Method for predicting impurity concentrations of molten iron, method for producing molten iron, device for predicting impurity concentrations of molten iron, and method for generating impurity concentration prediction model
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- JFE STEEL CORP
- Filing Date
- 2025-12-19
- Publication Date
- 2026-08-06
Smart Images
Figure JP2025044473_06082026_PF_FP_ABST
Abstract
Description
Method for predicting impurity concentration of molten iron, method for producing molten iron, apparatus for predicting impurity concentration of molten iron, and method for generating impurity concentration prediction model
[0001] The present invention relates to a method for predicting the impurity concentration of molten iron, a method for producing molten iron, an apparatus for predicting the impurity concentration of molten iron, and a method for generating an impurity concentration prediction model.
[0002] In an electric arc furnace facility, a main raw material composed of iron-based scrap (cold iron source) etc. is melted by arc heat to produce molten iron. The types of iron-based scrap are defined by the unified inspection standard for iron-based scrap, and there are various types such as heavy scrap, new chips, shredders, etc., and the price is determined by the quality of the scrap. The most widely circulated in the market is heavy scrap (H1 to H4). Heavy scrap mixed with iron sources disassembled in the market has a risk of containing impurities such as copper (Cu) and tin (Sn) that are difficult to remove by melting and refining, so-called trump elements.
[0003] The price of iron-based scrap is determined taking into account such scrap quality. For example, new chips generated when punching steel plates have a lower risk of containing trump elements compared to heavy scrap, and due to the limited circulation volume, they are traded at a high price. Therefore, if a large amount of new chips is used to reduce the risk of containing trump elements, the production cost of molten iron will increase. Conventionally, the blending ratio of the iron-based scrap has been determined based on the past knowledge and situation judgment of the operator within a practically operable range so that the produced molten iron satisfies the target values of chemical components, and the blending ratio of heavy scrap and new chips has been determined. For this reason, there has been a problem that the blending ratio of heavy scrap and new chips is greatly affected by the proficiency of the operator.
[0004] In response to such problems, Patent Document 1 discloses a method for predicting the impurity concentration of molten iron by inputting the charging amount of iron-based scrap, the impurity concentration and the remaining amount of molten iron in the previous charge into an impurity concentration prediction model and outputting the impurity concentration of the molten iron in the subsequent charge. According to Patent Document 1, by inversely analyzing the impurity concentration prediction model used for predicting the impurity concentration of molten iron, it is possible to specify the blending of iron-based scrap in which the impurity concentration is within the target range.
[0005] Patent No. 7207624
[0006] In Japan, where high-grade steel is manufactured with strict control over trump elements such as copper, domestically generated low-grade scrap (heavy scrap) carries a high risk of containing trump elements and therefore cannot be 100% recycled. As a result, 8 to 9 million tons of iron scrap are exported overseas annually.
[0007] Compared to molten iron produced by the blast furnace-converter method, molten iron produced by the electric furnace method consumes less energy and CO2. ２ The amount of waste generated is also low. From the perspective of environmental impact, it is desirable that 100% of the iron scrap generated domestically be recycled domestically. In response to this requirement, the method disclosed in Patent Document 1 can produce molten iron that satisfies the chemical composition standards to some extent, and can be used in the blending of various iron scraps, thereby contributing to the effective utilization of low-grade scrap domestically.
[0008] However, the method disclosed in Patent Document 1 for predicting impurity concentrations was inaccurate, and in actual molten iron production, the upper limit of the copper component standard, which is a trump element, was frequently exceeded. As a result, it became necessary to reduce the amount of heavy scrap used and increase the amount of high-grade scrap such as new scrap, which ultimately led to an increase in the cost of molten iron production.
[0009] This invention has been made in view of the problems of the prior art, and its objective is to provide a method for predicting the impurity concentration of molten iron and a device for predicting the impurity concentration of molten iron that can improve the accuracy of predicting the impurity concentration of molten iron that will be used as a trump element. Another objective of this invention is to provide a method for producing molten iron in which the impurity concentration is within a predetermined range, and a method for generating an impurity concentration prediction model used for predicting the impurity concentration of molten iron.
