JP7617414B2 - Slag removal system, method for generating guidance for slag removal system, and method and program for automatically controlling slag removal system - Google Patents

Description

The present invention relates to a slag removal system, a method for generating guidance for a slag removal system, and an automatic control method and program for a slag removal system.

Patent Document 1 describes a melting operation device that includes a drive mechanism and a plurality of work tools operated by the drive mechanism. In the technology described in Patent Document 1, slag, which is impurities separated from metal by melting, is discharged from a melting furnace. The melting operation device performs operations on molten metal produced by melting material in a melting furnace. The melting operation device includes a teach pendant and an input device, and uses these to perform various operations such as removing slag, which is impurities separated from metal by melting.
More specifically, in the technology described in Patent Document 1, a program created by a teach pendant is stored in a control device. An operator operates the drive mechanism via the teach pendant to perform teaching, and the control device learns and records the operation, and the operation of the drive mechanism is stored in the control device. The control device controls the drive of the drive mechanism by, for example, reproducing the operation learned by teaching.
However, in the technology described in Patent Document 1, the slag removal operation is learned without taking into consideration the amount of sulfur recovery, and therefore, the technology described in Patent Document 1 cannot realize slag removal while suppressing the amount of sulfur recovery and the amount of molten iron loss.

Patent Document 2 describes an industrial robot that removes slag (impurities such as metal oxides, slag) that occurs in molten metal in a melting furnace. In the technology described in Patent Document 2, the industrial robot corrects the working position of a hand device based on information about the height of the molten metal surface measured by a radar. The industrial robot also samples the molten metal, roughly removes the slag, and removes the molten metal from the melting furnace.
However, in the technology described in Patent Document 2, slag is removed without considering the amount of re-sulfurization. Therefore, even with the technology described in Patent Document 2, it is not possible to realize slag removal while suppressing the amount of re-sulfurization and the amount of molten iron loss.

Patent Document 3 describes an automatic slag removal device that automatically removes slag generated in molten metal in a melting furnace. In the technology described in Patent Document 3, a camera photographs the entire surface of the molten metal in a melting furnace from a predetermined angle, and an arm is operated by electronic control to remove the slag in the melting furnace. The control unit determines whether or not slag removal is necessary based on the image data photographed by the camera, determines the slag removal content if necessary, and electronically controls the arm based on the determined content to perform the slag removal process.
In detail, in the technology described in Patent Document 3, machine learning is performed by using correct answer data indicating whether or not slag is present so as to be able to determine whether or not slag has occurred in each of multiple areas contained in a captured image.
However, in the technology described in Patent Document 3, machine learning and slag removal are performed without taking into consideration the amount of re-Sulfurization. Therefore, even with the technology described in Patent Document 3, it is not possible to realize slag removal that suppresses the amount of re-Sulfurization and the amount of molten iron loss.

In typical slag removal operations, the operator of the slag removal equipment operates the equipment while watching the surface of the molten metal on a display monitor or directly with the naked eye, and the slag removal rate and molten iron yield during slag removal operations tend to depend on the skill of the operator. For this reason, variations in the slag removal rate and molten iron yield occur due to the presence of operators with different levels of proficiency and the influence of the operating environment.

JP 2019-105389 A JP 2018-179348 A JP 2020-085395 A JP 2020-112429 A

In view of the above, the present invention aims to provide a slag disposal system that can realize slag disposal with reduced S and molten iron loss, a method for generating guidance for the slag disposal system, and an automatic control method and program for the slag disposal system.

One aspect of the present invention is a slag removal system comprising a vessel for storing molten metal, a first actuator for at least tilting the vessel, a first detection unit for detecting at least the tilt angle of the vessel, a slag removal member for removing slag from the surface of the molten metal in the vessel, a second actuator for moving the slag removal member, a second detection unit for detecting the amount of displacement of the second actuator, an operation unit for receiving an operation from an operator, a control unit for controlling the first actuator and the second actuator in response to the operation received by the operation unit, an imaging unit for capturing an image of the surface of the molten metal in the vessel, a weight distribution calculation unit for calculating a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit, a guidance generation unit for generating guidance for the operation of the operator with respect to the operation unit, and a guidance output unit for outputting the guidance generated by the guidance generation unit, wherein the guidance generation unit generates a guidance for the operation of the operator with respect to the operation unit after the slag removal member has removed slag. The system includes a learning model unit that calculates a slag removal operation schedule, which is a schedule for the operation of the first actuator and the second actuator to remove slag using the slag removal member, based on a target value for the amount of slag remaining after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel, and an AI model unit that generates guidance for the operator's operation on the operation unit to realize the slag removal operation schedule calculated by the learning model unit based on the target value for the amount of slag remaining after slag removal. The learning model unit performs supervised learning using a set of past operation data of the slag removal system and the slag removal performance when the past operation data of the slag removal system was obtained as teacher data, and the AI model unit modifies the guidance for the operator's operation on the operation unit based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator in response to the operation received by the operation unit.

One aspect of the present invention is a guidance generation method for a slag removal system comprising a vessel for containing molten metal, a first actuator for at least tilting the vessel, a first detection unit for detecting at least the inclination angle of the vessel, a slag removal member for removing slag from the surface of the molten metal in the vessel, a second actuator for moving the slag removal member, a second detection unit for detecting the amount of displacement of the second actuator, an operation unit for receiving an operation from an operator, a control unit for controlling the first actuator and the second actuator in response to the operation received by the operation unit, an imaging unit for imaging the surface of the molten metal in the vessel, a weight distribution calculation unit for calculating a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit, a guidance generation unit for generating guidance for the operation of the operator with respect to the operation unit, and a guidance output unit for outputting the guidance generated by the guidance generation unit, the guidance generation unit being configured to generate guidance for the operation of the operator with respect to the operation unit, and The method includes a calculation step of calculating a slag removal operation schedule, which is a schedule of the operation of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of the amount of slag remaining after slag removal, which is the amount of residual slag remaining on the surface of the molten metal in the slag removal system; a generation step of generating guidance for the operator's operation on the operation unit to realize the slag removal operation schedule calculated in the calculation step based on the target value of the amount of slag remaining after slag removal; a learning step of performing supervised learning using a set of past operation data of the slag removal system and the slag removal performance when the past operation data of the slag removal system was obtained as teacher data; and a correction step of correcting the guidance for the operator's operation on the operation unit based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator in response to the operation received by the operation unit.

One aspect of the present invention is a slag removal system including a container for storing molten metal, a first actuator for at least tilting the container, a first detection unit for detecting at least the tilt angle of the container, a slag removal member for removing slag from the surface of the molten metal in the container, a second actuator for moving the slag removal member, a second detection unit for detecting the amount of displacement of the second actuator, an operation unit for receiving an operation from an operator, a control unit for controlling the first actuator and the second actuator in response to the operation received by the operation unit, an imaging unit for capturing an image of the surface of the molten metal in the container, a weight distribution calculation unit for calculating a weight distribution of residual slag remaining on the surface of the molten metal in the container based on an image of the surface of the molten metal in the container captured by the imaging unit, a guidance generation unit for generating guidance for the operation of the operator with respect to the operation unit, and a guidance output unit for outputting the guidance generated by the guidance generation unit. A program for executing a calculation step of calculating a slag removal operation schedule, which is a schedule of the operation of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of the amount of slag remaining after slag removal, which is the amount of residual slag remaining on the surface of the molten metal in the vessel; a generation step of generating guidance for the operator's operation on the operation unit to realize the slag removal operation schedule calculated in the calculation step based on the target value of the amount of slag remaining after slag removal; a learning step of performing supervised learning using a set of past operational data of the slag removal system and the slag removal performance when the past operational data of the slag removal system was obtained as teacher data; and a correction step of correcting the operator's operation guidance on the operation unit based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator in response to the operation received by the operation unit.

One aspect of the present invention is a slag removal system comprising a vessel for containing molten metal, a first actuator for at least tilting the vessel, a first detection unit for detecting at least the tilt angle of the vessel, a slag removal member for removing slag from the surface of the molten metal in the vessel, a second actuator for moving the slag removal member, a second detection unit for detecting the amount of displacement of the second actuator, a control unit for controlling the first actuator and the second actuator, an imaging unit for imaging the surface of the molten metal in the vessel, a weight distribution calculation unit for calculating a weight distribution of the residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel imaged by the imaging unit, and a control schedule generation unit for generating a control schedule for the first actuator and the second actuator by the control unit, wherein the control schedule generation unit is configured to generate a control schedule for the first actuator and the second actuator by using a control schedule for the control schedule, the control schedule generation unit being ... The slag removal system includes a learning model unit that calculates a slag removal operation schedule, which is a schedule for the operation of the first actuator and the second actuator to remove slag using the slag removal member, based on a target value of the slag residual amount after slag removal, and an AI model unit that generates a control schedule for the first actuator and the second actuator that realizes the slag removal operation schedule calculated by the learning model unit based on the target value of the slag residual amount after slag removal. The learning model unit performs supervised learning using a set of past operation data of the slag removal system and the slag removal performance when the past operation data of the slag removal system was obtained as teacher data, and the AI model unit modifies the control schedule for the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator based on the control schedule generated by the learning model unit.

One aspect of the present invention is an automatic control method for a slag removal system comprising a vessel for containing molten metal, a first actuator for at least tilting the vessel, a first detection unit for detecting at least the tilt angle of the vessel, a slag removal member for removing slag from the surface of the molten metal in the vessel, a second actuator for moving the slag removal member, a second detection unit for detecting the amount of displacement of the second actuator, a control unit for controlling the first actuator and the second actuator, an imaging unit for imaging the surface of the molten metal in the vessel, a weight distribution calculation unit for calculating a weight distribution of the residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel imaged by the imaging unit, and a control schedule generation unit for generating a control schedule for the first actuator and the second actuator by the control unit, the control schedule being based on a target value of a slag remaining amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after slag removal by the slag removal member. The automatic control method for a slag removal system includes a calculation step of calculating a slag removal operation schedule, which is a schedule of the operation of the first actuator and the second actuator for removing slag by the slag removal member; a generation step of generating a control schedule for the first actuator and the second actuator that realizes the slag removal operation schedule calculated in the calculation step based on a target value of the amount of slag remaining after the slag removal; a learning step of performing supervised learning using a set of past operation data of the slag removal system and the slag removal performance when the past operation data of the slag removal system was obtained as teacher data; and a correction step of correcting the control schedule for the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator based on the control schedule generated in the generation step.

One aspect of the present invention is a slag removal system including a container for storing molten metal, a first actuator for at least tilting the container, a first detection unit for detecting at least the tilt angle of the container, a slag removal member for removing slag from the surface of the molten metal in the container, a second actuator for moving the slag removal member, a second detection unit for detecting the amount of displacement of the second actuator, a control unit for controlling the first actuator and the second actuator, an imaging unit for imaging the surface of the molten metal in the container, a weight distribution calculation unit for calculating a weight distribution of the residual slag remaining on the surface of the molten metal in the container based on an image of the surface of the molten metal in the container imaged by the imaging unit, and a control schedule generation unit for generating a control schedule for the first actuator and the second actuator by the control unit, the control unit being configured to generate a control schedule for the first actuator and the second actuator, the control unit being configured to generate a control schedule for the first actuator and the second actuator, the control unit being configured to generate a control schedule for the first actuator and the second actuator, the control unit being configured to generate a control schedule for the first actuator and the second actuator by the control unit, the control unit being configured to generate a control schedule for the first actuator and the second actuator, the control unit being configured to generate a control schedule for the first actuator and the second actuator, the control unit being configured to generate a control schedule for the second actuator and ... a calculation step of calculating a slag removal operation schedule, which is a schedule of the operation of the first actuator and the second actuator for removing slag by the slag removal member, based on the target value of the slag remaining amount after the slag removal; a generation step of generating a control schedule for the first actuator and the second actuator that realizes the slag removal operation schedule calculated in the calculation step based on the target value of the slag remaining amount after the slag removal; a learning step of performing supervised learning using a set of past operation data of the slag removal system and the slag removal performance when the past operation data of the slag removal system was obtained as teacher data; and a correction step of correcting the control schedule for the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator based on the control schedule generated in the generation step.

According to the present invention, it is possible to provide a slag removal system that can achieve slag removal while suppressing the amount of sulfur return and molten iron loss, a method for generating guidance for the slag removal system, and an automatic control method and program for the slag removal system.

