JP2022000535A - Method for generating coating weight prediction model, method for predicting plating coating weight, method for controlling plating coating weight, method for manufacturing hot-dip metal coated steel sheet, device for performing them and method for generating quality prediction model - Google Patents

Abstract

To provide a technique capable of improving the prediction accuracy of a plating coating weight.SOLUTION: A method for generating a coating weight prediction model comprises: acquiring a plurality of sets of input operation information data including data selected from steel sheet information on a steel sheet 10 to be plated and nozzle operation information data on a wiping nozzle 6, and learning data having actual result data of a plating coating weight using the input operation information data; and including at least a part of the input operation information data in input data by machine learning using the acquired learning data to generate a coating weight prediction model 34 using the plating coating weight as output data. One of the input data obtained by the machine learning including one or more data selected from the steel sheet information and one or more data selected from the nozzle operation information in the input data may include the calculated plating coating weight using a numerical calculation formula on the basis of a physical model for calculating the plating coating weight as the output data.SELECTED DRAWING: Figure 4

Description

The present invention relates to a technique for predicting the quality of manufactured products. The present invention is a technique used in a technique for manufacturing a hot-dip galvanized steel sheet such as a hot-dip galvanized steel sheet.

Conventionally, the prediction of the adhesion amount of a galvanized steel sheet is calculated using a prediction model formula (numerical calculation formula) based on a physical model (for example, Non-Patent Document 1 and Patent Document 1).
Further, in Patent Document 2, various actual data are used as input items, and data with wiping nozzle output (nozzle operation conditions) as an output item are stored in a database, and the required points for the material to be plated are selected. Create a local model (multiple regression model) based on the selected neighborhood data. Then, it is described that the plating adhesion amount is controlled by substituting the required point into the created local model and determining the wiping nozzle output for the steel sheet to be plated.

Examination of gas drawing control mechanism of continuous hot-dip galvanizing: Takeshi Yasutaya et al., Iron and Steel Vol. 66, No. 7, 1980

Japanese Unexamined Patent Publication No. 5-171395 JP-A-2007-262503

When the amount of adhesion is predicted by a prediction model (numerical calculation formula) based on a physical model, the model itself has an error, so that further improvement in prediction accuracy is required.
For example, in the production of a continuous galvanized steel sheet, the plating process of the steel sheet is continuously executed while welding the tail end portion of the leading material and the tip end portion of the trailing material. Then, in the vicinity immediately after the change in the adhesion amount, for example, the learning value of the trailing material with respect to the preceding material is used to correspond immediately after the change, but the larger the prediction error is, the more likely the adhesion amount error is to occur. At this time, it may be possible to solve the problem by creating a detailed preset table that matches the operating conditions. However, this requires a large amount of development time and maintenance in response to changes in operations.

Further, in recent years, for the purpose of beautifying the surface appearance, there is a technique for performing plating adhesion treatment at a wiping nozzle output with a high wiping temperature (a temperature exceeding the melting point of zinc (about 500 ° C.)) and a low discharge pressure. .. The physical model formula corresponding to this technique has not been established, and it is being tried to obtain the predicted value of the adhesion amount by changing a part of the constants for high temperature based on the conventional physical model.

The present invention has been made by paying attention to the above points, and a technique capable of improving the prediction accuracy of the plating adhesion amount even in the production of a continuous hot-dip galvanized steel sheet having a high temperature and low pressure wiping output. The purpose is to provide.

In order to solve the problem, one aspect of the present invention is hot-dip plating provided with an adhesion adjustment step of injecting a wiping gas from a wiping nozzle onto the surface of a steel plate to be plated after being immersed in a plating bath to adjust the amount of plating adhesion. It is a method of generating an adhesion amount prediction model for predicting the plating adhesion amount after the adhesion adjustment step in a steel plate manufacturing facility, and one or more data selected from the steel plate information regarding the steel plate to be plated and a nozzle operation related to a wiping nozzle. Acquire a plurality of sets of learning data having input operation information data including one or more data selected from the information and actual data of the plating adhesion amount after the adhesion adjustment step using the input operation information data. The gist is to generate an adhesion amount prediction model that includes at least a part of the above input operation information data in the input data and uses the plating adhesion amount as output data by machine learning using the acquired learning data. do.
As one of the input data by the machine learning, one or more data selected from the steel plate information and one or more data selected from the nozzle operation information are included in the input data, and the plating adhesion amount is calculated as output data. It may have a plating adhesion amount calculated by using a numerical calculation formula based on a physical model.

Further, an aspect of the present invention is in a hot-dip plated steel sheet manufacturing facility provided with an adhesion adjusting step of injecting a wiping gas from a wiping nozzle onto the surface of a steel sheet to be plated after being immersed in a plating bath to adjust the amount of plating adhesion. It is a method for predicting the amount of plating adhesion after the above-mentioned adhesion adjustment step, and predicts the amount of plating adhesion by using the adhesion amount prediction model generated by the method for generating the adhesion amount prediction model of the above-described embodiment. The gist is that.

Further, an aspect of the present invention is in a hot-dip plated steel plate manufacturing facility provided with an adhesion adjusting step of injecting a wiping gas from a wiping nozzle onto the surface of the steel plate to be plated after being immersed in a plating bath to adjust the amount of plating adhesion. It is a plating adhesion amount control method that controls the plating adhesion amount after the adhesion adjustment step. In the production of a hot-dip plated steel plate, the tail end of the preceding material and the tip of the following material are connected by welding and covered. Before the plated steel plate is continuously conveyed and the adhesion adjusting step for the trailing material is executed, the set operation information set for the trailing material including at least a part of the steel plate information of the trailing material and the adhesion of the above-mentioned embodiment are performed. Using the adhesion amount prediction model generated by the amount prediction model generation method, the prediction value calculation process for calculating the prediction value of the plating adhesion amount to the trailing material, the calculated adhesion amount prediction value, and the preset adhesion The gist is to execute the resetting step of resetting the operation information regarding the wiping nozzle for the trailing material in the direction in which the deviation becomes smaller based on the deviation from the target value.

Further, an aspect of the present invention is a method for manufacturing a hot-dip galvanized steel sheet, which comprises an adhesion adjusting step of injecting a wiping gas from a wiping nozzle onto the surface of a steel sheet to be plated after being immersed in a plating bath to adjust the amount of plating adhesion. Therefore, the gist is to adjust the plating adhesion amount by the adhesion amount control method of the above aspect.

Further, the aspect of the invention is selected from a plurality of qualities required for the manufactured product in the processing process of processing the workpiece into a product by executing the operation of the manufacturing apparatus with the data of the set operating conditions. It is a method of generating a quality prediction model for predicting the selected quality, which is the selected quality, and is input operation information including one or more data selected from the information on the workpiece and one or more data selected from the above operating conditions. A plurality of sets of learning data having the data of the above and the actual data regarding the selection quality of the product after the processing process using the input operation information data are acquired, and the machine learning using the acquired learning data. As a result, a quality prediction model is generated in which at least a part of the data of the input operation information is included in the input data and the data related to the selected quality is used as the output data, and one or more data selected from the information related to the workpiece and the above are described. The gist is to include one or more data selected from the operating conditions in the input data.
The data related to the selected quality calculated by using the numerical calculation formula based on the physical model for calculating the data related to the selected quality as the output data may be used as one of the input data by the machine learning.

