JPH05293516A - Method for estimating rolling load of rolling mill - Google Patents

Abstract

PURPOSE:To improve the estimating accuracy of a rolling load and to obtain better estimating accuracy with a smaller number of data used in time of estimation in a method for estimating the rolling load used by a rolling mill. CONSTITUTION:In a neural net port A, rolling conditions (kind of WR, rolling speed, thickness of a mother plate, plate thickness on the outlet side, draft) are inputted, and mu and km, being adjusting parameters, are estimated. These estimates and rolling conditions (plate thicknesses on inlet and outlet sides, tensions on inlet and outlet sides, diameter of WR) are inputted into a mathematical model B to estimate a rolled load value. In this case, the neural net work is learnt in advance by the rolling conditions and the rolling load at that time. That is, in this invention, the input and the output are considered as a certain rolling condition and a rolling load value to the rolling condition, and a rolling load value as a teacher signal, respectively.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a rolling load estimation method used in a rolling mill.

[0002]

2. Description of the Related Art In a general rolling mill, when rolling to a desired strip thickness or shape, the rolling load required at that time is estimated from the rolling conditions, and each value is used to set each operation amount. Therefore, if the rolling load estimated value is largely deviated, it will have an adverse effect on the shape and plate thickness after rolling.

The conventional rolling load estimation method in a rolling mill is based on the following (1) obtained from the equilibrium of stress during rolling.
The rolling condition was input to a mathematical model such as (5) and the rolling load q was estimated by the convergence calculation.

[0004] (the Hill equation) load estimation equation _{q = k m √ ((k} m Δh)) {1.08 + 1.79μr√ ((R '/ H)) - 1.02r} [kg / mm] ( _{1) k m = g (H0} , H, h, σb, σf, n, k o, c) (2) However, Δh = H-h (3 ) H0: Mother board thickness [mm] n: Strain hardening coefficient
[−] H: Inlet plate thickness [mm] _ko : Deformation resistance offset [kg / mm ² ] h: Outlet plate thickness [mm] c: Deformation resistance coefficient
[Kg / mm ^2] μ: Friction coefficient [-] k _m: Mean constrained yield stress ^{[kg / mm 2] σb,} σf: inlet and outlet tension ^{[kg / mm 2] R '} : flat WR diameter [mm]

Roll flat type (Hitchcock type) R: WR diameter [mm] E: Young's rate of WR [Kg / mm ² ] ι: Poisson ratio [-] (WR: Abbreviation for work roll)

Here, the friction coefficient μ included in the mathematical model
It is or the average constrained yield stress k _m accurate measurement is difficult physical property values.

The average constrained yield stress is a parameter (n, k _o ,
By introducing c), the value is more accurately estimated by calculation.

The parameters (n, k _o , c) and the friction coefficient μ introduced for the sake of convenience in the calculation of the average constrained yield stress have been determined in advance by a table using the following rolling conditions as input.

Then, given a certain rolling condition, μ and (n, k _o , c) under that condition are obtained from the table, and by using these values, the rolling load is calculated from the above equations (1) to (5). The value q was introduced.

Accordingly, when such a physical property values μ and k _m is referred to as adjustment parameter, since the adjustment parameters include discrete values is a table value, the value is varied, the adjustment parameter precision rolling load Affects the estimation accuracy,
It was always difficult to accurately estimate the rolling load.

Further, the adjustment parameter should essentially take a continuous value, and in order to obtain the target accuracy, the rolling condition which is the input of the table must be finely divided. Therefore, the table becomes huge and it takes a very long time to adjust the table value.

[0012]

As described above, in the conventional rolling load estimation method for a rolling mill, there are variations in the adjustment parameter values, and the accuracy of the adjustment parameters affects the rolling load estimation accuracy, and the accuracy is always high. There is a problem that it is difficult to estimate the rolling load, and in order to obtain the target accuracy, the rolling condition that is the input of the table must be finely divided and the table that inputs the rolling condition becomes huge. There was also a problem that it took a very long time to adjust the table values.

It is an object of the present invention to provide a new rolling load estimation method for a rolling mill which solves the above problems and improves the rolling load estimation accuracy.