FIG. 1 is a diagram showing an example of an outline of a dross removal system according to a first embodiment. FIG. 2 is a diagram showing an example of the flow of data and control signals in the dregs removal system of the first embodiment. FIG. 2 is a diagram showing in further detail the flow of data and control signals in the tailings removal system of the first embodiment. This figure shows an example of the slag disposal performance of the slag disposal system and past operational data of the slag disposal system (operational performance, equipment performance, and operating performance) that constitute the teacher data (learning data) used in supervised learning performed by the learning model unit. 4A to 4C are diagrams showing examples of correct answer data and incorrect answer data used as teacher data (learning data) in supervised learning performed by a learning model unit. 4 is a flowchart for explaining an example of processing executed in the dregs removal system of the first embodiment. 7 is a diagram for explaining an example of processing by a learning model unit in step S12 of FIG. 6. FIG. 13 is a diagram showing an example of an image captured by the imaging unit when raking of molten iron is occurring. A figure showing an example of a factor taken into account in the slag removal system of the first embodiment as a factor that may affect the slag removal operation schedule (objective variable) predicted (calculated) by the learning model unit. FIG. 2 is a diagram for explaining an example of a slag removal operation in the slag removal system of the first embodiment. FIG. 2 is a diagram for explaining an example of a slag removal operation in the slag removal system of the first embodiment. FIG. 13 is a diagram showing an example of an outline of a tailings removal system according to a second embodiment. FIG. 11 is a diagram showing an example of the flow of data and control signals in the dregs removal system of the second embodiment. FIG. 11 is a diagram showing in more detail the flow of data and control signals in the tailings removal system of the second embodiment. 13 is a flowchart for explaining an example of a process executed in the dregs removal system of the second embodiment. 16 is a diagram for explaining an example of the process of calculating an error, updating parameters, expressions, etc., performed in step S29 of FIG. 15. 16 is a diagram for explaining an example of the process of calculating an error, updating parameters, expressions, etc., performed in step S29 of FIG. 15.

The following describes embodiments of the slag removal system, guidance generation method for the slag removal system, automatic control method for the slag removal system, and program of the present invention.

First Embodiment
Fig. 1 is a diagram showing an example of an outline of the slag removal system 1 of the first embodiment. Fig. 2 is a diagram showing an example of the flow of data and control signals in the slag removal system 1 of the first embodiment. Fig. 3 is a diagram showing the flow of data and control signals in the slag removal system 1 of the first embodiment in more detail.
In the example shown in Figures 1 to 3, the slag removal system 1 includes a container 1A, a first actuator 1B, a first detection unit 1C, a slag removal member 1D, a second actuator 1E, a second detection unit 1F, an operation unit 1G, a control unit 1H, an imaging unit 1I, a weight distribution calculation unit 1J, a guidance generation unit 1K, a guidance output unit 1L, and an operation database 1M.
The vessel 1A is configured similarly to the vessel described in, for example, US Pat. No. 5,399,992 and contains molten metal.
The first actuator 1B changes the position, posture, etc. of the container 1A. The first actuator 1B includes a tilting device 1B1 and a container cart 1B2. The tilting device 1B1 is equipped with a tilting cylinder 1B11, and has the function of adjusting the tilt angle of the container 1A by operating the tilting cylinder 1B11. The container cart 1B2 supports the container 1A and the tilting device 1B1. The position of the container 1A can be changed by moving the container cart 1B2.
The first detection unit 1C includes a container tilt angle meter 1C1 and a container cart position detection device 1C2. The container tilt angle meter 1C1 detects the tilt angle of the container 1A. The container cart position detection device 1C2 detects the position of the container cart 1B2. The position of the container A1 can be calculated by using the position of the container cart 1B2 detected by the container cart position detection device 1C2.

The slag removal member 1D removes slag from the surface of the molten metal in the vessel 1A. The slag removal member 1D includes a scraper 1D1 and an arm 1D2. The scraper 1D1 is configured similarly to the scraper described in, for example, Patent Document 4. The arm 1D2 is configured similarly to the arm described in, for example, Patent Document 4. The scraper 1D1 is attached to the tip of the arm 1D2.
1 to 3, the slag removal member 1D includes a scraper 1D1 and an arm 1D2, but in other examples, the slag removal member 1D may be configured similarly to the slag removal tool described in Patent Document 1. In still another example, the slag removal member 1D may be configured similarly to the slag removal tool described in Patent Document 2.

In the example shown in Figures 1 to 3, the second actuator 1E moves the slag removal member 1D. The second actuator 1E includes a cylinder 1E1 and a traveling cart 1E2. The cylinder 1E1 includes a lifting cylinder 1E11 and a swivel cylinder 1E12. The lifting cylinder 1E11 moves the arm 1D2 in the vertical direction. The swivel cylinder 1E12 moves the arm 1D2 in the swivel direction. The traveling cart 1E2 supports the arm 1D2 to which the scraper 1D1 is attached and the cylinder 1E1. The traveling cart 1E2 is equipped with a motor 1E21 that drives the wheels of the traveling cart 1E2, and the arm 1D2, etc. can be moved in the horizontal direction by operating the motor 1E21.
The slag removal device is composed of a slag removal member 1D and a second actuator 1E.
The second detection unit 1F detects the displacement amount of the second actuator 1E. The second detection unit 1F includes a linear encoder 1F1, a rotary encoder 1F2, an arm imaging camera 1F3, and a scraper coordinate calculation unit 1F4. The linear encoder 1F1 detects the stroke amount of the cylinder 1E1. That is, the linear encoder 1F1 detects the stroke amount of the lifting cylinder 1E11 and the stroke amount of the turning cylinder 1E12. The rotary encoder 1F2 detects the number of revolutions of the wheels of the traveling carriage 1E2. The position of the traveling carriage 1E2 can be calculated by using the number of revolutions of the wheels of the traveling carriage 1E2 detected by the rotary encoder 1F2. The arm imaging camera 1F3 captures an image of the arm 1D2. The scraper coordinate calculation unit 1F4 calculates the coordinates (position) of the scraper 1D1 based on the image of the arm 1D2 captured by the arm imaging camera 1F3.

The operation unit 1G accepts operations by an operator.
The control unit 1H controls the first actuator 1B and the second actuator 1E in response to the operation received by the operation unit 1G. The control unit 1H includes a tilting device control unit 1H1 and a slag removal device control unit 1H2. The tilting device control unit 1H1 controls the first actuator 1B in response to the operation received by the operation unit 1G. The slag removal device control unit 1H2 controls the second actuator 1E in response to the operation received by the operation unit 1G.
The imaging unit 1I captures an image of the surface of the molten metal in the vessel 1A.
The weight distribution calculation unit 1J calculates the weight distribution of the residual slag remaining on the surface of the molten metal in the vessel 1A based on the image of the surface of the molten metal in the vessel 1A captured by the imaging unit 1I. In detail, the weight distribution calculation unit 1J calculates the weight distribution of the residual slag remaining on the surface of the molten metal in the vessel 1A by using, for example, the technology described in FIG. 2 of Patent Document 4.

1 to 3, the guidance generating unit 1K generates guidance for the operator's operation on the operating unit 1G. The guidance generating unit 1K includes a learning model unit 1K1, an AI model unit 1K2, a scraper state quantity acquiring unit 1K3, a vessel state quantity acquiring unit 1K4, a molten metal surface state quantity acquiring unit 1K5, a slag removal operation schedule acquiring unit 1K6, and a slag removal performance database 1K7.
In the example shown in Figure 2, not only are an arm imaging camera 1F3, a linear encoder 1F1, a rotary encoder 1F2, an imaging unit 1I, and a container tilt angle meter 1C1 connected to the guidance generation unit 1K, but also a setting switch and a process computer.
In the example shown in Figures 1 to 3, the learning model unit 1K1 calculates a slag removal operation schedule, which is a schedule of the operation of the first actuator 1B and the second actuator 1E for removing slag by the slag removal member 1D, based on a target value for the slag remaining amount after slag removal, which is the amount of residual slag remaining on the surface of the molten metal in the container 1A after the slag removal member 1D has been performed.
In detail, the learning model unit 1K1 performs supervised learning using a pair of past operational data of the slag disposal system 1 and the slag disposal performance at the time when the past operational data of the slag disposal system 1 was obtained as teacher data (learning data).

Figure 4 is a diagram showing an example of the slag disposal performance of the slag disposal system 1 and past operational data (operational performance, equipment performance, and operating performance) of the slag disposal system 1, which constitute the teacher data (learning data) used in supervised learning performed by the learning model unit 1K1.
In the example shown in FIG. 4, the slag removal performance of the slag removal system 1, which constitutes the teacher data (learning data) used in supervised learning performed by the learning model unit 1K1, includes the amount of slag remaining after slag removal [t], the amount of restored S [×10 ⁻³ %], and the amount of molten iron loss [t/ch].
The amount of sulfur [×10 ⁻³ %] is calculated by the following formula.
Return S amount [×10 ^-3 %] = S after KR (Kanbara Reactor) - Final S after blowing
The amount of molten iron loss [t/ch] is calculated by the following formula.
Hot metal loss amount [t/ch] = Charging amount - (Output amount - Output alloy amount) / Converter yield

5 is a diagram showing an example of correct data and incorrect data used as teacher data (learning data) in supervised learning performed by the learning model unit 1K1. The vertical axis of Fig. 5 shows the amount of return sulfur [×10 ⁻³ %], and the horizontal axis of Fig. 5 shows the amount of molten iron loss [t/ch].
In the example shown in Figure 5, a set of the slag waste performance of the slag waste system 1 in which the amount of S returned is smaller than the S returned threshold T1 and the amount of molten metal loss is smaller than the molten metal loss threshold T2, and the past operating data of the slag waste system 1 (operational performance, equipment performance, and operating performance (see Figure 4)) when the slag waste performance of the slag waste system 1 in which the amount of S returned is smaller than the S returned threshold T1 and the amount of molten metal loss is smaller than the molten metal loss threshold T2 is obtained, is used as correct answer data for the teacher data (learning data).
In addition, in the example shown in Figure 5, a pair of the slag disposal record of the slag disposal system 1 in which the amount of S returned is equal to or greater than the S returned threshold value T1 or the amount of molten iron loss is equal to or greater than the molten iron loss threshold value T2, and the past operating data (operational record, equipment record, and operating record) of the slag disposal system 1 when the amount of S returned is equal to or greater than the S returned threshold value T1 or the amount of molten iron loss is equal to or greater than the molten iron loss threshold value T2 is obtained, is used as incorrect answer data for the teacher data (learning data). That is, in the example shown in Figure 5, a set of the slag waste performance record of the slag waste system 1 in which the amount of S returned is smaller than the S returned threshold T1 and the amount of molten metal loss is equal to or greater than the molten metal loss threshold T2, and the past operational data (operational performance, equipment performance, and operating performance) of the slag waste system 1 when the slag waste performance record of the slag waste system 1 in which the amount of S returned is smaller than the S returned threshold T1 and the amount of molten metal loss is equal to or greater than the molten metal loss threshold T2 is obtained, is used as incorrect answer data for the teacher data (learning data). In addition, a set of the slag discharge performance of the slag discharge system 1 in which the amount of S returned is equal to or greater than the S returned threshold T1 and the amount of molten iron loss is smaller than the molten iron loss threshold T2, and the past operating data (operational performance, equipment performance, and operating performance) of the slag discharge system 1 when the amount of S returned is equal to or greater than the S returned threshold T1 and the amount of molten iron loss is smaller than the molten iron loss threshold T2, is used as incorrect answer data for the teacher data (learning data). Furthermore, a set of the slag discharge performance of the slag discharge system 1 in which the amount of S returned is equal to or greater than the S returned threshold T1 and the amount of molten iron loss is equal to or greater than the molten iron loss threshold T2, and the past operating data (operational performance, equipment performance, and operating performance) of the slag discharge system 1 when the slag discharge performance of the slag discharge system 1 in which the amount of S returned is equal to or greater than the S returned threshold T1 and the amount of molten iron loss is equal to or greater than the molten iron loss threshold T2 is obtained, is used as incorrect answer data for the teacher data (learning data).

In the example shown in FIG. 4, the past operation data of the slag removal system 1 constituting the teacher data (learning data) used in the supervised learning performed by the learning model unit 1K1 includes data on the molten metal handled by the slag removal system 1 and the slag removed from the surface of the molten metal (shown as "operational performance" in FIG. 4). In detail, the data on the molten metal handled by the slag removal system 1 and the slag removed from the surface of the molten metal (operational performance) includes "blow method [-]", "slag removal level [-]", "presence or absence of slag removal performance before KR treatment [-]", "slag properties [-]", "estimated slag amount before slag removal [t]", "molten iron temperature before slag removal treatment [°C]", "slag removal time before KR treatment [min]", "C, Si, Mn, P, S [%] after KR treatment", "KR treatment agent input amount [t]", and "molten iron amount [t]".

