JP3467677B2 - Learning control method of rolling load in rolling mill - 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 load learning control method for a rolling mill, which is particularly suitable for setting up a rolling load on a cold rolling mill.

[0002]

2. Description of the Related Art Conventionally, in rolling control of a cold rolling mill, a rolling mathematical model is used to predict a rolling load for the next rolled material and to set the rolling load, and the setting accuracy at that time is improved. Therefore, it is essential to predict rolling load more accurately.

In general, in order to improve the prediction accuracy of the mathematical model (rolling load model) as described above, a method of performing learning for correcting the model using the latest actual value is adopted. As the most general learning method performed to improve the accuracy of such a rolling load model, as disclosed in, for example, Japanese Patent Laid-Open No. 50-108150, a rolling load actual value and a rolling actual result are used for calculation. The ratio (or difference) between the calculated rolling load model and the calculated value is taken as the following (1)
There is known a method of using a correction coefficient (learning coefficient) Zp for predicting and calculating the next rolling load by the predictive expression of the expression. As the learning coefficient Zp, the "difference" between the actual load and the model calculation value can be used, but the case where "ratio" is used will be mainly described below.

P = Zp.P0 (1) Here, P: rolling load predicted value Zp: Weighting ratio learning coefficient P0: Rolling load model calculation value

However, in the case of predicting and calculating the next rolling load by the formula (1), the load ratio learning coefficient Zp is usually the current value of the ratio as shown in the following formula (2). , The value smoothed by using the previous weighting ratio learning coefficient is used.

[0006]     Zp = (1-α) Zp-1 + α · Pact / Pcal (2) Where Zp-1: previous value of weight ratio learning coefficient Pact: Actual load value Pcal: Rolling load model calculation value α: constant of 0 <α <1

When the next rolling load is predicted and calculated using the above equation (1), the factors that cause the prediction error can be divided into two factors, an error peculiar to the rolled material and an error peculiar to the rolling mill. it can.

The former error peculiar to the rolled material appears in the prediction error of the deformation resistance of the rolled material and depends on the composition of the rolled material, the manufacturing conditions in the hot rolling process, etc. It is considered to have random errors (to some extent).

On the other hand, the latter error peculiar to the rolling mill appears in the error in predicting the friction coefficient and is considered to change with a certain tendency over time depending on the roll condition and the rolling oil condition.

[0010]

However, in the conventional learning method of the rolling load model, the error component, which is considered to be due to the above two factors, is corrected by one correction as shown in the equation (1). Since learning was performed only with the coefficient (learning coefficient) Zp, there were the following problems.

For example, even if the correction coefficient is divided into learning categories for each steel type of the rolled material, it is not possible to learn the error caused by the change over time of the rolling mill. Further, as a learning classification of the correction coefficient, it is possible to learn a decrease in roll surface roughness that occurs as rolling progresses, and a change over time on the rolling mill side that is considered to be caused by a change in the concentration of rolling oil used. Even if there are divisions, it is not possible to use the correction factors used up to that point for changes in the steel type of the rolled material.

The present invention has been made to solve the above-mentioned conventional problems. The rolling load to be set at the next rolling is
An object of the present invention is to provide a technique relating to learning control of rolling load, which enables highly accurate rolling control from the beginning by enabling accurate prediction calculation using rolling results.

[0013]

