JP2022124231A - Slag outflow determination method, converter operation method, molten steel production method, slag outflow determination device, converter operation equipment, and molten steel production equipment - Google Patents

Abstract

The present invention improves the accuracy of determining slag contamination while reducing the maintenance load by using captured infrared images during converter operation. [Solution] A discharge stream 1A discharged from a tapping port 3a of a converter 3 toward a ladle 4 is imaged with an infrared camera 15, and based on the captured image, presence or absence of slag mixing in the discharge stream 1A is determined. The apparatus includes an image acquisition unit 20A that acquires an image of the discharge stream 1A taken by the infrared camera 15, and a machine learning model using the infrared image of the discharge stream. and a determination section 20B that determines whether or not slag is mixed in from the captured image acquired by the image acquisition section 20A using the learned model 50 that recognizes whether or not slag is mixed in. It is preferable that the captured image is an image that includes a range from the position of the tapping port 3a to the position directly above the ladle 4 in the discharge stream 1A. [Selection diagram] Figure 1

Description

The present invention uses a slag outflow determination technique for detecting slag mixing in a discharge flow (molten steel flow) discharged from a converter when discharging molten steel from the converter to a ladle, and the technique. It is a technology related to converter operation and molten steel production.
Here, in this specification, machine learning includes deep learning.

Molten steel refined in the converter is discharged from the molten steel mouth of the converter toward the ladle by tilting the converter. When the timing of ending the tapping to the ladle (stopping tapping) is late, the slag in the converter is discharged from the tapping port and flows into the ladle. When the amount of slag flowing out to the ladle is large, impurities such as P in the slag return to the molten steel, affecting the processing time of the next step and the auxiliary materials used. In addition, since the injected alloy is trapped in the slag, it also causes a decrease in the yield of the alloy. On the other hand, if the tapped steel to the ladle is rounded up too early, a large amount of steel remains in the converter, which also leads to a decrease in yield. Therefore, it is important to determine whether slag is mixed in the discharge stream discharged from the converter.

Conventionally, when a skilled worker observes the color of the discharge stream discharged from the converter and determines that slag is mixed in the discharge stream, the tapping is stopped. However, there were problems such as the skill in judgment and the limited position where the discharge flow can be seen well.

As a conventional technique for solving this problem, for example, there is a determination method using an infrared method (see Patent Document 1).

The infrared method is a method in which the discharge flow (molten steel flow) discharged from the converter is imaged with an infrared camera and the presence or absence of slag contamination is determined based on the captured image. Specifically, this method is based on the energy difference between the radiant energy of the molten steel and the radiant energy of the slag in the discharge stream, which is measured by the infrared camera, by photographing the discharge stream from the converter with an infrared camera. , the presence or absence of slag in the discharge stream is determined based on the difference in brightness between slag and molten steel. Then, when it is determined that slag is mixed in, the discharge of molten steel from the converter (steel tapping) is stopped.

Japanese Patent No. 5444692

Slag outflow determination by the infrared method is based mainly on the difference in radiant energy (appearing as a difference in brightness) and the determination of slag with a certain threshold, or based on the change in the average value of measured energy values up to the determination point (threshold). A method of determination is employed. However, whichever method is adopted, in order to avoid erroneous determination or to deal with differences between steel grades, the determination threshold must be set with a certain amount of leeway. For this reason, there is a problem that the judgment timing is delayed accordingly.

In addition, in this infrared method, the influence of dust caused by the discharge of molten steel is suppressed and the determination is stable. It is necessary to make a determination based on an infrared image that captures the lower position (for example, about one-third from the bottom). However, at the lower part of the discharged steel stream, if the discharged stream diverges or narrows, the ratio of slag pixels included in the infrared image decreases, so there are cases where slag inclusion is not detected. The inventor found out that there is a problem. In addition, there is also a problem that the detection of slag mixing is delayed because the determination is made at the discharge flow part.

Furthermore, in the infrared method, many settings such as a plurality of initial values and threshold values (for example, setting of 14 set values) are required for determination. Therefore, it is necessary to adjust the plurality of set values according to the operating environment (steel type, furnace conditions, etc.) that changes over time. However, this adjustment of the setting value requires a large amount of test work because it is an event that cannot be obtained by calculation.

