Learning control device for rolling process

Abstract

One embodiment of the present invention relates to a learning control device for a rolling process, which calculates a predicted value based on an actual value measured in the rolling process, calculates an instantaneous value of a learning coefficient based on a difference between the calculated predicted value and the actual value of the rolling process, calculates an updated value of the learning coefficient based on the instantaneous value of the learning coefficient and a previous value of a unit of a rolling condition coincidence of a learning coefficient table, updates the learning coefficient of the unit of the rolling condition coincidence of the learning coefficient table, determines a time of occurrence of a learning coefficient mutation based on the instantaneous value of the learning coefficient saved in a rolling information database, detects a level deviation of the instantaneous value of the learning coefficient before and after the time of occurrence of the learning coefficient mutation as a learning coefficient mutation component, and renews the learning coefficient of the learning coefficient table by correction based on the instantaneous value of the learning coefficient saved in the rolling information database after the determined time of occurrence of the learning coefficient mutation and the learning coefficient mutation component.

Description

Technical Field

[0001] The invention relates to a learning control device for a rolling process. Background Art

[0002] In rolling mills, non-ferrous materials such as steel, aluminum, and copper are rolled to produce metal strips used in the manufacture of automobiles, electric motors, and other products. The rolling process (rolling process) includes various types of rolling mills, including hot sheet rolling mills, heavy plate rolling mills, cold rolling mills, and rolling mills for rolling wire rods.

[0003] Furthermore, in any rolling process, the finished product is controlled so that target values ​​such as temperature, which affect desired dimensions, shapes, and mechanical properties, are maintained. Generally speaking, there are two types of rolling process control: setpoint control and dynamic control.

[0004] In setting control, mathematical models that reproduce the phenomena during the rolling process are used to determine the set values ​​for various equipment, such as the rolling mill speed, cooling water flow rate, and roll gap, so that the rolled material reaches the desired size and temperature. To reduce the computational load, these mathematical models are often simplified.

[0005] Therefore, there may be a discrepancy between the predicted values ​​of the rolling phenomena calculated by the mathematical model and the actual values ​​measured by the measuring instruments installed in the equipment. This discrepancy manifests as errors in the target dimensions and temperature of the product, resulting in defective products that fall outside the tolerances for quality assurance.

[0006] Furthermore, in recent years, requirements for product specifications have become increasingly sophisticated and diversified, and strict management has been required for product quality assurance. Therefore, there is a demand for improvement and stabilization of the prediction accuracy of calculation models.

[0007] Furthermore, in rolling process control, a learning coefficient is set for the calculation model, and the learning coefficient is adjusted based on the deviation between the predicted value and the actual value, thereby improving the accuracy of the prediction based on the calculation model and stabilizing it.

[0008] Generally speaking, the learning coefficient is determined by comparing the predicted value of the prediction target obtained from the actual value with the actual value of the prediction target. The learning coefficient obtained here is the learning coefficient relative to the rolled material, that is, the instantaneous value.

[0009] Furthermore, the instantaneous value of the learning coefficient may vary greatly due to factors omitted from the simplified calculation model, measurement errors of the measuring instrument, and various disturbances in the rolling process. Therefore, the instantaneous value of the learning coefficient is smoothed and applied as an updated value.

[0010] The updated value of the learning coefficient is generally recorded in the corresponding division (hereinafter referred to as "unit") in the learning table divided based on processing conditions, i.e., rolling conditions, such as product target thickness, width, temperature, material composition, reduction rate, and number of processing passes.

[0011] Specifically, the rolling process learning control device uses a table divided according to rolling conditions to obtain appropriate learning coefficients corresponding to the rolling conditions. By using appropriate learning coefficients, the rolling process learning control device improves the accuracy of rolling phenomenon predictions based on the algorithmic model and ensures rolling stability.

[0012] Even when simply updating the learning coefficient of one cell in the table that matches the rolling conditions, a large amount of actual rolling results is required to sufficiently update and converge the learning coefficients of all cells in the table that can be handled operationally.

[0013] As a countermeasure, for example, as described in Patent Document 1, there is a method of updating the learning coefficient of a cell in a table that matches the rolling conditions while also updating the learning coefficients of cells with similar rolling conditions. Here, multiple cells adjacent to the matching cell are updated.

[0014] According to this method, the learning coefficient can be fully updated with fewer actual rolling results. Therefore, even under rolling conditions for which there are no actual rolling results, a sudden decrease in prediction accuracy can be prevented, and stability of rolling performance can be ensured.

