Method for generating shape control actuator setting model of rolling equipment, method for setting shape control actuator of rolling equipment, method for controlling shape of steel sheet, and shape control system and method for manufacturing same

Abstract

Provided is a method for generating a shape control actuator setting model of rolling equipment with which it is possible to generate a shape control actuator setting model for controlling the shape of a sheet accurately over the longitudinal direction. The method for generating a shape control actuator setting model of rolling equipment provided with a shape control actuator and a shape detector includes: a data acquisition step for acquiring operation result data selected from rolling operation parameters of the rolling equipment, and shape result data acquired from the shape detector, of the equipment; and a step for generating a shape control actuator setting model using a neural network composed of an input layer having 8-256 nodes, an intermediate layer having 16-2048 nodes, and an output layer having 1-256 nodes, using the acquired data as input result data and a shape control actuator setting value as output result data, with the number of items of training data being 5 × 104 to 1 × 106.

Description

Method for generating a shape control actuator setting model for rolling equipment, method for setting a shape control actuator, method for controlling the shape of a steel plate, shape control system, and method for manufacturing the same

[0001] The present invention relates to a method for generating a shape control actuator setting model for rolling equipment, a method for setting a shape control actuator, a method for controlling the shape of a steel plate, a shape control system, and a manufacturing method.

[0002] In rolling mills for reducing the thickness of thin steel sheets, the steel sheet is loaded into the rolling mill as a coiled steel strip, continuously discharged, and rolled while being transported. The rolling process requires not only controlling the steel sheet to a predetermined size but also appropriately controlling the steel sheet shape. Here, the "steel sheet shape" refers to flatness, and typical examples include edge waves and center elongation. The shape of the steel sheet manifests as a shape defect, such as a wave shape, due to buckling caused by the inability to maintain the elongation strain distributed across the width of the steel sheet during rolling, i.e., longitudinal strain, as in-plane stress. Shape defects in steel sheets not only degrade product quality, but also force a reduction in the rolling speed (i.e., line speed) of the rolling equipment, reducing production efficiency. Furthermore, shape defects in steel sheets can cause operational problems such as steel sheet breakage and under-sizing. Therefore, the rolling process requires controlling the shape of the steel sheet to a desired target shape along the longitudinal direction.

[0003] Meanwhile, rolling mills that make up rolling equipment are equipped with shape control actuators that control the elastic deformation of the work rolls to adjust the elongation strain distributed in the width direction of the steel plate to a desired distribution. Different shape control actuators are used depending on the mill type of the rolling mill. A typical example is a work roll bender, which applies a bending force to the work rolls to adjust the deflection of the work rolls. Other methods include roll shifting, which moves the work rolls or intermediate rolls in the axial direction, and split backup rolls to adjust the deflection of the work rolls. Furthermore, rolling equipment may be equipped with a steel plate shape detector, and the shape control actuators are set so that the steel plate shape measured by the shape detector matches the target shape, i.e., the target shape.

[0004] On the other hand, it is known that the quality of neural network learning is determined by the data input to the neural network and the learning calculation conditions. However, it is difficult to comprehensively change all conditions to determine the quality of learning, and identifying the learning conditions to generate a good model largely depends on know-how.

[0005] Shape control devices that set such shape control actuators are equipped with a setting calculation function that performs initial settings before the start of a rolling pass of a steel plate. Furthermore, they often have a feedback control function that corrects, i.e., resets, the setting values ​​of the shape control actuators by referring to the output from the shape detector after the start of a rolling pass. By performing appropriate initial settings before the start of a rolling pass of a steel plate, it is possible to improve the shape of the steel plate from its leading edge. Furthermore, the feedback control function can improve the shape of the steel plate throughout its entire length, even if disturbances occur. However, it can be difficult to quantitatively predict the effect of the shape control actuators on the deformation of the work rolls of a rolling mill. For example, it is necessary to predict the behavior of thermal expansion of various rolls in the rolling mill, which can lead to various errors. For this reason, techniques have been proposed for appropriately setting the manipulated variable of the shape control actuators of rolling equipment.

[0006] Specifically, Patent Document 1 describes a method for appropriately initializing a shape control actuator. More specifically, the method described in Patent Document 1 configures a neural network using influencing factors related to the flatness of a rolled material, i.e., the shape, as input data. In this method, an optimal setting value calculation device calculates optimal setting values ​​for each control element for an actual shape value obtained when rolling is performed at a certain setting value, and paired data between the influencing factors and the setting values ​​of each control element at this time is accumulated and used as learning data for the neural network. In this case, the neural network has a hierarchical structure, and learning is performed by error backpropagation learning, with the influencing factors as inputs and the setting values ​​of each control element as outputs.

[0007] On the other hand, Patent Document 2 also describes a method for appropriately initializing a shape control actuator. In detail, the method described in Patent Document 2 accumulates the rolling conditions of a steel plate, the setting values ​​of the shape control actuator, and shape data as shape control performance data in advance, and extracts typical data, i.e., a prototype, by learning using a neural network. Then, before executing a rolling pass of the steel plate, the similarity between the rolling conditions of the steel plate and the typical data is calculated. Based on the calculated similarity, the setting values ​​of the shape control actuator according to the typical data are combined to perform initial setting of the shape control actuator.

[0008] Furthermore, Patent Document 3 exemplifies shape control of a steel plate by rolling equipment, and describes a method of configuring a neural network that inputs shape deviation data, which is the difference between the actual shape detected by a shape detector and the target shape, and outputs the operation amount of a shape control actuator. Patent Document 3 also describes shape feedback control that resets the shape control actuator in accordance with the detected shape deviation. Furthermore, Patent Document 3 describes a method of optimizing the control output of the feedback control by providing a control result judgment unit that judges whether the control output of the shape control actuator is good or bad.

[0009] Japanese Patent Laid-Open No. 7-246408 Japanese Patent Laid-Open No. 2020-134967 Japanese Patent Laid-Open No. 2018-180799

[0010] However, the method described in Patent Document 1 relates to a method for initially setting optimal rolling conditions, and does not specifically describe the conditions under which the neural network should be trained. Therefore, there is a drawback in that the accuracy and stability of the initial setting method using the trained neural network obtained are not guaranteed.