According to the aspect of the present invention, it is possible to improve the prediction accuracy of the plating adhesion amount even in the production of a continuous hot-dip galvanized steel sheet in which the temperature of the wiping gas is set to a high temperature equal to or higher than the melting point of the plating solution. .. As a result, according to the aspect of the present invention, it is possible to manufacture a high quality continuously hot-dip galvanized steel sheet.
Here, when the calculation result of the physical model formula is adopted as one explanatory variable (input data) of the machine learning model for predicting the amount of hot-dip plating adhesion (selective quality), the machine learning is regarded as a complete black box calculation. Instead, it is possible to generate a prediction model by machine learning including physical grounds.

In addition, when the calculation result of the physical model formula is adopted as one explanatory variable of the machine learning model, even if the physical model itself has an error, it is possible to obtain a machine learning model with the error reduced. Become. Since the generation of the machine learning model is an offline process, it is possible to set and use a detailed physical model that takes time.
Further, according to the aspect of the present invention, for example, it is possible to form a machine learning model while analyzing what kind of component is insufficient in the physical model formula, and the physical model formula also has a certain degree of accuracy. Therefore, other explanatory variables can be used as corrections, and as a result, improvement in prediction accuracy can be expected.
In addition, when the warp component of the galvanized steel sheet and the wiping gas temperature are included in the explanatory variables, it becomes possible to incorporate components that are likely to be related to the amount of adhesion, although it cannot be expressed by the conventional physical model.

It is a figure explaining the manufacturing equipment of the continuous hot-dip galvanized steel sheet which concerns on embodiment based on this invention. It is a schematic diagram explaining the structure around a nozzle. It is a figure which shows the structure of a plating control device. It is a figure which shows the structure of the plating control apparatus which concerns on embodiment based on this invention. It is a figure explaining the structure of the reset part. It is a figure which shows the correlation of the adhesion amount in the case of using the comparative example 1. FIG. It is a figure which shows the correlation of the adhesion amount when the comparative example 2 is used. It is a figure which shows the correlation of the adhesion amount in the case of using an Example.

Next, an embodiment of the present invention will be described with reference to the drawings.
In the present embodiment, a continuous hot-dip galvanized steel sheet manufacturing facility is taken as an example as a manufacturing facility having a processing process for processing a workpiece into a product by executing the operation of the manufacturing apparatus with the data of the set operating conditions. The present invention is also applicable to a manufacturing facility using a manufacturing apparatus for processing a product other than a hot-dip galvanized steel sheet, as long as the process of processing the workpiece into a product can be expressed by a physical model. The present invention is a technique suitable for manufacturing a hot-dip galvanized steel sheet.
Further, in the present embodiment, hot-dip galvanizing will be described as an example of hot-dip plating.

(composition)
FIG. 1 is a diagram showing a configuration example around an apparatus for executing a process of an adhesion adjusting step in a continuous galvanized steel sheet manufacturing facility. The adhesion adjusting step (processing step) is a step of executing a processing process of injecting a wiping gas from the wiping nozzle 6 onto the surface of the steel sheet 10 to be plated after being immersed in the plating bath to adjust the amount of plating adhesion.
As shown in FIG. 1, the continuous galvanized steel sheet manufacturing equipment includes a welding device 1, a quenching furnace 3, a plating bath 4 accommodating a plating bath 5, a wiping nozzle 6, and a plating adhesion meter 7. Further, an alloying furnace and a cooling zone are arranged between the wiping nozzle 6 and the plating adhesion amount meter 7 along the pass line for transporting the steel plate.

The welding device 1 is a device for welding the tail end portion of the leading material (leading coil) and the tip end portion of the trailing material (following coil). By sequentially connecting the materials in the welding apparatus 1, the steel plate 10 to be plated is continuously formed, and the steel plate 10 to be plated can be continuously conveyed to the adhesion adjusting step.
At the time of welding, the transfer of the positions of the tail end of the leading material and the tip of the trailing material is temporarily stopped, but the stop is absorbed by the looper device and heat treatment or plating adheres to the steel sheet. Etc. can be continuously executed.

The steel sheet 10 to be plated is continuously overheated in the baking furnace 3 and then immersed in the plating bath 5, so that the plating adheres to the surface of the steel sheet 10 to be plated to form a plating layer. Subsequently, the steel sheet 10 to be plated is conveyed above the surface of the molten metal, and the plating adhesion amount (thickness of the plating layer) is controlled to the target adhesion amount by the wiping gas from the wiping nozzle 6 above the bathtub 4. To. After that, the steel sheet 10 to be plated is heated in an alloying furnace to be alloyed, and then the cooling treatment is executed in the cooling zone to fix the plating, and then the steel sheet is plated with the plating adhesion meter 7. The amount of adhesion is measured.

As shown in FIG. 2, the wiping nozzle 6 is arranged at a position facing both sides of the steel plate 10 to be plated whose transport direction is changed upward by the roll 4a in the bathtub 4. The wiping nozzle 6 can inject the wiping gas toward the surface of the steel sheet.
Further, as a driving unit of the wiping nozzle 6, a height adjusting actuator 20, a nozzle spacing adjusting actuator 21 for adjusting the distance to the steel plate, and a nozzle angle adjusting actuator 22 are provided.

The height adjusting actuator 20 is a device that moves the wiping nozzle 6 in the vertical direction to adjust the height of separation from the upper surface of the plating bath 5. The nozzle spacing adjusting actuator 21 is a device that adjusts the amount of separation from the steel plate by moving the wiping nozzle 6 in the lateral direction to approach and separate from the steel plate. The height adjusting actuator 20 and the nozzle spacing adjusting actuator 21 may be composed of one actuator. The nozzle angle adjusting actuator 22 is an actuator that adjusts the inclination of the injection angle with respect to the direction orthogonal to the steel plate surface by changing the angle (injection angle) in the injection direction with respect to the steel plate surface in the wiping nozzle 6 in the vertical direction. ..

A pressure gauge 23 and a thermometer 24 are interposed in the supply pipe for supplying gas to the wiping nozzle 6, and a gas pressure adjusting device 25 for controlling the pressure of the gas supplied through the supply pipe. Has. It also has a heating device that adjusts the gas temperature to the target temperature.
Here, in the present embodiment, it is assumed that the wiping temperature is controlled to a high temperature (a temperature exceeding the melting point of zinc (about 500 ° C.)).

(Plating control device 30)
Further, the continuous hot-dip galvanized steel sheet manufacturing equipment of the present embodiment has a plating control device 30.
The plating control device 30 executes plating adhesion control on continuously conveyed steel sheets.
As shown in FIG. 3, the plating control device 30 of the present embodiment includes an adhesion amount prediction model generation device 31, a plating adhesion amount prediction device 32, and a plating adhesion amount control device 33.

(Attachment amount prediction model generation device 31)
The generator 31 of the adhesion amount prediction model is a hot-dip plated steel sheet provided with an adhesion adjustment step of injecting a wiping gas from the wiping nozzle 6 onto the surface of the steel sheet 10 to be plated after being immersed in the plating bath to adjust the plating adhesion amount. This is a device for generating an adhesion amount prediction model 34 for predicting the plating adhesion amount after the adhesion adjustment step in a manufacturing facility.
As shown in FIG. 4, the adhesion amount prediction model generation device 31 includes a learning data acquisition unit 31A, an adhesion amount calculation unit 31B, and a prediction model generation unit 31C.