[0010] The means for solving the above problems are as follows: [1] A method for predicting the impurity concentration of molten iron produced by melting iron scrap in a smelting furnace, comprising: a measurement step of measuring the amount of iron scrap to be charged into the smelting furnace; an imaging step of imaging the iron scrap being transported to the smelting furnace to generate image data; and an impurity concentration prediction step of inputting input data including the amount of iron scrap charged and the image data into an impurity concentration prediction model to output the impurity concentration of molten iron. [2] The method for predicting the impurity concentration of molten iron according to [1], wherein the input data further includes the impurity concentration of molten iron from the previous charge and the amount of molten iron remaining in the charge. [3] A method for producing molten iron by melting iron scrap in a smelting furnace, comprising: a charging step of charging the iron scrap into the smelting furnace so that the impurity concentration of the molten iron predicted in the impurity concentration prediction step of the method for predicting the impurity concentration of molten iron described in [1] or [2] is within a predetermined range; and a melting step of melting the iron scrap charged into the smelting furnace. [4] The method for producing molten iron according to [3], wherein the iron scrap is charged into the smelting furnace in multiple batches, the impurity concentration is predicted each time the iron scrap is charged in the impurity concentration prediction step, and the iron scrap is charged in the charging step so that the impurity concentration of the molten iron, which is obtained by accumulating the impurity concentrations predicted each time the iron scrap is charged, is within a predetermined range. [5] A device for predicting the impurity concentration of molten iron produced by melting iron scrap in a smelting furnace, comprising: a conveying device for conveying the iron scrap to the smelting furnace; a measuring device for measuring the amount of iron scrap to be charged into the smelting furnace; an imaging device for imaging the iron scrap during conveyance and generating image data; and a calculation device for predicting the impurity concentration of the molten iron, wherein the calculation device has an impurity concentration prediction unit that inputs input data including image data acquired from the imaging device and the amount of iron scrap acquired from the measuring device into an impurity concentration prediction model to output the impurity concentration of the molten iron.[6] The molten iron impurity concentration prediction device according to [5], wherein the input data further includes the impurity concentration of the molten iron from the previous charge and the amount of molten iron remaining from the charge. [7] The molten iron impurity concentration prediction device according to [5] or [6], wherein the calculation device has a charging control unit that controls the charging of the iron-based scrap into the smelting furnace so that the impurity concentration of the molten iron predicted by the impurity concentration prediction unit is within a predetermined range. [8] The molten iron impurity concentration prediction device according to [7], wherein the iron-based scrap is charged into the smelting furnace in multiple batches, the impurity concentration prediction unit predicts the impurity concentration each time the iron-based scrap is charged, and the charging control unit controls the charging of the iron-based scrap so that the impurity concentration of the molten iron, which is obtained by accumulating the impurity concentrations predicted each time the iron-based scrap is charged, is within a predetermined range. [9] A method for generating an impurity concentration prediction model used to predict the impurity concentration of molten iron produced by melting iron scrap in a smelting furnace, the method comprising: training a machine learning model with multiple datasets, each set consisting of image data of the iron scrap in past molten iron production, actual values of the amount of iron scrap charged, and actual values of the impurity concentration of the molten iron, as training data; and generating an impurity concentration prediction model that takes input data including the image data and the amount charged as input and outputs the impurity concentration of molten iron.
[0011] By implementing the method for predicting the impurity concentration of molten iron according to the present invention, it becomes possible to predict with high accuracy the impurity concentration of molten iron produced by melting iron-based scrap. In this way, by controlling the amount of iron-based scrap charged using the highly accurate predicted impurity concentration of molten iron, it becomes possible to produce molten iron with an impurity concentration within a predetermined range using iron-based scrap with high accuracy.
[0012] Figure 1 is a schematic diagram showing an example configuration of a molten iron manufacturing facility including a device for predicting the impurity concentration of molten iron according to this embodiment. Figure 2 is a schematic diagram showing an example configuration of a calculation device. Figure 3 is a schematic diagram showing data stored in the storage unit. Figure 4 is a graph showing the relationship between the upper limit of the Cu concentration and the actual value of the Cu concentration (average value) of the manufactured molten iron. Figure 5 is a graph showing the error range (2σ) of the Cu concentration of the manufactured molten iron. Figure 6 is a graph showing the charge ratio when the iron-based scrap is ranked up. Figure 7 is a graph showing the charge ratio when the iron-based scrap is ranked down. Figure 8 is a graph showing the error range (2σ) of the Cu concentration of the molten iron manufactured in Invention Example 1 and Invention Example 2. Figure 9 is a graph showing the charge ratio when the iron-based scrap is ranked up. Figure 10 is a graph showing the charge ratio when the iron-based scrap is ranked down.
[0013] The present invention will be described in detail below through embodiments of the present invention. The following embodiments are preferred examples of the present invention, and the present invention is not limited in any way by these embodiments. Figure 1 is a schematic diagram showing an example of the configuration of a molten iron manufacturing facility 100 including a molten iron impurity concentration prediction device 12 according to this embodiment. The molten iron manufacturing facility 100 according to this embodiment includes a molten iron impurity concentration prediction device 12 and an arc electric furnace 50.
[0014] In the molten iron production facility 100 according to this embodiment, iron-based scrap 40, which is ranked according to its impurity concentration, is stored in the scrap yard 10. The iron-based scrap 40 stored in the scrap yard 10 is ranked, for example, from A to N ranks, with A-rank iron-based scrap having the highest impurity concentration and N-rank iron-based scrap having the lowest impurity concentration. On the other hand, the cost of iron-based scrap is inversely proportional to the impurity concentration. Therefore, A-rank iron-based scrap is the cheapest, and N-rank iron-based scrap is the most expensive.
[0015] In the production of molten iron using iron-based scrap 40, impurities are trump elements that are difficult to remove by refining in the arc electric furnace 50, such as copper (Cu), tin (Sn), nickel (Ni), chromium (Cr), molybdenum (Mo), cobalt (Co), or bismuth (Bi). The embodiments and examples will be described below using an example where the impurity of the produced molten iron is Cu.
[0016] The molten iron impurity concentration prediction device 12 includes a lifting magnet 14, a crane scale 16, an imaging device 18, and a calculation device 20. The lifting magnet 14 lifts the iron-based scrap 40 by magnetic force and transports it to the arc electric furnace 50. In this embodiment, the lifting magnet 14 is an example of a transport device that transports the iron-based scrap 40 to the arc electric furnace 50.
[0017] The crane scale 16 is attached to the lifting magnet 14 and measures the amount of ferrous scrap 40 to be charged into the arc electric furnace 50. Measuring the amount of ferrous scrap 40 charged using this crane scale 16 is a measurement step in the method for predicting the impurity concentration of molten iron. The crane scale 16 is an example of a measuring device for measuring the amount of ferrous scrap charged into the arc electric furnace 50. The crane scale 16 outputs the measured amount of ferrous scrap 40 charged to the calculation device 20.
[0018] The imaging device 18 is installed in the transport path of the iron-based scrap 40 carried by the lifting magnet 14. The imaging device 18 is a digital camera equipped with a color image sensor (CCD or CMOS). The imaging device 18 captures images of the iron-based scrap 40 being transported to the arc electric furnace 50 by the lifting magnet 14 and generates image data. This process is the imaging step in the method for predicting the impurity concentration of molten iron. The imaging device 18 outputs the generated image data to the arithmetic unit 20.