[0011] The method described in Patent Document 2 also relates to the initial setting of rolling conditions, and is an initial setting method characterized in that the set values ​​before the production of the rolled material and the set values ​​manually set by an operator are input. However, with this method, the corrected values ​​set by the operator are not necessarily obtained when the rolled material is rolled, and it cannot be said to be an optimal initial setting method when the plate crown of the rolled material changes in the longitudinal direction.

[0012] The method described in Patent Document 3 can be applied to shape feedback control by determining whether the control output of a shape control actuator is good or bad. However, this method also does not mention the specific conditions under which the neural network should be trained to maximize the effect obtained, and there is room for improvement in the prediction accuracy and stability of control using the trained neural network.

[0013] The present invention has been made to solve the above problems, and an object of the present invention is to provide a method for generating a shape control actuator setting model for rolling equipment, which is capable of generating a shape control actuator setting model that accurately controls the shape of a steel plate along the longitudinal direction. Another object of the present invention is to provide a setting method for shape control actuators of rolling equipment, which is capable of accurately controlling the shape of a steel plate along the longitudinal direction, as well as a shape control method and shape control system for a steel plate. Still another object of the present invention is to provide a steel plate manufacturing method that is capable of manufacturing steel plates having a desired shape along the longitudinal direction with a high yield.

[0014] The gist of the present invention that advantageously solves the above-mentioned problems is as follows: [1] A method for generating a shape control actuator setting model for rolling equipment, which is provided with a shape control actuator that controls the shape of a steel sheet and a shape detector that detects the shape of the steel sheet, for determining setting values ​​of the shape control actuator, comprising: a data acquisition step of acquiring one or more operation record data selected from rolling operation parameters of the rolling equipment and shape record data acquired from the shape detector; ４ ~1 x 10 ６ [2] In the above [1], the neural network learning is performed using a batch size of 8 to 1024, 1 x 10 －３ ~1 x 10 －６ [3] A method for generating a shape control actuator setting model for rolling equipment, the method being performed using a learning rate between

[0001] and

[0002] . [3] A method for setting shape control actuators for rolling equipment, comprising a set value updating step of determining and updating set values ​​of the shape control actuators using a shape control actuator setting model generated according to the method for generating a shape control actuator setting model for rolling equipment described in [1] or [2] above. [4] A method for controlling the shape of a steel plate, comprising a step of updating set values ​​of the shape control actuators during rolling of the steel plate, using the method for setting shape control actuators for rolling equipment described in [3] above. [5] A shape control system for a steel plate, comprising means for updating set values ​​of the shape control actuators during rolling of the steel plate, using the method for setting shape control actuators for rolling equipment described in [3] above. [6] A method for producing a steel plate, comprising a rolling step of rolling the steel plate using the method for controlling the shape of a steel plate described in [4] above.

[0015] According to the method for generating a shape control actuator setting model for rolling equipment of the present invention, it is possible to generate a shape control actuator setting model that accurately controls the shape of a steel plate along the longitudinal direction. Furthermore, according to the method for setting the shape control actuators for rolling equipment and the method and system for controlling the shape of a steel plate according to the present invention, it is possible to accurately control the shape of a steel plate along the longitudinal direction. Furthermore, according to the method for manufacturing a steel plate according to the present invention, it is possible to manufacture a steel plate having a desired shape along the longitudinal direction with a high yield.

[0016] FIG. 1 is a side view showing the configuration of rolling equipment according to one embodiment of the present invention. FIG. 2 is a front view showing the general configuration of a shape control actuator (actuator) and a rolling mechanism according to the embodiment. FIG. 3 is a schematic diagram for explaining a method for generating a shape control actuator setting model according to one embodiment of the present invention. FIG. 4 is a diagram showing an example of shape actual data. FIG. 5 is a schematic diagram showing the configuration of a multilayer neural network used in the shape control actuator setting model according to the embodiment. FIG. 6 is a diagram showing an example of calculation for calculating intermediate layer values ​​using input layer data. FIG. 7 is a diagram showing an example of calculation for calculating output layer values ​​using final intermediate layer data. FIG. 8 is a diagram showing a calculation method for a batch normalization layer. FIG. 9 is a diagram showing a calculation method for a dropout layer. FIG. 10 is a diagram showing the configuration of a steel plate shape control system according to one embodiment of the present invention. FIG. 11 is a diagram showing the correspondence relationship between the number of nodes in the input layer, intermediate layer, and output layer. FIG. 12 is a diagram showing a comparison result between the output value of the shape control actuator setting value and the actual value in an example.

[0017] The following describes in detail the embodiments of the present invention. Note that the drawings are schematic and may differ from the actual embodiments. Furthermore, the following embodiments exemplify facilities, devices, and methods for embodying the technical concept of the present invention, and are not intended to limit the configuration to those described below. In other words, the technical concept of the present invention can be modified in various ways within the technical scope described in the claims.

[0018] [Rolling Equipment] First, the configuration of a steel sheet rolling equipment according to one embodiment of the present invention will be described with reference to Figures 1 and 2. In this embodiment, equipment including at least a rolling mill and a shape detector will be referred to as rolling equipment. In Figures 1 and 2, symbols UP represent upward, DN represent downward, FD represent the rolling direction, and WD represent the width direction of the steel sheet.

[0019] FIG. 1 is a side view showing the configuration of rolling equipment according to this embodiment. As shown in FIG. 1, the rolling equipment 1 according to this embodiment is composed of a 12-high cluster-type rolling mill. The 12 rolls include a pair of work rolls 2 (upper and lower), two pairs of intermediate rolls 3 (upper and lower), a pair of small backup rolls 4 (upper and lower), two pairs of upper backup rolls 5, and two pairs of lower backup rolls 6. A steel sheet S, which is the material to be rolled, is discharged from a payoff roll 7. The steel sheet S is supplied from one of a pair of tension reels 8 and 9 that sandwich the rolling mill, rolled, and wound onto the tension reel on the opposite side. A shape detector 10 detects the difference in elongation of the steel sheet S in the width direction WD.

