Apparatus and method for determining path, program, driller and path determining model
The trajectory determination device uses a machine learning model to automate slag removal in molten metal furnaces, addressing inefficiencies by generating optimal trajectories for slag removal, thereby improving operational efficiency.
Patent Information
- Application Number
- JP2023082825
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-19
- Publication Date
- 2024-11-29
AI Technical Summary
Existing slag removal processes in molten metal melting furnaces rely heavily on skilled operators and lack efficient automation using artificial intelligence, leading to inefficiencies in removing floating impurities.
A trajectory determination device and method using a machine learning model to analyze molten metal surface images and generate trajectory information for a slag removal member, enabling automated and efficient slag removal operations.
The system effectively removes slag from molten metal surfaces by determining optimal trajectories for the slag removal member, enhancing efficiency and reducing reliance on skilled operators.
Smart Images

Figure 2024166619000001_ABST
Abstract
Description
[Technical field]
[0001] The present disclosure relates to an orbit determination device, an orbit determination method, a program, a training device, and an orbit determination model. [Background technology]
[0002] Impurities (which may be called "slag" or "slag") separate from the molten metal in the melting furnace and float on the surface of the molten metal in the melting furnace. In order to remove the floating impurities (hereinafter referred to as "slag"), a slag removal operation is performed. In a typical slag removal operation, an operator operates a slag remover while visually checking the surface of the molten metal in the melting furnace or while checking an image of the molten metal surface transmitted from a camera that captures the surface of the molten metal in the melting furnace, and the slag floating in the melting furnace is scraped out of the furnace by a dragger attached to the slag remover.
[0003] To achieve efficient slag removal work, the slag remover needs to be operated by a skilled operator, and technology for automating the operation of the slag remover has been proposed. [Prior art documents] [Patent documents]
[0004] [Patent Document 1] JP 2019-105389 A [Patent Document 2] JP 2018-179348 A [Patent Document 3] JP 2020-085395 A Summary of the Invention [Problem to be solved by the invention]
[0005] With the recent advancement of deep learning technology, artificial intelligence (AI) technology has been used in various technical fields. A technology for realizing efficient slag removal work by using a machine learning model trained on the operation of a slag remover by a skilled operator is desirable.
[0006] One objective of the present disclosure is to provide a technique for efficiently removing slag from the surface of molten metal. [Means for solving the problem]
[0007] One aspect of the present disclosure relates to an trajectory determination device having a state acquisition unit that acquires image data showing a surface of molten metal in a container, and an trajectory determination unit that uses a trajectory determination model to generate trajectory information showing a trajectory of a slag removal member for scraping off slag on the molten metal based on the image data. Effect of the Invention
[0008] According to the present disclosure, slag on the surface of molten metal can be efficiently removed. [Brief description of the drawings]
[0009] [Figure 1] FIG. 1 is a schematic diagram illustrating a slag removal system according to one embodiment of the present disclosure. [Diagram 2] 2A-2D are schematic diagrams illustrating the operation of a dragger for scraping slag out of a melting furnace according to one embodiment of the present disclosure. [Diagram 3] FIG. 3 is a block diagram illustrating a hardware configuration of an orbit determination device according to an embodiment of the present disclosure. [Figure 4] FIG. 4 is a block diagram illustrating a functional configuration of an orbit determination device according to an embodiment of the present disclosure. [Diagram 5] FIG. 5 is a schematic diagram illustrating an orbit determination model according to one embodiment of the present disclosure. [Figure 6] 6A-6C are diagrams illustrating example control signals for a dessicator according to one embodiment of the present disclosure. [Figure 7] 7A-7C are diagrams illustrating a central slag discharge, a left-turn slag discharge, and a right-turn slag discharge according to one embodiment of the present disclosure. [Figure 8] 8A-8C are diagrams illustrating a trajectory pattern according to one embodiment of the present disclosure. [Figure 9] FIG. 9 is a diagram illustrating a section of the surface area of molten metal in a melting furnace according to one embodiment of the present disclosure. [Figure 10] FIG. 10 is a block diagram illustrating a functional configuration of a training device according to an embodiment of the present disclosure. [Figure 11] FIG. 11 is a flowchart illustrating an orbit determination process according to one embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0010] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.
[0011] In the following embodiments, a trajectory determination device is disclosed that determines the trajectory of a dragger for scraping slag of molten metal from a melting furnace.