The present invention has been made with a focus on the above points, and aims to improve the accuracy of determining slag contamination while suppressing the maintenance load by using the captured infrared image in the operation of the converter. .

The inventor used an infrared camera to photograph the tapped stream (exhaust stream) flowing out of the tapping port of the converter, and separated the obtained infrared image into an image with slag mixed in and an image without slag mixed in, and labeled them. It was found that slag contamination can be determined with high accuracy by using a trained model in which the training data is used as learning data (teacher data) and trained by CNN (convolutional neural network).

Furthermore, the inventor used Grad-CAM (gradient load class activation mapping ), it was found that the grounds for classification judgment extended to the upper part of the steel flow. They also found that by including the upper region as well as the conventional lower region as the region of the target image, it is possible to generate a model with further improved accuracy.

Then, in order to solve the problem, one aspect of the present invention is to capture an image of the discharge flow discharged from the tapping port of the converter toward the ladle with an infrared camera, and based on the captured image, the discharge flow A slag contamination determination method for determining the presence or absence of slag contamination in a machine, comprising: an image acquisition step of continuously acquiring captured images of the discharged flow captured by the infrared camera; and a determination step of determining presence or absence of slag contamination from the captured image acquired in the image acquisition step, using a learned model that recognizes the presence or absence of slag contamination. .

Further, according to one aspect of the present invention, an image of a discharge flow discharged from a tapping port of a converter toward a ladle is imaged with an infrared camera, and the presence or absence of slag mixing in the discharge flow is determined based on the captured image. A slag contamination determination device for judging slag contamination, comprising: an image acquisition unit that acquires an image of the discharge stream captured by the infrared camera; and a determination unit that determines presence/absence of slag mixture from the captured image acquired by the image acquisition unit using a trained model that recognizes presence/absence of slag.
In addition, it is preferable that the captured image is an image including a range from the position of the tapping port in the discharge flow to the position directly above the ladle.

ADVANTAGE OF THE INVENTION According to the aspect of this invention, in converter operation, it becomes possible using the imaged infrared image to perform determination of slag mixing accurately, suppressing a maintenance load.

BRIEF DESCRIPTION OF THE DRAWINGS It is a schematic diagram which shows the converter operating equipment which concerns on embodiment based on this invention. It is a figure which shows the structural example of a slag outflow determination apparatus. FIG. 4 is a diagram illustrating a configuration example of a trained model generation unit; FIG. 4 is a diagram for explaining an example of a captured image that constitutes unique data; FIG. 4 is a diagram for explaining the configuration of a learned model update unit; FIG. 4 is a flow diagram showing an example of processing for determining an image region;

Next, embodiments of the present invention will be described with reference to the drawings.
When the refining of the molten steel in the converter is completed, the ladle is moved to the lower part of the converter and the converter 3 is tilted as shown in FIG. Steel is tapped from the steel mouth 3a toward the ladle 4.

(Constitution)
In this embodiment, a slag stopper 10 is installed near the tapping port 3 a of the converter 3 . The slag stopper 10 includes a rotatable arm 11 , a cast iron stopper 12 attached to the tip of the arm 11 , and a hydraulic cylinder 13 for driving the arm 11 . The hydraulic cylinder 13 is controlled by a slag stopper control device 16, and by closing the tapping port 3a with the stopper portion 12, discharge of the molten steel 1 from the converter 3 is stopped.
Here, the discharge stream 1A discharged from the steel tapping port 3a is usually called "steel tapping stream".
Further, in this embodiment, an infrared camera 15 and a slag outflow determination device 20 are provided.

<Infrared camera 15>
The infrared camera 15 is a camera that captures an image of the discharge stream 1A that is discharged from the tapping port 3a toward the ladle 4. As shown in FIG.

In the conventional infrared method, as described above, of the discharge flow 1A in the range from the tapping port 3a to the height position of the upper end surface 4a of the ladle 4 (the position directly above the ladle 4), the lower three An area near one minute was imaged.

On the other hand, when we tried to visualize the grounds for judgment with Grad-CAM technology for the trained model 50 machine-learned by CNN, we found new knowledge that the upper region of the discharge stream 1A is emphasized. Obtained.