[0020] FIG. 2 is a front view showing a schematic configuration of a shape control actuator (operating element) and a rolling mechanism. The steel sheet S is pressed down and elongated from above and below by the work rolls 2. In the example of FIG. 2, the upper backup roll 5 is divided into seven parts in the axial direction. The roll crown can be adjusted by extruding the center 51a, quarter-in 51b, quarter-out 51c, and edge 51d in that order from the axial center position. The lower backup roll 6 is divided into six parts in the axial direction in the example of FIG. 2. The roll crown can be adjusted by extruding the center 61a, quarter 61b, and edge 61c in that order from the axial center position.

[0021] The rolling equipment is equipped, as shape control actuators, with a roll crown adjustment function for the upper backup roll 5 and the lower backup roll 6, an intermediate roll bender 11, and a work roll bender 12 for applying bending force to the upper and lower work rolls 2. Note that the shape control actuator may also be equipped with a leveling mechanism for tilting the axial center of the upper backup roll 5 or the lower backup roll 6 in the width direction. In addition to these shape control actuators, the rolling equipment may also be equipped with an intermediate roll shifter for shifting the upper and lower intermediate rolls 3 in opposite directions relative to the roll axial direction, and a work roll shifter for shifting the upper and lower work rolls 2 in opposite directions relative to the upper and lower work rolls 2.

[0022] Returning to FIG. 1, the shape detector 10 is a measuring instrument for measuring the shape of the steel sheet S rolled by a rolling mill. As the shape detector 10, any means that is normally used for measuring the shape of the steel sheet S may be selected. The shape of the steel sheet S is generated by the distribution of the lengths of the longitudinal line segments of the steel sheet S in the width direction, and the difference in length in the width direction is expressed as the elongation difference rate (%) or the elongation difference (I-unit (1×10 －５ )) is usually used to evaluate the elongation difference. Specifically, if the length of the steel sheet S is distributed in the width direction, the contact pressure of the roll that contacts the steel sheet S at a constant wrap angle, i.e., the vertical load on the contact roll, will be distributed in the width direction. The shape detector 10 calculates the elongation difference by measuring the distribution of the contact pressure. At this time, the width direction distribution of the contact pressure is detected by load detectors that are divided in the axial direction and arranged inside the contact roll. This is also called a divided load cell method.

[0023] However, the shape detector 10 may also use a method in which the out-of-plane displacement of the steel sheet S is detected by a non-contact distance meter, such as a laser distance meter, and converted into differential elongation, or a method in which the differential elongation is calculated from moire fringes irradiated onto the steel sheet S. Using any of these shape measuring instruments, the differential elongation rate or differential elongation (I-unit) is measured from the line segment length in the width direction of the steel sheet S. Note that the number of channels for detecting differential elongation in the width direction, i.e., the number of divisions in the width direction, is preferably greater than the number of shape control actuators used for shape control. Specifically, the shape detector 10 is preferably divided into 24 to 64 sections in the width direction, and is capable of measuring the differential elongation rate or differential elongation at each width direction position.

[0024] [Shape Control System] Next, a shape control system for the steel sheet S provided in the rolling facility 1 will be described with reference to FIG.

[0025] As shown in FIG. 1 , the shape control system for a steel sheet S includes a control computer 100 and a rolling controller 110. The control computer 100 acquires information on the base material dimensions of the steel sheet S, such as thickness, width, and length, as well as information on the deformation resistance and product thickness of the steel sheet S, from a host computer. The control computer 100 then calculates a predicted value of the rolling load for each rolling pass when rolling the number of rolling passes set according to the manufacturing specifications of the steel sheet S. The control computer 100 then sends an initial setting value for the shape control actuator for each rolling pass to the rolling controller 110 based on the predicted value of the rolling load. The rolling controller 110 then operates each device of the shape control actuator based on the acquired initial setting value. The rolling controller 110 also acquires information on the target shape of the steel sheet S for each rolling pass from the control computer 100.

[0026] After the initial setting values ​​of the shape control actuators are set and the rolling pass starts, the rolling controller 110 compares the actual shape of the steel sheet S obtained from the shape detector 10 with the target shape of the steel sheet S obtained from the control computer 100. The rolling controller 110 then resets the setting values ​​of the shape control actuators as needed, that is, executes shape feedback control. By executing shape feedback control, even if there is a disturbance such as a thickness variation of the base material in the longitudinal direction of the steel sheet S, it is possible to make the shape of the steel sheet S approach the target shape, and it is possible to produce a steel sheet S having a uniform shape in the longitudinal direction.

[0027] [Shape Control Actuator Setting Model] Next, a shape control actuator setting model according to this embodiment will be described with reference to FIGS.

[0028] The shape control actuator setting model according to this embodiment is provided in the rolling controller 110 shown in FIG. 1 and is used to send appropriate setting values ​​to the shape control actuators by referring to the actual shape of the steel sheet S acquired from the shape detector 10. As shown in FIG. 3, the shape control actuator setting model M is generated by a shape control actuator setting model generation unit 130. In this embodiment, the data acquisition unit 120 acquires, in the rolling pass of the steel sheet S, actual shape data 121 of the steel sheet S acquired from the shape detector 10 during rolling, one or more operation actual data 122 selected from the rolling operation parameters of the rolling equipment 1, and actual data 123 of the setting values ​​of the shape control actuators. The actual data 123 of the setting values ​​of the shape control actuators refers to the setting values ​​of at least one or more shape control actuators selected from the above-mentioned shape control actuators. The various data acquired by the data acquisition unit 120 constitute a set of data sets and are stored in a database unit 131 of the shape control actuator setting model generation unit 130, and a shape control actuator setting model M is generated by a machine learning unit 133.