[0015] On the other hand, learning control using tables divided according to rolling conditions has the problem of difficulty following time-series changes in the rolling process. For example, even if the learning coefficients in the table are fully updated to meet operational requirements, if there are cells with actual rolling results under the corresponding rolling conditions, if the rolling process changes, the learning coefficients will become inappropriate, and there is a risk of significantly reducing prediction accuracy under these rolling conditions.

[0016] A solution to this problem has been proposed, for example, based on the method disclosed in Patent Document 2. For example, Patent Document 2 proposes separating a time series learning coefficient that compensates for errors caused by time series variations in the deviation between a predicted value based on a formula model and an actual value from a learning coefficient corresponding to the rolling conditions, and correcting the predicted value based on these two learning coefficients.

[0017] The time-series variation here refers to, for example, linear variation in behavior such as the effect of a reduction in roll diameter due to wear caused by rolling friction. This method can appropriately obtain a learning coefficient for each rolling condition that eliminates this time-series rolling process variation.

[0018] Prior art literature

[0019] Patent Literature

[0020] Patent Document 1: Japanese Patent Application Laid-Open No. 6-259107

[0021] Patent Document 2: Japanese Patent Application Laid-Open No. 4-367901 Summary of the Invention

[0022] Problems to be solved by the invention

[0023] Prediction errors in the set calculations, a factor contributing to poor hot rolling quality, include prediction errors in the mathematical model representing the deformation characteristics of the rolled material, as well as mechanical and measurement errors. These errors can occur unexpectedly due to equipment failure, roll replacement, repair or replacement, poor calibration, operator error, or changes in weather conditions.

[0024] However, it is extremely difficult to immediately identify the main causes of these sudden errors from the information measured by sensors on the rolling line. In many cases, the main causes of the errors are not determined until several to dozens of rolls have been produced after the main causes of the errors have occurred.

[0025] In this case, from the time these error factors occur until these error factors are identified and appropriate responses are completed, the prediction errors of the set calculations detected by sensors on the rolling line are learned as prediction errors based on the formula model.

[0026] In the learning methods of each volume in the past, and the learning methods based on past time series changes, if the errors of these different main factors are regarded as the prediction errors of the formula model and learned, there is a possibility that the errors of the formula model will interfere with the learning of the original object phenomenon, the learning coefficients will be updated inaccurately, and the prediction accuracy of the set calculation will be reduced.

[0027] Furthermore, inaccurate learning results during this period are sequentially written into learning tables divided by rolling conditions. Therefore, their impact varies with each learning division, depending on the update frequency. Even if the main error factors and their impact can be identified after a certain level of learning, it is difficult to correct the learning table. This means that continued use of inaccurate learning results can lead to quality defects.

[0028] The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to provide a learning control device capable of correcting a subsequent learned rolling process even when an error factor due to a sudden fluctuation occurs.

[0029] Means for solving problems

[0030] An embodiment of the present invention relates to a learning control device for a rolling process, characterized in that the learning control device for the rolling process saves and controls the rolling process while updating the learning coefficient of the formula model used for calculating the set value for the rolling process through a learning coefficient table composed of a plurality of units divided according to rolling conditions. The learning control device for the rolling process comprises: a predicted value calculation unit for calculating a predicted value based on an actual value measured in the rolling process; an instantaneous value calculation unit for calculating an instantaneous value of the learning coefficient based on a difference between the predicted value calculated by the predicted value calculation unit and the actual value of the rolling process; an updating unit for calculating an updated value of the learning coefficient based on the instantaneous value of the learning coefficient calculated by the instantaneous value calculation unit and a previous value of a unit in the learning coefficient table that meets the rolling conditions, and updates the learning coefficient of the unit in the learning coefficient table that meets the rolling conditions; An information database stores the instantaneous value of the learning coefficient calculated by the instantaneous value calculation unit, the previous value of the learning coefficient, the updated value of the learning coefficient, the date and time information for determining the rolled material, the rolling conditions, the coordinates of the units of the learning coefficient table based on the rolling conditions, the actual values ​​in the rolling process, and the date and time history of events in the rolling process; a mutation detection unit determines the occurrence time of the mutation of the learning coefficient based on the instantaneous value of the learning coefficient stored in the rolling information database, and detects the deviation in the level of the instantaneous value of the learning coefficient before and after the occurrence time of the mutation of the learning coefficient as the learning coefficient mutation component; and a re-updating unit re-updates the learning coefficient of the learning coefficient table by correcting the instantaneous value of the learning coefficient stored in the rolling information database after the occurrence time of the mutation of the learning coefficient determined by the mutation detection unit and the learning coefficient mutation component.