The blowing method [-] is classified into, for example, "MURC (Multi refining Converter) (see FIG. 9)", "LD-ORP (LD converter-Optimized Refining Process) (see FIG. 9)", etc.
The slag level [-] is classified, for example, into "S", "A", "B", and "C" (see FIG. 9).
"Whether or not slag removal before KR processing has been performed [-]" indicates whether or not pre-slag removal has been performed.
"Slag property [-]" is classified into, for example, "hard", "soft", "sticky" (see FIG. 9), etc.
The estimated amount of slag before discharge [t] is classified, for example, into "0.5 to 1.0 t", "1.0 to 1.5 t", "1.5 to 2.0 t", "2.0 to 2.5 t", "2.5 to 3.0 t" (see Figure 9), etc.
The "molten iron temperature before slag removal treatment [°C]" is obtained using a known method.
The slag discharge time before KR treatment [min] is the time from the start to the end of the slag discharge before KR treatment when the slag discharge before KR treatment is performed.
C, Si, Mn, P, S [%] after KR treatment are the contents of C, Si, Mn, P, and S after KR treatment.
The KR treating agent dosage [t] is the amount of KR treating agent used.

In the example shown in FIG. 4, the past operational data of the slag removal system 1, which constitutes the teacher data used in the supervised learning performed by the learning model unit 1K1, includes data on the shape of the slag removal member 1D (shown as "equipment performance" in FIG. 4). In detail, the data on the shape of the slag removal member 1D (equipment performance) includes the "shape [score] of the scraper 1D1 (dragger)" and the "shape [number of uses] of the KR impeller."

In the example shown in FIG. 4, the past operation data of the slag removal system 1 constituting the teacher data used in the supervised learning performed by the learning model unit 1K1 includes data on the movement of the container 1A and the slag removal member 1D (shown as "operational performance" in FIG. 4). In detail, the data on the movement of the container 1A and the slag removal member 1D (operational performance) includes "slag removal time after KR treatment [min]", "processing pattern of slag removal stage N [-]", "processing time of slag removal stage N [min]", "tilting angle range of the ladle (container 1A) [°]", "scraping speed [times/min]", "operator [name]", "scraping interruption time ratio [%]", and "irregular operation frequency [%]".

Examples of "processing pattern [-] for slag removal stage N" are, for example, "1st to 20th scraping: scraping slag from the center of the pot (container 1A)", "21st to 60th scraping: scraping slag around the pot (container 1A)", "61st to 80th scraping: scraping together slag on the left and right sides of the pot (container 1A) and removing them all at once", and "81st to 100th scraping: following the slag position".
"Tilt angle range of the pot (container 1A) [°]" is the range of tilt angle of the container 1A from the start to the end of the slag removal operation.
"Scraping speed [times/min]" is the number of times scraping is performed by scraper plate 1D1 per minute.
"Operator [name]" is the name of the operator who operates the operation unit 1G.
The "scraping interruption time ratio [%]" is the percentage of the time during which scraping by scraper plate 1D1 is not performed from the start to the end of the slag removal operation.

In the example shown in Figures 1 to 3, the AI model unit 1K2 generates guidance for the operator's operation on the operation unit 1G that realizes the slag removal operation schedule calculated by the learning model unit 1K1 based on the target value of the slag remaining amount after slag removal.
In addition, the AI model unit 1K2 has a function of modifying guidance for the operator's operation on the operation unit 1G based on real-time data obtained by the first detection unit 1C, the second detection unit 1F, and the imaging unit 1I when the control unit 1H is controlling the first actuator 1B and the second actuator 1E in response to the operation received by the operation unit 1G.

The scraper state quantity acquisition unit 1K3 acquires, as scraper state quantities, the coordinates (position) of the scraper 1D1 calculated by the scraper coordinate calculation unit 1F4, the stroke amount of the lifting cylinder 1E11 and the stroke amount of the turning cylinder 1E12 detected by the linear encoder 1F1, the number of revolutions of the wheels of the traveling carriage 1E2 detected by the rotary encoder 1F2, and the state of the second actuator 1E that is the object of control by the slag removal device control unit 1H2. The data acquired as the scraper state quantity by the scraper state quantity acquisition unit 1K3 is sent to the AI model unit 1K2.
The container state quantity acquisition unit 1K4 acquires, as container state quantities, the tilt angle of the container 1A detected by the container tilt angle meter 1C1 and the state of the first actuator 1B that is the target of control by the tilting device control unit 1H1. The data acquired as the container state quantities by the container state quantity acquisition unit 1K4 is sent to the AI model unit 1K2.
The molten metal surface state quantity acquisition unit 1K5 acquires the weight distribution of the residual slag remaining on the surface of the molten metal in the vessel 1A calculated by the weight distribution calculation unit 1J as the molten metal surface state quantity. The data acquired as the molten metal surface state quantity by the molten metal surface state quantity acquisition unit 1K5 is sent to the AI model unit 1K2.
The slag removal operation schedule acquisition unit 1K6 acquires the slag removal operation schedule calculated by the learning model unit 1K1. The slag removal operation schedule acquired by the slag removal operation schedule acquisition unit 1K6 is sent to the AI model unit 1K2.
The slag disposal record database 1K7 stores data on the disposal record of the slag disposal system 1, for example, as shown in Fig. 4. The slag disposal record of the slag disposal system 1 stored in the slag disposal record database 1K7 is sent to the learning model unit 1K1 and used for supervised learning performed in the learning model unit 1K1.

The guidance output unit 1L outputs the guidance generated by the guidance generating unit 1K (in the example shown in FIG. 2, displays it on a display monitor).
The operation database 1M stores past operation data (operational performance, equipment performance, and operating performance) of the slag disposal system 1, for example, as shown in Fig. 4. The past operation data of the slag disposal system 1 stored in the operation database 1M is sent to the learning model unit 1K1 and used for supervised learning performed in the learning model unit 1K1.

FIG. 6 is a flowchart for explaining an example of a process executed in the dregs removal system 1 of the first embodiment.
In the example shown in FIG. 6, in step S11A, the operation database 1M outputs past operation data (operational performance, equipment performance, and operating performance) (constraint conditions) of the tailings removal system 1 to the learning model unit 1K1.
Also, in step S11A, the slag disposal performance database 1K7 outputs the slag disposal performance (constraint conditions) of the slag disposal system 1 to the learning model unit 1K1.
Next, in step S11B, the learning model unit 1K1 performs supervised learning (prediction using a learning model) using a set of past operational data (operational performance, equipment performance, and operating performance) of the slag disposal system 1 and the slag disposal performance at the time when the past operational data (operational performance, equipment performance, and operating performance) of the slag disposal system 1 was obtained as teacher data (learning data).

Next, in step S12, a target value for the amount of slag remaining after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel 1A after the slag has been removed by the slag removal member 1D, is set, for example, by an operator (user of the slag removal system 1).
In another example, in step S12, the learning model unit 1K1 may set a target value for the amount of slag remaining after slag removal.

In the example shown in FIG. 6, in step S12, the learning model unit 1K1 calculates a slag removal operation schedule, which is a schedule of the operation of the first actuator 1B and the second actuator 1E for removing slag using the slag removal member 1D, based on the set target value for the amount of slag remaining after slag removal.

FIG. 7 is a diagram for explaining an example of the processing by the learning model unit 1K1 in step S12 of FIG.
In the example shown in Fig. 7, the "operational performance" and "equipment performance" shown in Fig. 4 are set in the input layer of the learning model unit 1K1. Also, a part of the "operational performance" shown in Fig. 4 is set in the output layer of the learning model unit 1K1 as an "operational performance prediction", and the "equipment performance" shown in Fig. 4 is set in the output layer of the learning model unit 1K1 as an "equipment performance prediction".
In the example of the slag removal operation schedule shown in Figure 7, "Elapsed time" indicates the time elapsed from the start time of the slag removal operation of the slag removal system 1, and "Pot tilt angle" indicates the tilt angle of the container 1A. "Number of scrapings" indicates the number of scrapings by the scraper plate 1D1 at each stage. Each example of "operation pattern" corresponds to each example of the "Slag removal stage N processing pattern [-]" described above.
In the example of the slag removal operation schedule shown in Fig. 7, during the period when the elapsed time is 0 to 2 [min] (during "Stage 1"), the tilt angle of the container 1A is set to 30 [°] by the tilting device 1B1 of the first actuator 1B, and the "number of scrapings" is set to 20. In addition, the slag removal member 1D is moved by the second actuator 1E so that the "operation pattern" of the scraper 1D1 becomes "slag removal from the center of the pot (container 1A)".
Next, during the period of 2 to 6 minutes (during "Stage 2"), the tilting device 1B1 of the first actuator 1B sets the tilt angle of the container 1A to 30° and the "number of scrapings" to 40. In addition, the second actuator 1E moves the slag removal member 1D so that the "operation pattern" of the scraper 1D1 becomes "slag removal around the pot (container 1A)."
Next, during the period when the elapsed time is 6 to 8 minutes (during "Stage 3"), the tilting device 1B1 of the first actuator 1B sets the tilt angle of the container 1A to 31° and the "number of scrapings" to 20. In addition, the slag removal member 1D is moved by the second actuator 1E so that the "operation pattern" of the scraper 1D1 becomes "scrape up the slag on the left and right sides of the pot (container 1A) and remove the slag."
Next, during the period of 8 to 10 minutes, the tilt angle of the container 1A is set to 32° by the tilting device 1B1 of the first actuator 1B, and the "number of scrapings" is set to 20. In addition, the slag removal member 1D is moved by the second actuator 1E so that the "operation pattern" of the scraper 1D1 becomes "slug position tracking."

In other words, in the example shown in FIG. 7, the schedule of operation of the first actuator 1B and the second actuator 1E calculated in step S12 of FIG. 6 includes a time period (a time period of 0 to 6 minutes elapsed) during which the first actuator 1B and the second actuator 1E operate without being based on the image of the surface of the molten metal in the vessel 1A captured by the imaging unit 1I, and a time period (a time period of 6 to 10 minutes elapsed) during which the first actuator 1B and the second actuator 1E operate based on the image of the surface of the molten metal in the vessel 1A captured by the imaging unit 1I.

In the example shown in FIG. 7, the operations of the first actuator 1B and the second actuator 1E that are not based on the image of the surface of the molten metal in the container 1A captured by the imaging unit 1I (operations during a time period when the elapsed time is 0 to 6 minutes) are executed before the operations of the first actuator 1B and the second actuator 1E that are based on the image of the surface of the molten metal in the container 1A captured by the imaging unit 1I (operations during a time period when the elapsed time is 6 to 10 minutes).

In the example shown in FIG. 6, next, in step S13, the learning model unit 1K1 outputs the slag removal operation schedule calculated in step S12, and the slag removal operation of the slag removal system 1 starts.

Next, in step S14, the guidance generating unit 1K generates guidance for the operator's operation on the operating unit 1G based on the slag removal operation schedule output in step S13. That is, the guidance generating unit 1K generates guidance for the operator's operation on the operating unit 1G that realizes the slag removal operation schedule calculated based on the target value of the slag remaining amount after slag removal set in step S12.
Also, in step S14, the guidance output unit 1L outputs the guidance generated by the guidance generating unit 1K.
Furthermore, in step S14, the operator performs an input operation on the operation unit 1G according to the guidance outputted by the guidance output unit 1L (for example, displayed on a display monitor).
That is, in step S14, the operation unit 1G accepts an input operation by the operator following the guidance output by the guidance output unit 1L.
In the example shown in Fig. 7, when step S14 in Fig. 6 is executed for the first time, the guidance generating unit 1K generates guidance for the operator's operation on the operating unit 1G to realize the slag removal operation schedule during the period when the "elapsed time" is 0 to 2 min. The operating unit 1G accepts the operator's operation to set the inclination angle of the container 1A to 30° and execute the "number of scrapings" of 20, and the operator's operation to execute "removal of slag from the center of the pot (container 1A)" as the "operation pattern" of the scraper 1D1.
The guidance generating unit 1K generates guidance for the operator's operation on the operating unit 1G to realize the slag removal operation schedule during the "elapsed time" period of 2 to 6 minutes. The operating unit 1G accepts the operator's operation to set the inclination angle of the container 1A to 30° and to perform 40 "scraping times," and the operator's operation to perform "slag removal around the pot (container 1A)" as the "operation pattern" of the scraper 1D1.

In the example shown in FIG. 6, next, in step S15, the AI model unit 1K2 acquires the characteristics of the discharged slag from the scraper state quantity acquisition unit 1K3, the container state quantity acquisition unit 1K4, etc.
In step S15, the AI model unit 1K2 acquires the molten metal surface state quantity (weight distribution of the residual slag remaining on the surface of the molten metal in the vessel 1A calculated by the weight distribution calculation unit 1J) from the molten metal surface state quantity acquisition unit 1K5. That is, the AI model unit 1K2 acquires the amount of slag remaining on the surface of the molten metal in the vessel 1A calculated by the weight distribution calculation unit 1J.
Next, in step S16, the AI model part 1K2 determines whether the amount of slag remaining on the surface of the molten metal in the container 1A obtained in step S15 has become less than the target value of the amount of slag remaining after slag removal set in step S12.
Furthermore, in step S16, the AI model unit 1K2 determines whether the elapsed time after the start of the dross removal operation of the dross removal system 1 has exceeded a time limit.
Furthermore, in step S16, the AI model unit 1K2 determines whether or not molten iron scraping is occurring based on the image captured by the imaging unit 1I.

