Cold-rolled steel sheet material control system and cold-rolled steel sheet material prediction model generation method

Abstract

The cold-rolled steel sheet material control system according to one embodiment of the present invention comprises: a programmable logic controller for acquiring process data of a produced cold-rolled steel sheet; the process controller is used for controlling the manufacturing process of the cold-rolled steel sheet according to the process data of the programmable logic controller; a model generation unit that generates a prediction model for predicting the material quality in the width direction of the cold-rolled steel sheet to be produced by learning the learning data set in advance; and a production management system that transmits and receives information to and from the process controller, and that manages the production of the cold-rolled steel sheet by receiving the predicted value of the prediction model from the model generation unit. Comprising the following steps: a model generation part generates a learning data set; a preset model is trained through the learning data set to generate a prediction model for predicting the material of the cold-rolled steel sheet in the width direction, and the learning data set comprises multiple pieces of data including hot rolling process data in process data provided by a programmable logic controller; material data and component data provided from a production management system; and mapping data of measurement positions between cold rolling process data and annealing process data in the process data from the programmable logic controller, and the process data provided by the programmable logic controller is the process data of the produced cold-rolled steel sheet acquired by a sensor in advance.

Description

Technical Field

[0001] The present invention relates to a cold-rolled steel sheet material control system and a method for generating a cold-rolled steel sheet material prediction model. Background Art

[0002] Recently, various environmental regulations such as carbon dioxide (CO2) emission restrictions and energy usage regulations have been introduced in countries around the world. Therefore, improving fuel efficiency and durability has become an important issue that automotive companies need to address. To this end, if thin high-strength steel is used, various problems such as the environment, fuel efficiency, collision resistance, and durability can be improved simultaneously. In order to achieve vehicle body lightweighting, the development of high-strength steel is ongoing. However, in order to exhibit high-strength steel, it is necessary to increase the content of components such as carbon (C), manganese (Mn), and silicon (Si). The more components there are within the steel type, the greater the material sensitivity (material change) to process temperature changes. Therefore, the more severe the material deviation within the coil for gigapascal high-strength steel. If material deviation occurs, springback deviation within a single coil will occur during customer forming, resulting in continuous poor forming quality.

[0003] (Patent Document 1) Korean Registered Patent Bulletin No. 10-0858902 Summary of the Invention

[0004] (1) Technical Problems to be Solved

[0005] According to an embodiment of the present invention, there is provided a cold-rolled steel sheet material control system and a method for generating a cold-rolled steel sheet material prediction model that use artificial intelligence to predict the material according to the total width of the cold-rolled steel sheet.

[0006] (2) Technical Solutions

[0007] To solve the above problems of the present invention, a cold-rolled steel sheet material control system according to an embodiment of the present invention may include: a programmable logic controller that acquires process data of the produced cold-rolled steel sheet; a process controller that controls the manufacturing process of the cold-rolled steel sheet according to the process data of the programmable logic controller; a model generation unit that generates a prediction model for predicting the material in the width direction of the cold-rolled steel sheet to be produced by learning a learning data set in advance; and a production management system that transmits and receives information with the process controller and receives the prediction value of the prediction model of the model generation unit to manage the production of the cold-rolled steel sheet. The learning data set may include: hot rolling process data in the process data from the programmable logic controller; material data and component data from the production management system; and mapping data of the measurement position between the hot rolling process data and the cold rolling and annealing process data in the process data from the programmable logic controller.

[0008] A method for generating a cold-rolled steel sheet material prediction model according to an embodiment of the present invention may include the following steps: a model generation unit generates a learning data set; and a preset model is trained through the learning data set to generate a prediction model for predicting the material in the width direction of the cold-rolled steel sheet. The multiple data included in the learning data set may include: hot rolling process data in the process data provided by a programmable logic controller; material data and composition data provided by a production management system; and mapping data of the measurement position between the hot rolling process data and the cold rolling and annealing process data in the process data from the programmable logic controller. The process data provided by the programmable logic controller may be the process data of the produced cold-rolled steel sheet obtained in advance through sensors.

