JPH04232214A - Method and device for controlling cooling of steel plate - Google Patents

Abstract

PURPOSE:To accurately predict the coefficient of heat transfer at the time of cooling a steel plate and to obtain the desired material quality in the steel plate after cooling. CONSTITUTION:Sensors 14, 16 for detecting temp. of the steel plate at inlet and outlet in this cooling device and the actual value to the coefficient of heat transfer is calculated with an arithmetic part 36 for the actual value to the coefficient of heat transfer based on control data inputted from an input device 35 disposed separately and temps. detected with these sensors 14, 16. Then, new continuous functional values of components in the steel plate, outer shape of the steel plate and temp. of the steel plate having the typical coefficient of heat transfer shown by neural net work model so that the predictive value to the coefficient of heat transfer approaches to the actual value, is calculated with a learning part 37 for the coefficient of heat transfer and the cooling at the next time is executed based on this functional value.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] The present invention is a method for cooling a steel plate, which controls cooling of a steel plate based on a heat transfer coefficient expressed as a continuous function value of steel plate components, steel plate outer shape, and steel plate temperature using a neural network model. The present invention relates to a control method and device.

0002

[Prior Art] Currently, a process called controlled cooling is used to obtain a quenching effect by rapidly cooling a high-temperature thick steel plate immediately after hot rolling with water cooling, thereby imparting high strength properties to the steel plate. Steel sheet manufacturing equipment is now in operation. This cooling is
Cooling is continued until the steel plate reaches the target temperature, but in order to accurately cool the steel plate, it is necessary to set the degree of cooling of the steel plate as a cooling control factor. This degree of cooling can be expressed by the amount of heat per unit temperature, unit time, and unit area, and is equivalent to the heat transfer coefficient.

[0003] Generally, this heat transfer coefficient is difficult to formulate because of its strong nonlinearity. For this reason, the heat transfer coefficient is treated as a discrete table value classified into types depending on factors such as the composition, external shape, and temperature of the steel plate to be cooled. When cooling a steel plate, the predicted value of the heat transfer coefficient for the next cooling is estimated from the previous actual value of the heat transfer coefficient for each applicable type classification of the steel plate to be cooled.

0004

[Problem to be Solved by the Invention] However, since the table values of this heat transfer coefficient are discrete, the heat transfer coefficient is discontinuous between the steel plate type classifications, and the heat transfer coefficient is discontinuous between the steel plate type classifications, and the heat transfer coefficient is discontinuous between the steel plate type classifications. A uniform heat transfer coefficient is not necessarily applied to steel plates. If a steel plate is cooled by applying an inappropriate heat transfer coefficient, there is a risk that the desired material quality will not be obtained in the steel plate after cooling. The present invention was made to solve these conventional problems, and uses a neural network model to express the heat transfer coefficient as a continuous function value based on factors such as steel plate composition, steel plate outer shape, and steel plate temperature. By setting the table values to the table values, the occurrence of the above-mentioned problems can be avoided, and by updating the table values sequentially based on the actual values of steel plate cooling, it is possible to improve the cooling control accuracy of the steel plate. The purpose of this invention is to provide a cooling control method and device.

[0005]

[Means for Solving the Problems] One of the present inventions for achieving the above object is to measure the temperature of a steel plate detected at the inlet and outlet of a cooling device, as well as physical elements such as the composition and external shape of the steel plate. calculate the actual value of the heat transfer coefficient of the steel plate based on the actual value, calculate the predicted value of the heat transfer coefficient during the next cooling of the steel plate based on the actual value, and control the cooling of the steel plate based on the predicted value. In the steel plate cooling control method, the heat transfer coefficient is determined as a continuous function value of steel plate composition, steel plate outer shape, and steel plate temperature using a neural network model based on the detected temperature and physical factors of the steel plate. This is a cooling control method for a steel plate, characterized in that the steel plate is cooled based on the function value. The predicted value of the heat transfer coefficient of the steel plate is calculated based on the standard heat transfer coefficient expressed as a continuous function value, and the temperature of the steel plate detected at the inlet and outlet of the cooling device, as well as the temperature of the steel plate in question are calculated. The actual value of the heat transfer coefficient of the steel plate is calculated based on physical elements such as composition and external shape, and the predicted value is expressed by a neural network model using a learning function so that the predicted value approaches the actual value. The continuous function values of the steel plate composition, steel plate outer shape, and steel plate temperature of the standard heat transfer coefficient are learned and updated sequentially, and the next steel plate is cooled based on the updated continuous function values. This is a method for controlling cooling of a steel plate.

