JP7140073B2 - LEARNING MODEL GENERATION METHOD, DATABASE CONSTRUCTION METHOD, MILL SETUP SETTING METHOD, ROLLED MATERIAL MANUFACTURING METHOD, PROCESSING TARGET MANUFACTURING METHOD, AND LEARNING MODEL GENERATING DEVICE - Google Patents

Description

The present invention relates to a learning model generation method, a database construction method, a mill setup setting method, a rolled material manufacturing method, a processing target manufacturing method, and a learning model generation apparatus.

The mill setup in a tandem rolling mill with multiple rolling stands determines the strip thickness and shape of the rolled material on the delivery side of each rolling stand, and the optimization of the set values is important from the viewpoint of stable operation and quality assurance. be. Against this background, various mill setup setting methods have been proposed. Specifically, Patent Literatures 1 and 2 describe a method of calculating a mill set-up setting value using a trained neural network. In addition, Non-Patent Document 1 describes a method of correcting mill setup setting values by performing numerical optimization by sequential quadratic programming using a mathematical model (rolling model).

JP-A-5-38511 JP-A-8-90020

Akira Murakami et al., "Optimization of Pass Schedule for Cold Tandem Rolling Mill," Tetsu-to-Hagane, The Iron and Steel Institute of Japan, 2004, Vol. 90, No. 11 K. Nakai et al., "Mill Setup Model Adjustment Support System", Hitachi Hyoron, Hitachi Hyoronsha, 1985, Vol. 67, No. 4

The methods described in Patent Documents 1 and 2 and Non-Patent Document 1 are based on the concept of calculating mill setup setting values based on a large amount of operational performance data, and all operational performance data are available regardless of whether the rolling efficiency is high or low. are subject to learning. However, in the actual operation data, even if the same product is manufactured, there are some that have different mill setup setting values depending on the operator, and this difference in mill setup setting values may result in a difference in rolling efficiency. . Therefore, according to the methods described in Patent Literatures 1 and 2 and Non-Patent Literature 1, the rolling efficiency is affected by variations among operators, and past operation results indicating high rolling efficiency may not necessarily be reproduced.

The present invention has been made in view of the above problems, and an object of the present invention is to provide a learning model generating method and a learning model generating apparatus capable of generating a learning model that outputs setting values of production equipment that satisfy desired operation indices. is to provide Another object of the present invention is to provide a database construction method capable of constructing a database of operation performance data that satisfies desired operation indexes. Another object of the present invention is to provide a mill setup setting method capable of setting a mill setup set value that satisfies a desired operational index. Furthermore, another object of the present invention is to provide a method for producing a rolled material that can produce a rolled material while satisfying desired operational indices. Another object of the present invention is to provide a method for manufacturing an object to be treated, which is capable of producing an object to be treated while satisfying desired operational indices.

The learning model generation method according to the present invention uses production facility operation performance data including the operating conditions of the production facility, the actual values of the variables that serve as the operation indicators of the production facility, and the actual values of the set values of the production facility. , a first learning step of learning a learning model of the variables that inputs the operating conditions and outputs predicted values of the variables; a selection step of calculating a predicted value of the variable and selecting operation result data corresponding to a processing target based on a comparison between the predicted value and the actual value of the variable; and a second learning step of learning a learning model of the setting value, in which a condition is input and a predicted value of the setting value is output.

In the learning model generation method according to the present invention, in the above invention, the first learning step classifies the operation performance data into a plurality of groups for each specification of the product produced in the production facility, and generating a learning model of said variables for group operational performance data.

The learning model generation method according to the present invention is characterized in that, in the above invention, the operation index is an index relating to the efficiency of the production equipment.

The database construction method according to the present invention uses production facility operation performance data including operating conditions of the production facility, actual values of variables that serve as operational indicators of the production facility, and actual values of set values of the production facility, A predicted value of the variable is calculated by inputting the operating condition included in the operation performance data to a learning model of the variable that has been learned to input the operating condition and output the predicted value of the variable. and classifying the operation result data based on the comparison between the predicted value and the actual value of the variable to create a database of the operation result data.

A mill setup setting method according to the present invention is a mill setup by inputting operating conditions for a rolled material to be processed into a learning model of mill setup setting values for rolling equipment generated using a learning model generating method according to the present invention. and calculating a predicted set value, and setting the mill setup set value of the rolling equipment according to the calculated predicted mill setup set value.

