to become a finish rolled material. A cooling device 7 is provided in a facility called a run-out table downstream of the finishing rolling mill 6, and the finish rolled material is cooled to a predetermined temperature and then wound into a coil by a coiler 8. In addition, a plurality of width gauges are installed as width measuring means in the middle of the conveying process of the hot rolling line 1. In the example shown in FIG. 1, a rough exit width gauge 11 is installed at the exit side of the rough rolling mill 5, and a finish exit width gauge 12 is installed at the exit side of the finishing rolling mill 6. In addition, a pre-coiler width gauge (coiler entry width gauge) 13 that measures the width of the steel sheet before coiling is installed at the exit side of the cooling device 7. Hereinafter, the width of the rough rolled material measured by the rough exit width gauge 11 may be referred to as the rough exit width, the width of the finish rolled material measured by the finish exit width gauge 12 may be referred to as the finish exit width, and the width of the steel sheet measured by the pre-coiler width gauge 13 may be referred to as the pre-coiler width.

The hot rolling line 1 is equipped with a programmable logic controller (PLC) 90 that controls each device constituting the hot rolling line 1, a control computer (process computer) 91 that gives control commands to the control controller 90, and a host computer 92 that gives manufacturing instructions to the hot rolling line 1. The width control of the steel plate in the hot rolling line 1 is performed by the control computer 91 setting the control target value of the rough exit width (rough control target width), the control target value of the finish exit width (finish control target width), and the control target value of the pre-coiler width (pre-coiler control target width) based on the host computer 92 or manufacturing instructions from the host computer 92, and setting the operating conditions of the rough rolling mill 5 and the finish rolling mill 6. Specifically, the host computer 92 or the control computer 91 sets the target width (finishing target width) of the finish rolled material, taking into consideration the amount of width change of the steel sheet that occurs between the exit side of the finishing mill 6 and the pre-coiler width measuring device 13, based on the pre-coiler target width (pre-coiler target width) determined from the product specifications of the hot-rolled steel sheet. Furthermore, the host computer 92 or the control computer 91 sets the rough target width (hereinafter sometimes simply referred to as the target width), which is the target width of the rough rolled material, based on the set finishing target width, taking into consideration the amount of width change of the steel sheet at the finishing mill 6. In this case, the target values of the width control (rough control target width, finishing control target width, pre-coiler control target width) may be set by providing an extra width (margin) in advance for the rough target width, the finishing target width, and the pre-coiler target width. Then, the host computer 92 or the control computer 91 sets the rolling conditions in each pass of the rough rolling so that the rough exit width coincides with the rough control target width. The host computer 92 or the control computer 91 also sets the rolling conditions in each rolling stand of the finish rolling mill 6 so that the finish exit width coincides with the finish control target width. Furthermore, the host computer 92 or the control computer 91 may set the tension between the finish rolling mill 6 and the coiler 8 and the cooling conditions of the cooling device 7 so that the coiler pre-width coincides with the coiler pre-width. In this case, the finish rolling mill 6 may perform dynamic width control while referring to the actual measured values of the rough exit width and the finish exit width. The control controller 90 has a function of collecting information acquired from various sensors (thickness gauge, thermometer, etc.) in addition to information acquired from a width gauge installed in the hot rolling line 1 at a predetermined sampling period and outputting them to the control computer 91.

One embodiment of the present invention, a method for predicting the width of a rough rolled material, is a method for predicting the rough exit side width. Another embodiment of the present invention, a method for controlling the width of a rough rolled material, is a method for controlling the width of a rough rolled material so that the rough exit side width satisfies a predetermined relationship with respect to a rough target width.

[Heating furnace]
FIG. 2 is a schematic diagram showing an example of the configuration of the heating furnace 2 shown in FIG. 1. As shown in FIG. 2, in this embodiment, the slab SA is charged into the heating furnace 2 from the left side of FIG. 2. The temperature of the slab SA charged into the heating furnace 2 may be cooled to about room temperature in the slab yard after casting, or may be about 600 ° C during cooling. In addition, the slab SA may be charged at a temperature of about 600 to 800 ° C without passing through the slab yard after casting. The inside of the heating furnace 2 is divided into a plurality of zones, and generally, on the upstream side, a heating zone divided into 2 to 8 zones and 1 to 3 soaking zones are provided. In the example shown in FIG. 2, five heating zones and one soaking zone are provided, and here, both are collectively referred to as the "heating furnace zone". Each heating furnace zone is set to a different atmospheric temperature so that the average temperature of the slab SA charged into the heating furnace 2 gradually increases to a predetermined target heating temperature (the target value of the average temperature of the slab SA when it is extracted from the heating furnace 2). Further, a thermometer 21 for measuring the atmospheric temperature in each heating furnace zone is provided at the upper portion of each heating furnace zone.

Figure 3 is a schematic diagram of the inside of the heating furnace 2 shown in Figure 1, seen from above. As shown in Figure 3, the slab SA loaded into the heating furnace 2 passes through each heating furnace zone in sequence inside the heating furnace 2 by a transport device called a walking beam 22. In addition, multiple slabs SA are loaded into the heating furnace 2 at the same time, and are extracted from the extraction side outlet of the heating furnace 2 in the order in which they were loaded into the heating furnace 2, and hot rolling is performed. The inside of the walking beam 22 is water-cooled, and there are parts where the temperature rise of the slab SA is locally inhibited by a member called a skid that is in direct contact with the slab SA. The part of the slab SA in contact with the skid is called a skid mark, and has a lower temperature than the part not in contact with the skid. Skid marks are one of the causes of variation in the rough output side width.

[Width reduction press device]
Fig. 4 is a schematic diagram showing a configuration example of the width reduction press device 4 shown in Fig. 1. After the primary scale formed on the surface of the slab SA heated by the heating furnace 2 is removed by the descaling device 3, the slab SA is width-reduced to a predetermined set width by the width reduction press device 4. As shown in Fig. 4, the width reduction press device 4 includes a pair of width reduction dies 41. The width reduction dies 41 reduce the slab SA in the width direction. The width reduction press device 4 drives the width reduction dies 41 by a drive device 42 while transporting the slab SA, and intermittently reduces the width of the slab SA from both sides in the width direction of the slab SA.

In the width reduction press device 4, the slab SA is transported using pinch rolls 43 and the like. The width reduction press device 4 can change the feed pitch of the slab SA between width reduction passes by changing the driving amount of the pinch rolls 43. The feed pitch means the transport distance of the slab SA for each width reduction pass in the width reduction press device 4. The driving amount of the pinch rolls 43 is controlled by a controller 90 that controls the width reduction press device 4. On the surface of the width reduction die 41 that contacts the slab SA, a parallel portion 41a parallel to the transport direction of the slab SA and an inclined portion 41b that expands in the width direction in the opposite direction to the transport direction of the slab SA are formed in sequence from the tip side of the transport direction of the slab SA. In the width reduction die 41, one or more parallel portions 41a may be provided between the inclined portions 41b in order to suppress the occurrence of slippage with respect to the slab SA. The shape of the width reduction die 41 changes the deformation state of the slab SA and affects the width of the rough output side.