<Learning data collection unit 31A>
The learning data collection unit 31A includes input operation information data including one or more data selected from the steel plate information regarding the steel plate to be plated 10 and one or more data selected from the nozzle operation information regarding the wiping nozzle 6, and the input operation information thereof. A process of acquiring a plurality of sets of learning data having the actual data of the plating adhesion amount after the adhesion adjustment step using the data of the above is executed. The acquired learning data is sequentially stored in the database 36 (learning storage unit).
That is, the learning data collection unit 31A uses the operation information actually used as the input operation information data for each steel material, and the actual data of the plating adhesion amount in the operation (the adhesion amount measured by the plating adhesion amount meter 7). ) And is used as learning data. The measurement with the plating adhesion amount meter 7 is performed a plurality of times, and the average value is used.

The input operation information is set based on the setting information supplied from the process computer which is a higher-level computer. The setting information includes the target zinc adhesion amount.
Here, the steel plate information includes the plate width, plate thickness, and steel type of the steel plate to be plated 10.
The nozzle operation information includes the nozzle spacing, the injection angle with respect to the steel plate surface, the height from the plating bath, the pressure of the wiping gas, and the temperature of the wiping gas in the wiping nozzle 6. The nozzle spacing can be considered as the distance between two nozzles facing each other across the steel plate 10, but in the present application, it represents the distance from the nozzle to the surface of the steel plate.
Further, in the present embodiment, the line speed and the temperature of the plating bath are included as the input operation information.
The learning data in the database 36 is sequentially accumulated and updated.

Here, in order to improve the accuracy of the prediction model, the training data may be executed in consideration of processing such as data cleansing.
An example will be described.
(1) Nozzle spacing data The nozzle spacing is the distance from the nozzle to the surface of the steel sheet as described above. Originally, measurement is possible by installing a rangefinder to the steel plate. However, in the process targeted by the present application, it may be difficult to permanently install a rangefinder because the temperature around the wiping nozzle is high (atmospheric temperature close to the melting point of zinc). In such a case, the distance between the nozzles facing each other across the steel plate 10 can be used. For example, the nozzle spacing may be obtained by subtracting (thickness of the steel plate / 2) from 1/2 of the distance between the opposing nozzles as the distance between the nozzle and the steel plate. Then, the value is used for physical model calculation and explanatory variables.
In this case, it is assumed that the pass line of the steel plate (the position where the steel plate passes) is the center between the nozzles, but the position of the steel plate may be biased to one nozzle side. If the data at that time is used as it is as the learning data of the adhesion amount prediction model, the prediction accuracy of the adhesion amount prediction model deteriorates.

When the pass line of the steel sheet is biased to one of the front and back surfaces of the steel sheet, there is a difference in the amount of adhesion between the front and back surfaces of the steel sheet. Therefore, the difference in the amount of adhesion between the front and back of the steel sheet at the same position in the longitudinal direction is evaluated, and the data is excluded when the difference exceeds a predetermined range (for example, when there is a difference of 3% or 8%). It may be used as data. Alternatively, only the data set when the zinc adhesion ratio on the front and back of the steel sheet is 0.9 to 1.1 may be used as the learning data. Further, the above-mentioned predetermined range and the range of the zinc adhesion amount ratio on the front and back of the data set may be determined in consideration of the prediction error of the adhesion amount prediction model.

Alternatively, the nozzle spacing on the front and back of the steel sheet may be corrected and stored according to the difference in the amount of adhesion between the front and back. In this case, a function for calculating the correction amount of the nozzle interval according to the adhesion amount may be constructed by collecting the measured value of the nozzle interval and the actual value of the adhesion amount, and used.
This makes it possible to construct an accurate learning model even if the path line of the steel sheet fluctuates.

(2) How to save the learning data The predicted adhesion amount is distributed in the range of ^{about 20 mg / m 2 to} 300 g / m ^{2 depending on the steel type and the size of the steel plate.} If the amount of adhesion of the training data is biased, the prediction accuracy of the amount of attachment with a large number of data is high, and the prediction accuracy of the amount of attachment with a small number of data deteriorates. Therefore, in advance predetermined amount deposited amount (e.g., a suitable value in the range of ^{50g / m 2 ~100g / m 2} ) is divided for each, the number of data of the divided ranges are set to be more than a predetermined number of data The training data may be saved and the prediction model may be trained. Here, the predetermined number of data may be about 2000 sets, for example, as long as it can be learned to some extent.
By dividing and storing the learning data in this way and securing the amount of data in each division, it is possible to maintain the accuracy of the prediction model for the overall amount of adhesion. In addition, by holding such data, it is possible to build a model for each divided division.

(3) Others In addition, when comparing the zinc adhesion amount prediction calculation result by the physical model and the zinc adhesion amount actual result, the difference tends to be remarkably different depending on the nozzle-steel plate distance and the gas pressure. Therefore, if it is determined that it is difficult to predict the amount of adhesion of all operation patterns by one machine learning model, multiple machine learning models are constructed according to the operation conditions such as gas pressure and nozzle spacing, and which model is selected from the operation conditions. You may choose whether to use it.

<Database 36>
The database 36 contains input operation information including one or more data selected from the steel plate information regarding the steel plate to be plated 10 and one or more data selected from the nozzle operation information regarding the wiping nozzle 6 as actual data in the manufacture of the hot-dip plated steel plate. , A plurality of sets of learning data having the actual data of the plating adhesion amount measured by the plating adhesion amount meter 7 after the adhesion adjustment step using the input operation information data are stored.
At that time, the storage amount of the data is adjusted according to the above-mentioned data cleansing and the adhesion amount classification.
In addition, the generated adhesion amount prediction model 34 (model formula) is also stored in the database 36.

<Adhesion amount calculation unit 31B>
The adhesion amount calculation unit 31B is based on each learning data stored in the database 36, and for each learning data for each data, one or more data selected from the steel plate information and one or more data selected from the nozzle operation information. Is included in the input data, and the plating adhesion amount (selection quality) is calculated using a numerical calculation formula based on a physical model for calculating the plating adhesion amount as output data. The adhesion amount calculation unit 31B stores the calculated plating adhesion amount (also referred to as the calculated adhesion amount) in the database 36 as one data of the input operation information in the corresponding learning data. As a result, each learning data is updated to a state including the calculated adhesion amount.

As the numerical calculation formula based on the physical model, for example, known model formulas described in Patent Documents 1 and 2, Non-Patent Document 1 and the like may be adopted.
An example of the model formula is shown in the following formula (1).
W = f (LS, P, D, t) ... (1)
here,
LS: Line speed P: Wiping gas pressure D: Nozzle interval t: Steel plate thickness W: Expected adhesion value by physical model.

<Prediction model generation unit 31C>
The prediction model generation unit 31C includes at least a part of the input operation information data in the input data by machine learning using a plurality of sets of learning data processed by the adhesion amount calculation unit 31B, and outputs the plating adhesion amount. The adhesion amount prediction model 34 is generated.
Here, in the present embodiment, the input data includes at least the nozzle spacing, the plate thickness, the line speed, the pressure of the wiping gas, and the temperature of the wiping gas.
Further, it is preferable that the input data includes at least one of the nozzle angle (injection angle with respect to the steel plate surface), the steel type, the plate width, the temperature of the plating bath, and the height of the nozzle from the plating bath.

In the present embodiment, as described above, as one of the input data by machine learning, the calculated adhesion amount, which is the estimated adhesion value calculated by the adhesion amount calculation unit 31B, is used as the input data for machine learning.
In the present embodiment, the input data in the adhesion amount calculation unit 31B is included as the input data for machine learning. However, the configuration may include only a part of the input data of the adhesion amount calculation unit 31B as the input data for machine learning.
Further, the input may not include the output of the physical model.

