JPH10180321A - Learning control method in rolling mill - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve thickness precision and shape precision even in the case of rolling a thick plate by learning a model error every the kind of steel, a rolling dimension and a work roll to be used and reflecting the error in the initial rolling schedule of the subsequent rolling stock. SOLUTION: The load model error from actual operating data is successively learned on not only the model error among respective paths but also the high reproducible factor of a rolling load model, i.e., every the kind of steel, the rolling size and the work roll, and the error is reflected in the subsequent rolling stock. By this way, the setup precision of the rolling mill is improved, consequently, thickness precision and shape precision are improved even in the case of rolling the thick plate. Also, since the hysteresis temperature is not varied in the rolling under the same condition, both temperature model error and correction taken in are simultaneously executed.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a learning control method for a rolling mill that corrects a model based on a difference between a calculated value calculated using a setup model of a rolling mill and an actual value obtained as a result of actual rolling. .

[0002]

2. Description of the Related Art When rolling with a rolling mill, a calculation of a manipulated variable is performed by a computer in order to obtain an appropriate roll opening, a rolling speed, and the like, and each manipulated variable is set based on the predicted calculated value. Like that. Whether or not a predetermined thickness accuracy required for a product can be obtained in rolling with the operation amount set in this way depends on the accuracy of a setup model used for the prediction calculation.

In general, as a method of improving the accuracy of a setup model of a rolling mill, learning control is performed in which correction is performed at the time of calculation by the model based on actual values.
Among the above-mentioned models, there is a rolling load equation as a mathematical model that greatly affects the thickness accuracy, and a method of learning a correction coefficient K shown in equation (1) is used as a learning control method of the rolling load equation.

P = K · P ₀ (1) where P is a predicted rolling load value, P ₀ is a mathematical model calculation value, and K is a correction coefficient.

A method of learning the correction coefficient K is disclosed in
As described in Japanese Patent Application Laid-Open No. 2-26709, a method of performing learning control between slabs in addition to learning control between paths is known.

More specifically, learning is performed by classifying into the deformation resistance and the friction coefficient according to the steel type, and when the steel type of the learning target material is different from the previous one, the error between the calculated value and the actual value is set within the same steel type. Learning as an error of deformation resistance with little variation,
At the same time as the previous time, the error between the calculated value and the actual value is learned as the error of the friction coefficient, which has little variation due to the change of the steel type, to improve the accuracy of the setup model.

[0007]

When rolling rolled materials of various sizes, such as thick plate rolling, the rolled size may vary greatly.

However, according to the learning method as described above, when the rolling size greatly changes, not only the variation in the error due to the size of the rolling load equation cannot be corrected, but also for the steel type with the originally poor load prediction accuracy. Since the influence of the correction amount in the actual rolling is large, the deviation from the initial rolling schedule becomes large, and there is a problem that a thickness accuracy defect and a shape defect occur.

SUMMARY OF THE INVENTION The present invention has been made in order to solve such a problem, and an object of the present invention is to provide a learning control method for a rolling mill capable of improving thickness accuracy and shape accuracy even in the case of thick plate rolling. And

[0010]

In order to achieve the above object, a learning control method for a rolling mill according to the present invention obtains a calculated value calculated by using a setup model of a rolling mill and an actual rolling result. In the learning control method of a rolling mill that corrects the model by a difference from an actual value, the error of the model is learned for each steel type, rolling dimension, and work roll used,
The error is reflected in the initial rolling schedule of the succeeding rolling material.

The mathematical model calculation value (P ₀ ) of the rolling load in the above equation (1) is generally expressed by the following equation. P ₀ = K ₀ · W · ld · Q _p (2) where K ₀ is the average deformation resistance, W is the plate width, ld is the contact projection length, and Q _p is the rolling force function.