[0012] [Removal system] 1, a slag removal system 10 according to an embodiment of the present disclosure includes a vessel 20 for accommodating molten metal such as a melting furnace, a slag remover 30 for removing slag from the molten metal in the vessel 20, a camera 40 for capturing an image of the surface of the molten metal in the vessel 20, an operation terminal 50 for operating the slag remover 30, and a trajectory determination device 100 for determining the trajectory of a dragger 31 of the slag remover 30. In the following embodiment, the slag remover 30 includes a dragger 31, a vessel tilting unit 32, and a scraper 33, and the dragger 31 is a scraping-type slag dragger equipped with the scraper 33 at its tip.
[0013] In the slag removal system 10, the trajectory determination device 100 acquires image data showing the surface of the molten metal in the vessel 20 captured by the camera 40, and generates trajectory information showing the trajectory of the dragger 31 based on the acquired image data using the trajectory determination model 60. For example, the trajectory determination model 60 may be any machine learning model trained to receive image data showing the surface of the molten metal and determine the trajectory of the dragger 31 for effectively scraping out the floating slag from the vessel 20 based on the distribution state of the slag in the image data. As an example, the trajectory determination model 60 is trained to reproduce the trajectory of the dragger 31 operated by a skilled operator who operates the dragger 31 of the slag remover 30 to efficiently remove the slag floating on the surface of the molten metal while checking the captured image of the surface of the molten metal in the vessel 20 tilted by the vessel tilting unit 32 of the slag remover 30.
[0014] For example, for image data showing slag 21 floating on the surface of the molten metal as shown in FIG. 2A, the trajectory information output from the trajectory determination model 60 may first instruct the dragger 31 to extend in the Y direction as shown in FIG. 2B. Next, the trajectory information may instruct the dragger 31 to rotate counterclockwise as shown in FIG. 2C, and finally instruct the dragger 31 to return the extended dragger 31 to scrape the slag 21 forward as shown in FIG. 2D. Here, the dragger 31 is a scraping type dragger, and a scraper plate 33 for scraping the slag 21 is provided at the tip of the dragger 31. The scraper plate 33 lowered to the surface of the molten metal moves the slag 21 in the contraction direction of the dragger 31 in accordance with the return operation of the dragger 31 as shown in FIG. 2D, and finally discharges the slag 21 into a slag pan (not shown) outside the vessel 20.
[0015] In the illustrated embodiment, the orbit determination model 60 is mounted on the orbit determination device 100, but the orbit determination model 60 according to the present disclosure is not limited to this and may be disposed in another device communicatively connected to the orbit determination device 100. In this case, the orbit determination device 100 may transmit image data to the device and receive orbit information obtained from the orbit determination model 60 for the image data from the device.
[0016] When the orbit information is acquired, the orbit determination device 100 may present the orbit information acquired from the orbit determination model 60 to the operation terminal 50 as guidance information for the operator of the operation terminal 50. When the guidance information is acquired, the operator can operate the operation terminal 50 according to the acquired guidance information.
[0017] Alternatively, the orbit determination device 100 may transmit the orbit information acquired from the orbit determination model 60 to the skimmer 30 as a control signal for controlling the operation of the skimmer 30. When the control signal is acquired, the dragger 31 can be automatically operated according to the acquired control signal.
[0018] In this way, the trajectory determination device 100 can use the trajectory determination model 60 to determine trajectory information indicating the trajectory of the dragger 31 suitable for the distribution state of the slag 21 captured in the image data from the image data showing the surface of the molten metal in the vessel 20. This makes it possible to realize the operation of the dragger 31 suitable for the distribution state of the slag 21.
[0019] Here, the orbit determination device 100 may be realized by a computing device such as a personal computer, and may have, for example, a hardware configuration as shown in Fig. 3. That is, the orbit determination device 100 has a storage device 101, a processor 102, a user interface (UI) device 103, and a communication device 104, which are interconnected via a bus B.
[0020] The programs or instructions for implementing various functions and processes in the orbit determination device 100, which will be described later, may be downloaded from an external device via a network or the like, or may be provided from a removable storage medium such as a CD-ROM (Compact Disk-Read Only Memory) or a flash memory.
[0021] Storage device 101 may be implemented by random access memory, flash memory, a hard disk drive, etc., and stores installed programs or instructions as well as files, data, etc. used in the execution of the programs or instructions. Storage device 101 may also include a non-transitory storage medium.
[0022] The processor 102 may be realized by one or more central processing units (CPUs), graphics processing units (GPUs), processing circuitry, etc., which may be composed of one or more processor cores, and performs various functions and processes of the orbit determination device 100 described below in accordance with programs, instructions, and data such as parameters necessary to execute the programs or instructions stored in the memory device 101.