[0019] The imaging device 18 may continuously capture images from multiple points in time or from multiple angles during the transport of the iron-based scrap 40 to generate multiple image data. In this case, the imaging device 18 outputs the continuously captured and generated image data as time-series data to the processing unit 20. This makes it possible to obtain image data showing the time-series changes in the distribution of foreign matter, which can be used for feature extraction and impurity concentration prediction described later.
[0020] The arc electric furnace 50 melts the iron-based scrap 40 charged into the furnace by heating with an arc generated between the electrodes and the iron-based scrap 40. Molten iron is produced using the iron-based scrap 40. In this embodiment, the arc electric furnace 50 is an example of a smelting furnace. When using other smelting furnaces, molten iron may be produced by melting the iron-based scrap 40 by heating through an oxidation reaction between oxygen supplied from an oxygen blowing lance and carbon supplied from a carbon blowing lance.
[0021] In the molten iron manufacturing method of this embodiment, a hot-heel operation is performed in which a portion of the molten iron produced in the previous molten iron manufacturing (hereinafter, "molten iron manufacturing" may be referred to as "charge"), which was melted in the arc electric furnace 50, is left in the furnace and carried over to the next charge. By leaving a portion of the molten iron from the previous charge in the furnace and carrying it over, the melting efficiency of the iron-based scrap 40 charged into the next charge can be increased.
[0022] The computing unit 20 is connected to the crane scale 16 and the imaging device 18 via wired or wireless communication. The computing unit 20 obtains the amount of ferrous scrap to be charged into the arc electric furnace 50 from the crane scale 16. The computing unit 20 obtains image data of the ferrous scrap to be charged into the arc electric furnace 50 from the imaging device 18.
[0023] The calculation unit 20 is also connected to a process computer (not shown) that controls the operation of the arc electric furnace 50, enabling communication between the two. The calculation unit 20 obtains the operating conditions for the arc electric furnace 50 from the process computer. The operating conditions for the arc electric furnace 50 include the target range of Cu concentration of the molten iron to be produced, the amount of molten iron remaining from the previous charge, and the Cu concentration of the molten iron from the previous charge.
[0024] The calculation device 20 acquires image data of the iron scrap, the amount of iron scrap charged, the amount of molten iron remaining from the previous charge, and the Cu concentration of the molten iron from the previous charge. It then inputs this data, including the image data, into an impurity concentration prediction model to output the Cu concentration of the molten iron. This process is the impurity concentration prediction step in the method for predicting the impurity concentration of molten iron. In this way, the calculation device 20 predicts the Cu concentration of the molten iron produced by melting the iron scrap 40.
[0025] The impurity concentration prediction model used to predict the impurity concentration of molten iron is a pre-trained machine learning model that has been trained using multiple datasets, each dataset consisting of actual values of the input data and actual values of the Cu concentration of the molten iron. The impurity concentration prediction model is generated in advance and stored in the computing device 20.
[0026] The calculation device 20 preferably controls the charging of iron-based scrap 40 into the arc electric furnace 50 so that the Cu concentration of the molten iron produced is within a target range. The target range is an example of a predetermined range. The calculation device 20 predicts the Cu concentration of the molten iron using an impurity concentration prediction model and controls the charging of iron-based scrap 40 so that the predicted Cu concentration of the molten iron is within the target range. This process is the charging step in the method for producing molten iron.
[0027] The charging of iron-based scrap 40 into the arc electric furnace 50 may be carried out in multiple stages. The number of times the iron-based scrap 40 is charged is predetermined based on the amount of iron-based scrap transported by the lifting magnet 14 and the target amount of molten iron produced in the arc electric furnace 50.
[0028] When iron scrap is charged in multiple stages, the calculation device 20 predicts the Cu concentration for each stage of iron scrap 40 charging and calculates the Cu concentration of the molten iron by accumulating the predicted Cu concentrations. The calculation device 20 controls the charging of iron scrap 40 so that the Cu concentration of the molten iron calculated in this way falls within the target range. This process is also a charging step in the molten iron manufacturing method.
[0029] After the iron-based scrap 40 is charged into the arc electric furnace 50, the iron-based scrap 40 charged into the furnace is melted by heating caused by an arc generated between the electrodes and the iron-based scrap 40, thereby producing molten iron. This process is the melting step in the method for producing molten iron. In this way, molten iron with a Cu concentration within the target range is produced by melting the iron-based scrap 40 using the arc electric furnace 50.
[0030] Next, the arithmetic unit 20 will be described. Figure 2 is a schematic diagram showing an example configuration of the arithmetic unit 20. The arithmetic unit 20 is a general-purpose computer such as a workstation or personal computer. The arithmetic unit 20 has a control unit 22, an input unit 24, an output unit 26, a storage unit 28, and a communication unit 30. The control unit 22 is, for example, a CPU, and by executing a program stored in the storage unit 28, it functions as an impurity concentration prediction unit 32, a charging control unit 34, and an impurity concentration prediction model generation unit 36.
[0031] The input unit 24 is, for example, a keyboard, a touch panel integrated with a display, etc. The output unit 26 is, for example, an LCD or CRT display, etc. The storage unit 28 is, for example, an information recording medium such as a flash memory that can be updated and recorded, a hard disk that is built-in or connected via a data communication terminal, a memory card, etc., and a device for reading and writing them. The storage unit 28 stores programs, data, calculation formulas, machine learning models before training, etc., for realizing each function of the arithmetic unit 20.
