JPH05119807A - Identifying method for control model in continuous rolling mill and control method for continuous rolling mill - Google Patents

Abstract

PURPOSE:To execute rolling with high precision even in the case a rolling phenomenon is non-liner by correcting a coupling coefficient of a hierarchical type neural network by learning, and identifying a control model. CONSTITUTION:A plate crown/shape control model in a k-th stand is constituted of plate width, rolling data in each stand extending from a first stand to a k-th stand, and a neural network of a three-layer structure for inputting actual result data in a k-th stand, and outputting bender force (pressure) in a k-th stand. Subsequently, with respect to this network, learning is executed by inputting plate width, and plate thickness in each stand, a rolling load, a gage meter constant, and an actual result plate crown quantity and an actual result plate shape in a k-th stand, and also, inputting an actual result value of bender force as a teacher signal, so that bender pressure in the case a satisfactory plate crown-shape is obtained in a final stand is outputted, by which this control model is identified.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for identifying a control model in a continuous rolling mill and a method for controlling the continuous rolling mill,
Particularly, even when the rolling phenomenon is non-linear and complicated, a control model identification method capable of easily identifying a control model for rolling, and a continuous rolling mill control method capable of easily and accurately controlling a continuous rolling mill Regarding

[0002]

2. Description of the Related Art Continuous rolling is a state in which materials are simultaneously rolled by two or more continuous rolling mills ("Theory and practice of sheet rolling", Japan Iron and Steel Institute Joint Research Group, Rolling Theory Section). Ed., 1984).

In this continuous rolling, each factor of rolling influences each other via tension between stands. In the case of strip rolling, generally 6 to 7 rolling mill stands are arranged in tandem for hot rolling and 5 to 6 rolling mill stands are arranged for cold rolling. The rolling factors such as the friction coefficient, roll gap, roll speed, motor characteristics, roll diameter, and material deformation characteristics affect each other.

Since such rolling factors influence each other over all stands, the rolling characteristics in continuous rolling should be examined by considering all rolling factors by one rather than examining the relation of each rolling factor individually. It is better to think comprehensively as one system. Therefore, specifically, it is common to obtain the characteristics of the entire rolling mill by solving the rolling factors of all the stands in a continuous rolling mill based on the rolling characteristics of the single rolling mill.

The control in the conventional continuous rolling mill is performed by using the result obtained by solving the group of equations (rolling model) obtained from the continuous rolling theory by the linear calculation method or the non-linear calculation method based on the above concept. Is common.

[0006]

However, depending on the item of control, the influence of the rolling factor is not sufficiently reflected, so that the precision of control is insufficient or there is no control at all.

[0007] For example, a rolling model used for strip crown / shape control in hot rolling is composed of models relating to roll deformation of rolling mill, thermal expansion of roll, wear of roll and deformation of rolled material. The model related to thermal expansion, the model related to roll wear, the model related to deformation of rolled material, etc. are not based on sufficient theoretical analysis, and the character as a statistical model obtained by statistically analyzing experimental data is strong.

The above experimental data can be obtained by directly conducting an experiment on a production line, but in reality, only a limited amount of data can be obtained at a limited time in consideration of the production plan and the like. Therefore, the accuracy of the model deteriorates with the aging of the characteristics of the rolling mill or the like, and for the rolled material having different rolling characteristics from the rolled material of the experiment when obtaining the experimental data, the experimental data is Since a statistical model obtained by statistical analysis may not provide sufficient accuracy, the conventional rolling model may not perform good strip crown / shape control.

In such a case, manual control is often performed by the intuition and experience of a skilled operator, but this manual control is sufficiently performed due to individual differences of the skilled operator and artificial errors. There is a problem that uniform and standardized controllability cannot be obtained, and the operation itself will not be realized unless excellent skilled operators are constantly trained and secured.

The present invention has been made to solve the above-mentioned conventional problems, and easily identifies a control model that enables highly accurate rolling even when the rolling phenomenon in a continuous rolling mill is nonlinear. It is a first object to provide a method of identifying a control model in a continuous rolling mill that can be performed.

The present invention also provides a control method for a continuous rolling mill, which uses the control model identified by the above identification method and can control the coating control amount such as the plate crown and shape in the continuous rolling with high accuracy. The second issue is to provide.