FIG. 8 is a diagram showing an example of an image captured by the imaging unit 1I when raking of molten iron is occurring.
In the example shown in Fig. 8, a state in which molten iron that should not be scraped out is being scraped out (a dripping state) is captured by the imaging unit 1I. The image shown in Fig. 8 (i.e., an image captured by the imaging unit 1I) is used to evaluate the characteristics of the slag. The imaging unit 1I functions as a molten iron loss monitoring unit.
Returning to the explanation of step S16 in Fig. 6, if the amount of residual slag becomes less than the target value of the amount of residual slag after slag removal, the slag removal operation of the slag removal system 1 is terminated and the process proceeds to step S18. If the elapsed time after the start of the slag removal operation of the slag removal system 1 exceeds the time limit, the slag removal operation of the slag removal system 1 is also terminated and the process proceeds to step S18. If molten iron is scraped out, the slag removal operation of the slag removal system 1 is also terminated and the process proceeds to step S18.
On the other hand, if the amount of residual slag is not less than the target amount of residual slag after slag removal, and the elapsed time since the start of the slag removal operation of the slag removal system 1 has not exceeded the time limit, and no molten iron has been scraped out, proceed to step S17.

In step S17, the learning model unit 1K1 modifies the operator's operation guidance for the operation unit 1G (e.g., the guidance generated by the guidance generation unit 1K when step S14 is executed for the first time) based on real-time data obtained by the first detection unit 1C, the second detection unit 1F, and the imaging unit 1I when the control unit 1H is controlling the first actuator 1B and the second actuator 1E in response to the operator's operation received by the operation unit 1G (e.g., when step S14 is executed for the first time).
In detail, in step S17, the learning model unit 1K1 calculates the error between the real-time data (actual measured values) obtained by the first detection unit 1C, the second detection unit 1F, and the imaging unit 1I when step S14 is first executed, and the guidance (indicated values) generated by the guidance generation unit 1K when step S14 is first executed.
Furthermore, in step S17, the learning model unit 1K1 modifies parameters, expressions, etc. used by the guidance generating unit 1K to generate guidance when step S14 is executed for the second time.
Then, the process returns to step S14.
In step S14 which is executed for the second time, the guidance generating unit 1K generates guidance for the operator's operation of the operating unit 1G based on the parameters, expressions, etc. corrected in step S17.
In the example shown in Fig. 7, when step S14 in Fig. 6 is executed for the second time, the guidance generating unit 1K generates guidance for the operator's operation on the operating unit 1G to realize the slag removal operation schedule during the period of "elapsed time" of 6 to 8 minutes. The operating unit 1G accepts the operator's operation to set the inclination angle of the container 1A to 31 degrees and perform the "number of scrapings" of 20, and the operator's operation to move the scraper 1D1 to scrape up and remove the slag from the left and right sides of the pot (container 1A) according to the guidance generated by the guidance generating unit 1K.
In the example shown in Fig. 7, when step S16 in Fig. 6 is executed for the second time, the result is determined to be NO, and when step S14 in Fig. 6 is executed for the third time, the guidance generating unit 1K generates guidance for the operator's operation on the operating unit 1G to realize the slag removal operation schedule during the period when the "elapsed time" is 8 to 10 minutes. The operating unit 1G accepts the operator's operation to set the inclination angle of the container 1A to 32 degrees and to perform the "scraping number" of 20 times, and the operator's operation to move the scraper 1D1 to follow the slag position according to the guidance generated by the guidance generating unit 1K.

In step S18, the slag removal record database 1K7 acquires and stores the amount of slag remaining after slag removal that was compared with the target value of the amount of slag remaining after slag removal in step S16.
In addition, in step S18, the operation database 1M acquires and stores the operation record, equipment record, and operating record from the start to the end of the slag removal operation of the slag removal system 1.
Furthermore, in step S18, the AI model unit 1K2 calculates a matching score indicating whether the operator's operation was in accordance with the guidance generated by the guidance generation unit 1K, and the guidance output unit 1L outputs the matching score calculated by the AI model unit 1K2 (for example, displays it on a display monitor).

Next, in step S19, the learning model unit 1K1 calculates the error between the real-time data (actual measurement value) and the guidance (indicated value) generated by the guidance generation unit 1K based on the real-time data obtained by the first detection unit 1C, the second detection unit 1F, and the imaging unit 1I when the control unit 1H is controlling the first actuator 1B and the second actuator 1E in response to the operator's operation received by the operation unit 1G (i.e., when step S14 is executed), and updates the parameters, formulas, etc. used by the guidance generation unit 1K to generate guidance the next time the slag removal operation of the slag removal system 1 is performed.

Figure 9 is a diagram showing an example of a factor taken into consideration in the slag removal system 1 of the first embodiment as a factor that may affect the slag removal operation schedule (objective variable) predicted (calculated) by the learning model unit 1K1.
In the example shown in Fig. 9, the blowing method is considered to be MURC or LD-ORP. Also, the steel type (slag removal level) is considered to be classified as "S", "A", "B" or "C". Furthermore, whether pre-slag removal is performed or not is considered. Also, whether the slag property is hard, soft or sticky is considered.
In addition, in the example shown in Figure 9, it is taken into consideration whether the estimated slag amount before slag discharge is classified into 0.5 to 1.0 t, 1.0 to 1.5 t, 1.5 to 2.0 t, 2.0 to 2.5 t, or 2.5 to 3.0 t.

10 and 11 are diagrams for explaining an example of the slag removal operation in the slag removal system 1 of the first embodiment. In detail, FIG. 10(A) is a time chart of the operation of the tilting device 1B1, FIG. 10(B) is a time chart of the forward operation of the traveling carriage 1E2, FIG. 10(C) is a time chart of the backward operation of the traveling carriage 1E2, FIG. 10(D) is a time chart of the operation of the lifting cylinder 1E11 lifting the arm 1D2, FIG. 10(E) is a time chart of the operation of the lifting cylinder 1E11 lowering the arm 1D2, FIG. 10(F) is a time chart of the operation of the turning cylinder 1E12 turning the arm 1D2 to the right, and FIG. Figure 10(G) is a time chart of the operation of the rotating cylinder 1E12 rotating the arm 1D2 to the left, Figure 10(H) is a time chart of the inclination angle of the container 1A detected by the container inclination angle meter 1C1, Figure 10(I) is a time chart of the x coordinate of the scraper 1D1 calculated by the scraper coordinate calculation unit 1F4, Figure 10(J) is a time chart of the y coordinate of the scraper 1D1 calculated by the scraper coordinate calculation unit 1F4, and Figure 10(K) is a time chart of the z coordinate of the scraper 1D1 calculated by the scraper coordinate calculation unit 1F4.
Figure 11(A) is a view of the container 1A, scraper 1D1, and arm 1D2 at time t1 as viewed from above in the vertical direction (the positive side of the z-axis). Figure 11(B) is a view of the container 1A, scraper 1D1, and arm 1D2 at time t1 as viewed from the right side in the horizontal direction (the positive side of the x-axis).
Figure 11(C) is a view of the container 1A, scraper 1D1, and arm 1D2 at time t3 as viewed from above in the vertical direction (the positive side of the z-axis). Figure 11(D) is a view of the container 1A, scraper 1D1, and arm 1D2 at time t3 as viewed from the right side in the horizontal direction (the positive side of the x-axis).

10 and 11, the forward movement of the traveling carriage 1E2 (movement toward the positive side of the y-axis) starts at time t1, and the arm 1D2 starts turning left. As a result, at time t1, the value of the x-coordinate of the scraper 1D1 starts to decrease, and the value of the y-coordinate of the scraper 1D1 starts to increase.
Next, at time t2, the left rotation of the arm 1D2 ends, and as a result, the decrease in the x-coordinate value of the scraper 1D1 ends at time t2.
Next, at time t3, the forward movement of the traveling carriage 1E2 (movement toward the positive side of the y-axis) ends. As a result, at time t3, the increase in the y-coordinate value of the scraper 1D1 ends.
The trajectory of the scraper 1D1 from time t1 to time t3 is shown by the dashed line in Figures 11(A) and 11(B).
At time t3, the arm 1D2 starts to descend, causing the z-coordinate value of the scraper 1D1 to start decreasing.

Next, at time t4, the descent of the arm 1D2 ends, and as a result, the decrease in the z-coordinate value of the scraper 1D1 ends at time t4.
At time t4, the traveling carriage 1E2 starts moving backward (toward the negative side of the y-axis), and the arm 1D2 starts rotating to the right. As a result, at time t4, the x-coordinate value of the scraper 1D1 starts to increase, and the y-coordinate value of the scraper 1D1 starts to decrease.
Next, at time t5, the rightward rotation of the arm 1D2 ends, and as a result, the increase in the x-coordinate value of the scraper 1D1 ends at time t5.
Next, at time t6, the rearward movement of the traveling carriage 1E2 (movement toward the negative side of the y-axis) ends. As a result, at time t6, the decrease in the y-coordinate value of the scraper 1D1 ends.
The trajectory of scraper 1D1 from time t3 to time t6 is shown by the dashed lines in Figures 11(C) and 11(D).

As shown in FIG. 10(A), during the period from time t1 to time t6, the tilting device 1B1 does not change the tilt angle of the container 1A. Therefore, as shown in FIG. 10(H), the tilt angle of the container 1A detected by the container tilt angle meter 1C1 is maintained at a constant value.

When the slag removal equipment is operated based on the operator's perception and judgment, the slag removal efficiency (how much slag can be removed in a limited time) and the amount of molten iron loss (the amount of molten iron scraped out) will depend on the operator's level of proficiency and the operating environment, etc., which may result in variations in the molten iron yield and components.
Therefore, in the slag removal system 1 of the first embodiment, the learning model unit 1K1 performs analysis (learning) using past operation data of the molten metal to be removed and past data on the results of slag removal to calculate an optimal slag removal operation schedule. Furthermore, the AI model unit 1K2 performs analysis using real-time data (instrumentation and image data) obtained while the slag removal operation is being performed, and predicts the optimal operation of the scraper 1D1. The calculation results of the learning model unit 1K1 and the prediction results of the AI model unit 1K2 are output (displayed on a display monitor) by the guidance output unit 1L as guidance for the operator's operation.
For example, in the example shown in Figure 7, the total time required for slag removal is set to 10 minutes. During the first 10 minutes, from 0 to 2 minutes, slag removal is performed from the center of the pot (container 1A). During 2 to 6 minutes, slag removal is performed from the periphery of the pot (container 1A). During 6 to 8 minutes, the slag on the left and right sides of the pot (container 1A) is scraped together and removed together. During 8 to 10 minutes, slag position tracking control is performed.

Second Embodiment
Below, a second embodiment of the slag removal system, the guidance generating method for the slag removal system, the automatic control method for the slag removal system, and the program of the present invention will be described.
The slag discharge system 1 of the second embodiment is configured similarly to the slag discharge system 1 of the first embodiment described above, except for the points described below. Therefore, according to the slag discharge system 1 of the second embodiment, the same effects as the slag discharge system 1 of the first embodiment described above can be achieved, except for the points described below.

Fig. 12 is a diagram showing an example of an outline of the slag removal system 1 of the second embodiment. Fig. 13 is a diagram showing an example of the flow of data and control signals in the slag removal system 1 of the second embodiment. Fig. 14 is a diagram showing the flow of data and control signals in the slag removal system 1 of the second embodiment in more detail.
In the example shown in Figures 12 to 14, the slag removal system 1 includes a container 1A, a first actuator 1B, a first detection unit 1C, a slag removal member 1D, a second actuator 1E, a second detection unit 1F, a control unit 1H, an imaging unit 1I, a weight distribution calculation unit 1J, a control schedule generation unit 1N, a display unit 1P, and an operation database 1M.
In other words, in the example shown in Figures 1 to 3, the slag removal system 1 is equipped with an operation unit 1G, a guidance generating unit 1K, and a guidance output unit 1L, but in the example shown in Figures 12 to 14, the slag removal system 1 does not include an operation unit 1G, a guidance generating unit 1K, or a guidance output unit 1L.
In the example shown in Figures 12 to 14, similarly to the example shown in Figures 1 to 3, the container 1A contains molten metal. The first actuator 1B changes the position, attitude, etc. of the container 1A (i.e., tilts the container 1A, etc.). The first actuator 1B includes a tilting device 1B1 having a function of adjusting the tilt angle of the container 1A, and a container cart 1B2 that can move while supporting the container 1A and the tilting device 1B1. The first detection unit 1C detects the tilt angle of the container 1A, etc. The first detection unit 1C includes a container tilt angle meter 1C1 that detects the tilt angle of the container 1A, and a container cart position detection device 1C2 that detects the position of the container cart 1B2.