[0029] The shape performance data 121 acquired by the data acquisition unit 120 is shape data measured by the shape detector 10 (see FIG. 1). For example, the shape data is as shown in FIG. 4. The shape data is information specified by the widthwise position of the steel sheet S and the differential elongation rate or differential elongation (I-UNIT) at that position. The differential elongation rate is a value expressed in strain format of the line segment length of the steel sheet S at other widthwise positions, based on the position where the line segment length in the widthwise direction of the steel sheet S is shortest. Furthermore, since the shape data becomes a very small value even when the differential elongation rate is expressed as a percentage, the differential elongation rate (I-UNIT) is conventionally expressed by converting the differential elongation rate to a value 100,000 times the value and expressed in units called I-UNIT. In this case, when the differential elongation rate or differential elongation at the widthwise end of the steel sheet S is larger than that of other parts, it is often referred to as a "semi-stretched" wave. When the differential elongation rate or differential elongation at the widthwise center is larger than that of other parts, it is often referred to as a "center elongation." Furthermore, when the elongation difference rate or difference between the widthwise end portion and the widthwise center portion is large, it is called quarter elongation, and when the above shapes are combined, it is called compound elongation.

[0030] The operation performance data 122 acquired by the data acquisition unit 120 together with the shape performance data 121 of the steel sheet S is one or more pieces of operation performance data 122 selected from rolling operation parameters. The rolling operation parameters are parameters that represent the rolling state when a rolling pass is executed. These parameters represent factors that can affect the shape of the steel sheet S, excluding the shape performance data 121 of the steel sheet S and the performance data 123 of the setting values ​​of the shape control actuators. The rolling operation parameters may, for example, use information on the thickness and width of the base material held by a host computer that issues instructions to the control computer 100 as dimensional information of the steel sheet S. Furthermore, the entry thickness, exit thickness, and width of the rolling pass for which the shape performance data is acquired may also be used. Another example of the rolling operation parameters is attribute information of the base material. The attribute information of the base material may, for example, use indexes that represent the characteristics of the base material, such as the deformation resistance of the steel sheet S, the steel type, the chemical composition, the surface roughness of the base material, and the distribution state of surface oxides. Furthermore, the attribute information of the base material may also include manufacturing conditions in a process (hereinafter also referred to as a previous process) prior to the rolling process by the rolling equipment 1. For example, if the previous process is an annealing process, information such as the annealing temperature, soaking time, and cooling rate may be included in the attribute information of the base material, and if the previous process is a hot rolling process, information such as the coiling temperature in the hot rolling line may also be included in the attribute information of the base material. This is because these are parameters that indirectly affect the rolling load when shape control is performed by the rolling equipment 1.

[0031] Another example of the rolling operation parameter is rolling condition information in the rolling pass for which the shape actual data 121 is acquired. Examples of the rolling condition information include rolling load, rolling torque, rolling speed, entry tension, exit tension, and work roll diameter. Furthermore, parameters related to lubrication conditions in rolling, such as coolant flow rate, coolant temperature, and coolant concentration, may also be used as the rolling condition information. Note that, for each parameter of the dimensional information of the steel sheet S, the attribute information of the base material, and the rolling condition information, either a set value preset by the host computer or the control computer 100 or an actual value actually measured in the rolling equipment 1 may be used. When the rolling equipment 1 is equipped with measuring instruments such as a thickness gauge for measuring the thickness of the steel sheet S, a load cell for measuring the rolling load, and a tension meter for measuring the tension applied to the steel sheet S, it is preferable to use the actually measured values.

[0032] The rolling operation parameters preferably include data on the target shape of the steel sheet S, which is set in the rolling control controller 110. The target shape of the steel sheet S may be a target value of the differential elongation rate at each position in the width direction of the steel sheet S, as exemplified as the actual shape data in FIG. 4. However, as the target shape of the steel sheet used in the rolling operation parameters of this embodiment, a polynomial approximation may be performed on the target differential elongation as a function of the width direction position, and the coefficient values ​​of the polynomial may be used. For example, as the target shape of the steel sheet S, the width direction distribution of the target differential elongation (I-unit (target)) is approximated by a quartic function and expressed as shown in the following mathematical formula (1). However, z in mathematical formula (1) is the width direction position obtained by normalizing the sheet width of the steel sheet S in the range of -1 to +1. In this case, the coefficients λ1 to λ4 may be included in the rolling operation parameters as data on the target shape of the steel sheet. [Mathematical formula (1)] I-unit (target) = λ １ z+λ ２ z ２ +λ ３ z ３ +λ ４ z ４

[0033] In this embodiment, the operation performance data 122 is set by selecting at least one parameter from the above-described rolling operation parameters, and the operation performance data 122 related to this parameter is used. On the other hand, the rolling operation parameters may include identification data of the operator who manually operated the shape control actuator of the rolling equipment 1. When the shape of the steel sheet S is controlled by manual operation by an operator, the rolling operation parameters may include numbers, symbols, names, etc. as identification data that can identify the operator who performed the operation. This is because the shape control of the steel sheet S exhibits unique control performance depending on the characteristics of the shape control actuator of the rolling equipment 1, and therefore, differences arise in the quality of the shape control of the steel sheet S between highly skilled operators and less skilled operators. For this reason, the shape control performance of the steel sheet S can be separately evaluated based on the difference between the target shape and the actual shape, etc., and a shape control actuator setting model M can be generated that selectively outputs the operation data of operators who achieved favorable evaluation results.

[0034] The result data 123 of the setting values ​​of the shape control actuators of the rolling equipment 1 refers to the setting information of the shape control actuators specified according to the rolling mill used. When the rolling mill is equipped with multiple shape control actuators, any shape control actuator may be selected from among them and used as the result data 123 of the setting values ​​of the shape control actuators in this embodiment. The setting information of the shape control actuators may be a leveling setting value that controls the tilt state of the axial center of the top backup roll 5 or the bottom backup roll 6. Setting values ​​for the bender force of the intermediate roll bender 11 or the work roll bender 12 may also be used. Furthermore, roll shift setting values ​​that represent the axial shift amount of the intermediate roll 3 or the work roll 2 may also be used. Furthermore, since the 12-high cluster-type rolling mill shown in FIG. 1 is equipped with shape control actuators that adjust the position of the split backup rolls, the setting information of the shape control actuators may include the setting position information of each backup roll, such as the center position, quarter-in position, quarter-out position, and edge position of the top backup roll 5, and the center position, quarter position, and edge position of the bottom backup roll 6. The actual data 123 of the setting values ​​of the shape control actuators may be the setting values ​​(command values) of each control element exemplified above, or may be actual values ​​if they are actually measured. Furthermore, if the bender force or the like can be calculated from the measured value of the hydraulic pressure or the like, it is preferable to use the actual measured value rather than the set command value.