Furthermore, an inlet temperature detection means for detecting the temperature of the steel plate at the inlet of the cooling device, an outlet temperature detection means for detecting the temperature of the steel plate at the outlet of the cooling device, and information on the composition, external shape, etc. of the steel plate to be cooled. an input means for inputting physical elements; a storage means for storing a standard heat transfer coefficient expressed by a neural network model as a continuous function value of steel sheet composition, steel sheet outer shape, and steel sheet temperature; and said input means. predicted value calculation means for calculating a predicted value of the heat transfer coefficient based on the physical elements of the steel plate inputted from the storage means and the standard heat transfer coefficient in the storage means; and the input side and outlet temperature detection means, respectively. Actual value calculation means for calculating the actual value of the heat transfer coefficient based on the detected temperature of the steel plate and the physical elements of the steel plate input from the input means; and the actual value of the heat transfer coefficient calculated by the prediction calculation means. a standard heat transfer coefficient steel plate component represented by a neural network model stored in the storage means so that the predicted value approaches the actual value of the heat transfer coefficient calculated by the actual value calculation means; This is a cooling control device for a steel plate, characterized in that it has updating means for calculating and updating new continuous function values of the steel plate outer shape and steel plate temperature.

[0007]

[Operation] According to the method of the present invention as described above, the heat transfer coefficient is expressed by a neural network model as a continuous function value of the steel plate composition, the steel plate outer shape, and the steel plate temperature. Therefore, it is possible to obtain an optimal predicted value of the heat transfer coefficient depending on the physical factors of the heat transfer coefficient. Since the steel plate is cooled based on this predicted value, it is possible to always obtain a steel plate of a desired material. Furthermore, the heat transfer coefficient is sequentially learned and updated based on the actual value of the steel plate cooled according to the predicted value, so that the predicted value and the actual value match, so as the number of times the steel plate is cooled increases, The optimum heat transfer coefficient can now be obtained.

Further, according to the apparatus of the present invention as described above, the updating means changes the predicted value of the heat transfer coefficient calculated by the prediction calculating means to the actual value of the heat transfer coefficient calculated by the actual value calculating means. A new continuous function value of the heat transfer coefficient that approaches the new one is calculated, and the continuous function value obtained as a result of this calculation is updated and stored in the storage means. Therefore, the storage means always stores the latest data regarding the heat transfer coefficient, and as this data is updated, accurate predicted values can be calculated.

[0009]

Embodiments Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a schematic configuration diagram of the apparatus of the present invention. On the inlet side of the cooling device 12 that forcibly cools the steel plate 10 by injecting water 11,..., 11, there is an inlet temperature sensor 14 that detects the temperature of the steel plate 10 before cooling. Each outlet temperature sensor 16 is provided to detect the temperature after cooling. The inlet temperature sensor 14 and the outlet temperature sensor 16 function as an inlet temperature detector and an outlet temperature detector, respectively. The steel plate 10 is the motor 18
The paper is conveyed at a predetermined speed in the direction A in the drawing by drive rollers 20A to 20H driven by. The temperature of the steel plate 10 detected by the entrance temperature sensor 14 is determined by the A/D
The temperatures detected by the converter 21 and the output temperature sensor 16 are converted into digital signals by the A/D converter 22 and output to the industrial computer 25. The industrial computer 25 individually controls on/off a plurality of water injection valves, which will be described later, provided in the cooling device 12, and also controls the rotation speed of the motor 18. A vector calculation device 30 is connected to this industrial computer 25.
A neural network model is stored that expresses the heat transfer coefficient of as a continuous function value of steel plate components, steel plate outer shape, and steel plate temperature. The function value of the heat transfer coefficient represented by this neural network model is successively updated to the optimum value based on the actual value and predicted value of the heat transfer coefficient calculated by the industrial computer 25. There is.