[0012]

As shown in FIG. 1 (in the case of k-th stand), the present invention provides a control model for describing a rolling phenomenon in a single stand included in a continuous rolling mill, a rolling factor and a rolling model. It is composed of a hierarchical neural network that inputs a target controlled variable and outputs a control operation amount for obtaining the target controlled variable under the rolling factor, and the target single stand Alternatively, the control model obtained by inputting the rolling factor between the stands, the actual value of the controlled amount in the target single stand, and the rolling factor between the single stand or the stands located upstream of the target single stand. The output of is in agreement with the control operation amount of the target single stand when a good controlled variable is obtained at the final stand when actually rolling under the same rolling factor. The coupling coefficient of the layer neural network was modified by learning, by identifying the control model is obtained by achieving the first object.

The present invention also provides a method for identifying a control model in the continuous rolling mill, wherein the rolling phenomenon is
It is a shape change, the rolling factor is the strip width, strip thickness, rolling load and gauge meter learning term, the controlled amount is the strip crown amount and strip shape, and the control operation amount is the bender pressure (force). In addition, the first object is achieved.

The present invention also uses the control model identified by the control model identification method in the continuous rolling mill,
The second problem is achieved by controlling the continuous rolling mill.

The present invention further uses a plate crown / shape control model identified by the control model identification method,
The second subject is similarly achieved by performing the plate crown / shape control.

[0016]

In the present invention, a control model describing a rolling phenomenon in an arbitrary single stand included in a continuous rolling mill is inputted with a rolling factor and a target controlled amount, and a target model is set under the rolling factor. It is composed of a hierarchical neural network that outputs the control operation amount for obtaining the controlled variable.

As a result of actually controlling the continuous rolling mill, if the actually controlled amount to be controlled is y ₀ at the final stand, the actually controlled amount to be controlled is y ₀ . The difference Δy between the target controlled variable and the actually obtained controlled variable is given by the following equation (1).

Δy = y−y ₀ (1)

Now, let λ be the allowable accuracy of the product quality, and learn the neural network by using each rolling factor, the actual value of each control manipulated variable and the actual value of each controlled variable when the following equation (2) is satisfied. I do.

│Δy │ <λ (2)

In the following, the learning principle of the neural network will be described by taking a case where the control model is configured by a hierarchical neural network having a three-layer structure with m inputs and n outputs and the number of intermediate layers is 1 as shown in FIG. explain.

Now, in the above neural network, the coupling coefficient from the i-th unit of the input layer to the j-th unit of the intermediate layer is V _ij , and from the i-th unit of the intermediate layer to the j-th unit of the output layer. The coupling coefficient of W _ij ,
The input / output function of the unit of the middle layer and the output layer is f (X + θ)
And

The θ included in this input function is different for each unit, and is corrected together with Vij and Wij by the following learning. Input u _i to the i-th unit of the input layer
The input from this unit to the j-th unit in the hidden layer is given by the following equation (3). Therefore, _assuming that the output of the j-th unit in the intermediate layer is h _j , this output can be expressed by the following equation (4).

X _j = ΣV _ij · u _i (3) h _j = f (X _j ) (4)

[0025] The input Z _j for the j-th unit of the output layer, since it is given by the following equation (5), the output of the unit When y _j, can be expressed by the following equation (6).

Z _j = ΣW _ij · h _i (5) y _j = f (Z _j ) (6)

Also, input to the i-th unit of the input layer
The desired output (teacher signal) of the j-th unit in the output layer when u _i (i = 1, ..., m) is input is d _j
(J = 1, ..., N), and the error evaluation function E is defined as the following expression (7).

E = Σ (y _j −d _i ) ² (7)

At this time, the correction amount ΔW of the coupling coefficient W _ij
ij is given by the following equation (8) to ε as positive constant, also the correction amount [Delta] V _ij of V _ij is given by the following equation (9).

[0030]

[Equation 1]

According to the learning method described above, based on the evaluation value of the error when the learning data is input to the control model, the coupling coefficient between the units of the neural network is corrected so as to reduce the error. Control model identification for a single stand can be performed.

In the above learning, as the input u _i to the input layer, various control results data obtained by a skilled operator when good control is performed, or good calculation is performed by the conventional rolling model. The control model that outputs the control manipulated variable y _i to obtain a good controlled variable can be identified by using various actual rolling data in the case of breakage.

Further, by performing the rolling control of the continuous rolling mill using the identified control model, the controlled variable can be controlled with high accuracy even if the rolling phenomenon is non-linear.

Next, the case of a control model (plate crown / shape control model) used for plate crown / shape control will be described.

For example, as shown in FIG. 3, the plate crown / shape control model in the k-th stand is defined as the plate width and the first model.
For each stand from stand to k-th stand,
Input the plate thickness, rolling load and gauge meter constant, the actual plate crown amount and actual plate shape at the k-th stand,
The k-th stand is composed of a three-layered neural network that outputs the bender pressure.