Further, the load model element (K ₀ ,
ld, Q _p ) can be represented by the following parameters. K ₀ = f (T _k , ε, component) (3) ld = g (H, h, R ′) (4) Q _p = k (H, h, R ′) (5) _Tk is temperature, ε is strain, H is the thickness of the inlet plate, h is the thickness of the outlet plate, and R 'is the flat roll diameter.

Average deformation resistance K expressed by equation (3)
₀ changes depending on the steel type of the rolled material, and the variation within the same steel type is relatively small. However, when rolling rolled materials having various thicknesses and widths such as thick plate rolling, if the load prediction model is not corrected by the size in each rolling pass for the same steel type, the rolling must be performed. The setup accuracy of the machine does not improve.

Further, when the grinding condition of the work roll during rolling changes, the friction coefficient changes and the amount of load changes, so that it becomes necessary to correct the load prediction model according to the type of work roll used.

Therefore, in the present invention, the error of the load prediction model is learned for each of the elements with high reproducibility, ie, steel type, rolling size, and work roll, and the correction is reflected in the initial rolling schedule for the next and subsequent materials.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is an explanatory diagram for explaining a learning control method for a rolling mill, which is an example of an embodiment of the present invention. FIG. 2 is a graph showing the relationship between steel types and load prediction coefficients before and after online learning control. FIG. 3 is a graph showing the relationship between the steel type and the FCF average value before and after the online learning control.

As shown in FIG. 1, the steel type (composition) and the rolling dimensions (thickness and width categories) are obtained from actual rolling results in actual rolling of thick plate rolling.
And load model error (FCF)
= Actual load / predicted load in load model) is learned and updated for each mesh as follows.

FCF _Tb (new) = FCF _Tb (old) + α (FCF _mo (instantaneous) −FCT _Tb (old)) (6) where FCF _Tb is an FCF learning table value and FCF
_mo (instantaneous) indicates the instantaneous error value of the load model, and α indicates the influence coefficient.

The reflection of rolling on the next slab is as follows:
In the case of a slab where the steel type, rolling size, and work roll are the same, use the steel type table corresponding to the slab and use the learning value of the mesh corresponding to the thickness and width of the rolled material in each pass to initialize. Create a rolling schedule.

As described above, in this embodiment, not only learning of the model error between each pass, but also the rolling load model is obtained by using a highly reproducible element such as a steel type, a rolling size and a work roll. Since the error is sequentially learned and the error is reflected on the succeeding rolled material, the setup accuracy of the rolling mill is improved, and as a result, even in the case of thick plate rolling, the thickness accuracy and the shape accuracy are reduced. Improvement can be achieved. Further, since the hysteresis temperature does not change in the rolling under the same conditions, it is possible to simultaneously perform the correction incorporating the error of the temperature model.

FIGS. 2 and 3 show the relationship between the steel type and the load prediction coefficient and the relationship between the steel type and the FCF average value before and after the learning control according to this embodiment. 2 and 3
As is clear from FIG. 7, after the learning control according to the present embodiment is performed online, the load prediction accuracy is improved, and the FCF average value is also stable.

[0022]

As is apparent from the above description, according to the present invention, the load model error is sequentially learned from the actual operation data for each of the steel type, roll size and work roll, which are factors with high reproducibility. And the error is reflected in the succeeding rolled material, so that the setup accuracy of the initial rolling schedule is improved, and the thickness accuracy and shape accuracy can be improved even in the case of thick plate rolling. Is obtained.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram for describing a learning control method for a rolling mill as an example of an embodiment of the present invention.

FIG. 2 is a graph showing a relationship between a steel type and a load prediction coefficient before and after online learning control.

FIG. 3 is a graph showing a relationship between a steel type and an FCF average value before and after online learning control.

Claims (1)

[Claims] 1. A learning control method for a rolling mill, which corrects a model based on a difference between a calculated value calculated using a setup model of a rolling mill and an actual value obtained as a result of actual rolling. A learning control method for a rolling mill, wherein an error of the model is learned for each work roll used, and the error is reflected in an initial rolling schedule of a subsequent rolling material.