[Roughing mill]
Return to FIG. 1. The roughing mill 5 includes a reversible mill 5a capable of reverse rolling and a non-reversible mill 5b capable of rolling only in the downstream conveying direction. The arrows (solid lines) illustrated under the roughing mill 5 shown in FIG. 1 represent rolling passes (rolling passes for thinning the thickness). In the reversible mill 5a, about 5 to 11 rolling passes are usually performed in the reversible direction (from the upstream side to the downstream side or from the downstream side to the upstream side). In the final rolling pass, rolling and conveying to the next rolling mill are performed simultaneously, so the number of rolling passes of the reversible mill is always an odd number, and the steel sheet is conveyed to the rolling mill located downstream while being rolled. At this time, the time from when the tail end of the steel sheet leaves the rolling mill in the current rolling pass until the rolling direction is reversed and the steel sheet is bitten into the rolling mill in the next rolling pass is called the inter-pass time of the reversible passes. In addition, in the final rolling pass of the reversing passes, the time from when the tail end of the steel sheet leaves the reversing mill 5a until it is engaged in the non-reversing mill 5b is called the interpass time of successive passes. Furthermore, the interpass time of the reversing passes and the interpass time of the successive passes are collectively called the interpass air-cooling time. The interpass air-cooling time indicates the time during which the slab is air-cooled during transportation, and affects the temperature change of the steel sheet.

Figure 5 is a schematic diagram showing the configuration of the rolling stand constituting the roughing mill 5 shown in Figure 1. As shown in Figure 5, the roughing mill 5 is equipped with a horizontal rolling mill 51 that reduces the thickness of the steel plate SB and an edger (vertical rolling mill) 52 that reduces the width of the steel plate SB. The edger 52 is a rolling mill in which a pair of rolls are arranged vertically, and is installed adjacent to the horizontal rolling mill 51. The width reduction (width rolling) of the steel plate SB using the edger 52 is usually performed before the horizontal rolling in each rolling pass of the rough rolling. Therefore, when one edger 52 is arranged for the reversing rolling mill 5a, the width reduction by the edger 52 is performed only in the forward direction, i.e., the odd-numbered passes, and not in the even-numbered passes, i.e., the reverse passes. However, when edgers 52 are arranged on both sides of the reversing rolling mill 5a, the width reduction may be performed in either rolling pass. The roll opening (roll gap) of the horizontal rolling mill 51 and the opening of the edger 52 in each pass of rough rolling are set by the control computer 91. The rough rolling mill 5 is provided with a descaling header for spraying descaling water toward the steel sheet SB, and descaling is performed at the inlet side of the horizontal rolling mill 51. However, descaling water is not necessarily sprayed in all rolling passes of rough rolling, and a predetermined descaling pattern is set according to the material of the steel sheet SB, for example, spraying only in the first pass or spraying only in odd passes. The steel sheet SB in a state in which all rolling passes preset by the rough rolling mill 5 have been completed is called a rough bar or sheet bar, and is referred to as a rough rolled material in this specification.

[Width meter]
Return to FIG. 1. The width of the steel sheet in the hot rolling line 1 is measured by a rough-exit side width gauge 11, a finish-exit side width gauge 12, and a pre-coiler width gauge 13. An optical width measuring method is often used for these width gauges. In the optical width measuring method, a light source is arranged below the pass line along which the steel sheet is transported, and an image sensor is arranged above the pass line, and the width of the steel sheet is measured based on the shadow length of the steel sheet in the width direction of the light emitted from the light source while the steel sheet passes through. Some width gauges use a camera to identify the position of the width direction end of the steel sheet and measure the width. FIGS. 6(a) and (b) are schematic diagrams showing a configuration example of a rough-exit side width gauge using a camera. In the example shown in FIG. 6(a), a pair of cameras 15a and 15b provided in the rough-exit side width gauge take an image of the steel sheet SB including the width direction end of the steel sheet SB. CMOS or CCD sensors are used for the cameras 15a and 15b. Then, an image processing unit provided in the rough-out side width gauge identifies the positions of the widthwise ends of the steel sheet SB from the images captured by the cameras 15a and 15b, and calculates the width W of the steel sheet S based on the installation interval of the cameras 15a and 15b. Reference numeral 14 in the figure indicates a pass line. However, since this width gauge images the widthwise ends of the steel sheet SB from an oblique direction, a width measurement error is likely to occur when the steel sheet SB floats up from the pass line 14. Therefore, the width gauge often has a function of correcting the width measurement error in response to the floating of the steel sheet SB from the pass line. Specifically, as shown in FIG. 6(b), when the steel sheet SB is transported at a height of floating amount H from the pass line 14, the width measurement error is corrected as follows. That is, first, a pair of cameras 15a and 15b arranged in the width direction of the steel sheet SB identifies both widthwise ends of the steel sheet SB, thereby identifying the measured width value W1. Next, another pair of cameras 15c, 15d arranged in the width direction of the steel plate SB also identify both ends of the steel plate SB in the width direction, thereby identifying a measured width value W2. Then, based on the positional relationship between the two pairs of cameras (in the example shown in FIG. 6(b), the distance D between the cameras 15a, 15b in the width direction and the distance L between the cameras 15a (15b) and 15c (15d) in the width direction), the actual width W of the steel plate SB is calculated by the following formula (2), which becomes the measured width value of the steel plate SB.

However, as another width measurement method, a method is sometimes used in which a laser beam is emitted in the width direction of the steel plate, the reflected light from the end faces of the steel plate is received, and the width of the steel plate is measured based on the distance to both end faces of the steel plate. Some width gauges are equipped with a thermal expansion correction function that converts the width of the steel plate after cooling based on the temperature of the steel plate. Since the width of the steel plate is measured using a width gauge during the steel plate transportation process, the width measurement value of the steel plate obtained by the width gauge becomes time-series numerical information corresponding to the sampling pitch of the width gauge. In addition, the relationship between the longitudinal position of the steel plate and the actual value of the width of the steel plate is converted using information on the transport speed of the steel plate when it passes the position of the width gauge. Then, in the control computer 91, a representative value of the width of the steel plate is calculated based on the acquired actual width measurement value of the steel plate. The representative value of the width of a steel plate may be, for example, the average value of the width of the steel plate in the longitudinal direction of the steel plate (average width), the measured value of the width of the steel plate in the steady portion excluding the leading and trailing ends of the steel plate (steady width), the measured value of the width of the steel plate at the leading end of the steel plate (leading end width), or the measured value of the width of the steel plate at the trailing end of the steel plate (tailing end width). In addition, the minimum value (minimum width) or maximum value (maximum width) of the width of the steel plate in the longitudinal direction of the steel plate may be calculated. The width of the steel plate measured by a width gauge may be expressed as a deviation from a preset target width.

[Gaussian process regression]
A rough rolled material width prediction method according to one embodiment of the present invention predicts the rough exit side width in the hot rolling line 1. The rough rolled material width prediction method according to one embodiment of the present invention uses a width prediction model trained by a Gaussian process regression technique, which includes one or more operation parameters selected from the operation parameters of the rough rolling mill 5 as input data and uses statistical information on the rough exit side width as output data. The Gaussian process regression technique applied to the rough rolled material width prediction method according to one embodiment of the present invention will be described below.