Then, the bender pressure when a good plate crown / shape is obtained at the final stand is output to the neural network forming the k-th stand control model according to the above-mentioned learning principle. , Strip width, strip thickness, rolling load and gauge meter constant at each stand from the 1st stand to the kth stand, and the kth stand.
The control model is identified by inputting the actual plate crown amount and the actual plate shape in the stand and by inputting the actual value of the bender pressure as a teacher signal for learning.

When the identified control model is applied to the continuous rolling mill, the target of the plate to be actually rolled is the plate width, plate thickness, rolling load, plate crown amount, plate shape and calculation. By inputting the obtained gauge meter constant, the accurate bender pressure of the k-th stand is output.Therefore, by controlling the rolling mill based on this output value, highly accurate rolling control of the strip crown and shape can be performed. It becomes possible to do it.

[0038]

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

In this embodiment, a plate crown / shape control model in a continuous rolling mill is identified, and the plate crown / shape is controlled using the control model.

FIG. 4 is an overall configuration diagram of a control system showing an embodiment in which the present invention is applied to the final seventh stand in a hot finish rolling mill (continuous rolling mill).

In FIG. 4, reference numerals 1 to 7 denote rolling mills of the first stand to the seventh stand, respectively. Above the backup rolls (indicated by great circles) constituting these stands, the rolling load applied to the rolling mill is shown. The load cells 11 to 17 for detecting are installed.

Further, the work rolls (shown by small circles) of each stand are provided with bender devices 21 to 27 for applying a bender force (bender pressure).
Bender force detecting device 3 for detecting the applied bender force from the hydraulic pressure for applying the bender force in
1 to 37 are installed, and a strip width gauge 4 is provided on the exit side of the finishing rolling mill.
1, profile meter 42, shape detector 43 and plate thickness meter 4
4 are arranged respectively.

The control system further includes a control result determination device 51, a parameter calculation device 52, a learning data storage device 53, a model identification device 54, a rolling condition calculation device 55, and a bender control device 56, which will be described in detail below. ing.

The control result judging device 51 only operates when the plate crown amount and plate shape measured by the profile meter 42 and the shape detector 43 are within the allowable range compared with the target plate crown amount and plate shape. Send the output signal.

The parameter calculator 52 has a function of calculating the plate thickness based on the law of constant mass flow and a function of calculating the plate thickness by the gauge meter formula, and is a coefficient for correcting the error of the gauge meter formula by learning. It has a function of calculating a certain gauge meter constant (gauge meter learning term) from the difference in plate thickness calculated by both of the above functions.

When the learning data storage device 53 receives the output signal of the control result determination device 51, the strip width meter 41, the profile meter 42, and the shape detector 43 on the exit side of the finishing rolling mill.
And the strip width, the strip crown amount, the strip shape, and the strip thickness measured by the strip thickness meter 44, and the rolling loads of the first to seventh stands 1 to 7 detected by the load cells 11 to 17, respectively.
Each bender force of the first to seventh stands 1 to 7 detected by the bender force detection devices 31 to 37, and the parameter calculation device 5
2, the output plate thicknesses of the first to sixth stands 1 to 6 calculated based on the constant law of mass flow, and the device 52.
It has a function of accumulating the gauge meter constant calculated by the above as learning data for making the neural network shown in FIG. 3 learn.

The model identification device 54 is composed of the above-mentioned neural network, and includes the learning data storage device 53.
It has a function of identifying a plate crown / shape control model expressing the rolling phenomenon of the seventh stand based on the present invention, using the learning data received from.

When performing rolling, the rolling condition calculation device 55 uses the plate crown / shape control model identified by the model identification device 54 to obtain the bender force setting value of the seventh stand, thereby performing the plate crown / shape control. It has a function of calculating rolling conditions input to the model, and the bender control device 56 has a function of controlling the bender device 27 based on the bender force setting value.

According to the present embodiment, the learning data storage device 5
In No. 3, the rolling factor, the controlled amount, and the control operation amount when a good strip crown amount and strip shape (within an allowable range with respect to the target value) are obtained by the seventh stand 7 are accumulated as learning data. Therefore, by using the learning data created by the storage device 53 and applying the above-described learning method, the model identification device 54 expresses the rolling phenomenon of the seventh stand, which has the same configuration as that shown in FIG. A plate crown / shape control model consisting of a hierarchical neural network can be identified.