[0032] Figure 3 is a schematic diagram showing the data stored in the storage unit 28. As shown in Figure 3, the storage unit 28 stores programs for realizing each function of the computing device 20, as well as an impurity concentration prediction model 37 and a database 38. In this embodiment, the impurity concentration prediction model 37 is a trained machine learning model that takes input data including image data of iron scrap 40, the amount of iron scrap 40 charged, the amount of molten iron remaining from the previous charge, and the Cu concentration of the molten iron from the previous charge as input, and outputs the Cu concentration of the molten iron.
[0033] In the impurity concentration prediction model 37, it is preferable to use a convolutional neural network method that compresses the image data of the iron-based scrap 40 into one-dimensional data by convolutional processing when inputting the image data into the impurity concentration prediction model 37. As the convolutional neural network method, known models such as GoogleNet, VGG16, MOBILENET, and EFFICIENTNET may be used.
[0034] When iron scrap 40 contains a large amount of steel bars and motors, the Cu concentration of iron scrap 40 tends to be high. When it contains a large amount of copper-colored components, the Cu concentration of iron scrap 40 tends to be high. When it contains a large amount of red, blue, or other plastic-colored components, the Cu concentration of iron scrap 40 tends to be high. Thus, the shape information corresponding to steel bars and motors in iron scrap 40, as well as the above-mentioned color information, correlates with the Cu concentration. Therefore, by including image data containing this information as input data for the impurity concentration prediction model 37, the accuracy of predicting the Cu concentration of molten iron can be improved.
[0035] In addition to image data, shape information and color information extracted from the image data may be input into the impurity concentration prediction model 37 to predict the Cu concentration of molten iron. As shape information, numerical values of the outline (perimeter), area, aspect ratio (long side / short side), and circularity ratio of the iron scrap 40 may be obtained from the image data by edge detection or contour detection. The outline of the iron scrap 40 can be detected by detecting the outline of the iron scrap by performing edge detection on the image data. The area of the iron scrap 40 can be detected from the number of pixels included in the outline of the iron scrap. The aspect ratio of the iron scrap 40 can be detected from the length and width of the outline. The circularity ratio of the iron scrap 40 can be calculated by dividing the area of the iron scrap 40 by its perimeter.
[0036] For steel bars, there is a correlation with the aspect ratio of the iron scrap 40, and for motors, there is a correlation with the circularity ratio of the iron scrap 40. Therefore, by including this shape information in the input data of the impurity concentration prediction model 37, the accuracy of predicting the Cu concentration of molten iron is improved.
[0037] As color information, for example, a histogram of the RGB values of the image data may be obtained. An RGB histogram is information obtained by aggregating the red, green, and blue luminance values of each pixel in the image data into 256 levels and creating a histogram (256 x 3D vector). Alternatively, RGB may be converted to HSV values and a histogram of the resulting HSV values may be obtained. Since HSV expresses the color space using "hue, saturation, and lightness," it is less affected by lighting conditions and is superior to RGB values in color classification. For this reason, using HSV values allows for the detection of a higher proportion of copper-based colors in the image data than using RGB values.
[0038] Thus, the color and / or shape information extracted from image data correlates with the content, distribution, particle size, and adhesion state of non-iron foreign substances, such as copper-colored components, motors, and red / blue plastics. By using this information in addition to image data, it becomes possible to predict impurity concentrations with greater accuracy.
[0039] Database 38 stores more than 100 sets of datasets that serve as training data for machine learning models. The datasets stored in database 38 are datasets for molten iron manufactured in the past, consisting of image data of iron scrap 40, actual values of the amount of iron scrap 40 charged, actual values of the amount of molten iron remaining in the previous charge, actual values of the Cu concentration of the molten iron in the previous charge, and actual values of the Cu concentration of the molten iron as one set. Preferably, the number of datasets stored in database 38 is 500 sets or more, and more preferably 1000 sets or more.
[0040] Refer again to Figure 2. The communication unit 30 includes at least one of a communication module for wired communication and a communication module for wireless communication. The computing unit 20 communicates with the crane scale 16, the imaging device 18, and the process computer via the communication unit 30.
[0041] Next, the processes executed by the impurity concentration prediction unit 32, the charging control unit 34, and the impurity concentration prediction model generation unit 36 will be described. The impurity concentration prediction unit 32 acquires the charging amount and image data of the ferrous scrap from the crane scale 16 and the imaging device 18 via the communication unit 30. The impurity concentration prediction unit 32 acquires the remaining amount of molten iron in the previous charge and the Cu concentration of the molten iron in the previous charge from the process computer.
[0042] When the impurity concentration prediction unit 32 acquires these input data, it reads out the impurity concentration prediction model 37 from the storage unit 28. The impurity concentration prediction unit 32 inputs the image data and charging amount of the ferrous scrap 40, the remaining amount of molten iron in the previous charge, and the Cu concentration into the impurity concentration prediction model 37 to output the Cu concentration of the molten iron.
[0043] In this way, the impurity concentration prediction unit 32 predicts the Cu concentration of the molten iron produced by melting the ferrous scrap 40. The impurity concentration prediction unit 32 outputs the predicted Cu concentration of the molten iron to the charging control unit 34. The impurity concentration prediction unit 32 may cause the output unit 26 to display the predicted Cu concentration of the molten iron.
[0044] As described above, in the molten iron impurity concentration prediction device 12 according to this embodiment, the Cu concentration of the molten iron is predicted using an impurity concentration prediction model that includes, as input data, the image data generated by imaging the ferrous scrap 40 during transportation. Thereby, the impurity concentration of the molten iron produced in a refining furnace such as the arc furnace facility 50 using the ferrous scrap 40 as a raw material can be predicted with high accuracy.