[0023] The interface device 103 realizes an interface with a user of the orbit determination device 100. For example, the user operates a keyboard, a mouse, or the like to use a GUI (Graphical User Interface) displayed on a display or a touch panel, and transmits and receives various information, data, instructions, and the like between the orbit determination device 100 and the interface device 103.
[0024] The communication device 104 is realized by various communication circuits that execute communication processes with external devices, the Internet, and communication networks such as a LAN (Local Area Network).
[0025] However, the above-described hardware configuration is merely an example, and the orbit determination device 100 according to the present disclosure may be realized by any other appropriate hardware configuration.
[0026] [Trajectory determination device] Next, the orbit determination device 100 according to an embodiment of the present disclosure will be described with reference to Fig. 4. Fig. 4 is a block diagram showing a functional configuration of the orbit determination device 100 according to an embodiment of the present disclosure.
[0027] 4, the orbit determination device 100 includes a state acquisition unit 110 and an orbit determination unit 120. Each functional unit of the state acquisition unit 110 and the orbit determination unit 120 may be realized by causing the processor 102 to execute a computer program stored in the storage device 101 of the orbit determination device 100.
[0028] The state acquisition unit 110 acquires image data showing the surface of the molten metal in the vessel 20. Specifically, the state acquisition unit 110 acquires image data showing the surface of the molten metal in the vessel 20 captured by the camera 40 from the camera 40. Here, the image data may be image frames sampled from video data of the surface of the molten metal captured by the camera 40. The sampling of the image frames may be performed, for example, at the start of the slag removal operation by the dragger 31, and the sampled image frames may show the distribution state of the slag 21 floating on the surface of the molten metal at the start of the slag removal operation. When the image data is acquired from the camera 40, the state acquisition unit 110 provides the acquired image data to the trajectory determination unit 120. In addition, the state acquisition unit 110 may transfer the video data captured by the camera 40 to the operation terminal 50 so that the operator can check the state of the surface of the molten metal and / or the movement of the dragger 31 during the slag removal operation on the display of the operation terminal 50.
[0029] The trajectory determination unit 120 generates trajectory information indicating the trajectory of the slag removal member for scraping out the slag 21 on the molten metal based on the image data, using the trajectory determination model 60. Specifically, when the image data is acquired from the state acquisition unit 110, the trajectory determination unit 120 inputs the acquired image data to the trajectory determination model 60, and acquires from the trajectory determination model 60 trajectory information indicating the trajectory of the dragger 31 for efficiently scraping out the slag 21 floating on the surface of the molten metal in the image data.
[0030] Here, the trajectory information may indicate the two-dimensional movement of the dragger 31 relative to the surface of the molten metal, i.e., the extension and rotation of the dragger 31 as shown in FIG. 2. The trajectory information may also indicate the vertical movement of the dragger 31 relative to the surface of the molten metal, i.e., the lifting and lowering of the scraper 33 at the tip of the dragger 31. The height of the surface of the molten metal may vary depending on the remaining amount of molten metal in the container 20 tilted by the container tilting unit 32, the tilt angle of the container 20, and the like. For this reason, the position of the scraper 33 in the vertical direction in the trajectory information provided by the trajectory determination model 60 may enable efficient removal of the slag 21 and minimize unnecessary scraping of the molten metal.
[0031] In one embodiment, the trajectory determination unit 120 may generate guidance information for operating the slag removal member based on the trajectory information, and present the guidance information to the operator. Specifically, when the trajectory information is acquired from the trajectory determination model 60, the trajectory determination unit 120 may generate guidance information for assisting the operator in operating the operation terminal 50 based on the acquired trajectory information, and may display the guidance information on the display of the operation terminal 50, and / or may notify the guidance information by voice through the speaker of the operation terminal 50. With the assistance of the guidance information, the operator can operate the operation terminal 50 to operate the dragger 31 according to a trajectory suitable for the current distribution state of the slag 21.
[0032] In addition, in one embodiment, the trajectory determination unit 120 may output a control signal for the slag removal member based on the trajectory information. Specifically, when the trajectory information is acquired from the trajectory determination model 60, the trajectory determination unit 120 may generate a control signal for operating the dragger 31 of the slag remover 30 based on the acquired trajectory information, and transmit the control signal to the slag remover 30. The control signal allows the slag remover 30 to automatically operate the dragger 31 according to a trajectory suitable for the current distribution state of the slag 21.