FIG. 2 is a schematic diagram of a portion where the inlet or outlet temperature sensors 14 and 16 are installed. As shown,
When the steel plate 10 is cooled by the water 11, ..., 11 from the cooling device 12, water vapor is generated to the extent that visibility is obstructed.
Since this water vapor will interfere with temperature measurement by the inlet or outlet temperature sensor 14 or 16, devices for blowing off this water vapor are provided on the inlet and outlet sides. This device has the same function as a blower, and the air sent by the blower 19 is blown out from the duct 17 toward the steel plate 10. A radiation thermometer is used as the inlet or outlet temperature sensor 14 or 16, and this radiation thermometer is installed inside the duct 17. Therefore, water vapor that would interfere with temperature measurement is reliably removed by the air from the duct 17, making it possible to detect temperature under a favorable measurement environment.

FIG. 3 is an internal configuration diagram of the cooling device 12. Inside the cooling device 12, a water injection device 2 is installed at predetermined intervals.
3a to 23n are provided, and these water injection devices 23a to 23
Water is injected toward the steel plate 10 from n. Each water injection device is provided with water injection valves 24a to 24n, and the shutoff and opening of these water injection valves 24a to 24n is controlled via control lines 26a to 26n connected to the respective water injection valves, as shown in FIG.
This is carried out by the industrial computer 25 shown in FIG.

FIG. 4 is a specific configuration diagram of the vector calculation device 30. A modem 31 is provided to exchange data with the industrial computer 25, and data input from the modem 31 or output to the modem 31 is exchanged via a data transmission and collection computer 32, and the data is stored in the data storage. It is stored in the container 33. Specifically, this data is a function value of a heat transfer coefficient expressed by a neural network model. The vector calculator 34 is capable of performing matrix calculations at high speed, and calculates the heat transfer coefficient to be output to the industrial computer 25 based on the data stored in the data storage 33, and vice versa. The result calculated based on the data from 25 is updated and stored in the data storage 33.

FIG. 5 is a block diagram of the control system of the apparatus of the present invention. Inlet temperature sensor 14 and outlet temperature sensor 16
As described above, these are temperature sensors that measure the temperature of the steel plate 10 at the inlet and outlet sides of the cooling device 12, and radiation thermometers are used. The data detected by the inlet temperature sensor 14 and the outlet temperature sensor 16 and data regarding the physical elements of the steel plate and the steel plate temperature (coiling temperature) input from the steel plate data input unit 35 are sent to the heat transfer coefficient actual value calculation unit 36. and is input to the heat transfer coefficient learning section 37. The steel plate data input unit 35 is for inputting the steel plate components of the steel plate to be cooled, physical elements such as the steel plate outer shape, and the steel plate temperature (winding temperature). This input may be performed by a keyboard or automatically by a host computer. The steel plate data input section 35 functions as an input means, the heat transfer coefficient actual value calculation section 36 functions as an actual value calculation means, and the heat transfer coefficient learning section 37 functions as a predicted value calculation means and an updating means. The steel plate data input section 35 is provided either as an accessory to the industrial computer 25 or within its main body. Further, the heat transfer coefficient learning section 37 is provided in the vector calculation device 30.

The heat transfer coefficient actual value calculation unit 36 inputs the temperatures at the inlet and outlet sides of the steel plate detected by the inlet temperature sensor 14 and the outlet detection sensor 16, and the steel plate data input unit 35.
The actual heat transfer coefficient of the steel plate is calculated based on the data regarding the steel plate composition, external shape, and other physical elements as well as the steel plate temperature input from the industrial computer 25. ing. Heat transfer coefficient learning part 3
7 is a heat transfer coefficient actual value calculation unit 36 in addition to data regarding the inlet and exit temperatures of the steel plate, physical elements of the steel plate, and steel plate temperature that are input to the heat transfer coefficient actual value calculation unit 36.
Input the actual heat transfer coefficient of the steel plate calculated by and the cooling time of the steel plate calculated by the cooling time calculation unit 38 described later, and calculate a new heat transfer coefficient such that the predicted value of the heat transfer coefficient matches the actual value. Compute continuous function values for . The calculated function value is then updated and stored in the data storage device 33, which functions as a storage means. Specifically, this function value is a continuous function value expressed by a neural network model using parameters such as steel sheet composition, steel sheet outer shape, and steel sheet temperature (coiling temperature). The data accumulator 33 is provided within the vector calculation device 30 as described above.