[0031] Furthermore, the rolling process learning control device according to one embodiment of the present invention further includes a notification unit configured to notify the necessity of maintenance as an event when the sudden change detection unit identifies the occurrence time of the sudden change in the learning coefficient.

[0032] In addition, with regard to the learning control device for the rolling process of one embodiment of the present invention, after the mutation detection unit determines the occurrence time of the mutation of the learning coefficient, when there is no date and time history maintained as an event in the rolling information database, the re-updating unit re-updates the learning coefficient of the learning coefficient table based on the instantaneous value of the learning coefficient in the rolling after the occurrence time of the mutation of the learning coefficient, the mutation component of the learning coefficient, and the previous value of the learning coefficient of the unit based on the rolling condition of the learning coefficient table.

[0033] Effects of the Invention

[0034] According to the present invention, even if an error factor due to a sudden change occurs, subsequent learning can be corrected. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a diagram illustrating the configuration of a learning control device for a rolling process according to the first embodiment.

[0036] Figure 2 This is a diagram illustrating data stored in the rolling information database.

[0037] Figure 3 This is a diagram illustrating a learning coefficient table and data stored in the learning coefficient table for each event.

[0038] Figure 4 This is a flowchart illustrating the processing performed by the mutation detection unit.

[0039] Figure 5 This is a diagram illustrating the results of detection by the sudden change detection unit identifying a sudden change in the learning coefficient.

[0040] Figure 6 (a) is a graph illustrating the results of change point detection using changes in the instantaneous value of the learning coefficient. (b) is a graph illustrating the likelihood trend in the change point detection method using maximum likelihood and least squares. (c) is a graph illustrating the trend of the absolute value of the degree of change in change point detection using cumulative sum.

[0041] Figure 7 This is a flowchart showing a specific example of the processing performed by the re-updating unit.

[0042] Figure 8 This is a diagram illustrating the configuration of a learning control device for a rolling process according to a second embodiment. DETAILED DESCRIPTION

[0043] Hereinafter, embodiments of a learning control device for a rolling process will be described with reference to the drawings. Figure 1 This is a diagram illustrating the configuration of a learning control device 1 for a rolling process according to the first embodiment.

[0044] The learning control device 1 has a computer function including a CPU and memory (not shown), and is a device that controls the rolling process while learning a rolling mill (equipment) that rolls non-ferrous materials such as steel, aluminum, and copper.

[0045] Furthermore, the learning control device 1 controls the rolling process by updating and storing the learning coefficients of the formula model used to calculate the rolling process setting values ​​using a learning coefficient table composed of multiple units divided according to rolling conditions. Specifically, the learning control device 1 includes, for example, a storage unit 2, a learning unit 3, a setting calculation unit 4, and a learning coefficient re-updating unit 5.

[0046] The storage unit 2 is a device that stores, for example, a rolling information database (DB) 20 , a learning coefficient table 22 , and a learning coefficient table 24 for each event.

[0047] For example, Figure 2 As shown, the rolling information database 20 is a database that stores the manufacturing number, manufacturing date and time, and rolling conditions for each rolled material, and stores the instantaneous value of the learning coefficient of the unit that is the update object, as well as the previous value, updated value, and coordinate information of the unit based on the rolling conditions in the learning coefficient table.

[0048] In addition, the rolling information database 20 can also save information on the manufacturing date and time of the rolled material (or the manufacturing number of the rolled material), a history of the date and time of events such as maintenance, regular inspection or equipment replacement of the rolling process, and actual values ​​in the rolling process used when predicting the rolling phenomenon of the object.

[0049] like Figure 3 As shown, the learning coefficient table 22 is a table composed of a plurality of cells divided according to rolling conditions, and the learning coefficient is recorded (stored) in the cell that meets the rolling conditions.

[0050] As described above, the instantaneous value of the learning coefficient varies greatly due to various disturbances, etc. Therefore, the updated value of the learning coefficient is obtained using the smoothed instantaneous value of the learning coefficient. For example, the updated value of the learning coefficient is calculated using the following formula (1).

[0051] [Formula 1]

[0052] Z (i，j)NEW =Z (i，j)USED ·(1-α)+Z (i，j)CURRENT ·α···(1)

[0053] Z (i，j)NEW : The updated value of the learning coefficient of the unit

[0054] Z(i，j)USED : The previous value of the learning coefficient of this unit

[0055] Z (i，j)CURRENT : The instantaneous value of the learning coefficient of the unit

[0056] α: smoothing coefficient

[0057] i, j: the coordinates of the unit in the learning coefficient table

[0058] The coordinates of the cell in the learning coefficient table 22 are described here as two variables, but the number of variables is not limited to this. For example, if the steel grade is used as a classification in addition to the target product width and target product thickness values, the coordinates of the cell are determined by three variables.