Gaussian process regression, also called Gaussian process regression or Gaussian process, is a type of nonlinear regression model that estimates a function from an input variable to an output variable. The output is a probability distribution, and the use of a Gaussian distribution specified by two parameters, the mean value and the variance, is called a Gaussian process, and the probability distribution can be obtained using a Bayesian estimation technique to represent the reliability and uncertainty of the estimation. For example, when m variables are selected as inputs to a width prediction model, the input variables are represented by an input vector x. Also, the output variable associated with the input vector x as learning data is represented by y. Hereinafter, a method of obtaining statistical information y ^* of the rough output side width for a new input vector x ^* by the Gaussian process regression technique using n pieces of learning data x ⁽¹⁾ to x ⁽ⁿ⁾ , y ^{(1) to} y(n) will be specifically described. Since the m variables constituting the input vector x each represent a different physical quantity, the m variables may be standardized (normalized) in advance. Specifically, for each of the m variables, the mean value and standard deviation may be calculated from the n pieces of learning data, and each variable may be standardized using the calculated mean value and standard deviation. This is because standardizing the m variables in advance makes learning of hyperparameters, which will be described later, more efficient. In this case, to convert the m variables into physical quantities, it is sufficient to perform an inverse conversion using the calculated mean value and standard deviation.

In Gaussian process regression, a probability model is used in which a function f(x) representing an output variable y is estimated from an input variable x, and Gaussian noise is added to this. For example, the probability model is expressed as the following formula ( ³ ). In this case, it is assumed that the function f(x) follows a multivariate Gaussian distribution. It is also assumed that the Gaussian noise ε ⁽ⁱ⁾ follows a Gaussian distribution with a mean value of zero and a variance of σ _e (i)2. However, the Gaussian noise ε ⁽ⁱ⁾ may be a value that depends on the n pieces of training data x ⁽¹⁾ to x ⁽ⁿ⁾ , or may be a constant noise that does not depend on the n pieces of training data x ⁽¹⁾ to x ⁽ⁿ⁾ .

The Gaussian distribution refers to a distribution whose probability density N is expressed by the following formula (4). In formula (4), μ indicates the mean value, and σ indicates the standard deviation ( ^σ2 is the variance). In other words, the Gaussian distribution is a probability density specified by the mean value μ and the standard deviation σ or variance ^σ2 . Gaussian process regression uses a multivariate normal distribution, which is a multidimensional extension of such a Gaussian distribution.

In Gaussian process regression, the mean function (mean vector) representing the multivariate Gaussian distribution is a constant (for example, zero), and the covariance matrix is expressed by a kernel function. The kernel function is a function for calculating the similarity of data. The kernel function is expressed as k(x( ⁱ⁾ , x ^(j) ) using input vectors x( ⁱ⁾ and x ^(j ) as arguments, and outputs the similarity between input vectors x ⁽ⁱ⁾ and x ^(j) . Known kernel functions such as a white kernel, a linear kernel, a polynomial kernel, a Gaussian kernel, and a Mater kernel can be used as the kernel function. Some examples of kernel functions are expressed as shown in the following formulas (5) to (7) using a parameter θ. Formula (5) shows a linear kernel, formula (6) shows a quadratic polynomial kernel, and formula (7) shows a Gaussian kernel.

Based on the above assumptions, the function f(x) that represents the output variable y from the input variable x is expressed as shown below in equation (8) using the probability density N.

In this case, when the Gaussian noise is a constant value σ _e ² regardless of the learning data, if the covariance matrix K _n specified by the kernel function and the covariance matrix Σ _n including the Gaussian noise are defined by the following formulas (9) and (10), formula (8) can be expressed as the following formula (11) or formula (12), where I represents a unit matrix.

The covariance matrices K _n and Σ _n included in the right-hand sides of the formulas (11) and (12) include the learning data x ⁽¹⁾ to x ⁽ⁿ⁾ that are input, and the left-hand sides of the formulas (11) and (12) include the learning data y ⁽¹⁾ to y ⁽ⁿ⁾ that are output. Therefore, the hyperparameters (parameter θ and Gaussian noise σ _e ² ) included in the kernel function may be determined so that the relationship shown in the formula (11) or the formula (12) holds. The method of determining the hyperparameters may be selected from among known methods. For example, the hyperparameters may be calculated by calculating the likelihood function of the learning data and maximizing the log-likelihood represented by the logarithm of the calculated likelihood function. In this case, the calculation method for maximizing the log-likelihood may be an optimization method such as the Monte Carlo method or the conjugate gradient method. In addition, a method such as a cross-validation method or a marginal likelihood maximization method may be used.

Next, a method for estimating statistical information y ^* of the rough output width for a new input vector x ^* using a function f(x) for which the hyperparameters of the kernel function have been determined will be described. An estimate for an unknown input vector x ^* that is not included in the learning data can be expressed by applying Bayesian estimation as shown in the following formula (13). In this case, a vector of a newly specified kernel function k ^* is defined as shown in the following formula (14).

As a result, formula (13) can be expressed as formula (15) shown below. Then, statistical information y ^* of the coarse output side width for the input vector x ^* can be calculated by formulas (16) and (17) shown below, with the average value being _Wm and the variance being _Wσ . For details about the Gaussian process regression technique, refer to known literature (e.g., Non-Patent Document 1).

In this embodiment, the step of specifying the hyperparameters so as to express the relationship of the above formula (11) or formula (12) is called the model generation step. Specifically, in the model generation step, as shown in FIG. 7, first, n pieces of learning data x ⁽¹⁾ to x ⁽ⁿ⁾ , y ⁽¹⁾ to y ⁽ⁿ⁾ are acquired from a data set stored in a database or the like (step S1). Next, a kernel function to be used for machine learning is selected from, for example, formulas (5) to (7) (step S2). Next, a Gaussian distribution with a mean value of zero and a variance of σ _e ² is set as the Gaussian noise ε(i) (step S3). Next, the covariance matrices K _n and Σ _n specified by the kernel function are calculated using formulas (9) and (10) (step S4). Then, using formulas (11) and (12) that express the relationship between the input and output of the learning data, the parameter θ and the Gaussian noise σ _e ² are determined as hyperparameters by a learning method using a likelihood function (step S5). A kernel function that expresses the input/output relationship of the learning data is specified by the hyperparameters determined in this way. The determined hyperparameters may be stored in a storage device of a computer that executes the model generation step. On the other hand, a step of calculating an average value W _m and a standard deviation W _σ as statistical information y ^* of the rough output side width for the input vector x ^* using the hyperparameters determined by the model generation step is called a prediction step. Specifically, in the prediction step, as shown in FIG. 8, a new input vector x ^* is first obtained (step S11). Next, the hyperparameters stored in the storage device of the computer that executes the model generation step are obtained, and a kernel function k ^* for the new input vector x ^* is specified using Equation (14) (step S12). Then, using the relationship shown in Equation (16) and Equation (17), a predicted value of the statistical information y ^* of the rough output side width for the input vector x ^* is calculated as an average value W _m and a standard deviation W _σ .

[Method of generating width prediction model]
Next, an embodiment in which the above-mentioned Gaussian process regression technique is applied will be described as a method for generating a width prediction model for a roughly rolled material according to one embodiment of the present invention.

Figure 9 is a block diagram showing the configuration of a width prediction model generation unit according to one embodiment of the present invention. As shown in Figure 9, the width prediction model generation unit 100 of this embodiment includes a database unit 101 and a machine learning unit 102. The database unit 101 accumulates one or more operation performance data selected from the operation performance data of the rough rolling mill 5 and performance data of the rough output side width. The database unit 101 may accumulate, as necessary, one or more operation performance data selected from the operation performance data of the width reduction press device 4, one or more operation performance data selected from the operation performance data of the heating furnace 2, and one or more performance data selected from the performance data of the attribute information of the slab SA. The specific performance data accumulated in the database unit 101 will be described later.