Here, the physical model may be used by constructing a model including other than the inputs described here. For example, a model that considers the high temperature of the wiping gas may be used.
Further, since the processing of the adhesion amount calculation unit 31B and the prediction model generation unit 31C can be executed offline, the numerical calculation formula based on the physical model has a margin in processing time. Therefore, it is possible to set the physical model as a more detailed model and spend more time on the calculation.

(Plating adhesion prediction device 32)
The plating adhesion amount prediction device 32 is a hot-dip plated steel sheet provided with an adhesion adjustment step of injecting a wiping gas from a wiping nozzle 6 onto the surface of the steel sheet 10 to be plated after being immersed in the plating bath 4 to adjust the plating adhesion amount. It is a device that predicts the amount of plating adhesion after the adhesion adjustment process in the manufacturing equipment.
The plating adhesion amount prediction device 32 of the present embodiment includes an adhesion amount prediction unit 32A. The adhesion amount prediction unit 32A predicts the plating adhesion amount by using the adhesion amount prediction model 34 generated by the generation device 31 of the adhesion amount prediction model.

The plating adhesion amount prediction device 32 of the present embodiment inputs, for example, the input operation information in the setting information supplied from the process computer 40 to the trailing material into the adhesion amount prediction model 34 as the operation setting information. Calculate the predicted plating adhesion amount when operating using the set setting information.
In the present embodiment, as shown in FIG. 4, the plating adhesion amount prediction device 32 constitutes a part of the plating adhesion amount control device 33.

(Plating adhesion amount control device 33)
The plating adhesion control device 33 manufactures a hot-dip plated steel sheet having an adhesion adjustment step of injecting a wiping gas from a wiping nozzle 6 onto the surface of the steel sheet 10 to be plated after being immersed in the plating bath 5 to adjust the plating adhesion amount. It is a device that controls the amount of plating adhesion after the adhesion adjustment process in the equipment.
In the present embodiment, as described above, in the production of the hot-dip galvanized steel sheet, the tail end portion of the preceding material and the tip end portion of the trailing material are connected by welding, and the steel sheet to be plated 10 is continuously conveyed and plated. Processing is executed continuously. Further, the control is executed for each steel material (preceding material and trailing material).
The plating adhesion amount control device 33 of the present embodiment includes a predicted value calculation unit 32, a resetting unit 37, a nozzle operation unit 38, and a gas control unit 39.

<Predicted value calculation unit 32>
The predicted value calculation unit 32 includes at least a part of the steel plate information of the trailing material before the adhesion adjusting step for the trailing material is executed, that is, before the plating process of each steel material is executed. Using the set operation information set in the above and the adhesion amount prediction model 34 generated by the adhesion amount prediction model generation device 31, a process of calculating the predicted value of the plating adhesion amount to the trailing material is executed.
The prediction value calculation unit of the present embodiment is composed of a plating adhesion amount prediction device 32.

<Resetting unit 37>
The resetting unit 37 has a small deviation based on the deviation between the predicted value of the adhesion amount calculated by the predicted value calculation unit 32 and the preset target value for adhesion before executing the adhesion adjusting step for the trailing material. In this direction, the process of resetting the operation information regarding the wiping nozzle 6 for the trailing material is executed.
In the present embodiment, the operation information regarding the wiping nozzle 6 to be reset is the position of the wiping nozzle 6. In the present embodiment, a case where the distance between the tip of the wiping nozzle 6 and the surface of the steel plate is reset will be described as an example. The position of the wiping nozzle 6 to be reset may be the height of the wiping nozzle 6 or the injection angle. Further, the operation information to be reset may be two or more operation information.
As shown in FIG. 5, the resetting unit 37 of the present embodiment includes a deviation amount calculation unit 37A, an influence coefficient calculation unit 37B, and a change amount calculation unit 37C.

<Deviation amount calculation unit 37A>
The deviation amount calculation unit 37A calculates the deviation between the adhesion amount target value and the predicted value obtained by the predicted value calculation unit 32.
If the deviation calculated by the deviation amount calculation unit 37A is equal to or less than the preset threshold value, the processing of the resetting unit 37 is terminated and the process proceeds to the next processing unit.

<Impact coefficient calculation unit 37B>
The impact coefficient calculation unit 37B uses the adhesion amount prediction model 34 generated by the generation device 31 of the adhesion amount prediction model to obtain the coefficient of the change in the operation information regarding the wiping nozzle 6 with respect to the change in the plating adhesion amount in the trailing material. Calculate a certain impact factor.
The impact coefficient calculation unit 37B of the present embodiment calculates the impact coefficient based on the change in the plating adhesion amount predicted by the adhesion amount prediction model 34 by changing the set operation information set in the trailing material.
The influence coefficient calculation unit 37B calculates the influence coefficient based on the perturbation of the manufacturing specifications including at least the position of the wiping nozzle 6 in the vicinity of the specifications of the setting conditions for manufacturing the trailing material which is the manufacturing target material.

The influence coefficient Kwd is expressed by, for example, the following equation (2).
Kwd = δW / δD ・ ・ ・ (2)
δD is the perturbation amount of the manufacturing specifications.
δW is the amount of change in the amount of adhesion with respect to the perturbation of the manufacturing specifications.
δD is determined, for example, as follows.

The operation amount of the wiping nozzle 6 set for the trailing material to be controlled (set value input from the process computer 40) or the operation amount under the operating conditions of the wiping nozzle 6 at the time of the current operation is used as the reference operation amount. do. Next, the operation amount in which the reference operation amount is slightly shifted, that is, the perturbation operation amount is defined as the perturbation operation amount. The operation information regarding the wiping nozzle 6 that changes the operation amount may be one variable or two or more types.
In the present embodiment, the operation information regarding the wiping nozzle 6 for changing the separation position of the nozzle 6 with respect to the steel plate is used. In this case, the amount of change is set in the range of, for example, 2 mm to 6 mm.
Then, let δD be the above-mentioned change amount (= reference operation amount-perturbation operation amount).

Further, δW is set as follows.
First, the predicted value (first predicted value) of the adhesion amount is obtained by the adhesion amount prediction model 34 using the reference operation amount as input information. When the reference predicted value is the operation amount of the wiping nozzle 6 set for the trailing material to be controlled, the first predicted value is a value calculated by the predicted value calculation unit 32.
Next, the predicted value (second predicted value) of the adhesion amount is obtained by the adhesion amount prediction model 34 using the perturbation operation amount (the operation amount in which a part of the reference operation is changed) as input information.
Then, the difference between the first predicted value and the second predicted value is defined as δW.
This influence coefficient Kwd is, for example, a coefficient indicating the amount of change in the amount of adhesion with respect to the amount of unit change (for example, 1 mm) of the nozzle 6 position.

<Change amount calculation unit 37C>
The change amount calculation unit 37C calculates the change amount of the operation information regarding the wiping nozzle 6 with respect to the trailing material from the calculated influence coefficient Kwd and the deviation.
The change amount calculation unit 37C of the present embodiment divides the deviation amount obtained by the deviation amount calculation unit 37A by the influence coefficient Kwd obtained by the influence coefficient calculation unit 37B, that is, obtains the value of (deviation amount / Kwd). The value is used as the correction amount for the position of the nozzle 6.
The change amount calculation unit 37C updates the set operating conditions according to the correction amount of the obtained position of the nozzle 6.

