Prediction device, learning device, prediction system, prediction method, learning method, and program
The prediction system uses machine learning to forecast unsteady bulging in continuous casting, addressing the challenge of predicting fluctuations in molten steel levels and enhancing slab quality and production efficiency.
Patent Information
- Application Number
- JP2021017982
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2021-02-08
- Publication Date
- 2024-11-14
- Estimated Expiration
- 2041-02-08
AI Technical Summary
Existing technologies struggle to predict unsteady bulging in continuous molten metal casting, which leads to fluctuations in molten steel levels, causing powder entrainment and slab quality deterioration.
A prediction system using machine learning methods, including random forest, support vector machine, and logistic regression, to analyze variables such as slab size, casting speed, molten steel temperature, and cooling water parameters to predict unsteady bulging before or during casting.
Enables accurate prediction of unsteady bulging, preventing slab quality issues and optimizing casting conditions to maintain continuous operation and improve production efficiency.
Smart Images

Figure 0007583264000001 
Figure 0007583264000002 
Figure 0007583264000003
Abstract
Description
[Technical field]
[0001] The present invention relates to a prediction device, a learning device, a prediction system, a prediction method, a learning method, and a program. [Background technology]
[0002] In a continuous casting machine for molten metal, the molten metal is supplied from a ladle to a tundish and stored therein, and then continuously poured into a mold through a submerged nozzle attached to the bottom of the tundish. The molten metal in the mold is cooled by the water-cooled mold, and a solidified shell is formed from the surface of the molten metal toward the inside, and the molten metal is continuously withdrawn from the bottom of the mold while being sandwiched between multiple pairs of rolls arranged opposite to each other. As it is withdrawn, the molten metal is cooled by a water spray, and the molten metal solidifies to the inside to produce a cast piece.
[0003] Since the rolls are set at a fixed interval, the slab containing unsolidified molten steel may bulge between the rolls, a phenomenon known as "bulging."
[0004] The area where bulging occurs does not return to its original position along the roll, and instead part of the solidified shell deforms, causing the slab to be pushed back into the unsolidified molten steel, leaving the area where bulging occurred. This phenomenon is called "unsteady bulging."
[0005] When unsteady bulging occurs along the rolls that are set at regular intervals, the solidified shell of the slab is pushed back toward the unsolidified molten steel and pushed toward the rolls periodically. This periodic disturbance causes fluctuations in the molten steel level in the mold.
[0006] Powder is scattered on the surface of the mold as a lubricant between the mold and the solidified shell. Fluctuations in the level of the molten steel in the mold entrain the powder on the surface into the molten steel. The entrained powder becomes inclusions inside the slab after solidification, which can lead to breakouts and reduced quality of the slab due to instability in the initial solidification.
[0007] Measures for preventing molten metal level fluctuations resulting from the occurrence of unsteady bulging (hereinafter referred to as "unsteady bulging-related molten metal level fluctuations") include a measure for slowing down the casting speed and increasing the thickness of the solidified shell between the rolls where bulging is occurring, a molten metal level control measure for adjusting the amount of molten steel supplied so as to counteract the molten metal level fluctuations, and a measure for increasing the strength of the solidified shell by increasing the cooling capacity by increasing the flow rate of water spray. [Prior art documents] [Patent documents]
[0008] [Patent Document 1] Japanese Patent Application Publication No. 5-57414 [Patent Document 2] JP 2006-68747 A Summary of the Invention [Problem to be solved by the invention]
[0009] However, in the past, although it was possible to take prescribed measures when unsteady bulging occurred during the casting of each charge in continuous casting, it was not possible to predict whether unsteady bulging would occur before or during the casting of each charge in continuous casting.
[0010] In view of the above circumstances, the present invention aims to provide a prediction device, a learning device, a prediction system, a prediction method, a learning method, and a program that are capable of predicting whether or not unsteady bulging will occur before or during the casting of each charge in continuous casting. [Means for solving the problem]
[0011] One aspect of the present invention is a prediction device that includes a communication unit that acquires at least one of the size of a slab, the casting speed, the temperature of molten steel, the number of continuous operations of continuous casting, the outside air temperature during casting, the composition of molten steel, the operating pattern of a group of water sprays for cooling the molten steel, and the temperature of cooling water as explanatory variables for prediction before or during casting of each charge of continuous casting, and a prediction unit that inputs the explanatory variables for prediction into a trained model that is generated based on a correlation between explanatory variables for learning corresponding to the explanatory variables for prediction and an objective variable for learning, and predicts whether or not unsteady bulging-related molten steel level fluctuations will occur based on the output of the trained model before or during casting of each charge of continuous casting, wherein the explanatory variables for learning are data including at least one of the size of the slab, the casting speed, the temperature of molten steel, the number of continuous operations of continuous casting, the outside air temperature during casting, the composition of molten steel, the operating pattern of a group of water sprays, and the temperature of cooling water in past continuous casting, and the objective variable for learning is data indicating whether or not unsteady bulging-related molten steel level fluctuations have occurred in past continuous casting.