According to the present invention, the rolling load prediction value obtained by applying the learning coefficient to the rolling load model calculation value obtained by substituting the rolling condition for the next rolling condition into the rolling load model In the rolling load learning control method in a rolling mill that performs rolling control by setting, the rolling load actual value and the rolling load model calculation value obtained by substituting the rolling condition actual value into the rolling load model are determined. Learning coefficient
Separated into a component that learns the error peculiar to the rolled material and a component that learns the error due to aging of the rolling mill, and learn both these components individually, and reflect the learning result in the learning coefficient, next time When predicting the rolling load to be set during rolling, the first component that learns the error peculiar to rolled material
1 Learning coefficient: Learn the error due to Zpk and rolling mill aging
The second learning coefficient of the component to be learned: Zpm and Zpk = (1-αk) Zpk-1 + αk · Pact / (Zpm-1 · Pc, respectively.
al) Zpm = (1-αm) Zpm-1 + αm · Pact / (Zpk · Pca
L) , and the next rolling load P is P = Zpk · Zpm · P0 (Zpk-1: previous value of the first learning coefficient Zpm-1: previous value of the second learning coefficient Pact: actual load value Pcal: rolling load The above problem is solved by predicting with model calculation values αk, αm: 0 <αk, αm <1 constant P0: rolling load model calculation value) . The present invention also sets the next rolling condition in the rolling load model.
Apply learning coefficient to rolling load model calculation value obtained by substitution
The rolling load prediction value obtained by
Rolling load learning control method in rolling mill for rolling control
The actual rolling load value and the rolling load model
Rolling load model calculated value obtained by substituting actual value of rolling condition
The learning coefficient determined based on the
Learn the components to be learned and the errors due to changes over time in the rolling mill.
Component and separate each of these components individually.
The learning results are reflected separately in the learning coefficient.
In order to predict the rolling load set for the next rolling,
First learning coefficient of component for learning error peculiar to rolled material: Zpk
And the second component that learns the error due to the change over time of the rolling mill.
Learning coefficient: Zpm and Zpm = (1-αm) Zpm-1 + αm · Pact / (Zpk-1 · Pc, respectively.
al) Zpk = (1-αk) Zpk-1 + αk · Pact / (Zpm · Pca
L) , and the next rolling load P is P = Zpk · Zpm · P0 (Zpk-1: previous value of the first learning coefficient Zpm-1: previous value of the second learning coefficient Pact: actual load value Pcal: rolling load Similarly, the above problem was solved by predicting with model calculation values αk, αm: 0 <αk, αm <1 constant P0: rolling load model calculation value) .
It is a thing.

That is, in the present invention, the learning coefficient applied to the rolling load model calculation value is separated into a component for learning the error peculiar to the rolled material and a component for learning the error due to the change over time of the rolling mill. Since the learning results of both components are individually learned and reflected in the learning coefficient used for the next rolling as a multiplication value of both, it is possible to improve the accuracy of the rolling load prediction value. .

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below with reference to the drawings.

First, the principle of a rolling load learning control method according to an embodiment of the present invention will be described.

In the present embodiment, the error of the rolling load model is separated into a component peculiar to the rolled material (corresponding to deformation resistance) and a time-varying component on the rolling mill side (corresponding to friction coefficient), and the following method is used. Learn by.

In the present embodiment, since the error component is learned by dividing it into two as described above, as the learning coefficient (load ratio learning coefficient) of the rolling load model, the following first and first values corresponding to the respective components are used. Two coefficients of 2 are used.

[0019] Zpk: First learning coefficient (corresponding to load ratio corresponding to deformation resistance) Zpm: Second learning coefficient (corresponding to the load ratio corresponding to the friction coefficient)

Since the first learning coefficient Zpk has different deformation resistance depending on the steel type (component) of the rolled material, a different value is used for each steel type group that is appropriately determined in advance for the rolled material. Further, this Zpk uses a different value for each stand to absorb the model error at each stand.

Further, since the friction coefficient of the second learning coefficient Zpm is greatly influenced by the roll property and the rolling oil property, the rolling is classified into a thick material group, a thin material group and the like depending on the difference in the thickness of the rolled material. Different values are used for each pattern and for each stand, and learning is performed without depending on the steel type of the rolled material.

In the present embodiment, the rolling load P set at the time of the next rolling is obtained by using the first and second learning coefficients Zpk and Zpm
The rolling load model calculation value P0 is multiplied by the following formula (3) for prediction.