The mill setup setting method according to the present invention uses the operation performance data of the rolling equipment including the operating conditions of the rolling equipment, the actual values of variables that serve as operational indicators of productivity, and the actual values of the mill setup setting values of the rolling equipment. input the operating conditions and output the predicted values of the variables by inputting the operating conditions included in the operation performance data into the learning model of the variables learned to output the predicted values of the variables. A selection step of selecting operation performance data that contributes to the productivity of the rolled material to be processed based on a comparison between the predicted value and the actual value of the variable, and using the selected operation performance data, the operation Mill setup setting of the rolling equipment by inputting the operating conditions to be processed to the learning model of the mill setup setting value that has been learned so as to input the conditions and output the predicted value of the mill setup setting value and a setting step of calculating a predicted value of the mill setup setting value and setting the mill setup setting value of the rolling equipment according to the calculated predicted value of the mill setup setting value.

A method for producing a rolled material according to the present invention is characterized by including a step of producing a rolled material while controlling rolling equipment according to the mill setup setting method according to the present invention.

A method for manufacturing a target object according to the present invention uses operation performance data of a production facility including operating conditions of the production facility, actual values of variables serving as operation indicators of the production facility, and actual values of set values of the production facility. input the operating conditions and output the predicted values of the variables by inputting the operating conditions included in the operation performance data into the learning model of the variables learned to output the predicted values of the variables. a selection step of selecting operation result data corresponding to the object to be processed based on a comparison between the predicted value and the actual value of the variable; By inputting the operating conditions to be processed to the learning model of the setting value that has been learned to output the predicted value of the value, the predicted value of the setting value of the production equipment is calculated, and the calculated setting value The method is characterized by comprising a setting step of setting a setting value of the production equipment according to the predicted value, and a step of manufacturing the processing target while controlling the production equipment according to the set setting value.

A learning model generation device according to the present invention uses operation performance data of a production facility including operating conditions of the production facility, actual values of variables that serve as operation indicators of the production facility, and actual values of set values of the production facility. , means for learning a learning model of the variables that inputs the operating conditions and outputs predicted values of the variables; A means for calculating a predicted value and selecting operation result data corresponding to a processing target based on a comparison between the predicted value and the actual value of the variable, and inputting the operating conditions using the selected operation result data, and means for learning a learning model of the setting value that outputs a predicted value of the setting value.

According to the learning model generation method and the learning model generation device according to the present invention, it is possible to generate a learning model that outputs setting values for production equipment that satisfy desired operation indices. Further, according to the database construction method of the present invention, it is possible to construct a database of operation result data that satisfies desired operation indexes. Further, according to the mill setup setting method of the present invention, it is possible to set the mill setup setting value that satisfies the desired operation index. Furthermore, according to the method for manufacturing a rolled material according to the present invention, it is possible to manufacture a rolled material while satisfying the desired operational index. Moreover, according to the manufacturing method of the object to be treated according to the present invention, the object to be treated can be produced while satisfying the desired operational index.

FIG. 1 is a block diagram showing the configuration of a rolling control system that is one embodiment of the present invention. FIG. 2 is a flow chart showing the flow of the reduction rate setting process, which is one embodiment of the present invention.

Hereinafter, the configuration and operation of a rolling control system that is an embodiment of the present invention will be described with reference to the drawings.

〔Constitution〕
First, referring to FIG. 1, the configuration of a rolling control system that is an embodiment of the present invention will be described.

FIG. 1 is a block diagram showing the configuration of a rolling control system that is one embodiment of the present invention. As shown in FIG. 1, a rolling control system 1, which is one embodiment of the present invention, rolls a steel plate S to a desired thickness and shape by controlling a tandem rolling mill having a plurality of rolling stands 2a to 2e. The system comprises a mill setup setting calculator 3, a process calculator 4, a rollover position controller 5, and a shift controller 6 as main components.