The performance data stored in the database unit 101 can be appropriately acquired from the control controller 90, the control computer 91, or the host computer 92. In addition, a data acquisition unit 103 may be provided to collect these performance data, and the performance data may be temporarily stored in the data acquisition unit 103, and a data set in which multiple types of performance data are associated may be generated and then accumulated in the database unit 101. Since the data stored in the database unit 101 may have different acquisition timings, it is easy to configure a data set in which the multiple types of performance data are associated with each other in the data acquisition unit 103. For the data set stored in the database unit 101, at least one performance data is acquired for one steel plate manufactured from one slab. For example, when the average width of the steel plate is used as the performance data of the rough output side width, the representative value of the operation performance data of the rough rolling mill 5 may be used as the performance data. In this case, the representative value for one steel plate may be used as the performance data for the operation performance data of the width reduction press device 4, the operation performance data of the heating furnace 2, and the performance data of the attribute information of the slab SA.

On the other hand, the data acquisition unit 103 may generate multiple data sets for one steel plate manufactured from one slab, and store them in the database unit 101. For example, when acquiring performance data on the width of three locations, the front end, steady state, and tail end of the rough rolled material, as performance data on the rough exit side width, the operation performance data acquired at the front end, steady state, and tail end of the rough rolled material may be associated with the performance data on the rough exit side width at the corresponding positions for the operation performance data of the rough rolling mill 5. However, for performance data that is specified regardless of the longitudinal position of the rough rolled material, such as performance data of the attribute information of the slab, the performance data of the same attribute information is associated with the performance data on the rough exit side width at the front end, steady state, and tail end of the rough rolled material.

Furthermore, the data acquisition unit 103 may acquire actual data on the rough exit side width for each position divided in the longitudinal direction for one rough rolled material, and the operational performance data acquired for each position in the longitudinal direction of the rough rolled material may be stored in the database unit 101 in association with the actual data on the rough exit side width measured at each position. That is, the number of divisions in the longitudinal direction of the rough rolled material is set to, for example, about 20 to 200, and the actual data on the rough exit side width in each division section and the operational performance data corresponding to each position are assigned. In this case, although the length of the steel plate SB rough rolled by the rough rolling mill 5 changes for each rough rolling pass, if operational performance data is acquired at a position corresponding to the division in the longitudinal direction of the rough rolled material, a data set corresponding to each division section can be configured in the data acquisition unit 103. When a data set corresponding to multiple positions divided in the longitudinal direction of the rough rolled material is accumulated in the database unit 101, the machine learning unit 102 can also generate a different width prediction model for each position in the longitudinal direction of the rough rolled material.

The width prediction model generating unit 100 can be provided in a control computer 91 for controlling the production of steel sheets by the hot rolling line 1. The width prediction model generating unit 100 may be provided in a host computer 92 that gives production instructions to the control computer 91, or in an independent computer that can communicate with other devices. The machine learning unit 102 may be configured in a device separate from the database unit 101 using a device that can receive the data set stored in the database unit 101. The database unit 101 stores 100 or more data sets. Preferably, 10,000 or more data sets, more preferably 100,000 or more data sets, are stored in the database unit 101. Screening may be performed on the data stored in the database unit 101 as necessary.

The machine learning unit 102 uses the data set stored in the database unit 101 to generate a width prediction model M through machine learning using the Gaussian process regression technique. The learning data used by the machine learning unit 102 are multiple data sets including one or more operation performance data selected from the operation performance data of the roughing rolling mill 5 and performance data of the rough exit side width, which are stored in the database unit 101. The machine learning unit 102 uses the learning data to perform machine learning using the Gaussian process regression technique, which includes one or more operation performance data selected from the operation performance data of the roughing rolling mill 5 as input performance data and uses statistical information of the rough exit side width as output data, to generate a width prediction model M. In addition, the machine learning unit 102 may use the data set accumulated in the database unit 101 to perform machine learning using a Gaussian process regression technique using, as input performance data, one or more operation performance data selected from the operation performance data of the width reduction press device 4, one or more operation performance data selected from the operation performance data of the heating furnace 2, and one or more performance data selected from the performance data of the attribute information of the slab SA, to generate a width prediction model M.

In this case, machine learning refers to identifying hyperparameters to be applied to Gaussian process regression by the model generation step shown in FIG. 7. The width prediction model M refers to the hyperparameters identified in this manner. By identifying the hyperparameters, the input/output relationship of the learning data is identified, and it becomes possible to predict statistical information on the rough exit side width for an unknown input. In this embodiment, the statistical information on the rough exit side width that is the output of the width prediction model M learned by Gaussian process regression includes the predicted result of the average value of the width of the rough rolled material and the standard deviation or variance, which is an index representing the variance thereof. This makes it possible to predict the variance of the rough exit side width together with the average value of the rough exit side width.

On the other hand, in the conventional technology, as described in Patent Document 3, for example, the average value or representative value of the width change of the rough rolled material is predicted based on a database generated using actual data on the width change of the rough rolled material. However, in the conventional technology, information on the variation in the width change of the rough rolled material cannot be obtained. For this reason, the target value of the width of the rough rolled material must be set by adding a preset extra width (margin). Specifically, in the conventional width control method, as shown in FIG. 10, an extra width (margin) _Wr is set in advance for the target width (target width) _Wt of the rough exit side width, and the sum of the target width _Wt and the extra width _Wr is set as the control target value _Wc of the rough exit side width. The reason for setting the extra width _Wr is to prevent the rough exit side width from falling below the target width _Wt , since a certain variation occurs in the rough exit side width. If the rough exit side width falls below the target width Wt, the width of the steel plate after finish rolling may fall below the product target width, and the product cannot be obtained due to the insufficient width. Therefore, the hot-rolled steel sheet produced may be scrapped or the shipping destination of the product may be changed, resulting in a decrease in product yield and a delay in delivery. On the other hand, if the set extra width _Wr is too large, the width of the steel sheet after finish rolling becomes excessively large compared to the product target width, and it becomes necessary to trim the steel sheet to obtain the product, which reduces the product yield. That is, if an attempt is made to prevent the occurrence of insufficient width as a hot-rolled steel sheet, the rough-end width becomes excessive, and if an attempt is made to suppress the decrease in yield due to the edge trimming, the width shortage is likely to occur. For this reason, in the conventional technology, actual data on the rough-end width is collected for each thickness and width classification of the hot-rolled steel sheet, the extra width is set in advance according to the variation, and the factory personnel periodically monitor whether the setting of the extra width is appropriate.

In contrast, according to the present embodiment, the average value and statistical variation of the rough exit width can be predicted according to the input operating conditions of the hot rolling line, so that an appropriate excess width can be set according to the operating conditions of each rough rolled material, rather than for each thickness or width category of the hot rolled steel sheet as in the conventional technology. This makes it possible to suppress a decrease in product yield due to insufficient width or excess width of the hot rolled steel sheet caused by the variation in the rough exit width.

[Slab attribute information]
The attribute information of the slab that can be used as an input for the width prediction model M refers to information on the slab dimensions that affect the width change of the slab in the width reduction press device 4 and the rough rolling mill 5, and information on the composition of the slab. The information on the slab dimensions is information on the thickness, width, length, and weight of the slab. The information on the composition of the slab is information on the content of the components contained in the slab, such as the C content, Si content, Mn content, P content, S content, Nb content, Ti content, Cu content, Ni content, Mo content, and B content of the slab. The information on the slab dimensions affects the temperature change of the slab in the hot rolling line 1, and therefore affects the variation in the rough output side width. In addition, the information on the composition of the slab affects the deformation resistance of the slab and the composition and thickness of the oxide film generated on the surface of the slab. This affects the friction force at the interface between the rolling roll and the slab, and changes the deformation state of the slab, which affects the variation in the rough output side width.