[0012] One aspect of the present invention is a learning device that includes a communication unit that acquires explanatory variables for learning including at least one of the size of a cast piece, the casting speed, the temperature of molten steel, the number of continuous operations of continuous casting, the outside air temperature during casting, the composition of molten steel, the operating pattern of a group of water sprays for cooling the molten steel, and the temperature of cooling water, and an objective variable for learning that indicates whether or not unsteady bulging-related molten steel level fluctuations have occurred, and a learning unit that generates a trained model based on the correlation between the explanatory variables for learning and the objective variable for learning.
[0013] In one aspect of the present invention, the learning unit generates the trained model using a machine learning technique.
[0014] In one aspect of the present invention, the machine learning technique is at least one of a random forest, a support vector machine, a logistic regression analysis, and a polynomial logistic regression classification.
[0015] One aspect of the present invention is a prediction system including a learning device and a prediction device, wherein the learning device has a communication unit that acquires explanatory variables for learning including at least one of a size of a slab, a casting speed, a temperature of molten steel, a number of continuous operations of continuous casting, an outside air temperature during casting, a composition of molten steel, an operating pattern by a group of water sprays for cooling the molten steel, and a temperature of cooling water, and an objective variable for learning indicating whether or not unsteady bulging-related molten steel level fluctuation has occurred, and a learning unit that generates a trained model based on a correlation between the explanatory variables for learning and the objective variable for learning, and the prediction device acquires at least one of the size of the slab, the casting speed, the temperature of molten steel, the number of continuous operations of continuous casting, the outside air temperature during casting, the composition of molten steel, the operating pattern by a group of water sprays for cooling the molten steel, and the temperature of cooling water before or after the casting of each charge of continuous casting. the explanatory variables for prediction are input to a trained model generated based on a correlation between the explanatory variables for learning corresponding to the explanatory variables for prediction and the objective variables for learning, and a prediction unit that predicts whether or not unsteady bulging-related molten metal level fluctuations will occur based on the output of the trained model before or during casting of each charge of continuous casting, wherein the explanatory variables for learning are data including at least one of the size of the slab, the casting speed, the temperature of molten steel, the number of continuous operations of continuous casting, the outside air temperature during casting, the composition of the molten steel, the operating pattern by a water spray group, and the temperature of cooling water in past continuous casting, and the objective variable for learning is data indicating whether or not unsteady bulging-related molten metal level fluctuations have occurred in past continuous casting.
[0016] One aspect of the present invention is a prediction method executed by a prediction device, comprising a communication step of acquiring at least one of the size of a slab, the casting speed, the temperature of molten steel, the number of continuous cast runs, the outside air temperature during casting, the composition of molten steel, the operation pattern of a water spray group for cooling the molten steel, and the temperature of cooling water as an explanatory variable for prediction before or during casting of each charge of continuous casting, and inputting the explanatory variables for prediction into a trained model generated based on a correlation between explanatory variables for learning corresponding to the explanatory variables for prediction and a target variable for learning, thereby predicting the unsteady bulging molten steel surface temperature. and a prediction step of predicting whether or not level fluctuations will occur based on the output of the trained model before or during casting of each charge in continuous casting, wherein the explanatory variables for learning are data including at least one of the size of the slab, the casting speed, the temperature of the molten steel, the number of continuous operations of continuous casting, the outside air temperature during casting, the composition of the molten steel, the operating pattern of the water spray group, and the temperature of the cooling water in past continuous castings, and the objective variable for learning is data indicating whether or not unsteady bulging-related molten steel level fluctuations have occurred in past continuous castings.
[0017] One aspect of the present invention is a learning method executed by a learning device, the learning method including a communication step of acquiring explanatory variables for learning including at least one of the size of a cast piece, the casting speed, the temperature of molten steel, the number of continuous operations of continuous casting, the outside air temperature during casting, the composition of molten steel, the operating pattern of a group of water sprays for cooling the molten steel, and the temperature of cooling water, and an objective variable for learning indicating whether or not unsteady bulging-related molten steel level fluctuations have occurred, and a learning step of generating a learned model based on the correlation between the explanatory variables for learning and the objective variable for learning.
[0018] One aspect of the present invention is a program for causing a computer to function as a prediction device.