The mill setup setting calculation device 3 comprises an information processing device such as a computer, and includes an operation performance data database (operation performance data DB) 3a in which past operation performance data of the tandem rolling mill is stored. In the present embodiment, the operation performance data includes the rolling conditions of the steel sheet S, the actual values of information indicating the rolling efficiency of the tandem rolling mill such as the maximum rolling speed, and the actual values of the set values of the mill setup process (mill setup set values). contains information about Further, as the information about the rolling conditions of the steel sheet S, the composition, thickness, width, steel type, roll specifications, etc. of the steel sheet S can be exemplified. Although details will be described later, the mill setup setting calculation device 3 inputs the rolling conditions of the steel plate S, learns a mill setup model that outputs the predicted value of the mill setup setting value, and sets the mill setup of the steel plate S to be processed. Calculation of values and construction of high-efficiency database are executed.

The above-mentioned mill setup process means a process of setting mill setup set values such as reduction ratio distribution, front tension, rear tension, bending force, shift amount, etc. of each rolling stand according to the rolling conditions of the steel plate S. . In the mill setup process, for example, according to the flow described in Non-Patent Document 2, the rolling load, rolling torque, etc. are predicted and calculated based on the setting of the rolling reduction and tension of each rolling stand, and the running strip thickness is changed. Calculate the reduction position and roll peripheral speed of each rolling stand for. In addition, in the mill setup process, setting values for bending force and shift amount, which greatly affect the quality of steel plate S (for example, plate shape and off-gauge), are determined based on material specifications and predicted values of rolling load. It is common to use the value adjusted in 1 as the final mill set-up setting value.

The process computer 4 outputs information about the rolling conditions of the steel plate S to be processed to the mill setup setting calculation device 3, and performs each rolling according to the mill setup setting values of the steel plate S to be processed calculated by the mill setup setting calculation device 3. Calculate the rolling schedule for the stand. The process computer 4 then controls the operations of the roll position control device 5 and the shift control device 6 according to the calculated rolling schedule.

The roll-down position control device 5 controls the roll-down position, which is the gap between the pair of work rolls provided in each rolling stand, in accordance with the control signal from the process computer 4, thereby controlling the rolling position of the steel sheet S to be processed on the delivery side of each rolling stand. Control the plate thickness to the target thickness.

The shift control device 6 controls the bending force and shift amount of the pair of work rolls provided in each rolling stand according to the control signal from the process computer 4, thereby controlling the shape of the steel sheet S to be processed on the delivery side of each rolling stand. is controlled to the target shape.

As described above, once the rolling conditions for the steel sheet S are determined, the mill setup setting values are calculated, and the reduction position, maximum rolling speed, etc. of each rolling stand are calculated from the mill setup setting values using physical models and various machine learning models. be done. At this time, the operator often manually adjusts the mill set-up value, and the mill set-up value varies greatly between operators. As a result, the values calculated through physical models, various machine learning models, etc. also vary among operators, resulting in differences in rolling efficiency.

Therefore, in the rolling control system 1 of the present embodiment, the mill setup setting calculation device 3 calculates the mill setup setting values as follows so as not to cause a difference in rolling efficiency. do. The operation of the mill setup setting calculation device 3 when calculating the rolling reduction setting value will be described below with reference to the flowchart shown in FIG. In the present embodiment, the calculation process of the rolling reduction setting value (rolling reduction setting process) will be described, but the mill setup setting values other than the rolling reduction such as tension can be similarly calculated.

[Reduction rate setting process]
FIG. 2 is a flow chart showing the flow of the reduction rate setting process, which is one embodiment of the present invention. The flowchart shown in FIG. 2 starts at the timing when a mill setup set value setting command is input to the mill setup setting calculation device 3, and the rolling reduction rate setting process proceeds to the process of step S1.

In the process of step S1, the mill setup setting calculation device 3 uses all the operation result data stored in the operation result data DB 3a to set the rolling condition of the steel plate S as an input variable and the maximum rolling speed of the tandem rolling mill. Machine-learning a prediction model for the maximum rolling speed with the predicted value as an output variable. Examples of machine learning techniques include deep learning, logistic regression analysis, decision trees, neural networks, and the like. In this process, the mill setup setting calculation device 3 classifies the operation result data into a plurality of groups for each specification of the steel plate S, and creates a maximum rolling speed prediction model for the classified operation result data of each group. You can learn. According to such a process, a prediction model of the maximum rolling speed can be generated according to the difference in specifications of the steel sheet S. As a result, the process of step S1 is completed, and the rolling reduction rate setting process proceeds to the process of step S2.