For this reason, in the present embodiment, the imaging range of the infrared camera 15 is set to the position of the discharge flow 1A directly above the ladle 4 receiving the discharge flow 1A from the position of the tapping port 3a (the height of the upper end 4a of the ladle 4). ) is set to include the range. The imaging range and the hot water surface in the ladle 4 may also be included. This is because the state of the hot water surface changes depending on the presence or absence of slag.

As described above, in the present embodiment, the infrared camera 15 continuously captures images of the discharge stream 1A located at least in the range directly above the ladle 4 from the tapping port 3a, and the captured images are captured by the slag outflow determination device 20. are sending to

<Slag outflow determination device 20>
The slag outflow determination device 20, as shown in FIG. 2, includes an image acquisition section 20A, a determination section 20B, and a transmission section 20C.

[Image Acquisition Unit 20A]
The image acquisition unit 20A performs a process of continuously acquiring captured images captured by the infrared camera 15 from the infrared camera 15 at a predetermined sampling period. Alternatively, the infrared camera 15 may take a moving image, and the image acquisition unit 20A may appropriately extract a still image from the moving image as the captured image.

A captured image obtained by clipping a part of the infrared image captured by the infrared camera 15 may be used. The image is set so as to capture the discharged flow 1A.

[Determination unit 20B]
The determination unit 20B uses the learned model 50 to perform processing for determining whether or not slag is mixed in from the captured image acquired by the image acquisition unit 20A.

The learned model 50 is a model machine-learned using an infrared image of the discharge flow, and is a model that recognizes the presence or absence of slag contamination. The trained model 50 uses captured images as input data, and determines (classifies) whether or not slag is present as output data.
The determination unit 20</b>B of the present embodiment stores the acquired captured image and the determination result using the captured image in the storage unit 21 .
Note that the storage unit 21 is composed of a database and the like, but does not have to be physically one medium.

[Transmitting unit 20C]
20 C of transmission parts perform the process which transmits the determination result of the determination part 20B. When the determination unit 20B determines that slag is mixed in, the transmission unit 20C of the present embodiment outputs a command to the slag stopper control device 16 to close the tapping port 3a. Upon receiving the closing command, the slag stopper control device 16 drives the hydraulic cylinder 13 to close the tapping port 3a.

Here, the process of stopping discharge from the tapping port 3a is not limited to the process using the slag stopper. For example, the transmission unit 20C outputs a notification command to a notification device such as a buzzer, and the operator stops the tilting of the furnace body of the converter 3 according to the notification command, thereby stopping the discharge of the molten steel 1. Also good.

<Generation of trained model 50>
Generation of the learned model 50 described above will be described.
The slag outflow determination device 20 has a learned model generator 22 as shown in FIG. The trained model generation unit 22 performs processing for generating a trained model 50 .
The trained model generation unit 22 includes a learning data storage unit 22A, a first learning unit 22B, a unique data acquisition unit 22C, and a second learning unit 22D.

[Learning data storage unit 22A]
The learning data storage unit 22A performs a process of storing in the storage unit 21 learning data 51 (teacher data) obtained by labeling the captured image captured by the infrared camera 15 as to whether or not slag is present.
The image pick-up range of this picked-up image is also assumed to be an area including the discharged flow 1A located in the range directly above the ladle 4 from the tapping port 3a.

For example, for the initial learning data 51, the entire discharge stream 1A is continuously imaged by the infrared camera 15, and an expert judges the presence or absence of slag contamination in the captured images to determine whether slag is present or not. classify. In addition, for the discharge stream 1A, the conventional infrared method is used for determination, and the captured image before determining that slag is present is labeled as not containing slag. label. After that, data may be generated by learning by correcting the judgment result by a person for data that is difficult to judge from before and after the start of slag outflow to complete slag, or for peculiar data such as branching.

It is assumed that the number of learning models used for learning is 3000, for example.

[First learning unit 22B]
The first learning unit 22B performs machine learning (deep learning) using the learning data 51 accumulated in the accumulation unit 21, and executes processing for obtaining a first learning model 52 that recognizes the presence or absence of slag contamination.