[Heating furnace operating parameters]
The operation parameters of the heating furnace that can be used as inputs to the width prediction model M are parameters that represent the operation conditions of the heating furnace 2 when the slab is heated in the heating furnace 2, and refer to information that influences the width change of the slab in the width reduction press device 4 and the rough rolling mill 5. The operation parameters of the heating furnace 2 can include the temperature of the slab when it is charged into the heating furnace 2, the residence time of the slab in a specific heating furnace zone in the heating furnace 2, the atmospheric temperature of the final heating furnace zone of the heating furnace 2, and the temperature of the slab extracted from the heating furnace 2. These parameters affect the temperature drop during rough rolling of the slab, and therefore affect the rough exit width.

Information on the charging position of the slab in the heating furnace 2 and the positional relationship between the slab and other slabs in the heating furnace 2 may also be used. The charging position of the slab in the heating furnace 2 may be information on the furnace entry position P, which represents the distance between the walking beam 22 located at one end of the heating furnace 2 and the longitudinal end of the slab SA, as shown in FIG. 3. The furnace entry position P changes the position of the skid mark in the longitudinal direction of the slab SA, which affects the width fluctuation in the longitudinal direction of the slab SA, and thus affects the variation in the rough output side width. For the charging position of the slab SA, the distance D1 between the longitudinal end of the slab SA and the furnace wall of the heating furnace 2, and a parameter representing the interval between the walking beam (fixed skid) 22 or the moving skid 23 of the heating furnace 2 in which the slab SA is charged may be used. In addition, the distance (charging interval) D2 between the slab and other adjacent slabs in the heating furnace 2 may be used as information on the positional relationship between the slab and other slabs in the heating furnace 2. When the charging interval D2 changes in the heating furnace 2, the temperature of the end face of the slab SA changes. This affects the temperature distribution of the slab SA in the hot rolling line 1, and affects the variation in the rough output side width. The information on the positional relationship between the slab and other slabs in the heating furnace 2 may include the length or thickness of the other slab, and the difference in distance between the tip of the slab to be predicted and the tip of the other slab to the wall of the heating furnace.

[Operation parameters of width reduction press device]
The operation parameters of the width reduction press device 4 that can be used as input for the width prediction model M refer to the operation conditions when performing width reduction on a heated slab. The operation parameters of the width reduction press device 4 can be operation parameters related to the width reduction amount of the slab. The operation parameters related to the width reduction amount of the slab include the width reduction amount of a representative position in the longitudinal direction of the slab and the feed pitch of the slab between width reduction passes. The operation parameters related to the width reduction amount of the slab may include the width reduction start position, which is the length at which the parallel part 41a of the width reduction die 41 contacts the slab in the first width reduction pass on the front end of the slab. The operation parameters related to the width reduction amount of the slab affect the dogbone shape (thickness distribution in the width direction) formed in the slab after width reduction, and thus affect the variation in the rough output side width. In addition, even if the width reduction amount of the slab is constant in the longitudinal direction of the slab, the dogbone shape is different at the steady part, front end, and tail end of the slab. This causes the average width of the rough end to fluctuate, affecting the variation of the rough end width. Even if the feed pitch of the slab between width reduction passes is constant, a difference in the dog-bone shape occurs between the stationary part and the leading and trailing ends of the slab. This affects the variation of the rough end width.

Furthermore, the operation parameters of the width reduction press device 4 can be operation parameters related to the die shape applied to the width reduction press device 4. The operation parameters related to the die shape are representative values fixed with respect to the longitudinal direction of the slab. For example, the length of the parallel part 41a and the angle of the inclined part 41b of the width reduction die 41 shown in FIG. 4 may be used. The shape of the width reduction die 41 causes differences in the deformation state of the slab, which affects the variation in the rough-side width. In addition, the operation parameters of the width reduction press device 4 may be the total weight and cumulative value of the slab length of the slab that has been width-reduced using the width reduction die 41 that has been installed in the width reduction press device 4 after grinding offline. This is because the wear and damage of the width reduction die 41 progresses as the total weight and cumulative value of the slab length increase, and an error occurs between the set value and the actual value of the width reduction amount, causing variation in the rough-side width.

[Roughing mill operating parameters]
The operation parameters of the roughing mill used to input the width prediction model M refer to the rolling operation conditions that affect the width of the steel plate in any rolling pass of the roughing rolling by the roughing mill 5. The operation parameters of the roughing mill 5 preferably include the rolling conditions by the horizontal rolling mill 51 and the edger 52 that constitute the roughing mill 5. The rolling conditions of the horizontal rolling mill 51 may be the roll opening, the work roll diameter, the entry thickness, the exit thickness, the reduction ratio, the rough target width, the rolling load, and the steel plate temperature in any rolling pass. These affect the width spread behavior of the steel plate in horizontal rolling, thereby affecting the variation in the rough exit side width. The rolling conditions of the edger 52 may be the edger opening, the edger roll diameter, the plate thickness, the width reduction ratio, and the width reduction load in any rolling pass. These rolling conditions affect the width spread behavior in the horizontal rolling of the steel plate, and affect the variation in the rough exit side width. In addition, even if the operating conditions of width rolling are constant, the behavior of dog-bones formed differs between the steady portion and the leading and tailing ends of the steel plate, which affects the width distribution in the longitudinal direction of the steel plate, and thus affects the variation in width of the rough exit side.

In addition, the operation parameters of the rough rolling mill 5 may be the total weight and cumulative length of the steel plate that has been rough rolled using the work rolls and edger rolls used in the horizontal rolling mill 51 and the edger rolls used in the edger 52, which have been ground offline and then installed in the rough rolling mill 5. This is because the wear and damage of the work rolls and edger rolls progresses as the cumulative value of the total weight and length of the steel plate increases, and an error occurs between the set value and the actual value of the reduction rate in horizontal rolling and the width reduction amount by the edger, resulting in variation in the width of the steel plate. Since the rolling conditions, which are the operation parameters of the rough rolling mill 5, have a large effect on the width of the steel plate, it is preferable to use operation parameters selected from the operation parameters of all rolling passes in the rough rolling process as inputs to the width prediction model M. Specifically, the operation parameters of the rough rolling mill 5 may include the set value or actual value of the outlet plate thickness from the first pass to the final pass of rough rolling. This is the so-called rough rolling pass schedule. If the pass schedule differs, the thickness of the steel sheet when transported between the rough passes changes, and the temperature distribution during air cooling changes. This causes differences in the deformation behavior of the steel sheet in the width direction depending on the longitudinal position of the steel sheet, and affects the variation in the rough exit side width. Furthermore, the operation parameters of the rough rolling mill 5 may include information on the air cooling time between rolling passes of the rough rolling and whether or not descaling water is injected in any rolling pass. This is because the temperature fluctuation of the steel sheet affects the variation in the rough exit side width.

[Width prediction method for rough rolled material]
The method for predicting the width of a roughly rolled material according to one embodiment of the present invention includes a prediction step of predicting statistical information of the rough delivery side width using the width prediction model M generated as described above. The width prediction unit that executes the prediction step can be provided in a control computer 91 for controlling the hot rolling line 1. The width prediction unit may be provided in a host computer 92 that gives manufacturing instructions to the control computer 91, or may be provided in an independent computer that can communicate with other devices. Hereinafter, the operation of the width prediction unit according to one embodiment of the present invention will be described with reference to FIG. 11.