[0009] (III) Beneficial effects

[0010] According to an embodiment of the present invention, it can exhibit precise and accurate prediction capabilities, and has the effect of being able to help solve factors that reduce production capacity such as die modification and forming condition changes caused by the sheet material part inside the coil during forming. Description of the drawings

[0011] Figure 1 is a schematic configuration diagram of a cold-rolled steel sheet material control system according to an embodiment of the present invention.

[0012] Figure 2 is a diagram showing the process data detected in a cold-rolled steel sheet material control system according to an embodiment of the present invention, Figure 3 is an analysis result showing which variable among the X variables (process variables) is more emphasized in the predicted values calculated by the model generated through a cold-rolled steel sheet material control system according to an embodiment of the present invention and a method for generating a cold-rolled steel sheet material prediction model according to an embodiment of the present invention to derive the predicted values.

[0013] Figure 4 is a schematic flowchart of a method for generating a cold-rolled steel sheet material prediction model according to an embodiment of the present invention.

[0014] Figure 5 is a diagram showing the learning data variables of a method for generating a cold-rolled steel sheet material prediction model according to an embodiment of the present invention.

[0015] Figure 6 and Figure 7 is a diagram showing the setting of weights of a method for generating a cold-rolled steel sheet material prediction model according to an embodiment of the present invention.

[0016] Figure 8 is a diagram showing an exemplary computing environment capable of implementing a cold-rolled steel sheet material control system according to an embodiment of the present invention. Detailed Embodiment

[0017] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art to which the present invention pertains can easily implement the present invention.

[0018] First, in a cold-rolled steel sheet material control system and a cold-rolled steel sheet material prediction model generation method according to an embodiment of the present invention, in the case of a cold-rolled steel sheet, for example, a cold-rolled steel sheet for automobiles, after manufacturing a steel billet, it is heated. By weight%, the steel billet contains: C: 0.001 - 0.4%, Si: 0.1 - 2.5%, Mn: 1 - 3%, Al: 0.1% or less, Cr: 1.0% or less, Ti: 0.2% or less, Nb: 0.1% or less, V: 0.2% or less, Mo: 0.5% or less, B: 30 ppm or less, the balance being Fe and other inevitable impurities. This is to smoothly perform the subsequent hot rolling process and obtain a material corresponding to the target, and it is preferably heated to a temperature range of 1100 - 1300 °C. Hot rolling is performed by rough rolling followed by finish rolling, and before entering the coiling (CT) step after finish rolling, it will go through a cooling process. By cold rolling the coiled hot-rolled steel sheet at a predetermined reduction rate at room temperature, a cold-rolled steel sheet can be manufactured. The cold-rolled steel sheet after cold rolling is preferably then subjected to continuous annealing. The continuous annealing can be composed of a heating section (Heating Section, HS), a soaking section (Soaking Section, SS), a slow cooling section (Slow Cooling Section, SCS), a rapid cooling section (Rapid Cooling Section, RCS), a reheating section (Reheating Section, RHS), an overaging section (Overaging Section, OAS), and a final cooling section (Final Cooling Section, FCS).

[0019] Figure 1 It is a schematic configuration diagram of a cold-rolled steel sheet material control system according to an embodiment of the present invention.

[0020] Refer to Figure 1 , a cold-rolled steel sheet material control system according to an embodiment of the present invention may include a sensor 110, a programmable logic controller (Programmable Logic Controller, PLC) 120, a process controller (Process Controller, PC) 130, a manufacturing execution system (Manufacturing Execution System, MES) 140, and a model generation unit 150.

[0021] The sensor 110 can be respectively arranged in the above-mentioned cold-rolled steel sheet manufacturing process to detect process data. The programmable logic controller 120 can obtain the process data from the sensor 110 and transmit it to the process controller 130. The process controller 130 can communicate with the production management system 140 and control the programmable logic controller 120 according to the control of the production management system 140 and the obtained process data, so as to control the above-mentioned cold-rolled steel sheet manufacturing process. The programmable logic controller 120 can transmit the process data obtained from the sensor 110 to the model generation unit 150.

[0022] The model generation unit 150 can generate a learning data set based on the process data and train a preset model, so as to generate a prediction model capable of predicting the material according to the width direction (total width) of the cold-rolled steel sheet.

[0023] The model generation unit 150 can include a first database (DB) 151, a second database 152, and a model generator 153.