[0033] The state acquisition unit 110 may acquire image data showing the surface of the molten metal after the slag removal operation is completed each time the slag removal member completes the slag removal operation. Specifically, when one slag removal operation by the dragger 31 as described above is completed, the state acquisition unit 110 may acquire image data showing the surface of the molten metal after the slag removal operation and pass the acquired image data to the trajectory determination unit 120. When the image data after the slag removal operation is acquired, the trajectory determination unit 120 may use the trajectory determination model 60 to determine the next trajectory information of the dragger 31 suitable for the distribution state of the slag 21 floating on the surface of the molten metal shown in the acquired image data. The acquisition of the image data does not necessarily need to be performed in response to the completion of one slag removal operation by the dragger 31. In response to the completion of a predetermined number of slag removal operations, the state acquisition unit 110 may acquire image data showing the surface of the molten metal after a predetermined number of slag removal operations and pass the acquired image data to the trajectory determination unit 120.
[0034] [Orbit determination model] As described above, the trajectory determination model 60 may be realized by any machine learning model that receives image data showing the surface of the molten metal in the vessel 20 and outputs trajectory information showing the trajectory of the dragger 31 as a slag removal member for removing the slag 21 on the molten metal. Specifically, the trajectory determination model 60 may be trained to generate trajectory information based on the distribution state of the slag 21 floating on the molten metal in the vessel 20. For example, the trajectory determination model 60 may be realized by a neural network such as a convolutional neural network, or any other type of machine learning model.
[0035] For example, as shown in FIG. 5, the trajectory determination model 60 may be trained by the training device 70 using training data stored in a training data database (DB) 80, and the trained model may be used as the trajectory determination model 60. When the trajectory determination model 60 receives image data showing the surface of the molten metal in the container 20 as an input and generates trajectory information showing the trajectory of the dragger 31 as an output, each training data may be composed of a pair of image data showing the surface of the molten metal and correct trajectory information showing the correct trajectory of the dragger 31. For example, when the trajectory determination model 60 is realized by a neural network, the training device 70 may input training image data to an input layer of the neural network to be trained, and update parameters of the neural network to be trained according to a comparison result between the trajectory information acquired from the output layer and the corresponding correct trajectory information.
[0036] Such correct trajectory information may be generated based on operation data of the dragger 31 by a skilled operator. For example, the operation data may be acquired as three-dimensional time-series data as shown in FIG. 6. The time-series data shown in FIG. 6A indicates the movement of the dragger 31 in the extension direction (Y direction). That is, the operator extends the dragger 31 at time t1, and instructs the operation terminal 50 to stop the extension movement at time t2. Then, the operator returns the dragger 31 at time t4, and instructs the operation terminal 50 to stop the return movement at time t5. The time-series data shown in FIG. 6B indicates the rotation movement (X direction) of the dragger 31. That is, the operator rotates the dragger 31 counterclockwise at time t1, and instructs the operation terminal 50 to stop the rotation movement at time t3. Then, the operator rotates the dragger 31 clockwise at time t4, and instructs the operation terminal 50 to stop the rotation movement at time t6. Also, the time series data shown in FIG. 6C instructs the operation terminal 50 to lower the dragger 31 in the vertical direction (Z direction) at time t3 and to stop the lowering operation at time t4. Then, the operator instructs the operation terminal 50 to raise the dragger 31 at time t5 and to stop the raising operation at time t6. The dragger 31 is controlled so as to only descend to the height of the dragger 31 set by the operation parameters. That is, even if the operator operates the dragger 31 to descend, the dragger 31 is set so as to only descend to the height indicated by the operation parameters. The training image data may be associated with operation data indicating the two-dimensional operation of the dragger 31 as shown in FIGS. 6A and 6B, or may be further associated with operation data indicating the vertical operation of the dragger 31 as shown in FIG. 6C.
[0037] In one embodiment, the orbit determination model 60 may receive luminance distribution data indicating the luminance distribution in the image data. The temperature of the molten metal is significantly higher than the temperature of the slag 21, and a significant luminance difference may occur between the molten metal and the slag 21 in the image data acquired by imaging the surface of the molten metal. For this reason, the orbit determination model 60 may be trained by training data consisting of luminance distribution data indicating the luminance distribution in the image data and correct trajectory information indicating the correct trajectory of the dragger 31. In this case, the orbit determination unit 120 may derive luminance distribution data from the image data as a preprocessing step and input the derived luminance distribution data to the orbit determination model 60 in order to use the orbit determination model 60 trained in this way. In this way, by using the luminance distribution data, the orbit determination model 60 may be able to better identify the distribution state of the slag 21 and better determine the trajectory of the dragger 31.