[0045] When acquiring image data as time-series data from the imaging device 18, the arithmetic unit 20 applies an object detection algorithm to the plurality of image data, and extracts foreign object candidates such as copper-colored members, motors, and plastic pieces. In the object detection algorithm, foreign object candidates are specified using rectangular regions or masks. The arithmetic unit 20 records the coordinates and sizes of the specified foreign object candidates. The arithmetic unit 20 grasps the time-series data of the position of the foreign object by tracking the coordinates and sizes of the foreign objects recorded in the image data, which is time-series data, and this may be included in the input data of the impurity concentration prediction model 37 as statistical quantities (average, variance, maximum value) or a time-series model. The time-series model is an integrated feature amount output by subjecting time-series data to time-series processing by LSTM.
[0046] The charging control unit 34 acquires the target range of the Cu concentration of the molten iron from the process computer. The target range of the Cu concentration of the molten iron to be produced may be acquired in advance and stored in the storage unit 28. In this case, the charging control unit 34 reads out the target range of the Cu concentration of the molten iron from the storage unit 28.
[0047] When the charging control unit 34 acquires the Cu concentration of the molten iron predicted by the impurity concentration prediction unit 32, it compares the predicted Cu concentration of the molten iron with the target range of the Cu concentration, and determines whether the predicted Cu concentration of the molten iron is within the target range. When the charging control unit 34 determines that the predicted Cu concentration of the molten iron is within the target range, it continues the conveyance of the iron scrap 40 and charges the iron scrap 40 into the arc furnace 50.
[0048] The arithmetic unit 20 may be provided with control logic for automatically executing scrap rank control, charging order, and cost optimization based on the image feature amount. Thereby, automation and optimization of the operation can be achieved, contributing to reduction of manufacturing costs and stabilization of quality.
[0049] On the other hand, if the charging control unit 34 determines that the predicted Cu concentration is outside the target range, it stops transporting the iron-based scrap 40 and returns the iron-based scrap 40 to the scrap yard 10. If the predicted Cu concentration exceeds the target range, the charging control unit 34 upgrades the iron-based scrap 40 and charges the higher-ranked iron-based scrap 40 with a lower Cu concentration into the arc electric furnace 50.
[0050] If the predicted Cu concentration falls below the target range, the charging control unit 34 downgrades the rank of the iron-based scrap 40 charged into the arc electric furnace 50, charging the arc electric furnace 50 with iron-based scrap 40 of a lower rank and with a higher Cu concentration. In this way, the charging control unit 34 controls the rank of the iron-based scrap 40 charged into the arc electric furnace 50 so that the Cu concentration of the molten iron is within the target range. This makes it possible to reduce the cost of the iron-based scrap 40 used while keeping the Cu concentration of the molten iron produced by melting the iron-based scrap 40 within the target range. As a result, it becomes possible to produce molten iron at a lower cost than before while satisfying the target value of the Cu concentration of the molten iron.
[0051] When iron-based scrap 40 is charged into the arc electric furnace 50 in multiple batches, the impurity concentration prediction unit 32 predicts the Cu concentration of the molten iron each time a predetermined number of iron-based scraps are charged. The charging control unit 34 then calculates the Cu concentration of the molten iron by integrating the Cu concentrations output from the impurity concentration prediction unit 32. The charging control unit 34 controls the charging of the iron-based scrap 40 into the arc electric furnace 50 so that the Cu concentration of the molten iron falls within the target range.
[0052] If the charging control unit 34 determines that the accumulated Cu concentration is within the target range, it continues transporting the iron scrap and charges the iron scrap 40 into the arc electric furnace 50. On the other hand, if it determines that the predicted Cu concentration is outside the target range, the charging control unit 34 stops transporting the iron scrap and returns the iron scrap to the scrap yard 10.
[0053] If the accumulated Cu concentration exceeds the target range, the iron scrap charged into the arc electric furnace 50 is upgraded to a higher-ranked iron scrap with a lower Cu concentration. If the accumulated Cu concentration of the molten iron falls below the target range, the iron scrap charged into the arc electric furnace 50 is downgraded to a lower-ranked iron scrap with a higher Cu concentration. This operation is carried out until a predetermined number of iron scraps have been charged. This reduces the cost of the iron scrap 40 used while keeping the Cu concentration of the molten iron within the target range, similar to charging the iron scrap 40 once. As a result, molten iron can be produced at a lower cost than before while satisfying the target value for Cu concentration of the molten iron.
[0054] Next, the impurity concentration prediction model generation unit 36 will be described. The impurity concentration prediction model generation unit 36 uses the dataset stored in the database 38 to generate an impurity concentration prediction model 37, which takes the following input data as input and outputs the Cu concentration of molten iron. The input data may include color features and / or shape features extracted from image data of iron scrap 40.
[0055] <Input Data> Image data of iron scrap Amount of iron scrap charged Amount of molten iron remaining from the previous charge Cu concentration of molten iron from the previous charge
[0056] The impurity concentration prediction model generation unit 36 uses the dataset stored in the database 38 as training data to train a machine learning model and generates a trained machine learning model. This trained machine learning model becomes the impurity concentration prediction model 37. LightGBM can be used as the machine learning model. Not limited to LightGBM, a multilayer perceptron or a convolutional neural network (CNN) may also be used as the machine learning model.
[0057] The impurity concentration prediction model generation unit 36 may perform machine learning by dividing the dataset stored in the database 38 into training data and test data. By dividing the dataset into training data and test data in this way, the impurity concentration prediction model generation unit 36 can learn weight coefficients using the training data. Furthermore, the impurity concentration prediction model generation unit 36 can generate the impurity concentration prediction model 37 while changing the structure of the machine learning model (number of hidden layers and number of nodes) so that the accuracy of Cu concentration in the test data is high. By generating the impurity concentration prediction model 37 by dividing the dataset into training data and test data in this way, the prediction accuracy of Cu concentration by the impurity concentration prediction model 37 can be improved.