[0024] Figure 2 It is a diagram showing the process data detected in the cold-rolled steel sheet material control system according to an embodiment of the present invention. Figure 3 It is an analysis result showing which variable in the X variable (process variable) is more emphasized in the predicted value calculated by the model generated by the cold-rolled steel sheet material control system according to an embodiment of the present invention and the cold-rolled steel sheet material prediction model generation method according to an embodiment of the present invention to derive the predicted value.

[0025] Referring to Figure 2 , the shown process data are the data of yield strength (YP), tensile strength (TS), and elongation (EL) measured in the width direction of the cold-rolled steel sheet.

[0026] The learning data selects the factors that have a greater impact on the material and loads the component data for each roll of cold-rolled steel sheet, and the data related to the width direction (total width) in the hot rolling, cold rolling, and annealing processes can be stored in the first database 151. Referring to Figure 2 and Figure 3 , more specifically, the temperature before hot rolling finish (FET), the temperature after hot rolling finish (FDT), and the coiling temperature (CT) can be measured and collected, and these width direction data can be used as hot rolling data for learning data. The data of the hot rolling process can be collected for the total width of the coil at a predetermined interval, for example, data measured at equal intervals within 10m.

[0027] By connecting the data between the hot rolling process and the subsequent cold rolling annealing process, as shown in Table 1 below, the matching of the material prediction model may be more excellent compared to the case of using the average value in each process as each variable of the hot rolling process. That is, as variables related to hot rolling, the data of the material measurement part is used instead of the average value of the total width, so that the prediction matching can be excellent.

[0028] [Table 1]

[0029]

[0030] Figure 4 is a schematic flowchart of a method for generating a cold rolled steel sheet material prediction model according to an embodiment of the present invention.

[0031] Referring to Figure 1 and Figure 4 , the model generator 153 generates a learning data set based on the process data stored in the first database 151, where the process data stored in the first database 151 can be queried (S1), merged with the material data (S2), a learning data set is generated, and a preset model is trained to generate a prediction model (S3).

[0032] First, the step of generating learning data may include the following steps: querying the coil composition data of the slab (S1a); querying the hot rolling process variables in the width direction (S1b); querying the cold rolling and annealing process data (S1c); and querying the skin pass process data (S1d), and since the dimensional change of the coil between the hot rolling process and the cold rolling process is too large, a measurement position mapping operation between the hot rolling data and the cold rolling / annealing data is required. Therefore, the hot rolling data can be extended by the length caused by cold rolling and merged with the cold rolling and annealing data.

[0033] Regarding the cold rolling related data in detail, first, the data related to the rolling force generated during cold rolling will be measured and collected, and the passing speed of the steel sheet in the continuous annealing furnace from the heating section to the final cooling section and in the annealing furnace will be measured and collected. Since the lengths of the hot rolled steel sheet and the cold rolled steel sheet are different, a position mapping step for each data of hot rolling / cold rolling and annealing can be further performed.

[0034] Figure 5 is a diagram showing the learning data variables of a method for generating a cold rolled steel sheet material prediction model according to an embodiment of the present invention, Figure 6 and Figure 7 is a diagram showing the setting of weights of a method for generating a cold rolled steel sheet material prediction model according to an embodiment of the present invention.

[0035] First, referring simultaneously to Figure 1 and Figure 5, including the step of querying the measured and collected data from the first database 151 through the model generator 153, which is data related to process variables and corresponds to the X data in the model. The output, that is, the data related to Y is the material measurement value, corresponding to the yield strength, tensile strength, and elongation. For each coiled sheet, the above-mentioned process variables will be stored in the second database 152. The information of newly produced coiled sheets is periodically stored in the first database 152 (for the variables as described above), and the period can be one week or one month. Re-query the learning data stored in the second database 152 as described above through the model generator 153, and apply the outlier removal logic for the loaded data. The outlier logic is to exclude the data below or above the 1st quartile and 3rd quartile in the way of the common 1.5*IQR standard, or to function as filtering the error data when collecting data.

[0036] As Figure 5 shown, for the XY pairs in the generated learning data, the training and test sets are separately made, and the training of the model is only performed on the learning data. The trained model is evaluated through the test set, and the ways to evaluate the matching of the model can be the Root Mean Square Error (RMSE) and the Mean Absolute Percentage Error (MAPE).