[0019] One aspect of the present invention is a program for causing a computer to function as a learning device. Effect of the Invention
[0020] According to the present invention, it is possible to predict whether or not unsteady bulging will occur before or during the casting of each charge in continuous casting. [Brief description of the drawings]
[0021] [Figure 1] FIG. 1 is a diagram showing a configuration example of a continuous casting device in an embodiment. [Diagram 2] 4 is a flowchart showing an example of the operation of the learning device in the embodiment. [Diagram 3] 1 is a flowchart illustrating an example of the operation of a prediction device in an embodiment. [Figure 4] FIG. 13 is a diagram illustrating an example of evaluation of a pre-prediction result of logistic regression classification in an embodiment. [Diagram 5] FIG. 13 is a diagram illustrating an example of evaluation of a pre-prediction result of polynomial logistic regression classification in an embodiment. [Figure 6] FIG. 13 is a diagram illustrating an example of evaluation of a pre-prediction result of class classification (linear kernel) using a support vector machine in an embodiment. [Figure 7] FIG. 13 is a diagram showing an example of evaluation of a pre-prediction result of class classification using a support vector machine (non-linear kernel) in an embodiment. [Figure 8] FIG. 13 is a diagram showing an example of evaluation of prior prediction results using random forest in an embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0022] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described in detail with reference to the drawings. (First embodiment) 1 is a diagram showing an example of the configuration of a continuous casting apparatus 1 in an embodiment. The continuous casting apparatus 1 includes a continuous casting machine 2, a control device 3, a driving device 4, a communication line 5, and a prediction system 6. The continuous casting machine 2 includes a tundish 20, a sliding nozzle 21, a mold 22, a roll group 23, a spray group 24, a temperature sensor 25, and a level sensor 26.
[0023] The prediction system 6 includes a learning device 60 and a prediction device 61. The learning device 60 includes a communication unit 600, a storage unit 601, and a learning unit 602. The prediction device 61 includes a communication unit 610, a storage unit 611, a prediction unit 612, and a display unit 613.
[0024] Some or all of the devices of the prediction system 6 are realized as software by a processor such as a CPU (Central Processing Unit) executing a program deployed from a storage unit, which is a non-volatile recording medium (non-transient recording medium), to a memory. The program may be recorded on a computer-readable recording medium. The computer-readable recording medium is, for example, a portable medium such as a flexible disk, a magneto-optical disk, a ROM (Read Only Memory), a CD-ROM (Compact Disc Read Only Memory), or a non-transient recording medium such as a storage device built into a computer system, such as a hard disk. The program may be transmitted to the prediction system 6 via a communication line 5.
[0025] The communication line 5 is a line such as a LAN (Local Area Network) or a WAN (Wide Area Network), etc. The communication line 5 transmits communication data (e.g., control signals, numbers, explanatory variables, and objective variables) between the control device 3, the learning device 60, and the prediction device 61.
[0026] Some or all of the devices in the prediction system 6 may be realized using hardware including electronic circuits (electronic circuits or circuitry) using, for example, an LSI (Large Scale Integration circuit), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array).
[0027] The temperature sensor 25 is a sensor that measures temperature, such as a thermo camera. The temperature sensor 25 measures the temperature of the molten steel 100 before it is poured into the tundish 20 during casting of each charge. The temperature sensor 25 transmits the measured value of the temperature of the molten steel 100 to the control device 3 at regular intervals.
[0028] The level sensor 26 measures the surface level of the molten steel 100 in the mold 22. The level sensor 26 outputs the measurement value of the molten steel surface level "h" to the control device 3 at regular intervals.
[0029] The control device 3 is a device that controls the operation of each part of the continuous casting machine 2. The control device 3 determines whether or not unsteady bulging-related molten metal level fluctuation has occurred based on the measurement value (sampling result) of the molten metal level "h" (Reference document 1: JP 2011-143414 A). The control device 3 transmits the casting number of each charge in the continuous casting to the prediction system 6 during (immediately after) casting of each charge.
[0030] In a learning stage prior to the prediction stage, the control device 3 transmits to the learning device 60 explanatory variables for learning corresponding to the explanatory variables for prediction, objective variables for learning, and the casting number of each charge in the continuous casting.
[0031] In the prediction stage, the control device 3 transmits explanatory variables for prediction and the casting number of each charge in the continuous casting to the prediction device 61.
[0032] The control device 3 obtains the prediction result of whether or not unsteady bulging-related molten metal level fluctuation will occur from the prediction device 61 during the casting of each charge.
[0033] The control device 3 changes the operating pattern of the water spray group for cooling the molten steel and the casting speed based on the prediction results, thereby preventing the occurrence of unsteady bulging before the casting of each charge.
[0034] In the first embodiment, the prediction system 6 is a system that predicts whether or not unsteady bulging-related molten metal level fluctuation will occur during the casting of each charge in continuous casting.
