JPH0661566B2 - Rolling mill setup device - Google Patents

Description

TECHNICAL FIELD The present invention relates to a set-up device for a cold or hot rolling mill, and in particular, has a function of correcting a set-up model using actual data during rolling. The present invention relates to a set-up device.

[Conventional technology]

One of the roles of a rolling mill control computer is to determine a set-up value (set value) of a controlled variable for each rolling coil based on rolling conditions and the like. This set-up value is usually calculated by a set-up model which is a mathematical model adapted online.

However, since the characteristics of the rolling mill slightly differ depending on each plant or each plant, it is necessary to make adjustments and corrections to adapt the set-up model to the characteristic peculiar to the rolling mill. This is because the rolling oil concentration, roll surface roughness, motor torque cross of each stand, etc. are different for each plant, and it is necessary to adjust the constants related to the model in accordance with this.

On the other hand, conventionally, the set-up model is corrected by online learning control, and the set-up value is determined by reflecting the correction result. Therefore, the advanced rate, the friction coefficient, the deformation resistance, the rolling load, the rolling position, and the like used in the set-up model are adjusted by learning control in accordance with the actual operation.

However, since the set-up model is a theoretical formula and includes an approximate formula suitable for online, the so-called conventional learning control may not be able to absorb the model error.

Therefore, conventionally, in order to absorb such a model error, a set-up value such as sheet thickness, rolling speed, rolling load, or the like,
The actual values of the control variables for determining it and the model calculated values corresponding to them are plotted, their correlation functions are analyzed, and the set-up model formula is corrected based on this result, and this is put online. When applied and compared again with the actual value and the calculated value, adjustment was made by a trial and error method.

[Problems to be solved by the invention]

However, according to the above-described trial and error method, a great deal of time is spent on adjustment, and the adjustment work often depends on the experience and intuition of the adjuster. For this reason, it may be difficult to complete the adjustment within a certain limited period of time, or erroneous adjustment may not always be suitable for the actual rolling mill characteristics, making it difficult to perform stable adjustment. There is a problem that is.

An object of the present invention is to solve the above-mentioned conventional problems, in other words, to provide a set-up device for a rolling mill capable of automatically adjusting a model error of the set-up model.

[Means for solving problems]

The present invention, in order to achieve the above object, based on input data including a given control variable, a set-up calculation means for calculating a set-up value of a rolling control amount by a preset set-up model and outputting the set-up value to a rolling control device. The actual calculated value of the set-up value is obtained by the set-up model using the actual value corresponding to the control variable output from the rolling mill, and the actual calculated value and the actual value output from the rolling mill corresponding to the actual calculated value. Error calculating means for obtaining the error of the error, a correction function represented by the weighting coefficient indicating the degree of influence of the control variable on the error, and a plurality of the error calculating means obtained by the error calculating means. From the error data of and the actual value of the control variable corresponding to this,
Regression analysis means for obtaining the weighting coefficient by regression analysis and specifying a correction function, and the set-up model is corrected by the correction function specified by the regression analysis means, and the corrected set-up model is used to correct the control variable. Obtaining a calculated result value for test using the actual value, a simulation error calculating means for obtaining an error between the actual calculated value for test and the actual value output from the corresponding rolling mill,
Test accuracy evaluation means for determining whether or not the error calculated by the simulation error calculation means is within a predetermined range, and outputting a command for correcting the set-up model by the specified correction function when the error is within the predetermined range. And a set-up device for a rolling mill.

[Action]

With such a configuration, the regression analysis means analyzes the cause of the error based on the error obtained by the error calculation means. By this analysis, it is possible to know the degree to which the control variable relating to the set-up model that determines the set-up value affects the model error, and a correction function that can adjust the equation of the set-up model corresponding to each control variable is obtained.

Then, the simulation using the online data is executed by the set-up model corrected by the correction function obtained by the simulation error calculating means, and the model error is obtained based on the simulation. Further, the accuracy of the above correction is verified by the test accuracy evaluation means, and the result of this is that the set-up model is corrected.

