JP2677964B2 - Rolling mill shape control initial setting method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a rolling mill for rolling a rolled material such as a steel plate, in which the shape (flatness, steepness) of the rolled material is used.
The present invention relates to an initial setting method of a shape control operation end for controlling a shape.

[0002]

2. Description of the Related Art In rolling a steel sheet or the like, a rolling mill is provided with several shape operation ends (actuators) in order to obtain a desired plate shape (flatness, steepness), but generally, it is a target. It is very difficult to theoretically obtain the initial setting value of each operation end for obtaining the shape.

Therefore, conventionally, a method of indexing from a preset set value table is often used. This table system is a system in which an optimum operating end set value is determined by classifying rolling conditions (steel type, rolling reduction, rolling roll diameter, etc.) for rolled materials based on past rolling results.

On the other hand, Japanese Unexamined Patent Publication No. 4-167908 discloses a method of calculating an initial set value by a neural network using an influencing factor relating to the flatness (shape) of a rolled material as input data. In this method, for the actual shape value when rolling with a certain set value, the optimum set value calculation device calculates the optimum set value of each operating end, and the pair of the influencing factor and each operating end set value at this time is calculated. Accumulate and use for learning of neural network. The neural network has a hierarchical structure and learns by error backpropagation learning with an influencing factor as an input and each operating end set value as an output.

[0005]

However, in the table method, depending on whether the division is appropriate or not,
There is a problem that the accuracy is greatly affected. That is, if a rough division is made, it becomes difficult to obtain the optimum value, and if a fine division is made, the table size becomes large, and a great effort is required for its preparation and maintenance.

Further, the neural network of the type disclosed in Japanese Patent Laid-Open No. 4-167908 (hierarchical type, error backpropagation learning) has a problem of convergence of learning. This is because the learning is performed so that the output of the neural network and the teacher signal are the same, the difference (error) between the two is used as the evaluation function (loss function), and the convergence calculation is performed to minimize the evaluation function. Because of this, it takes a lot of time to learn,
Alternatively, there is a problem that the local minimum solution falls into and does not converge.

Further, as the data for learning, the rolling result data is not directly used but the value calculated by the optimum set value calculating means is used, but it is originally difficult to create the optimum set value calculating means. In addition to the above problem, the value calculated by the optimum set value calculation means naturally contains an error, and the neural network learns the teacher data including the error, so that the accuracy decreases.

Furthermore, in the neural network of the type used here, the internal structure that connects the relationship between the input and the output is made into a black box, and the so-called inverse problem of inversely calculating the input from the output cannot be solved. Therefore, for example, in a neural network that outputs the operating end set value using the influencing factors (rolling conditions) and the desired shape as input data, when the error in the output value with respect to certain input data becomes large, There is a problem that it is not possible to verify the shape by back-calculating the shape from the influencing factors and the setting value of the operating end.

The present invention has been made to solve the above-mentioned conventional problems, and it is possible to easily and optimally set the optimum initial setting value of the shape control operation end based on the rolling record without the need to create a table. An object of the present invention is to provide a shape control initial setting method for a rolling mill, which can be accurately determined and set.

[0010]

According to the present invention, a rolling condition for rolling material is provided in a shape control initial setting method for a rolling mill, in which an operating end for controlling the shape of the rolling material is set before rolling the rolling material. , The set of shape operation end and the shape index value of the rolled material are used as the shape control record data, and a plurality of typical shape control record data is extracted from the shape control record data group based on the past rolling record and the shape control standard is extracted. Hold as a data group,
Before rolling, the rolling condition of the rolled material and the target shape index value are given, the similarity between the rolling condition and the shape index value is calculated with the shape control standard data, and the shape control standard is determined according to the similarity. The object is achieved by synthesizing and calculating the shape operation end set value of the data.

The present invention also achieves the above-mentioned object by using, as the shape index value, the elongation distribution in the plate width direction measured for each channel width of a constant width.

The present invention also achieves the above-mentioned object in the same manner by using, as the shape index value, a value obtained by normalizing the elongation distribution in the plate width direction by the plate width, which is measured for each constant width channel width. It is a thing.

Further, according to the present invention, when the shape operation end used in the initial setting is limited to the movement symmetrical with respect to the plate width, by using the value of only the symmetric component from which the asymmetric component is removed as the shape index value, Similarly, the above-mentioned object is achieved.