[0058] Furthermore, it is preferable to obtain the actual values of the input data and the actual values of the Cu concentration of the molten iron from the process computer each time molten iron is produced in the arc electric furnace 50, and record them in the database 38. In this case, the database 38 may be updated with a certain upper limit on the number of data sets.
[0059] Using the updated dataset in this way, the impurity concentration prediction model generation unit 36 may update the impurity concentration prediction model 37 by machine learning, for example, every six months or every year. By updating the impurity concentration prediction model using a dataset containing the latest data, the latest status of the molten iron impurity concentration prediction device 12 can be reflected in the impurity concentration prediction model 37. As a result, the Cu concentration of molten iron can be predicted with even higher accuracy using the impurity concentration prediction model 37.
[0060] As described above, the molten iron impurity concentration prediction device 12 according to this embodiment predicts the Cu concentration of molten iron using an impurity concentration prediction model 37 that includes image data of the iron-based scrap 40 transported to the smelting furnace as input data. This makes it possible to predict the Cu concentration of molten iron with higher accuracy than conventional technology, which predicted the Cu concentration of molten iron without using image data of the iron-based scrap 40. In this way, by controlling the amount of iron-based scrap charged using the highly accurate predicted impurity concentration of molten iron, it becomes possible to manufacture molten iron with a Cu concentration within the target range with high accuracy using iron-based scrap.
[0061] The embodiments of the present invention are not limited to those described above and can be modified in various ways. In the above embodiments, an example was shown in which a hot-heel operation is performed in which a portion of the molten iron from the previous charge melted in the arc electric furnace 50 is left in the furnace and carried over to the next charge, but the invention is not limited to this. If a hot-heel operation is not performed, the input data of the impurity concentration prediction model 37 does not need to include the amount of molten metal remaining from the previous charge and the Cu concentration of the molten iron from the previous charge. Even in this case, by using an impurity concentration prediction model that includes image data of iron-based scrap 40 as input data, the Cu concentration of the molten iron can be predicted with high accuracy.
[0062] In the above embodiment, an example was shown in which the calculation device 20 has an impurity concentration prediction unit 32, a charging control unit 34, and an impurity concentration prediction model generation unit 36, but it is not limited to this. For example, if an impurity concentration prediction model is generated by an external calculation device and stored in the storage unit 28 via the communication unit 30, the calculation device 20 does not need to have an impurity concentration prediction model generation unit 36. Furthermore, if the calculation device 20 only needs to predict the Cu concentration of molten iron, the calculation device 20 does not need to have a charging control unit 34.
[0063] In the above embodiment, an example was shown in which the charging of iron-based scrap 40 is controlled by determining whether the predicted Cu concentration of the molten iron is within the target range each time the iron-based scrap 40 is charged in multiple batches, but the invention is not limited to this. When the iron-based scrap 40 is charged in multiple batches, the Cu concentration is predicted each time, but the control of the charging of iron-based scrap does not necessarily have to be performed each time. That is, when the iron-based scrap 40 is charged in multiple batches, the control of the charging of iron-based scrap may be performed every other time, or it may be performed only for the last batch of iron-based scrap 40 to be charged. This makes it possible to charge the iron-based scrap 40 into the arc electric furnace 50 quickly.
[0064] (Example 1) Next, Example 1, in which molten iron was produced using an arc electric furnace and hot-heel operation, will be described. In Example 1, the Cu concentration of the molten iron was predicted using the impurity concentration prediction method of molten iron in Comparative Example 1, Comparative Example 2, and the Inventive Example, and molten iron was produced by controlling the rank of the iron-based scrap charged so that the predicted Cu concentration was within the target range. In Comparative Examples 1 and 2 and the Inventive Example, approximately 12 tons of iron-based scrap were charged into the arc electric furnace 12 times in one charge. The Cu concentration prediction methods for Comparative Example 1, Comparative Example 2, and the Inventive Example are as follows.
[0065] Comparative Example 1: The Cu concentration of molten iron was predicted using a regression calculation of the amount of iron scrap of each rank charged × the Cu coefficient of the iron scrap of each rank. The Cu coefficient was calculated using the actual Cu concentration of molten iron produced in the past and the above regression equation. Comparative Example 2: The Cu concentration of molten iron was predicted using an impurity concentration prediction model that takes the amount of iron scrap charged, the amount of molten iron remaining from the previous charge, and the Cu concentration of molten iron from the previous charge as input data, and outputs the Cu concentration of molten iron. Example of Invention: The Cu concentration of molten iron was predicted using an impurity concentration prediction model that takes image data of iron scrap, the amount of iron scrap charged, the amount of molten iron remaining from the previous charge, and the Cu concentration of molten iron from the previous charge as input data, and outputs the Cu concentration of molten iron.
[0066] Figure 4 is a graph showing the relationship between the upper limit of the Cu concentration and the actual value of the Cu concentration (average value) of the molten iron produced. The horizontal axis of Figure 4 represents the upper limit of the Cu concentration of the molten iron (target value: mass%), and the vertical axis represents the average value of the Cu concentration of the molten iron (actual value: mass%). As shown in Figure 4, it can be seen that the Cu concentration of the molten iron produced in the invention example is controlled to a range that does not exceed the upper limit of the target Cu concentration, and is not too low.
[0067] On the other hand, the Cu concentration of the molten iron produced in Comparative Example 1 sometimes matched the upper limit of the target Cu concentration, and sometimes the Cu concentration was too low, resulting in large fluctuations in the Cu concentration of the molten iron and instability. In Comparative Example 2, although the fluctuation in Cu concentration was suppressed more than in Comparative Example 1, the fluctuation in the Cu concentration of the molten iron was larger than in the inventive example. Overall, the Cu concentration of the molten iron in Comparative Example 2 was lower than in the inventive example.