[0059] The updated value of the learning coefficient is recorded by updating the learning coefficient of the cell to be updated in the learning coefficient table 22. Furthermore, the coordinate information of the cell with respect to the rolling conditions in the learning coefficient table 22, the previous value of the learning coefficient, and the actual value used when calculating the predicted value of the predicted object are recorded in the rolling information database 20.

[0060] In the process of updating the learning coefficient, a plurality of cells adjacent to the cell to be updated may be updated simultaneously to promote the updating of the entire learning coefficient table 22. For example, as shown in the following equation (2), the learning coefficients of adjacent cells may be updated.

[0061] [Formula 2]

[0062] Z (i′，j′)NEW =Z (i′，j′)USED (1-β)+Z (i，j)CURRENT ·β ···(2)

[0063] Z (i′，j′)NEW : Updated value of the learning coefficient of the adjacent unit

[0064] Z (i′，j′)USED : Previous value of the learning coefficient of the adjacent unit

[0065] β: Smoothing coefficient in the update of learning coefficients of adjacent units

[0066] i', j': coordinates of the cell adjacent to this cell in the learning coefficient table

[0067]

[0068] In addition, the learning coefficients recorded in the learning coefficient table 22 are copied to the learning coefficient table 24 for each event according to the maintenance events such as the regular inspection and equipment replacement of each device. Figure 3The configuration of the learning coefficient table 22 shown is the same.

[0069] Learning Department 3( Figure 1 ) is a device having a predicted value calculation unit 30, an instantaneous value calculation unit 32 and an updating unit 34.

[0070] The predicted value calculation unit 30 calculates a predicted value of a prediction target based on an actual value measured during the rolling process for the calculation model used in the setting calculation, and outputs the calculated value to the instantaneous value calculation unit 32 .

[0071] The instantaneous value calculation unit 32 calculates the instantaneous value of the learning coefficient based on the actual value of the prediction target, outputs the calculated instantaneous value to the updating unit 34, and stores the instantaneous value together with the rolling conditions in the rolling information database 20. For example, the instantaneous value calculation unit 32 calculates the instantaneous value of the learning coefficient based on the difference between the predicted value calculated by the predicted value calculation unit 30 and the actual value of the rolling process.

[0072] The updating unit 34 calculates an updated value of the learning coefficient based on the instantaneous value calculated by the instantaneous value calculating unit 32, and outputs the calculated updated value to the rolling information database 20 and the learning coefficient table 22. For example, the updating unit 34 calculates an updated value of the learning coefficient based on the instantaneous value of the learning coefficient calculated by the instantaneous value calculating unit 32 and the learning coefficient (previous value) of the cell to be updated corresponding to the rolling condition in the learning coefficient table 22, and updates the learning coefficient of the cell in the learning coefficient table 22 that meets the rolling condition.

[0073] The setting calculation unit 4 includes a learning coefficient reading unit 40 and a setting calculation unit 42 , and is a device that determines setting values ​​for each device using a calculation model that reproduces the phenomenon of the rolling process.

[0074] In order to improve the prediction accuracy, the learning coefficient reading unit 40 reads the learning coefficient of the cell that matches the rolling condition in the learning coefficient table 22 and outputs it to the setting calculation unit 42 .

[0075] The setting calculation unit 42 corrects the setting value for each device using the learning coefficient output by the learning coefficient reading unit 40 , and outputs the corrected setting value to each device.

[0076] The learning coefficient re-updating unit 5 includes a mutation detection unit 50, a determination unit 52, and a re-updating unit 54, which detects sudden changes in the learning coefficient, determines the time of occurrence, and calculates the mutation component. After the time of occurrence of the mutation of the learning coefficient, the mutation component (learning coefficient mutation component) is used to correct the learning coefficient stored in the learning coefficient table 22.

[0077] For example, the mutation detection unit 50 has a change point detection function for detecting the mutation of the learning coefficient based on the instantaneous value of the learning coefficient stored in the rolling information database 20, detecting the sudden change (mutation) of the learning coefficient, determining the time when the mutation occurs, and calculating the deviation in the level of the instantaneous value of the learning coefficient before and after the time when the mutation occurs as the learning coefficient mutation component.

