KR20030052754A - Control method for asymetric shape in stainless sendzimir mill using neural network - Google Patents

Abstract

PURPOSE: A method for controlling an asymmetric shape in a stainless sendzimir mill is provided to prevent a meandering of a strip and to reduce defects of the strip by correcting the asymmetric shape in short time. CONSTITUTION: A present shape of a stainless sendzimir mill is detected by using a shape detector including a plurality of cell aligned widthwise a strip(S210). A representative shape pattern is represented based on the present shape of the stainless sendzimir mill. A difference between the present shape of the stainless sendzimir mill and the representative shape pattern is calculated, thereby extracting asymmetric components(S230). Then, a present asymmetric component is corrected while controlling a position of an as-u-roller based on the asymmetric components(S240).

Description

Control method for asymetric shape in stainless sendzimir mill using neural network

The present invention relates to a shape control method in stainless cold rolling (cold rolling), and particularly provides a control method of the asymmetrical shape using a neural network in the Zhenmime mill.

A general zenmemere mill supports a work roll for rolling an input steel sheet, a first intermediate roll used to control the shape of an edge portion of the steel sheet while supporting the work roll, and the first intermediate roll. The secondary intermediate roll, the supporting roll supporting the secondary intermediate roll, the ed-yu roll for controlling the width direction of the steel plate, the shape detector for detecting the shape of the steel plate, and controlling the shape of the steel plate. And a hydraulic system for operating the primary intermediate roll and the egg-roll.

A representative problem among the operation instability factors using the Jenjime mill may include asymmetry of the plate shape. Due to such asymmetry, the shape becomes poor during rolling operation, meandering phenomenon occurs, and the likelihood of accidents such as plate breakage increases. In addition, the rolling speed can not be increased, resulting in reduced production. Therefore, it is very important and urgent problem to eliminate such asymmetry of plate shape in rolling operation.

In order to solve this problem, Japanese Patent Laid-Open No. 9-85309 "Rolling method of a stainless steel sheet by a Zhenmier rolling mill" has a shape of a rolled material by the Eze-U-roll while vibrating the primary intermediate roll during rolling. How to control it, and

Japanese Patent Laid-Open Publication No. 5-317940 "Method for Controlling Rolling Material Shape of Zhenmier Rolling Mill" classifies the measured shape into several steepness patterns by neural network, and uses the fuzzy inference for the ez-yu- The method of determining the control ring of the roll and the primary intermediate roll is used.

However, this method classifies the current shape into a specific basic pattern irrespective of symmetry asymmetry to determine the position of the Eze-U roll, so that the meandering phenomenon caused by the failure to remove the plate-shaped asymmetry component in a short time is There is a problem that does not effectively prevent the operation instability (such as the phenomenon of shaking from side to side).

The present invention aims to produce a high-quality product by correcting such asymmetry of the plate shape in a short time, to prevent meandering phenomenon during operation, to reduce the defect of the plate shape.

1 is a block diagram of a Zenmime mill device applied to the implementation of the present invention,

2 is a flowchart illustrating a shape control method according to the present invention;

Figure 3 is a representative shape pattern of the second, fourth, sixth function form used in the present invention.

※ Description of the main parts of the drawings ※

1: steel plate 11a, 11b: working roll

12a, 12b, 12c, 12d: 1st intermediate roll

13a, 13b, 13c, 13d, 13e, 13f: secondary intermediate roll

14a, 14d, 14e, 14f, 14g, 14h: support rolls 14b, 14c: ed-you-roll

20: hydraulic system 30a, 30b: shape detector

40: control unit

According to the present invention for achieving the above object, a polynomial in which the preceding term is an even order from a first step of detecting a current shape from a shape detector formed of a plurality of cells in the width direction of the steel sheet and the current shape obtained in the first step. A second step of classifying the representative shape pattern, a third step of extracting an asymmetric component by obtaining a difference between the shape value obtained in the first step and the representative shape pattern obtained in the second step, and the third step A stainless cold rolled shape control method is provided, comprising a fourth step of compensating a component for the current asymmetry by controlling the position of the egg-roll with an asymmetrical component.

Hereinafter, a configuration of an apparatus for performing a shape control method according to the present invention will be described with reference to the accompanying drawings.