The cooling time calculation unit 38 calculates the optimum temperature for the steel plate 10 based on the heat transfer coefficient calculated by the heat transfer coefficient learning unit 37 and the inlet temperature of the steel plate 10 detected by the inlet temperature sensor 14. The cooling time is calculated and the calculation result is sent to the water injection valve opening/closing control section 39 and the steel plate conveyance speed control section 4.
Output to 0. The water injection valve opening/closing control section 39 controls the opening and closing of each water injection valve based on the result of this calculation, and the steel plate conveyance speed control section 40 controls the rotational speed of the motor 18 based on the result of this calculation. This allows the steel plate to be cooled appropriately. The result of this calculation is also input to the heat transfer coefficient learning section 37, and is referred to during the learning calculation of the heat transfer coefficient. The cooling time calculation section 38, the water injection valve opening/closing control section 39, and the steel plate conveyance speed control section 40 are provided within the industrial computer 25.

The steel plate cooling control device configured as described above calculates and learns the heat transfer coefficient as described below based on the flowcharts shown in FIGS. 7 to 9. FIG. 6 is a structural diagram of a backpropagation neural network model that implements the steel plate cooling control method of the present invention, FIG. 7 is a main flowchart showing the heat transfer coefficient calculation process, and FIG. 8 is Step 3 of the main flowchart. FIG. 9 is a subroutine flowchart of the processing of
is a subroutine flowchart of the process of step 6 of the main flowchart. In FIG. 6, i is the input layer, j is the first hidden layer, l is the second hidden layer, k is the output layer, Wji is the coupling coefficient matrix of the input layer i and the first hidden layer j, and θ
j is the threshold vector of the first hidden layer j, Rlj is the coupling coefficient matrix of the first hidden layer j and the second hidden layer l, φl is the threshold vector of the second hidden layer l, and Vkl is the coupling coefficient matrix of the second hidden layer l and the output layer k
The coupling coefficient matrix of rk is the threshold vector of output layer k.

First, when the cooling control device is activated, the heat transfer coefficient learning section 37 inputs each coupling coefficient matrix Wji, Rlj, Vkl and threshold vectors θj, φl rk stored in the data storage 33. Note that the initial values of these coupling coefficient matrices and threshold vectors may be arbitrarily set at the beginning because they are successively modified to the optimum values through the processing of subsequent steps. Note that this coupling coefficient matrix and threshold vector are continuous function values of the heat transfer coefficient described above (step 1). Next, the heat transfer coefficient learning unit 37 inputs the physical elements of the steel plate input by the steel plate data input unit 35, that is, the steel plate composition, the steel plate outer shape, and the steel plate temperature (hereinafter, these data are collectively referred to as control). data). This input may be performed arbitrarily using a keyboard or the like, or automatically using data sent from a host computer. The steel sheet components mentioned above are carbon content, silicon content, and manganese content, the steel sheet outer shape is the steel sheet thickness, and the steel sheet temperature is the temperature after cooling by the cooling device (winding temperature). These data input here will be input to the input layer i in FIG. 6 (step 2). Next, the heat transfer coefficient learning unit 37 calculates an estimated value, in other words, a predicted value, of the heat transfer coefficient based on each data input in the above steps. Specifically, these data are input to the input layer i in FIG. 6, and calculations are performed to obtain the estimated value of the heat transfer coefficient to be output to the output layer k. Note that this series of processing is signal processing from the input layer to the output layer, and henceforth will be referred to as forward calculation. Further, the specific process in this step will be explained in detail later based on the subroutine flowchart of FIG. 8 (step 3).

