JP2024031854A - Composition estimation method for estimating the chemical composition of scrap materials - Google Patents

Abstract

[Problem] A composition estimation method, a scrap material management method, a composition estimation program, a composition estimation device, a melting furnace equipment, which can efficiently manufacture metal products having a desired chemical composition using scrap materials as raw materials; Provides a scrapyard and a method for generating a trained model. A component estimation method for estimating the chemical components of scrap material 16, in which a plurality of images obtained by dividing an image 18 obtained through photographing a predetermined scrap material 16 into predetermined image sizes is provided. A pre-processing step S2 of creating an image group consisting of all or part of the divided images 18b, and a pre-processing step S2 for each divided image 18b included in the image group, using the trained model 19 with the divided image 18b as an input image. , a chemical component output step S4 of outputting a chemical component corresponding to the predetermined scrap material 16 in the input image. [Selection diagram] Figure 7

Description

The present invention relates to a component estimation method for estimating the chemical components of scrap materials.

BACKGROUND ART A technique is known in which a desired metal product is obtained by supplying various raw materials including scrap materials to a melting furnace and melting them (see, for example, Patent Document 1). Furthermore, a technique for estimating the grade of scrap material from an image of the scrap material is known (for example, see Patent Document 2).

Japanese Patent Application Publication No. 2020-185573 Patent No. 7036296

If the chemical composition of scrap material can be estimated before it is supplied to the melting furnace, metal products having desired chemical compositions can be efficiently produced.

Therefore, the object of the present invention is to provide a component estimation method, a scrap material management method, a component estimation program, a component estimation device, and a component estimation method that can efficiently manufacture metal products having desired chemical components using scrap materials as raw materials. The object of the present invention is to provide melting furnace equipment, a scrap yard, and a method for generating a trained model.

One aspect of the present invention is as follows.

[1]
A composition estimation method for estimating the chemical composition of scrap materials, the method comprising:
a pre-processing step of creating an image group consisting of all or part of a plurality of divided images obtained by dividing an image of a prescribed scrap material into a prescribed image size;
For each of the divided images included in the image group, the divided image is used as an input image, and a chemical component output step of outputting a chemical component corresponding to the predetermined scrap material in the input image using a learned model;
A component estimation method having

[2]
The component estimation method according to [1], wherein the predetermined image size is 2000 mm or less on all sides on an actual scale.

[3]
In the chemical component output step, for each divided image included in the image group, the divided image is used as an input image, and the learned model outputs product category information corresponding to the predetermined scrap material in the input image. [1] or [2], further comprising a step of outputting product category information, wherein the product category information obtained through the step of outputting product category information is input, and a chemical component corresponding to the input is outputted from a database. The component estimation method described.

[4]
The component estimation method according to [3], wherein in the product category information output step, one product category is output as the product category information for the input image.

[5]
The method according to any one of [1] to [4], further comprising a post-processing step of estimating the chemical components of the original image using the chemical components output for each input image by the chemical component output step. Component estimation method.

[6]
Estimating the chemical composition of the scrap material to be managed using the composition estimation method described in any one of [1] to [5],
If the chemical components estimated by the component estimation method do not satisfy the predetermined conditions, based on the estimated chemical components, a portion of the scrap material to be managed is removed, or the scrap material to be managed is removed. A scrap material management method that adds another scrap material to the.

[7]
A part or all of the scrap material whose chemical composition has been estimated in advance by the composition estimation method according to any one of [1] to [5] is melted based on the chemical composition estimated by the composition estimation method. A method of operating a melting furnace comprising a step of supplying scrap material to the furnace.

[8]
A method for operating a melting furnace, comprising a scrap material supply step of supplying a part or all of the scrap material to be managed by the scrap material management method according to [6] to the melting furnace.

[9]
A component estimation program that causes a computer to execute the component estimation method according to any one of [1] to [5].

[10]
A component estimation device for estimating the chemical components of scrap materials,
a pre-processing unit that creates an image group consisting of all or part of a plurality of divided images obtained by dividing an image of a prescribed scrap material into a prescribed image size;
a chemical component output unit that takes the divided images included in the image group as an input image and outputs a chemical component corresponding to the predetermined scrap material in the input image using a learned model;
A component estimation device having:

[11]
[10] Melting furnace equipment comprising the component estimating device according to item [10] and a melting furnace.

[12]
A scrapyard having the component estimating device according to [10].

[13]
A method for generating a trained model that uses an image of a scrap material as an input and outputs product category information of the scrap material, the method comprising:
Teacher data including a combination of a product category and a teacher image containing scrap materials corresponding to the product category, and size change, movement, rotation, partial cropping, or reversal of the teacher image by enlarging or reducing the teacher image, or A method for generating a trained model, which generates the trained model by learning using expanded training data generated by performing arbitrary combinations.

[14]
A trained model generated by the trained model generation method described in [13].

According to the present invention, a component estimating method, a scrap material management method, a component estimating program, a component estimating device, a melting Furnace equipment, scrapyards, and methods for generating trained models can be provided.

FIG. 1 is a block diagram showing a component estimating device according to an embodiment of the present invention. FIG. 2 is a schematic diagram showing a trained model used in the component estimating device shown in FIG. 1. FIG. It is a schematic diagram which shows an example of the image obtained by photographing a predetermined scrap material. 4 is a schematic diagram showing an example of a background-removed image formed by a preprocessing unit from the image shown in FIG. 3. FIG. FIG. 2 is a schematic diagram showing an example of melting furnace equipment having the component estimating device shown in FIG. 1. FIG. FIG. 6 is a schematic diagram showing a state in which moving scrap material is photographed in the melting furnace equipment shown in FIG. 5; FIG. 2 is a flow diagram showing an example of a method for operating a melting furnace using the component estimating device shown in FIG. 1. FIG. FIG. 2 is a flowchart showing a first specific example of a melting furnace operating method using the component estimating device shown in FIG. 1. FIG. FIG. 2 is a flowchart showing a second specific example of a melting furnace operating method using the component estimating device shown in FIG. 1; FIG. 2 is an explanatory diagram illustrating an example of a database in the chemical component output section shown in FIG. 1. FIG. 3 is an explanatory diagram illustrating an example of teacher data for generating the learned model shown in FIG. 2. FIG. 12 is an explanatory diagram illustrating an example of expanded teacher data generated from the teacher data shown in FIG. 11. FIG.

Embodiments of the present invention will be described below by way of example with reference to the drawings.

As shown in FIG. 1, in an embodiment of the present invention, a component estimating device 1 includes a pre-processing section 2, a chemical component output section 4 (including a product category information output section 3), a post-processing section 5, an imaging device 6, and a chemical component output section 4 (including a product category information output section 3). It has a notification section 7. The pre-processing section 2, the chemical component output section 4, and the post-processing section 5 are each constituted by a part of the processing section 9 of the computer 8. The computer 8 includes a central processing unit 10, a main memory 11, and an auxiliary memory 12. A photographing device 6, an input device 13, and an output device 14 are communicably connected to the computer 8 by wire or wirelessly. The notification section 7 is constituted by an output device 14.

The central processing unit 10 is configured by, for example, a CPU (Central Processing Unit). Note that the central processing unit 10 may be configured with an accelerator having a large number of calculation cores such as a GPU (Graphics Processing Unit) for high-speed calculation. The main storage device 11 is composed of, for example, a memory. The auxiliary storage device 12 is configured by, for example, an SSD (Solid State Drive) or an HDD (Hard Disk Drive). The photographing device 6 is composed of, for example, a camera. The input device 13 includes, for example, a keyboard, a mouse, or a touch panel. The output device 14 is configured by, for example, a display or a speaker.

The component estimating device 1 of this embodiment estimates the chemical components of the scrap material 16 in order to efficiently operate the melting furnace 15 for manufacturing ferrous metal products such as steel (see FIG. 5). Note that the component estimating device 1 of this embodiment can also be applied to the operation of the melting furnace 15 for manufacturing metal products other than iron-based metal products. In this embodiment, an electric furnace having an electrode 17 for melting the scrap material 16 is used as the melting furnace 15, but a melting furnace 15 other than an electric furnace may be used. The photographing device 6 photographs a predetermined scrap material 16 and generates an image 18 of the scrap material 16 (see FIG. 3). The auxiliary storage device 12 stores an image 18 generated by the photographing device 6, data obtained through arithmetic processing by the processing section 9, and the like. The central processing unit 10 processes the component estimation program.

In order to reduce the cost of molten iron in electric furnaces, it is necessary to use various iron-based scrap materials 16 such as low-grade iron-based scrap materials 16 called H2 and H3 (referred to as low-grade scrap materials). be. Low-grade scrap materials are often collected from automobiles, construction waste, electrical appliances, etc., and it is unknown what kind of chemical composition of iron they contain, so the chemical composition of low-grade scrap materials is not constant. Further, the low-grade scrap material may include metal materials other than the iron-based scrap material 16, such as electric motors and electric wires, whose main component is other than iron (Fe). Copper (Cu) in particular is difficult to remove using current steelmaking methods, and if it is contained in molten iron, it must be diluted to below a specified value with other iron-based scrap materials (referred to as high-grade scrap materials) that do not contain Cu. There is a need.

If the chemical composition of the molten metal of the iron-based scrap material 16 is only measured, it is necessary to input a large amount of additional material when the chemical composition significantly deviates from the target chemical composition, which increases the cost of the additional material. Furthermore, if there is a large amount of undesirable chemical components such as Cu, it is necessary to dilute it with high-grade scrap material, which means that a large amount of high-grade scrap material must be used. Since it takes a lot of time to melt a large amount of high-grade scrap material, the time required to produce molten metal with a desired chemical composition increases, leading to a significant deterioration in the operational efficiency of the melting furnace 15. In order to reduce the cost of additional materials and improve the operational efficiency of the melting furnace 15, it is necessary to estimate the chemical composition of the scrap material 16 supplied to the melting furnace 15 in advance and adjust the supplied material in advance based on the estimated value. is preferred.

Therefore, the component estimation program of this embodiment includes a photographing step S1, a preprocessing step S2, a chemical component output step S4 (including a product category information output step S3), a postprocessing step S5, and a notification step S6 (see FIG. 7). The computer 8 is caused to execute the following in this order. Therefore, in the component estimation method of this embodiment, the photographing step S1, the pre-processing step S2, the chemical component output step S4 (including the product category information output step S3), the post-processing step S5, and the notification step S6 are executed in this order. Further, in the melting furnace operating method of this embodiment, a photographing step S1, a pre-processing step S2, a chemical component output step S4 (including a product category information output step S3), a post-processing step S5, a notification step S6, a scrap material inspection step S7 and scrap material supply step S8 are executed in this order (see FIG. 7). The component estimation program of this embodiment can be changed to any configuration that causes the computer 8 to execute the preprocessing step S2 and the chemical component output step S4 in this order. In addition, the component estimation method and the melting furnace operation method can be changed accordingly.

In the photographing step S1, the processing unit 9 photographs a predetermined scrap material 16 using the photographing device 6 based on settings or instructions via the input device 13, and generates an image 18 of the scrap material 16 (see FIG. (See 3). In the photographing step S1, for example, an image 18 is photographed with the photographing device 6 such as a still camera or a video camera so that the aggregate of the scrap materials 16 is captured.

The preprocessing step S2 includes an image segmentation step. In the image division step, the preprocessing unit 2 divides the image 18 (referred to as the whole image 18a) obtained through photographing the predetermined scrap material 16 into a predetermined image size, and divides the image 18 (referred to as the whole image 18a) into a plurality of divided images 18 (divided An image 18b) is generated, and an image group consisting of all or part of the plurality of divided images 18b is created. At that time, for example, it is preferable to divide the entire image 18a into a grid pattern. For example, it is more preferable to divide one divided image 18b into a parallelogram (including a rectangle, a rhombus, and a square). In this case, one divided image 18b has four sides. Further, as shown in FIG. 3, it is more preferable to divide the image so that one divided image 18b becomes a rectangle (including a square).

Further, it is preferable that the divided image 18b includes as few types of scrap materials 16 as possible. In other words, the divided images 18b are preferably divided into predetermined image sizes such that one divided image 18b includes a size that allows the product category of the scrap material 16 to be determined. Furthermore, it is more preferable that the divided images 18b are divided into a predetermined image size in which each divided image 18b includes a single type of scrap material 16. Therefore, since the scrap material 16 is often 600 mm or more and 2000 mm or less, the image size of one divided image 18b may be set so that all sides of the divided image 18b are 2000 mm or less in actual scale. preferable. The shapes of the scrap materials 16 vary, but most of them are several mm or more in size, so the image size of one divided image 18b is such that all sides of the divided image 18b are 10 mm or more and 2000 mm or less. It is more preferable to set it so that Further, considering the installation distance of the camera, it is more preferable to set all sides of one divided image 18b to be 100 mm or more and 2000 mm or less. For example, when the entire image 18a is divided such that one divided image 18b is a parallelogram, it means that the sizes of all four sides of the divided image 18b satisfy the above conditions. The image segmentation step may be performed only when the image 18 includes a certain number or more (for example, a certain number or more) of scrap materials 16.

The preprocessing step S2 may include a background separation step. In the background separation step, the preprocessing unit 2 separates the background, which is a portion other than the scrap material 16, from the image 18 to generate an image 18 from which the background is at least partially removed (see FIG. 4). The background separation step may be performed only if the image 18 contains more than a certain amount of background. To separate the background, for example, an image 18 including the scrap material 16 is input, and image segmentation that is trained in advance using an annotation image in which each pixel indicating the background and the pixel indicating the scrap material 16 in the image 18 is labeled is performed. can be used. In the background separation step, if the hue is clearly different from the scrap material 16, it may be determined by color and removed. For example, when the sky is reflected as a background, the scrap material 16 is often gray or brown, so the divided image 18b containing a large amount of blue may be removed as the background. Further, since the scrap material 16 exists vertically below the image 18, the divided image 18b that is located at the top of the image 18 and has a different hue from the bottom of the image 18 may be removed as a background. The divided image 18b other than the scrap material 16 may be removed as the background by determining the scrap material 16 and the background using a machine learning model trained as teacher data. It is preferable to attach a removal flag to the divided image 18b to be removed so that it can be identified in future processing. The background separation step can be performed after the image segmentation step. In this case, the preprocessing unit 2 creates a part of the plurality of divided images 18b obtained by dividing the image 18 into predetermined image sizes, that is, an image group consisting of the divided images 18b other than the background portion. . The background separation step may be performed before the image segmentation step. In this case, the preprocessing unit 2 creates an image group consisting of all or part of the plurality of divided images 18b obtained by dividing the background-separated image 18 into predetermined image sizes.

In the product category information output step S3, the product category information output unit 3 inputs the image 18 obtained by photographing a predetermined scrap material 16, and uses the learned model 19 (see FIG. 2) to determine the image of the scrap material 16. Product category information 20 is output. In the product category information output step S3 of the present embodiment, the product category information output unit 3 uses the divided image 18b as an input image for each divided image 18b included in the image group created by the preprocessing unit 2, and uses the learned The model 19 outputs product category information 20 corresponding to the predetermined scrap material 16 in the input image.

As shown in FIG. 2, the trained model 19 is a trained machine learning model, which inputs the image 18 of the scrap material 16 and outputs the product category information 20 of the scrap material 16. The trained model 19 is obtained by machine learning using a combination of the image 18 of the scrap material 16 and the product category information 20 corresponding to the scrap material 16 as training data. For example, the trained model 19 is configured with a convolutional neural network, etc., and includes a combination of a product category and a teacher image containing scrap material 16 that corresponds to the product category (i.e., It can be generated by learning in advance using teacher data (the input is the image 18 including the scrap material 16 to which it belongs, and the output is the product category). In the learning to obtain the trained model 19, for example, as shown in FIG. 12, the teacher image is augmented by resizing the teacher image by enlarging or reducing it, moving, rotating, partially cropping or reversing, or any combination thereof. Extended training data generated by performing a simulation may also be used. Although it is preferable to prepare 10 times or more of the training data as the number of parameters of the machine learning to be used, it is generally physically difficult to prepare a large amount of training data in many cases. Therefore, it is preferable to use expanded teacher data obtained by expanding the teacher data as the teacher data. In data expansion, expanded teacher data is created by randomly combining enlargement, reduction, movement, rotation, partial cropping, and reversal of the teacher image as shown in FIG. By using data expansion such as enlargement, reduction, movement, rotation, partial cropping, and reversal, it becomes possible to perform learning using a teacher image that is close to an image of the actual scrap material 16. Multiple selections may be made for enlarging, reducing, moving, rotating, partially cropping, and reversing. Since the scrap material 16 is photographed in various postures, estimation accuracy can be improved by performing augmentation. Preferably, the machine learning model includes multiple neural networks or one or more decision trees to classify multiple classes. The teacher image corresponds to a product category, the scrap material 16 corresponding to the product category is reflected, and the number of pixels indicated by the corresponding product category among the pixels of the teacher image is greater than the number of pixels indicating other scrap materials 16. It may be something that is visible a lot. The teacher data used for learning the machine learning model may include teacher data in which a background image that is an image other than the scrap material 16 is associated with a category other than scrap materials.

The scrap materials 16 come in a variety of colors and shapes, and when considering the combinations of how they overlap, the required training data is almost infinite, and it is generally difficult to create a machine learning model that operates stably. However, as in the present embodiment, by dividing the image 18 and generating the divided image 18b such that preferably a single type of scrap material 16 is reflected, the teacher image of the prepared training data is a single type of scrap material 16. Only scrap material 16 is sufficient, and the required training data can be significantly reduced. Further, the scrap materials 16 may be randomly arranged and photographed at various angles and distances. By applying data expansion, it becomes possible to efficiently train a machine learning model with a small amount of data in estimating the product category of the scrap material 16. It is more effective to mix not only the scrap material 16 but also images other than the scrap material 16 as the background in the training data. In the preprocessing step S2, a removal flag may be attached to the divided images 18b that clearly do not include the scrap material 16, but even in this case, there is a possibility that the divided images 18b include images with a large amount of background. Therefore, by learning the background image, it becomes possible to distinguish the divided images 18b other than the scrap material 16 from the divided images 18b, and the accuracy of the estimated chemical components can be improved.

The product category information 20 is information indicating a product category that is a category of the original product of the scrap material 16 having various shapes such as shaped steel, bar shape, and wire shape included in the image 18. Product categories can be obtained by classifying various products that can become scrap materials 16 into several categories in advance. For example, product categories are classified into automobiles, building materials, electrical appliances, etc. Product categories may be further subdivided. For example, steel frames, reinforcing bars, etc. can be used instead of building materials, and refrigerators, microwave ovens, etc. can be used instead of electrical appliances. When the image 18 includes a plurality of scrap materials 16, the product category information 20 corresponding to the image 18 may be information indicating a plurality of product categories. In this case, the product category information 20 may be information such as 60% automobiles, 20% building materials, 10% electrical appliances, etc., as shown in FIG. 2, for example. At this time, the numerical value indicating the proportion of each product category included in the product category information 20 (a numerical value such as "60%" for automobiles) represents the possibility that the image 18 corresponds to the product category (certainty of estimation). It may be a numerical value. In this case, for example, the trained model 19 outputs the estimated probability of each product category for the divided image 18b using a softmax function. The product category information output step S3 may output one product category as the product category information 20 for the input image. That is, the product category information 20 may be information indicating only a specific product category to which the image 18 most likely corresponds. In this case, for example, the product category information output step S3 estimates that the product category with the highest probability among the estimated probabilities of each product category for the divided image 18b is the product category of the divided image 18b. The product category information 20 may indicate categories other than product categories. Such categories may include, for example, background (portions other than scrap material 16 in image 18).

In the present embodiment, in the chemical component output step S4, the chemical component output unit 4 inputs the product category information 20 obtained via the product category information output unit 3 for each divided image 18b included in the image group, and inputs the product category information 20 obtained through the product category information output unit 3. Outputs the chemical components corresponding to the input. In the chemical component output step S4 of the present embodiment, the chemical component output unit 4 receives the product category information 20 obtained through the product category information output step S3 as input, and uses the database as shown in FIG. Output the corresponding chemical components. More specifically, for each product category information 20 output by the product category information output unit 3, the product category information 20 is input, and the chemical components corresponding to the input are outputted from the database. That is, the chemical component output unit 4 outputs the chemical components corresponding to the divided image 18b for each product category information 20 corresponding to the divided image 18b.

If the scrap materials 16 belong to the same product category, they often include metal materials with similar chemical components. Therefore, a table (database) that specifies chemical components for each product category is created in advance. For example, a car can be specified as having A% iron and B% copper, and a building material can be specified as having C% iron and D% copper. By comparing with the table, the chemical components corresponding to each divided image 18b can be output. In this case, for example, in the chemical component output step S4, the chemical components of the image 18 may be predicted by taking a weighted average of the chemical components in the chemical component table associated with the product category using the estimated probability of the product category as a weight. good. For example, if the product category information 20 of the image 18 input to the chemical component output unit 4 is 50% automobiles and 50% building materials, the chemical components output by the chemical component output unit 4 are (A+C)/2% iron, copper becomes (B+D)/2%.

In post-processing step S5, the post-processing unit 5 averages the chemical components output by the chemical component output unit 4. The averaging process may be performed using an arithmetic average or a harmonic average, or may be performed using a weighted average that weights the divided images 18b, for example. The post-processing unit 5 estimates the chemical components of the original image 18 using the chemical components output for each input image in the chemical component output step S4. That is, the post-processing unit 5 outputs the chemical component corresponding to the entire image 18a (that is, the chemical component of the scrap material 16 included in the entire image 18a) as an estimated value through the above-mentioned averaging process. The chemical component output unit 4 attaches an exclusion flag to the divided image 18b for which the product category is determined to be the background by the chemical component output unit 4 in the chemical component output step S4, and the post-processing unit 5 attaches the exclusion flag to the divided image 18b in which the product category is determined to be the background by the chemical component output unit 4 in the chemical component output step S4. The attached divided image 18b may be excluded from the averaging process. If the total weight of the scrap material 16 is known, the estimated chemical components may be multiplied by the total weight to output the estimated weight of each chemical component.

In this way, by executing the photographing step S1, the pre-processing step S2, the chemical component output step S4 (as described above, in this embodiment, the product category information output step S3 is included) and the post-processing step S5, The chemical composition of a given scrap material 16 can be estimated. Therefore, by performing the scrap material supply step S8 (see FIG. 7) in which the scrap material 16 is supplied to the melting furnace 15 based on the result of the estimation, efficient manufacturing of metal products can be realized. Note that details of the scrap material supply step S8 will be described later.

After the product category information output step S3, a product category information averaging step is executed to obtain product category information 20 corresponding to the entire image 18a by averaging the product category information 20 output for each divided image 18b. A product category information averaging processing section is provided in the processing section 9, and after the product category information averaging processing step, in a chemical component output step S4, the chemical component output section 4 inputs the product category information 20 corresponding to the entire image 18a. The chemical component corresponding to the input may be output. With such a configuration and method as well, the chemical composition of a predetermined scrap material 16 can be estimated.

In the notification step S6, the notification unit 7 issues a notification requesting inspection of the photographed predetermined scrap material 16 when the product category information output unit 3 outputs the product category information 20 including the predetermined product category. The notification can be made via the output device 14, for example, by images or sounds. According to the notification step S6, for example, a predetermined product category (e.g. When the product category information output unit 3 outputs product category information 20 including product categories (such as product categories including electric motors that are likely to contain a large amount of copper), in the scrap material inspection step S7, for example, a person or a machine, etc. Based on the notification, the photographed predetermined scrap material 16 is inspected before being supplied to the melting furnace 15, and scrap materials 16 belonging to the predetermined product category can be removed from among them as necessary. Therefore, according to the notification step S6, it is possible to manufacture metal products even more efficiently.

The chemical component output unit 4 attaches an inspection flag to the divided image 18b on which the product category information 20 as described above has been outputted by the chemical component output unit 4 in the chemical component output step S4, and the divided image with the inspection flag is attached in the notification step S6. 18b may be displayed for notification. According to such notification, the inspection can be easily performed in the inspection step.

If the scrap material 16 containing undesirable chemical components is removed by the notification step S6 and the scrap material inspection step S7, it is possible to reduce the need to dilute the molten metal with additional high-quality scrap material 16, thereby reducing manufacturing costs. Can be reduced. Moreover, since the melting time of the additional high-quality scrap material 16 can be reduced, the operating time can be shortened.

Note that the notification step S6 (notification unit 7) and the scrap material inspection step S7 may not be provided. The pre-processing step S2 (pre-processing section 2) and the post-processing step S5 (post-processing section 5) may not be provided. The process may be started by a person taking a photograph using the photographing device 6 and inputting the obtained image 18 to the computer 8.

The chemical components estimated in the post-processing step S5 can be output to the output device 14 or stored in the auxiliary storage device 12, for example, to notify a person. Furthermore, when an inspection flag is attached, the notification unit 7 of the output device 14 can notify that an inspection is necessary. The divided image 18b and the corresponding product category information 20, inspection flag, and exclusion flag may be stored in the auxiliary storage device 12 and output to the output device 14 or the like according to instructions from the input device 13 when necessary.

In the scrap material supply step S8, for example, a person or a machine calculates the estimation result of the chemical composition of the scrap material 16 by the component estimation method of the present embodiment (that is, the estimated value of the chemical composition of the scrap material 16 included in the entire image 18a, or , the estimated chemical composition of the scrap material 16 included in the overall image 18a), the scrap material 16 is supplied to the melting furnace 15. By supplying the scrap material 16 whose approximate chemical composition is known in advance to the melting furnace 15, metal products can be efficiently manufactured. In other words, the melting furnace operating method according to the present embodiment uses a part or all of the scrap material 16 whose chemical composition has been estimated in advance by the composition estimation method according to the present embodiment based on the chemical composition estimated by the composition estimation method. and a step of supplying scrap material to the melting furnace.

As another method for operating the melting furnace, it is also possible to use a method for managing scrap material 16, which will be described later. That is, the melting furnace operating method according to another embodiment includes a scrap material supply step of supplying part or all of the scrap material 16 to be managed by the scrap material management method described below to the melting furnace. In this case, the chemical composition of the scrap material 16 is controlled in advance under predetermined conditions, and the scrap material 16 can be supplied to the melting furnace 15 without estimating the chemical composition again, so metal products can be manufactured efficiently. be able to. In this case, a plurality of parts of the managed scrap material 16 may be prepared and further mixed before being supplied. Furthermore, before supplying the mixed scrap material 16, the chemical composition can be estimated again using the composition estimation method of this embodiment, and then the scrap material 16 can be supplied to the melting furnace.

Furthermore, the chemical components of the scrap material 16 may be managed based on the chemical components of the scrap material 16 estimated in advance by the component estimation method of this embodiment. In this case, the scrap material management method estimates the chemical components of the scrap material 16 to be managed using the component estimation method of this embodiment, and if the chemical components estimated by the component estimation method do not satisfy the predetermined conditions, Based on the estimated chemical composition, a part of the scrap material 16 to be managed is removed, or another scrap material 16 is added to the scrap material 16 to be managed. The predetermined conditions are set in advance, for example.

When a part of the scrap pile 21 (see FIG. 5) formed by stacking the scrap materials 16 is photographed, the estimated value of the chemical composition of the scrap materials 16 obtained from the acquired image 18 is used as the chemical composition of the scrap pile 21. It may be used as a representative value. In that case, the above-mentioned representative value is used as an estimated value of the chemical composition of any scrap material 16 taken out from the scrap pile 21. After measuring the weight of the scrap material 16 taken out from the scrap pile 21 using a weighing device, the amount of chemical components of the scrap material 16 to be supplied to the melting furnace 15 is estimated by multiplying the weight by a representative value of the chemical components. be able to. Based on the estimated amount of chemical components, check the difference between the chemical components of the target molten metal and determine the scrap pile 21 from which to take out the scrap material 16 to be supplied next and the amount of scrap material 16 to be taken out. Metal products with target components can be obtained in a short amount of time. By adjusting the supply amount of auxiliary raw materials such as ore, it is possible to get the chemical composition of the molten metal closer to the desired chemical composition, thereby reducing time and improving the quality of metal products. If molten metal with the desired chemical composition can be produced in a short time, the energy cost for operating the melting furnace 15 can be reduced, and therefore the cost of metal products can be reduced.

In a melting furnace facility 25 having a scrapyard 24 having a component estimating device 1 and a melting furnace 15, as shown in FIG. When the image 18 is acquired, the chemical components obtained from the image 18 may be used as representative values of the chemical components of the scrap material 16 being transported. After measuring the weight of the scrap material 16 being transported, the amount of chemical components of the scrap material 16 being transported to be supplied to the melting furnace 15 can be estimated by multiplying the weight by the representative value of the chemical component. . By measuring each transport device, it is possible to estimate the chemical components with higher accuracy than for each scrap pile 21.

The photographing device 6 is preferably arranged to photograph the moving scrap material 16. For example, when the photographing device 6 is moving the scrap material 16 with the lifting magnet 22 and transferring the scrap material 16 to another transport device by tilting the transport device, the photographing device 6 can record the scrap material 16 being thrown into the melting furnace 15 from the transport device, An image 18 is generated by photographing the scrap material 16 while the scrap material 16 is moving, such as when the material 16 is descending. By photographing the moving scrap material 16, it is possible to photograph the scrap material 16 in a more scattered state, so that the estimation accuracy of the product category information 20 estimated from the obtained image 18 can be improved. In a stationary state, only the surface of the lump of scrap material 16 (such as the scrap pile 21) can be photographed, but if the moving scrap material 16 is photographed, it is easier to photograph the inside of the lump of scrap material 16, so the product category information 20 The estimation accuracy can be improved.

The melting furnace operating method of this embodiment may be performed, for example, as in the first specific example shown in FIG. In the first specific example, conditions such as the target chemical composition of the metal product (molten metal) are first set, and then the photographing device 6 photographs an image 18 including the scrap material 16 (photographing step S1). Then, the preprocessing unit 2 divides the image 18 and creates divided images 18b (preprocessing step S2). Then, the product category information output section 3 of the chemical component output section 4 outputs the product category information 20 for each divided image 18b using the trained model 19 (product category information output step S3). Then, the chemical component output section 4 outputs chemical components from the product category information 20 output by the product category information output section 3 using a table for each divided image 18b (chemical component output step S4). Then, the chemical component output unit 4 attaches an exclusion flag to the divided image 18b estimated by the product category information output unit 3 to be of something other than the scrap material 16 (background). Then, the post-processing unit 5 averages the chemical components of the divided images 18b to which no exclusion flag has been attached, and outputs the averaged chemical components as an estimated value of the chemical components of the entire image 18a (post-processing step S5). Then, according to the condition settings, the chemical components are outputted to the output device 14 as necessary. Then, according to the condition settings, the processing results by the processing unit 9 are stored in the auxiliary storage device 12 as necessary. Then, if the process can be terminated (such as when molten metal with the target component is obtained), the process is terminated.

Moreover, the melting furnace operating method of this embodiment may be performed, for example, as in the second specific example shown in FIG. In the second specific example, conditions such as the target chemical composition of the metal product (molten metal) are first set, and then the photographing device 6 photographs an image 18 including the scrap material 16 (photographing step S1). Then, the preprocessing unit 2 divides the image 18 and creates divided images 18b (preprocessing step S2). Then, the product category information output section 3 of the chemical component output section 4 outputs the product category information 20 for each divided image 18b using the trained model 19 (product category information output step S3). Then, the chemical component output section 4 outputs chemical components from the product category information 20 output by the product category information output section 3 using a table for each divided image 18b (chemical component output step S4). Then, the chemical component output unit 4 attaches an exclusion flag to the divided image 18b estimated by the product category information output unit 3 to be of something other than the scrap material 16 (background). In the second specific example, the divided image 18b in which the chemical component output unit 4 outputs the product category information 20 including product categories that may contain a large amount of chemical components that are not preferable to be supplied to the melting furnace 15 is If there is, an inspection flag is attached to the divided image 18b. Then, the post-processing unit 5 averages the chemical components of the divided images 18b to which no exclusion flag has been attached, and outputs the averaged chemical components as an estimated value of the chemical components of the entire image 18a (post-processing step S5). If there is a test flag, the test flag is output to the output device 14. Then, according to the condition settings, the chemical components are outputted to the output device 14 as necessary. Then, according to the condition settings, the processing results by the processing unit 9 are stored in the auxiliary storage device 12 as necessary. Then, if the process can be terminated (such as when molten metal with the target component is obtained), the process is terminated.

Note that in the embodiments described so far, an example has been described in which the learned model 19 outputs the product category information 20 of the scrap material 16, but the present invention is not limited thereto. For example, the trained model 19 may be a trained machine learning model that receives the image 18 of the scrap material 16 as an input and outputs a chemical component corresponding to the product category information 20 of the scrap material 16. That is, in the above embodiment, the learning model 19 has the function of combining the trained model 19 that outputs the product category information 20 and the database that receives the output product category information 20 as input and outputs the chemical components corresponding to the input. The completed model 19 may also be used. In this case, the learned model 19 is obtained by machine learning using a combination of the image 18 of the scrap material 16 and the chemical component corresponding to the scrap material 16 as training data.

The flow of the component estimation method in this case is shown in FIG. 7 without the product category information output step S3. In other words, there is a chemical component output step S4 that does not include the product category information output step S3. In this case, in the chemical component output step S4, the chemical component output unit 4 uses the trained model 19 as an input image, for each divided image 18b included in the image group, to generate a response corresponding to the input image. Output chemical components.

Further, in this case, the component estimation program of this embodiment causes the computer 8 to execute the chemical component output step S4, which does not include the product category information output step S3.

In order to reduce CO ₂ , the use of electric furnaces, which are said to emit about 1/4 the amount of CO ₂ of blast furnaces, is expected to expand. The main raw materials used in electric furnaces are waste materials from steel mills and various types of steel scraps from automobiles, construction, and electrical appliances. In an electric furnace, the composition of molten steel tapped from the electric furnace tends to be directly linked to the quality of the final product, so adjusting the composition of the molten metal produced in the electric furnace is important in electric furnace operation. In most cases, when operating an electric furnace, the chemical composition of the molten metal is measured and the desired molten iron composition is adjusted by adding necessary additives such as coke. In order to obtain the desired components, it is necessary to repeatedly detect the chemical components of the molten metal and add additives, which is a labor-intensive task. According to the embodiment described above, the chemical composition of scrap material to be input into an electric furnace or the like can be estimated in advance, and the composition of molten iron manufactured in the electric furnace can be adjusted in a short time. The work load on such an electric furnace can be significantly reduced. Further, according to this embodiment, the time for operating the electric furnace can be shortened, the chemical composition of the molten metal in the electric furnace is stabilized, so the quality can be improved, and the manufacturing cost of the molten metal manufactured in the electric furnace can be reduced. can.

The present invention is not limited to the embodiments described above, and can be modified in various ways without departing from the gist thereof.

1 Component estimation device 2 Pre-processing section 3 Product category information output section 4 Chemical component output section 5 Post-processing section 6 Photographing device 7 Notification section 8 Computer 9 Processing section 10 Central processing unit 11 Main storage device 12 Auxiliary storage device 13 Input device 14 Output device 15 Melting furnace 16 Scrap material 17 Electrode 18 Image 18a Whole image 18b Divided image 19 Learned model 20 Product category information 21 Scrap pile 22 Lifting magnet 23 Dolly 24 Scrap yard 25 Melting furnace equipment S1 Photographing step S2 Pre-processing step S3 Product Category information output step S4 Chemical component output step S5 Post-processing step S6 Notification step S7 Scrap material inspection step S8 Scrap material supply step

Claims (14)

A composition estimation method for estimating the chemical composition of scrap materials, the method comprising:
a pre-processing step of creating an image group consisting of all or part of a plurality of divided images obtained by dividing an image of a prescribed scrap material into a prescribed image size;
a chemical component output step for each of the divided images included in the image group, using the divided image as an input image and outputting a chemical component corresponding to the predetermined scrap material in the input image using a learned model; ,
A component estimation method having 2. The component estimation method according to claim 1, wherein the predetermined image size is 2000 mm or less on all sides in real scale. In the chemical component output step, for each divided image included in the image group, the divided image is used as an input image, and the learned model outputs product category information corresponding to the predetermined scrap material in the input image. The component estimation according to claim 1, further comprising a step of outputting product category information, wherein the product category information obtained through the step of outputting product category information is input, and a chemical component corresponding to the input is outputted from a database. Method. 4. The component estimation method according to claim 3, wherein in the product category information output step, one product category is output as the product category information for the input image. 2. The component estimation method according to claim 1, further comprising a post-processing step of estimating the chemical components of the original image using the chemical components outputted for each input image by the chemical component output step. Estimating the chemical composition of the scrap material to be managed by the composition estimation method according to any one of claims 1 to 5,
If the chemical components estimated by the component estimation method do not satisfy the predetermined conditions, based on the estimated chemical components, a portion of the scrap material to be managed is removed, or the scrap material to be managed is removed. A scrap material management method that adds another scrap material to the. A part or all of the scrap material whose chemical composition has been estimated in advance by the composition estimation method according to any one of claims 1 to 5 is put into a melting furnace based on the chemical composition estimated by the composition estimation method. A method of operating a melting furnace comprising a step of supplying scrap material. A melting furnace operating method comprising a scrap material supply step of supplying a part or all of the scrap material to be managed by the scrap material management method according to claim 6 to the melting furnace. A component estimation program that causes a computer to execute the component estimation method according to any one of claims 1 to 5. A component estimation device for estimating the chemical components of scrap materials,
a pre-processing unit that creates an image group consisting of all or part of a plurality of divided images obtained by dividing an image of a prescribed scrap material into a prescribed image size;
a chemical component output unit that takes the divided images included in the image group as an input image and outputs a chemical component corresponding to the predetermined scrap material in the input image using a learned model;
A component estimation device having: A melting furnace facility comprising the component estimating device according to claim 10 and a melting furnace. A scrap yard comprising the component estimating device according to claim 10. A method for generating a trained model that uses an image of a scrap material as an input and outputs product category information of the scrap material, the method comprising:
Teacher data including a combination of a product category and a teacher image containing scrap materials corresponding to the product category, and size change, movement, rotation, partial cropping, or reversal of the teacher image by enlarging or reducing the teacher image, or A method for generating a trained model, which generates the trained model by learning using expanded training data generated by performing arbitrary combinations. A trained model generated by the trained model generation method according to claim 13.

Images