In the process of step S2, the mill setup setting calculation device 3 applies machine-learning in the process of step S1 to the prediction model of the maximum rolling speed for all the operation result data stored in the operation result data DB 3a. By inputting the rolling conditions, the predicted value of the maximum rolling speed is calculated. As a result, the process of step S2 is completed, and the rolling reduction rate setting process proceeds to the process of step S3.

In the processing of step S3, the mill setup setting calculation device 3 selects operation performance data in which the actual value of the maximum rolling speed is greater than the predicted value of the maximum rolling speed calculated in the processing of step S2. In this process, the mill setup setting calculation device 3 stores the selected operation result data in the high efficiency database and the unselected operation result data in the low efficiency database, thereby constructing the operation result data database. You may Further, the processing of steps S1 to S3 does not necessarily have to be executed each time the rolling reduction rate setting process is executed, but may be executed at a predetermined timing, such as each time the operation performance data is updated. As a result, the process of step S3 is completed, and the rolling reduction rate setting process proceeds to the process of step S4.

In the process of step S4, the mill setup setting calculation device 3 uses the operation performance data selected in the process of step S3 to set the rolling condition of the steel sheet S as an input variable and the predicted value of the reduction ratio as an output variable. Off-line or on-line machine learning of the prediction model of the reduction rate setting value. Examples of machine learning techniques include deep learning, logistic regression analysis, decision trees, neural networks, and the like. As a result, the process of step S4 is completed, and the rolling reduction rate setting process proceeds to the process of step S5.

In the process of step S5, the mill setup setting calculation device 3 inputs the rolling conditions of the steel sheet S to be processed to the prediction model of the rolling reduction set value machine-learned in the process of step S4. Calculate the reduction rate setting value for Then, the mill setup setting calculator 3 outputs information on the calculated rolling reduction set value to the process computer 4, thereby controlling the rolling reduction of each rolling stand according to the calculated rolling reduction setting value. As a result, the process of step S5 is completed, and the series of reduction rate setting processes is completed.

As is clear from the above description, in the rolling reduction setting process, which is an embodiment of the present invention, the mill setup setting calculation device 3 uses the operation performance data of the tandem rolling mill to input the rolling conditions of the steel sheet S, A maximum rolling speed prediction model that outputs the maximum rolling speed prediction value is learned, and the maximum rolling speed prediction model is input with the rolling conditions of the steel plate S included in the operation performance data to obtain the maximum rolling speed prediction value. Calculate, select the actual operation data corresponding to the object to be processed based on the comparison between the predicted value and the actual value of the maximum rolling speed, and input the rolling conditions of the steel plate S using the selected operational result data, and the reduction rate Learn a prediction model of the reduction rate setting value that outputs the predicted value of the setting value. According to such a process, since the operation result data selected based on the comparison between the predicted value and the actual value of the maximum rolling speed is used as the learning target, compared with the case where all the operation result data are used as the learning target. can be used to generate a learning model that outputs a set reduction ratio that satisfies the desired maximum rolling speed.

Although the embodiments to which the invention made by the present inventors is applied have been described above, the present invention is not limited by the descriptions and drawings forming a part of the disclosure of the present invention according to the present embodiments. For example, the above-described embodiment is an example in which the present invention is applied to the setting value calculation of a rolling mill, but the scope of application of the present invention is not limited to the above-described embodiment. It can be applied to the setting value calculation of various production equipment whose operation index changes depending on the situation. In the above embodiment, the actual operation data used for the learning calculation was selected based on the actual value of the rolling efficiency. You may select the operation result data used for learning calculation by using. That is, in the present invention, predicted values of variables obtained from a learning model are used to select operation performance data necessary for improving productivity, quality, etc. from a large amount of past operation performance data. In the embodiment, the rolling speed was selected as this variable with focus on productivity, but it is possible to appropriately select a variable for bringing about manufacturing merits such as quality improvement and defect rate. Thus, other embodiments, examples, operation techniques, etc. made by those skilled in the art based on this embodiment are all included in the scope of the present invention.