[0035] The shape performance data 121 of the steel sheet S acquired by the data acquisition unit 120 and one or more operation performance data 122 selected from the rolling operation parameters are associated with each other based on the manufacturing control number and rolling pass information of the steel sheet S. These are then stored as a set of data sets in the database unit 131. Multiple data sets stored in the database unit 131 may be generated along the longitudinal direction of the steel sheet S, even for the same rolling pass of the same steel sheet S. For example, if performance data is acquired at a predetermined interval during rolling, e.g., every 100 ms, the number of data sets acquired per rolling pass will be enormous. Therefore, when storing data sets in the database unit 131, it is preferable to perform a screening process to extract only the data sets necessary for generating the shape control actuator setting model M. This reduces the number of data sets stored in the database unit 131.

[0036] For example, it takes approximately 0.25 to 10 seconds from the time the shape control actuators of the rolling equipment 1 are reset until the steel sheet S reaches the shape detector 10. Therefore, the sampling period for acquiring the data set may be set in the range of 0.2 to 10 seconds, and only the data set extracted in this manner may be stored in the database unit 131. In this case, the sampling period may be changed depending on the rolling speed. Furthermore, in regions where the thickness and deformation resistance of the base material vary significantly, such as near the leading and trailing ends of the steel sheet S, the rolling state may fluctuate significantly over time even between the time the shape control actuators of the rolling equipment 1 are reset and the time the steel sheet S reaches the shape detector 10. In this case, the correspondence between the setting values ​​of the shape control actuators and the actual shape may become unclear. In such cases, a process may be performed to exclude the data set acquired at the leading or trailing end of the steel sheet S from the database unit 131. Furthermore, if the output of the shape detector 10 regarding the actual shape fluctuates significantly in a short period of time, a detection error in the shape detector 10 may have occurred. Therefore, if the shape performance fluctuates in a short cycle, a screening process may be performed to remove it from the database unit 131 .

[0037] The shape control actuator setting model generation unit 130 generates a shape control actuator setting model M in a machine learning unit 133 using learning data stored in a database unit 131. The shape control actuator setting model generation unit 130 may be located inside the control computer 100 that controls the rolling process of the steel sheet S by the rolling equipment 1, or may be configured as hardware separate from the control computer 100. Alternatively, the shape control actuator setting model generation unit 130 may be located in a shape control actuator setting unit 140, which will be described later.

[0038] The number of learning data in the data set stored in the database unit 131 is 5×10 ４ ~1 x 10 ６ The number of data sets must be between 1 and 2. This is necessary for efficient learning of the neural network; if the number of data sets is small, sufficient learning cannot be performed, and if the number of data sets is too large, sufficient learning cannot be performed with the combination of parameters described in this specification. Furthermore, the data sets may be classified based on the attribute information and rolling pass number of the steel sheet S, and the shape control actuator setting model M may be generated according to these classifications. The number of data sets stored in the database unit 131 may be limited to a certain number, and the data sets stored in the database unit 131 may be updated as appropriate within this limit. For example, old information that has passed a predetermined time may be deleted from the learning data.

[0039] After the learning data is accumulated in the database unit 131, the preliminary processing unit 132 performs preliminary processing on the data set accumulated in the database unit 131 as necessary before performing machine learning to generate the shape control actuator setting model M. Examples of preliminary processing performed by the preliminary processing unit 132 include normalization of the shape data and performance data of the setting values ​​of the shape control actuators, data shuffling, etc.

[0040] The preliminary processing unit 132 may perform a normalization process on the shape actual data, which is the input actual data stored in the database unit 131, the operation actual data selected from the rolling operation parameters, and the setting actual data of the shape control actuator, which is the output actual data. The normalization process is a process for efficiently updating the weight coefficients during network learning, and is effective in avoiding the gradient vanishing problem when learning a neural network, and can improve the prediction accuracy of the model for the same number of epochs. When normalizing the input variables and output variables, it is preferable to specify the minimum and maximum values ​​of each variable and normalize them in the range of 0 to 1. Alternatively, the average value of the variable may be calculated and normalized in the range of -1 to +1.

[0041] Data sets made up of shape actual data 121, which is input actual data, operation actual data 122 selected from rolling operation parameters, and actual data 123 of setting values ​​of shape control actuators, which is output actual data, are stored in the database unit 131 in the order in which the data was acquired. In addition, data shuffling may be performed to randomly rearrange the order of the data sets stored in the database unit 131. Changing the order of the data sets acquired in chronological order is effective in improving the generalization performance of the neural network and suppressing overlearning.

[0042] The data set stored in the database unit 131 may be divided into training data and test data. The training data, which accounts for 70 to 80% of the entire data set, is used for training the neural network. On the other hand, the remaining data set is used to confirm whether the neural network obtained after training by the machine learning unit 133 can also perform well on the test data.

[0043] The machine learning unit 133 uses the data set accumulated in the database unit 131 to perform machine learning using a plurality of learning data, in which the shape actual data 121 and the operation actual data 122 of the rolling operation parameters are used as input actual data, and the actual data 123 of the setting values ​​of the shape control actuators is used as output actual data, to generate a shape control actuator setting model M that calculates the setting values ​​of the shape control actuators from the shape actual data of the steel sheet S. Any machine learning model can be used as the machine learning model for generating the shape control actuator setting model M, as long as it can obtain sufficient estimation accuracy for the setting values ​​of the shape control actuators for practical use. In this embodiment, it is preferable to use a commonly used fully connected neural network in order to maximize its effects.