The present invention also uses a simulation model that models physical phenomena of rolling as a group of shape control result data based on past rolling results, and pseudo-calculates the shape data from the rolling conditions and the shape control initial setting amount. The above-mentioned object was similarly achieved by adding the above data.

The present invention further provides a rolling mill shape control initial setting method for setting the operating end for controlling the shape of the rolled material before rolling the rolled material, in which the rolling condition of the rolled material and the setting of the shape operating end are set. A set of the value and the shape index value of the rolled material is used as the shape control result data, and a plurality of typical shape control result data is extracted from the shape control result data group based on the past rolling results and retained as the shape control standard data group, Before rolling, the rolling condition of the rolled material and the target shape index value are given, the similarity between the rolling condition and the shape index value and the shape control standard data is obtained, and the shape control standard data is calculated according to the similarity. The shape operation end set value is calculated by combining, and the similarity between the rolling condition and the shape operation end set value is obtained with the shape control standard data, and the shape index of the shape control standard data is calculated according to the similarity. Combining values When the difference between the index value of the actual shape measured by rolling out and actual rolling exceeds a certain value, a plurality of typical shapes are added from the shape control actual data group with new rolling results. The above-mentioned object is similarly achieved by extracting the control performance data and holding it as a shape control standard data group.

[0016]

In order to control the shape of the rolled material by the shape operation ends of the rolling mill, (1) feedback control is performed to operate these operation ends based on the actual shape values detected during rolling. (2) Before starting the rolling, the optimum setting of the operating end for the rolling conditions is initialized. However, the present invention is based on the above method (2).

That is, according to the present invention, a set of rolling conditions of rolled material, set values of shape operation ends, and shape index values of rolled material is used as shape control actual data, and the shape control actual data group based on past rolling actual results is used. A plurality of typical shape control record data is extracted and stored as a shape control standard data group, and the rolling condition of the rolled material and the target shape index value are given before rolling, and the rolling condition and the shape index value are given. On the other hand, the similarity with the shape control standard data is obtained, and the shape control end set value of the shape control standard data is combined and calculated according to the similarity. It is possible to determine and set an optimum initial setting value based on the rolling record without having to create a table for

Further, in the learning of the actual data, since the convergence calculation is not performed so as to minimize the error evaluation function, the learning does not take a very long time, or the local minimum solution does not occur and the convergence does not occur. , Easy to learn.

Further, since it is not necessary to prepare an optimum set value calculating means as data for learning, and the rolling result data is directly used, the optimum set value calculating means is generated in the teaching data for learning of the neural network. It is possible to accurately calculate the optimum setting value without including an error.

Further, when the output value for a certain input data becomes unsuitable, the shape can be back-calculated from the influencing factor and the operating end set value for verification, so that the optimum set value can be obtained with high accuracy. Can be calculated.

The same result can be obtained when the elongation index distribution in the plate width direction measured for each constant width channel width is used as the shape index value. When the width is standardized, the shape index value that does not depend on the number of channels can be set.

When the shape operation end used in the initial setting is limited to the movement symmetrical with respect to the plate width, the same result can be obtained by using the value of only the symmetric component from which the asymmetric component is removed as the shape index value. Obtainable.

As a shape control result data group based on past rolling results, data obtained by pseudo-calculating the shape data from the rolling conditions and the shape control initial setting amount is used by using a simulation model that models a physical phenomenon of rolling. In the case of adding, the shape control initial setting can be accurately performed even if the actual data cannot be collected sufficiently.

Further, a set of rolling conditions of the rolled material, set values of the shape operation end, and shape index values of the rolled material is used as shape control record data, and a plurality of typical shapes are selected from the shape control record data group based on past rolling records. Control result data is extracted and retained as a shape control standard data group, and before rolling, the rolling condition of the rolled material and the target shape index value are given, and the shape control standard data is applied to the rolling condition and the shape index value. Of the shape control end data of the shape control standard data is calculated according to the similarity, and the similarity between the shape control standard data and the rolling condition and the shape operation end set value is calculated. Seeking
When the shape index value of the shape control standard data is synthesized and calculated according to the degree of similarity, and the difference between the index value of the actual shape measured by actual rolling is greater than a certain value, When extracting a plurality of typical shape control result data from the shape control result data group with a new rolling result and holding it as the shape control standard data group, the predicted shape obtained by "back calculation" and the actual rolling The accuracy can be maintained by re-learning when there is a large error in the actual shape data measured at the time.