For example, the first learning unit 22B sequentially inputs the learning data 51 accumulated in the accumulation unit 21, and learns the input image with a CNN (convolutional neural network), so that the captured image is used as input data, and the slug 2 A trained model 50 is generated whose output data is the classification of presence/absence of mixture.

Here, the convolutional neural network (CNN) superimposes multiple convolution layers and pooling layers on the input image in multiple stages to extract the feature amount of the image, and in the final stage, the image is divided into categories and the output is constructed. It is a neural network that

[Specific data acquisition unit 22C]
The singular data acquisition unit 22C uses evaluation learning data 51 prepared in advance, uses the evaluation learning data 51 as input data, and uses the first learning model 52 obtained by the first learning unit 22B. The presence or absence of slag contamination is determined, and processing is performed to select and acquire specific data 53, which is an infrared image that has been erroneously determined by the determination and is labeled as to whether or not slag is included. The acquired unique data 53 is stored in the storage unit 21 .

The learning data 51 for evaluation is an infrared image prepared for evaluation and labeled as to whether or not slag is present.
A peculiar captured image that is likely to be erroneously determined by the conventional infrared method is used as a captured image for evaluation. For example, as shown in FIG. 4, an image of the branched discharge stream 1A and an image when the discharge stream 1A is narrowed are examples of peculiar captured images. Such peculiar captured images are used as learning data 51 for evaluation.
Alternatively, the learning data 51 used in the first learning section 22B may be used as the learning data 51 for evaluation.

[Second learning unit 22D]
The second learning unit 22D performs additional learning in CNN using the singular data 53 acquired by the singular data acquisition unit 22C for the first learning model 52 obtained by the first learning unit 22B, and adds The model after learning is stored in the accumulation unit 21 as a learned model 50 .

In this way, the accuracy of the trained model 50 is improved by performing additional learning using the erroneously determined learning data 51 . Note that the first learned model 52 obtained by the first learning unit 22B may be used as the learned model 50 .

<Learned model update unit 23>
This embodiment includes a learned model update unit 23 .
The learned model update unit 23 appropriately updates the learned model 50 according to accumulation of new learning data 51 .

Incidentally, as described above, the peculiar data 53 acquired by the peculiar data acquisition section 22C is in a state of being accumulated in the accumulation section 21. FIG.
The learned model update unit 23 includes a learning data update unit 23A and a re-learning unit 23B, as shown in FIG.

[Learning data update unit 23A]
The learning data update unit 23A generates learning data 51 from the captured image determined by the determination unit 20B and the determination result, stores the generated learning data 51 in the storage unit 21, and stores the learning data 51 in the storage unit 21. update the existing learning data 51; If the number of learning data stored in the storage unit 21 is limited to a finite range due to capacity restrictions, etc., the timing of storage may be changed as appropriate, for example, so that the number of data falls within a preset range. Old learning data may be deleted sequentially. Alternatively, other criteria may be defined to determine the data to be deleted. As a result, the learning data is updated according to the state of the converter 3 and the usage environment.

[Re-learning unit 23B]
The relearning unit 23B uses the learning data 51 of the storage unit 21 updated by the learning data updating unit 23A and the specific data 53 stored in the storage unit 21 to obtain a learning model that recognizes the presence or absence of slag contamination. , modifies the trained model 50 .

The re-learning unit 23B executes, for example, every predetermined period of time, such as every three months or half a year, or relearns a predetermined number of data (for example, 1/3 or more data) of the accumulated learning data 51 for new learning. When data 51 is obtained, a new trained model 50 is generated and the trained model 50 is updated.

The model generation processing executed by the re-learning unit 23B is basically the same as that of the first learning unit 22B, but differs in that the singular data 53 is included from the beginning as part of the learning data.

When a preset condition is satisfied, the learned model update unit 23 automatically acquires and updates a new learned model 50, thereby automating the update of the learned model 50. FIG. Also, by using the unique data 53 obtained in advance, the accuracy of the model is good.

<Regarding the Range of the Image Area Acquired by the Image Acquisition Unit 20A>
The area of the discharge stream 1A imaged by the infrared camera 15 is expanded compared to the conventional infrared method, and is set to include at least the range from the tapping port 3a to just above the ladle 4, and the basis for image prediction by CNN was investigated by Grad-CAM (gradient weighted class activation mapping). As a result, it was found that the basis for classification judgment extended to the upper part of the steel stream. Then, the inventors have found that by including the upper region as well as the conventional lower region as the target image region, it becomes possible to generate and use a model with further improved accuracy.

That is, in the present embodiment, Grad-CAM technology is used to determine the region of the captured image to be acquired in the image acquisition step through the steps shown in FIG.

First, in a learning model generation step S10, a learning model for recognizing the presence or absence of slag contamination is generated by learning by CNN using learning data 51 in which infrared images are labeled with slag contamination. For example, the learning data 51 is generated using 1000 learning data 51 . The process of generating may be executed by the same process as that of the first learning unit 22B, for example.

Next, in a convolutional layer influence evaluation step S20, influence coefficients of changes in each pixel of a plurality of convolutional layers of the generated learning model on the slug 2 determination output are calculated.

Next, in an input evaluation image generation step S30, the infrared image captured by the infrared camera 15 is used as an input image using the influence coefficient obtained in step S20 and a plurality of convolution layers used for calculating the influence coefficient. Evaluate the effect of each pixel in the input image on the Slug 2 decision output.

Then, from the input evaluation image created in the input evaluation image creation step S30, an infrared image region having a high degree of influence on the slag 2 determination output is extracted, and the image is acquired in the image acquisition step so as to include the extracted region. Determine the area of the captured image to be captured.

Here, Grad-CAM is an abbreviation for Gradient-weighted Class Activation Mapping, and is a technique for visualizing important pixels (pixels with large gradients) by weighting gradients for predicted values. Grad-CAM processing is performed, for example, as follows. The input image is input in the forward direction, and the convolutional layer and the classification result are taken out. Error backpropagation is then performed based on the classification results to form the gradients of the convolutional layers. Next, the GAP of the gradient of the convolutional layer is taken, the weighted sum of the convolutional layers weighted by the GAP is taken, the original image size is resized, a Grad-CAM image is obtained, and the basis of the image is visualized.

(Modification)
Here, the generation of the trained model 50 is not limited to the method described above.
For example, for a known model consisting of a neural network that has been trained for general events, a new layer is added to the final stage of the model by fine tuning technology, and re-learning for this judgment is performed. You may generate the trained model 50 adopted in .

For example, only four final layers (Flatten, Dense, Dropout, Dense) are added to an existing stable model and re-learned. Fine-tuning replaces the final output layer of the trained model 50 and changes the parameters of the part near the input layer and the output layer. This makes it possible to adapt to a new task by making it possible to adapt to a new task that does not exist in existing learning.

Here, the Flatten layer is the one-dimensionalization layer of the data for the fully connected layer input. A Dense (fully connected) layer is a layer that combines image data from which feature portions are extracted and outputs values (feature variables) transformed by an activation function. The dropout layer is a layer that prevents overfitting by randomly disconnecting some of the connections between the fully connected layer and the output layer.

(Other operations)
The slag outflow determination device 20 of the present embodiment can accurately determine slag contamination by determining the presence or absence of slag contamination with the trained model 50 learned by CNN for continuously captured infrared images. It becomes possible.

As a result, in the operation of the converter 3, it becomes possible to suppress the amount of the slag 2 mixed in the ladle 4, and it becomes possible to improve the alloy yield.

In addition, as a result of investigating the grounds for judgment by Grad-CAM technology, based on new knowledge that the range of grounds for classification judgment spreads to the upper part of the steel flow, in this embodiment, when judging slag contamination based on infrared images Secondly, the region of the infrared image (captured image) used for judgment, which reflects the discharged flow 1A, was made wider than before, and was set to include the area from the tapping port 3a to the upper part of the ladle 4. This further improves the determination accuracy of the trained model 50 . Here, in the learning model learned by CNN, the knowledge that importance is placed from the upper part of the steel output flow is a new knowledge.

In the infrared method, a judgment area is provided at the bottom of the tapping flow in order to judge where the slag 2 has fallen and the ratio of the slag 2 has increased, leading to a delay in judgment. On the other hand, in the present embodiment, the judgment is based on the condition immediately after coming out of the tapping port 3a (the upper part of the tapping flow), and it was confirmed that slag contamination can be judged early on, as with the expert. ing.

Further, when the trained model 50 is generated, additional learning is performed using a specific image (specific data 53) that is likely to be erroneously determined in the conventional infrared method, thereby further improving the determination accuracy of the trained model 50.

In addition, it has a system that regenerates the learned model 50 as appropriate from the newly acquired learning data 51 and the previously obtained specific data 53 (fixed learning data), and automatically updates the learned model 50. As a result, deterioration of the determination accuracy of the learned model 50 can be suppressed even if the steel type or operating conditions change over time.

That is, without adjusting the many set values necessary for the infrared method, the current operating conditions are automatically calculated by automatically using the CNN using the updated learning data 51 (teacher data). The system automatically learns the parameters you are interested in, which reduces the maintenance load.

(others)
The present disclosure can also take the following configurations.
(1) A slag contamination determination method for determining the presence or absence of slag contamination in the discharge flow based on the imaged image of the discharge flow discharged from the tapping port of the converter toward the ladle. An image acquisition step of continuously acquiring captured images of the discharge stream captured by the infrared camera; and a determination step of determining presence/absence of slag mixture from the captured image acquired in the image acquisition step, using a trained model to be recognized.

In this slag outflow determination method, for example, the discharge stream discharged from the tapping port of the converter toward the ladle is imaged with an infrared camera, and based on the captured image, the presence or absence of slag mixing in the discharge stream is determined. A slag contamination determination device for judging slag contamination, comprising: an image acquisition unit that acquires an image of the discharge stream captured by the infrared camera; and a determination unit that determines presence/absence of slag contamination from the captured image acquired by the image acquisition unit using a trained model that recognizes presence/absence of slag outflow determination device.
According to this configuration, in the operation of the converter, it is possible to accurately determine whether or not slag is present while suppressing the maintenance load by using the captured infrared image.

(2) The infrared image used for the learning and the captured image used for determining whether slag is present are images including the range from the tapping port position to the position directly above the ladle in the discharge flow.
For example, the captured image acquired by the image acquisition unit is an image including a range from the tapping port position to the position directly above the ladle in the discharge flow.

According to this configuration, the determination accuracy using the learned model is improved by including the upstream discharge flow position, which is not used in the conventional infrared method, in the region of the captured image.
In addition, by including the position of the discharged stream on the upstream side in the area of the captured image, the determination speed is improved compared to the conventional infrared method.

(3) Store learning data labeled with presence or absence of slag in the infrared image captured by the infrared camera in a storage unit, perform machine learning using the learning data stored in the storage unit, and determine whether slag is mixed. A first learning model for recognizing the presence or absence of slag is obtained, and the obtained first learning model is used for the infrared image prepared for evaluation and labeled for the presence or absence of slag contamination. Determine presence/absence of slag, acquire specific data consisting of an infrared image labeled as to presence/absence of slag contamination, which is an infrared image erroneously determined in the determination, and compare the obtained first learning model with the acquired peculiar data. Additional learning is performed using the unique data, and the learning model after the additional learning is defined as the learned model.

This method has, for example, a trained model generating unit that generates the trained model, and the trained model generating unit generates training data labeled with presence/absence of slag inclusion in the captured image captured by the infrared camera. a learning data accumulation unit that accumulates in the accumulation unit, a first learning unit that performs machine learning using the learning data accumulated in the accumulation unit and obtains a first learning model that recognizes the presence or absence of slag contamination; For the infrared image prepared for evaluation and labeled with the presence or absence of slag, the presence or absence of slag is determined using the first learning model obtained by the first learning unit, and the determination is performed. A specific data acquisition unit that selects specific data that is an infrared image that is erroneously determined in and is an infrared image labeled with the presence or absence of slag contamination, and for the first learning model obtained in the first learning unit, a second learning unit that performs additional learning using the unique data acquired by the unique data acquisition unit and updates the model after the additional learning as the learned model.
According to this configuration, it is possible to generate a highly accurate trained model.

(4) Store the acquired specific data in the storage unit in advance, and store the captured image for which the presence or absence of slag contamination is determined in the determination step as learning data in the storage unit 21 and store it in the storage unit. The stored learning data is updated, and the trained model for recognizing presence/absence of slag contamination is re-learned using the updated learning data in the storage unit and the specific data stored in the storage unit.

In this method, for example, the peculiar data acquired by the peculiar data acquisition unit is stored in advance in the storage unit, and learning data is generated from the captured image and the determination result determined by the determination unit, and the generation of learning data a learning data updating unit for storing the obtained learning data in the accumulating unit and updating the learning data accumulated in the accumulating unit; the learning data in the accumulating unit updated by the learning data updating unit; a slag outflow determination device comprising: .

According to this configuration, even if the condition of the state of the converter is changed, the learned model is automatically updated according to the condition. In addition, since the update learning is performed including the peculiar data for which the correct determination was initially not possible, highly accurate learning determination performance is maintained.

(5) The learning is performed using a convolutional neural network (CNN).
According to this configuration, it is possible to reliably generate a trained model.

(6) Recognizing the presence or absence of slag contamination by using learning data labeled with the presence or absence of slag contamination for the infrared image including at least the range from the tapping port position to the position directly above the ladle in the discharge stream. a learning model generation step for generating a learning model to be used as a learning model; and a convolutional layer impact evaluation for calculating an influence coefficient of a change in each pixel of a plurality of convolution layers in the learning model generated in the learning model generation step on a slag determination output. An infrared image captured by the infrared camera is used as an input image, and each pixel of the input image is used as a slag determination output, using the step, the influence coefficient, and a plurality of convolution layers used to calculate the influence coefficient. an input evaluation image creating step for evaluating the influence of the input evaluation image, extracting an infrared image region having a high degree of influence on the slag determination output from the input evaluation image created in the input evaluation image creating step; The region of the captured image to be acquired in the image acquisition step is determined so as to include the region.
According to this configuration, it is possible to set the optimum range of the imaging region that can improve the determination accuracy.

(6) A method of operating a converter, comprising operating a converter using the method for determining slag outflow described above.
According to this configuration, it is possible to provide a method of operating a converter with a good yield.
(7) A molten steel manufacturing method comprising the converter operating method described above.
According to this configuration, it is possible to provide a molten steel manufacturing method with a good yield.

(8) A converter operating facility comprising the slag outflow determination device described above.
According to this configuration, it is possible to provide a converter operation facility with a good yield.
(9) Molten steel production equipment comprising the slag outflow determination device described above.
According to this configuration, it is possible to provide a molten steel manufacturing facility with a good yield.

1 Molten steel 1A Discharge flow 2 Slag 3 Converter 3a Steel output port 4 Ladle 4a Upper end of ladle (directly above position)
15 Infrared camera 20 Slag outflow determination device 20A Image acquisition unit (image acquisition step)
20B Determination unit (determination step)
20C transmission unit 21 storage unit 22 trained model generation unit 22A learning data storage unit 22B first learning unit 22C specific data acquisition unit 22D second learning unit 23 trained model update unit 23A learning data update unit 23B re-learning unit 50 Trained model 51 Learning data 52 First learning model 53 Specific data

Claims (15)