[0078] Figure 4 This is a flowchart illustrating the processing performed by the sudden change detection unit 50. First, after rolling of the rolled material is completed, the sudden change detection unit 50 obtains the instantaneous value of the learning coefficient corresponding to the rolling conditions I and J from the rolling information database 20 (S100).

[0079] Then, the sudden change detection unit 50 determines whether the number of acquired instantaneous value data has reached N ( S102 ). If it has not reached N ( S102 : No), the process returns to S100 . If it has reached N ( S102 : Yes), the process proceeds to S104 .

[0080] That is, the sudden change detection unit 50 has traditionally obtained instantaneous values ​​of the learning coefficient corresponding to the number of rolled bars N from the rolling information database 20. Alternatively, the sudden change detection unit 50 may obtain instantaneous values ​​of the learning coefficient for not only one segment that meets the rolling conditions but also adjacent segments.

[0081] In this case, the conditions for obtaining the instantaneous value of the learning coefficient are as shown in the following formula (3).

[0082] [Formula 3]

[0083] Z (I，J)CURRENT (n)(n=1 to N)···(3)

[0084]

[0085] Z (I，J)CURRENT (n): The instantaneous value of the learning coefficient for the rolling condition and its adjacent conditions

[0086] I, J: Coordinates of the rolling condition and its adjacent conditions

[0087] i, j: coordinates of the rolling condition

[0088] s l , s u : The adjacent condition range of the rolling condition

[0089] n E : The time at which the event occurred is assigned

[0090] The number N of rolled pieces from which the sudden change detection unit 50 obtains the instantaneous value of the learning coefficient is preferably included from the event occurrence time n E Here, the acquired condition is defined as the rolling condition and its adjacent segments. This is to allow for time series analysis of learning coefficients at the same level, even if learning coefficients for segments not adjacent to the rolling condition have completely different values. If learning coefficients are at the same level regardless of the rolling condition, the sudden change detection unit 50 may also acquire all learning coefficients in a time series.

[0091] Using the instantaneous value of the learning coefficient thus obtained, the sudden change detection unit 50 detects a sudden change in the learning coefficient. Furthermore, the sudden change detection unit 50 uses a conventional change point detection method to determine the presence and timing of a sudden change in the learning coefficient. Examples of change point detection methods include methods using maximum likelihood and least squares, and methods using cumulative sums. These methods will be described below.

[0092] The method for detecting change points using maximum likelihood and least squares is as follows: when the change in the instantaneous value of the learning coefficient of the rolling condition and its adjacent conditions is divided into τth, the moment when the likelihood in the interval before and after τ becomes maximum or minimum is found.

[0093] The specific details are shown below: Here, the change point τ and the transition of the instantaneous value of the learning coefficient of the rolling condition and its adjacent conditions are defined as shown in the following formula (4).

[0094] [Formula 4]

[0095]

[0096] At this time, μ1, μ2, and τ are determined by the least squares method as follows so as to minimize the likelihood U. Here, the residual sum of squares is used as the likelihood, but the present invention is not limited to this. In addition, yk represents the kth Z CURRENT.

[0097] [Formula 5]

[0098]

[0099] [Formula 6]

[0100]

[0101] [Formula 7]

[0102]

[0103] [Formula 8]

[0104]

[0105] [Formula 9]

[0106] Learning coefficient mutation component:

[0107] In addition, the change point detection method using cumulative sums is a method that accumulates the degree of change between numerical values ​​according to time or time series, and determines an abnormality when the cumulative sum exceeds a threshold. The degree of change Sc is calculated as shown in the following formula (10).

[0108] [Equation 10]

[0109] S c (n) = S c (n-1)+(Z (I，J)CURRENT (n)-μ Z )···(10)

[0110] S c (n): Degree of change under rolling number n

[0111] S c (0) = 0: Initial value of the degree of change

[0112] μ Z : The average value of the learning coefficients obtained

[0113] At this time, as shown in the following formula (11), the absolute value of Sc(n) at the time of change becomes the maximum.

[0114] [Formula 11]

[0115]

[0116] Furthermore, the learning coefficient sudden change component (difference) in this method is calculated as shown in the following formula (12).

[0117] [Equation 12]

[0118]

[0119] In addition, if Figure 5 As shown, the deviation of the average value of the learning coefficient before and after the change point is detected when the instantaneous value of the learning coefficient suddenly changes.

[0120] In this way, the sudden change detection unit 50 obtains the time of change of the learning coefficient and the deviation of the average value of the learning coefficient before and after the change point (S104: Figure 4 ).