The cooling time calculation unit 38 calculates the cooling time of the steel plate to be cooled based on the predicted value of the heat transfer coefficient calculated in step 3, and the cooling time calculation unit 38 calculates the cooling time of the steel plate to be cooled based on the predicted value of the heat transfer coefficient calculated in step 3. It is output to the conveyance speed control section 40. The water injection valve opening/closing control unit 39 controls the opening and closing of each water injection valve based on this cooling time, and controls the steel plate conveyance speed control unit 4.
0 controls the rotational speed of the motor 18 based on this cooling time so that the steel plate is transported at an optimal transport speed. As a result, the steel plate is transported within the cooling device 12 at an optimal transport speed and cooled under optimal cooling conditions. Note that the desired cooling conditions can also be obtained by keeping the conveyance speed of the steel plate constant and changing the actual cooling length by varying the number of water injection valves that are opened (step 4).
). The temperature of the steel plate after this cooling is detected by the outlet temperature sensor 16, and the heat transfer coefficient actual value calculation unit 3
6 is the detected temperature and the inlet temperature detection sensor 1
The temperature detected by 4, and the steel plate data input section 3
Based on the input data from step 5, the actual value of the heat transfer coefficient of the steel plate is calculated (step 5). Next, the heat transfer coefficient learning unit 37 performs step 1 so that the predicted value of the heat transfer coefficient calculated based on the control data input in step 2 matches the actual value calculated in step 5.
A correction operation is performed on the input coupling matrix and threshold vector. In other words, each coupling coefficient matrix Wji, Rlj, Vkl and threshold vector θj
, φl, and rk are performed. Note that since this series of processing is signal processing from the output layer to the input layer, it will be referred to as backward calculation from now on. Further, the specific processing process in this step will be explained in detail later based on the subroutine flowchart of FIG. The calculation in this step is a learning calculation using the backpropagation method (step 6). After the heat transfer coefficient learning unit 37 performs this calculation, the calculation result is updated and stored in the data storage unit 33. Then, the next steel plate cooling is performed based on this updated data (step 7).

Next, the processing in step 3 of the main flowchart will be explained in more detail based on the subroutine flowchart of FIG. This step concerns forward calculations to calculate the estimated heat transfer coefficient,
This is performed by the heat transfer coefficient learning section 37. First, the control data input in step 2 of the main flowchart is set to a vector value Ii in a form that can be input to the input layer i (step 10). Then, the signal Ii output from the input layer i has a coupling coefficient matrix W representing the coupling relationship between the input layer i and the first intermediate layer j before reaching the first intermediate layer j.
After being transformed by ji, the intermediate layer threshold θj is added. That is, the signal Uj reaching the first intermediate layer j
is given by the following equation (step 11).

[0020]

[Math 1]

Next, the signal U input inside the first intermediate layer j
Processing is performed to convert j to a signal Hj and output it. The conversion here means that the signal Uj is converted using a monotone non-decreasing function f.
is normalized to a signal Hj having a value from 0 to 1. More specifically, the value of the signal Uj is to be associated with a value between 0 and 1, and the signal Hj is Hj
=f(Uj). The value of Hj is a value that satisfies 0<Hj<1. In this embodiment, a function expressed by the following equation was used as the monotonically non-decreasing function f. f(x)=1/{1+exp (-x)}
(Step 12). Next, the signal Hj output from the first intermediate layer j is
After being transformed by a coupling coefficient matrix Rlj representing the coupling relationship between the first intermediate layer j and the second intermediate layer l before reaching the intermediate layer l, the threshold value φl of the second intermediate layer l is added. That is, the signal Uj reaching the first intermediate layer j is given by the following equation (step 13).

[0022]

[Math 2]

Processing is performed to convert the signal tl input into the second intermediate layer l into a signal Gl and output it. The conversion here means converting the signal tl from 0 to 1 using a monotonically non-decreasing function f.
The purpose is to normalize the signal Gl having values up to . More specifically, the value of the signal tl is associated with a value between 0 and 1, and the signal Gl can be determined by the formula Gl=f(tl). The value of Gl satisfies 0<Gl<1. In this embodiment, an equation similar to that used in step 12 was used as the monotonically non-decreasing function f (step 14). Further, before reaching the output layer k, the signal Gl output from the second intermediate layer l is converted by a coupling coefficient matrix Vkl representing the coupling coefficient between the second intermediate layer l and the output layer k, and then A threshold rk of k is added. That is, the signal Sk reaching the output layer k is given by the following equation (step 15).

[0024]

[Math 3]

Then, the signal Sk input inside the output layer k
Processing is performed to convert and output the signal Ok. The conversion here means normalizing the signal Sk to a signal Ok having a value from 0 to 1 using a monotonically non-decreasing function f. More specifically, the value of the signal Sk is to be associated with a value from 0 to 1, and the signal Ok is
=f(Sk). The value of Ok is a value that satisfies 0<Ok<1. Note that the monotonically non-decreasing function f used in this case is the same function as that used in the normalization in step 12 (step 16). Finally, take out the signal Ok that has reached the output layer k as the estimated value of the heat transfer coefficient,
The process proceeds to step 4 of the main flowchart (step 17).