P = Zpk.Zpm.P0 (3)

Regarding the first and second learning coefficients Zpk and Zpm applied to this equation (3), the following (4)
The calculation is performed in the order of the formula and the formula (5), and then both are multiplied by the rolling load model calculation value to predict the next rolling load by the formula (3).

[0025]     Zpk = (1−αk) Zpk−1 + αk · Pact / (Zpm−1 · Pcal) (4)     Zpm = (1-αm) Zpm-1 + αm · Pact / (Zpk · Pcal) (5) Where Zpk-1: the previous value of the first learning coefficient Zpm-1: Previous value of the second learning coefficient αk, αm: constants of 0 <αk, αm <1

The concept of learning by the above equations (4) and (5) is that the first learning coefficient corresponding to the load error corresponding to the deformation resistance initially divided into each steel type group, assuming that the rolling change of the rolling mill is small. Zpk is learned by the above equation (4) using the previous value as the first and second learning coefficients.

On the other hand, the remaining error is considered to be a load error that appears due to the change with time of the rolling mill, and this is considered as the second learning coefficient Zpm.
For this purpose, the learning is performed in the rolling order irrespective of the steel type group by the above equation (5) including the current value calculated by the above equation (4) as the first learning coefficient.

The first learning coefficient Zpk and the second learning coefficient Z
The learning order of pm may be reversed as shown in the following expressions (6) and (7).

[0029]   Zpm = (1-αm) Zpm-1 + αm · Pact / (Zpk-1 · Pcal) (6)   Zpk = (1-αk) Zpk-1 + αk · Pact / (Zpm · Pcal) (7)

Which of the learning by the equations (4) and (5) or the learning method by the equations (6) and (7) is selected may be determined by the relationship between the smoothing coefficients αk and αm. .

According to the learning method of the present embodiment described in detail above, the load error component due to the rolled material and the load error component due to the aging of the rolling mill can be learned with different learning coefficients. Highly accurate rolling load prediction calculation became possible.

Next, an example which is a specific example of this embodiment will be described.

[0033]

EXAMPLE FIG. 1 is a flow chart showing the procedure of the prediction calculation of rolling load when the learning control method of this embodiment is applied to an actual machine. However, here, the above equation (6),
The learning method based on equation (7) is used.

First, in step 1, rolling results such as rolling load, entry side plate thickness, exit side plate thickness, entry side tension, exit side tension, etc. are loaded into a control device (not shown), and these step results are rolled in step 2. The rolling load model calculation value Pcal is calculated by substituting it in the load model. As the rolling load model used here, an arbitrary model formula such as the well-known Hill formula or the Brond & Ford formula can be used.

Then, in step 3, the second learning coefficient Zpm is learned from the equation (6), and this learning value is used to
The first learning coefficient Zpk is learned and determined by the equation (7), and in step 4, these learning values are written in the respective storage tables.

Next, the same rolling load model formula as that used in step 2 above is substituted with the rolling conditions similar to the rolling results shown in step 1 which are set at the time of the next rolling, and the rolling load model calculation value is obtained. The next rolling load was predicted by calculating P0 and substituting the calculated value P0 into the equation (3) in which the learned first and second learning coefficients are applied.

FIG. 2 shows a rolling load prediction value (calculation value) when the learning control method of the present embodiment described above is applied to an actual machine.
In addition to showing the transition of rolling load actual values over time, the learning classification also shows the predicted load value when learning is performed by the conventional method set for each steel type group.

As shown in FIG. 2, in the learning control method of this embodiment, even when the roll property immediately after roll replacement changes significantly with time, the change is separated from the learning coefficient for each steel type group of the conventional method. I tried to make them learn,
It can be seen that a good load prediction result is obtained.