In the example shown in FIGS. 12 to 14, similarly to the example shown in FIGS. 1 to 3, a slag remover 1D removes slag from the surface of the molten metal in the vessel 1A.
12 to 14, the slag removal member 1D includes a scraper 1D1 and an arm 1D2 having the scraper 1D1 attached to the tip thereof, but in other examples, the slag removal member 1D may be configured similarly to the slag removal tool described in Patent Document 1. In still another example, the slag removal member 1D may be configured similarly to the slag removal tool described in Patent Document 2.

In the example shown in Figures 12 to 14, the second actuator 1E moves the slag removal member 1D, as in the example shown in Figures 1 to 3. The second actuator 1E includes a cylinder 1E1 that moves the arm 1D2 vertically or in a rotational direction, and a traveling carriage 1E2 that moves the arm 1D2 horizontally. The slag removal device is composed of the slag removal member 1D and the second actuator 1E.
The second detection unit 1F detects the amount of displacement of the second actuator 1E. The second detection unit 1F includes a linear encoder 1F1 that detects the stroke amount of the cylinder 1E1, a rotary encoder 1F2 that detects the number of rotations of the wheels of the traveling carriage 1E2, an arm imaging camera 1F3 that captures an image of the arm 1D2, and a scraper coordinate calculation unit 1F4 that calculates the coordinates of the scraper 1D1 based on the image of the arm 1D2 captured by the arm imaging camera 1F3.

In the example shown in FIGS. 1 to 3, an operation unit 1G receives an operation from an operator, and a control unit 1H controls a first actuator 1B and a second actuator 1E in accordance with the operation received by the operation unit 1G.
12 to 14, the control unit 1H autonomously controls the first actuator 1B and the second actuator 1E without the need for an operator's operation. The control unit 1H includes a tilting device control unit 1H1 that controls the first actuator 1B and a slag removal device control unit 1H2 that controls the second actuator 1E.

In the example shown in Figures 12 to 14, similar to the example shown in Figures 1 to 3, the imaging unit 1I images the surface of the molten metal in the vessel 1A. The weight distribution calculation unit 1J calculates the weight distribution of the residual slag remaining on the surface of the molten metal in the vessel 1A based on the image of the surface of the molten metal in the vessel 1A captured by the imaging unit 1I.

12 to 14, a control schedule generating unit 1N generates a control schedule for the first actuator 1B and the second actuator 1E by the control unit 1H. The control schedule generating unit 1N includes a learning model unit 1N1, an AI model unit 1N2, a scraper state quantity acquiring unit 1N3, a vessel state quantity acquiring unit 1N4, a molten metal surface state quantity acquiring unit 1N5, a slag removal operation schedule acquiring unit 1N6, and a slag removal performance database 1N7.
In the example shown in Figure 13, not only are an arm imaging camera 1F3, a linear encoder 1F1, a rotary encoder 1F2, an imaging unit 1I, and a container tilt angle meter 1C1 connected to the control schedule generation unit 1N, but also a setting switch and a process computer.
In the example shown in Figures 12 to 14, the learning model unit 1N1 calculates a slag removal operation schedule, which is a schedule for the operation of the first actuator 1B and the second actuator 1E for removing slag by the slag removal member 1D, based on a target value for the slag remaining amount after slag removal, which is the amount of residual slag remaining on the surface of the molten metal in the container 1A after the slag removal member 1D has removed the slag.
In detail, the learning model unit 1N1 performs supervised learning using a set of past operational data of the slag disposal system 1 (operational performance, equipment performance, and operating performance (see Figure 4)) and the slag disposal performance at the time when the past operational data of the slag disposal system 1 was obtained as teacher data (learning data).

In the examples shown in Figures 12 to 14, the slag discharge performance of the slag discharge system 1 and the past operational data of the slag discharge system 1 (operational performance, equipment performance and operating performance) that constitute the teacher data (learning data) used in supervised learning performed by the learning model unit 1N1 can be, for example, the slag discharge performance of the slag discharge system 1 and the past operational data of the slag discharge system 1 (operational performance, equipment performance and operating performance) shown in Figure 4.
In this example, the slag removal record constituting the teaching data used in the supervised learning performed by the learning model unit 1N1 includes the amount of slag remaining after slag removal [t].
In this example, the slag removal record constituting the teacher data (learning data) used in the supervised learning performed by the learning model unit 1N1 includes the amount of sulfur recovery [×10 ⁻³ %] and the amount of molten iron loss [t/ch].

In the examples shown in FIGS. 12 to 14, correct answer data and incorrect answer data such as those shown in FIG. 5 can be used as teacher data (learning data) in supervised learning performed by the learning model unit 1N1.
In this example, a set of the slag waste performance record of the slag waste system 1 in which the amount of S returned is smaller than the S returned amount threshold T1 and the amount of molten iron loss is smaller than the molten iron loss amount threshold T2, and the past operating data (operating record, equipment record, and operating record) of the slag waste system 1 when the slag waste performance record of the slag waste system 1 in which the amount of S returned is smaller than the S returned amount threshold T1 and the amount of molten iron loss is smaller than the molten iron loss amount threshold T2 is obtained, is used as correct answer data for the teacher data (learning data) in supervised learning performed by the learning model unit 1N1.
In addition, in this example, a pair of the slag discharge performance of the slag discharge system 1 in which the amount of S returned is equal to or greater than the S returned threshold value T1 or the amount of molten iron loss is equal to or greater than the molten iron loss threshold value T2, and the past operating data (operating performance, equipment performance, and operating performance) of the slag discharge system 1 at the time when the slag discharge performance of the slag discharge system 1 in which the amount of S returned is equal to or greater than the S returned threshold value T1 or the amount of molten iron loss is equal to or greater than the molten iron loss threshold value T2 was obtained, is used as incorrect answer data for the teacher data (learning data) in supervised learning performed by the learning model unit 1N1.

In the example shown in Figures 11 to 13, the past operational data (operational performance, equipment performance, and operating performance) of the slag removal system 1 that constitutes the teacher data (learning data) used in the supervised learning performed by the learning model unit 1N1 includes data on the molten metal handled by the slag removal system 1 and the slag removed from the surface of the molten metal, such as that shown in "operational performance" in Figure 4.

In the example shown in Figures 11 to 13, the past operational data (operational performance, equipment performance, and operating performance) of the slag removal system 1 that constitutes the teacher data (learning data) used in supervised learning performed by the learning model unit 1N1 includes data on the shape of the slag removal member 1D, for example, as shown in "equipment performance" in Figure 4.

In the example shown in Figures 12 to 14, the past operational data (operational performance, equipment performance, and operating performance) of the slag removal system 1 that constitutes the teacher data (learning data) used in supervised learning performed by the learning model unit 1N1 includes data on the movement of the container 1A and the slag removal member 1D, for example, as shown in "operational performance" in Figure 4.

In the example shown in Figures 12 to 14, the AI model unit 1N2 generates a control schedule for the first actuator 1B and the second actuator 1E that realizes the slag removal operation schedule calculated by the learning model unit 1N1 based on the target value of the slag remaining amount after slag removal.
In addition, the AI model unit 1N2 has a function of modifying (updating) the control schedule of the first actuator 1B and the second actuator 1E based on real-time data obtained by the first detection unit 1C, the second detection unit 1F, and the imaging unit 1I when the control unit 1H is controlling the first actuator 1B and the second actuator 1E based on the control schedule generated by the learning model unit 1N1.
When the control unit 1H controls the first actuator 1B and the second actuator 1E based on the control schedule generated by the learning model unit 1N1, the real-time data obtained by the first detection unit 1C, the second detection unit 1F, and the imaging unit 1I includes the amount of slag remaining on the surface of the molten metal in the container 1A acquired by the AI model unit 1N2 (the amount of slag remaining on the surface of the molten metal in the container 1A calculated by the weight distribution calculation unit 1J).
In detail, when the control unit 1H controls the first actuator 1B and the second actuator 1E based on the control schedule generated by the learning model unit 1N1, the real-time data obtained by the first detection unit 1C, the second detection unit 1F, and the imaging unit 1I includes features of slag discharge acquired by the AI model unit 1N2 from the scraper state quantity acquisition unit 1N3, the vessel state quantity acquisition unit 1N4, etc., and molten metal surface state quantities acquired by the AI model unit 1N2 from the molten metal surface state quantity acquisition unit 1N5 (weight distribution of residual slag remaining on the surface of the molten metal in the vessel 1A calculated by the weight distribution calculation unit 1J).

The scraper state quantity acquisition unit 1N3 acquires, as scraper state quantities, the coordinates (position) of the scraper 1D1 calculated by the scraper coordinate calculation unit 1F4, the stroke amount of the lifting cylinder 1E11 and the stroke amount of the turning cylinder 1E12 detected by the linear encoder 1F1, the number of rotations of the wheels of the traveling carriage 1E2 detected by the rotary encoder 1F2, and the state of the second actuator 1E that is the object of control by the slag removal device control unit 1H2. The data acquired as the scraper state quantity by the scraper state quantity acquisition unit 1N3 is sent to the AI model unit 1N2.
The container state quantity acquisition unit 1N4 acquires, as container state quantities, the tilt angle of the container 1A detected by the container tilt angle meter 1C1 and the state of the first actuator 1B that is the target of control by the tilting device control unit 1H1. The data acquired as the container state quantities by the container state quantity acquisition unit 1N4 is sent to the AI model unit 1N2.
The molten metal surface state quantity acquisition unit 1N5 acquires the weight distribution of the residual slag remaining on the surface of the molten metal in the vessel 1A calculated by the weight distribution calculation unit 1J as the molten metal surface state quantity. The data acquired as the molten metal surface state quantity by the molten metal surface state quantity acquisition unit 1N5 is sent to the AI model unit 1N2.
The slag removal operation schedule acquisition unit 1N6 acquires the slag removal operation schedule calculated by the learning model unit 1N1. The slag removal operation schedule acquired by the slag removal operation schedule acquisition unit 1N6 is sent to the AI model unit 1N2.
The slag disposal record database 1N7 stores data on the disposal record of the slag disposal system 1, for example, as shown in Fig. 4. The slag disposal record of the slag disposal system 1 stored in the slag disposal record database 1N7 is sent to the learning model unit 1N1 and used for supervised learning performed in the learning model unit 1N1.

The display unit 1P displays, for example, the current status of the control schedule generated by the control schedule generating unit 1N.
The operation database 1M stores past operation data (operational performance, equipment performance, and operating performance) of the slag disposal system 1, for example, as shown in Fig. 4. The past operation data of the slag disposal system 1 stored in the operation database 1M is sent to the learning model unit 1N1 and used for supervised learning performed in the learning model unit 1N1.

FIG. 15 is a flowchart for explaining an example of a process executed in the tailings removal system 1 of the second embodiment.
In the example shown in FIG. 15, in step S21A, the operation database 1M outputs past operation data (operational performance, equipment performance, and operating performance) (constraint conditions) of the tailings removal system 1 to the learning model unit 1N1.
Also, in step S21A, the slag disposal performance database 1N7 outputs the slag disposal performance (constraint conditions) of the slag disposal system 1 to the learning model unit 1N1.
Next, in step S21B, the learning model unit 1N1 performs supervised learning (prediction using a learning model) using a set of past operational data (operational performance, equipment performance, and operating performance) of the slag disposal system 1 and the slag disposal performance at the time when the past operational data (operational performance, equipment performance, and operating performance) of the slag disposal system 1 was obtained as teacher data (learning data).

Next, in step S22, a target value for the amount of slag remaining after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel 1A after the slag has been removed by the slag removal member 1D, is set, for example, by an operator (user of the slag removal system 1).
In another example, in step S22, the learning model unit 1N1 may set a target value for the amount of slag remaining after slag removal.

In the example shown in Figure 15, in step S22, the learning model unit 1N1 calculates a slag removal operation schedule (for example, a slag removal operation schedule as shown in Figure 7), which is a schedule of the operation of the first actuator 1B and the second actuator 1E for removing slag using the slag removal member 1D, based on the set target value for the amount of slag remaining after slag removal.
In addition, in step S22, the control schedule generation unit 1N generates a control schedule for the first actuator 1B and the second actuator 1E by the control unit 1H that realizes the slag removal operation schedule calculated based on the target value of the slag remaining amount after slag removal.

In the example shown in FIG. 7, the schedule of operation of the first actuator 1B and the second actuator 1E calculated in step S22 of FIG. 15 includes a time period (a time period of 0 to 6 minutes elapsed) during which the first actuator 1B and the second actuator 1E operate without being based on an image of the surface of the molten metal in the vessel 1A captured by the imaging unit 1I, and a time period (a time period of 6 to 10 minutes elapsed) during which the first actuator 1B and the second actuator 1E operate based on an image of the surface of the molten metal in the vessel 1A captured by the imaging unit 1I.

In the example shown in FIG. 7, the operations of the first actuator 1B and the second actuator 1E that are not based on the image of the surface of the molten metal in the container 1A captured by the imaging unit 1I (operations during a time period when the elapsed time is 0 to 6 minutes) are executed before the operations of the first actuator 1B and the second actuator 1E that are based on the image of the surface of the molten metal in the container 1A captured by the imaging unit 1I (operations during a time period when the elapsed time is 6 to 10 minutes).