[0038] In one embodiment, the trajectory determination unit 120 may extract an image data portion showing the internal region of the vessel 20 from the image data, and input the image data portion to the trajectory determination model 60. That is, the trajectory determination unit 120 may perform preprocessing to extract an image data portion showing the internal region inside the vessel 20 from the image data acquired by the state acquisition unit 110, and input the image data portion showing the internal region including the surface of the molten metal to the trajectory determination model 60. Although the distribution of the molten metal and slag 21 is important in trajectory determination, foreign objects such as the inner wall of the vessel 20 and surrounding structures may be included in the image data depending on the installation conditions of the camera 40. If such foreign objects are included in the training data, the accuracy of the trajectory determination model 60 may decrease. Therefore, by training the trajectory determination model 60 as a model that inputs only data of the internal region of the pot, a more suitable trajectory may be output.
[0039] In one embodiment, the trajectory information may indicate one of a plurality of predetermined trajectory patterns of the slag removal member. That is, the trajectory information may indicate any of a plurality of predefined trajectory patterns of the dragger 31. In actual operation, a skilled operator may select a trajectory pattern suitable for the current distribution state of the slag 21 from a plurality of trajectory patterns and operate the dragger 31 according to the selected trajectory pattern. For example, the trajectory patterns may be classified into a central slag discharge, a left-turn slag discharge, and a right-turn slag discharge. The central slag discharge is a trajectory pattern in which the dragger 31 arranged in the central region of the container 20 is extended and retracted to scrape out the slag 21, as shown in FIG. 7A. The left-turn slag discharge is a trajectory pattern in which the dragger 31 arranged in the left region of the container 20 is extended and retracted to scrape out the slag 21, as shown in FIG. 7B. Also, the right-turn slag removal is a trajectory pattern in which the dragger 31 arranged in the right region of the container 20 is extended and retracted to scrape out the slag 21, as shown in FIG. 7C. Furthermore, each trajectory pattern may also indicate the timing of raising and lowering the dragger 31. In this case, the trajectory determination model 60 may be trained to receive image data or brightness distribution data as input and output any one of the three trajectory patterns. The trajectory determination unit 120 may use the trajectory determination model 60 trained in this way to determine the trajectory pattern of the dragger 31 from the image data or brightness distribution data.
[0040] The classification of the trajectory patterns as training data may be performed by an operator or the like, but the classification of the trajectory patterns according to the present disclosure is not limited to this and may be performed by clustering the operation data. For example, the correct trajectory patterns as training data may be clustered by any known unsupervised learning such as the k-means method. For example, a large amount of operation data as shown in FIG. 6 may be collected, and the collected operation data may be clustered into k patterns. The average data of the operation data of each cluster may be derived, and the derived average data may be determined as the trajectory pattern of the cluster.
[0041] In one embodiment, the trajectory determination model 60 may output trajectory information indicating a plurality of trajectories for one image data. That is, the trajectory determination model 60 may be trained to output trajectory information indicating two or more trajectories for one input image data. For example, when image data immediately after scraping out the slag 21 in the center of the container 20 is input to the trajectory determination model 60, the trajectory determination model 60 may suit both the left and right sides of the container 20 as the next scraping operation. In such a case, depending on the operator, the left side may be scraped and the right side may be scraped. Therefore, by training the trajectory determination model 60 using both the left and right sides as correct answer data, trajectory information that reduces the adverse effects of bias in the operator's operation data may be generated. When such a trajectory determination model 60 is used, it is necessary to finally determine one operation to be executed. For this reason, when using a trajectory determination model 60 that outputs two or more trajectories, one operation may be determined using a logic different from that of the trajectory determination model 60. For example, the orbit determination unit 120 may calculate the frequency of occurrence of each previously selected motion, and select a motion with a low frequency of occurrence from among multiple trajectory candidates in the orbit information output from the orbit determination model 60. This configuration has the effect of constructing a system that reduces bias in the output trajectories. The bias in the trajectory here means, for example, a tendency to select the left side more than the right side.
[0042] Alternatively, the operation data may be a two-dimensional or three-dimensional trajectory of the scraper 33 at the tip of the dragger 31 as shown in Fig. 8, and trajectory data indicating shape features of the two-dimensional or three-dimensional trajectory may be clustered by any known unsupervised learning. For example, the shape features may be the length, width, angle, maximum and minimum coordinates on the X-axis, maximum and minimum coordinates on the Y-axis, etc. of the trajectory shape of the scraper 33. Average data of the operation data of each cluster may be derived, and the derived average data may be determined as the trajectory pattern of the cluster.