[0035] Next, the learning device 60 (learning stage) will be described in detail. (A) The larger the size of the slab in the mold 22, the higher the possibility of insufficient cooling of the molten steel 100. Therefore, there is a correlation between the size of the slab and the occurrence of unsteady bulging. (B) The faster the casting speed, the higher the possibility of insufficient cooling of the molten steel 100. Therefore, there is a correlation between the casting speed and the occurrence of unsteady bulging. (C) The higher the temperature of the molten steel 100, the greater the possibility of insufficient cooling of the molten steel 100. Therefore, there is a correlation between the temperature of the molten steel 100 and the occurrence or non-occurrence of unsteady bulging. (D) The more the number of continuous operations of continuous casting is increased, the higher the temperatures of the peripheral equipment such as the roll group 23 and the cooling water become, and the higher the possibility of insufficient cooling of the molten steel 100 becomes. Therefore, there is a correlation between the number of continuous operations of continuous casting and the occurrence of unsteady bulging. (E) The higher the outside air temperature, the slower the cooling rate of the molten steel 100 and the surrounding equipment becomes, and therefore the higher the possibility of insufficient cooling of the molten steel 100 increases. Therefore, there is a correlation between the outside air temperature and the occurrence of unsteady bulging. (F) The possibility of insufficient cooling of the molten steel 100 changes due to, for example, changes in the solidification temperature depending on the composition (steel type) of the molten steel 100. Therefore, there is a correlation between the composition of the molten steel 100 and the occurrence or non-occurrence of unsteady bulging. (G) The possibility of insufficient cooling of the molten steel 100 varies depending on the pattern of the spray group 24 that cools the molten steel 100. Therefore, there is a correlation between the pattern of the spray group 24 that cools the molten steel 100 (e.g., the arrangement pattern of multiple sprays, the amount of cooling water) and the occurrence of unsteady bulging. (H) The higher the temperature of the cooling water, the higher the possibility of insufficient cooling of the molten steel 100. Therefore, there is a correlation between the temperature of the cooling water and the occurrence of unsteady bulging.
[0036] In the learning stage, the communication unit 600 acquires from the control device 3 explanatory variables for learning, objective variables for learning, and the casting number of each charge in the continuous casting.
[0037] The storage unit 601 stores a learning model of machine learning. The storage unit 601 stores explanatory variables and objective variables for each casting number of a charge. The storage unit 601 stores algorithm data representing a machine learning technique as a program executed by the learning unit 602. The storage unit 601 may store the learning model updated by the learning unit 602 as a learned model.
[0038] The learning unit 602 updates the learning model using a machine learning technique based on the correlation between the explanatory variables for learning and the target variables for learning, thereby generating a trained model. The machine learning technique is not limited to a specific machine learning technique. For example, the machine learning technique may be any of a random forest, a support vector machine, a logistic regression analysis, and a polynomial logistic regression classification.
[0039] The learning unit 602 acquires explanatory variables for learning from the storage unit 601. The learning unit 602 inputs the explanatory variables for learning to a learning model. The learning unit 602 derives an error between the output of the learning model and correct answer data. The learning unit 602 updates the learning model using, for example, an error backpropagation method so as to reduce the error.
[0040] Next, the prediction device 61 (prediction stage) will be described in detail. In the prediction stage, the communication unit 610 acquires explanatory variables for prediction from the control device 3 during the casting of each charge in the continuous casting. In the first embodiment, the operation data is data representing at least one of the set values of the operation conditions and the measured values (actual values) of the operation. The communication unit 610 acquires a machine learning trained model from the communication unit 600.
[0041] The storage unit 611 stores a learned model of machine learning. The storage unit 611 stores each casting number of the charge. The storage unit 611 stores algorithm data representing the machine learning method as a program executed by the prediction unit 612.
[0042] The prediction unit 612 inputs explanatory variables for prediction into the trained model and predicts whether or not unsteady bulging-related molten metal level fluctuations will occur (prediction result) based on the output of the trained model (0: will not occur, 1: will occur) during casting of each charge in continuous casting.
[0043] The display unit 613 is a display device such as a liquid crystal display. For example, the display unit 613 displays an image indicating whether or not unsteady bulging-related molten metal level fluctuation will occur (prediction result) during casting of each charge. The display unit 613 may include a touch panel. The touch panel accepts predetermined operations. A signal corresponding to the operation accepted by the touch panel is transmitted to the learning unit 602 and the prediction unit 612 via the communication unit 610 as, for example, an instruction signal.
[0044] Next, an example of the operation of the prediction system 6 will be described. FIG. 2 is a flowchart showing an example of the operation of the learning device 60 in an embodiment. The operation shown in FIG. 2 is executed in a learning stage prior to the prediction stage. The communication unit 600 judges whether the executed charge casting is a learning target charge or not. Whether or not it is a learning target is determined in advance by, for example, a person in charge of operating the continuous casting system (step S101). If it is determined that the executed charge casting is a charge other than the learning target charge (step S101: NO), the communication unit 600 re-executes the process of step S101 after a predetermined time has elapsed.
[0045] If it is determined that the executed charge casting is the casting of the charge to be learned (step S101: YES), the communication unit 600 acquires explanatory variables for learning from the control device 3 (step S102). The communication unit 600 acquires objective variables for learning from the control device 3 (step S103).
[0046] The storage unit 601 acquires explanatory variables for learning and objective variables for learning from the communication unit 600. The learning unit 602 acquires explanatory variables for learning and objective variables for learning from the storage unit 601. The learning unit 602 inputs explanatory variables for learning to a learning model (step S104). The learning unit 602 derives an error between the output of the learning model and the objective variables for learning (step S105). The learning unit 602 updates the learning model using, for example, an error backpropagation method so as to reduce the error (step S106).