The operation of the width prediction unit 110 shown in FIG. 11 is performed before the rough-out side width of the steel plate manufactured in the hot rolling line 1 is measured by the rough-out side width meter 11. The operation of the width prediction unit 110 can be performed, for example, at the stage when the steel plate to be predicted is charged into the heating furnace 2 as a slab. At the stage when the slab is charged into the heating furnace 2, information on the slab dimensions and the component composition of the slab are specified by the upper computer 92 as attribute information of the slab. In addition, at the stage when the slab is charged into the heating furnace 2, the manufacturing specifications of the hot-rolled steel plate are set, and the corresponding standard operating parameters are specified. Therefore, the prediction step can be performed by inputting the actual data of the attribute information of the slab and the set values of other operating parameters specified as standard conditions corresponding to the manufacturing specifications of the hot-rolled steel plate. In addition, the prediction step can be performed at the stage when the steel plate to be predicted is extracted from the heating furnace 2 as a slab. In this case, actual data of the operating parameters of the heating furnace 2 can be obtained and can be used as input for the width prediction model M. Furthermore, the prediction step can be executed, for example, at a stage after the steel plate to be predicted has been width-reduced by the width reduction press device 4. For the slab after width reduction, performance data of the operation parameters of the width reduction press device 4 is acquired by the control computer 91 or the host computer 92. The acquired performance data is used as an input of the width prediction model M, and the setting values of the operation parameters of the roughing mill 5 are set by the control computer 91, so that the input of the width prediction model M is specified.

The operation of the width prediction unit 110 can be performed, for example, during the rolling pass of the steel plate by the rough rolling mill 5. This is because the actual data of the operation parameters of the rough rolling mill 5 in the rolling pass prior to the current rolling pass is acquired by the control computer 91 or the host computer 92, and the setting values of the operation parameters of the rough rolling mill 5 in the rolling pass after the current one can be acquired by the control computer 91 or the host computer 92. In any case, the width prediction unit 110 can predict the statistical information of the rough output width by inputting the actual values of the operation parameters in the process or pass upstream of the current process and the setting values of the operation parameters until the rough rolling is completed into the width prediction model M.

The input data acquisition unit 111 of the width prediction unit 110 shown in FIG. 11 acquires the actual values or set values of the operation parameters of the hot rolling line 1 held by the control computer 91 or the upper computer 92 in the above manner. The width prediction unit 110 inputs the input data acquired by the input data acquisition unit 111 to the prediction unit 112. The prediction unit 112 acquires the hyperparameters determined by the width prediction model generation unit 100. Then, the prediction unit 112 calculates a kernel function for the input data (new input vector) by the prediction step shown in FIG. 8, and calculates statistical information on the rough exit side width, which is output data. As described above, the operation of the width prediction unit 110 can be executed at each stage where the steel sheet passes through each process up to the exit side of the rough rolling mill 5 of the hot rolling line 1, so it may be executed multiple times in the process of manufacturing one steel sheet. The statistical information on the rough exit side width output in the above manner may be displayed on a monitor or the like connected to the width prediction unit 110. Based on the output display of statistical information on the rough output width, at least one of the operating parameters of the roughing mill 5 and the finishing mill 6 can be reset, thereby preventing width defects from occurring in the hot-rolled steel sheet.

[Width control method for rough rolled material]
In one embodiment of the present invention, the width control method for rough rolled material resets one or more operation parameters selected from the operation parameters of the rough rolling mill 5 based on the statistical information of the rough exit side width predicted as described above so as to reduce the probability that the rough exit side width falls below the target width Wt. In the width prediction method for rough rolled material described above, the statistical information of the rough exit side width, which is the output of the width prediction model M, is specified as, for example, the average value _Wm and standard deviation _Wσ of the rough exit side width. In this case, the rough exit side width W is predicted to follow the probability density distribution g(W) shown in the following formula (18).

FIG. 12 is an example that shows a schematic diagram of a probability density distribution g(W) of the rough exit side width W predicted by the width prediction model M together with the target width _Wt . In the example shown in FIG. 12, the average value _Wm of the rough exit side width W predicted by the width prediction model M is smaller than the target width _Wt , and it is predicted that the rough exit side width W will be smaller than the target width _Wt with a high probability due to the variation of the rough exit side width W. That is, in this case, it is predicted that the rough exit side width W will be insufficient with a high probability from the operation conditions of the hot rolling line 1 set at the present time. Therefore, in this example, one or more operation parameters selected from the operation parameters of the rough rolling mill 5 are reset so that the probability that the predicted value W of the rough exit side width W falls below the target width _Wt is reduced. Specifically, the operation parameters of the rough rolling mill 5 are corrected so that the probability density distribution g(W) shown by the solid line in FIG. 12 becomes the probability density distribution shown by the dashed line. In this case, as the operation parameters of the roughing mill 5 to be reset, the operation parameters of the rolling pass in which rolling is not currently being performed are selected. In addition, it is preferable to select the operation parameters to be reset from the operation parameters used as input to the width prediction model M. In the example shown in FIG. 12, the reset operation parameters of the roughing mill 5 are input again to the width prediction model M to output statistical information on the rough exit side width W, and it is confirmed whether the probability that the output statistical information on the rough exit side width W falls below the target width _Wt is reduced. This makes it possible to determine whether the appropriate operation conditions have been reset.

13, when the probability density distribution g(W) of the rough exit width W predicted by the width prediction model M predicts that the rough exit width W is more likely to fall below the target width _Wt , the control target value _Wc for the rough width control may be set so that the probability of the rough exit width W falling below the target width _Wt is reduced. By performing the rough width control of the hot rolling line 1 using the set control target value _Wc , the probability of the rough exit width W being insufficient for the target width _Wt can be reduced even if there is variation in the rough exit width W. Furthermore, it is preferable to reset one or more operation parameters selected from the operation parameters of the rough rolling mill 5 so that the target width _Wt satisfies the following mathematical formula (1).

FIG. 14 is a schematic diagram showing the relationship between the probability density distribution g(W) of the rough exit side width predicted by the width prediction model M and the target width _Wt . The above formula (1) means that the target width _Wt is in the range of _Wm -2.5Wσ and _Wm - _1.5Wσ specified by the average value _Wm and standard deviation _Wσ of the rough exit side width W output from the width prediction model M. In this case, it is preferable to reset one or more operation parameters selected from the operation parameters of the rough rolling _mill 5 so that the probability density distribution g(W) of the rough exit side width W output from the width prediction model M satisfies the above formula (1) with the target _width Wt being constant. FIG. 15 is a diagram showing the relationship between the deviation of the rough exit side width W from the target width _Wt and the amount of cut-off of the steel plate. As can be seen from FIG. 15, even if the rough exit side width W is too large or too small for the target width Wt, the amount of cut-off of the steel plate increases and the product yield of the hot-rolled steel plate decreases. In detail, if the rough-side width W is excessively large relative to the target width _Wt , it is necessary to perform edge trimming in a subsequent process so that the hot-rolled steel sheet meets the width specification. On the other hand, if the rough-side width W is excessively small relative to the target width _Wt , it is not possible to produce a steel sheet product, and some of the steel sheet must be scrapped. However, the amount of cut-off of the steel sheet is suppressed more when the rough-side width W is excessively large relative to the target width _Wt than when it is excessively small relative to the target width _Wt . For this reason, the above formula (1) sets the operation parameters of the rough rolling mill 5 so that the probability of the rough-side width W being excessively small is low even if the rough-side width W varies relative to the target width _{Wt. That is, the lower limit value for the target width Wt} in formula (1) is intended to reduce the probability that the rough-side width W is excessively small, and the upper limit value for the target width _Wt is intended to prevent the rough-side width W from being excessively large relative to the target width _Wt .