[0121] Figure 6This is a diagram illustrating the results of detecting change points using the data of a hot rolling mill according to the above-mentioned method. Figure 6 (a) is a diagram illustrating the result of detecting a change point based on the change in the instantaneous value (average value) of the learning coefficient. Figure 6 (b) is a diagram illustrating the trend of likelihood in a change point detection method using maximum likelihood and the least squares method. Figure 6 (c) is a graph illustrating the trend of the absolute value of the degree of change in change point detection using the cumulative sum. The target of change point detection is the instantaneous value of the learning coefficient in the product width prediction formula model.

[0122] exist Figure 6 In (a), not only one division that meets arbitrary rolling conditions is obtained, but also the learning coefficients of about 10,000 divisions including adjacent divisions are obtained, and the trend is shown. Figure 6 (a) shows that a sudden change in the learning coefficient occurs near the center of the trend.

[0123] In addition, if Figure 6 As shown in (b) and (c), the minimum likelihood value for change point detection using maximum likelihood and least squares, and the maximum absolute value of the degree of change for change point detection using cumulative sums, are shown at the same time. These coincide with the moment of sudden change in the learning coefficient, appropriately capturing the moment of the sudden change.

[0124] Note that the change point detection method described here is merely an example, and the change point detection method applicable to the present invention is not limited thereto.

[0125] In this way, the mutation detection unit 50 determines the occurrence time of the mutation of the learning coefficient based on the instantaneous value of the learning coefficient stored in the rolling information database 20, and detects the deviation in the level of the instantaneous value of the learning coefficient before and after the occurrence time of the mutation of the learning coefficient as the learning coefficient mutation component.

[0126] Determination unit 52 ( Figure 1 ) It is determined whether the learning coefficient mutation component at the change point detected by the mutation detection unit 50 is greater than the mutation determination threshold ε.

[0127] When the determination unit 52 determines that the learning coefficient mutation component is greater than or equal to the mutation determination threshold ε, the re-updating unit 54 re-updates the learning coefficients in the learning coefficient table 22. For example, the re-updating unit 54 re-updates the learning coefficients after the mutation time based on the change time detected by the mutation detection unit 50 and the learning coefficient mutation component. In this case, the re-updating unit 54 may also store the learning coefficient mutation component separately in advance.

[0128] Figure 72 is a flowchart showing a specific example of the processing performed by the re-updating unit 54. First, the re-updating unit 54 obtains the learning coefficient table 22 from the storage unit 2 (S200).

[0129] Next, the re-updating unit 54 acquires the instantaneous value of the learning coefficient and the coordinate information of the cell with respect to the rolling condition in the learning coefficient table 22 from the rolling information database 20 ( S202 ).

[0130] Then, based on the history of event occurrence dates and times, the re-updating unit 54 determines whether the instantaneous value of the learning coefficient is an instantaneous value after the sudden change time (S204). If the instantaneous value is an instantaneous value after the sudden change time (S204: Yes), the re-updating unit 54 proceeds to S206. If the instantaneous value is not an instantaneous value after the sudden change time (S204: No), the re-updating unit 54 proceeds to S208.

[0131] In S206 , the re-updating unit 54 performs correction by adding a sudden change component of the learning coefficient to the instantaneous value of the learning coefficient, and calculates an updated value.

[0132] Furthermore, in S208 , the re-updating unit 54 calculates an updated value using the instantaneous value of the learning coefficient.

[0133] Then, the re-updating unit 54 re-updates the corresponding cell of the learning coefficient table 22 using the learning coefficient stored in the learning coefficient table 24 for each event as the previous value ( S210 ).

[0134] The updating order performed by the re-updating unit 54 satisfies the conditions shown in the following equation (13).

[0135] [Formula 13]

[0136]

[0137] Z E_(i，j)NEW : Updated value of the learning coefficient of the corresponding unit

[0138] Z E_(i，j)USED : The learning coefficient of the corresponding unit in the learning coefficient table of each event

[0139] Z (i，j)CURRENT (n'): The instantaneous value of the learning coefficient of the corresponding unit after the event date and time

[0140] α: smoothing coefficient

[0141] i, j: The coordinates of the cell in the learning coefficient table for each event

[0142] Learning coefficient mutation component

[0143] T CURRENT: The manufacturing date and time of the rolled material at which the instantaneous value of the learning coefficient of the corresponding unit was obtained

[0144] Moments of change Manufacturing date of the rolled material

[0145] When updating the learning coefficients in the learning coefficient table 22, when a plurality of adjacent cells are updated simultaneously, the plurality of adjacent cells are updated in the same order as shown in the following equation (14).