<Nozzle operation unit 38>
When the nozzle operation unit 38 determines that the adhesion amount control of the steel material (following material) to be controlled is executed, the change amount calculation unit 37C supplies each command of the nozzle 6 to each actuator for the updated operating condition.
For example, when the welding points of the leading material and the trailing material pass through the plating bath 5, the process of changing the nozzle position to the target nozzle position is executed.

<Gas control unit 39>
The gas control unit 39 measures the pressure and temperature of the gas supplied to the nozzle 6 and adjusts the pressure and temperature of the gas supplied to the nozzle 6 so as to be a preset pressure and temperature.

(Operation and others)
In this embodiment, machine learning is performed using learning data consisting of past operation record data, and when the adhesion amount prediction model 34 for predicting the plating adhesion amount is generated, it is obtained by a physical model based on the operation record data. The predicted value of the adhesion amount is adopted as one of the input data of machine learning.
As a result, the adhesion amount prediction model 34 of the present embodiment improves the adhesion amount prediction accuracy.
That is, in this embodiment, the calculation result of the physical model formula is adopted as one explanatory variable of the machine learning model for predicting the zinc adhesion amount. For this reason, instead of making machine learning a complete black box calculation, it is possible to build a machine learning model while analyzing what components are lacking in the physical model formula, including the physical basis. ..

Further, since the physical model formula also has a certain degree of accuracy, other explanatory variables can be used as corrections, and as a result, the accuracy of the generated adhesion amount prediction model 34 can be expected to improve.
Further, for example, by including the warp component of the galvanized steel sheet and the wiping gas temperature in the explanatory variables (input data of the adhesion amount prediction model 34), components that cannot be expressed by the conventional physical model but are likely to be related to the adhesion amount can be obtained. Can be captured. From this point as well, the adhesion amount prediction model 34 improves the adhesion amount prediction accuracy.

As described above, in the present embodiment, by adopting the calculation result of the physical calculation model formula as one explanatory variable of the machine learning model, the other explanatory variables are corrected in the form of correcting the error of the calculation result of the physical calculation model formula. Since machine learning can be performed using the above, it is possible to predict the amount of adhesion with high accuracy.

(Modification example)
In the above embodiment, the resetting unit 37 automatically updates the nozzle operation information, but the present invention is not limited to this.
For example, the adhesion amount of the next coil (following material) is calculated in the same manner as above, and the adhesion amount prediction value obtained by the adhesion amount prediction unit 32A is displayed. Then, based on this, the operator may operate the nozzle 6 to improve the adhesion amount disengagement. At this time, for example, a table showing the relationship between the nozzle pressure / nozzle spacing and the adhesion amount calculated using a machine learning model is used as a guideline for the operation amount.

(Example)
In this embodiment, the above equation is adopted as the physical model, and the adhesion amount prediction model 34 is generated by machine learning using a neural network. The plating was hot dip galvanizing.
In the embodiment, the input data of machine learning are the calculation result by the physical model, the nozzle wiping gas pressure, the wiping gas temperature, the nozzle height, the nozzle spacing, the steel grade, the plate thickness, the line speed, and the plating temperature in the plating bath 4. , The output data was the amount of zinc adhered.
The steel type is a kind of code categorized according to the composition of the steel sheet to be manufactured and the difference in the manufacturing method, and is treated like a label.
Further, the temperature of the nozzle gas was operated at a temperature of about 500 degrees, which exceeds the melting point of zinc. For temperature detection, the directly measured temperature or the set temperature of the gas temperature adjustment system can be used.

<Comparative Example 1>
In Comparative Example 1, the predicted adhesion amount was obtained by a calculation formula based on a physical model.
FIG. 6 shows the correlation between the predicted adhesion amount and the actual adhesion amount in Comparative Example 1.
In Comparative Example 1, RMSE (Root Mean Square Error) was 21%.

<Example 1>
Next, in the first embodiment, the adhesion amount prediction model is generated by performing machine learning using only the raw operation performance data as input data (explanatory variable) without using the calculation result by the physical model at the time of machine learning. , The predicted adhesion amount was obtained by the adhesion amount prediction model.
FIG. 7 shows the correlation between the predicted adhesion amount and the actual adhesion amount in Example 1.
In Example 1, the RMSE (Root Mean Square Error) was 13%.

<Example 2>
On the other hand, in the second embodiment, as one of the explanatory variables (input data) of the machine learning in the first embodiment, the calculation result by the physical model is further added to generate the adhesion amount prediction model 34, and the adhesion amount thereof. The predicted adhesion amount was obtained by the prediction model 34.
FIG. 8 shows the correlation between the predicted adhesion amount and the actual adhesion amount in Example 2.
In Example 2, the RMSE (Root Mean Square Error) was 10%.

<Evaluation>
From the comparison between Comparative Example 1 and Example 1, it was found that the error was smaller when the prediction model by machine learning was used than the predicted value of the zinc adhesion amount in the physical model.
In this example, since the temperature of the wiping gas exceeds the melting point of zinc, the density, viscosity, and flow velocity of the wiping gas may change, which may affect the amount of adhesion. Is not easy. In this regard, in Example 1, it is considered that the accuracy was improved because these effects could be effectively expressed by machine learning.

Further, from Example 1 and Example 2, it was found that the error is further reduced by including the calculation result based on the physical model as the input data of the machine learning.
As described above, it can be seen that the prediction accuracy is improved based on the present invention.
Here, by including the calculation result based on the physical model as the input data of machine learning, the operating conditions change drastically because the model based on the physical phenomenon is used, and even if there is not enough data, it is reasonable. There is a possibility that the operation can be carried out, and it is expected that it can be used even when operating special materials. Further, since the physical model makes a reasonable prediction and the machine learning corrects the predicted value based on the data for the knowledge not included in the physical model, the accuracy of the second embodiment is further improved as compared with the first embodiment. Conceivable.

(effect)
This embodiment has the following effects.
(1) The present embodiment is a hot-dip plated steel sheet manufacturing facility provided with an adhesion adjusting step of injecting wiping gas from a wiping nozzle 6 onto the surface of the steel plate to be plated 10 after being immersed in a plating bath to adjust the amount of plating adhesion. In the method for generating an adhesion amount prediction model for predicting the plating adhesion amount after the adhesion adjustment step, one or more data selected from the steel plate information regarding the steel plate to be plated 10 and the nozzle operation information regarding the wiping nozzle 6 are used. A plurality of sets of learning data having input operation information data including one or more selected data and actual data of the plating adhesion amount after the adhesion adjustment step using the input operation information data are acquired. By machine learning using the acquired learning data, an adhesion amount prediction model 34 is generated in which at least a part of the input operation information data is included in the input data and the plating adhesion amount is used as output data.

The apparatus is, for example, in a hot-dip plated steel sheet manufacturing facility provided with an adhesion adjusting step of injecting wiping gas from a wiping nozzle 6 onto the surface of a steel plate to be plated 10 after being immersed in a plating bath to adjust the amount of plating adhesion. An adhesion amount prediction model generation device 31 for predicting the plating adhesion amount after the adhesion adjustment step, which is selected from one or more data selected from the steel plate information regarding the steel plate to be plated 10 and nozzle operation information regarding the wiping nozzle 6. A plurality of sets of learning data having input operation information data including one or more data and actual data of the plating adhesion amount after the adhesion adjustment step using the input operation information data are stored for learning. By machine learning using the data storage unit 35 and a plurality of sets of learning data stored in the learning data storage unit 35, at least a part of the input operation information data is included in the input data, and the plating adhesion amount is calculated. The configuration includes a prediction model generation unit 31C that generates an adhesion amount prediction model 34 as output data.