The rough rolled material width control method in this embodiment outputs statistical information on the rough output width in response to different operating conditions for each rough rolled material, in addition to classifications such as the steel type and size of the slab and the product dimensions of the hot-rolled steel plate, so that it is possible to predict the variation in the rough output width for each rough rolled material. This eliminates the need to set the extra width in advance according to the steel type and size classification of the steel plate, as in the conventional technology, and allows an appropriate extra width to be applied to each rough rolled material produced in the hot rolling line, thereby improving the product yield of the hot-rolled steel plate.

Example 1
As an embodiment of the present invention, a rough rolled material width prediction and width control were performed in a hot rolling line 1 including a width reduction press device 4 arranged downstream of a heating furnace 2, a rough rolling mill 5 consisting of four reversible rolling mills 5a and one non-reversible rolling mill 1b, and a finish rolling mill 6 consisting of seven rolling stands. In this embodiment, a slab having a slab thickness of 250 to 270 mm and a slab width of 600 to 1600 mm was heated in the heating furnace 2 by the above-mentioned hot rolling line 1, and a hot rolled steel sheet having a plate thickness of 30 to 35 mm on the delivery side of the rough rolling mill 5 and a plate thickness of 2 to 3 mm on the delivery side of the finish rolling mill 6 was produced. In addition, the hot rolling line 1 is equipped with a rough delivery side width meter 11, a finish delivery side width meter 12, and a coiler front width meter 13.

The control computer 91 and the host computer 92 of the hot rolling line 1 collect the actual values of the operating parameters of the rough rolled material produced in the hot rolling line 1, and the data acquisition unit 103 acquires the actual data. The data acquisition unit 103 acquires the edger opening, edger roll diameter, and total rolling length after grinding of the edger roll in all rolling passes of rough rolling as the operation actual data of the rough rolling mill 5. In addition, the data acquisition unit 103 acquires the work roll diameter, entry side plate thickness, exit side plate thickness, and total rolling length after grinding of the work roll of the horizontal rolling mill in all rolling passes of rough rolling as the operation actual data of the rough rolling mill 5. In addition, the data acquisition unit 103 acquires the thickness and width data of the slab as the actual data of the attribute information of the slab.

On the other hand, the data acquisition unit 103 calculated the average width of the steady part from the actual value of the rough-exit side width measured by the rough-exit side width meter 11, and used this as actual data of the rough-exit side width. The actual data of the rough-exit side width was associated with the above-mentioned operation actual data by the data acquisition unit 103 to form one data set for one rough rolled material, and was stored in the database unit 101. Then, at the stage where 30,000 data sets were stored in the database unit 101, these data sets were divided into 20,000 pieces of learning data and 10,000 pieces of test data, and a width prediction model M was generated by the machine learning unit 102 using the learning data. In this embodiment, a radial basis function (RBF) kernel was used as the kernel function of the Gaussian process regression. In addition, noise having a constant variance σ _e ² was used as the Gaussian noise regardless of the learning data. The kernel function used in this embodiment is expressed by the following mathematical formula (19). Here, ∥x ⁽ⁱ⁾ −x ^(j) ∥ represents the Euclidean distance between the input vectors.

In this embodiment, the hyperparameters of the width prediction model M were specified by a Gaussian process regression method using the learning data. Then, the operation record data of the test data was input to the prediction section 112 of the width prediction unit 110, and the average value _Wm and standard deviation _Wσ of the rough exit side width W, which is the output of the width prediction model M, were obtained. In addition, the RMSE (root mean square error) was calculated from the deviation between the actual data Wa of the rough exit side width W, which is the test data _, and the average value _Wm . Furthermore, the number of test data pieces in which the actual data _Wa of the rough exit side width W falls within the range of _Wm ± _Wσ and the range of _Wm ± _1.96Wσ was obtained, and the ratio of the number of test data pieces falling within these ranges to all the test data pieces was calculated.

As a result, the RMSE calculated from the deviation between the actual data W _a and the average value W _m of the rough exit side width W was 0.1 mm, which was good. Furthermore, the probability that the rough exit side width W was within the range of W _m ±W _σ was 67.3%, and the probability that it was within the range of W _m ±1.96W _σ was 95.3%. This means that if the variation in the rough exit side width W is assumed to follow a normal distribution, the probabilities are 68.3% and 95.0%, respectively, and therefore the variation in the rough exit side width W can be accurately predicted by the width prediction model M of this embodiment. On the other hand, the width prediction model M generated as described above was stored in the prediction section 112 of the width prediction section 110, and width control was performed on a rough rolled material having a slab width in the range of 1000 to 1200 mm. In this case, the statistical information of the rough output width W was calculated using the width prediction model M at the timing when the slab was extracted from the heating furnace 2 for each steel plate, and the operation parameters of the rough rolling mill 5 were reset so that the target width _Wt of the rough rolled material set for each steel plate satisfied the relationship shown in the mathematical formula (1). The edge opening of the rough rolling mill 5 was selected as the operation parameter of the rough rolling mill 5 to be reset. The rough rolled material thus subjected to rough rolling was subsequently subjected to finish rolling to produce a hot rolled steel plate. The hot rolled steel plate produced was 400 coils. As a result, the proportion of coils whose width of the hot rolled steel plate was below the target width before the coiler was reduced by 35% compared to the conventional example in which the extra width was set in advance. In addition, the edge cut allowance, which was cut off due to the width becoming excessively larger than the target width before the coiler, was reduced by 0.8 mm on average by this embodiment. From the above, it was confirmed that this embodiment makes it possible to predict statistical information on the rough output side width, and that by applying the predicted statistical information on the rough output side width to width control of rough rolled material, width defects of hot-rolled steel sheets can be reduced and product yield can be improved.

Example 2
As an embodiment of the present invention, another example of the width prediction method for rough rolled material will be described. In the above embodiment 1, when the performance data is accumulated in the database unit 101, the data acquisition unit 103 acquires the operation performance data of the width reduction press device 4 and the operation performance data of the heating furnace 2, and accumulates them in association with the operation performance data of the rough rolling mill 5 and the performance data of the attribute information of the slab. In this embodiment, the input data used in the width prediction model M is changed using the performance data accumulated in the database unit 101 in this manner, and the width prediction accuracy of the rough rolled material is evaluated.

The operational performance data of the width reduction press device 4 stored in the database unit 101 are the width reduction amount SPW of the slab, the feed pitch SPP of the slab between width reduction passes, and the width reduction start position SPS. The width reduction amount and feed pitch in the steady part of the slab were used as the width reduction amount SPW of the slab and the feed pitch SPP of the slab between width reduction passes. In addition, the operational performance data of the heating furnace 2 stored in the database unit 101 are the residence time IFT from when the slab is charged into the heating furnace 2 to when it is extracted from the heating furnace 2, the temperature (extraction temperature) ET of the slab extracted from the heating furnace 2, and the distance D1 between the longitudinal end of the slab and the furnace wall of the heating furnace 2, which is an operational parameter related to the charging position of the slab in the heating furnace 2.