[0047] The learning unit 602 determines whether to end the learning stage based on a predetermined instruction signal (step S107). If it is determined that the learning stage should be continued (step S107: NO), the learning unit 602 returns the process to step S101. If it is determined that the learning stage should be ended (step S107: YES), the learning unit 602 records the updated explanatory variables for learning and the trained model in the storage unit 601 (step S108).
[0048] Fig. 3 is a flowchart showing an example of the operation of the prediction device 61 in the embodiment. The operation shown in Fig. 3 is executed in a prediction stage after the learning stage, for example, before the casting of each charge.
[0049] The communication unit 610 acquires the trained model from the learning device 60 (step S201). The communication unit 610 determines whether the casting of the charge in the continuous casting to be performed is the casting of the charge to be predicted. Whether or not it is the prediction target is determined in advance by, for example, a person operating the continuous casting apparatus 1 (step S202). If it is determined that the casting of the charge to be performed is the casting of a charge other than the prediction target (step S202: NO), the communication unit 610 re-executes the process of step S202 after a predetermined time has elapsed.
[0050] When it is determined that the casting of the charge in the continuous casting to be executed is the casting of the charge to be predicted (step S202: YES), the communication unit 610 acquires explanatory variables for prediction from the control device 3. The storage unit 611 stores the acquired explanatory variables for prediction. The prediction unit 612 acquires the stored explanatory variables for prediction from the storage unit 611 (step S203).
[0051] The prediction unit 612 acquires the trained model from the storage unit 611. The prediction unit 612 may acquire the trained model from the learning device 60 using the communication unit 610. The prediction unit 612 inputs explanatory variables for prediction to the trained model (step S204). The prediction unit 612 predicts whether or not unsteady bulging-related molten metal level fluctuations will occur (whether or not unsteady bulging will occur) based on the output of the trained model (0: will not occur, 1: will occur) (step S205). If it is predicted that unsteady bulging-related molten metal level fluctuations will not occur (unsteady bulging will not occur) (step S205: NO), the prediction unit 612 returns the process to step S202.
[0052] When it is predicted that unsteady bulging-related molten metal level fluctuation will occur (unsteady bulging will occur) (step S205: YES), the prediction unit 612 notifies, for example, the control device 3 or an operator, that unsteady bulging will occur. The prediction unit 612 may cause the display unit 613 to display that unsteady bulging will occur (step S206).
[0053] The control device 3 may control the operation of the drive device 4 based on the notification. For example, the control device 3 may change the operation pattern of the spray group 24 for cooling the molten steel 100 during the casting of each charge so as to obtain appropriate operating conditions. The control device 3 may present appropriate operating conditions (e.g., appropriate operating pattern and casting speed) to the operator based on the notification. If appropriate operating conditions do not exist, the control device 3 may issue a warning to the operator based on the notification.
[0054] Next, we will explain an example of evaluating prediction results using machine learning techniques (supervised learning, classification problems).
[0055] The predicted value is 1 when it is predicted that unsteady bulging-type mold level fluctuations will occur, and is 0 when it is predicted that unsteady bulging-type mold level fluctuations will not occur. The actual measured value (measured value) is 1 when unsteady bulging-type mold level fluctuations actually occur, and is 0 when unsteady bulging-type mold level fluctuations do not actually occur.
[0056] In what follows, a true (correct) predicted value (the predicted value and the actual measured value are the same) will be denoted as "T". A false (incorrect) predicted value (the predicted value and the actual measured value are different) will be denoted as "F". A positive value (=1) will be denoted as "P". A negative value (=0) will be denoted as "N". In what follows, a true positive will be denoted as "TP". A false negative will be denoted as "FN". A false positive (overdetection) will be denoted as "FP". A true negative will be denoted as "TN".
[0057] As an example, the evaluation data including the combination of explanatory variables for learning and objective variables for learning is 378 points (=TP+FN+FP+TN).
[0058] The objective variable for learning is the occurrence or non-occurrence of unsteady bulging-related molten metal level fluctuation in past continuous castings (0: no occurrence, 1: occurrence). The explanatory variables for learning are the size of the slab (mm 2), casting speed (m / min), temperature of the molten steel 100 (°C), number of continuous operations of continuous casting (number of continuous castings), outside air temperature (°C), composition of the molten steel 100 (carbon, silicon, manganese, phosphorus, sulfur), and an identification number of an operation pattern for the spray group 24 that cools the molten steel 100 with water. The explanatory variables for learning may further include the temperature of the cooling water (mist) and the amount of cooling water. The explanatory variables for learning may be prepared for each season (each predetermined period in the year).