In the example shown in FIG. 15, next, in step S23, the learning model unit 1N1 outputs the slag removal operation schedule calculated in step S22, and the slag removal operation of the slag removal system 1 starts.

Next, in step S24, the control section 1H controls the first actuator 1B and the second actuator 1E based on the control schedule generated in step S22.
In the example shown in Fig. 7, when step S24 in Fig. 14 is executed for the first time, the control unit 1H controls the first actuator 1B and the second actuator 1E to realize a slag removal operation schedule during the period when the "elapsed time" is 0 to 2 min. Specifically, the control unit 1H controls the first actuator 1B to set the inclination angle of the container 1A to 30° without being based on an image of the surface of the molten metal in the container 1A captured by the imaging unit 1I, and controls the second actuator 1E to set the "number of scrapings" to 20 and execute "slag removal from the center of the pot (container 1A)" as the "operation pattern" of the scraper 1D1.
The control unit 1H also controls the first actuator 1B and the second actuator 1E to realize a slag removal operation schedule during the "elapsed time" period of 2 to 6 minutes. The control unit 1H controls the first actuator 1B to set the inclination angle of the container 1A to 30° without being based on the image of the surface of the molten metal in the container 1A captured by the imaging unit 1I, and controls the second actuator 1E to set the "number of scrapings" to 40 and to perform "slag removal around the pot (container 1A)" as the "operation pattern" of the scraper 1D1.
Also, in step S24, the display unit 1P displays, for example, on a display monitor, the current state of control of the first actuator 1B and the second actuator 1E by the control unit 1H.

In the example shown in FIG. 15, next, in step S25, the AI model unit 1N2 acquires the characteristics of the discharged scum from the scraper state quantity acquisition unit 1N3, the container state quantity acquisition unit 1N4, etc.
In step S25, the AI model unit 1N2 acquires the molten metal surface state quantity (the weight distribution of the residual slag remaining on the surface of the molten metal in the vessel 1A calculated by the weight distribution calculation unit 1J) from the molten metal surface state quantity acquisition unit 1N5. That is, the AI model unit 1N2 acquires the amount of slag remaining on the surface of the molten metal in the vessel 1A calculated by the weight distribution calculation unit 1J.

Next, in step S26, the AI model unit 1N2 determines whether the amount of slag remaining on the surface of the molten metal in the container 1A obtained in step S25 has become less than the target value of the amount of slag remaining after slag removal set in step S22.
Furthermore, in step S26, the AI model unit 1N2 determines whether the elapsed time since the start of the dross removal operation of the dross removal system 1 has exceeded a time limit.
Also, in step S26, the AI model unit 1N2 determines whether or not molten iron scraping is occurring based on the image captured by the imaging unit 1I.
If the amount of residual slag becomes less than the target amount of residual slag after slag removal, the slag removal operation of the slag removal system 1 is terminated and the process proceeds to step S28. If the elapsed time after the start of the slag removal operation of the slag removal system 1 exceeds the time limit, the slag removal operation of the slag removal system 1 is also terminated and the process proceeds to step S28. If scraping of molten iron occurs, the slag removal operation of the slag removal system 1 is also terminated and the process proceeds to step S28.
On the other hand, if the amount of residual slag is not less than the target amount of residual slag after slag removal, and the elapsed time since the start of the slag removal operation of the slag removal system 1 has not exceeded the time limit, and no molten iron has been scraped out, proceed to step S27.

In step S27, the learning model unit 1N1 modifies (updates) the control schedule for the first actuator 1B and the second actuator 1E based on real-time data obtained by the first detection unit 1C, the second detection unit 1F, and the imaging unit 1I when the control unit 1H is controlling the first actuator 1B and the second actuator 1E based on the control schedule generated in step S22 (for example, when step S24 is executed for the first time).
The real-time data obtained by the first detection unit 1C, the second detection unit 1F, and the imaging unit 1I while the control unit 1H is controlling the first actuator 1B and the second actuator 1E based on the control schedule generated in step S22 includes the amount of slag remaining on the surface of the molten metal in the container 1A acquired in step S25 (the amount of slag remaining on the surface of the molten metal in the container 1A calculated by the weight distribution calculation unit 1J).
In detail, in step S27, the learning model unit 1N1 calculates the error between the real-time data (actual values) obtained by the first detection unit 1C, the second detection unit 1F, and the imaging unit 1I when step S24 is executed for the first time, and the control schedule (indicated values) generated in step S22.
Furthermore, in step S27, the learning model unit 1N1 updates the parameters, equations, etc. used by the control unit 1H to control the first actuator 1B and the second actuator 1E when step S24 is executed for the second time.
Next, the process returns to step S24.
In step S24 which is executed for the second time, the control section 1H controls the first actuator 1B and the second actuator 1E based on the parameters, equations, etc. updated in step S27.
In the example shown in Fig. 7, when step S24 in Fig. 15 is executed for the second time, the control unit 1H controls the first actuator 1B and the second actuator 1E to realize a slag removal operation schedule during the period when the "elapsed time" is 6 to 8 minutes. Specifically, the control unit 1H controls the first actuator 1B to set the inclination angle of the container 1A to 31°, and controls the second actuator 1E to set the "number of scrapings" to 20 and to perform "slag removal by scraping up slag on the left and right sides of the pot (container 1A)" as the "operation pattern" of the scraper 1D1.
In the example shown in Fig. 7, when step S26 in Fig. 15 is executed for the second time, the result is determined to be NO, and when step S24 in Fig. 15 is executed for the third time, the control unit 1H controls the first actuator 1B and the second actuator 1E to realize the slag removal operation schedule during the period when the "elapsed time" is 8 to 10 minutes. Specifically, the control unit 1H controls the first actuator 1B to set the inclination angle of the vessel 1A to 32°, and controls the second actuator 1E to set the "number of scrapings" to 20 and move the scraper 1D1 to follow the slag position based on the image of the surface of the molten metal in the vessel 1A captured by the image capture unit 1I.

In step S28, the slag removal record database 1N7 acquires and stores the amount of slag remaining after slag removal that was compared with the target value of the amount of slag remaining after slag removal in step S26.
In addition, in step S28, the operation database 1M acquires and stores the operation record, equipment record, and operating record from the start to the end of the slag removal operation of the slag removal system 1.

Next, in step S29, the learning model unit 1N1 calculates the error between the real-time data (actual values) and the control schedule (indicated values) generated by the control schedule generation unit 1N based on the real-time data obtained by the first detection unit 1C, the second detection unit 1F, and the imaging unit 1I when the control unit 1H is controlling the first actuator 1B and the second actuator 1E based on the control schedule generated in step S22 (i.e., when step S24 is executed), and updates the parameters, formulas, etc. used by the control schedule generation unit 1N to generate a control schedule when the slag removal operation of the slag removal system 1 is next performed.

16 and 17 are diagrams for explaining an example of the calculation of errors and the updating of parameters, expressions, etc., performed in step S29 of FIG.
In the example shown in Figure 16, during execution of the process with the "number of strokes" set to "15th" (when the "elapsed time" is 5.8 min), molten iron loss occurs as shown in the image in the upper right of Figure 16, and dripping is detected by the imaging unit 1I.
Therefore, the result of the determination in step S26 in FIG. 15 is NO, and step S27 in FIG. 15 is executed.
In step S27, the schedule for the "stroke count" from the "16th time" onwards is revised from the schedule shown in the upper left of FIG. 16 to the schedule shown in the lower left of FIG.
Furthermore, as shown in the graph at the bottom right of Figure 16, the slag removal operation of the slag removal system 1 ends at the "elapsed time" of 5.8 min, when the occurrence of molten iron loss is detected (i.e., the curve shown by "actual results" ends).
As shown in the graph at the bottom right of Fig. 16, in the example shown in Fig. 16, at the "elapsed time" of 5.8 [min], the amount of remaining slag is almost equal to the target value, and the elapsed time does not exceed the time limit. In other words, in the example shown in Fig. 16, as described above, at the "elapsed time" of 5.8 [min], molten iron loss occurs, and therefore the determination in step S26 in Fig. 15 is NO.

In the example shown in Figure 17, "elapsed time [min]", "residual slag amount [t]", "whether or not molten iron loss has occurred", "immersion position of scraper 1D1 [mm]", "tilting angle of ladle (container 1A)", and "scraping speed [times/min]" are set in the input layer of AI model section 1K2, and the control schedule is predicted (changed) by AI model section 1N2.

As described above, when the slag removal device is operated based on the perception and judgment of the operator, the actual slag removal efficiency (how much slag can be removed in a limited time) and the amount of molten iron loss (the amount of molten iron scraped out) will depend on the operator's level of proficiency and the operating environment, etc., which may result in variations in the molten iron yield and components.
Therefore, in the second embodiment of the slag removal system 1, the learning model unit 1N1 performs analysis (learning) using past operation data (operational performance, equipment performance, and operating performance) of the molten metal to be slag removed and data on past slag removal performance to generate an optimal control schedule for the first actuator 1B and the second actuator 1E. Furthermore, the AI model unit 1N2 performs analysis using real-time data (instrumentation and image data) obtained during the slag removal operation to predict the optimal operation of the scraper 1D1. The calculation results of the learning model unit 1N1 and the prediction results of the AI model unit 1N2 are reflected in the generated control schedule, and the control unit 1H automatically controls the first actuator 1B and the second actuator 1E based on the generated control schedule.

Although the above describes the form for carrying out the present invention using the embodiments, the present invention is not limited to these embodiments in any way, and various modifications and substitutions can be made without departing from the scope of the present invention. The configurations described in the above-mentioned embodiments and examples may be combined.

In addition, the whole or part of the functions of each part of the slag removal system 1 in the above-mentioned embodiment may be realized by recording a program for realizing these functions in a computer-readable recording medium, reading the program recorded in the recording medium into a computer system, and executing it. Note that the "computer system" here includes hardware such as an OS and peripheral devices.
Furthermore, "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, and storage units such as hard disks built into computer systems. Furthermore, "computer-readable recording medium" may also include those that dynamically hold a program for a short period of time, such as a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line, and those that hold a program for a certain period of time, such as volatile memory inside a computer system that serves as a server or client in such cases. Furthermore, the above program may be one that realizes part of the above-mentioned functions, or may be one that can realize the above-mentioned functions in combination with a program already recorded in the computer system.

1...slag removal system, 1A...container, 1B...first actuator, 1B1...tilting device, 1B11...tilting cylinder, 1B2...container cart, 1C...first detection unit, 1C1...container tilt angle meter, 1C2...container cart position detection device, 1D...slag removal member, 1D1...scraper, 1D2...arm, 1E...second actuator, 1E1...cylinder, 1E11...lifting cylinder, 1E12...rotating cylinder, 1E2...traveling cart, 1E21...motor, 1F...second detection unit, 1F1...linear encoder, 1F2...rotary encoder, 1F3...arm imaging camera, 1F4...scraper coordinate calculation unit, 1G...operation unit, 1H...control unit, 1H1...tilting device control unit, 1 H2... Slag removal device control unit, 1I... Imaging unit, 1J... Weight distribution calculation unit, 1K... Guidance generation unit, 1K1... Learning model unit, 1K2... AI model unit, 1K3... Scraper state quantity acquisition unit, 1K4... Vessel state quantity acquisition unit, 1K5... Mold surface state quantity acquisition unit, 1K6... Slag removal operation schedule acquisition unit, 1K7... Slag removal performance database, 1L... Guidance output unit, 1M... Operation database, 1N... Control schedule generation unit, 1N1... Learning model unit, 1N2... AI model unit, 1N3... Scraper state quantity acquisition unit, 1N4... Vessel state quantity acquisition unit, 1N5... Mold surface state quantity acquisition unit, 1N6... Slag removal operation schedule acquisition unit, 1N7... Slag removal performance database, 1P... Display unit