[0044] The neural network of this embodiment will now be described with reference to FIG. 5 . Input data acquired from the data acquisition unit 120 is assigned to the input layer and output layer shown in FIG. 5 . More specifically, the data acquisition unit 120 assigns operational performance data for the current time step, shape performance data for the current time step, and performance data for the shape control actuator settings for the current time step to the input layer of the neural network, and performance data for the shape control actuator settings for the next time step to the output layer of the neural network. The neural network of this embodiment acquired from the data acquisition unit 120 must have an input layer with 8 to 256 nodes for this input data and intermediate layers with 1 to 11 nodes each having 16 to 2048 nodes. If these numbers are too small, the model's feature extraction capability will be reduced, resulting in poor prediction accuracy. If these numbers are too large, overlearning due to the complexity of the model will become more pronounced, resulting in poor prediction accuracy. The number of nodes in the output layer must be between 1 and 256. The reason for setting upper and lower limits on the number of nodes in the output layer is determined by the number of targets to be predicted in the target phenomenon, regardless of prediction accuracy. In this embodiment, the number of actuators is equal to the number of shape control actuators of the rolling mill.

[0045] The batch size is 8 to 1024, and the learning rate is 1 × 10 －３ ~1 x 10－６ From the perspective of efficient model training, it is preferable that the batch size is small. If the batch size is small, the number of data considered in each training calculation is too small, which can prevent the entire network from training correctly, resulting in a decrease in prediction accuracy. If the batch size is large, the network will not be able to learn from sudden data included in the training data, which can result in a decrease in prediction accuracy. The learning rate is a hyperparameter used in methods such as gradient descent; if it is too small, learning progress will be slow, and if it is too large, the system will fluctuate around the optimal solution.

[0046] In the data acquired from the data acquisition unit 120, the actual data for the setting values ​​of the shape control actuators at the current time step and the actual data for the setting values ​​of the shape control actuators at the next time step are set according to the following relationship: That is, the actual data for the setting values ​​of the shape control actuators at the current time step after a time difference has elapsed is used as the actual data for the setting values ​​of the shape control actuators at the next time step. This is to take into account the time difference between when the shape control actuators are operated and when the shape is actually measured by a shape meter on the delivery side of the rolling mill. The time difference is preferably determined between 1 and 10 seconds, and is more preferably set by dividing the distance from the rolling mill to the shape meter by the rolling speed at each instant. Furthermore, the actual data for the setting values ​​of the shape control actuators at the next time step assigned to the output layer of the neural network may be the difference from the actual data for the setting values ​​of the shape control actuators at the current time step divided by the time difference.

[0047] The actual data of the setting values ​​of the shape control actuators at the current time step and the actual data of the setting values ​​of the shape control actuators at the next time step acquired from the data acquisition 120 are the operation amounts of the shape control actuators operated by the operator during operation of the rolling mill. The data can be identified by linking a flag indicating the operation by the operator to the data. If there is too little data operated by the operator in the accumulated data, the flag may be used to delete the actual data that the operator did not operate.

[0048] As shown in Figure 6, in the above neural network, the input data arranged in the input layer is multiplied by the weighting coefficients of the intermediate layer and then added to the bias coefficients. This is the same calculation method as commonly known in neural networks. The calculated result is then input into an activation function to obtain the output from the intermediate layer. The activation function here is necessary to give the neural network output nonlinearity, making it possible to accurately predict even nonlinear phenomena.

[0049] These calculation methods are similar to those in the output layer shown in Fig. 7. However, in the output layer, no activation function is used because the output values ​​are expressed as positive, negative, and continuous values.

[0050] From the viewpoint of prediction accuracy, it is preferable to use the ReLU function as the activation function of the intermediate layer. Commonly used functions, such as the PReLU function and the LeakyReLU function, which are derivatives of the ReLU function, as well as the tanh function and the Swish function, can also be used.

[0051] It is preferable to provide a batch normalization layer immediately before the activation function calculation in each hidden layer from the viewpoint of improving prediction accuracy and stabilizing learning calculations. As shown in Figure 8, the batch normalization layer is a calculation that normalizes data using the mean and variance calculated for each batch. This calculation causes the value input to the activation function to fall between 0 and 1, which is the range with a steep slope of the activation function, making it possible to efficiently learn the weight coefficients.

[0052] From the perspective of improving prediction accuracy, it is preferable to provide a dropout layer immediately after the calculation of the activation function in each hidden layer. As shown in Figure 9, a dropout layer is a calculation that invalidates the output with a certain probability. Here, it is preferable to select the probability between 0.1 and 0.5. This calculation enables the network to learn so that it can make accurate predictions without relying on the weight coefficients of specific nodes, and it is possible to suppress overlearning.

[0053] It is extremely difficult to accurately predict the amount of operation of a shape control actuator in a rolling line, even if the rolling conditions and operator are identified. There are three possible reasons for this: (1) Not all information that an operator can recognize is obtained as input data, and predictions may be made using only limited information. (2) Even if the rolled material is the same, it is unlikely that the exact same type of shape control actuator will be operated at the same time. (3) The number of data points in which an operator operates a shape control actuator out of the total number of data points is extremely small, which is unfavorable for neural network learning.

[0054] In this situation, the inventors thought that there must be learning conditions under which neural network learning can be completed stably and with high accuracy. As a result, they investigated the conditions under which neural network learning can be completed stably and with high accuracy by changing various learning conditions. They found that there is an optimal value for the influence of various learning conditions on the model's prediction accuracy, and that if the value is too small or too large, the model's performance will deteriorate.

[0055] Furthermore, in this embodiment, the input data includes not only the shape record data of the steel sheet S but also operational record data selected from the rolling operation parameters because the rolling operation parameters affect the rolling load when rolling the steel sheet S, and the rolling load causes complex changes in the deflection deformation of the work rolls. Furthermore, when performing shape control in the rolling equipment 1, even if the set values ​​of the shape control actuators are the same, if the rolling conditions such as the thickness, width, reduction rate, and deformation resistance of the steel sheet S are different, it is widely known from experience that the shape of the steel sheet S observed as a result of rolling will differ.