[0025]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 is a side view showing the outline of the structure of a 12-high cluster type rolling mill in one embodiment of the present invention.

In FIG. 1, 12-stage rolls are a work roll 10 consisting of one set of upper and lower, an intermediate roll 12 consisting of two sets of the upper and lower, and a small backup roll 1 consisting of one set of upper and lower.
4, upper two pairs of upper backup rolls 16 and lower two pairs of lower backup rolls 18. The rolled material 20 is supplied from the pay-off roll 7, the right tension reel 24, or the left tension reel 26, rolled, and wound on the tension reel on the opposite side.
The shape detector 28 detects the shape of the rolled material in the width direction.

FIG. 2 is a front view showing a schematic configuration of the shape control operation end and the rolling mechanism. The rolled material 20 is rolled from above and below by the work roll 10 and thinly stretched. 12
Is an intermediate roll, 16 is an upper backup roll, and 18 is a lower backup roll. Upper backup roll 16
Is divided into seven, and the roll crown can be adjusted by extruding the center 16A, quarter-in 16B, quarter-out 16C, and edge 16D from the center while maintaining symmetry. Lower backup roll 1
8 is divided into six, and the roll crown can be adjusted by extruding the center 18A, the quarter 18B, and the edge 18C from the center while maintaining symmetry. As the shape operation end, the roll crown adjustment of the upper and lower backup rolls and the intermediate roll bender 30 are provided.

FIG. 3 is a block diagram showing an example of the data flow in the shape control initial setting process. The output of the shape detector 28 is subjected to data processing by the shape data processing mechanism 32 to become shape data (shape index value), which is stored in the shape data storage mechanism 34 at any time. Here, it is assumed that an elongation distribution in the plate width direction (divided into M channels (1 <M) in the plate width direction) is used as the shape data.

Similarly, the rolling condition data is stored in the rolling condition data storage mechanism 36, and the shape operation end data is stored in the shape data operation end storage mechanism 38. Here, in the present embodiment, the rolling conditions are the plate width, the predicted rolling load, and the work roll diameter, but in addition to this, steel types, pre-processes, product target thickness, inlet side temperature, outlet side target temperature, etc. are added as rolling conditions. Of course it is possible. Further, in the present embodiment, the upper backup roll edge position, the quarter out position, the upper backup roll quarter in position, the lower backup roll edge position, the lower backup roll quarter position, and the intermediate roll bender value are used as the shape operation ends. . However, the position of each divided backup roll is based on the center.

In the shape control record data accumulating mechanism 40, the rolling condition data accumulating mechanism 36, the shape operating end data accumulating mechanism 38, the rolling condition data accumulated in the shape data accumulating mechanism 34, the shape operating end data, and the shape at that time. The data is combined and stored as shape control record data. The typical data extraction mechanism 42 extracts typical data (prototype) representative of the distribution of these data based on the shape control record data. The obtained typical data is stored in the typical data group storage mechanism 44. From the accumulation of each data to the extraction and accumulation of typical data will be referred to as "learning". This learning may be executed at any time if a sufficient number of data are collected.

The rolling condition input mechanism 46 and the target shape input mechanism 48 respectively input rolling condition data and target shape data (desired shape) to the initial set value calculation mechanism 50 for the next rolled material. The initial set value calculation mechanism 50 obtains the degree of similarity of all typical data in the typical data group accumulating mechanism 44 with respect to the rolling condition data and the target shape data, and the shape control operation end set value of the typical data according to the degree of similarity. Are calculated to calculate initial setting values for the input rolling condition data and target shape data, that is, for the next rolled material. The initial set value output mechanism 52 outputs the calculated shape control operation end initial set value. The mechanism of this part will be collectively referred to as "calculation".

After that, the shape control operation end is initialized based on the value output by the initial set value output mechanism 52, and the rolling of the next rolled material is started.