As described in detail above, according to this embodiment,
Since the rolling load prediction accuracy in the setup of the cold rolling mill can be improved, the accuracy of the setting change for the next rolled material can be improved. As a result, it is possible to reduce the off-gauge length of the sheet thickness at the coil tip for any rolled material, and also the unstable rolling phenomenon (the seizure phenomenon between the roll and the rolled material) due to the setting error of the target rolling pass schedule. And slip) can be prevented.

Further, since the prediction accuracy of the rolling load to be set is improved and the prediction accuracy of the rolling power is also improved, there is an effect that it is possible to prevent the occurrence of the overload state of the motor due to the misprediction of the power. .

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

For example, in the above-described embodiment, the case where the weight "ratio" is used as the learning coefficient has been mainly described, but the weight "difference" may be used.

[0043]

As described above, according to the present invention,
Since the rolling load set at the time of the next rolling can be accurately predicted and calculated by using the actual rolling results, the rolling control can be performed with high accuracy from the beginning.

[Brief description of drawings]

FIG. 1 is a flowchart showing a learning procedure according to an embodiment of the present invention.

FIG. 2 is a diagram showing the effect of the invention.

─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-60-15010 (JP, A) JP-A-62-167009 (JP, A) JP-A-56-160815 (JP, A) (58) Field (Int.Cl. ⁷ , DB name) B21B 37/00-37/78

Claims (2)

(57) [Claims] 1. A rolling control method in which a rolling load prediction value obtained by applying a learning coefficient to a rolling load model calculation value obtained by substituting a rolling load model into a rolling load model is set at the time of the next rolling. In the learning control method of the rolling load in the rolling mill, the learning coefficient determined based on the rolling load actual value and the rolling load model calculated value obtained by substituting the rolling condition actual value into the rolling load model The error learning component of the rolling mill and the error learning component of the rolling mill over time are separated, and these two components are individually learned, and the learning results are reflected in the learning coefficient, and set at the next rolling. per to predict the rolling load that, first learning coefficient of the components to learn rolled material inherent errors: ZPK
And the second component that learns the error due to the change over time of the rolling mill.
Learning coefficient: Zpm and Zpk = (1-αk) Zpk-1 + αk · Pact / (Zpm-1 · Pc, respectively
al) Zpm = (1-αm) Zpm-1 + αm · Pact / (Zpk · Pca
L) , and the next rolling load P is P = Zpk · Zpm · P0 (Zpk-1: previous value of the first learning coefficient Zpm-1: previous value of the second learning coefficient Pact: actual load value Pcal: rolling load A method of learning control of rolling load in a rolling mill, characterized by predicting with model calculation values αk, αm: 0 <αk, αm <1, a constant P0: rolling load model calculation value) . 2. Substituting the next rolling condition into the rolling load model
Obtained by applying the learning coefficient to the calculated rolling load model value.
Rolling control by setting the estimated rolling load value for the next rolling
On the learning control method of rolling load in rolling mill
The rolling load actual value and the rolling condition actual in the rolling load model.
Based on the rolling load model values calculated by substituting the values
The learning coefficient to be determined is used to learn the error peculiar to rolled material.
Minute and the component that learns the error due to the aging of the rolling mill.
Separated and train both of these components individually, and the learning result
Rolling that is set in the next rolling by reflecting
In predicting the load, the first learning coefficient of the component for learning the error peculiar to the rolled material: Zpk
And the second component that learns the error due to the change over time of the rolling mill.
Learning coefficient: Zpm and Zpm = (1-αm) Zpm-1 + αm · Pact / (Zpk-1 · Pc, respectively.
al) Zpk = (1-αk) Zpk-1 + αk · Pact / (Zpm · Pca
L) , and the next rolling load P is P = Zpk · Zpm · P0 (Zpk-1: previous value of the first learning coefficient Zpm-1: previous value of the second learning coefficient Pact: actual load value Pcal: rolling load A method for learning control of rolling load in a rolling mill, characterized by predicting with model calculation values αk, αm: 0 <αk, αm <1 constant P0: rolling load model calculation value).