1 rolling control system 2a, 2b, 2c, 2d, 2e rolling stand 3 mill setup setting calculator 3a operation performance data database (operation performance data DB)
4 process computer 5 screw position control device 6 shift control device

Claims (8)

Using the operating conditions of the production equipment, the actual value of the index related to the efficiency of the production equipment, and the actual value of the setting value of the production equipment, input the operating conditions and the index of the efficiency a first learning step of learning a learning model of the efficiency index outputting a predicted value;
A predicted value of the efficiency index is calculated by inputting the operating conditions included in the operation result data into the learning model of the efficiency index , and operation result data having a larger actual value than the predicted value of the efficiency index is selected. a selection step to
a second learning step of learning a learning model of the set values using the selected operational performance data, the operating conditions being input and the predicted values of the set values being output;
A learning model generation method, comprising: The first learning step classifies the operational performance data into a plurality of groups for each specification of the product produced in the production facility, and applies a learning model of the index related to efficiency to the operational performance data of each classified group. 2. The learning model generation method according to claim 1, further comprising the step of generating. Using the operating conditions of the production equipment, the actual value of the index related to the efficiency of the production equipment, and the actual value of the setting value of the production equipment, input the operating conditions and the index of the efficiency A predicted value of the efficiency index is calculated by inputting operating conditions included in the operation result data into a learning model of the efficiency index that has been learned to output the predicted value, and the efficiency index is calculated. A database construction method, comprising the step of classifying operation result data having an actual value larger than a predicted value of and creating a database of the operation result data. By inputting the operating conditions of the rolled material to be processed into the learning model of the mill setup setting values of the rolling equipment generated using the learning model generation method according to claim 1 or 2 , the predicted value of the mill setup setting value is calculated. calculating and setting the mill setup settings for the rolling equipment in accordance with the calculated predicted values of the mill setup settings. Using the operating conditions of the rolling equipment, the actual value of the index related to the rolling efficiency of the rolling equipment , and the actual value of the mill setup setting value of the rolling equipment, the operating conditions are input, and the rolling Calculate the predicted value of the index of rolling efficiency by inputting the operating conditions included in the operation result data into the learning model of the index of rolling efficiency that has been learned to output the predicted value of the index of efficiency. a selection step of selecting operation performance data that contributes to the productivity of the rolled material to be processed , the actual value being larger than the predicted value of the index related to the rolling efficiency ;
Using the selected operation performance data, the operating conditions to be processed are applied to the learning model of the mill setup setting values that has been learned to input the operating conditions and output the predicted values of the mill setup setting values. a setting step of calculating a predicted value of the mill setup setting value of the rolling equipment by inputting and setting the mill setup setting value of the rolling equipment according to the calculated predicted value of the mill setup setting value;
A mill setup setting method comprising: A method for manufacturing a rolled material, comprising the step of manufacturing a rolled material while controlling rolling equipment according to the mill setup setting method according to claim 4 or 5 . Using the operating conditions of the production equipment, the actual value of the index related to the efficiency of the production equipment, and the actual value of the setting value of the production equipment, input the operating conditions and the index of the efficiency A predicted value of the efficiency index is calculated by inputting operating conditions included in the operation result data into a learning model of the efficiency index that has been learned to output the predicted value, and the efficiency index is calculated. A selection step of selecting operation performance data whose actual value is larger than the predicted value of
By inputting the operating conditions to be processed to the learning model of the setting values that has been learned to input the operating conditions and output the predicted values of the setting values using the selected operation performance data a setting step of calculating a predicted value of the setting value of the production equipment and setting the setting value of the production equipment according to the calculated predicted value of the setting value;
a step of manufacturing the processing target while controlling the production equipment according to the set values;
A method of manufacturing an object to be processed, comprising: Using the operating conditions of the production equipment, the actual value of the index related to the efficiency of the production equipment, and the actual value of the setting value of the production equipment, input the operating conditions and the index of the efficiency means for learning a learning model of the efficiency index that outputs a predicted value;
A predicted value of the efficiency index is calculated by inputting the operating conditions included in the operation result data into the learning model of the efficiency index , and operation result data having a larger actual value than the predicted value of the efficiency index is selected. means to
means for learning a learning model for the set values using the selected operational performance data, the operating conditions being input and the predicted values of the set values being output;
A learning model generation device characterized by comprising:

Images