FIG. 4 is a block diagram showing the flow of data when the predicted shape value is obtained from the rolling condition data and the shape control operation end set value. The shape data back calculation mechanism 54
With respect to the rolling condition data and the shape control operation end set value, the similarity regarding the rolling condition data and the shape control operation end set value of the typical data in the typical data group accumulating mechanism 44 created by “learning” is obtained. Shape data is calculated according to the similarity. This shape data corresponds to a predicted shape that will be obtained when rolling is performed with the rolling condition data and the shape control operation end set value.

The "learning" and "calculation" will be described in detail below.

When it is assumed that there are J shape control record data in total, the j-th shape control record data Dj is given by
Define with an expression.

Dj = {Smj, A1j, A2j, A3j, B1j, B2j, B3j, B4j, B5j, B6j} (1)

However, the meanings of the variables are as follows. (Shape data) Smj: Elongation rate of m channel in the strip width direction (1 <m ≤ M) M: Number of effective channels (rolling condition data) A1j: Strip width A2j: Predicted rolling load A3j: Work roll diameter (Shape operation end) Setting value) B1j: Upper backup roll / edge position B2j: Upper backup roll / quarter out position B3j: Upper backup roll / quota position B4j: Lower backup roll / edge position B5j: Lower backup roll / quota position B6j: Intermediate roll vendor value

In the typical data extracting mechanism 42, typical data (prototype) representative of the distribution of this data is based on the shape control record data {Dj | 1 <j≤J}.
Is extracted. One example of this "learning" can be represented by a network structure as shown in FIG. That is, D
Each element of j {Smj, A1j, A2j, A3j, B1j,
There is an input unit to which B2j, B3j, B4j, B5j, B6j} is input, and a prototype unit corresponding to typical data, and each prototype unit is weighted with each input unit. A prototype unit Pk and each input unit {Sm, A1, A2,
The coupling coefficient (weight) between A3, B1, B2, B3, B4, B5, B6} is {PSmk, PA1k, PA2k, PA3.
k, PB1k, PB2k, PB3k, PB4k, PB5
k, PB6k}. This coupling coefficient corresponds to each attribute value of typical data. Therefore, extraction of typical data can be interpreted as creating this prototype unit. The input unit appropriately performs conversion such as standardization on the input data.

The prototype unit is created by the following procedure.

(1) An input unit suitable for the input dimension is placed. In this example, the number of input units is (9 + M) from the rolling condition 3, the shape data M, and the shape control operation end 6. Initially, no prototype unit is placed. Also, the coefficient TH is determined.

(2) The first shape control result data D1 is used as the coupling coefficient P1 of the first prototype unit.
Also, the counter C1 of the prototype unit P1 is set to 1. P1 = D1 That is, {PSm 1, PA11, PA21, PA31, P
B11, PB21, PB31, PB41, PB51, P
B61} = {Sm 1, A11, A21, A31, B1
1, B21, B31, B41, B51, B61} C1 = 1

(3) Regarding the second shape control result data D2, the distance d (D2, D2 from the prototype unit P1)
P1) is calculated. Here, this distance is defined by the following equation (2).

[0044]

(Equation 1)

When (3-a) d (D2, P1)> TH, the second shape control result data D2 is set as the coupling coefficient P2 of the second prototype unit. Also, the counter C2 of the prototype unit P2 is set to 1. When P2 = D2 C2 = 1 (3-b) d (D2, P1) ≤TH Correct the coupling coefficient of the first prototype unit P1.

However, here, the modified prototype P1
new is expressed as P1new = {C1 · P1 + D2) / (C1 + 1). Also, the counter C1 of the prototype unit P1 is set to 1
Add. C1new = C1 + 1

(4) Similarly, with respect to the j-th shape control result data Dj, the distances to all the prototype units created so far are found, and the prototype unit Pk which is the minimum distance is selected.

When (4-a) d (Dj, Pk)> TH, the coupling coefficient _{Pk + 1} of the (k + 1) th prototype unit of the jth shape control result data Dj is set. Also, the counter C _{k + 1} of the prototype unit P _{k + 1} is set to 1. When P _{k + 1} = Dj C _{k + 1} = 1 (4-b) d (Dj, Pk) ≦ TH The coupling coefficient of the prototype unit Pk is modified. However, the modified prototype Pknew is set as Pknew = {Ck.multidot.Pk + Dj) / (Ck _{+ 1} ).
Also, 1 is added to the counter Ck of the prototype unit Pk. Cknew = Ck + 1

(5) All the shape control result data {Dj
By performing on | 1 <j ≦ J}, K prototype units are generated.