A slag contamination determination method for determining the presence or absence of slag contamination in the discharge flow based on the imaged image, wherein the discharge flow discharged from the tapping port of the converter toward the ladle is imaged with an infrared camera. ,
an image acquisition step of continuously acquiring captured images of the discharged flow captured by the infrared camera;
Judgment to determine the presence or absence of slag contamination from the captured image acquired in the image acquisition step using a machine-learned model that recognizes the presence or absence of slag contamination, which is a model machine-learned using the infrared image of the discharge flow a step;
A method for determining slag outflow, comprising: The infrared image used for the learning and the captured image used for determining whether slag is mixed are images including a range from the tapping port position in the discharge flow to a position directly above the ladle. The slag outflow determination method according to claim 1. accumulating learning data labeled with presence or absence of slag in the accumulation unit for the infrared image captured by the infrared camera,
Machine learning is performed using the learning data accumulated in the accumulation unit to obtain a first learning model that recognizes the presence or absence of slag contamination,
The presence or absence of slag contamination is determined using the obtained first learning model for the infrared image prepared for evaluation and labeled as to whether or not slag is present, and an erroneous determination is made. obtaining singular data consisting of an infrared image labeled with the presence or absence of slag contamination,
Additional learning is performed on the obtained first learning model using the acquired singular data, and the learning model after additional learning is the learned model.
The slag outflow determination method according to claim 1 or 2, characterized in that: accumulating the acquired specific data in advance in the accumulation unit,
storing the captured image for which presence/absence of slag contamination is determined in the determination step in the accumulation unit 21 as learning data, and updating the learning data accumulated in the accumulation unit;
relearning the trained model for recognizing the presence or absence of slag contamination using the updated learning data of the storage unit and the unique data stored in the storage unit;
The slag outflow determination method according to claim 3, characterized in that: The slag outflow determination method according to any one of claims 1 to 4, wherein the learning is performed using a convolutional neural network (CNN). A learning model that recognizes the presence or absence of slag contamination by using learning data labeled with the presence or absence of slag contamination for an infrared image including at least the range from the steel tapping port position to the position directly above the ladle in the discharge stream. a learning model generation step that generates
A convolutional layer influence evaluation step of calculating influence coefficients of changes in each pixel of a plurality of convolutional layers of the learning model generated in the learning model generation step on a slag determination output;
Using the influence coefficient and the plurality of convolution layers used to calculate the influence coefficient, the infrared image captured by the infrared camera is used as an input image, and the influence of each pixel of the input image on the slag determination output is calculated. an input evaluation image creation step to be evaluated;
with
From the input evaluation image created in the input evaluation image creation step, an infrared image region having a high degree of influence on the slag determination output is extracted, and the captured image acquired in the image acquisition step so as to include the extracted region. determine the area of
The slag outflow determination method according to claim 5, characterized in that: A converter operating method, comprising operating a converter using the slag outflow determination method according to any one of claims 1 to 6. A method for producing molten steel, comprising the method for operating a converter according to claim 7 . A slag contamination determination device that captures an image of a discharge flow discharged from a steel tapping port of a converter toward a ladle with an infrared camera and determines whether or not slag is mixed in the discharge flow based on the captured image. ,
an image acquisition unit that acquires an image of the discharged flow captured by the infrared camera;
Using a machine-learned model that recognizes the presence or absence of slag contamination, the presence or absence of slag contamination is determined from the captured image acquired by the image acquisition unit. a determination unit;
A slag outflow determination device, comprising: 10. The slag outflow determination according to claim 9, wherein the captured image acquired by the image acquisition unit is an image including a range from the tapping port position to the position directly above the ladle in the discharge flow. Device. Having a trained model generation unit that generates the trained model,
The trained model generation unit is
a learning data accumulation unit that accumulates in the accumulation unit learning data labeled with the presence or absence of slag inclusion in the captured image captured by the infrared camera;
a first learning unit that performs machine learning using the learning data accumulated in the accumulation unit and obtains a first learning model that recognizes the presence or absence of slag contamination;
For the infrared image prepared for evaluation and labeled with the presence or absence of slag, the presence or absence of slag is determined using the first learning model obtained by the first learning unit, and the determination is performed. A specific data acquisition unit that selects specific data that is an infrared image that is erroneously determined in and is an infrared image labeled with the presence or absence of slag inclusion;
The first learning model obtained by the first learning unit is subjected to additional learning using the singular data acquired by the singular data acquisition unit, and the model after additional learning is updated as the learned model. 2 learning units;
The slag outflow determination device according to claim 9 or 10, comprising: The singular data acquired by the singular data acquisition unit is stored in advance in the storage unit,
Furthermore,
A learning data update unit that generates learning data from the captured image determined by the determination unit and the determination result, stores the generated learning data in the storage unit, and updates the learning data stored in the storage unit. When,
Using the learning data in the storage unit updated by the learning data update unit and the specific data stored in the storage unit, a learning model for recognizing the presence or absence of slag contamination is obtained, and the obtained learning model is used in the learning. a re-learning unit that updates as a finished model;
The slag outflow determination device according to claim 11, comprising: The slag outflow determination device according to any one of claims 9 to 12, wherein the learning is performed by a convolutional neural network (CNN). A converter operating facility comprising the slag outflow determination device according to any one of claims 9 to 13. Molten steel production equipment comprising the slag outflow determination device according to any one of claims 9 to 13.

Images