JP7056401B2 - Boil detection method in continuous casting mold, quality judgment method of continuous casting slab, monitoring method of continuous casting equipment, boil detection device in continuous casting mold - Google Patents

Description

The present invention uses a method for detecting boil in a mold for continuous casting, which monitors the molten metal surface of the mold for continuous casting during continuous casting and detects the boil of gas bubbles discharged from the mold for continuous casting. The present invention relates to a method for determining the quality of continuously cast slabs, a method for monitoring continuous casting equipment, and a boil detection device in a mold for continuous casting.

In continuous metal casting, molten metal is continuously supplied into the continuous casting mold via a dipping nozzle, a solidified shell is grown on the mold wall of the continuous casting mold, and the obtained slab is continuously cast. It is configured to continuously produce slabs by pulling out from the mold.

At the time of this continuous casting, the mold powder is added on the molten metal surface in the continuous casting mold. This mold powder melts on the surface of the molten metal in the continuous casting mold and flows between the mold wall and the solidified shell. Then, the mold powder forms a powder film between the mold wall and the solidified shell, and exerts a lubricating action between the mold wall and the solidified shell.
Further, at the time of continuous casting, an inert gas is supplied to the molten metal in the mold for continuous casting via a dipping nozzle in order to float and separate non-metal inclusions and the like in the molten metal.

Here, during continuous casting, the inert gas supplied into the continuous casting mold may aggregate and coarsen the gas bubbles, which may be released on the molten metal surface. This phenomenon is called "boil". When the boil of gas bubbles is generated, the molten mold powder on the molten metal surface may be caught in the molten metal, and the internal quality of the slab may be deteriorated.
Further, when a hole or a crack is generated in the dipping nozzle, the molten steel passing through the dipping nozzle entrains the atmospheric gas, and this atmospheric gas is discharged from the inside of the mold for continuous casting onto the molten metal surface. "Boil" will occur.
When the gas bubbles discharged from the continuous casting mold are boiled, the mold powder on the molten metal surface is removed and the molten metal is instantaneously exposed. Therefore, when observing the mold for continuous casting, when boil occurs, a region having high brightness is instantaneously observed.

For this reason, during continuous casting, the surface quality and internal quality of the slab can be determined by monitoring the molten metal surface of the continuous casting mold and detecting the boil of gas bubbles discharged from the mold, and equipment. It is possible to detect anomalies.
For this reason, conventionally, there has been provided a means for monitoring the molten metal surface of a continuous casting mold and detecting a boil of gas bubbles during continuous casting.

For example, in Patent Document 1, the inside of a mold is imaged, the obtained image signal is arithmetically processed, the number of bubbles floating on the molten metal surface and the size of the flame generated on the molten metal surface are detected, and the size of the gas is detected. A method of controlling the blowing amount has been proposed.
Further, in Patent Document 2, a method of obtaining an image by a television camera, fixing the stored image, taking a difference for each pixel from a newly input image, detecting the presence or absence of boil, and performing determination processing. Has been proposed.

Further, in Patent Document 3, one and the other in the long side direction of the mold are imaged by two cameras centering on the immersion nozzle, and the obtained image is subjected to an average luminance value per pixel by an image processing device. Is calculated, and a method of detecting one-sided boiling from the difference between the average brightness value of one in-mold image and the other in-mold image has been proposed.
Further, in Patent Document 4, the molten steel surface in the mold is imaged to obtain original image data, and the boil generation value and the non-boil generation value are binar values according to the brightness value of each pixel of the original image data. A method has been proposed in which the boil phenomenon is determined by performing the conversion process and comparing the determination image data with a predetermined determination pattern.

Further, in Patent Document 5, an image pickup unit that captures an image in a continuously cast mold and a flame portion generated inside the mold are detected by analyzing color distribution information based on the image in the mold. , An in-mold monitoring device including an abnormality detection unit for detecting an abnormality in the mold has been proposed.

Japanese Unexamined Patent Publication No. 02-235556 Japanese Unexamined Patent Publication No. 05-07715 Japanese Unexamined Patent Publication No. 08-11943 Japanese Unexamined Patent Publication No. 10-034305 Japanese Unexamined Patent Publication No. 2007-275944

By the way, on the molten metal surface of the mold for continuous casting, a flame is generated by burning the added mold powder. Therefore, in the image of the molten metal surface of the continuous casting mold, the combustion flame of the mold powder and the boil of the gas bubbles discharged from the inside of the continuous casting mold are recognized as high-brightness portions. , It was difficult to accurately distinguish between the combustion flame of the mold powder and the boil of gas bubbles emitted from the mold for continuous casting.
In Patent Document 4, the purpose is to discriminate between the flame and the boil to determine the boil, but in this method, the combustion flame of the mold powder and the mold for continuous casting are discharged with sufficient accuracy. It was difficult to distinguish it from the boil of gas bubbles.

The present invention has been made in view of the above-mentioned situation, in which the molten metal surface of the continuous casting mold is monitored, the combustion flame of the mold powder, the boil of gas bubbles discharged from the inside of the continuous casting mold, and the boil of gas bubbles. Boil detection method in a continuous casting mold that can accurately identify and reliably detect boil, quality judgment method of continuously cast slab using this boil detection method in a continuous casting mold, continuous It is an object of the present invention to provide a monitoring method for casting equipment and a boil detection device in a mold for continuous casting.

In order to solve the above problems, the boil detection method in the continuous casting mold according to the present invention detects the boil in the continuous casting mold that detects the boil of gas bubbles discharged from the continuous casting mold during continuous casting. The method is an imaging step of acquiring moving image data by imaging the inside of the continuous casting mold, a still image data acquisition step of cutting out still image data from the moving image data, and moving averaging processing of the still image data. A moving average processing step for obtaining moving average image data, and a binarization processing step for grayscale the moving average image data and binarize the moving average image data into a low-brightness portion and a high-brightness portion according to the brightness to obtain a binarized image. Then, based on the cut-out image extraction step of extracting the contour data of the high-brightness portion from the binarized image as a cut-out image and the extracted cut-out image, the cut-out image is released from the continuous casting mold. It is provided with a boil determination step for determining whether or not it corresponds to the boil of the gas bubble, and the determination logic in the boil determination step includes the cutout image and a teacher label having a correct answer label. In addition, it is characterized by being constructed by a supervised learning method .

According to the boil detection method in the mold for continuous casting having this configuration, the cutout image extraction step of extracting the contour data of the high brightness portion as a cutout image from the binarized image binarized according to the brightness and the extraction are performed. Based on the cut-out image, the mold powder is provided with a boil determination step for determining whether or not the cut-out image corresponds to the boil of the gas bubbles discharged from the continuous casting mold. It is possible to accurately distinguish between the combustion flame and the boil of gas bubbles discharged from the mold for continuous casting, and it is possible to reliably detect the boil of gas bubbles.

Here, since the boil detection method in the continuous casting mold according to the present invention has a moving average processing step for obtaining the moving average image data by performing the moving average processing of the still image data , the moving average processing step. This makes it possible to remove noise in the image and stably detect boil of gas bubbles.
Further, since the determination logic in the boil determination step is constructed by a supervised learning method based on the cutout image and the teacher label having the correct answer label, the combustion flame of the mold powder and the mold for continuous casting are used. It is possible to more accurately identify the boil of gas bubbles emitted from the inside.

Further, in the method for detecting a boil in a mold for continuous casting according to the present invention, in the boil determination step, the cut-out image is obtained from the inside of the mold for continuous casting by using the feature amount of the extracted cut-out image. It is preferable that the configuration is such that it is determined whether or not it corresponds to the boil of the released gas bubbles.
In this case, the combustion flame of the mold powder and the boil of the gas bubbles discharged from the mold for continuous casting can be discriminated more accurately, and the boil of the gas bubbles can be reliably detected.

Further, in the boil detection method in the continuous casting mold according to the present invention, a pixel vector may be used as the feature amount of the cutout image.
In this case, since the pixel vector is used as the feature amount of the extracted cut-out image, it is determined whether or not the cut-out image corresponds to the boil of the gas bubbles discharged from the continuous casting mold. It is possible to make a more accurate judgment.

Further, in the method for detecting boil in the mold for continuous casting according to the present invention, it is preferable that the feature amount of the cut-out image is a pixel vector processed into a HOG feature amount.
In this case, as the feature amount of the extracted image, the pixel vector processed into the HOG feature amount is used, so that the cutout image is the gas bubble emitted from the continuous casting mold. It is possible to determine whether or not it corresponds to a boil with even higher accuracy.

The quality determination method for the continuously cast slab of the present invention is obtained by using the above-mentioned boil detection method in the continuous casting mold according to the detection status of the boil of gas bubbles discharged from the continuous casting mold. It is characterized by determining the quality of the continuously cast slabs produced.

If gas bubble boil occurs, mold powder may be caught in it, so it is possible to worry about the effect on the internal quality of the continuously cast slab depending on the detection status of the gas bubble boil. Become.
Therefore, in the method for determining the quality of the continuously cast slab of the present invention, the boil of the gas bubbles released from the inside of the continuous casting mold is detected by the above-mentioned boil detection method in the continuous casting mold. It is possible to detect the boil of bubbles with high accuracy, and it is possible to accurately determine the quality of continuously cast slabs.

The monitoring method of the continuous casting equipment of the present invention uses the above-mentioned boil detection method in the continuous casting mold, and the equipment abnormality is caused according to the detection status of the boil of the gas bubbles discharged from the continuous casting mold. It is characterized by determining the presence or absence.

In the event of equipment abnormalities such as holes, defects, or cracks in the dipping nozzle, gas in the atmosphere may be caught in the molten metal supplied into the continuous casting mold, or molten steel in the continuous casting mold may occur. Since the flow is different from the expected state, more boil of gas bubbles discharged from the continuous casting mold will be generated as compared with the healthy state.
Therefore, in the monitoring method of the continuous casting equipment of the present invention, the boil of the gas bubbles released from the inside of the continuous casting mold is detected by the above-mentioned method of detecting the boil in the continuous casting mold. It is possible to detect the boil with high accuracy, and it is possible to accurately determine the presence or absence of an abnormality in the continuous casting equipment.

The boil detection device in the continuous casting mold of the present invention is a boil detection device in the continuous casting mold that detects the boil of gas bubbles discharged from the continuous casting mold during continuous casting, and is for continuous casting. An imaging means for capturing moving images in a mold to obtain moving image data, a still image data acquisition means for cutting out still image data from the moving image data, and a moving averaging process for moving averaging the still image data to obtain moving average image data. The means, a binarization processing means that grayscales the moving average image data and binarizes the moving average image data into a low-brightness portion and a high-brightness portion according to the brightness to obtain a binarized image, and the binarization processing means from the binarized image. Based on the cut-out image extraction means that extracts the contour data of the high-brightness portion as a cut-out image and the extracted cut-out image, the cut-out image corresponds to the boil of the gas bubbles discharged from the continuous casting mold. It is provided with a boil determination means for determining whether or not to perform, and the determination logic in the boil determination means is constructed by a supervised learning method based on the cutout image and a teacher label having a correct answer label. It is characterized by being done.

According to the gas bubble detection device in the mold for continuous casting having this configuration, the contour data of the high-brightness portion is cut out as an image from the binarized image obtained by binarizing the low-brightness portion and the high-brightness portion according to the brightness. A boil determination means for determining whether or not the cut-out image corresponds to a boil of the gas bubbles discharged from the continuous casting mold based on the cut-out image extraction means to be extracted and the extracted cut-out image. Since it is equipped with, it is possible to accurately distinguish between the combustion flame of the mold powder and the boil of gas bubbles discharged from the mold for continuous casting, and it is possible to reliably detect the boil of gas bubbles. It will be possible.

Here, since the boil detection device in the mold for continuous casting of the present invention is configured to include a moving average processing means for moving average processing the still image data to obtain moving average image data, the moving average processing is provided. By means, the noise of the image can be removed, and the boil of gas bubbles can be stably detected.
Further, since the determination logic in the boil determination means is constructed by a supervised learning method based on the cutout image and the teacher label having the correct answer label, the combustion flame of the mold powder and the mold for continuous casting are used. It is possible to more accurately identify the boil of gas bubbles emitted from the inside.

Further, in the boil detection device in the continuous casting mold of the present invention, the boil determination means uses the extracted feature amount of the cutout image to discharge the cutout image from the continuous casting mold. It is preferable to determine whether or not the gas bubble boil.
In this case, the combustion flame of the mold powder and the boil of the gas bubbles discharged from the mold for continuous casting can be discriminated more accurately, and the boil of the gas bubbles can be reliably detected.

Further, in the boil detection device in the mold for continuous casting of the present invention, a pixel vector may be used as the feature amount of the cutout image.
In this case, since the pixel vector is used as the feature amount of the extracted cut-out image, it is determined whether or not the cut-out image corresponds to the boil of the gas bubbles discharged from the continuous casting mold. It is possible to make a more accurate judgment.

Further, in the boil detection device in the mold for continuous casting of the present invention, it is preferable that the feature amount of the cut-out image is a pixel vector processed into a HOG feature amount.
In this case, as the feature amount of the extracted image, the pixel vector processed into the HOG feature amount is used, so that the cutout image is the gas bubble discharged from the mold for continuous casting. It is possible to more accurately determine whether or not the image corresponds to the boil of.

As described above, according to the present invention, the molten metal surface of the continuous casting mold is monitored, and the combustion flame of the mold powder and the boil of gas bubbles discharged from the inside of the continuous casting mold are accurately discriminated. , A method for detecting boil in a mold for continuous casting that can reliably detect boil, a method for determining the quality of continuously cast slabs using this method for detecting boil in a mold for continuous casting, a method for monitoring continuous casting equipment, Further, it is possible to provide a boil detection device in a mold for continuous casting.

It is a schematic diagram which shows an example of the continuous casting equipment which carries out the boil detection method in the mold for continuous casting which is an embodiment of this invention. It is a schematic diagram which shows an example of the boil detection apparatus in the mold for continuous casting which is an embodiment of this invention. It is a flow chart of discriminative model learning used for the boil detection method in the mold for continuous casting which is an embodiment of this invention. It is a flow chart of the boil detection method in the mold for continuous casting which is an embodiment of this invention. In the embodiment of the present invention, it is a figure which shows an example of the image data of the molten metal surface in a mold, a gray scale image, a binarized image, and a contour extraction result. It is explanatory drawing of HOG feature amount. It is a figure which shows the relationship between the detection state of the boil of a gas bubble in a mold for continuous casting, and the defect generation state of a slab. It is explanatory drawing which shows the observation situation for obtaining FIG. 7. It is a figure which shows the relationship between the detection state of the boil of a gas bubble in a mold for continuous casting, and the occurrence state of an equipment abnormality.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. The present invention is not limited to the following embodiments.

First, FIG. 1 shows an example of a continuous casting facility that implements a boil detection method in a continuous casting mold according to an embodiment of the present invention.
In this continuous casting facility 1, molten steel is transferred from a converter by a ladle 2, molten steel is transferred to a tundish 4 via a long nozzle 3, large inclusions are levitated and separated in the tundish 4, and then through a dipping nozzle 20. The molten metal is supplied into the continuous casting mold 30 to continuously cast the continuously cast slab.

As shown in FIG. 2, the continuous casting facility 1 of the present embodiment is not suitable for the continuous casting mold 30, the dipping nozzle 20 for discharging molten steel into the continuous casting mold 30, and the molten steel in the dipping nozzle 20. The inert gas introducing means 10 for introducing the active gas is provided.
Here, as shown in FIG. 2, the tundish 4 and the immersion nozzle 20 are connected to each other via the upper nozzle 6 and the sliding nozzle 7.

As shown in FIG. 2, the inert gas introducing means 10 in the present embodiment supplies the gas sleeve nozzle 11 provided on the side wall portion of the upper nozzle 6 and the immersion nozzle 20 and the inert gas to the gas sleeve nozzle 11. The gas supply pipe 12 is provided. In this embodiment, argon gas is introduced as the inert gas.

As shown in FIG. 2, the immersion nozzle 20 in the present embodiment has a substantially bottomed cylindrical shape, and has a vertical hole portion 21 and a pair of discharge holes 22 (22A) provided below the vertical hole portion 21. , 22B) and. Here, the discharge hole 22 is gradually inclined downward as it goes outward in the radial direction, and the pair of discharge holes 22 (22A, 22B) with respect to the axis of the immersion nozzle 20 (vertical hole portion 21). Are arranged symmetrically.

In the continuous casting facility 1 having such a configuration, molten steel is supplied to the casting space of the continuous casting mold 30 via the dipping nozzle 20.
Argon gas, which is a kind of inert gas, is blown into the molten steel supplied through the immersion nozzle 20 by the inert gas introducing means 10. Further, the mold powder 9 is supplied onto the molten steel stored in the mold space in order to keep the molten steel warm and to ensure the lubricity between the solidified shell S and the inner wall of the continuous casting mold 30.

Here, the inert gas introduction means 10 may aggregate the inert gas supplied into the continuous casting mold 30 to coarsen the bubbles, which may be released on the molten metal surface. This phenomenon is called "boil". When the boil of gas bubbles is generated, the molten mold powder 9 on the molten metal surface is caught in the molten steel, and there is a possibility that the internal quality of the slab for continuous casting is deteriorated.

Further, when a hole or a crack is generated in the dipping nozzle 20, atmospheric gas is entrained in the molten steel passing through the dipping nozzle 20, and the atmospheric gas is transferred from the inside of the continuous casting mold 30 onto the molten metal surface. It will be released and boil will be generated.

Here, when the gas bubbles discharged from the continuous casting mold 30 are boiled, the mold powder is removed and the molten steel is instantaneously exposed. Therefore, when observing the inside of the continuous casting mold 30, when boil occurs, a region having high brightness is instantaneously observed.
Therefore, in the present embodiment, the boil detection device 50 in the continuous casting mold for detecting the boil of the gas bubbles discharged from the continuous casting mold 30 is arranged.

Since a flame is generated by burning the mold powder 9 on the molten metal surface of the continuous casting mold 30, both the combustion flame of the mold powder 9 and the boil of gas bubbles are observed as a region having high brightness. It will be. Therefore, it is necessary to accurately distinguish between the combustion flame of the mold powder 9 and the boil of gas bubbles.

The boil detection device 50 in the continuous casting mold according to the present embodiment includes an imaging camera 51 that obtains moving image data by taking a moving image in the continuous casting mold 30, and a still image data acquisition means that cuts out still image data from the moving image data. 52, a moving averaging processing means 53 that performs moving averaging processing of image data to obtain moving averaging image data, greyscaling the moving averaging image data, and binarizing the moving average image data into a low-brightness portion and a high-brightness portion according to the brightness. A binarization processing means 54 for obtaining a binarized image, a cutout image extraction means 55 for extracting contour data of a high-brightness portion from the binarized image as a cutout image, a feature amount of the extracted cutout image, and a machine in advance. Based on the identification model obtained by performing the training, and the boil determination means 56 for determining whether or not the cutout image corresponds to the boil of gas bubbles discharged from the continuous casting mold 30. I have.

In the present embodiment, the computer terminal 59 includes each of the above means. Specifically, it is as follows. The computer terminal 59 is an information processing device including a processor and a memory. The computer terminal 59 functions as a device including image data cutting means 52, moving average processing means 53, binarization processing means 54, contour extraction means 55, and boil determination means 56 by executing a program by the processor.

A part or all of the above-mentioned functions included in the computer terminal 59 may be realized by using hardware such as an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), and an FPGA (Field Programmable Gate Array). The program executed by the computer terminal 59 may be recorded on a computer-readable recording medium.

Computer-readable recording media include, for example, flexible disks, magneto-optical disks, ROMs, CD-ROMs, portable media such as semiconductor storage devices (eg SSD: Solid State Drive), hard disks built into computer systems, and semiconductor storage. It is a storage device such as a device. The above program may be transmitted via a telecommunication line. Note that some or all of the above functions included in the computer terminal 59 may be executed by an information processing device different from that of the computer terminal 59. For example, a part or all of the above functions may be implemented in a server device or a cloud connected to the image pickup camera 51 or the computer terminal 59 via a network.

Here, the discriminative model learning used in the boil detection method in the continuous casting mold according to the present embodiment will be described with reference to FIG.

(Imaging step S101)
First, the image pickup camera 51 is used to take a moving image of the molten metal surface of the continuous casting mold 30, and the moving image data is stored in the moving image recording unit. In this embodiment, as shown in FIG. 2, the continuous casting mold 30 is configured to shoot moving images by two imaging cameras 51 on one side and the other side of the immersion nozzle 20 in the long side direction. ..

(Cutout image acquisition step S102)
Next, the image of the candidate boil is cut out from the image pickup data stored in the moving image recording unit. In the cutout image acquisition step S102, the following still image data acquisition step S121, moving average processing step S122, binarization processing step S123, and cutout image extraction step S124 are carried out.

(Still image data acquisition step S121)
The moving image data stored in the moving image recording unit is cut out as still image data (hereinafter, simply referred to as “image data”) for each frame.

(Moving average processing step S122)
Next, a moving average process is performed on the obtained image data. The moving average image data in the previous stage was multiplied by the weighting coefficient α, the obtained image data was multiplied by the weighting coefficient (1-α), and the sum of these was used as the moving average image data this time. In this embodiment, the weighting coefficient α is set to be in the range of 0.8 or more and 0.99 or less.

(Battering process step S123)
Next, an image of the molten metal surface portion in the mold for continuous casting (hereinafter referred to as "partial image") is extracted from the moving average image data, this is grayscaled, and binar values are divided into a low-brightness portion and a high-brightness portion according to the brightness. The binarization process is performed to obtain a binarized image. That is, the partial image shown in FIG. 5A is generated by extracting only the molten metal surface portion in the mold for continuous casting from the moving average image data, and the partial image is grayscaled as shown in FIG. 5B. As shown in FIG. 5 (c), a binarized image was obtained by binarizing the low-luminance portion and the high-luminance portion according to the brightness. It should be noted that the process of extracting only the molten metal surface portion in the mold for continuous casting may be implemented in any way. For example, a partial image may be generated by extracting an image of a predetermined region defined in advance from the image data cut out from the moving image data. The predetermined area may be defined in advance by a designer or the like according to the image pickup area of the image pickup camera 51 or the like.

(Cutout image extraction step S124)
Next, the contour data of the high-luminance portion is extracted from the binarized image. In this embodiment, the algorithm of Suzuki85 is used to recognize the contour of the high-luminance portion from the binarized image, and as shown in FIG. 5D, the smallest rectangle (extrinsic rectangle) capable of including the recognized contour. ) Extracted as a shape. The contour recognition method does not have to be limited to the Suzuki 85, and may be realized by using another algorithm or the like.
Then, the coordinates of the vertices of the extracted rectangle (extrinsic rectangle) are calculated, and the image of the high-brightness portion (hereinafter referred to as a cutout image) is cut out from the still image data based on the vertices coordinates of the extrinsic rectangle.

(Teacher Labeling Step S103)
Since the above-mentioned cutout image includes the combustion flame of the mold powder 9 in addition to the boil of gas bubbles, it is necessary to create a discriminative model for determining the boil and the flame from the cutout image. Therefore, first, for each of the plurality of cutout images extracted in the cutout image extraction step S124, a person determines whether or not the cutout image corresponds to a boil of gas bubbles and assigns the cutout image as a teacher label. A set of the cutout image and the teacher label (hereinafter referred to as teacher data) is stored in the teacher data storage unit.

(Feature quantity calculation step S104)
Create the features of the cutout image to be used as the input of the discriminative model. It is also possible to use the pixel vector of the cutout image or the area of the contour portion as the feature amount of the cutout image, but in the present embodiment, as the feature amount of the cutout image, the gradient direction and the intensity of the brightness are set for each pixel. The calculated one is used, and more specifically, the HOG (Histogram of Oriented Gradient) feature amount is used.

Here, the HOG feature amount will be described with reference to FIG.
First, one cropped image is reduced or enlarged to a predetermined size (in this embodiment, 40 pixels wide x 30 pixels high). The relationship between the cut-out image, the block, the cell, and the pixel is shown in FIG. 6A. In this embodiment, the cell size is 5 × 5 pixels and the block size is 3 × 3 cells.

Next, the gradient direction and intensity of the luminance are calculated for each pixel of one cell. Hereinafter, the position coordinates of the pixels in the vertical direction / horizontal direction will be described as x and y.

First, the luminance value vector [f _x (x, y), f _y (x, y)] at the coordinates (x, y) is calculated based on the following equation.

Note that L (x, y) represents the luminance value of the pixel at the coordinates (x, y). f _x (x, y) is the difference between the brightness values of the left and right pixels of the pixel of the coordinate (x, y), and f _y (x, y) is the brightness of the pixels above and below the pixel of the coordinate (x, y). The difference between the values.

Next, the intensity m (x, y) and the gradient θ of the luminance value vector [f _x (x, y), f _y (x, y)] are calculated based on the following equation. The result of visualizing the luminance gradient vector is shown in FIG. 6 (b).

Then, as shown in FIG. 6C, a histogram is created for each cell with the gradient direction calculated for each pixel as the horizontal axis and the gradient intensity as the vertical axis. In the present embodiment, the gradient direction θ is divided into 9 divisions at intervals of 20 ° from 0 to 180 °, and the sum of the intensities of each division region is defined as the gradient intensity. When the gradient direction θ exceeds 180, it is set to be within the range of 0 to 180 ° as θ-180. Then, nine gradient intensities v (n) exist in one cell.

Here, as shown in FIG. 6D, since one block is a 3 × 3 cell, there are 3 × 3 × 9 = 81 gradient intensities in one block. Then, as shown in the following equation, the HOG feature amount is obtained by normalizing the gradient intensity for each block.

Here, b is the block size and N is the number of gradient divisions, and in this embodiment, b = 3 and N = 9, respectively. ε is a constant prepared so that the denominator of the equation 4 does not become 0, and ε = 1. Since the number of blocks per cut-out image is 24, the number of HOG features per cut-out image is 81 × 24 = 1944.

(Discriminative model creation step S105)
Whether the cut-out image corresponds to the boil of gas bubbles discharged from the continuous casting mold 30 or the combustion flame by inputting the feature amount of the cut-out image (HOG feature amount in this embodiment) (hereinafter, cut-out). A discriminative model that outputs (called an image class) is constructed by supervised learning, which is one of the machine learning methods, based on a plurality of teacher data stored in the teacher data storage unit, and stored in the discrimination model storage unit. .. Specific examples of supervised learning include a K-nearest neighbor method, a support vector machine, and deep learning, but in this embodiment, the K-nearest neighbor method is used.

Hereinafter, the procedure for detecting the boil using the created discriminative model will be described with reference to FIG.

(Imaging step S01)
Similar to the image pickup step S101 described above, the image pickup camera 51 is used to take a moving image of the molten metal surface of the continuous casting mold 30 and store the moving image data in the moving image recording unit.

(Cutout image acquisition step S02)
Next, similarly to the cutout image acquisition step S102 described above, the image of the candidate boil is cut out from the image pickup data stored in the moving image recording unit. In this cut-out image acquisition step S02, a still image data acquisition step S21, a moving average processing step S22, a binarization processing step S23, and a cut-out image extraction step S24 are carried out.

Here, the still image data acquisition step S21, the moving average processing step S22, the binarization processing step S23, and the cutout image extraction step S24 are the still image data acquisition step S121 and the moving average processing step S122, respectively. 2. The same processing is performed in the binarization processing step S123 and the cutout image extraction step S124.
As a result, the cut-out image of the high-luminance portion is acquired from the still image data.

(Feature quantity calculation step S03)
Create the features of the cutout image to be used as the input of the discriminative model. In the feature amount calculation step S03, the same processing as in the feature amount calculation step S104 described above is performed.

(Boil determination step S04)
Next, the class of the cut-out image is determined using the feature amount of the cut-out image and the identification model stored in the discrimination model storage unit. As the determination result, the image pickup time / the coordinates of the cutout image (eg, the apex of the rectangle, the center of gravity, etc.) / the class of the cutout image are stored in the boil judgment result storage unit.

The quality determination method of the continuously cast slab using the boil detection method in the continuous casting mold according to the present embodiment will be described below.
When the inert gas is supplied into the continuous casting mold 30 by the inert gas introducing means 10, the bubbles of the inert gas are aggregated and coarsened, and the coarsened gas bubbles may be released on the molten metal surface. .. At this time, the molten mold powder 9 on the molten metal surface is caught in the molten steel, and the internal quality of the slab for continuous casting deteriorates.

Therefore, by detecting the boil of gas bubbles discharged from the mold 30 for continuous casting, it is possible to determine the internal quality of the slab for continuous casting. For example, as shown in FIG. 7, it can be confirmed that a defect of the slab for continuous casting has occurred at the position where the boil of gas bubbles is detected.

FIG. 7 shows the results of photographing the mold molten metal surface and investigating the number of boil detected using the discriminative model. In the example of FIG. 7, as shown in FIG. 8, the mold surface was imaged from the north side and the south side by two cameras, and each imaging area was bisected in the mold thickness direction and divided into a total of four areas. FIG. 7 shows graphs of the four regions arranged horizontally and vertically arranged in a time series. Further, in each graph, each region whose horizontal axis is divided into four is divided into 20 in the mold width direction, and the position thereof is shown. The vertical axis represents the frequency of boil occurrence within a specific time (1 minute in the figure) at the position.

Next, a monitoring method of the continuous casting facility 1 using the boil detection method in the continuous casting mold according to the present embodiment will be described.
When a hole or a crack is generated in the dipping nozzle 20, atmospheric gas is entrained in the molten steel passing through the dipping nozzle 20, and this atmospheric gas is discharged from the inside of the continuous casting mold 30 onto the molten metal surface, and boil of gas bubbles is generated. It will occur.

Therefore, by detecting the boil of gas bubbles discharged from the mold for continuous casting, it is possible to detect an abnormality in the continuous casting equipment. For example, FIG. 9 is a diagram showing the boil generation state of the north L plane and the F plane at a certain time, as in FIG. 7. As shown in FIG. 9, the frequency of boil generation gradually increases after 10 minutes on the north L surface. As a result of investigating this as well, it was found that a hole was formed in the immersion nozzle 20.

According to the boil detection method in the continuous casting mold according to the present embodiment having the above configuration, the binarized image obtained by binarizing the low-brightness portion and the high-brightness portion according to the brightness has high brightness. The contour data is continuously cast based on the cutout image extraction step S24 for extracting the contour data of the portion as a cutout image, the feature amount of the extracted cutout image, and the identification model obtained by performing machine learning in advance. Since the boil determination step S04 for determining whether or not the gas bubble is boiled from the inside of the mold 30 is provided, the mold powder is based on the identification model obtained by performing machine learning in advance. It is possible to accurately distinguish between the combustion flame of 9 and the boil of the gas bubble, and it is possible to reliably detect the boil of the gas bubble.

Further, in the present embodiment, as the feature amount of the extracted cutout image, the one obtained by calculating the gradient direction and the intensity of the brightness for each pixel is used, and more specifically, the HOG feature amount is used. Based on the identification model obtained by performing machine learning in advance, it is possible to more accurately determine whether or not the contour data corresponds to the boil of gas bubbles discharged from the continuous casting mold 30. It will be possible.

Further, according to the quality determination method for continuously cast slabs of the present embodiment, the boil of gas bubbles discharged from the inside of the continuous casting mold 30 is detected by the boil detection method in the continuous casting mold of the present embodiment. Since it is detected, it is possible to accurately detect the boil of gas bubbles. Then, the quality of the continuously cast slab can be accurately determined according to the boil generation condition.

Further, according to the monitoring method of the continuous casting equipment of the present embodiment, the boil of gas bubbles discharged from the inside of the continuous casting mold 30 is detected by the boil detection method in the continuous casting mold of the present embodiment. Therefore, even if a hole, a crack, or the like is generated in the dipping nozzle 20, it is possible to accurately detect it, and it is possible to accurately determine whether or not there is an abnormality in the continuous casting equipment 1.

According to the boil detection device in the mold for continuous casting according to the present embodiment, the contour data of the high-brightness portion is cut out as an image from the binarized image obtained by binarizing the low-brightness portion and the high-brightness portion according to the brightness. Based on the cutout image extraction means 55 to be extracted, the feature amount of the extracted cutout image, and the identification model obtained by performing machine learning in advance, the cutout image is a gas released from the inside of the continuous casting mold 30. Since it is provided with a boil determination means 56 for determining whether or not it corresponds to a bubble boil, the combustion flame of the mold powder 9 and the boil of gas bubbles can be determined by the identification model obtained by performing machine learning in advance. , It becomes possible to identify accurately, and it becomes possible to reliably detect the boil of gas bubbles.

The boil detection method in the mold for continuous casting, the quality determination method of the continuous cast slab, the monitoring method of the continuous casting equipment, and the boil detection device in the mold for continuous casting, which are the embodiments of the present invention, have been described above. The invention is not limited to this, and can be appropriately modified without departing from the technical idea of the invention.

For example, in the present embodiment, in the cutout image extraction step S24, the algorithm of Suzuki85 is used to extract the contour data of the high-luminance portion from the binarized image, but the present invention is not limited to this. The contour data of the high-luminance portion may be extracted by using another algorithm.
Further, although the description has been made assuming that the HOG feature amount is used as the feature amount of the cut-out image, the present invention is not limited to this, and other feature amounts may be used.

Further, in the present embodiment, the continuous casting equipment shown in FIGS. 1 and 2 has been described as an example, but the present invention is not limited to this, and may be applied to continuous casting equipment having other configurations.

30 Mold for continuous casting 50 Boil detection device in mold for continuous casting 51 Imaging camera (imaging means)
52 Still image data acquisition means 53 Moving average processing means 54 Binarization processing means 55 Contour extraction means 56 Boil determination means S01 Imaging step S21 Still image data acquisition step S22 Moving average processing step S23 Binarization processing step S24 Cut out image extraction step S04 Boyle judgment step

Claims (10)

It is a boil detection method in a continuous casting mold that detects the boil of gas bubbles released from the continuous casting mold during continuous casting.
An imaging step of acquiring moving image data by taking a moving image of the inside of the continuous casting mold,
A still image data acquisition step for cutting out still image data from the moving image data, and
The moving average processing step of performing the moving average processing of the still image data to obtain the moving average image data, and
A binarization processing step of grayscale the moving average image data and binarizing the low-luminance portion and the high-luminance portion according to the brightness to obtain a binarized image.
A cut-out image extraction step for extracting the contour data of the high-luminance portion as a cut-out image from the binarized image, and
A boil determination step for determining whether or not the cutout image corresponds to a boil of the gas bubbles discharged from the continuous casting mold based on the extracted cutout image.
Equipped with
A method for detecting a boil in a mold for continuous casting , wherein the determination logic in the boil determination step is constructed by a supervised learning method based on the cutout image and a teacher label having a correct answer label . In the boil determination step, it is determined whether or not the cutout image corresponds to the boil of the gas bubbles discharged from the continuous casting mold by using the feature amount of the cutout image extracted. The method for detecting a boil in a mold for continuous casting according to claim 1 . The method for detecting boil in a mold for continuous casting according to claim 1 or 2, wherein a pixel vector is used as the feature amount of the cut-out image. The boil detection method in a mold for continuous casting according to any one of claims 1 to 3, wherein the feature amount of the cut-out image is a pixel vector processed into a HOG feature amount. Using the boil detection method in the continuous casting mold according to any one of claims 1 to 4, the boil of gas bubbles discharged from the continuous casting mold can be detected according to the detection status. A method for determining the quality of continuously cast slabs, which comprises determining the quality of continuously cast slabs. The boil detection method in the continuous casting mold according to any one of claims 1 to 4 is used to continuously detect the boil of gas bubbles discharged from the continuous casting mold. A method for monitoring a continuous casting facility, which comprises determining the presence or absence of an abnormality in the casting facility. A boil detection device in a continuous casting mold that detects the boil of gas bubbles released from the continuous casting mold during continuous casting.
An imaging means for obtaining moving image data by taking a moving image of the inside of the continuous casting mold,
Still image data acquisition means for cutting out still image data from the moving image data,
A moving average processing means for obtaining moving average image data by performing moving average processing on the still image data,
A binarization processing means for obtaining a binarized image by grayscale the moving average image data and binarizing the low-luminance portion and the high-luminance portion according to the brightness.
A cut-out image extraction means for extracting the contour data of the high-luminance portion as a cut-out image from the binarized image, and
A boil determination means for determining whether or not the cutout image corresponds to a boil of the gas bubbles discharged from the continuous casting mold based on the extracted cutout image.
Equipped with
A boil detection device in a mold for continuous casting , wherein the determination logic in the boil determination means is constructed by a supervised learning method based on the cutout image and a teacher label having a correct answer label . The boil determination means determines whether or not the cutout image corresponds to the boil of the gas bubbles discharged from the continuous casting mold by using the feature amount of the cutout image extracted. The boil detection device in a continuous casting mold according to claim 7 . The boil detection device in a mold for continuous casting according to claim 7 or 8, wherein a pixel vector is used as the feature amount of the cut-out image. The boil detection device in a mold for continuous casting according to any one of claims 7 to 9, wherein the feature amount of the cut-out image is a pixel vector processed into a HOG feature amount.

Images