The coupling coefficient of this prototype unit is
Since it corresponds to the attribute value of typical data, K typical data has been extracted. At this time, if the k-th data of the K typical data is Pk, Pk is expressed by the following equation (3).

Pk = {PSmk, PA1k, PA2k, PA3k, PB1k, PB2k, PB3k, PB4k, PB5k, PB6k} (3)

However, the meaning of each variable is as follows. (Shape data) PSmk: Elongation rate of m channel in strip width direction (rolling condition data) PA1k: Strip width PA2k: Predicted rolling load PA3k: Work roll diameter (shape operation end set value) PB1k: Upper backup roll / edge position PB2k : Upper backup roll / quarter out position PB3k: Upper backup roll / quota position PB4k: Lower backup roll / edge position PB5k: Lower backup roll / quarter position PB6k: Intermediate roll vendor value

Where PSk = {PSmk} PAk = {PA1k, PA2k, PA3k} PBk = {PB1k, PB2k, PB3k, PB4k,
PB5k, PB6k}.

Next, "calculation" will be described with reference to a network diagram showing an example of calculation in FIG. In the “calculation”, the degree of similarity between the rolling condition and the target shape of the next rolled material and the typical data is obtained, and the shape operation end set amount corresponding to the degree of similarity is calculated. Therefore, in FIG. 6, the rolling condition data and the target shape data are input, and via the prototype unit,
The shape control operating end set value is output.

The rolling conditions and target shape of the next rolled material are rolling conditions: XA = {XA1, XA2, XA3}, where XA1: strip width, XA2: predicted rolling load, XA3: work roll diameter, and target shape. : XS = {XSma}, where XSm is the elongation of the m channel in the plate width direction, and the desired shape operation end setting amount is given by the following equation (4).

Shape operation end set amount: XB = {XB1k, XB2k, XB3k, XB4k, XB5k, XB6k} (4) XB1: Upper backup roll / edge position XB2: Upper backup roll / quarter out position XB3: Upper backup roll -Quarter-in position XB4: Lower backup roll-Edge position XB5: Lower backup roll-Quarter position XB6: Intermediate roll vendor value

First, the rolling condition XA of the next rolled material, the target shape XS, and the similarity between the rolling condition PAk and the shape PSk of all prototype units (typical data) are represented by the following formula (5): distance d 1 It is defined by (X, Pk).

[0058]

(Equation 2)

This distance d 1 (X, Pk) becomes an input to the prototype unit Pk. The output PY1k of the prototype unit Pk at this time is defined by the following expression (6).

PY1k = f (d1 (X, Pk)) (6) However, f (•) is an input / output function of the prototype unit, and becomes larger as the distance is smaller (the similarity is larger). In addition, for example, a Gaussian function such as f (x) = exp (−x ² / 2σ ² ) (where σ: constant) is used.

The shape operation end set amount XB is calculated according to the degree of similarity with the typical pattern. That is, the combination of the coupling coefficient PBk of the prototype unit according to the output PY1k of the prototype unit is the shape operation end set amount XB.
Output. Therefore, the following expression (7) is established.

[0062]

(Equation 3)

However, XB = {XB1, XB2, XB3, XB4, XB5, X
B6} PBk = {PB1k, PB2k, PB3k, PB4k,
PB5k, PB6k} In this way, the shape operation end optimum setting amount can be calculated. However, in the case of "learning", if conversion such as normalization is performed on the input data, the output value is inversely transformed.

Next, the "back calculation" for obtaining the predicted shape from the rolling conditions and the shape operation end set amount will be described with reference to the network diagram showing the back calculation in FIG. In the "back calculation", the degree of similarity with the typical data regarding the rolling condition of the next rolled material and the shape operation end setting amount is obtained, and the shape value according to the degree of similarity is calculated.
Therefore, the rolling condition data and the shape control operation end data are input to the network of FIG. 7, and the shape data is output via the prototype unit.

The rolling conditions of the next rolled material, the shape control operation end, and the predicted shape to be obtained are defined in the following (8), (9), and
It is determined by the equation (10).