Claims (36)

a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
a second detection unit that detects a displacement amount of the second actuator;
an operation unit that accepts operations by an operator;
a control unit that controls the first actuator and the second actuator in response to an operation received by the operation unit;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
a guidance generating unit that generates guidance for the operation of the operator with respect to the operation unit;
A guidance output unit that outputs the guidance generated by the guidance generating unit.
The guidance generation unit
a learning model unit that calculates a slag removal operation schedule, which is a schedule for the operation of the first actuator and the second actuator to remove slag using the slag removal member, based on a target value for a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after slag removal using the slag removal member;
and an AI model unit that generates guidance for the operator's operation on the operation unit that realizes the slag removal operation schedule calculated by the learning model unit based on the target value of the slag residual amount after slag removal;
The learning model unit includes:
A supervised learning is performed using a pair of the past operation data of the slag discharge system and the slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
The AI model unit includes:
correcting a guidance for an operation of the operator with respect to the operation unit based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator in response to an operation received by the operation unit;
The slag removal record constituting the teacher data used in the supervised learning performed by the learning model unit includes the amount of S and the amount of molten iron loss.
Drainage system. In the supervised learning performed by the learning model unit,
A combination of a slag disposal record in which the amount of sulfur return is smaller than the sulfur return threshold and the amount of molten iron loss is smaller than the molten iron loss threshold, and past operation data of the slag disposal system when the amount of sulfur return is smaller than the sulfur return threshold and the amount of molten iron loss is smaller than the molten iron loss threshold is obtained is used as correct answer data of the teacher data,
A pair of a slag disposal record in which the amount of sulfur removal is equal to or greater than the sulfur removal threshold value, or the amount of molten iron loss is equal to or greater than the molten iron loss threshold value, and past operation data of the slag disposal system when the amount of sulfur removal is equal to or greater than the sulfur removal threshold value, or the amount of molten iron loss is equal to or greater than the molten iron loss threshold value, is used as incorrect answer data of the teacher data.
The tailings removal system of claim 1 . a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
an operation unit that accepts operations by an operator;
a control unit that controls the first actuator and the second actuator in response to an operation received by the operation unit;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
a guidance generating unit that generates guidance for the operation of the operator with respect to the operation unit;
A guidance output unit that outputs the guidance generated by the guidance generating unit.
The guidance generation unit
a learning model unit that calculates a slag removal operation schedule, which is a schedule for the operation of the first actuator and the second actuator to remove slag using the slag removal member, based on a target value for a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after slag removal using the slag removal member;
and an AI model unit that generates guidance for the operator's operation on the operation unit that realizes the slag removal operation schedule calculated by the learning model unit based on the target value of the slag residual amount after slag removal;
The learning model unit includes:
A supervised learning is performed using a pair of the past operation data of the slag discharge system and the slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
The AI model unit includes:
correcting a guidance for an operation of the operator with respect to the operation unit based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator in response to an operation received by the operation unit;
The past operation data of the slag removal system constituting the teacher data used in the supervised learning performed by the learning model unit includes data on the shape of the slag removal member.
Drainage system. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
an operation unit that accepts operations by an operator;
a control unit that controls the first actuator and the second actuator in response to an operation received by the operation unit;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
a guidance generating unit that generates guidance for the operation of the operator with respect to the operation unit;
A guidance output unit that outputs the guidance generated by the guidance generating unit.
The guidance generation unit
a learning model unit that calculates a slag removal operation schedule, which is a schedule for the operation of the first actuator and the second actuator to remove slag using the slag removal member, based on a target value for a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after slag removal using the slag removal member;
and an AI model unit that generates guidance for the operator's operation on the operation unit that realizes the slag removal operation schedule calculated by the learning model unit based on the target value of the slag residual amount after slag removal;
The learning model unit includes:
A supervised learning is performed using a pair of the past operation data of the slag discharge system and the slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
The AI model unit includes:
correcting a guidance for an operation of the operator with respect to the operation unit based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator in response to an operation received by the operation unit;
Data relating to the molten metal and the slag removed from the surface of the molten metal handled by the slag removal system and data relating to the shape of the slag removal member are set in an input layer of the learning model unit;
The prediction of the movement of the container and the slag removal member and the prediction of the slag removal performance are set in the output layer of the learning model unit.
Drainage system. The slag removal performance constituting the teacher data used in the supervised learning performed by the learning model unit includes the amount of slag remaining after slag removal.
A tailings removal system according to any one of claims 1 to 4 . The past operation data of the slag removal system constituting the teaching data used in the supervised learning performed by the learning model unit includes data regarding the movement of the container and the slag removal member.
A tailings removal system according to any one of claims 1 to 4 . The past operation data of the slag removal system constituting the teaching data used in the supervised learning performed by the learning model unit includes data on the molten metal handled by the slag removal system and slag removed from the surface of the molten metal.
A tailings removal system according to any one of claims 1 to 4 . The first actuator includes:
A tilting device having a function of adjusting the tilt angle of the container;
a container carriage capable of supporting the container and the tilting device and being movable;
The first detection unit includes:
a container tilt angle meter for detecting a tilt angle of the container;
and a container cart position detection device for detecting the position of the container cart.
A tailings removal system according to any one of claims 1 to 4 . The slag removal member includes:
A scraper and
and an arm having the scraper attached to its tip,
The second actuator includes:
A cylinder for moving the arm in a vertical direction or a rotational direction;
a traveling carriage for moving the arm in a horizontal direction;
The second detection unit includes:
a linear encoder for detecting a stroke amount of the cylinder;
A rotary encoder for detecting the number of rotations of the wheels of the traveling carriage;
an arm imaging camera that captures an image of the arm;
A scraper coordinate calculation unit that calculates the coordinates of the scraper based on the image of the arm captured by the arm imaging camera is included.
A tailings removal system according to any one of claims 1 to 4 . The control unit includes:
a tilting device control unit that controls the first actuator in response to an operation received by the operation unit;
A dross removal device control unit that controls the second actuator in response to an operation received by the operation unit.
A tailings removal system according to any one of claims 1 to 4 . a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
an operation unit that accepts operations by an operator;
a control unit that controls the first actuator and the second actuator in response to an operation received by the operation unit;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
a guidance generating unit that generates guidance for the operation of the operator with respect to the operation unit;
A guidance generating method for a slag removal system comprising: a guidance output unit that outputs the guidance generated by the guidance generating unit,
a calculation step of calculating a slag removal operation schedule, which is a schedule of operations of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after the slag removal member has removed the slag;
A generating step of generating an operation guidance of the operator on the operating unit that realizes the slag removal operation schedule calculated in the calculating step based on the target value of the slag residual amount after the slag removal;
A learning step of performing supervised learning using a pair of past operation data of the slag discharge system and a slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
a correction step of correcting a guidance of an operation of the operator with respect to the operation unit based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator in response to an operation received by the operation unit ,
The schedule of the operations of the first actuator and the second actuator calculated in the calculation step includes:
a time period during which the first actuator and the second actuator operate without being based on an image of the surface of the molten metal in the vessel captured by the imaging unit; and
a time period during which the first actuator and the second actuator are operated based on an image of the surface of the molten metal in the container captured by the imaging unit,
The operation of the first actuator and the second actuator not based on an image of the surface of the molten metal in the container captured by the imaging unit is
The operation of the first actuator and the second actuator is performed prior to the operation of the first actuator and the second actuator based on an image of the surface of the molten metal in the container captured by the imaging unit.
A method for generating guidance for a tailings removal system. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
an operation unit that accepts operations by an operator;
a control unit that controls the first actuator and the second actuator in response to an operation received by the operation unit;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
a guidance generating unit that generates guidance for the operation of the operator with respect to the operation unit;
A guidance output unit that outputs the guidance generated by the guidance generating unit.
a calculation step of calculating a slag removal operation schedule, which is a schedule of operations of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after the slag removal member has removed the slag;
A generating step of generating an operation guidance of the operator on the operating unit that realizes the slag removal operation schedule calculated in the calculating step based on the target value of the slag residual amount after the slag removal;
A learning step of performing supervised learning using a pair of past operation data of the slag discharge system and a slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
a correction step of correcting a guidance of an operation of the operator with respect to the operation unit, based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator in response to an operation received by the operation unit,
The program, wherein the slag removal record constituting the teacher data used in the supervised learning performed in the learning step includes the amount of re-sulfurization and the amount of molten iron loss. In the supervised learning performed in the learning step,
A combination of a slag disposal record in which the amount of sulfur return is smaller than the sulfur return threshold and the amount of molten iron loss is smaller than the molten iron loss threshold, and past operation data of the slag disposal system when the amount of sulfur return is smaller than the sulfur return threshold and the amount of molten iron loss is smaller than the molten iron loss threshold is obtained is used as correct answer data of the teacher data,
A pair of a slag disposal record in which the amount of sulfur return is smaller than the sulfur return threshold or the amount of molten iron loss is smaller than the molten iron loss threshold, and past operation data of the slag disposal system when the amount of sulfur return is smaller than the sulfur return threshold or the amount of molten iron loss is smaller than the molten iron loss threshold is obtained is used as incorrect answer data of the teacher data.
The program according to claim 12. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
an operation unit that accepts operations by an operator;
a control unit that controls the first actuator and the second actuator in response to an operation received by the operation unit;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
a guidance generating unit that generates guidance for the operation of the operator with respect to the operation unit;
A guidance output unit that outputs the guidance generated by the guidance generating unit.
a calculation step of calculating a slag removal operation schedule, which is a schedule of operations of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after the slag removal member has removed the slag;
A generating step of generating an operation guidance of the operator on the operating unit that realizes the slag removal operation schedule calculated in the calculating step based on the target value of the slag residual amount after the slag removal;
A learning step of performing supervised learning using a pair of past operation data of the slag discharge system and a slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
a correction step of correcting a guidance of an operation of the operator with respect to the operation unit, based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator in response to an operation received by the operation unit,
A program in which the past operating data of the slag removal system that constitutes the teaching data used in the supervised learning performed in the learning step includes data regarding the shape of the slag removal member. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
an operation unit that accepts operations by an operator;
a control unit that controls the first actuator and the second actuator in response to an operation received by the operation unit;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
a guidance generating unit that generates guidance for the operation of the operator with respect to the operation unit;
A guidance output unit that outputs the guidance generated by the guidance generating unit.
a calculation step of calculating a slag removal operation schedule, which is a schedule of operations of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after the slag removal member has removed the slag;
A generating step of generating an operation guidance of the operator on the operating unit that realizes the slag removal operation schedule calculated in the calculating step based on the target value of the slag residual amount after the slag removal;
A learning step of performing supervised learning using a pair of past operation data of the slag discharge system and a slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
a correction step of correcting a guidance of an operation of the operator with respect to the operation unit, based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit controls the first actuator and the second actuator in response to an operation received by the operation unit,
In the learning step, data on the molten metal and the slag removed from the surface of the molten metal handled by the slag removal system and data on the shape of the slag removal member are set in an input layer;
A program in which predictions of the movement of the container and the slag removal member and predictions of slag removal performance are set in an output layer. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
A control unit that controls the first actuator and the second actuator;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
A slag removal system comprising: a control schedule generating unit that generates a control schedule for the first actuator and the second actuator by the control unit,
The control schedule generation unit
a learning model unit that calculates a slag removal operation schedule, which is a schedule for the operation of the first actuator and the second actuator to remove slag using the slag removal member, based on a target value for a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after slag removal using the slag removal member;
and an AI model unit that generates a control schedule for the first actuator and the second actuator to realize the slag removal operation schedule calculated by the learning model unit based on the target value of the slag residual amount after the slag removal,
The learning model unit includes:
A supervised learning is performed using a pair of the past operation data of the slag discharge system and the slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
The AI model unit includes:
correcting a control schedule for the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit while the control unit is controlling the first actuator and the second actuator based on the control schedule generated by the learning model unit ;
The slag removal record constituting the teacher data used in the supervised learning performed by the learning model unit includes the amount of S returned and the amount of molten iron loss.
Drainage system. In the supervised learning performed by the learning model unit,
A combination of a slag disposal record in which the amount of sulfur return is smaller than the sulfur return threshold and the amount of molten iron loss is smaller than the molten iron loss threshold, and past operation data of the slag disposal system when the amount of sulfur return is smaller than the sulfur return threshold and the amount of molten iron loss is smaller than the molten iron loss threshold is obtained is used as correct answer data of the teacher data,
A pair of a slag disposal record in which the amount of sulfur return is smaller than the sulfur return threshold or the amount of molten iron loss is smaller than the molten iron loss threshold, and past operation data of the slag disposal system when the amount of sulfur return is smaller than the sulfur return threshold or the amount of molten iron loss is smaller than the molten iron loss threshold is obtained is used as incorrect answer data of the teacher data.
17. The tailings removal system of claim 16 . a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
A control unit that controls the first actuator and the second actuator;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
A slag removal system comprising: a control schedule generating unit that generates a control schedule for the first actuator and the second actuator by the control unit,
The control schedule generation unit
a learning model unit that calculates a slag removal operation schedule, which is a schedule for the operation of the first actuator and the second actuator to remove slag using the slag removal member, based on a target value for a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after slag removal using the slag removal member;
and an AI model unit that generates a control schedule for the first actuator and the second actuator to realize the slag removal operation schedule calculated by the learning model unit based on the target value of the slag residual amount after the slag removal,
The learning model unit includes:
A supervised learning is performed using a pair of the past operation data of the slag discharge system and the slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
The AI model unit includes:
correcting a control schedule for the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit while the control unit is controlling the first actuator and the second actuator based on the control schedule generated by the learning model unit ;
The past operation data of the slag removal system constituting the teacher data used in the supervised learning performed by the learning model unit includes data on the shape of the slag removal member.
Drainage system. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
A control unit that controls the first actuator and the second actuator;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
A slag removal system comprising: a control schedule generating unit that generates a control schedule for the first actuator and the second actuator by the control unit,
The control schedule generation unit
a learning model unit that calculates a slag removal operation schedule, which is a schedule for the operation of the first actuator and the second actuator to remove slag using the slag removal member, based on a target value for a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after slag removal using the slag removal member;
and an AI model unit that generates a control schedule for the first actuator and the second actuator to realize the slag removal operation schedule calculated by the learning model unit based on the target value of the slag residual amount after the slag removal,
The learning model unit includes:
A supervised learning is performed using a pair of the past operation data of the slag discharge system and the slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
The AI model unit includes:
correcting a control schedule for the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit while the control unit is controlling the first actuator and the second actuator based on the control schedule generated by the learning model unit ;
Data relating to the molten metal and the slag removed from the surface of the molten metal handled by the slag removal system and data relating to the shape of the slag removal member are set in an input layer of the learning model unit;
The prediction of the movement of the container and the slag removal member and the prediction of the slag removal performance are set in the output layer of the learning model unit.
Drainage system. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
A control unit that controls the first actuator and the second actuator;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
A slag removal system comprising: a control schedule generating unit that generates a control schedule for the first actuator and the second actuator by the control unit,
The control schedule generation unit
a learning model unit that calculates a slag removal operation schedule, which is a schedule for the operation of the first actuator and the second actuator to remove slag using the slag removal member, based on a target value for a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after slag removal using the slag removal member;
and an AI model unit that generates a control schedule for the first actuator and the second actuator to realize the slag removal operation schedule calculated by the learning model unit based on the target value of the slag residual amount after the slag removal,
The learning model unit includes:
A supervised learning is performed using a pair of the past operation data of the slag discharge system and the slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
The AI model unit includes:
correcting a control schedule for the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit while the control unit is controlling the first actuator and the second actuator based on the control schedule generated by the learning model unit ;
When the control unit controls the first actuator and the second actuator based on the control schedule generated by the learning model unit, the real-time data obtained by the first detection unit, the second detection unit, and the imaging unit includes the amount of slag remaining on the surface of the molten metal in the vessel obtained by the AI model unit,
The amount of slag remaining on the surface of the molten metal in the container acquired by the AI model unit is calculated by the weight distribution calculation unit.
Drainage system. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
A control unit that controls the first actuator and the second actuator;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
A slag removal system comprising: a control schedule generating unit that generates a control schedule for the first actuator and the second actuator by the control unit,
The control schedule generation unit
a learning model unit that calculates a slag removal operation schedule, which is a schedule for the operation of the first actuator and the second actuator to remove slag using the slag removal member, based on a target value for a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after slag removal using the slag removal member;
and an AI model unit that generates a control schedule for the first actuator and the second actuator to realize the slag removal operation schedule calculated by the learning model unit based on the target value of the slag residual amount after the slag removal,
The learning model unit includes:
A supervised learning is performed using a pair of the past operation data of the slag discharge system and the slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
The AI model unit includes:
correcting a control schedule for the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit while the control unit is controlling the first actuator and the second actuator based on the control schedule generated by the learning model unit ;
The slag removal member includes:
A scraper and
and an arm having the scraper attached to its tip,
The second actuator includes:
A cylinder for moving the arm vertically or in a rotational direction;
a traveling carriage for moving the arm in a horizontal direction;
The second detection unit includes:
a linear encoder for detecting a stroke amount of the cylinder;
A rotary encoder for detecting the number of rotations of the wheels of the traveling carriage;
an arm imaging camera that captures an image of the arm;
A scraper coordinate calculation unit that calculates the coordinates of the scraper based on the image of the arm captured by the arm imaging camera is included.
Drainage system. The slag removal performance constituting the teacher data used in the supervised learning performed by the learning model unit includes the amount of slag remaining after slag removal.
22. A tailings removal system according to any one of claims 16 to 21 . The past operation data of the slag removal system constituting the teacher data used in the supervised learning performed by the learning model unit includes data regarding the movement of the container and the slag removal member.
22. A tailings removal system according to any one of claims 16 to 21 . The past operation data of the slag removal system constituting the teaching data used in the supervised learning performed by the learning model unit includes data on the molten metal handled by the slag removal system and slag removed from the surface of the molten metal.
22. A tailings removal system according to any one of claims 16 to 21 . The slag removal operation time, the amount of slag remaining after slag removal, the occurrence of molten iron loss, the immersion position of the slag removal member, the inclination angle of the vessel, and the scraping speed of the slag removal member are set in the input layer of the AI model section.
22. A tailings removal system according to any one of claims 16 to 21 . The first actuator includes:
A tilting device having a function of adjusting the tilt angle of the container;
a container carriage capable of supporting the container and the tilting device and being movable;
The first detection unit includes:
a container tilt angle meter for detecting a tilt angle of the container;
and a container cart position detection device for detecting the position of the container cart.
22. A tailings removal system according to any one of claims 16 to 21 . The control unit includes:
A tilting device control unit for controlling the first actuator;
and a slag removal device control unit that controls the second actuator.
22. A tailings removal system according to any one of claims 16 to 21 . a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least a tilt angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
A control unit that controls the first actuator and the second actuator;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
A method for automatically controlling a tailings removal system comprising: a control schedule generating unit for generating a control schedule for the first actuator and the second actuator by the control unit,
a calculation step of calculating a slag removal operation schedule, which is a schedule of operations of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after the slag removal member has removed the slag;
A generating step of generating a control schedule of the first actuator and the second actuator for realizing the slag removal operation schedule calculated in the calculating step based on the target value of the slag residual amount after the slag removal;
A learning step of performing supervised learning using a pair of past operation data of the slag discharge system and a slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
a correction step of correcting the control schedule of the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit while the control unit is controlling the first actuator and the second actuator based on the control schedule generated in the generation step.
A method for automatically controlling a tailings removal system. The schedule of the operations of the first actuator and the second actuator calculated in the calculation step includes:
a time period during which the first actuator and the second actuator are operated without being based on an image of the surface of the molten metal in the container captured by the imaging unit; and
and a time period during which the first actuator and the second actuator are operated based on an image of the surface of the molten metal in the container captured by the imaging unit.
29. A method for automatically controlling a tailings removal system according to claim 28 . The operation of the first actuator and the second actuator not based on an image of the surface of the molten metal in the container captured by the imaging unit is
The operation of the first actuator and the second actuator is performed prior to the operation of the first actuator and the second actuator based on an image of the surface of the molten metal in the container captured by the imaging unit.
30. The method for automatically controlling a tailings removal system according to claim 29 . a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
A control unit that controls the first actuator and the second actuator;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
A computer installed in the slag removal system includes a control schedule generating unit that generates a control schedule for the first actuator and the second actuator by the control unit,
a calculation step of calculating a slag removal operation schedule, which is a schedule of operations of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after the slag removal member has removed the slag;
A generating step of generating a control schedule of the first actuator and the second actuator for realizing the slag removal operation schedule calculated in the calculating step based on the target value of the slag residual amount after the slag removal;
A learning step of performing supervised learning using a pair of past operation data of the slag discharge system and a slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
a correction step of correcting a control schedule of the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit while the control unit is controlling the first actuator and the second actuator based on the control schedule generated in the generation step ,
The program, wherein the slag removal record constituting the teacher data used in the supervised learning performed in the learning step includes the amount of re-sulfurization and the amount of molten iron loss. In the supervised learning performed in the learning step,
A combination of a slag disposal record in which the amount of sulfur return is smaller than the sulfur return threshold and the amount of molten iron loss is smaller than the molten iron loss threshold, and past operation data of the slag disposal system when the amount of sulfur return is smaller than the sulfur return threshold and the amount of molten iron loss is smaller than the molten iron loss threshold is obtained is used as correct answer data of the teacher data,
A pair of a slag disposal record in which the amount of sulfur return is smaller than the sulfur return threshold or the amount of molten iron loss is smaller than the molten iron loss threshold, and past operation data of the slag disposal system when the amount of sulfur return is smaller than the sulfur return threshold or the amount of molten iron loss is smaller than the molten iron loss threshold is obtained is used as incorrect answer data of the teacher data.
32. The program of claim 31. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
A control unit that controls the first actuator and the second actuator;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
A computer installed in the slag removal system includes a control schedule generating unit that generates a control schedule for the first actuator and the second actuator by the control unit,
a calculation step of calculating a slag removal operation schedule, which is a schedule of operations of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after the slag removal member has removed the slag;
A generating step of generating a control schedule of the first actuator and the second actuator for realizing the slag removal operation schedule calculated in the calculating step based on the target value of the slag residual amount after the slag removal;
A learning step of performing supervised learning using a pair of past operation data of the slag discharge system and a slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
a correction step of correcting a control schedule of the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit while the control unit is controlling the first actuator and the second actuator based on the control schedule generated in the generation step ,
A program in which the past operating data of the slag removal system that constitutes the teaching data used in the supervised learning performed in the learning step includes data regarding the shape of the slag removal member. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
A control unit that controls the first actuator and the second actuator;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
A computer installed in the slag removal system includes a control schedule generating unit that generates a control schedule for the first actuator and the second actuator by the control unit,
a calculation step of calculating a slag removal operation schedule, which is a schedule of operations of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after the slag removal member has removed the slag;
A generating step of generating a control schedule of the first actuator and the second actuator for realizing the slag removal operation schedule calculated in the calculating step based on the target value of the slag residual amount after the slag removal;
A learning step of performing supervised learning using a pair of past operation data of the slag discharge system and a slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
a correction step of correcting a control schedule of the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit while the control unit is controlling the first actuator and the second actuator based on the control schedule generated in the generation step ,
In the learning step, data on the molten metal and slag removed from the surface of the molten metal handled by the slag removal system and data on the shape of the slag removal member are set in an input layer;
A program in which predictions of the movement of the vessel and the slag removal member and predictions of slag removal performance are set in an output layer. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
A control unit that controls the first actuator and the second actuator;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
A computer installed in the slag removal system includes a control schedule generating unit that generates a control schedule for the first actuator and the second actuator by the control unit,
a calculation step of calculating a slag removal operation schedule, which is a schedule of operations of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after the slag removal member has removed the slag;
A generating step of generating a control schedule of the first actuator and the second actuator for realizing the slag removal operation schedule calculated in the calculating step based on the target value of the slag residual amount after the slag removal;
A learning step of performing supervised learning using a pair of past operation data of the slag disposal system and a slag disposal performance when the past operation data of the slag disposal system was obtained as teacher data;
a correction step of correcting a control schedule of the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit while the control unit is controlling the first actuator and the second actuator based on the control schedule generated in the generation step ,
A program, wherein real-time data obtained by the first detection unit, the second detection unit, and the imaging unit when the control unit is controlling the first actuator and the second actuator based on the control schedule includes the amount of slag remaining on the surface of the molten metal in the container calculated by the weight distribution calculation unit. a vessel for containing molten metal;
A first actuator for at least tilting the container;
A first detection unit that detects at least an inclination angle of the container;
a slag removal member for removing slag from the surface of the molten metal in the vessel;
A second actuator for moving the slag removal member;
A second detection unit that detects a displacement amount of the second actuator;
A control unit that controls the first actuator and the second actuator;
an imaging unit that images a surface of the molten metal in the container;
a weight distribution calculation unit that calculates a weight distribution of residual slag remaining on the surface of the molten metal in the vessel based on an image of the surface of the molten metal in the vessel captured by the imaging unit;
A computer installed in the slag removal system includes a control schedule generating unit that generates a control schedule for the first actuator and the second actuator by the control unit,
a calculation step of calculating a slag removal operation schedule, which is a schedule of operations of the first actuator and the second actuator for removing slag by the slag removal member, based on a target value of a slag residual amount after slag removal, which is the amount of slag remaining on the surface of the molten metal in the vessel after the slag removal member has removed the slag;
A generating step of generating a control schedule of the first actuator and the second actuator for realizing the slag removal operation schedule calculated in the calculating step based on the target value of the slag residual amount after the slag removal;
A learning step of performing supervised learning using a pair of past operation data of the slag discharge system and a slag discharge performance when the past operation data of the slag discharge system was obtained as teacher data;
a correction step of correcting a control schedule of the first actuator and the second actuator based on real-time data obtained by the first detection unit, the second detection unit, and the imaging unit while the control unit is controlling the first actuator and the second actuator based on the control schedule generated in the generation step ,
The slag removal member includes:
A scraper and
and an arm having the scraper attached to its tip,
The second actuator includes:
A cylinder for moving the arm vertically or in a rotational direction;
a traveling carriage for moving the arm in a horizontal direction;
The second detection unit includes:
a linear encoder for detecting a stroke amount of the cylinder;
A rotary encoder for detecting the number of rotations of the wheels of the traveling carriage;
an arm imaging camera that captures an image of the arm;
The program includes a scratcher coordinate calculation unit that calculates the coordinates of the scratcher based on an image of the arm captured by the arm imaging camera.

Images