Therefore, by inputting the rolling condition calculated by the rolling condition calculation device 55 into the control model, an appropriate bender force setting value under the rolling condition can be obtained. Therefore, the bender control based on the bender force setting value is obtained. By controlling the bender device 27 by the device 56, a good plate crown amount and plate shape can be obtained.

FIG. 5 is a bender obtained by the control model of the seventh stand identified based on the present embodiment with respect to the actual value of the bender force of the seventh stand when a good plate crown amount and plate shape are obtained. The relationship between the magnitude of the error in the force setting value and its frequency is shown.

Similarly, FIG. 6 shows the setting of the bending force of the stand obtained by the conventional rolling model with respect to the actual value of the bending force of the seventh stand when a good strip crown amount and strip shape are obtained. The relationship between the magnitude of the value error and its frequency is shown.

As is apparent from FIGS. 5 and 6, it is understood that the bender force set value according to the present invention has a very small error setting rate as compared with the bender force set value according to the conventional method. Therefore, according to this embodiment, the plate crown
Since it is possible to perform shape control with high accuracy, the automation ratio of rolling control can be significantly improved compared to the conventional method, and therefore the rolling control manually performed by a skilled operator can be significantly reduced. ..

Although the present invention has been specifically described above, the present invention is not limited to the above-mentioned embodiments, and various modifications can be made without departing from the scope of the invention.

For example, in the above-mentioned embodiment, the continuous rolling mill having 7 stands is shown, but the number of stands may be at least larger than 7 or vice versa.

In the above embodiment, the case where the control model is set only for the seventh (final) stand has been described, but the stand for setting the control model is not limited to the last stand.

Further, independent control models may be set for all or some of the first to seventh stands.

[0058]

As described above, according to the present invention,
A control model that describes the rolling phenomenon on a single stand is configured by a hierarchical neural network, and high-precision rolling is performed by learning the above neural network using various rolling performance data when good control is performed. It is possible to identify a control model that outputs a control operation amount that can be controlled.

Further, by using the identified control model, it is possible to obtain an excellent effect that automatic control of the strip crown, shape, etc. in continuous rolling can be performed easily and with high accuracy.

[Brief description of drawings]

FIG. 1 is a flowchart showing the gist of a method for identifying a control model expressing a rolling phenomenon of a k-th stand according to the present invention.

FIG. 2 is a diagram showing a neural network for explaining the principle of the learning method of the neural network.

FIG. 3 is a diagram showing a neural network expressing a rolling phenomenon of a k-th stand.

FIG. 4 is a schematic side view including a partial block diagram showing an overall configuration of a hot finish rolling mill applied to the strip crown / shape control method of the one embodiment.

FIG. 5 is a diagram showing the magnitude and frequency of erroneous setting of the vendor force setting value according to the present invention.

FIG. 6 is a diagram showing the magnitude and frequency of erroneous setting of the vendor force setting value by the conventional method.

[Explanation of symbols]

 S ... Rolled material, 1-7 ... 1st-7th stand, 11-17 ... Load cell, 21-27 ... Bender device, 31-37 ... Bender force detection device, 41 ... Strip width meter, 42 ... Profile meter, 43 ... Shape detector, 44 ... Plate thickness gauge, 51 ... Control result determination device, 52 ... Parameter calculation device, 53 ... Learning data storage device, 54 ... Model identification device, 55 ... Rolling condition calculation device, 56 ... Bender control device .

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification code Internal reference number FI technical display location G05B 17/02 7740-3H

Claims (4)

[Claims] 1. A control model describing a rolling phenomenon in a target single stand included in a continuous rolling mill, inputs a rolling factor and a target controlled amount, and sets a target controlled amount under the rolling factor. And a rolling factor between the stands, the actual value of the controlled variable at the target single stand, and the target. The output of the control model obtained by inputting the rolling factors between the single stands located upstream of the single stand or between the stands is good at the final stand when actually rolling under the same rolling factor. The coupling coefficient of the hierarchical neural network is corrected by learning so as to match the control operation amount of the target stand when the controlled variable is obtained. A method for identifying a control model in a continuous rolling mill, which is characterized by identifying a roll. 2. The rolling phenomenon according to claim 1, wherein the rolling phenomenon is strip crown / shape change, the rolling factor is strip width, strip thickness, rolling load and gauge meter learning term, and the controlled amount is strip crown amount and strip shape. A method for identifying a control model in a continuous rolling mill, wherein the control operation amount is a bender pressure. 3. A control method for a continuous rolling mill, comprising controlling the continuous rolling mill using the control model identified by the identification method according to claim 1. 4. A control method for a continuous rolling mill, characterized by performing strip crown / shape control using the control model identified by the identification method according to claim 2.