[0059] FIG. 4 is a diagram showing an example of evaluation of prediction results of logistic regression classification in an embodiment. In FIG. 4, the regularization strength parameter "C" is 1. That is, the parameter "C" is a parameter that indicates the degree of misclassification that is tolerated. Therefore, the larger the value of the parameter "C", the higher the accuracy rate is likely to be.
[0060] The precision at the prediction stage is 0.567 (=Precision=TP / (TP+FP)=34 / (34+26)). The recall at the prediction stage is 0.548 (=Recall=TP / (TP+FN)=34 / (34+28)). The F1 value (harmonic mean of recall and precision) at the prediction stage is 0.557 (=2×Recall×Precision / (Recall+Precision)=2×0.548×0.567 / (0.548+0.567)). The F1 value at the training stage was 0.769.
[0061] FIG. 5 is a diagram showing an example of evaluation of prediction results of polynomial logistic regression classification in an embodiment. In FIG. 5, the regularization strength parameter "C" is 50. The precision rate at the prediction stage is 0.583 (=28 / (28+20)). The recall rate at the prediction stage is 0.451 (=28 / (28+34)). The F1 value at the prediction stage is 0.509 (=2×0.451×0.583 / (0.451+0.583)). The F1 value at the learning stage was 0.857.
[0062] FIG. 6 is a diagram showing an example of evaluation of prediction results of Support Vector Classification (SVC) (linear kernel) using a Support Vector Machine (SVM) in an embodiment. In FIG. 6, the regularization strength parameter "C" is 10. The precision rate at the prediction stage is 0.529 (=36 / (36+32)). The recall rate at the prediction stage is 0.581 (=36 / (36+26)). The F1 value at the prediction stage is 0.554 (=2×0.581×0.529 / (0.581+0.529)). The F1 value at the learning stage was 0.780.
[0063] FIG. 7 is a diagram showing an example of evaluation of prediction results of class classification using a support vector machine (nonlinear kernel) in an embodiment. In FIG. 7, the regularization strength parameter "C" is 10. A squared polynomial is used for class classification. The parameter "γ" indicates which of points in the point distribution map that are close to the line and points that are far from the line are emphasized. The value of the parameter "γ" is 0.01. The larger the value of "γ", the more importance is placed on points in the point distribution map that are close to the line.
[0064] The precision rate at the prediction stage is 0.396 (= 42 / (42 + 64)). The recall rate at the prediction stage is 0.677 (= 42 / (42 + 20)). The F1 score at the prediction stage is 0.500 (= 2 × 0.677 × 0.396 / (0.677 + 0.396)). The F1 score at the training stage was 0.783.
[0065] FIG. 8 is a diagram showing an example of evaluation of prediction results using random forest in an embodiment. The precision rate at the prediction stage is 1.000 (=18 / (18+0)). The recall rate at the prediction stage is 0.290 (=18 / (18+44)). The F1 value at the prediction stage is 0.450 (=2×1.000×0.290 / (1.000+0.290)). The F1 value at the learning stage was 0.977.
[0066] Even in a case where unsteady bulging-type molten metal level fluctuation does not occur, if unnecessary measures to prevent unsteady bulging-type molten metal level fluctuation are taken before casting of each charge, normal continuous operation of continuous casting will be erroneously stopped. For this reason, it is preferable that the number of false positives "FP" is small. In the prediction result using the random forest, the number of false positives "FP" is zero. Therefore, the prediction unit 612 may predict the occurrence or non-occurrence of unsteady bulging-type molten metal level fluctuation by using the random forest preferentially. This can reduce the possibility that the prediction unit 612 erroneously stops normal continuous operation of continuous casting.
[0067] From the above, it is possible to predict whether or not unsteady bulging-related molten metal level fluctuations will occur during the casting of each charge in continuous casting. This makes it possible to prevent breakouts and deterioration of the quality of the slab due to instability in initial solidification. In addition, by optimizing the casting conditions using the learning model, it becomes possible to increase the casting speed and the temperature of the slab at the end of the casting.
[0068] By making it possible to predict the unsteady bulging-related molten steel level fluctuation, it is possible to optimize the casting conditions according to the composition of the molten steel (for example, the carbon content) and the cooling water temperature.
[0069] Second embodiment The second embodiment differs from the first embodiment in that some of the explanatory variables for prediction are replaced with virtual data for prediction (virtual set values and virtual measured values). The virtual data is data (set values and measured values) that is determined in advance using a simulation or the like. The part of the explanatory variables for prediction is, for example, an explanatory variable among the explanatory variables for prediction whose value is predicted to change during casting. The second embodiment will mainly be described with respect to the differences from the first embodiment.
[0070] The temperature sensor 25 measures the temperature of the molten steel 100 poured into the tundish 20 during (immediately after) the casting of each charge. The communication unit 610 acquires explanatory variables for prediction from the control device 3 during the casting of each charge in continuous casting. The prediction unit 612 replaces some of the explanatory variables for prediction with virtual data for prediction. The prediction unit 612 inputs the partially replaced explanatory variables for prediction into the trained model, and predicts whether or not unsteady bulging-related molten steel level fluctuations will occur (prediction result) based on the output of the trained model (0: will not occur, 1: will occur) during the casting of each charge in continuous casting.