The steel sheet information includes, for example, the plate width, plate thickness, and steel type of the steel plate 10 to be plated.
The nozzle operation information includes, for example, the nozzle spacing, the injection angle with respect to the steel plate surface, the height from the plating bath, the pressure of the wiping gas, and the temperature of the wiping gas in the wiping nozzle 6.
Further, the input operation information includes, for example, the line speed and the temperature of the plating bath.
According to this configuration, even in the production of a continuous hot-dip galvanized steel sheet in which the temperature of the wiping gas is set to a high temperature equal to or higher than the melting point of the plating solution, it is possible to improve the prediction accuracy of the plating adhesion amount. As a result, according to the present embodiment, it is possible to manufacture a high-quality continuous hot-dip galvanized steel sheet.

(2) In the present embodiment, as one of the input data by the machine learning, one or more data selected from the steel plate information and one or more data selected from the nozzle operation information are included in the input data, and the plating adhesion amount is included. Has a plating adhesion amount calculated using a numerical calculation formula based on a physical model for calculating as output data.
The device includes, for example, one or more data selected from the steel plate information and one or more data selected from the nozzle operation information in the input data, and a numerical value based on a physical model for calculating the plating adhesion amount as output data. It is provided with an adhesion amount calculation unit 31B for calculating the plating adhesion amount using a calculation formula, and the plating adhesion amount calculated by the adhesion amount calculation unit 31B is used as one of the input data by the machine learning. ..
According to this configuration, in the present embodiment, the calculation result of the physical model formula is adopted as one explanatory variable of the machine learning model for predicting the adhesion amount of hot-dip plating. As a result, it becomes possible to generate a prediction model by machine learning including physical grounds, instead of making machine learning a complete black box calculation.

Further, by adopting the calculation result of the physical model formula as one explanatory variable of the machine learning model, even if the physical model itself has an error, it is possible to obtain a machine learning model in which the error is reduced. .. Since the generation of the machine learning model is an offline process, it is possible to set and use a detailed physical model that takes time.
Further, according to the present embodiment, for example, it is possible to build a machine learning model while analyzing what kind of components are insufficient in the physical model formula, and the physical model formula also has a certain degree of accuracy. Therefore, other explanatory variables can be used as corrections, and as a result, improvement in prediction accuracy can be expected.

In addition, since the model based on physical phenomena is used, the operating conditions may change drastically, and even if there is not enough data, it may be possible to carry out the operation as it is, and it can be used even when operating special materials. Be expected. In addition, the physical model makes reasonable predictions, and machine learning corrects the predicted values based on the data for knowledge that is not included in the physical model, so it is considered that the accuracy has been further improved.
In addition, by including the warp component of the galvanized steel sheet and the wiping gas temperature as explanatory variables, it becomes possible to incorporate components that are likely to be related to the amount of adhesion, which cannot be expressed by the conventional physical model.

(3) The present embodiment is a hot-dip plated steel sheet manufacturing facility including a bonding adjustment step of injecting a wiping gas from a wiping nozzle 6 onto the surface of a steel sheet 10 to be plated after being immersed in a plating bath to adjust the amount of plating adhesion. Is a method for predicting the amount of plating adhesion after the above-mentioned adhesion adjustment step, which is a method for predicting the amount of plating adhesion, using the adhesion amount prediction model 34 generated by the method for generating the adhesion amount prediction model of the present embodiment. Predict the amount.

The apparatus is, for example, in a hot-dip plated steel sheet manufacturing facility provided with an adhesion adjusting step of injecting a wiping gas from a wiping nozzle 6 onto the surface of a steel sheet 10 to be plated after being immersed in a plating bath to adjust the amount of plating adhesion. The plating adhesion amount is predicted by using the adhesion amount prediction model 34 generated by the generation device 31 of the adhesion amount prediction model of the embodiment, which is the plating adhesion amount prediction device 32 for predicting the plating adhesion amount after the adhesion adjustment step. The adhesion amount prediction unit 32A for predicting the above is provided.
According to this configuration, even in the production of a continuous hot-dip galvanized steel sheet in which the temperature of the wiping gas is set to a high temperature equal to or higher than the melting point of the plating solution, it is possible to improve the prediction accuracy of the plating adhesion amount. As a result, according to the aspect of the present invention, it is possible to manufacture a high quality continuously hot-dip galvanized steel sheet.

(4) The present embodiment is a hot-dip plated steel plate manufacturing facility including a bonding adjustment step of injecting a wiping gas from a wiping nozzle 6 onto the surface of the steel plate to be plated 10 after being immersed in a plating bath to adjust the plating adhesion amount. In the above-mentioned plating adhesion amount control method for controlling the plating adhesion amount after the adhesion adjustment step, in the production of a hot-dip plated steel plate, the tail end portion of the leading material and the tip end portion of the trailing material are connected by welding. The set operation information set for the trailing material including at least a part of the steel plate information of the trailing material and the book before executing the above-mentioned adhesion adjusting step for the trailing material by continuously transporting the steel plate 10 to be plated. Using the adhesion amount prediction model 34 generated by the method for generating the adhesion amount prediction model of the embodiment, the prediction value calculation step for calculating the prediction value of the plating adhesion amount to the trailing material, and the calculated adhesion amount prediction value. Based on the deviation from the preset target value for adhesion, the resetting step of resetting the operation information regarding the wiping nozzle 6 for the trailing material is executed in the direction in which the deviation becomes smaller.

The apparatus is, for example, in a hot-dip plated steel plate manufacturing facility provided with an adhesion adjusting step of injecting a wiping gas from a wiping nozzle 6 onto the surface of a steel plate to be plated 10 after being immersed in a plating bath to adjust the amount of plating adhesion. A plating adhesion amount control device 33 that controls the plating adhesion amount after the adhesion adjustment step. In the production of a hot-dip plated steel plate, the tail end portion of the leading material and the tip end portion of the trailing material are connected by welding. Before the steel plate 10 to be plated is continuously conveyed and the adhesion adjusting step for the trailing material is executed, the set operation information set for the trailing material including at least a part of the steel plate information of the trailing material and the present implementation Using the adhesion amount prediction model 34 generated by the generation device 31 of the adhesion amount prediction model of the form, the prediction value calculation unit 32 that calculates the predicted value of the plating adhesion amount to the trailing material, and the above-mentioned adhesion to the trailing material. Before executing the adjustment step, based on the deviation between the calculated estimated value of the amount of adhesion and the preset target value for adhesion, the operation information regarding the wiping nozzle 6 for the trailing material is reset in the direction in which the deviation becomes smaller. The configuration is such that the resetting unit 37 is provided.
According to this configuration, even in the production of a continuous hot-dip galvanized steel sheet in which the temperature of the wiping gas is set to a high temperature equal to or higher than the melting point of the plating solution, it is possible to improve the accuracy of the plating adhesion amount. As a result, according to the aspect of the present invention, it is possible to manufacture a high quality continuously hot-dip galvanized steel sheet.

(5) In the present embodiment, the temperature of the wiping gas is set to be equal to or higher than the melting temperature of the plating.
According to this configuration, it is possible to control the amount of plating adhesion with high accuracy even under operating conditions including errors in the prediction by the physical model.

(6) In the present embodiment, the operation information regarding the wiping nozzle 6 to be reset is the position of the wiping nozzle 6.
According to this configuration, the responsiveness to the change in the operating conditions to the target adhesion amount is good.