[0043] In one embodiment, the trajectory information may indicate a divided section of the surface of the molten metal. For example, the surface of the molten metal in the vessel 20 may be divided into a plurality of sections, and the trajectory information may indicate any one of these sections. In the example shown in FIG. 9, the surface area of the molten metal is divided into nine sections, A1, A2, . . . , A9. Note that A10 indicates the position of the dragger 31 after the slag removal operation, that is, the scraping position of the slag 21 outside the furnace. In this case, the trajectory determination model 60 may be trained to receive image data or brightness distribution data as an input and output any one of the nine sections. The trajectory determination unit 120 may use the trajectory determination model 60 trained in this way to determine the movement position of the dragger 31 corresponding to the section (e.g., the section to which the dragger 31 should move) from the image data or brightness distribution data.
[0044] In one embodiment, the trajectory information may indicate a series of trajectories of the slag removal member. Specifically, the trajectory information may indicate each trajectory of the dragger 31 for multiple slag removal operations. In the slag removal operation by a skilled operator, the operator may check the distribution state of the slag 21 floating on the surface of the molten metal, determine a series of trajectory patterns for multiple slag removal operations, and perform each slag removal operation according to the determined series of trajectory patterns. For example, the series of trajectory patterns may indicate a trajectory pattern of 10 slag removal operations of central slag removal → central slag removal → central slag removal → left turn slag removal → left turn slag removal → right turn slag removal → right turn slag removal → central slag removal. Alternatively, the series of trajectory patterns may indicate a section of 10 slag removal operations of A2 → A1 → A3 → A5 → A4 → A6 → A8 → A7 → A9 → A10. In this case, the orbit determination model 60 may be trained to receive image data or brightness distribution data as input and output any orbit pattern or sequence of segments. The orbit determination unit 120 may use the orbit determination model 60 trained in this way to determine an orbit pattern or sequence of segments from the image data or brightness distribution data.
[0045] In one embodiment, the state acquisition unit 110 may measure the amount of slag removed by the slag removal member, and the trajectory determination unit 120 may generate trajectory information for stirring the molten metal based on the measured amount of slag removed. Specifically, the state acquisition unit 110 may use a camera and / or a sensor to measure the amount of slag removed by the dragger 31 and notify the trajectory determination unit 120 of the measured amount of slag removed. When the amount of slag remaining in the vessel 20 decreases, the amount of slag scraped out in one slag removal operation may decrease. In this case, a skilled operator may often stir the molten metal in the vessel 20 to make it easier to collect the slag 21. When such a trajectory pattern for stirring is set and the amount of slag removed falls below a predetermined threshold, the trajectory determination unit 120 may select the trajectory pattern for stirring. When the stirring of the molten metal is completed, the status acquisition unit 110 may acquire image data showing the surface of the molten metal in the container 20, and perform a slag removal operation on the slag 21 floating on the surface of the molten metal after stirring.
[0046] [Training device] Next, a training device 70 according to an embodiment of the present disclosure will be described with reference to Fig. 10. The training device 70 can be realized as a computing device such as a server, and can be realized by a hardware configuration similar to that of the orbit determination device 100 shown in Fig. 3, without being limited thereto. Fig. 10 is a block diagram showing a functional configuration of the training device 70 according to an embodiment of the present disclosure. As shown in Fig. 10, the training device 70 has a data acquisition unit 71 and a training unit 72.
[0047] The data acquisition unit 71 acquires training data for training the trajectory determination model 60. The training data may be composed of, for example, image data showing the surface of the molten metal in the vessel 20 and correct trajectory information showing the correct trajectory of the slag removal member for scraping out the slag 21 on the molten metal. The training data may be preferably generated based on operation data of the slag remover 30 by a skilled operator, and may be composed of a trajectory of the dragger 31 for effectively scraping out the floating slag from the vessel 20 based on the surface area of the molten metal in the image data.
[0048] The training unit 72 uses the training data to train the orbit determination model 60. For example, when the orbit determination model 60 is realized by a neural network, the training unit 72 inputs image data of the training data to the orbit determination model 60 to be trained, and acquires an output result from the orbit determination model 60 to be trained. The training unit 72 then compares the acquired output result with the correct orbit information of the training data, and adjusts the parameters of the orbit determination model 60 according to the backpropagation method based on the error between the output result and the correct orbit information. The finally acquired orbit determination model 60 is provided to the orbit determination device 100.