[0071] The control device 3 obtains the prediction result of whether or not unsteady bulging-related molten metal level fluctuation will occur from the prediction device 61 during the casting of each charge. The control device 3 may present appropriate operating conditions (e.g., appropriate operation pattern and casting speed) to the operator based on the notification. The control device 3 may also issue a warning to the operator based on the notification. This prevents the occurrence of unsteady bulging during the casting of each charge.
[0072] In this way, some of the explanatory variables for prediction are replaced with virtual data for prediction. The set of explanatory variables for prediction with some of the variables replaced is input to the trained model. Such a trained model outputs a prediction result. This makes it possible to predict whether or not unsteady bulging-related molten metal level fluctuations will occur while each charge is being cast in continuous casting. During operation, it is possible to predict the occurrence of unsteady bulging due to active factors (e.g., an increase in the casting speed) and optimize the operating conditions.
[0073] Third embodiment The third embodiment differs from the first and second embodiments in that whether or not unsteady bulging-related molten metal level fluctuation will occur is predicted before casting of each charge in continuous casting. In the third embodiment, the differences from the first and second embodiments will be mainly described.
[0074] In the third embodiment, all explanatory variables for prediction are virtual data for prediction. In other words, the explanatory variables for prediction do not include operational data. In the third embodiment, the prediction system 6 is a system that predicts whether or not unsteady bulging-related molten metal level fluctuation will occur before casting of each charge in continuous casting. For example, the operation shown in FIG. 3 is performed before casting of each charge.
[0075] The communication unit 610 acquires explanatory variables that do not include operational data (measurement values, etc.) from the control device 3 as explanatory variables for prediction before each charge is cast in continuous casting. The explanatory variables that do not include operational data are determined in advance, for example, by using a simulation. The prediction unit 612 inputs the explanatory variables for prediction into the trained model, and predicts whether or not unsteady bulging-related molten metal level fluctuations will occur based on the output of the trained model (0: will not occur, 1: will occur) before each charge is cast in continuous casting.
[0076] The control device 3 obtains the prediction result of whether or not unsteady bulging-related molten metal level fluctuation will occur from the prediction device 61 before casting of each charge is started. The control device 3 may present appropriate operating conditions (e.g., appropriate operation pattern and casting speed) to the operator based on the notification. If appropriate operating conditions do not exist, the control device 3 may issue a warning to the operator based on the notification. This prevents the occurrence of unsteady bulging before casting of each charge is started.
[0077] In this manner, the explanatory variable set for simulation-based prediction is input to the trained model. The trained model outputs a prediction result using the explanatory variable set for simulation-based prediction. This makes it possible to predict whether or not unsteady bulging-related molten metal level fluctuations will occur before casting of each charge in continuous casting. It is possible to predict the occurrence of unsteady bulging due to passive factors (e.g., temperature fluctuations in the tundish 20 after ladle replacement) before operation and to reflect this in the operating conditions for the next operation (e.g., conditions for determining whether or not to slow down the casting speed).
[0078] Although an embodiment of the present invention has been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and designs that do not deviate from the gist of the present invention are also included. [Explanation of symbols]
[0079] 1...continuous casting device, 2...continuous casting machine, 3...control device, 4...drive device, 5...communication line, 6...prediction system, 20...tundish, 21...sliding nozzle, 22...mold, 23...roll group, 24...spray group, 25...temperature sensor, 26...level sensor, 60...learning device, 61...prediction device, 100...molten steel, 101...outer shell, 600...communication unit, 601...memory unit, 602...learning unit, 610...communication unit, 611...memory unit, 612...prediction unit, 613...display unit
Claims
1. a communication unit that acquires, as explanatory variables for prediction, the size of the slab, the casting speed, the temperature of the molten steel, the number of continuous operations of the continuous casting, the composition of the molten steel, and an operation pattern of a water spray group for cooling the molten steel before or during the casting of each charge of the continuous casting; a prediction unit that inputs the explanatory variables for prediction into a trained model that is generated based on a correlation between explanatory variables for learning corresponding to the explanatory variables for prediction and a target variable for learning, and predicts whether or not unsteady bulging-related molten metal level fluctuation will occur based on an output of the trained model before or during casting of each charge in continuous casting; Equipped with The explanatory variables for learning are data including a size of a cast piece, a casting speed, a temperature of molten steel, a number of continuous operations of continuous casting, a composition of molten steel, and an operation pattern of a water spray group in past continuous casting, The learning objective variable is data representing whether or not unsteady bulging-related molten metal level fluctuation has occurred in past continuous casting. Prediction device.
2. The communication unit further acquires at least one of an outside air temperature and a cooling water temperature during casting as an explanatory variable for the prediction. The prediction device according to claim 1 .