(7) In the present embodiment, in the resetting step, the adhesion amount prediction model 34 generated by the method for generating the adhesion amount prediction model of the present embodiment is used, and the wiping for the change in the plating adhesion amount with respect to the trailing material. From the impact coefficient calculation process that calculates the impact coefficient, which is the coefficient of change in the operation information related to the nozzle 6, and the calculated impact coefficient and the above deviation, the change amount that calculates the change amount of the operation information related to the wiping nozzle 6 for the trailing material. It includes a calculation process.
In the apparatus, for example, the resetting unit 37 uses the adhesion amount prediction model 34 generated by the generation device 31 of the adhesion amount prediction model of the embodiment, and the wiping nozzle for a change in the plating adhesion amount with respect to the trailing material. The impact coefficient calculation unit 37B, which calculates the impact coefficient, which is the coefficient of change in the operation information related to 6, and
It is configured to include a change amount calculation unit 37C for calculating the change amount of the operation information regarding the wiping nozzle 6 with respect to the trailing material from the calculated influence coefficient and the above deviation.
According to this configuration, it is possible to surely adjust by the position of the nozzle 6 which is the target adhesion amount.

(8) In the present embodiment, in the above-mentioned impact coefficient calculation step, the above-mentioned impact coefficient is calculated based on the change in the plating adhesion amount predicted by the adhesion amount prediction model 34 by changing the set operation information set in the trailing material. calculate.
In the apparatus, for example, the impact coefficient calculation unit 37B calculates the impact coefficient based on the change in the plating adhesion amount predicted by the adhesion amount prediction model 34 by changing the set operation information set in the trailing material. The configuration is to be used.
With this configuration, it is possible to surely adjust by the position of the nozzle 6 which is the target adhesion amount.

(9) In the present embodiment, a method for manufacturing a hot-dip plated steel sheet including a bonding adjustment step of injecting a wiping gas from a wiping nozzle 6 onto the surface of a steel sheet 10 to be plated after being immersed in a plating bath to adjust the amount of plating adhesion. The present invention provides a method (manufacturing apparatus) for manufacturing a hot-dip plated steel sheet, which comprises adjusting the plating adhesion amount by the adhesion amount control method (adhesion amount control device 33) of the present embodiment.
According to this configuration, it is possible to manufacture a high quality continuously hot-dip galvanized steel sheet.

(10) In the present embodiment, from the plurality of qualities required for the manufactured product in the processing process of processing the workpiece into a product by executing the operation of the manufacturing apparatus with the data of the set operating conditions. An input operation including one or more data selected from the information on the workpiece and one or more data selected from the above operating conditions, which is a method of generating a quality prediction model for predicting the selected quality, which is the selected quality. A machine that uses the acquired learning data by acquiring a plurality of sets of training data having information data and actual data related to the selection quality of the product after the processing process using the input operation information data. By learning, a quality prediction model is generated in which at least a part of the data of the input operation information is included in the input data and the data related to the selection quality is used as the output data.

At this time, it is based on a physical model for including one or more data selected from the information on the workpiece and one or more data selected from the operating conditions in the input data and calculating the data on the selected quality as output data. A configuration may be adopted in which the data related to the selection quality calculated by using the numerical calculation formula is used as one of the input data by the machine learning.
This method of generating a quality prediction model makes it possible to provide a quality prediction model with good quality accuracy and to manufacture a higher quality product.

1 Welding device 3 Baking furnace 4 Bath 5 Bath 6 Wiping nozzle 7 Plating adhesion meter 10 Steel plate 20 Height adjustment actuator 21 Nozzle interval adjustment actuator 22 Nozzle angle adjustment actuator 23 Pressure gauge 24 Thermometer 25 Gas pressure adjustment device 30 Plating control device 31 Generation device 31A Learning data collection unit 31B Adhesion amount calculation unit 31C Prediction model generation unit 32 Prediction device (prediction value calculation unit)
32A Adhesion amount prediction unit 33 Adhesion amount control device 34 Adhesion amount prediction model 35 Learning data storage unit 36 Database 37 Reset unit 37A Deviation amount calculation unit 37B Impact coefficient calculation unit 37C Change amount calculation unit 38 Nozzle operation unit 39 Gas control unit

Claims (21)