The structure of a typical Zenmime mill is shown in FIG. 1. The steel sheet 1 is rolled through the upper and lower work rolls 11a and 11b. The first intermediate rollers 12a, 12b, 12c, and 12d control the shape of the edge portion of the steel sheet 1 while supporting the work rolls 11a and 11b. Second intermediate rollers 13a, 13b, 13c, 13d, 13e, and 13f serve to support the primary intermediate rolls 12a, 12b, 12c, and 12d. The backup rollers 14a, 14d, 14e, 14f, 14g, and 14h support the secondary intermediate rolls 13a, 13b, 13c, 13d, 13e, and 13f. -u-roller) 14b and 14c are divided into eight segments in the longitudinal direction (the direction perpendicular to the drawing), and each segment is operable independently to control the width direction of the steel sheet. Used. The hydraulic system 20 operates the primary intermediate rolls 12a, 12b, 12c, 12d and the egg-rolls 14b, 14c, and the shape detectors 32a, 32b have 32 load cells in the longitudinal direction. The shape of the steel plate is detected by measuring the load applied to each cell, which is configured as a load cell. The controller 40 receives the current shape of the steel sheet from the shape detectors 32a and 32b, classifies the shape pattern, and then supplies the primary intermediate rolls 12a, 12b, 12c, and 12d to the hydraulic system 20. Direct operation of the egg-rolls 14b and 14c.

2 is a flowchart of a shape control method according to the present invention. In the control method according to the present invention, the first step (S210) of measuring the shape of the steel sheet 1 through which the shape detectors 30a and 30b pass therethrough with 32 load cells, and the control unit 40 to the shape detectors 30a and 30b The second step (S220) of receiving the measured shape from the) and classifies it into a representative shape pattern, by calculating the difference between the shape measured from the shape detectors (30a, 30b) and the shape pattern classified in the second step The third step (S230) of calculating the asymmetrical shape component, and using the calculated asymmetrical shape component as the E-roll (14b, 14c) position calculation logic input, the position of the egg-roll (14b, 14c) The fourth step 240 is to calculate.

As the representative shape pattern of the second step, three types of quadratic functions, quaternary functions, and sixth order functions shown in FIG. 3 are used. These patterns have a symmetrical shape, and the symmetry component of the shape being measured can be represented by various combinations of these patterns. In the present invention, when classifying the measured shape into a representative shape, a back-propagation learning method having a structure called multi layred perceptron and a momentum uses a learning algorithm of a neural network.

The neural network model is a model of neural network, which is a combination of neurons in the human brain. It simulates the information processing of the human brain and trains the system to perform certain tasks. Normal neural networks are highly applicable because they are learned through examples, and do not require a separate storage location because information is encoded and processed in a distributed form. Neural networks show very good performance in pattern recognition processing that requires many assumptions and knowledge to be processed by parallel, high-speed computation, and so on. There are many models of neural network such as perceptron, multilayer perceptron, optimal classifier, Gaussian classifier.

As shown in FIG. 4, the multilayer perceptron structure used in the present invention is composed of an input layer, an intermediate layer, and an output layer. The input layer is composed of 32 nodes, and the shape detector composed of the 32 load cells The current shape values detected at 30a and 30b) are input. The intermediate layer is composed of 16 nodes and serves as an intermediate connection between the input layer and the output layer. The output layer is composed of three nodes, and the output is a result of classifying the pattern, indicating the degree of three basic patterns included in the input shape. That is, the values detected at the three nodes of the output layer represent the weights of the second, fourth and sixth functions, respectively.

That is, make a function with 32 shape values input from the shape measurement roll, and The values obtained by approximating to are output from the output node as a, b, and c, respectively.

The calculation of the asymmetrical shape component of the third step uses Equation 1 below.

N (n) = F (n)-S (n)

n is an index from 1 to 32, N is an asymmetric shape component, F is a shape measured from a shape detector, and S is a symmetric component calculated from the pattern classified in the second step. The asymmetric component thus obtained is controlled to control the egg roll to remove the asymmetric component of the steel sheet.

As described above, only the asymmetric components of the steel sheet are first extracted, extracted, and corrected, thereby eliminating the plate-shaped asymmetric components in a short time, and effectively preventing operation instability such as meandering phenomenon of the steel sheet.

The technical spirit of the present invention has been described above with reference to the accompanying drawings, which describes exemplary embodiments of the present invention by way of example and does not limit the present invention. In addition, it is obvious that any person skilled in the art can make various modifications or imitations without departing from the technical scope of the present invention.

Claims (2)

In stainless cold rolling shape control method, A first step of detecting a current shape from a shape detector formed of a plurality of cells in the width direction of the steel sheet; A second step of expressing the representative shape pattern by a polynomial whose previous term is even order from the current shape obtained in the first step; A third step of extracting an asymmetric component by obtaining a difference between the shape value obtained in the first step and the representative shape pattern obtained in the second step; And A fourth step of controlling the position of the egg-roll with the asymmetric component obtained in the third step, thereby compensating for the current asymmetry component; Stainless cold rolling shape control method, characterized in that configured to include. The multi-layer perceptron structure of claim 1, wherein the representative shape of the second stage comprises a neural network having an input layer consisting of 32 nodes, an intermediate layer consisting of 16 nodes, and an output layer consisting of three nodes. And a quadratic function, quadratic function, and sixth order function, wherein the three nodes of the output layer represent weights of the functions.