[0056] Furthermore, the shape control actuators in the rolling equipment 1 directly or indirectly deflect the work rolls, which primarily apply a rolling force directly to the steel sheet S. The shape control actuators control the shape of the steel sheet S by actively stretching a certain range in the width direction of the steel sheet by changing the deflection of the work rolls. However, like the deflection of a beam, the elastic deformation of the work rolls can only be a smooth curve. The shape of the work rolls also has a curve that can be approximated by a quadratic to octagonal function. This is also true when the split backup rolls of a 12-high cluster-type rolling mill are operated individually. Even if the backup rolls are locally compressed, the intermediate rolls in contact with the backup rolls have a cylindrical shape that is continuous in the axial direction, so the deflection of the intermediate rolls will have a smooth curve, and the deflection of the work rolls in contact with them will also have a smooth curve. Therefore, even if the shape control actuators in the rolling equipment 1 are arbitrarily configured, the actual shape of the steel sheet S detected by the shape detector 10 will have the characteristic of being able to be approximated by a continuous curve in the width direction of the steel sheet S.

[0057] From the above, in this embodiment, it is essential to specify various learning conditions for the neural network in order to efficiently extract information on the rolled material and operation information and to realize stable predictions for the prediction of the operation amount of the shape control actuator in the rolling mill, which is difficult to predict accurately. ４ ~1 x 10 ６ It is noteworthy that favorable learning can be performed between 5×10 and 10. In this embodiment, in order to accurately predict the setting value of the shape control actuator, which is difficult to predict, a minimum of 5×10 ４ training data is required, and conversely, the training data is 1 × 10 ６ If it is more than this, it indicates that other learning conditions need to be changed from those described above.

[0058] The shape control actuator setting model M may be updated to a new model by re-learning, for example, every month or every year. This is because the more data stored in the database unit 131, the more accurate the shape control actuator setting becomes. Furthermore, by updating the shape control actuator setting model M based on the latest data, it is possible to generate a shape control actuator setting model M that reflects changes in operating conditions over time.

[0059] The model learning frequency is preferably 10 to 90 days, and more preferably, re-learning every day or every few hours is preferable from the viewpoint of the model's prediction accuracy.

[0060] [Method for Controlling the Shape of a Steel Sheet] Next, a method for controlling the shape of a steel sheet according to one embodiment of the present invention will be described with reference to FIG.

[0061] In the steel plate shape control method according to this embodiment, the shape control actuator setting model M generated by the machine learning unit 133 is used to determine the setting values ​​of the shape control actuators of the rolling equipment 1. The determined setting values ​​of the shape control actuators are commanded to the shape control actuators to control the shape of the steel plate S. FIG. 10 shows an example configuration of a shape control system for the steel plate S including a shape control actuator setting unit in the rolling equipment. The shape control actuator setting unit 140 shown in FIG. 10 may be provided in a general-purpose computer such as a workstation or a personal computer. The shape control actuator setting unit 140 also has an input unit that acquires the shape control actuator setting model M generated by the shape control actuator setting model generation unit 130 via a network and a memory unit that stores the model. However, the shape control actuator setting unit 140 may also be configured inside the control computer 100 or the rolling control controller 110 for controlling the rolling equipment 1.

[0062] In the shape control of the steel sheet S using the shape control actuator setting unit 140 shown in FIG. 10 , the control computer 100 calculates initial setting values ​​for rolling the steel sheet S, and the rolling controller 110 initializes the shape control actuators. Thereafter, a rolling pass is started. Then, during the execution of the rolling pass, the data acquisition unit 120 acquires operation record data 122 selected from the rolling operation parameters of the rolling equipment 1 and shape record data 121 acquired from the shape detector 10. The data acquired by the data acquisition unit 120 is sent to the shape control actuator setting unit 140. The shape control actuator setting unit 140 calculates the setting values ​​of the shape control actuators by acquiring the data from the shape control actuator setting model generation unit 130 and inputting it into the shape control actuator setting model M stored in the memory unit. Then, the calculated setting values ​​of the shape control actuators are sent to the rolling controller 110 as new shape control actuator setting command values. The rolling controller 110 updates the current shape control actuator settings with the shape control actuator setting command values ​​acquired from the shape control actuator setting unit 140. As a result, the rolling controller 110 operates the shape control actuators of the rolling equipment 1 with the new setting values.

[0063] By acquiring such shape control actuator setting command values ​​at each preset shape control control period and sending them to the rolling controller 110, the shape control actuator setting values ​​are updated as needed in the longitudinal direction of the steel sheet S. Shape control of the steel sheet S can then be performed. The shape control control period is preferably set to approximately 0.2 to 10 seconds, since it takes a certain amount of time from updating the shape control actuator setting values ​​of the rolling equipment 1 until the steel sheet S reaches the shape detector 10. This allows the production of steel sheets S with a uniform shape in the longitudinal direction. Furthermore, many conventional automatic shape feedback controls update the shape control actuator setting values ​​as needed using a shape control actuator setting model constructed using a physical model or the like based on actual shape data detected by the shape detector 10. Therefore, steel sheet shape control using the shape control actuator setting unit 140 of this embodiment can be achieved by replacing part of a system for performing conventional automatic shape feedback control. Furthermore, the shape control actuator setting unit 140 of this embodiment may be provided in parallel with a conventional automatic shape feedback control system, and switching may be performed as needed to perform shape control of the rolling equipment.

[0064] [Method for manufacturing steel sheet] A method for manufacturing a steel sheet according to another embodiment of the present invention includes a rolling step of rolling the steel sheet S while controlling the shape of the steel sheet S using the above-described method for controlling the shape of the steel sheet S. This makes it possible to manufacture steel sheets having a desired shape along the longitudinal direction of the steel sheet S with a good yield.

[0065] An example of the present invention is described below. In this example, a shape control actuator setting model was first generated for a 12-high cluster rolling mill. The rolling mill used in this example had a work roll diameter of 70 to 120 mm and a maximum rolling speed of 600 m / min. The steel plate to be rolled was high-carbon steel, with a base plate thickness of 1.2 to 2.3 mm and a plate width of 400 to 1070 mm. The number of rolling passes ranged from 1 to 15, resulting in a steel plate with a product thickness of 0.4 to 1.8 mm. The rolling mill used was a reversing mill, with a shape detector positioned between two tension reels on the left and right of the rolling mill's center, allowing for acquisition of actual shape data for the steel plate in both right- and left-going rolling passes. The shape detector was a split load cell type with 44 load cell channels built in along the roll axis, with a detection width of 25.4 mm per channel. The shape detector outputs the value of I-unit at each position in the width direction of the steel sheet as the difference in elongation in the width direction of the steel sheet. In this example, the above steel sheet was rolled into 420 coils, and during this rolling, an operator with extensive experience in operating this rolling mill manually operated the shape control actuator, and learning data was accumulated in the database.