[0068] These results confirm that the inventive example can control the Cu concentration of molten iron to the target range with higher accuracy than Comparative Examples 1 and 2. Thus, in order to control the Cu concentration of molten iron to the target range, it is necessary to predict the Cu concentration of the molten iron produced by melting iron-based scrap with high accuracy. Therefore, from the above results, it is considered that the inventive example can predict the Cu concentration with higher accuracy than Comparative Examples 1 and 2, and thereby was able to control the Cu concentration of molten iron to the target range with high accuracy.
[0069] Figure 5 is a graph showing the error range (2σ) of the Cu concentration of the manufactured molten iron. The horizontal axis of Figure 5 represents the upper limit of the Cu concentration (target value: mass%), and the vertical axis represents the error range (2σ) of the average Cu concentration. As shown in Figure 5, the error range of the Cu concentration of the manufactured molten iron was smallest for the inventive example at all Cu concentrations. From this result, it was confirmed that the inventive example can control the Cu concentration within the target range with smaller fluctuations than Comparative Examples 1 and 2. Thus, in order to control the Cu concentration of molten iron within the target range with small fluctuations, it is necessary to predict the Cu concentration of the molten iron manufactured by melting iron-based scrap with high accuracy. Therefore, from the above results, it is considered that the inventive example was able to predict the Cu concentration with higher accuracy than Comparative Examples 1 and 2, and thereby was able to control the Cu concentration of the molten iron within the target range with high accuracy.
[0070] Figure 6 is a graph showing the charge ratio when ferrous scrap is upgraded. The vertical axis of Figure 6 represents the percentage of charges (%) when ferrous scrap is upgraded. Charges with upgraded ferrous scrap refer to charges where the predicted Cu concentration of molten iron is judged to exceed the upper limit, and the ferrous scrap being charged is upgraded.
[0071] As described in the above embodiment, as the rank of the iron scrap increases, the Cu concentration of the iron scrap becomes lower than that of the iron scrap before the rank upgrade. On the other hand, as the rank of the iron scrap increases, the cost of the iron scrap becomes higher than that of the iron scrap before the rank upgrade.
[0072] As shown in Figure 6, the proportion of charge involving higher-grade iron-based scrap was lower in the inventive example compared to Comparative Examples 1 and 2. These results confirm that the inventive example can predict the Cu concentration with higher accuracy than Comparative Examples 1 and 2 in the production of molten iron produced by melting iron-based scrap, thereby suppressing the upgrading of iron-based scrap. As a result, it was confirmed that the upgrading of iron-based scrap can be suppressed in the production of molten iron produced by melting iron-based scrap, thereby suppressing the increase in the production cost of molten iron.
[0073] Figure 7 is a graph showing the charge ratio when iron scrap is downgraded. The vertical axis of Figure 7 represents the percentage of charges (%) when iron scrap is downgraded. Charges with downgraded iron scrap refer to charges where the iron scrap being charged is downgraded because the predicted Cu concentration of the molten iron fell below the target range.
[0074] As shown in Figure 7, the inventive example had a higher proportion of charge made from downgraded iron scrap compared to Comparative Examples 1 and 2. These results confirm that the inventive example can predict the Cu concentration with higher accuracy than Comparative Examples 1 and 2 in the production of molten iron produced by melting iron scrap, thereby promoting the downgrading of iron scrap. As a result, it has been confirmed that the downgrading of iron scrap can be promoted in the production of molten iron produced by melting iron scrap, thereby reducing the production cost of molten iron.
[0075] (Example 2) Next, we will describe Example 2, in which, in addition to image data of iron scrap, shape features and color features extracted from the image data were input into an impurity concentration prediction model to predict the Cu concentration of molten iron.
[0076] In Invention Example 2-2 of Example 2, iron-based scrap was loaded into an arc electric furnace multiple times, and each time the iron-based scrap was transported, it was imaged using an imaging device to generate image data. This image data was analyzed to extract aspect ratio and circularity as shape information, and color information (RGB and HSV histograms) of copper-colored components (copper color) and plastic colors (red and blue) was extracted as color information.
[0077] In Invention Example 2-2, a machine learning model combining a convolutional neural network and LightGBM was used to construct an impurity concentration prediction model that takes image data, extracted shape information, and color information as input data, and outputs the Cu concentration of molten iron. Using this impurity concentration prediction model, the Cu concentration of molten iron was predicted, and molten iron was produced by charging iron-based scrap multiple times.
[0078] On the other hand, in Invention Example 2-1, similar to Invention Example 2-2, a machine learning model combining a convolutional neural network and LightGBM was used to construct an impurity concentration prediction model that takes image data as input and outputs the Cu concentration of molten iron. Molten iron was produced by charging iron-based scrap into an electric furnace multiple times while predicting the Cu concentration of the molten iron using this impurity concentration prediction model.
[0079] Figure 8 is a graph showing the error range (2σ) of the Cu concentration of the molten iron produced in Invention Example 2-1 and Invention Example 2-2. The horizontal axis of Figure 8 represents the upper limit of the Cu concentration (target value: mass%), and the vertical axis represents the error range (2σ) of the Cu concentration. As shown in Figure 8, Invention Example 2-2, which predicts the Cu concentration by including shape information and color information in the image data, showed smaller errors at all Cu concentrations compared to Invention Example 2-1, which predicts the Cu concentration using image data.
[0080] From these results, it was confirmed that Invention Example 2-2, which predicts impurity concentration by including shape information and color information in image data, can control the Cu concentration within the target range with smaller fluctuations than Invention Example 2-1. Thus, in order to control the Cu concentration of molten iron within the target range with small fluctuations, it is necessary to predict the Cu concentration of molten iron produced by melting iron-based scrap with high accuracy. Therefore, from the above results, it is considered that Invention Example 2-2 can predict the Cu concentration with higher accuracy than Invention Example 2-1, and thereby was able to control the Cu concentration of molten iron within the target range with high accuracy.
[0081] Figure 9 is a graph showing the charge ratio when ferrous scrap is upgraded. The vertical axis of Figure 9 represents the percentage of charges (%) when ferrous scrap is upgraded. Charges when ferrous scrap is upgraded refer to charges where the predicted Cu concentration of molten iron is judged to exceed the upper limit, and the ferrous scrap being charged is upgraded.
[0082] As shown in Figure 9, Invention Example 2-2 had a lower ratio of charges involving higher-grade iron scrap than Invention Example 2-1. These results confirm that Invention Example 2-2 can predict the Cu concentration with higher accuracy than Invention Example 2-1 in the production of molten iron produced by melting iron scrap, thereby suppressing the upgrading of iron scrap. As a result, Invention Example 2-2 can suppress the upgrading of iron scrap in the production of molten iron produced by melting iron scrap, thereby suppressing the increase in the production cost of molten iron.
[0083] Figure 10 is a graph showing the charge ratio when iron scrap is downgraded. The vertical axis of Figure 10 represents the percentage of charges where the iron scrap being charged is downgraded. Charges where iron scrap is downgraded refer to charges where the iron scrap being charged is downgraded because the predicted Cu concentration of the molten iron fell below the target range.
[0084] As shown in Figure 10, Invention Example 2-2 showed a higher proportion of charge made from downgraded iron scrap compared to Invention Example 2-1. These results confirm that Invention Example 2-2 can predict the Cu concentration with higher accuracy than Invention Example 2-1 in the production of molten iron produced by melting iron scrap, thereby promoting the downgrading of iron scrap. As a result, Invention Example 2-2 can promote the downgrading of iron scrap in the production of molten iron produced by melting iron scrap, thereby reducing the production cost of molten iron.
[0085] 10 Scrap yard 12 Molten iron impurity concentration prediction device 14 Lifting magnet 16 Crane scale 18 Imaging device 20 Calculation unit 22 Control unit 24 Input unit 26 Output unit 28 Storage unit 30 Communication unit 32 Impurity concentration prediction unit 34 Charging control unit 36 Impurity concentration prediction model generation unit 37 Impurity concentration prediction model 38 Database 40 Iron-based scrap 50 Arc electric furnace 100 Molten iron manufacturing equipment
Claims
1. A method for predicting the impurity concentration of molten iron produced by melting iron scrap in a smelting furnace, comprising: a measurement step of measuring the amount of iron scrap to be charged into the smelting furnace; an imaging step of imaging the iron scrap while it is being transported to the smelting furnace and generating image data; and an impurity concentration prediction step of inputting the amount of iron scrap charged and the image data into an impurity concentration prediction model to output the impurity concentration of the molten iron.
2. The method for predicting the impurity concentration of molten iron according to claim 1, wherein the input data further includes the impurity concentration of the molten iron from the previous charge and the amount of remaining molten iron from the charge.
3. A method for producing molten iron by melting iron scrap in a smelting furnace, comprising: a charging step of charging the iron scrap into the smelting furnace such that the impurity concentration of the molten iron predicted in the impurity concentration prediction step of the method for predicting the impurity concentration of molten iron according to claim 1 or claim 2 falls within a predetermined range; and a melting step of melting the iron scrap charged into the smelting furnace.
4. The method for producing molten iron according to claim 3, wherein the iron scrap is charged into the smelting furnace in multiple batches, the impurity concentration is predicted each time the iron scrap is charged in the impurity concentration prediction step, and the iron scrap is charged in the charging step such that the impurity concentration of the molten iron, which is obtained by accumulating the impurity concentrations predicted each time the iron scrap is charged, falls within a predetermined range.
5. A device for predicting the impurity concentration of molten iron produced by melting iron scrap in a smelting furnace, comprising: a conveying device for conveying the iron scrap to the smelting furnace; a measuring device for measuring the amount of iron scrap to be charged into the smelting furnace; an imaging device for imaging the iron scrap during conveyance and generating image data; and a calculation device for predicting the impurity concentration of the molten iron, wherein the calculation device has an impurity concentration prediction unit that inputs input data including image data acquired from the imaging device and the amount of iron scrap acquired from the measuring device into an impurity concentration prediction model to output the impurity concentration of the molten iron.
6. The molten iron impurity concentration prediction device according to claim 5, wherein the input data further includes the impurity concentration of the molten iron from the previous charge and the remaining amount of molten iron from the charge.
7. The molten iron impurity concentration prediction device according to claim 5 or 6, wherein the calculation device has a charging control unit that controls the charging of iron-based scrap into the smelting furnace so that the impurity concentration of the molten iron predicted by the impurity concentration prediction unit falls within a predetermined range.
8. The device for predicting the impurity concentration of molten iron according to claim 7, wherein the iron scrap is charged into the smelting furnace in multiple batches, the impurity concentration prediction unit predicts the impurity concentration each time the iron scrap is charged, and the charging control unit controls the charging of the iron scrap so that the impurity concentration of the molten iron, which is obtained by accumulating the impurity concentrations predicted each time the iron scrap is charged, falls within a predetermined range.
9. A method for generating an impurity concentration prediction model used to predict the impurity concentration of molten iron produced by melting iron scrap in a smelting furnace, comprising: training a machine learning model with multiple datasets, each consisting of image data of the iron scrap in past molten iron production, actual values of the amount of iron scrap charged, and actual values of the impurity concentration of the molten iron, as training data; and generating an impurity concentration prediction model that takes input data including the image data and the amount charged as input and outputs the impurity concentration of the molten iron.