Further, the process of step 6 of the main flowchart will be explained in more detail based on the subroutine flowchart of FIG. This step relates to backward calculation for modifying the coupling coefficient matrix and the threshold vector, and is performed by the heat transfer coefficient learning unit 37. First, the actual value of the heat transfer coefficient calculated in step 5 of the main flowchart is set as the teacher signal value Tk (step 20). Then, the error δk between the actual value Tk of the heat transfer coefficient in the output layer k and the estimated value Ok of the heat transfer coefficient calculated by the forward calculation described above is calculated as δk = Ok - T
Calculate by the formula of k (step 21). Next, it is determined whether or not this error δk is within a preset tolerance range. If it is within the tolerance range, the process proceeds to step 7 of the main flowchart; if it is not within the tolerance range, processing from the next step onward is performed. Then, backward calculation is performed (step 22). In this step, the coupling coefficient matrix Vkl representing the coupling relationship between the second intermediate layer l and the output layer k and the threshold vector rk of the output layer k are modified using the delta rule so as to minimize the error δk. Here, modification by the delta rule means that the coupling coefficient matrix and threshold vector before modification are Vkl, rk, respectively, and the coupling coefficient matrix and threshold vector after modification are Vkl′, rk, respectively.
′, it is expressed by the following formula. Vkl′=Vkl+αδk Ok (
1-Ok)Gl rk'=rk
+βδk Ok (1-Ok) However, α
, β is a constant

(Step 23).

Next, the error εl between the actual value Tk of the heat transfer coefficient in the second intermediate layer l and the estimated value Ok of the heat transfer coefficient calculated by forward calculation is calculated using the following equation (step 24).

[0028]

[Math 4]

In order to minimize the error εl, the coupling coefficient matrix Rlj representing the coupling relationship between the first intermediate layer j and the second intermediate layer l and the threshold vector φl of the second intermediate layer l are modified by the delta rule. . Here, modification by the delta rule means that the coupling coefficient matrix and threshold vector before modification are Rlj and φl, respectively, and the coupling coefficient matrix and threshold vector after modification are Rlj′ and φl′, respectively.
It is expressed by the following formula. Rlj′=Rlj+αεl Gl (
1-Gl ) Hj φl ′=φ
l +βεl Gl (1-Gl) However, α and β are constants

(Step 25). Furthermore, the actual value Tk of the heat transfer coefficient in the first intermediate layer j
Then, the error σj of the estimated value Ok of the heat transfer coefficient calculated by the forward method calculation is calculated using the following equation (step 26).

[0030]

[Math 5]

Finally, in order to minimize the error σj,
The coupling coefficient matrix Wji representing the coupling relationship between the input layer i and the first intermediate layer j and the threshold vector θj of the first intermediate layer j are modified by the delta rule. Here, modification by the delta rule means that the coupling coefficient matrix and threshold vector before modification are Wji and θj, respectively, and the coupling coefficient matrix and threshold vector after modification are Wji′ and θj′, respectively. expressed. Wji′=Wji+ασj Hj (
1-Hj ) Ii θj ′=θj +
βσj Hj (1-Hj) However, α,
β is a constant

(Step 27). In this way, after performing steps 23, 25, and 27 to correct the coupling coefficient matrix and threshold vector, the process returns to step 22 and the error δk between the actual value Tk and estimated value Ok of the heat transfer coefficient is This will be repeated until it converges within the allowable range. Through the above processing, it is possible to always maintain a heat transfer coefficient having an optimal function value.

[0032] By performing forward and backward calculations in this way to constantly update the heat transfer coefficient to the optimum value, the predicted value of the heat transfer coefficient will gradually approach the actual value. The horizontal axis of the graph shown in FIG. 10 is the predicted value of the heat transfer coefficient, and the vertical axis is the actual value of the heat transfer coefficient. Moreover, the solid line is a straight line with an estimation error of 0. In this graph +
The points indicated by marks are the points determined by the conventional method, which treats the heat transfer coefficient as a discretized table value by classifying the steel plate type, and each point gradually deviates from the straight line of 0 for the estimation error. They exist in different positions, and their existence is also discrete. In addition, the points marked with □ are the points obtained by the method of the present invention, which is trained using the backpropagation neural network model, and are located relatively close to the straight line of zero estimation error. exists, and its existence is convergent. Therefore, it can be seen that the prediction accuracy of the heat transfer coefficient is improved by applying the present invention.