Rolling conditions: XA = {XA1, XA2, XA3} (8) where XA1: strip width XA2: predicted rolling load XA3: work roll diameter

Shape operation end set amount: XB = {XB1k, XB2k, XB3k, XB4k, XB5k, XB6k} (9) However, XB1: upper backup roll edge position XB2: upper backup roll quarter out position XB3: upper Backup roll / Quarter-in position XB4: Lower backup roll / edge position XB5: Lower backup roll / quarter position XB6: Intermediate roll vendor value

Shape: XS = {XSm} (10) where XSm: elongation rate of m channel in the plate width direction

For the rolling condition XA, the shape control operation end XB, and all prototype units (typical data), the similarity between the rolling condition PAk and the shape control operation end XBk is given by (1)
It is defined by the distance d 2 (X, Pk) represented by the equation 1).

D 1 (X, Pk) = ‖XA-PAk‖ + ‖XB-PBk‖ + {(XA1-PA1k) ² + (XA2-PA2k) ² + (XA3-PA3k) ² + (XB1-PB1k) ^{2 + (XB2-PB2k) 2} + (XB3-PB3k) 2 + (XB4-PB4k) 2 + (XB5-PB5k) 2 + (XB6-PB6k) 2} 1/2 ... (11)

This distance d 2 (X, Pk) becomes an input to the prototype unit Pk. The output PY2k of the prototype unit Pk at this time is determined by the following equation (12).

PY2k = f (d2 (X, Pk)) (12) However, f (•) is an input / output function of the prototype unit, and becomes larger as the distance is smaller (the similarity is larger). In addition, for example, a Gaussian function such as f (x) = exp (−x ² / 2σ ² ) (where σ: constant) is used.

The shape XS is calculated according to the degree of similarity with the typical pattern. That is, the output PY of the prototype unit
The combination of the coupling coefficient PSk of the prototype unit according to 2k is the output of the shape operation end set amount XS.
Therefore, the following expression (13) is established.

XS = Σ (PSk · PY2k) / ΣPY2k (13) where XS = {XSm} XSm: elongation rate of m channel in the plate width direction PSk = {PSmk} PSmk: plate width of the kth typical data Direction of m channel.

In this way, the predicted shape can be calculated backward. However, in the case of “learning”, if conversion such as normalization is performed on the input data, the reverse conversion is performed on the output value.

Although the Euclidean distance is used as the degree of similarity in the above embodiments, the present invention is not limited to this.

Further, in the above-mentioned embodiment, the elongation distribution in the plate width direction is used as the shape data (shape index value).
It is also possible to use the one in which the number of channels is made constant by standardizing this with the plate width.

When the shape operation end used in the initial setting is limited to the movement symmetrical with respect to the plate width, the shape data is
It is also possible to remove the asymmetric component and use only the symmetric component. As a method of removing the asymmetric component, for example, there is a method of approximating the shape data by an m-order function and removing the odd-order (first-order, third-order, etc.) terms that are the asymmetric components.

Further, as the data used for "learning", not only the actual data but also the physical phenomenon of rolling can be modeled and the data obtained by simulating the shape data from the rolling condition and the shape control set amount can be added. Is. As a result, the present invention can be used with high accuracy even when the actual data cannot be collected sufficiently.

Further, the accuracy can be maintained by checking the error between the predicted shape obtained by "back calculation" and the actual shape data measured when actually rolling and re-learning when the error is large. Becomes

[0081]

As described above, according to the present invention,
It is not necessary to create a table for indexing the setting value of the shape control operation end, and it is possible to determine and set the optimum initial setting value based on the rolling record.

Further, in the learning of the actual data, the convergence calculation that minimizes the error evaluation function is not performed.
It is easy to learn without taking a very long time to learn or without falling into a local minimum solution and not converging.

Further, as the data for learning, it is not necessary to prepare the optimum set value calculating means, and the rolling result data is directly used, and the error generated by the optimum set value calculating means is used as the teaching data for learning of the neural network. Since, is not included, the optimum setting value can be calculated accurately.

Further, when the output value for a certain input data becomes unsuitable, the shape can be back-calculated from the influencing factor and the operation end set value for verification.
There is also an effect that the optimum set value can be calculated with high accuracy.

[Brief description of the drawings]

FIG. 1 is a side view showing an outline of a configuration of a rolling mill according to an embodiment of the present invention.

FIG. 2 is a front view showing a schematic configuration of a shape control operation end and a rolling mechanism in one embodiment of the present invention.