[0146] [Equation 14]

[0147]

[0148] Z E_(i′，j′)NEW : Updated value of the learning coefficient of the adjacent unit

[0149] Z E_(i′，j′)USED : The learning coefficient of each event in the adjacent unit learning coefficient table

[0150] β: Smoothing coefficient in the update of learning coefficients of adjacent units

[0151] i', j': coordinates of the cells adjacent to this cell

[0152]

[0153] That is, the re-updating unit 54 re-updates the learning coefficients in the learning coefficient table 22 by correcting the instantaneous values ​​of the learning coefficients and the learning coefficient sudden change components stored in the rolling information database 20 after the occurrence time of the sudden change of the learning coefficients determined by the sudden change detecting unit 50 .

[0154] In this manner, the learning coefficient re-updating unit 5 rewrites the learning coefficient of the learning coefficient table 24 for each event after re-update into the learning coefficient table 22 , thereby correcting the learning coefficient sudden change component.

[0155] Next, a second embodiment of the learning control device for the rolling process will be described. Figure 8 This is a diagram illustrating the configuration of a learning control device 1 a for a rolling process according to a second embodiment.

[0156] For example, the learning control device 1a includes a storage unit 2, a learning unit 3, a setting calculation unit 4, and a learning coefficient re-updating unit 5a. Figure 8 In the learning control device 1a shown in FIG. Figure 1 In the learning control device 1 shown, substantially the same components are given the same reference numerals.

[0157] When the learning control device 1a detects a sudden change in the learning coefficient and updates the learning coefficient, it notifies the operator and prompts maintenance. Furthermore, regardless of whether maintenance is prompted, if maintenance is not completed, the learning control device 1a corrects the learning coefficient until the next event, such as maintenance, occurs. Furthermore, the learning control device 1a is equipped with a separate storage device T that records the date and time when a sudden change in the learning coefficient is detected. N function.

[0158] The learning coefficient re-updating unit 5a includes a mutation detection unit 50, a determination unit 52, a re-updating unit 54 and a notification unit 56, which detects sudden changes in the learning coefficient, determines the time of occurrence, and calculates the mutation component. After the time of occurrence of the mutation of the learning coefficient, if maintenance is not completed, the mutation component of the learning coefficient stored in the learning coefficient table 22 is corrected.

[0159] Notification unit 56 has the function of notifying the operator of a sudden change in the learning coefficient detected by sudden change detection unit 50. For example, if no sudden change in the learning coefficient is observed, notification unit 56 may output a message to a human-machine interface (not shown) indicating the need for maintenance such as equipment inspection or replacement. Furthermore, notification unit 56 may generate an alarm using an audible device or other method that is easily perceived by the operator.

[0160] That is, when the sudden change detection unit 50 identifies the occurrence time of the sudden change in the learning coefficient, the notification unit 56 notifies the operator of the necessity of maintenance such as inspection or replacement of the equipment as an event.

[0161] In this way, the learning control device la calculates the updated value of the learning coefficient relative to the next rolled material, detects the sudden change of the learning coefficient, and judges whether maintenance has been performed thereafter. If no maintenance has been performed, the instantaneous value of the learning coefficient is corrected based on the sudden change component of the learning coefficient, and the updated value of the learning coefficient is obtained as follows.

[0162] [Equation 15]

[0163]

[0164] Z (i，j)NEW :Updated value of the learning coefficient of the corresponding unit in the learning coefficient table

[0165] Z E_(i，j)USED : The previous value of the learning coefficient of the corresponding unit in the learning coefficient table

[0166] Z (i，j)CURRENT : instantaneous value of the learning coefficient

[0167] α: smoothing coefficient

[0168] i, j: the coordinates of the unit in the learning coefficient table

[0169] Learning coefficient mutation component

[0170] T N : The date and time of the rolling material when the sudden change of the learning coefficient was detected

[0171] T n′E : The date and time of the most recent final event

[0172] Specifically, after the mutation detection unit 50 determines the time when a mutation of the learning coefficient occurs, the re-updating unit 54 re-updates the learning coefficient in the learning coefficient table 22 based on the instantaneous value of the learning coefficient in rolling after the time when the mutation of the learning coefficient occurs, the mutation component of the learning coefficient, and the previous value of the learning coefficient of the unit based on the rolling conditions of the learning coefficient table 22, if there is no date and time record maintained as an event in the rolling information database 20.