As described above, according to the present invention, since the model error of the set-up model is automatically corrected by the correction function based on the control variable, the adjustment period of the set-up model is shortened and corrected by this. Since the accuracy of the set-up model is verified and then applied to the online set-up model, troubles such as rolling control due to erroneous adjustment can be prevented and safety can be improved.

〔Example〕

 Hereinafter, the present invention will be described based on examples.

FIG. 2 shows an overall configuration diagram of an embodiment to which the present invention is applied, and FIG. 1 shows a functional block diagram of main parts.

The set-up device is largely a process signal input device 1,
It is composed of a set-up device main body 2, a storage device 3, and a set-up value output device 4. The set-up device main body 2 is divided into an online system 2A and a test system 2B, and a back ground job such as simulation is performed by the test system 2B. Online system 2A, test system 2B
Can be realized even with a single computer if it has the function of separating jobs, and it is also possible to realize a test system 2B as a backup system of a multi-system consisting of two or more computers.

The process signal input device 1 includes rolling mills 11a and 11b.
(Only the Fth and F-1th stands are shown for simplification, and the upstream stand is omitted.) Actual data of various controlled variables are input. For example, as a typical one, a thickness gauge 5
The actual plate thickness value, the actual reduction position value from the reduction position detection devices 6a and 6b, the inter-stand tension actual value from the tensiometers 7a and 7b, and the rolling load actual value from the rolling load detection devices 8a and 8b are input. Reference numeral 9 in FIG. 2 indicates a winding reel of a rolling coil, and reference numerals 10a and 10b indicate rolling speed detectors.

The actual data input to the process signal input device 1 is fetched by the data collecting means 22 and stored in the shared file 31, as shown in FIG. However, the performance data is collected at the timing when the rolling condition of each rolling mill becomes steady. This steady state determination is performed by the steady state determination means 2
It is supposed to be done by 1. For example, the judgment criteria are that the final stand rolling speed is equal to or higher than a certain value, 1/2 of the corresponding coil length has been rolled, and that 10 seconds have elapsed since the start of holding the steady state. Because I was satisfied. The shared file 31 in which the collected performance data is stored is an online system 2
It is a file in an area accessible from both the A and the test system 2B.

When the actual data is collected, the learning control means 24A is activated, the accuracy of the set-up model is determined using the actual data, and the set-up model is corrected based on this learning. The set-up calculating means 25A obtains a predetermined set-up value of the next coil by the corrected set-up model and outputs it to the controller of the rolling mill through the output device 4. Other set-up values (eg tension, rolling speed, rolling position, etc.)
Are also output together.

The model accuracy determination means 26 uses a fixed number (N, for example, 10
It is started when the model error data for 0 coils are accumulated, and it is determined whether or not the average error is within a predetermined range. When the model error exceeds the predetermined range, the regression analysis means 27 of the test system 2B is activated.

The regression analysis unit 27 uses the model error data accumulated in the online system 2A stored in the shared file 31,
A regression analysis is performed from the actual data corresponding to these, a correction function of the set-up model is obtained, and the elements of the result are stored in the test system file 33.

The learning control means 24B and the set-up calculation means 25B of the test system 2B are activated by the simulation execution command output from the simulation execution command means 23 of the online system 2A. The simulation execution instruction means 23 outputs the execution instruction at the timing of the analysis end signal output from the regression analysis means 27.

The learning control means 24B and the set-up calculation means 25B of the test system 2B have the same functional configuration as the online system 2A, and the difference is that a set-up model corrected by regression analysis is applied to simulate. Is.

The result of this simulation in the set-up calculation means 25B is printed out for collation confirmation.

On the other hand, when the model error data obtained by the simulation error calculation means included in the learning control means 24B is accumulated in a fixed number (for example, 10 to 20 coils), the test accuracy evaluation means 28 is activated and the error is calculated. Is verified to be within a predetermined range, and if so, the elements of the correction function obtained by the regression analysis are stored in a predetermined area of the online system file 32. As a result, the set-up calculation means 25A of the online system 2A corrects the set-up model based on the elements of the correction function read from the same area and determines the set-up value of the next coil.