In this embodiment, when 30,000 data sets were accumulated in the database unit 101, these data sets were divided into 20,000 pieces of training data and 10,000 pieces of test data, and the machine learning unit 102 generated a width prediction model M using the training data. In this case, the machine learning unit 102 performed machine learning by changing the variables used in the input data of the width prediction model M, and generated a width prediction model M corresponding to each condition. In all conditions, a radial basis function (RBF) kernel was used as the kernel function for Gaussian process regression.

Table 1 shows the input data used for each width prediction model. All of the width prediction models No. 1 to 3 in Table 1 include the operation parameters of the roughing mill 5 as input data. The operation parameters of the roughing mill 5 used were the edger opening EG and the edger roll diameter ED of the steady part in the entire rolling pass of edger rough rolling. Furthermore, the entry plate thickness HI, exit plate thickness HO, and the work roll diameter HWD of the horizontal rolling mill in the steady part in the entire rolling pass of rough rolling were used as operation parameters of the roughing mill 5. However, in this embodiment, parameters of the slab attribute information were not used as input data for the width prediction model M.

The width prediction model No. 1 listed in Table 1 used only the parameters of the rough rolling mill 5 as input data. The width prediction model No. 2 used the operation parameters of the width reduction press device in addition to the parameters of the rough rolling mill 5 as input data. The operation parameters of the width reduction press device used were the width reduction amount SPW, the feed pitch SPP, and the width reduction start position SPS. The width prediction model No. 3 used the operation parameters of the heating furnace 2 in addition to the parameters of the rough rolling mill 5 as input data. The operation parameters of the heating furnace 2 used were the residence time IFT, the extraction temperature ET, and the distance D1 from the furnace wall representing the charging position.

In this embodiment, the hyperparameters of the width prediction models No. 1 to 3 were identified by a Gaussian process regression method using the learning data. Then, the operation performance data of the test data was input to the prediction section 112 of the width prediction section 110, and the average value Wm and standard deviation Wσ of the rough exit side width W, which is the output of the width prediction model, were obtained. In addition, the RMSE (root mean square error) was calculated from the deviation between the actual data Wa of the rough exit side width W, which is the test data, and the average value Wm. Furthermore, the number of test data in which the actual data Wa of the rough exit side width W falls within the range of Wm ± Wσ was obtained, and the ratio of the number of test data falling within these ranges to all test data was calculated.

Table 1 shows the results of the prediction accuracy. In No. 1, the RMSE calculated from the deviation between the actual data Wa of the rough exit side width W and the average value Wm was 1.0 mm. In addition, the probability that the rough exit side width W falls within the range of Wm ± Wσ was 70.0%, which is close to the probability of 68.3% when it is assumed that the variation of the rough exit side width W follows a normal distribution. In No. 2, the RMSE was 0.1 mm, which improved the prediction accuracy compared to No. 1, while the probability that the rough exit side width W falls within the range of Wm ± Wσ was 66.3%, which is close to the probability of 68.3% when it is assumed that the variation of the rough exit side width W follows a normal distribution. In No. 3, the RMSE was 0.0 mm, and high prediction accuracy was obtained for the average value of the rough exit side width W. In addition, the probability that the rough-side width W falls within the range of Wm±Wσ was 66.0%, which is close to the probability of 68.3% when it is assumed that the variation of the rough-side width W follows a normal distribution. From the above, it was confirmed that any of the width prediction models can accurately predict the average value Wm and standard deviation Wσ of the rough-side width, which are statistical information of the rough-side width.

Although the above describes an embodiment of the invention made by the present inventors, the present invention is not limited by the description and drawings that form part of the disclosure of the present invention according to this embodiment. In other words, other embodiments, examples, and operational techniques made by those skilled in the art based on this embodiment are all included in the scope of the present invention.

According to the present invention, a method for predicting the width of rough rolled material can be provided, which is capable of predicting statistical information including the variation in the width of the rough rolled material. According to the present invention, a method for controlling the width of rough rolled material can be provided, which is capable of precisely controlling the longitudinal width of the rough rolled material, taking into account the variation in the width of the rough rolled material. According to the present invention, a method for manufacturing hot rolled steel sheet can be provided, which is capable of improving the product yield of the hot rolled steel sheet. According to the present invention, a method for generating a width prediction model of rough rolled material can be provided, which is capable of generating a width prediction model that predicts statistical information including the variation in the width of the rough rolled material.

REFERENCE SIGNS LIST 1 Hot rolling line 2 Heating furnace 3 Descaling device 4 Width reduction press device 5 Rough rolling mill 5a Reversible rolling mill 5b Non-reversible rolling mill 6 Finishing rolling mill 7 Cooling device 8 Coiler (winder)
11 Rough exit width meter 12 Finish exit width meter 13 Coiler front width meter (coiler entry width meter)
14 Pass line 15a, 15b, 15c, 15d Camera 21 Thermometer 22 Walking beam (fixed skid)
23 Moving skid 41 Width reduction die 41a Parallel portion 41b Inclined portion 42 Drive device 43 Pinch roll 51 Horizontal rolling mill 52 Edger (vertical rolling mill)
90 Control controller 91 Control computer 92 Host computer 100 Width prediction model generation unit 101 Database unit 102 Machine learning unit 103 Data acquisition unit 110 Width prediction unit 111 Input data acquisition unit 112 Prediction unit M Width prediction model SA Slab SB Steel plate

Claims (5)

A rough rolled material width prediction method for predicting the width of the rough rolled material in a hot rolling line including a heating furnace for heating a slab, a rough rolling mill for rough rolling the heated slab to produce a rough rolled material, and a finish rolling mill for finish rolling the rough rolled material to produce a finish rolled material, the rough rolled material width prediction method including a prediction step of predicting statistical information of the width of the rough rolled material using a width prediction model learned by a Gaussian process regression technique, the width prediction model including one or more operation parameters selected from the operation parameters of the rough rolling mill as input data and output data being statistical information of the width of the rough rolled material, the rough rolled material width prediction method including a resetting step of predicting statistical information of the width of the rough rolled material using the rough rolled material width prediction method, and resetting one or more operation parameters selected from the operation parameters of the rough rolling mill based on the predicted statistical information so as to reduce the probability that the width of the rough rolled material falls below the target width of the rough rolled material,
The statistical information on the width of the rough rolled material includes an average value Wm _and a standard deviation Wσ _of the width of the rough rolled material , and the resetting step includes a step of setting one or more operation parameters selected from operation parameters of a rough rolling mill so that the target width Wt of the rough rolled material _satisfies the relationship shown in the following formula (1). A rough rolled material width control method.

2. The rough rolled material width control method according to claim 1, wherein the hot rolling line includes a width reduction press device arranged upstream of the rough rolling mill and intermittently width-reduces the slab after it has been heated by the heating furnace, and the width prediction model includes one or more operation parameters selected from operation parameters of the width reduction press device as input data. The method of claim 1 , wherein the width prediction model includes as input data one or more operational parameters selected from operational parameters of the heating furnace. The method of claim 1 , wherein the width prediction model includes one or more parameters selected from attribute information of the slab as input data. A method for producing a hot-rolled steel sheet, comprising the step of producing a hot-rolled steel sheet using the method for controlling the width of a rough rolled material according to any one of claims 1 to 4.

Images