[0049] The trajectory determination model 60 obtained in this manner can, when it acquires image data showing the surface of the molten metal in the container 20, determine trajectory information showing the trajectory of a slag removal member for scraping off the slag 21 on the molten metal based on the image data.
[0050] [Orbit determination process] Next, an orbit determination process according to an embodiment of the present disclosure will be described with reference to Fig. 11. The orbit determination process is executed by the above-mentioned orbit determination device 100, and more specifically, may be realized by one or more processors 102 of the orbit determination device 100 executing one or more programs or instructions stored in one or more storage devices 101. Fig. 11 is a flowchart showing the orbit determination process according to an embodiment of the present disclosure.
[0051] 11, in step S101, the orbit determination device 100 acquires image data of the surface of the molten metal in the vessel 20. Specifically, the orbit determination device 100 acquires image data showing the distribution state of the slag 21 floating on the surface of the molten metal in the vessel 20 from the camera 40 arranged above the vessel 20.
[0052] In step S102, the orbit determination device 100 generates orbit information of the slag removal member using the orbit determination model 60. Specifically, the orbit determination device 100 inputs the acquired image data and / or brightness distribution data derived from the image data to the orbit determination model 60, and acquires orbit information indicating the orbit of the dragger 31 from the orbit determination model 60. The orbit determination model 60 may be trained to generate orbit information based on the distribution state of the slag 21 floating on the molten metal in the container 20.
[0053] For example, the trajectory determination model 60 may be trained to output a trajectory pattern of the dragger 31 (e.g., central slag removal, left turn slag removal, right turn slag removal, etc.) when image data and / or brightness distribution data is received as input. The trajectory determination model 60 may be trained to output a section (e.g., sections A1 to A9 in FIG. 9 ) through which the dragger 31 should move when image data and / or brightness distribution data is received as input. The trajectory determination model 60 may be trained to output a trajectory pattern or a series of sections of the dragger 31 for multiple slag removal operations when image data and / or brightness distribution data is received as input. The trajectory information may indicate the timing of descent and / or ascent of the dragger 31 along with the trajectory pattern and / or section.
[0054] In step S103, the orbit determination device 100 may present the orbit information as guidance information for operating the dragger 31. Specifically, the orbit determination device 100 may generate guidance information for guiding the operator based on the acquired orbit information, and transmit the guidance information to the operation terminal 50. Upon acquiring the guidance information, the operation terminal 50 may display the guidance information on a display to guide the operation of the dragger 31 by the operator, and / or may notify the guidance information via a speaker.
[0055] Alternatively, the orbit determination device 100 may output a control signal for operating the dragger 31 to the desluicing machine 30. Specifically, the orbit determination device 100 may generate a control signal for operating the dragger 31 and transmit the control signal to the desluicing machine 30. Upon receiving the control signal, the desluicing machine 30 operates the dragger 31 according to the control signal. Here, the control signal may be represented by a three-dimensional timing chart that specifies the on / off timing of the extension operation, rotation operation, and lifting operation of the dragger 31 related to the slag removal operation of the dragger 31, for example, as shown in Figs. 6A to 6C.
[0056] In step S104, the orbit determination device 100 determines whether to end the slag removal operation. For example, the orbit determination device 100 may determine whether the slag removal operation has been completed a predetermined number of times, and may end the orbit determination process when the slag removal operation has been completed a predetermined number of times. Alternatively, the orbit determination device 100 may end the orbit determination process when a predetermined time has elapsed since the start of the slag removal operation. Alternatively, the orbit determination device 100 may end the orbit determination process when the amount of slag remaining in the container 20 and / or the amount of slag most recently removed becomes equal to or less than a predetermined threshold. On the other hand, if the slag removal operation has not been completed a predetermined number of times, if the predetermined time has not elapsed since the start of the slag removal operation, or if the amount of slag remaining in the container 20 and / or the amount of slag most recently removed does not become equal to or less than a predetermined threshold, the orbit determination device 100 returns to step S101 and continues the slag removal operation.
[0057] According to the above-described embodiment, it is possible to obtain trajectory information of the dragger 31 suitable for the distribution state of the slag 21 floating on the molten metal by using the trajectory determination model 60 trained based on the slag removal operation of the dragger 31 by a skilled operator. This allows even an operator with a low level of skill to operate the dragger 31 according to the trajectory information output from the trajectory determination model 60, and also allows the operation of the dragger 31 to be automatically controlled according to the trajectory information output from the trajectory determination model 60.