The details will be described below. In order to make the description concrete, a case where the rolling load is determined as the set-up value will be described as an example.

The basic equation of the rolling-up set-up model is represented by the following equation (1). In the equation, P _c represents a set-up value, and x _n (n = 1, 2, ..., N) is a control variable such as a plate thickness, a tension, and a rolling speed for determining the rolling load.

P _c = (x ₁ , x ₂ , ..., X _n ) ... (1) When the formula (1) is expressed by a concrete approximation formula, the following formula (2) is obtained.

Here, h: thickness at delivery side gamma: reduction ratio b: sheet width k: 2-dimensional constrained deformation resistance Q _t: tension correction term t _f: forward tension t _b: backward tension R ': flat roll radius mu: friction factor H : thickness at entrance side R: roll radius C _H: Hichikotsuku coefficient Furthermore, the actual learning control means 24A, 24B and Setsutoatsupu calculating unit 25A, excisional up-model used in 25B includes a correction factor _{z p} by the learning control, obtained by regression analysis The set-up value P to be output to the control device is obtained by the following equation (3) corrected by the correction function Q _S thus obtained. Incidentally, (3) except Q _S by the equation is the same as inclusive excisional up-model conventional learning control.

P = z _p · P _C · Q _S (3) First, the rolling of the first coil by the set-up device is
The learning control means 24A is started when the steady state is reached, starting with the initial values of z _p = 1 and Q _S = 1. The learning control obtains the accuracy determination index Z _PA shown in the following equation (4) by the error calculating means.

Here, the actual load P _A, as a result of sub connexion rolling was executed Setsutoatsupu value P set in equation (3), the rolling load detector 8a, the detection value by 8b. Also,
The actual calculation load P _CA is obtained by using the actual value obtained from each detector as the control variable in the equation (3) as shown in the following equation (5).

_{P CA = z p · (x} 1A, x 2A, ..., x nA) · Q S ... (5) In addition, (4) the value _{Z PA} is equivalent to model error of formula, which is close to 1.0 The better the accuracy of the set-up model can be judged.

Then, the learning control means 24A finds the correction coefficient z _p by the following equation (6) and rewrites the contents of the set-up model of equation (3) to make the correction.

z _p = z _p ⁻¹ + δ _z (Z _PA −z _p ⁻¹ ) (6) where z _p ⁻¹ is the previous value of z _p δ _z is an exponential smoothing coefficient (1.0 or less) Then, the set-up model of the online system 2A is modified by the learning control performed for each coil and adjusted so as to adapt the characteristics peculiar to each plant. The above is basically the same as the learning control that has been conventionally performed.

Next, matters relating to the features of the present invention will be described.

While modifying the set-up model by learning control,
When the rolling results reached coil N present in the same steel grade, start model accuracy determining means 26 calculates the average value _PA of the obtained N pieces of determination index Z _PA and its standard deviation sigma, the following equation
When either (7) 'or (8)' is established, the regression analysis means 27 is activated to perform the regression analysis.

| Z _PA −1 |> ε ₁ (7) ′ σ> ε ₂ (8) ′ The regression analysis means 27 associates the model error represented by the N judgment indexes Z _PA with the control variable. The correction function Q _S of the following equation (7), which is to be analyzed and is represented by the main control variable x _n as a factor, and which is represented by the weighting coefficient a _n indicating the degree of influence of the factor on Z _PA , is specified. .

Q _S = a ₀ + a ₁ x ₁ + ... + a _n x _n (7) Looking at the rolling load of the formula (2), the formula (7) is set as the following formula (8).

Q _S = a ₀ + a ₁ / h + a ₂ γ + a ₃ t (8) Here, specifying Equation (8) means that the weighting factor a _n
And the weighting coefficient a _n is N
_PA and N corresponding performance data sets h _A , γ _A ,
Using t _A , regression analysis is performed using the following equation (9).

Z _PA = a ₀ + a ₁ / h _A + a ₂ γ _A + a ₃ t _A (9) Note that the initial value of Q _S in the equation (3) is a ₀ = 1 and a ₁
~ A _n = 0.