[0037] At the same time referring to Figure 1 , Figure 4 and Figure 5 , the generated data is merged with the material data (output variable) to form a learning data set, and then the preset model of the model generator 153 is trained. The model can be an integrated model between the neural network algorithm composed of 3 hidden layers and the Extreme Gradient Boosting (XGBoost) and Random Forest algorithms as machine learning methods. The form of the learning data can be formed into a table as Figure 5 shown in each width direction. Obtain the material values in the width direction, and use the process data and material data in the width direction, so that the model can be separately formed in each width direction, and the prediction model (S3) can be generated using the learning data set.

[0038] More specifically, it is a method that uses learning data to separately generate regression models with the Extreme Gradient Boosting (XGBoost) algorithm, Random Forest algorithm, and Neural network algorithm, and then uses the average value of the output values derived (predicted) in each model as the final predicted value. It shows an effect that the matching performance is further improved by more than 10% compared with the existing when using this model. More specifically, in the case of a steel grade with a yield strength of 1000 MPa, as shown in Table 2 below. (Example of yield strength prediction performance)

[0039] In addition, referring to Figure 1 、 Figure 2 and Figure 6 、 Figure 7 , when integrating the Extreme Gradient Boosting (XGBoost) algorithm, Random Forest algorithm, and Neural network algorithm, if the weights of each model are different, the matching performance of the model can be improved. The logic at this time is to derive the weights of each model through the Gradient Boosting method, that is, arbitrarily assign initial weights, and train until the Root Mean Square Error (RMSE) decreases as shown in Figure 7 . Among them, when initializing the weights, if 1 is input for each model respectively, the training speed will be further improved.

[0040] [Table 2]

[0041]

[0042] Finally, it includes a step (S4) in which the model generator 153 derives the optimal annealing temperature of the target material through the Particle Swarm Optimization (PSO) algorithm based on the prediction model.

[0043] That is, after the above prediction model is completed, the next step will be to create a model for deriving the annealing temperature among the X variables required to obtain the target material based on the prediction model. More specifically, it is a method of using the composition, hot rolling variables, and cold rolling variables as fixed variables and only the annealing temperature as a variable to find the annealing temperature required to obtain the best material. The variable of the annealing temperature can use one or more variables as variables. At this time, the algorithm will find the best annealing temperature based on particle swarm optimization. The total length data from hot rolling to annealing used can be, for example, 100 to 2000 per coil (which can vary according to the production length and production speed of the coil), and the best annealing indication value can be derived separately for each point (i.e., data) along the total width of the coil. Therefore, for the material deviation according to the total width generated in the hot rolling process, the material deviation caused by the hot rolling process can be reduced by setting different annealing temperatures according to the total width in the continuous annealing furnace, and more stable material can be ensured by matching the central value of the material specifications. As Figure 5 shown, the annealing temperature value is given for each point along the total width. As described above, when the best annealing temperature is derived, this value can be stored in the second database 152, and the best annealing temperature value stored in the second database 152 can be transmitted to the production management system 140. The best annealing temperature transmitted to the production management system 140 can be reflected in the factory design value.

[0044] Such operations can form a prediction model and a best annealing temperature derivation model respectively at points separated by a predetermined unit with a preset distance in the width direction. For example, for each 5 to 15 points.

[0045] Figure 8 FIG. is a diagram showing an exemplary computing environment capable of implementing a cold-rolled steel sheet material control system according to an embodiment of the present invention.

[0046] In Figure 8 an example of a system 1000 including a computing device 1100 configured to implement a cold-rolled steel sheet material control system according to an embodiment of the present invention is shown. For example, the computing device 1100 includes a personal computer, a server computer, a handheld or notebook device, a mobile device (mobile phone, PDA, media player, etc.), a multiprocessor system, a consumer electronic device, a microcomputer, a mainframe computer, a distributed computing environment including any of the above systems or devices, etc., but is not limited thereto.

[0047] The computing device 1100 may include at least one processing unit 1110 and a memory 1120. Among them, the processing unit 1110 may include, for example, a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc., and may include multiple cores. The memory 1120 may be a volatile memory (e.g., RAM, etc.), a non-volatile memory (e.g., ROM, flash memory, etc.), or a combination thereof.