3. a communication unit that acquires explanatory variables for learning, including the size of the slab, the casting speed, the temperature of the molten steel, the number of continuous operations of the continuous casting, the composition of the molten steel, and an operating pattern of a group of water sprays for cooling the molten steel, and a target variable for learning that indicates whether or not unsteady bulging-related molten steel level fluctuation has occurred; a learning unit that generates a trained model based on the correlation between the explanatory variables for learning and the target variables for learning; A learning device comprising:
4. The communication unit further acquires at least one of an outside air temperature and a cooling water temperature during casting as an explanatory variable for the learning. The learning device according to claim 3.
5. The learning unit generates the trained model using a machine learning technique. The learning device according to claim 3 or 4.
6. The machine learning method is at least one of a random forest, a support vector machine, a logistic regression analysis, and a polynomial logistic regression classification. The learning device according to claim 5 .
7. A prediction system including a learning device and a prediction device, The learning device includes: a first communication unit that acquires explanatory variables for learning including a size of a slab, a casting speed, a temperature of molten steel, a number of continuous operations of continuous casting, a composition of molten steel, and an operating pattern of a group of water sprays for cooling the molten steel, and a target variable for learning that indicates whether or not unsteady bulging-related molten steel level fluctuation has occurred; A learning unit that generates a trained model based on a correlation between the learning explanatory variables and the learning objective variables, The prediction device comprises: a second communication unit that acquires, as explanatory variables for prediction, the size of the slab, the casting speed, the temperature of the molten steel, the number of continuous operations of the continuous casting, the composition of the molten steel, and an operation pattern of a group of water sprays for cooling the molten steel before or during the casting of each charge of the continuous casting; a prediction unit that inputs the explanatory variables for prediction into a trained model that is generated based on a correlation between the explanatory variables for learning corresponding to the explanatory variables for prediction and the target variables for learning, and predicts whether or not unsteady bulging-related molten metal level fluctuations will occur based on an output of the trained model before or during casting of each charge in continuous casting, The explanatory variables for learning are data including a size of a cast piece, a casting speed, a temperature of molten steel, a number of continuous operations of continuous casting, a composition of molten steel, and an operation pattern of a water spray group in past continuous casting, The learning objective variable is data representing whether or not unsteady bulging-related molten metal level fluctuation has occurred in past continuous casting. Prediction system.
8. The first communication unit and the second communication unit further acquire at least one of an outside air temperature and a cooling water temperature during casting as an explanatory variable for learning and an explanatory variable for prediction. The prediction system of claim 7.
9. A prediction method executed by a prediction device, comprising: A communication step of acquiring the size of the slab, the casting speed, the temperature of the molten steel, the number of continuous cast runs, the composition of the molten steel, and the operation pattern of the water spray group for cooling the molten steel as explanatory variables for prediction before or during the casting of each charge of the continuous casting; a prediction step of inputting the explanatory variables for prediction into a trained model that has been generated based on a correlation between explanatory variables for learning corresponding to the explanatory variables for prediction and a target variable for learning, and predicting whether or not unsteady bulging-related molten metal level fluctuation will occur based on an output of the trained model before or during casting of each charge in continuous casting; Including, The explanatory variables for learning are data including a size of a cast piece, a casting speed, a temperature of molten steel, a number of continuous operations of continuous casting, a composition of molten steel, and an operation pattern of a water spray group in past continuous casting, The learning objective variable is data representing whether or not unsteady bulging-related molten metal level fluctuation has occurred in past continuous casting. Forecasting methods.
10. The communication step further includes acquiring at least one of an outside air temperature and a cooling water temperature during casting as an explanatory variable for the prediction. The prediction method according to claim 9.
11. A learning method executed by a learning device, comprising: a communication step of acquiring explanatory variables for learning including the size of the slab, the casting speed, the temperature of the molten steel, the number of continuous operations of the continuous casting, the composition of the molten steel, and an operating pattern of a group of water sprays for cooling the molten steel, and an objective variable for learning indicating whether or not unsteady bulging-related molten steel level fluctuation has occurred; a learning step of generating a trained model based on the correlation between the explanatory variables for learning and the target variables for learning; Learning methods including:
12. The communication step further includes acquiring at least one of an outside air temperature and a cooling water temperature during casting as an explanatory variable for the learning. The learning method according to claim 11.
13. A program for causing a computer to function as the prediction device according to claim 1 or 2.
14. A program for causing a computer to function as the learning device according to any one of claims 3 to 6.
Citation Information
Patent Citations
Method for preventing bulging in continuously cast slab
JP1993057414A
Method for preventing variation of level of molten metal due to bulging in continuous casting mold
JP2006068747A
Method for detecting unsteady bulging in continuously cast slab
JP2007136480A
Method for controlling molten metal surface level within mold in continuous casting machine
JP2007275924A
Instrument and method for controlling molten metal surface level in continuous casting machine
JP2008290082A