[0066] In generating the shape control actuator setting model, the entry thickness, delivery thickness, and width were selected as the dimensional information of the steel plate from among the rolling operation parameters. In addition, the steel grade code of the steel plate was selected as the attribute information of the base material. Furthermore, the rolling pass number, work roll diameter, rolling speed, coolant flow rate, and coolant temperature were selected from the rolling condition information. In this example, in addition to these rolling operation parameters, the coefficient λ when the shape of the steel plate shown in Equation (1) is functionally approximated by a quartic equation was used as the target shape of the steel plate. １ ~λ ４ On the other hand, the I-unit output from the shape detector was used as the shape actual data when the steel plate was rolled with the above rolling operation parameters. However, for the width direction position of the steel plate, a value normalized by the maximum width that can be rolled by this rolling mill was used, and actual data of I-unit at each position was collected.

[0067] On the other hand, six shape control actuators, which are output data of the shape control actuator setting model, were selected: work roll bender, intermediate roll bender, leveling, edge position of the lower backup roll, quarter position, and center position, and setting record data for these were acquired. In this example, a data set consisting of the above record data and shape control actuator setting record data was stored in the database unit. The number of data sets stored in the database unit was approximately 1.4 × 10 ６ Once this number had been reached, the shape control actuator setting model generation unit performed preliminary processing to divide the data, and 70% of the entire data set was used as training data to train the neural network, with the remaining data set being used as test data to evaluate the prediction accuracy.

[0068] As shown in Figure 11, the neural network used in this example had an input layer with 64 nodes, seven hidden layers with 512 nodes, and an output layer with 6 nodes. The ReLU function was used as the activation function, and the output of the activation function was input to a dropout layer with a probability of 0.3. The explanatory variables are as shown in Figure 11. The manipulated variable of the shape control actuator was divided by a time difference of 6 seconds, then normalized by the maximum and minimum values ​​within the actual results, and set to -100 to 100% for each shape control actuator.

[0069] The loss function in the learning calculation was the mean square error calculated from the network's predicted value and actual measured value, and the optimization calculation method was ADAM optimization. In addition, mini-batch learning (mini-batch gradient descent) was used for machine learning, with a batch size of 512, epoch count of 500, and an initial learning rate of 1 x 10 －３ However, if the loss function for the test data did not improve over 30 epochs, the learning rate was reduced to 1 / 10 and learning was continued. The minimum value of the learning rate was 1 × 10 －６Using the shape control actuator setting model generated as described above, the output values ​​of the shape control actuator setting values ​​and the actual values ​​were compared for the test data. In contrast, as a comparative example, learning was performed using a feedforward neural network using the same data set as the data set stored in the database unit, and the output values ​​of the generated model were compared with the actual values. Note that for the feedforward neural network of the comparative example, learning data was set to 3 x 10 ４ and other parameters were the same.

[0070] Figure 12 shows a comparison of the output values ​​and actual values ​​of the shape control actuator setting values ​​in the example and the comparative example. Note that Figure 12 shows the average loss function for the six shape control actuators in the output layer, and the smaller the loss function, the better the prediction accuracy. As shown in Figure 12, the prediction results using the neural network in the example show a significant improvement in the loss function compared to the comparative example. From the above, it was confirmed that the shape control actuator setting values ​​manually set by the operator can be accurately reproduced using a trained model based on a neural network.

[0071] REFERENCE SIGNS LIST 1 Rolling equipment 2 Work roll 3 Intermediate roll 4 Small backup roll 5 Upper backup roll 6 Lower backup roll 7 Payoff reel 8, 9 Tension reel 10 Shape detector 11 Intermediate roll bender 12 Work roll bender 100 Control computer 110 Rolling control controller 120 Data acquisition unit 130 Shape control actuator setting model generation unit 131 Database unit 132 Preliminary processing unit 133 Machine learning unit 140 Shape control actuator setting unit M Shape control actuator setting model S Steel plate UP Upward DN Downward FD Rolling direction WD Width direction

Claims

1. A method for generating a shape control actuator setting model for rolling equipment, which is provided with a shape control actuator that controls the shape of a steel plate and a shape detector that detects the shape of the steel plate, for determining the setting values ​​of the shape control actuator, comprising: a data acquisition step of acquiring one or more operation record data selected from rolling operation parameters of the rolling equipment and shape record data acquired from the shape detector; and a neural network consisting of an input layer having 8 to 256 nodes, 1 to 11 intermediate layers having 16 to 2048 nodes, and an output layer having 1 to 256 nodes, the data acquired in the data acquisition step being input record data and the setting values ​​of the shape control actuator being output record data, the neural network being configured as 5x10 ４ ~1 x 10 ６ and generating the shape control actuator setting model using the number of learning data of the shape control actuator setting model.

2. The neural network training was performed with a batch size of 8 to 1024, 1 x 10 －３ ~1 x 10 －６ The method for generating a shape control actuator setting model for rolling equipment according to claim 1 , wherein the method is performed using a learning rate between 3. A method for setting shape control actuators for rolling equipment, comprising a setting value update step of determining and updating the setting values ​​of the shape control actuators using a shape control actuator setting model generated according to the method for generating a shape control actuator setting model for rolling equipment set forth in claim 1 or 2.

4. A method for controlling the shape of a steel plate, comprising a step of updating the setting value of the shape control actuator during rolling of the steel plate, using the method for setting the shape control actuator of rolling equipment according to claim 3.

5. A steel plate shape control system comprising means for updating the setting value of the shape control actuator during rolling of the steel plate using the method for setting the shape control actuator of rolling equipment according to claim 3.

6. A method for manufacturing a steel plate, comprising a rolling step of rolling a steel plate using the method for controlling the shape of a steel plate according to claim 4.