[0048] In addition, the computing device 1100 may include a separate storage device 1130. The storage device 1130 includes a magnetic storage device, an optical storage device, etc., but is not limited thereto. In the storage device 1130, computer-readable commands for implementing one or more embodiments improved in this specification may be stored, and other computer-readable commands for implementing an operating system, an application program, etc. may be stored. The computer-readable commands stored in the storage device 1130 may be loaded into the memory 1120 for execution by the processing unit 1110.

[0049] In addition, the computing device 1100 may include (a plurality of) input devices 1140 and (a plurality of) output devices 1150. Among them, the (a plurality of) input devices 1140 may include, for example, a keyboard, a mouse, a pen, a voice input device, a touch input device, an infrared camera, a video input device, or any other input device, etc. In addition, the (a plurality of) output devices 1150 may include, for example, one or more displays, speakers, printers, or any other output device, etc. In addition, the computing device 1100 may use an input device or an output device provided in another computing device as the (a plurality of) input devices 1140 or the (a plurality of) output devices 1150.

[0050] In addition, the computing device 1100 may include (a plurality of) communication connections 1160 for communicating with another device (e.g., the computing device 1300) via a network such as the Internet 1200. Among them, the (a plurality of) communication connections 1160 may include a modem, a network interface card (NIC), an integrated network interface, a radio frequency transmitter / receiver, an infrared port, a USB connection, or other interfaces for connecting the computing device 1100 to another computing device. In addition, the (a plurality of) communication connections 1160 may include a wired connection or a wireless connection.

[0051] Each configuration of the above-described computing device 1100 can be connected through various interconnections such as a bus (e.g., Peripheral Component Interconnect (PCI), USB, FireWire (IEEE 1394), optical bus structure, etc.), or can be interconnected through a network.

[0052] Terms such as "programmable logic controller (PLC)", "process controller (PC)", "manufacturing execution system (MES)", "model generation unit", etc. used in this specification generally refer to hardware, a combination of hardware and software, software, or a computer-related entity (entity) as software in execution. For example, components such as "programmable logic controller (PLC)", "process controller (PC)", "manufacturing execution system (MES)", "model generation unit", etc. can be a process in execution on a processor, a processor, an object, an executable, an execution thread, a program, and / or a computer, but are not limited thereto. For example, an application program driven on a controller and the controller can both be components. One or more components can exist within a process and / or an execution thread, and the components can be localized on one computer or distributed between two or more computers.

[0053] As described above, according to the present invention, since an artificial intelligence model is used, a wider and larger amount of data can be utilized compared to existing physical models, thereby exhibiting more precise and accurate prediction capabilities. Additionally, in the configuration of the online model, since the system continuously drives from data generation to calculation, it contributes to the realization of an intelligent factory. In the case of cold-rolled products, the hot-rolling process is a pre-process, and in the entire process from hot-rolling to cold-rolling, since the data at each position within the coil is matched in a through-type manner, a more accurate prediction model is provided. Additionally, according to the configuration of the present invention as described above, as Figure 2 shown, material information related to the total width can be obtained, and the reasons related to the material differences at each position within the coil can be known. By providing such total-width material data, factors that reduce production capacity, such as die modification and forming condition changes caused by the sheet metal part within the coil during forming, can be solved.

[0054] The present invention described above is not limited to the above-described embodiments and drawings, but is limited by the appended claims. It will be readily apparent to those of ordinary skill in the art to which the present invention pertains that various changes and modifications can be made to its configuration without departing from the technical idea of the present invention.

Claims

1. A cold-rolled steel sheet material control system, comprising: A programmable logic controller that acquires process data of the produced cold-rolled steel sheet; A process controller that controls the manufacturing process of the cold-rolled steel sheet according to the process data of the programmable logic controller; A model generation unit that generates a prediction model for predicting the material in the width direction of the cold-rolled steel sheet to be produced by pre-learning a learning data set; And A production management system that sends and receives information with the process controller and receives the prediction value of the prediction model of the model generation unit to manage the production of the cold-rolled steel sheet, The learning data set includes: The hot rolling process data in the process data from the programmable logic controller; The material data and composition data from the production management system; and The mapping data of the measurement position between the hot rolling process data and the cold rolling and annealing process data in the process data from the programmable logic controller.