[0058] Although the embodiments of the present disclosure have been described in detail above, the present disclosure is not limited to the specific embodiments described above, and various modifications and variations are possible within the scope of the gist of the present disclosure described in the claims. [Explanation of symbols]
[0059] 10. Slag removal system 20 containers 21 Slug 30 Slag remover 31 Dragger 32 Container tilting part 33 Scraping Board 40 Camera 50 Operation terminal 60 Orbit Determination Model 70 Training equipment 80 Training Data DB 100 Orbit determination device 110 Status acquisition unit 120 Orbit determination part
Claims
1. a state acquisition unit that acquires image data showing a surface of the molten metal in the vessel; a trajectory determination unit that uses a trajectory determination model to generate trajectory information indicating a trajectory of a slag removal member for scraping off slag on the molten metal based on the image data; An orbit determination device having an
2. The trajectory determination device according to claim 1 , wherein the trajectory determination unit generates guidance information for operating the slag removal member based on the trajectory information and presents the guidance information to an operator.
3. The orbit determination device according to claim 1 , wherein the orbit determination unit outputs a control signal to the slag removal member based on the orbit information.
4. The trajectory determination device according to claim 1 , wherein the trajectory information indicates one of a plurality of predetermined trajectory patterns for the slag removal member.
5. The trajectory determination apparatus of claim 1 , wherein the trajectory information indicates two-dimensional movement of the slag removal member relative to a surface of the molten metal.
6. The trajectory determination apparatus of claim 5 , wherein the trajectory information further indicates vertical movement of the slag removal member relative to a surface of the molten metal.
7. The orbit determination apparatus of claim 1 , wherein the orbit information indicates divided sections of a surface of the molten metal.
8. The trajectory determination apparatus of claim 1 , wherein the trajectory information indicates a sequence of trajectories of the slag removal member.
9. The state acquisition unit measures an amount of slag removed by the slag removal member, The orbit determination device according to claim 1 , wherein the orbit determination unit generates orbit information for stirring the molten metal based on the measured amount of slag removed.
10. The orbit determination apparatus of claim 1 , wherein the orbit determination model is trained to generate the orbit information based on a distribution of slag floating above the molten metal in the vessel.
11. The trajectory determination device according to claim 10, wherein the trajectory determination model is trained using training data consisting of brightness distribution data indicating a brightness distribution in image data showing a surface of the molten metal in a vessel and correct trajectory information indicating a correct trajectory of the slag removal member.
12. The trajectory determination device according to claim 11 , wherein the correct trajectory information indicates a correct trajectory pattern clustered by unsupervised learning.
13. The trajectory determination device according to claim 1 , wherein the state acquisition unit acquires image data showing a surface of the molten metal after completion of a slag removal operation each time the slag removal member completes a slag removal operation.
14. 2. The trajectory determination device according to claim 1, wherein the slag removal member is a scraping-type dragger having a scraping plate at a tip for scraping out the slag.
15. The orbit determination device according to claim 1 , wherein the orbit determination unit extracts an image data portion showing an internal region of the container from the image data, and inputs the image data portion to the orbit determination model.
16. The orbit determination device according to claim 1 , wherein the orbit determination model outputs orbit information indicating a plurality of orbits for one image data.
17. acquiring image data indicative of a surface of the molten metal within the vessel; generating trajectory information indicating a trajectory of a slag removal member for scraping off slag on the molten metal based on the image data using a trajectory determination model; 2. A computer implemented orbit determination method comprising:
18. acquiring image data indicative of a surface of the molten metal within the vessel; generating trajectory information indicating a trajectory of a slag removal member for scraping off slag on the molten metal based on the image data using a trajectory determination model; A program that causes a computer to execute the following.
19. A data acquisition unit that acquires training data including image data showing a surface of the molten metal in the container and correct trajectory information showing a correct trajectory of a slag removal member for scraping out slag on the molten metal; a training unit that uses the training data to train an orbit determination model; A training device having the above configuration.
20. acquiring image data indicative of a surface of the molten metal within the vessel; determining trajectory information indicating a trajectory of a slag removal member for scraping off slag on the molten metal based on the image data; 1. An orbit determination model, comprising: A trajectory determination model that is trained using training data consisting of image data showing the surface of the molten metal in the vessel and correct trajectory information showing the correct trajectory of a slag removal member for scraping off slag on the molten metal.
Citation Information
Patent Citations
Slag removing apparatus of melting furnace
JP2018179348A
Melting operation device and melting operation method
JP2019105389A
Automatic slag removal device and automatic slag removal program
JP2020085395A
Cited By
Robot system, robot control method, and robot control program
WO2026163743A1