[0173] Then, based on the updated value of the learning coefficient, the learning control device 1a updates the learning coefficient table 22. Thus, the learning control device 1a performs setting calculations using the same level of learning coefficients thereafter.

[0174] As described above, according to the present invention, even if an error factor due to sudden fluctuations occurs, subsequent learning can be corrected. For example, according to the present invention, even if an anomaly occurs due to mechanical factors such as poor calibration of equipment during maintenance such as regular equipment inspection or replacement, or a persistent measurement anomaly due to instrument abnormality, the learning coefficient can be corrected based on the learning coefficient stored before the anomaly occurred. Furthermore, by minimizing the impact of anomalies, the present invention can reduce the persistent decline in product accuracy and achieve stable rolling.

[0175] Learning control devices 1, 1a periodically analyze data from multiple rolled coils, detecting sudden changes in prediction errors (learning values) caused by mechanical factors or measurement anomalies, and determining the timing of these changes. Based on a learning table stored separately for each periodic event, such as repairs and roll replacements, learning control devices 1, 1a reset the learning table to the value stored immediately before the sudden change, compensating for the difference in learning values ​​caused by the sudden change, and re-updating the learned values ​​from the sudden change to the current rolled material.

[0176] Furthermore, part or all of the functions of the learning control devices 1 and 1 a may be implemented as hardware such as a PLD (Programmable Logic Device) or an FPGA (Field Programmable Gate Array), or as programs executed by a processor such as a CPU.

[0177] Description of Reference Numerals

[0178] 1. 1a Learning control device, 2 Storage unit, 3 Learning unit, 4 Setting calculation unit, 5. 5a Learning coefficient re-updating unit, 20 Rolling information database, 22 Learning coefficient table, 24 Learning coefficient table for each event, 30 Prediction value calculation unit, 32 Instantaneous value calculation unit, 34 Update unit, 40 Learning coefficient reading unit, 42 Setting calculation unit, 50 Sudden change detection unit, 52 Judgment unit, 54 Re-updating unit, 56 Notification unit.

Claims

1. A learning control device for a rolling process, characterized in that: The learning control device for the rolling process controls the rolling process by updating the learning coefficients of the formula model used for calculating the set values ​​for the rolling process through a learning coefficient table composed of a plurality of units divided according to rolling conditions, and storing the updated learning coefficients. The learning control device for the rolling process includes: a predicted value calculation unit that calculates a predicted value based on an actual value measured during a rolling process; an instantaneous value calculation unit that calculates an instantaneous value of a learning coefficient based on a difference between the predicted value calculated by the predicted value calculation unit and an actual value of the rolling process; an updating unit that calculates an updated value of the learning coefficient based on the instantaneous value of the learning coefficient calculated by the instantaneous value calculating unit and a previous value of the cell in the learning coefficient table that meets the rolling conditions, and updates the learning coefficient of the cell in the learning coefficient table that meets the rolling conditions; a rolling information database storing an instantaneous value of the learning coefficient calculated by the instantaneous value calculation unit, a previous value of the learning coefficient, an updated value of the learning coefficient, date and time information of determining a rolled material, rolling conditions, coordinates of cells of the learning coefficient table based on rolling conditions, actual values ​​in the rolling process, and a date and time history of events in the rolling process; a sudden change detection unit that determines a time when a sudden change in the learning coefficient occurs based on the instantaneous value of the learning coefficient stored in the rolling information database, and detects a deviation in the level of the instantaneous value of the learning coefficient before and after the time when the sudden change in the learning coefficient occurs as a learning coefficient sudden change component; as well as The re-updating unit re-updates the learning coefficient of the learning coefficient table by correcting the learning coefficient mutation component based on the instantaneous value of the learning coefficient stored in the rolling information database after the occurrence time of the mutation of the learning coefficient determined by the mutation detecting unit and the learning coefficient mutation component.

2. The learning control device for the rolling process according to claim 1, characterized in that: The method further includes a notification unit configured to notify the necessity of maintenance as an event when the sudden change detection unit identifies the occurrence time of the sudden change in the learning coefficient.

3. The learning control device for rolling process according to claim 2, characterized in that: After the mutation detection unit determines the occurrence time of the mutation of the learning coefficient, the re-updating unit re-updates the learning coefficient of the learning coefficient table based on the instantaneous value of the learning coefficient in rolling after the occurrence time of the mutation of the learning coefficient, the mutation component of the learning coefficient, and the previous value of the learning coefficient of the unit based on the rolling conditions in the learning coefficient table, if there is no date and time history maintained as an event in the rolling information database.

Images