2. The cold-rolled steel sheet material control system according to claim 1, wherein The model generation unit extends the length of the hot rolling process data extended due to cold rolling and merges it with the cold rolling and annealing process data to correct the learning data set.

3. The cold-rolled steel sheet material control system according to claim 1, wherein The data acquired by the programmable logic controller from the sensor includes: the material data measured at each preset predetermined unit along the width direction of the cold-rolled steel sheet.

4. The cold-rolled steel sheet material control system according to claim 3, wherein The model generation unit generates the prediction model at each of the predetermined units where the material data is measured along the width direction of the cold-rolled steel sheet.

5. The cold-rolled steel sheet material control system according to claim 1, wherein Based on the prediction model, the model generation unit further generates an annealing temperature derivation model for deriving the optimal annealing temperature for finding the target material through a preset particle swarm optimization algorithm, and the production management system controls the process controller according to the optimal annealing temperature of the annealing temperature derivation model to control the annealing temperature of the cold-rolled steel sheet.

6. The cold-rolled steel sheet material control system according to claim 1, wherein The model generation unit generates the prediction model including a preset extreme gradient boosting algorithm, a random forest algorithm, and a neural network algorithm, and averages the prediction values of the extreme gradient boosting algorithm, the random forest algorithm, and the neural network algorithm of the prediction model to output a final prediction value.

7. The cold-rolled steel sheet material control system according to claim 6, wherein The model generation unit assigns weights to the prediction values of the extreme gradient boosting algorithm, the random forest algorithm, and the neural network algorithm of the prediction model according to a preset gradient boosting.

8. A method for generating a cold-rolled steel sheet material prediction model, comprising the following steps: The model generation unit generates a learning data set; And Training a preset model through the learning data set to generate a prediction model for predicting the material in the width direction of the cold-rolled steel sheet, The multiple data included in the learning data set include: The hot rolling process data in the process data provided by the programmable logic controller; The material data and component data provided by the production management system; and Mapping data of the measurement positions between the hot rolling process data and the cold rolling and annealing process data in the process data from the programmable logic controller, The process data provided by the programmable logic controller is the process data of the cold rolled steel sheet produced in advance through sensors.

9. The method for generating a cold rolled steel sheet material prediction model according to claim 8, wherein, In the step of generating the learning data set, The model generation unit extends the length of the hot rolling process data extended due to cold rolling and merges it with the cold rolling and annealing process data to correct the learning data set.

10. The method for generating a cold rolled steel sheet material prediction model according to claim 8, wherein, In the step of generating the learning data set, The data obtained by the programmable logic controller from the sensor includes: material data measured at each preset predetermined unit along the width direction of the cold rolled steel sheet.

11. The method for generating a cold rolled steel sheet material prediction model according to claim 10, wherein, In the step of generating the prediction model, The model generation unit generates the prediction model at each of the predetermined units where the material data is measured along the width direction of the cold rolled steel sheet.

12. The method for generating a cold rolled steel sheet material prediction model according to claim 8, further comprising the following steps: The model generation unit further generates an annealing temperature derivation model for finding the optimal annealing temperature of the target material based on the prediction model through a preset particle swarm optimization algorithm, and the production management system controls the process controller according to the optimal annealing temperature of the annealing temperature derivation model to control the annealing temperature of the cold rolled steel sheet.

13. The method for generating a cold rolled steel sheet material prediction model according to claim 8, wherein, In the step of generating the prediction model, The model generation unit generates the prediction model including a preset extreme gradient boosting algorithm, random forest algorithm, and neural network algorithm, and averages the prediction values of the extreme gradient boosting algorithm, the random forest algorithm, and the neural network algorithm of the prediction model to output a final prediction value.

14. The method for generating a cold rolled steel sheet material prediction model according to claim 13, wherein, In the step of generating the prediction model, The model generation unit assigns weights to the prediction values of the extreme gradient boosting algorithm, the random forest algorithm, and the neural network algorithm of the prediction model according to a preset gradient boost.

Images