[0033]

[Effects of the Invention] As is clear from the above explanation, according to the present invention, the accuracy of predicting the heat transfer coefficient during cooling of a steel plate is improved, and optimal cooling can be performed for the steel plate, thereby reducing temperature defects. The steel sheet material quality after cooling can be stabilized.

[Brief explanation of the drawing]

FIG. 1 is a schematic configuration diagram of an apparatus of the present invention.

FIG. 2 is a schematic diagram of a portion where inlet or outlet temperature sensors 14, 16 are installed. FIG. 2 is a side view of FIG. 1;

FIG. 3 is an internal configuration diagram of the cooling device 12.

FIG. 4 is a specific configuration diagram of the vector calculation device 30.

FIG. 5 is a block diagram of a control system of the device of the present invention.

FIG. 6 is a structural diagram of a backpropagation neural network model that implements the steel plate cooling control method of the present invention.

FIG. 7 is a main flowchart showing the process of calculating a heat transfer coefficient.

FIG. 8 is a subroutine flowchart of the processing in step 3 of the main flowchart.

FIG. 9 is a subroutine flowchart of the process of step 6 of the main flowchart.

FIG. 10 is a graph showing a comparison of predicted values and actual values of heat transfer coefficients between the method of the present invention and the conventional method.

[Explanation of symbols]

10 Steel plate 12 Cooling device 14 Inlet temperature sensor (inlet temperature detection means) 16
Outlet temperature sensor (outlet temperature detection means) 18 Motor 25 Industrial computer 30 Vector calculation device 33 Data accumulator (storage means) 35 Steel plate data input section (input means) 36 Heat transfer coefficient actual value calculation section (actual value calculation means) 37 Heat transfer coefficient learning section (predicted value calculation means, updating means) 38 Cooling time calculation section 39 Water injection valve opening/closing control section 40 Steel plate conveyance speed control section

Claims (3)

[Claims] Claim 1: Calculating the actual value of the heat transfer coefficient of the steel plate based on the temperature of the steel plate detected at the inlet and outlet of the cooling device, and physical elements such as the composition and external shape of the steel plate, and calculating the actual value of the heat transfer coefficient of the steel plate. In the steel plate cooling control method, a predicted value of the heat transfer coefficient for the next steel plate cooling is calculated based on the calculated value, and cooling of the steel plate is controlled based on the predicted value. Based on the detected temperature and physical elements of the steel plate, a neural network model is used to represent the steel plate composition, steel plate outer shape, and steel plate temperature as continuous function values, and the steel plate is cooled based on the function values. Features of steel plate cooling control method. [Claim 2] Calculating the predicted value of the heat transfer coefficient of the steel plate based on a standard heat transfer coefficient expressed as a continuous function value of steel plate components, steel plate outer shape, and steel plate temperature using a neural network model, Calculate the actual value of the heat transfer coefficient of the steel plate based on the temperature of the steel plate detected at the inlet and outlet of the cooling device, as well as physical elements such as the composition and external shape of the steel plate, and calculate the predicted value to the actual value. In order to get closer to 1. A method for controlling cooling of a steel plate, characterized in that the next cooling of the steel plate is performed based on a function value. 3. Inlet temperature detection means for detecting the temperature of the steel plate at the inlet of the cooling device; outlet temperature detection means for detecting the temperature of the steel plate at the outlet of the cooling device; an input means for inputting physical elements; a storage means for storing a standard heat transfer coefficient expressed by a neural network model as a continuous function value of steel sheet composition, steel sheet outer shape, and steel sheet temperature; and said input means. predicted value calculation means for calculating a predicted value of the heat transfer coefficient based on the physical elements of the steel plate inputted from the storage means and the standard heat transfer coefficient in the storage means; and the input side and outlet temperature detection means, respectively. Based on the detected temperature of the steel plate and the physical elements of the steel plate input from the input means,
actual value calculation means for calculating the actual value of the heat transfer coefficient, and a predicted value of the heat transfer coefficient calculated by the prediction calculation means to approach the actual value of the heat transfer coefficient calculated by the actual value calculation means, updating means for calculating and updating new continuous function values of steel plate composition, steel plate outer shape, and steel plate temperature of the standard heat transfer coefficient expressed by the neural network model stored in the storage means; A steel plate cooling control device characterized by: