JP7372017B2 - Steel component learning device, steel component estimation device, steel type determination device, steel component learning method, steel component estimation method, steel type determination method, and program - Google Patents

Description

The present invention relates to a steel component learning device, a steel component estimating device, a steel type determining device, a steel component learning method, a steel component estimating method, a steel type determining method, and a program.

Patent Document 1 describes that a spark image is binarized for each pixel, the number of short lines and the number of bursts of sparks are calculated by template matching, and the result is called "burst density" (=number of bursts/number of short lines). A technique for determining the amount of carbon in steel materials based on evaluation criteria has been disclosed.

Patent Document 2 discloses a technology that processes a spark image to convert it into a rupture image, and then extracts and analyzes the feature amounts included in the rupture image to inspect the content of a predetermined element in a steel material. Disclosed.

JP2016-99126A Japanese Patent Application Publication No. 9-145599

However, the template matching used in Patent Document 1 is susceptible to disturbance noise such as changes in illuminance in the photographing environment and vibrations, and the accuracy of determining the amount of carbon may not be sufficient.

In addition, with the technology disclosed in Patent Document 2, it is necessary to perform image conversion on the spark image in advance so that features corresponding to rupture can be extracted, so alloy components that do not have characteristics in the rupture image may be overlooked. There is a risk of it getting lost.

The present invention has been made in view of the above problems, and its main purpose is to provide a steel component learning device, a steel component estimation device, a steel type determination device, and a steel component learning device that can improve the estimation accuracy of steel components. The object of the present invention is to provide a method, a steel material composition estimation method, a steel type determination method, and a program.

In order to solve the above-mentioned problems, a steel material composition learning device according to one aspect of the present invention includes an acquisition means for acquiring image data of sparks generated when a steel material is ground and composition information of the steel material, and inputting the image data. and a learning means for constructing a trained model for estimating the components of steel from image data by machine learning, using the component information as training data.

Further, a steel component estimating device according to another aspect of the present invention includes an acquisition unit that acquires image data of sparks generated when a steel material is ground, and a machine that uses the image data as input data and the component information of the steel material as training data. and an inference means for estimating the components of the steel material from the image data using a learned model built in advance through learning.

Further, a steel type determination device according to another aspect of the present invention includes an input receiving unit that receives input of steel type information or component information of a steel material, and information inputted to the input receiving portion and estimated by the steel material component estimating device. and a steel type determining section that determines the steel type based on the components of the steel material.

Further, a steel component learning method according to another aspect of the present invention acquires image data of sparks generated when a steel material is ground and component information of the steel material, uses the image data as input data, and uses the component information as a teacher. As data, a trained model for estimating the composition of steel material from image data is constructed using machine learning.

In addition, a steel component estimation method according to another aspect of the present invention acquires image data of sparks generated when a steel material is ground, uses the image data as input data, and uses composition information of the steel material as training data to perform pre-preparation by machine learning. The components of the steel material are estimated from the image data using the constructed learned model.

Further, a steel type determination method according to another aspect of the present invention accepts input of steel type information or component information of a steel material, and determines the steel type based on the input information and the steel component estimated by the above-mentioned steel material component estimation method. Determine.

Further, a program according to another aspect of the present invention includes an acquisition unit that acquires image data of sparks generated when a steel material is ground and component information of the steel material; The computer functions as a learning means that uses machine learning to construct a trained model for estimating the composition of steel from image data as training data.

Further, a program according to another aspect of the present invention includes an acquisition means for acquiring image data of sparks generated when a steel material is ground, and a machine learning method using the image data as input data and composition information of the steel material as training data. A computer is made to function as an inference means for estimating the composition of the steel material from the image data using a learned model built in advance.

Further, a program according to another aspect of the present invention includes an input receiving section that receives input of steel type information or composition information of steel materials, and an inference means that functions based on the information input to the input receiving section and the above program. The computer functions as a steel type determination section that determines the steel type based on the steel composition.

According to the present invention, it is possible to improve the estimation accuracy of steel material components.

FIG. 2 is a block diagram showing a configuration example of a steel material component learning estimation device. FIG. 3 is a diagram showing an example of a spark image. FIG. 3 is a diagram showing an example of correspondence between spark images and component information. FIG. 3 is a flow diagram showing an example of a procedure of a learning phase. FIG. 3 is a diagram for explaining an example of building a trained model. FIG. 3 is a flow diagram showing an example of a procedure of an inference phase. FIG. 3 is a diagram showing an example of correspondence between steel types and component information. It is a figure showing an example of arrangement of cameras. FIG. 3 is a diagram for explaining an example of building a learned model. FIG. 3 is a diagram for explaining an example of building a learned model. FIG. 3 is a diagram for explaining a three-dimensional convolution filter. FIG. 3 is a diagram showing an example of a mask image. FIG. 2 is a block diagram showing a configuration example of a steel material component learning estimation device. FIG. 3 is a diagram showing an example of correspondence between spark images and steel types. FIG. 3 is a flow diagram showing an example of a procedure of an inference phase. FIG. 3 is a diagram for explaining an example of a method for determining alloying elements. FIG. 3 is a diagram for explaining an example of a method for determining alloying elements. FIG. 3 is a diagram for explaining an example of a method for determining alloying elements.

Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. It should be noted that each embodiment shown below exemplifies a method and apparatus for embodying the technical idea of the present invention, and the technical idea of the present invention is not limited to the following. do not have. The technical idea of the present invention can be modified in various ways within the technical scope described in the claims.

FIG. 1 is a block diagram showing a configuration example of a steel component learning estimation device 1. As shown in FIG. The steel component learning estimation device 1 is one embodiment of a steel component learning device, one embodiment of a steel component estimation device, and one embodiment of a steel type determination device.

The steel component learning estimation device 1 is a computer including a control unit 10 including a CPU, RAM, ROM, nonvolatile memory, input/output interface, and the like. The control section 10 includes an acquisition section 11 , a learning section 13 , an inference section 15 , and a steel type determination section 17 . These functional units are realized by the CPU of the control unit 10 executing information processing according to a program loaded into the RAM from the ROM or nonvolatile memory.

The program may be supplied, for example, via an information storage medium such as an optical disc or a memory card, or may be supplied, for example, via a communication network such as the Internet.

The control unit 10 can access the database 2. The database 2 may be provided inside the steel component learning estimation device 1 or may be provided outside.

The camera 3 photographs sparks generated when the steel material S is ground by the grinder G, generates image data, and outputs the image data to the control unit 10. The camera 3 is arranged so as to look down on the steel material S from above and face the direction of seeing the sparks, and its focal length and depth of field are adjusted so that the entire sparks are captured. The method for generating sparks by grinding the steel material S is based on, for example, the method specified in the spark test (JIS G 0566).

The image data generated by the camera 3 is used as input data in a learning phase by a learning section 13 and an inference phase by an inference section 15, which will be described later. Hereinafter, image data generated by photographing sparks will be referred to as a "spark image."

FIG. 2 is a diagram showing an example of the spark image 4. As shown in FIG. The characteristics (for example, shape, amount, color, etc.) of the sparks generated when the steel material S is ground, such as streamlines and ruptures at the base, middle, and tip, change depending on the components of the steel material S. For example, the higher the amount of carbon in the steel, the faster the spark will explode and the more streamlines there will be.

The spark image 4 output from the camera 3 to the control unit 10 is stored in the database 2. Further, a table as shown in FIG. 3 in which the identification information of the spark image 4 is associated with the component information of the steel material S is also stored in the database 2. The component information of the steel material S is represented by, for example, the amount of carbon or alloy element contained in the steel material S. The component information of the steel material S is used as teacher data in a learning phase by the learning section 13, which will be described later.

If the pixel size of the spark image 4 is large, it will take time to learn, so it is preferable to reduce the size to an extent that no information is lost (for example, about 300 pixels vertically by 100 pixels horizontally). Further, it is preferable to prepare a large number of spark images 4 by changing grinding conditions such as the strength and position of pressing the steel material S against the grinder G, and environmental conditions such as temperature and humidity. Further, as will be described later, a mask image in which a partial mask is set may be added to the spark image 4.

[Learning phase]
FIG. 4 is a flow diagram showing an example of the procedure of the learning phase. The control unit 10 functions as an acquisition unit 11 and a learning unit 13 by executing the information processing shown in the figure according to a program.

First, the control unit 10 acquires the spark image 4 and the component information of the steel material S (S11, processing as the acquisition unit 11). In the learning phase, the spark image 4 and the component information of the steel material S associated therewith are read from the database 2.

Next, the control unit 10 extracts a portion of the plurality of prepared spark images 4 and component information sets as training data for machine learning (S12), and executes machine learning using the extracted training data (S12). S13). Machine learning is performed using the spark image 4 as input data and the component information as teacher data. As a result, a learned model for estimating the components of the steel material S from the spark image 4 is constructed.

Next, the control unit 10 extracts a part other than the training data from the plurality of prepared spark images 4 and component information sets as test data (S14), and uses the extracted test data to create a learned model. Evaluate (S15). After that, the control unit 10 stores the trained models whose evaluation is equal to or higher than a predetermined value in the database 2 (S16), and ends the learning phase.

FIG. 5 is a diagram for explaining an example of building a trained model. The trained model is a convolutional neural network and includes a convolution filter, a pooling layer, a fully connected layer, and an output layer. In particular, a deep neural network in which neurons are combined in multiple stages is suitable. If the spark image 4 is a color image of, for example, 300 pixels in the vertical direction and 100 pixels in the horizontal direction, the input will be 90,000 elements.

When estimating the carbon content of the steel material S, the output layer of the convolutional neural network becomes one element. That is, a convolutional neural network is used for regression of the carbon content of the steel material S. For example, SC steel, which is a type of carbon steel, is often produced with a carbon content of about 0.1 to 0.6%, so various SC steels with a carbon content of 0.1 to 0.6% are produced. By preparing a spark image 4 of steel and performing learning, a trained model for estimating the amount of carbon from the spark image 4 is constructed.

When estimating the alloying elements (N types) contained in the steel material S, the output layer of the convolutional neural network becomes N elements. That is, a convolutional neural network is used to classify alloy elements contained in the steel material S. For example, by preparing a spark image 4 of molybdenum steel or chromium steel, which are steel materials containing alloying elements, and training it, a trained model that estimates alloying elements such as molybdenum and chromium contained in the steel material from the spark image 4 can be created. Constructed.

A convolutional neural network that regresses the carbon content of steel material S and a convolutional neural network that classifies the alloying elements of steel material S may be constructed separately, or a convolutional neural network that performs regression of the carbon content of steel material S and a convolutional neural network that performs classification of alloying elements of steel material S may be constructed separately. A convolutional neural network may be constructed that simultaneously performs classification.

Note that the convolutional neural network shown in FIG. 5 is just an example, and the layer structure is not limited to this, and there may be a plurality of convolution layers, pooling layers, and fully connected layers. Furthermore, machine learning other than neural networks such as support vector machines and decision trees may be used.

[Inference phase]
FIG. 6 is a flow diagram illustrating an example of the procedure of the inference phase. The control unit 10 functions as an acquisition unit 11, an inference unit 15, and a steel type determination unit 17 by executing the information processing shown in the figure according to a program.

First, the control unit 10 acquires the spark image 4 (S21, processing as the acquisition unit 11). In the inference phase, the spark image 4 acquired by the acquisition unit 11 from the camera 3 is directly input to the inference unit 15. The present invention is not limited to this, and the spark image 4 may be temporarily stored in the database 2 and then read from the database 2.

Next, the control unit 10 uses the acquired spark image 4 as input data and estimates the components of the steel material S using the learned model created in the learning phase (S22, processing as the inference unit 15).

Next, the control unit 10 determines the steel type of the steel material S based on the estimated components of the steel material S (S23, processing as the steel type determination unit 17). The steel type is determined by averaging the estimation results over several seconds, for example. The database 2 stores a table as shown in FIG. 7 in which steel types and component information are associated with each other, and the control unit 10 reads out the steel type corresponding to the estimated composition of the steel material S from the table.

For example, when inspecting SC steel, which is a type of carbon steel, by estimating the carbon content of the steel material S from the spark image 4, it is possible to determine the steel type, such as S10C or S35C. For example, the control unit 10 receives input of upper standard information (for example, SC steel) of steel type standards such as JIS (processing as an input reception unit), and determines the steel types that belong to the lower level (for example, S10C, S35C, etc.). You may.

Furthermore, when inspecting the metals contained in the steel material, by estimating the alloy elements contained in the steel material S from the spark image 4, the type of steel, such as molybdenum steel or chromium steel, can be determined, for example. For example, the control unit 10 may receive information indicating that the steel material contains an alloying element as component information, and determine the type of steel, such as molybdenum steel or chromium steel, based on the estimation result of the alloying element.

Furthermore, by combining the estimation of carbon content and the estimation of alloying elements, the steel type can be determined even if the steel material S is unknown.

According to the embodiment described above, the estimation accuracy can be improved by estimating the components of the steel material S from the spark image 4 using a learned model created by machine learning. Further, by repeating machine learning using the spark image 4 and component information acquired for each test, it is possible to further improve the estimation accuracy.

Further, according to the embodiment, it is possible to estimate the components of the steel material S from the spark image 4 without performing the image processing, image conversion, etc. disclosed in Patent Document 1 and Patent Document 2.

Furthermore, it becomes possible to replace the sensory tests that have hitherto been dependent on the skill level of the inspector with the highly accurate and quantitative component estimation of the embodiment. Further, according to the embodiment, it is possible to maintain component estimation with high accuracy even if the grinding conditions and environmental conditions change.

Further, according to the embodiment, rapid inspection and determination can be performed on a factory line where inspection for foreign material contamination and steel type determination are required in a short time. Further, according to the embodiment, it is possible to easily determine the carbon content of steel materials, which is difficult to do with fluorescent X-ray analysis.

[Multiple cameras]
8 and 9 are diagrams showing a modification using a plurality of cameras 3, 31, 32. In this modification, a plurality of spark images 4, 41, 42 obtained by photographing sparks from a plurality of different directions using a plurality of cameras 3, 31, 32 are input data to a convolutional neural network.

In particular, since the sparks generated when grinding the steel material S have characteristics at the root, center, and tip, in addition to the spark image 4 obtained by photographing the entire spark, By using one or more partial images 41 and 42 obtained by photographing the tip or the tip, it is possible to further improve the accuracy of component estimation.

In the illustrated example, in addition to a spark image 4 obtained by photographing the entire spark with the camera 3, a partial image 41 obtained by photographing the middle part of the spark with the camera 31, and a partial image 41 obtained by photographing the middle part of the spark with the camera 32. A partial image 42 obtained by photographing is used. In the partial images 41 and 42, the portion of interest is enlarged more than in the spark image 4.

Note that the present invention is not limited to this, and the spark image 4 may be omitted and two or more partial images 41 and 42 obtained by photographing the root, middle, or tip of the spark may be used.

[Time series images]
FIG. 10 is a diagram showing a modification using a plurality of time-series spark images 4. In this modification, a plurality of time-series spark images 4 obtained by photographing sparks with the camera 3 are input data to the convolutional neural network. The plurality of time-series spark images 4 may be, for example, a plurality of still images included in video data, or may be a plurality of still images individually generated by capturing at predetermined time intervals.

Since the state of sparks changes over time, by using a plurality of time-series spark images 4 as input data, the state of sparks that changes over time is better than when using only one spark image 4 as input data. Features become easier to recognize, making it possible to further improve the accuracy of component estimation.

In particular, as shown in FIG. 11, in addition to the two-dimensional vertical and horizontal axes of the spark image 4, a three-dimensional convolution filter may be used that has the time axis of the plurality of time-series spark images 4 as the third dimension. This makes it easier to recognize the characteristics of sparks that change over time, making it possible to further improve the accuracy of component estimation.

[Mask image]
FIG. 12 is a diagram showing a modified example using mask images 4a to 4c. In this modification, mask images 4a to 4c, which are partially masked spark images 4, may be used as one of the spark images 4 used as input data for the convolutional neural network.

By randomly setting a mask on the spark image 4, the points of interest of the convolutional neural network can be distributed, so that it is possible to further improve the precision of component estimation and the generalization performance.

In particular, since the sparks generated when grinding the steel material S have characteristics at the root, center, and tip, mask image 4a has a mask set at the base of the spark, and a mask is set at the middle part of the spark. By creating the mask image 4b or the mask image 4c in which a mask is set at the tip of the spark, it becomes easier to distribute the points of interest of the convolutional neural network to the root, center, and tip.

Only the mask images 4a, 4b, or 4c may be created, or when creating a mask image with randomly set masks, the ratio of the mask images 4a, 4b, or 4c may be greater than the others.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above embodiments, and it goes without saying that various modifications can be made by those skilled in the art.

In the above embodiment, the steel material component learning estimation device 1 includes both the configuration related to the learning phase and the configuration related to the inference phase, but is not limited to this. It may also be realized by a separate device.

Further, in the above embodiment, the learned model that estimates the components of the steel material S from the spark image 4 is used to determine the steel type from the estimated components of the steel material S, but the invention is not limited to this. You may also construct a trained model that directly estimates .

[Other embodiments]
Other embodiments will be described below. For configurations and procedures that overlap with those of the above embodiment, detailed explanations may be omitted by assigning the same numbers.

FIG. 13 is a block diagram showing a configuration example of a steel component learning estimation device 1B according to another embodiment. The control unit 10 includes an acquisition unit 11 , a learning unit 13 , an inference unit 15 , a steel type determination unit 17 , a preliminary inference unit 18 , and a model selection unit 19 .

Steel materials (steel types) can be grouped into a plurality of steel material groups according to carbon content. For example, a low carbon content group with a carbon content of 0.3% or less, a medium carbon content group with a carbon content of 0.3% or more and 0.7% or less, and a high carbon content group with a carbon content of 0.7% or more. Can be grouped.

Each of the plurality of steel material groups grouped in this way includes various steel types. For example, each of the low carbon content group, medium carbon content group, and high carbon content group includes steel types such as sulfur, molybdenum, chromium, or nickel in addition to carbon steel.

As shown in FIG. 13, in the present embodiment, the control unit 10B (learning unit 13) uses a learned model for each steel material group (a trained model for the low carbon content group, a trained model for the medium carbon content group, and a trained model for the medium carbon content group) in the learning phase. and a trained model for the high carbon content group).

The trained model for the low carbon content group is constructed based on the steel materials belonging to the low carbon content group, and is applied to determine the alloying elements of the steel materials belonging to the low carbon content group. The same applies to the trained model for the medium carbon content group and the trained model for the high carbon content group.

In addition, the control unit 10B (learning unit 13) also constructs a learned model for carbon amount estimation in the learning phase. This trained model for carbon content estimation is a trained model for estimating the carbon content of steel materials as described above, and detailed description thereof will be omitted.

As shown in FIG. 14, a table that associates spark images with steel types (or alloy elements) is prepared for each steel material group. In the learning phase, the spark images are used as input data and the alloy elements are used as training data. Note that the steel type may be used as training data.

The illustrated example shows a correspondence example between a spark image related to a steel material belonging to a low carbon content group and a steel type representing an alloy element contained in the steel material. That is, the spark image included in the low carbon content group table is image data generated by photographing sparks generated when steel material belonging to the low carbon content group is ground.

The control unit 10B (learning unit 13) uses the data set including the spark image and steel type (or alloy element) prepared for each steel material group in this way to perform the processing of the learning phase shown in FIG. 4 for each steel material group. Build a trained model for each steel group by running

FIG. 15 is a flow diagram illustrating an example of the procedure of the inference phase. The control unit 10B functions as an acquisition unit 11, an inference unit 15, a steel type determination unit 17, a preliminary inference unit 18, and a model selection unit 19 by executing the information processing shown in the figure according to a program.

When the control unit 10B acquires the spark image (S21, processing as the acquisition unit 11), the control unit 10B estimates the carbon content of the steel material from the acquired spark image using the trained model for carbon content estimation (S25, preliminary inference unit 18).

Next, the control unit 10B selects a learned model for one steel group from the learned models for a plurality of steel groups based on the estimated carbon content of the steel material (S26, as the model selection unit 19). process).

Specifically, the control unit 10B controls whether the estimated carbon content of the steel material is in a low carbon content group (carbon content range: 0.3% or less), a medium carbon content group (carbon content range: 0.3% or more, 0. .7% or less) and high carbon content group (carbon content range: 0.7% or more), as well as a trained model for the low carbon content group and a trained model for the medium carbon content group. , and the trained model for the high carbon content group.

Next, the control unit 10B uses the learned model for the selected steel group to estimate the alloy element contained in the steel from the spark image acquired in S21 (S27, processing as the inference unit 15).

For example, when the learned model for the low carbon content group is selected, the control unit 10B uses the learned model for the low carbon content group to estimate the alloy element contained in the steel material from the spark image. The same applies to the trained model for the medium carbon content group and the trained model for the high carbon content group.

Next, the control unit 10B determines the steel type of the steel material based on the estimated alloying element (S23, processing as the steel type determining unit 17). With the above, the inference phase ends.

According to the embodiment described above, alloy elements contained in steel materials are estimated using trained models for steel material groups according to the carbon content of steel materials, so it is possible to improve the estimation accuracy of alloy elements. becomes.

In addition, by estimating the carbon content of steel materials using a trained model for carbon content estimation and selecting a trained model for a steel material group based on the estimation results, even if the carbon content of steel materials is unknown, the carbon content of steel materials can be estimated. It becomes possible to estimate the carbon content and further estimate the alloying elements contained in the steel material.

Note that the present invention is not limited to this, and a trained model for a steel material group may be selected according to a user's operation.

[Alloy element determination]
As described below, the alloying element contained in the steel material may be determined based on a plurality of estimation results estimated by a learned model from each of a plurality of time-series spark images.

Because the state of the spark changes over time, characteristics that allow identification of the alloying element do not always appear. For example, with SCM steel (molybdenum), features appear at the tip of the spark, but they do not always appear, so the features may not be captured with just a single spark image, and it may be determined that it is just SC steel. .

FIG. 16 is a diagram showing an example of estimation results for 30 spark images of SCR steel (chromium). According to this, although most of the results are estimated to be chromium, about half of them are estimated to be carbon steel, and there are also a few estimated results to be molybdenum, nickel, and sulfur.

Therefore, in this modification, alloy elements are determined by majority vote based on a plurality of estimation results estimated by a trained model from each of a plurality of time-series spark images. This makes it possible to capture characteristics of sparks that change over time and improve the accuracy of determining alloy elements.

Specifically, multiple time-series spark images are sequentially input into a trained model, which sequentially counts the estimated alloying elements (estimation results), and the alloying element that reaches the determination threshold earliest is adopted. do.

FIG. 17 is a diagram showing an example of estimation results for SNCM steel (nickel). FIG. 18 is a diagram showing an example of estimation results for SCR steel (chromium). The horizontal axis represents the shooting time of spark images (corresponding to the number of images input to the trained model), and the vertical axis represents the cumulative number of images to be determined.

According to this, in FIG. 17, nickel reaches the determination threshold earliest, and in FIG. 18, chromium reaches the determination threshold earliest. In the illustrated example, the determination threshold is 30 sheets. It is preferable to set a determination threshold that can be determined by taking an image for about 1 to 2 seconds.

However, the present invention is not limited to this, and the alloy element that is the most common among the alloy elements (estimation results) estimated by the learned model from each of a plurality of spark images taken in a predetermined period may be adopted.

In addition, alloy elements may be determined using machine learning methods such as the k-nearest neighbor method and SVM (support vector machine).

According to the modification described above, it is possible to improve the accuracy of determining alloy elements by capturing characteristics of sparks that change over time using a plurality of time-series spark images. Furthermore, it is possible to improve the accuracy of determining steel materials with a low content of alloying elements and steel types with similar spark characteristics.

1 Steel component learning estimation device (steel component learning device, steel component estimation device), 2 Database, 10 Control unit, 11 Acquisition unit, 13 Learning unit, 15 Inference unit, 17 Steel type determination unit, 3, 31, 32 Camera, 4 Spark image, 41, 42 partial image, 4a-4c mask image

Claims (7)

an acquisition means for acquiring image data of sparks generated when grinding a steel material;
an inference means for estimating the components of the steel material from the image data using a trained model built in advance by machine learning using image data as input data and component information of the steel material as teacher data;
Equipped with
For each of a plurality of steel material groups grouped according to carbon content, a trained model for one steel material group is selected from the plurality of trained models constructed using alloy elements contained in the steel material as training data. further equipped with selection means,
The inference means estimates an alloy element contained in the steel material from the image data acquired by the acquisition means using the learned model for the selected one steel group,
The carbon content of the steel material is estimated from the image data acquired by the acquisition means using the image data as input data and the carbon content of the steel material as training data using a trained model for carbon content estimation built in advance by machine learning. further comprising a preliminary inference means;
The selection means selects the learned model for the first steel group based on the carbon content of the steel estimated by the preliminary inference means ,
Further comprising a determining means for determining an alloy element contained in the steel material based on a plurality of estimation results estimated by the learned model from each of a plurality of time-series image data obtained by photographing the sparks,
Steel component estimation device. an input reception unit that receives input of steel type information or composition information of steel materials;
a steel type determining unit that determines the steel type based on the information input to the input receiving unit and the steel component estimated by the steel component estimating device according to claim 1 ;
A steel type determination device. The steel type information is upper standard information of the steel type standard,
The steel type determination unit determines a steel type belonging to a lower level of the upper standard information.
The steel type determination device according to claim 2 . Obtain image data of sparks generated when grinding steel materials,
Estimating the components of the steel material from the image data using a trained model built in advance by machine learning with the image data as input data and the component information of the steel material as teacher data;
A method for estimating steel material components,
Furthermore, for each of a plurality of steel groups divided according to carbon content, a trained model for one steel group is selected from the plurality of trained models constructed using alloy elements contained in the steel as training data. choose,
In estimating the components of the steel material, using the learned model for the selected one steel material group, estimating alloy elements contained in the steel material from the acquired image data,
Furthermore, estimating the carbon content of the steel material from the acquired image data using the image data as input data and a trained model for carbon content estimation built in advance by machine learning using the carbon content of the steel material as teaching data,
In the selection of the learned model, the learned model for the first steel group is selected based on the estimated carbon content of the steel material ,
Further, determining an alloy element contained in the steel material based on a plurality of estimation results estimated by the learned model from each of a plurality of time-series image data obtained by photographing the sparks,
Steel composition estimation method. Accepts input of steel type information or composition information of steel materials,
Determining the steel type based on the input information and the steel composition estimated by the steel composition estimation method according to claim 4 .
Steel type determination method. Acquisition means for acquiring image data of sparks generated when grinding a steel material, and
Inference means for estimating the composition of the steel material from the image data using a trained model built in advance by machine learning using image data as input data and composition information of the steel material as teacher data;
make the computer function as
For each of a plurality of steel material groups grouped according to carbon content, a trained model for one steel material group is selected from the plurality of trained models constructed using alloy elements contained in the steel material as training data. Making the computer more functional as a means of selection,
The inference means estimates an alloy element contained in the steel material from the image data acquired by the acquisition means using the learned model for the selected one steel group,
The carbon content of the steel material is estimated from the image data acquired by the acquisition means using the image data as input data and the carbon content of the steel material as training data using a trained model for carbon content estimation built in advance by machine learning. make the computer further function as a means of preliminary inference;
The selection means selects the trained model for the first steel group based on the carbon content of the steel estimated by the preliminary inference means ,
further comprising a computer as a determination means for determining an alloy element contained in the steel material based on a plurality of estimation results estimated by the learned model from each of a plurality of time-series image data obtained by photographing the sparks; make it work,
program. an input reception unit that receives input of steel type information or composition information of steel materials, and
a steel type determining unit that determines the steel type based on the information input to the input receiving unit and the composition of the steel material estimated by the inference means functioning according to the program according to claim 6 ;
A program that allows a computer to function as a computer.

Images