[0014]

As a configuration for achieving the above object, a rolling load estimation method for a rolling mill according to the present invention is a rolling machine for rolling a steel sheet, and is accurately included in a mathematical model by a neural network model. Is characterized by estimating physical properties that cannot be measured by a neural network, and using that value to estimate rolling load by a mathematical model.

[0015]

Since the neural network model has a learning function, it is known that an arbitrary function can be approximated by the network. Therefore, in the present invention, by utilizing this feature, the adjustment parameter value conventionally obtained by the table having the rolling condition as an input is estimated by the neural network having the condition similarly inputted, and the adjustment parameter is obtained. From the estimated value and the rolling conditions, the rolling load estimated value was obtained by a mathematical model.

The neural network is composed of points and branches as shown in FIG. 3. Each point is given a threshold value, each branch is given a weighting coefficient, and the points are connected to the network by the branches. Each point has multiple inputs and one output, and each output value is calculated by an arbitrary function with respect to the value obtained by adding the threshold value at that point to all the input values multiplied by the weight.

In the present invention, learning is performed by using an input as a rolling condition, an output as a rolling load value for the rolling condition, and an actual rolling load value as a teacher signal.

Learning is a method such as backpropagation in which a weighting coefficient and a threshold value on the network are determined in advance so that the output value of the entire neural network becomes equal to the teacher signal for a certain input pattern. When a new rolling condition is input to the learned neural network, a value similar to the rolling load value for the learned similar rolling condition can be output. Therefore, the rolling load can be accurately estimated by learning many input / output patterns.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The rolling load estimating method of the present invention is shown in FIG. 1 and will be specifically described with reference to an application example thereof. Here, an example in which the present invention is applied to a rolling mill for rolling a steel plate 2 as shown in FIG. 2 with upper and lower work rolls 1, 1 will be shown. By repeating such rolling, the steel plate is processed into a target plate thickness.

[0020] Figure 3 shows an embodiment of a neural network unit of Figure 1, in the present invention, where the average yield stress k _m and μ friction coefficient and the adjustment parameter, estimated by the neural network as shown in FIG. 3 To do.

An example of using the neural network and the mathematical model in the rolling load estimation method of FIG. 1 will be described. First, in the neural network section A, rolling conditions (WR type, rolling speed, mother plate thickness, delivery plate) are used. the thickness, the rolling reduction) estimates the adjustment parameters are μ and k _m as inputs, these estimates and rolling conditions (and out side thickness, and out-side tension, the WR diameter) inputted to the mathematical model B, rolling load value To estimate.

In this case, it is assumed that the neural network has been learned in advance by a pair of rolling conditions and rolling load values at that time.

That is, in the present invention, learning is performed by using an input as a rolling condition, an output as a rolling load value for the rolling condition, and a rolling load actual value as a teacher signal.

Although one embodiment of the present invention has been described above, the present invention is not limited to the above embodiment, and various modifications can be made within the scope of the technical idea of the present invention.

[0025]

As described above, according to the rolling load estimation method for a rolling mill of the present invention, it is necessary to accurately measure the rolling load, which is considered to be included in the mathematical model for estimating the rolling load and adversely affect the estimation accuracy. The rolling load estimation accuracy can be improved by continuously estimating the impossible physical property values by the neural network.

Further, the number of data required for the estimation has conventionally required a huge table value, whereas the neural network only requires the number of points and the number of branches. As a result, accurate estimation can be performed with a smaller number of data.

[Brief description of drawings]

FIG. 1 is an explanatory diagram showing a rolling load estimation method of the present invention.

FIG. 2 is a side view showing a rolling state of a rolling mill using the method of the present invention.

FIG. 3 is an explanatory diagram showing an embodiment of the neural network unit of FIG.

[Explanation of symbols]

 A Neural network part B Mathematical model 1 Work roll 2 Steel plate

Claims (1)

[Claims] 1. In a rolling machine for rolling a steel sheet, a physical property value that cannot be accurately measured included in a mathematical model is estimated by a neural network by the neural network model, and the rolling load is calculated by the mathematical model using the value. A rolling load estimating method characterized by estimating.