Thus the correction coefficient Q _S obtained is once stored in the test system file 33, in this embodiment, so as to store the weighting coefficients a _n.

When the regression analysis is completed, a simulation execution command is output and the learning control means 24B of the test system is activated. First, the online learning control means 24 described above is used here.
As in A, the explanation was given using equations (1) to (3), but the basic equation of the set-up model was used, and the set-up model of the test system was set by correcting according to the result of the regression analysis as described below. , And an accuracy determination index Z _PA ′ representing a model error is obtained based on the actual data. This Z _PA ′ is obtained by using the actual test calculation value P _CA ′ obtained by replacing the term of Q _{S in} the equation (5) with the result of the regression analysis, as shown in the following equation (10).

Then, based on this, the test correction coefficient z _p ′ is obtained by the following equation (11), and the set-up model of the test system used in the set-up calculation means 25B or the like is corrected.

z _p ′ = z _p ′ ⁻¹ + δ _z (Z _PA ′ −z _p ′ ⁻¹ (11) In this way, the set-up model for the test corrected by the correction coefficient Q _S is used, and the online system is used. Simulate run including learning control using the same online data as 2A, and print out the result for collation confirmation.On the other hand, a predetermined number of coils (for example, 10 to 20)
When the simulation run is executed, the accuracy evaluation means 28 is activated to verify the accuracy of the corrected set-up model. According to the model accuracy determination means 26, this verification is passed when both the following expressions (12) and (13) are satisfied.

If it passed, rewriting the contents of weighting factor a _n-line table 32 online system 2A is referring to the results of the regression analysis. As a result, the set-up value relating to the next coil is calculated by the highly accurate set-up model corrected by the regression analysis, and the accurate set-up value adapted to the characteristic peculiar to the plant is applied to the rolling. If the verification result is unacceptable, an alarm can be output to notify the operator.

〔The invention's effect〕

As described above, according to the present invention, a model error that cannot be absorbed by learning control is analyzed by regression analysis, and a set-up model corrected based on this is simulated using online data, and the result is verified. Since the model accuracy is confirmed and then applied as an online set-up model, the model error can be automatically adjusted. As a result, since no special experience or intuition is required for the adjustment work, stable adjustment can be performed regardless of the ability of the operator, and the adjustment period can be greatly shortened. In addition, since it is applied online after verification by simulation, there is no chance that wrong adjustment results will be output to the online system.
Excellent reliability.

[Brief description of drawings]

FIG. 1 is a functional block diagram of main parts of an embodiment of the present invention, and FIG. 2 is an overall configuration diagram including the embodiment of FIG. 2 ... Set-up device body, 2A ... Online system,
2B ... test system, 23 ... simulation execution command means, 24A, 24B ... learning control means, 25A, 25
B ... Set-up calculation means, 26 ... Model accuracy determination means, 27 ... Regression analysis means, 28 ... Accuracy determination means.

Claims (1)

[Claims] 1. Set-up calculation means for calculating a set-up value of a rolling control amount by a preset set-up model based on input data including a given control variable and outputting the set-up value to a rolling control device, and output from a rolling mill. Error calculation for obtaining an error between the actual calculated value of the set-up value by the set-up model using the actual value corresponding to the control variable, and the actual value output from the rolling mill corresponding to the actual calculated value. Means, a correction function that represents the error with a main variable of the control variable as a factor, and a weighting coefficient that indicates the degree of influence of the factor on the error, and a plurality of error data obtained by the error calculating means. And a regression analysis means for determining the weighting coefficient by regression analysis from the actual values of the control variables corresponding thereto and specifying a correction function, and the regression analysis. The set-up model is corrected by the correction function specified by the step, and the actual calculated value for the test is obtained by using the actual value of the control variable by the corrected set-up model. When the error calculated by the simulation error calculating means for obtaining an error from the actual value output from the rolling mill corresponding to, and the error calculated by the simulation error calculating means is within a predetermined range, And a test accuracy evaluation means for outputting a command for correcting the set-up model according to the specified correction function, and a set-up device for a rolling mill.