FIG. 3 is a block diagram showing a data flow in a shape control initial setting process according to an embodiment of the present invention.

FIG. 4 is a block diagram showing a data flow when a predicted shape value is obtained from rolling condition data and a shape control operation end set value in one embodiment of the present invention.

FIG. 5 is a network diagram showing an example of “learning” in one embodiment of the present invention.

FIG. 6 is a network diagram showing an example of “calculation” in one embodiment of the present invention.

FIG. 7 is a network diagram showing an example of “back calculation” in one embodiment of the present invention.

[Explanation of symbols]

 10 ... Work roll 12 ... Intermediate roll 14 ... Small backup roll 16 ... Upper backup roll 18 ... Lower backup roll 20 ... Rolled material 22 ... Payoff roll 24 ... Right tension reel 26 ... Left tension reel 28 ... Shape detector 30 ... Intermediate roll Vendor 32 ... Shape data processing mechanism 34 ... Shape data storage mechanism 36 ... Rolling condition data storage mechanism 38 ... Shape operation end data storage mechanism 40 ... Shape control record data storage mechanism 42 ... Typical data extraction mechanism 44 ... Typical data group storage mechanism 46 ... Rolling condition input mechanism 48 ... Target shape input mechanism 50 ... Initial setting value calculation mechanism 52 ... Initial setting value output mechanism 54 ... Shape data back calculation mechanism

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Masafumi Hoshino, 1st Kawasaki-cho, Chuo-ku, Chiba-shi, Chiba Kawasaki Steel Co., Ltd. Chiba Works (72) Riki Ozaka 1st Kawasaki-cho, Chuo-ku, Chiba, Chiba Prefecture Kawasaki Chiba Steel Works, Ltd.

Claims (6)

(57) [Claims] 1. A rolling mill shape control initial setting method in which an operating end for controlling the shape of a rolled material is set before rolling the rolled material. Using a set of shape index values of rolled material as shape control performance data, multiple typical shape control performance data are extracted from the shape control performance data group based on past rolling performance and stored as a shape control standard data group, before rolling. The rolling condition of the rolled material and the target shape index value are given to, the degree of similarity between the rolling condition and the shape index value is compared with the shape control standard data, and the shape control standard data is calculated according to the similarity. A shape control initial setting method for a rolling mill, characterized by combining and calculating shape operation end set values. 2. The rolling mill shape control initial setting method according to claim 1, wherein the shape index value is an elongation distribution in the strip width direction measured for each channel width of a constant width. 3. The rolling mill according to claim 1, wherein the shape index value is a value obtained by normalizing the elongation distribution in the strip width direction, which is measured for each channel width of a constant width, by the strip width. Shape control initial setting method. 4. When the shape operation end used in the initial setting is limited to a symmetric movement with respect to the plate width in claim 1, as the shape index value, a value of only a symmetric component from which an asymmetric component is removed is used. The initial setting method for the shape control of the rolling mill. 5. The shape data according to claim 1, wherein, as a shape control performance data group based on past rolling performance, a simulation model that models a physical phenomenon of rolling is used, and the shape data is simulated from the rolling condition and the shape control initial setting amount. A method for initializing the shape control of a rolling mill, which is characterized by adding the calculated data to. 6. A rolling mill shape control initial setting method for setting an operating end for controlling the shape of a rolled material before rolling the rolled material, comprising: rolling conditions of the rolled material, set values of the shaped operating end, and rolling. Using the set of shape index values of the material as shape control actual data, multiple typical shape control actual data are extracted from the shape control actual data group based on the past rolling results and stored as the shape control standard data group, before rolling. Given the rolling condition of the rolled material and the target shape index value, the degree of similarity between the rolling condition and the shape index value with the shape control standard data is obtained, and the shape manipulation of the shape control standard data is performed according to the similarity. The end setting values are combined and calculated, and the degree of similarity between the rolling condition and the shape operation end setting value is calculated with the shape control standard data, and the shape index value of the shape control standard data is combined according to the similarity. Calculated and calculated When the difference between the actual shape measured by rolling and the index value of the shape becomes larger than a certain value, a plurality of typical shape control results from the shape control result data group with new rolling results added A method for initializing the shape control of a rolling mill, which comprises extracting the data and holding it as a shape control standard data group.