Plating adhesion after the above-mentioned adhesion adjustment process in a hot-dip plated steel sheet manufacturing facility equipped with an adhesion adjustment process that adjusts the amount of plating adhesion by injecting wiping gas from the wiping nozzle onto the surface of the steel sheet to be plated after being immersed in the plating bath. It is a method of generating an adhesion amount prediction model for predicting the amount.
After the above adhesion adjustment step using the input operation information data including one or more data selected from the steel plate information related to the steel plate to be plated and one or more data selected from the nozzle operation information related to the wiping nozzle, and the input operation information data. Acquire a plurality of sets of learning data having the actual data of the plating adhesion amount of
By machine learning using the acquired learning data, an adhesion amount prediction model is generated in which at least a part of the input operation information data is included in the input data and the plating adhesion amount is used as output data.
A method for generating an adhesion amount prediction model. The above steel sheet information includes the plate width, plate thickness, and steel type of the steel plate to be plated.
The nozzle operation information includes the nozzle spacing, the injection angle with respect to the steel plate surface, the height from the plating bath, the pressure of the wiping gas, and the temperature of the wiping gas in the wiping nozzle.
The input operation information includes the line speed and the temperature of the plating bath.
The method for generating an adhesion amount prediction model according to claim 1. As one of the input data used in the machine learning, one or more data selected from the steel plate information and one or more data selected from the nozzle operation information are included in the input data, and the plating adhesion amount is calculated as output data. The method for generating an adhesion amount prediction model according to claim 1 or 2, further comprising a plating adhesion amount calculated using a numerical calculation formula based on the physical model of the above. Plating adhesion after the above-mentioned adhesion adjustment process in a hot-dip plated steel sheet manufacturing facility equipped with an adhesion adjustment process that adjusts the amount of plating adhesion by injecting wiping gas from the wiping nozzle onto the surface of the steel sheet to be plated after being immersed in the plating bath. It is a method of predicting the amount of plating adhesion, which predicts the amount.
A method for predicting the amount of plating adhesion, which comprises predicting the amount of plating adhesion by using the adhesion amount prediction model generated by the method for generating the adhesion amount prediction model according to any one of claims 1 to 3. .. Plating adhesion after the above-mentioned adhesion adjustment process in a hot-dip plated steel sheet manufacturing facility equipped with an adhesion adjustment process that adjusts the amount of plating adhesion by injecting wiping gas from the wiping nozzle onto the surface of the steel sheet to be plated after being immersed in the plating bath. It is a plating adhesion amount control method that controls the amount,
In the production of hot-dip galvanized steel sheets, the tail end of the leading material and the tip of the trailing material are connected by welding to continuously transport the steel sheet to be plated.
Before performing the above adhesion adjustment step on the trailing material,
The set operation information set for the trailing material including at least a part of the steel plate information of the trailing material, and the adhesion amount generated by the method for generating the adhesion amount prediction model according to any one of claims 1 to 3. A prediction value calculation process for calculating the prediction value of the amount of plating adhered to the trailing material using the prediction model, and
Based on the deviation between the calculated predicted value of the adhesion amount and the preset target value for adhesion, the resetting process for resetting the operation information regarding the wiping nozzle for the trailing material in the direction in which the deviation becomes smaller, and the resetting process.
A plating adhesion control method characterized by performing. The plating adhesion amount control method according to claim 5, wherein the temperature of the wiping gas is set to be equal to or higher than the melting temperature of the plating. The plating adhesion amount control method according to claim 5 or 6, wherein the operation information regarding the wiping nozzle to be reset is the position information of the wiping nozzle. The above resetting process is
Using the adhesion amount prediction model generated by the method for generating the adhesion amount prediction model according to any one of claims 1 to 3, the operation information regarding the wiping nozzle with respect to the change in the plating adhesion amount with respect to the trailing material. The impact factor calculation process that calculates the impact coefficient, which is the coefficient of change in
From the calculated impact coefficient and the above deviation, the change amount calculation process for calculating the change amount of the operation information regarding the wiping nozzle for the trailing material, and the change amount calculation process.
The plating adhesion amount control method according to any one of claims 5 to 7, further comprising. The above-mentioned impact factor calculation step is characterized in that the above-mentioned impact factor is calculated based on the change in the plating adhesion amount predicted by the adhesion amount prediction model by changing the set operation information set in the trailing material. 8. The plating adhesion amount control method according to 8. A method for manufacturing a hot-dip galvanized steel sheet, which comprises an adhesion adjustment step of injecting a wiping gas from a wiping nozzle onto the surface of a steel sheet to be plated after being immersed in a plating bath to adjust the amount of plating adhesion.
A method for manufacturing a hot-dip galvanized steel sheet, which comprises adjusting the plating adhesion amount by the plating adhesion amount control method according to any one of claims 5 to 9. Plating adhesion after the above-mentioned adhesion adjustment process in a hot-dip plated steel sheet manufacturing facility equipped with an adhesion adjustment process that adjusts the amount of plating adhesion by injecting wiping gas from the wiping nozzle onto the surface of the steel sheet to be plated after being immersed in the plating bath. It is a generator of the adhesion amount prediction model for predicting the amount.
After the above adhesion adjustment step using the input operation information data including one or more data selected from the steel plate information related to the steel plate to be plated and one or more data selected from the nozzle operation information related to the wiping nozzle, and the input operation information data. A learning data storage unit that stores a plurality of sets of learning data having the actual data of the plating adhesion amount of
By machine learning using a plurality of sets of learning data stored in the learning data storage unit, at least a part of the input operation information data is included in the input data, and the adhesion amount prediction using the plating adhesion amount as the output data. A predictive model generator that generates a model, and a predictive model generator
A device for generating an adhesion amount prediction model, which comprises. The above steel sheet information includes the plate width, plate thickness, and steel type of the steel plate to be plated.
The nozzle operation information includes the nozzle spacing, the injection angle with respect to the steel plate surface, the height from the plating bath, the pressure of the wiping gas, and the temperature of the wiping gas in the wiping nozzle.
The apparatus for generating an adhesion amount prediction model according to claim 11, wherein the input operation information includes a line speed and a temperature of the plating bath. The input data includes one or more data selected from the steel plate information and one or more data selected from the nozzle operation information, and a numerical calculation formula based on a physical model for calculating the plating adhesion amount as output data is used. Equipped with an adhesion amount calculation unit that calculates the plating adhesion amount,
As one of the input data by the machine learning, the plating adhesion amount calculated by the adhesion amount calculation unit is used.
11. The apparatus for generating an adhesion amount prediction model according to claim 11 or 12. Plating adhesion after the above-mentioned adhesion adjustment process in a hot-dip plated steel sheet manufacturing facility equipped with an adhesion adjustment process that adjusts the amount of plating adhesion by injecting wiping gas from the wiping nozzle onto the surface of the steel sheet to be plated after being immersed in the plating bath. It is a device for predicting the amount of plating adhesion, which predicts the amount.
It is characterized by comprising an adhesion amount prediction unit for predicting a plating adhesion amount by using the adhesion amount prediction model generated by the adhesion amount prediction model generation apparatus according to any one of claims 11 to 13. A device for predicting the amount of plating adhesion. Plating adhesion after the above-mentioned adhesion adjustment process in a hot-dip plated steel sheet manufacturing facility equipped with an adhesion adjustment process that adjusts the amount of plating adhesion by injecting wiping gas from the wiping nozzle onto the surface of the steel sheet to be plated after being immersed in the plating bath. It is a plating adhesion amount control device that controls the amount,
In the production of hot-dip galvanized steel sheets, the tail end of the leading material and the tip of the trailing material are connected by welding to continuously transport the steel sheet to be plated.
Prior to executing the above-mentioned adhesion adjusting step for the trailing material, the set operation information set for the trailing material including at least a part of the steel plate information of the trailing material, and any one of claims 11 to 13. Using the adhesion amount prediction model generated by the generation device of the adhesion amount prediction model described, the prediction value calculation unit for calculating the prediction value of the plating adhesion amount to the trailing material, and the prediction value calculation unit.
Regarding the wiping nozzle for the trailing material in the direction in which the deviation becomes smaller based on the deviation between the calculated predicted value of the adhesion amount and the preset target value for adhesion before executing the adhesion adjusting step for the trailing material. The resetting unit that resets the operation information,
A plating adhesion amount control device characterized by being provided with. The plating adhesion amount control device according to claim 15, wherein the temperature of the wiping gas is set to be equal to or higher than the melting temperature of the plating. The plating adhesion amount control device according to claim 15 or 16, wherein the operation information regarding the wiping nozzle to be reset is the position of the wiping nozzle. The above reset part
Operation information regarding the above-mentioned wiping nozzle with respect to a change in the plating adhesion amount with respect to the trailing material by using the adhesion amount prediction model generated by the generation device of the adhesion amount prediction model according to any one of claims 11 to 14. The impact coefficient calculation unit that calculates the impact coefficient, which is the coefficient of change in
From the calculated impact coefficient and the above deviation, the change amount calculation unit that calculates the change amount of the operation information regarding the wiping nozzle for the trailing material, and
The plating adhesion amount control device according to any one of claims 15 to 17, wherein the plating adhesion amount control device is provided. The above-mentioned impact factor calculation unit is characterized in that it calculates the above-mentioned impact coefficient based on the change in the plating adhesion amount predicted by the adhesion amount prediction model by changing the set operation information set in the trailing material. 18. The plating adhesion amount control device according to 18. It is a hot-dip galvanized steel sheet manufacturing apparatus equipped with an adhesion adjustment step of injecting a wiping gas from a wiping nozzle onto the surface of a steel sheet to be plated after being immersed in a plating bath to adjust the amount of plating adhesion.
A hot-dip galvanized steel sheet manufacturing apparatus comprising the plating adhesion amount control device according to any one of claims 15 to 19. In the processing process of processing the workpiece into a product by executing the operation of the manufacturing equipment with the data of the set operating conditions, the selected quality, which is the quality selected from the multiple qualities required for the manufactured product, is selected. It is a method of generating a quality prediction model for prediction.
Input operation information data including one or more data selected from the information about the workpiece and one or more data selected from the operation conditions, and the above-mentioned product after the processing process using the input operation information data. Acquire multiple sets of learning data with actual data on selection quality,
By machine learning using the acquired training data, a quality prediction model is generated in which at least a part of the input operation information data is included in the input data and the data related to the selection quality is output data.
A numerical calculation formula based on a physical model that includes one or more data selected from the information on the workpiece and one or more data selected from the operating conditions in the input data, and calculates the data on the selected quality as output data. The data related to the selection quality calculated using the above is used as one of the input data by the machine learning.
A method of generating a quality prediction model, which is characterized by the fact that.

Images