KR102272100B1 - Method and system for surface crack detection of continuous casting using deep learning images - Google Patents

Abstract

The present invention relates to a method and system for detecting vertical cracks in continuous casting through image learning, which improve the consistency of vertical crack detection by image learning to improve the quality of slabs through selective mill scale inspection. According to the present invention, the method for detecting vertical cracks in continuous casting through image learning may comprise the steps of: (a) collecting the temperature of a metal plate in a mold in real time in the continuous casting process; (b) generating a temperature map of the metal plate in units of slabs; (c) determining whether a vertical crack occurs in the temperature map based on deep learning; and (d) performing a detailed inspection on the slab determined to be vertically cracked.

Description

BACKGROUND ART Method and system for surface crack detection of continuous casting using deep learning images}

The present invention relates to a method and system for detecting longitudinal cracks in continuous casting through image learning, and more particularly, improving the consistency of longitudinal crack detection using image learning to improve the quality of cast steel through selective mill scale inspection It relates to a method and system for detecting longitudinal cracks in continuous casting through image learning.

A conventional continuous casting apparatus for continuously casting a slab can manufacture a slab having a predetermined width and thickness by simultaneously casting and rolling while passing molten steel through a plurality of roll segments arranged in a line.

The slab manufactured in the continuous casting device passes through the mold through the tundish that stores the molten steel in the liquid state from the casting furnace, and the surface layer forms a solid solidified shell by the cooling action. The solidification proceeds by cooling water while being guided by the guide segment roll installed in the lower segment to form a complete solid slab.

The solid slab is cut by a torch cutting machine (TCM) according to a predetermined length, and is quickly transferred to a post-process by a run out roller table.

1 is a graph conceptually illustrating a process in which a temperature difference is generated due to an air gap in a copper plate in a conventional mold, and FIG. 2 is a graph showing the copper plate temperature according to time measured by the temperature sensor of FIG. , FIG. 3 is a photograph showing an example of longitudinal cracks that may be generated by the air gap of FIG. 1 .

As shown in FIGS. 1 to 3 , in order to detect longitudinal cracks in the conventional continuous casting process, the possibility of occurrence of longitudinal cracks is detected in real time by using the temperature deviation measured in the mold.

This free vertical crack detection method measures the temperature from a temperature sensing means such as a plurality of temperature sensors installed in the mold, and the temperature difference occurs when the vertical crack occurs and does not occur due to the air gap. it will be used

That is, when a vertical crack occurs in the mold, the temperature of the copper plate adjacent to the crack generating part is decreased due to the air gap. This phenomenon can be used to predict the vertical crack.

The above-described technology has been described for understanding the background of the present invention, and does not mean the prior art widely known in the art to which the present invention pertains.

Korean Patent Laid-Open Publication No. 2010-0107805

However, the conventional method for detecting cracks in duty-free road is to apply a mathematical model when a temperature deviation occurs in the copper plate to determine it with a simple calculation algorithm, and to detect tax-free cracks that can occur in various forms according to various process specifications or various process environments. It was not possible to accurately reflect and discriminate, so there was a problem that the consistency was greatly reduced.

The present invention is to solve various problems, including the above problems, by forming a temperature map according to time of each row thermocouple installed in the copper plate of the mold, and deep learning it using a deep learning-based object detection model to obtain tax exemption An object of the present invention is to provide a duty-free furnace crack detection method and system through image learning that improves the consistency of furnace crack detection and enables selective mill scale inspection to significantly improve the cast steel quality. However, these problems are exemplary, and the scope of the present invention is not limited thereto.

A method for detecting longitudinal cracks in continuous casting through image learning according to the spirit of the present invention for solving the above problems, (a) collecting the temperature of the metal plate in the mold in real time in the continuous casting process; (b) generating a temperature map of the metal plate in units of slabs; (c) determining whether a crack occurs in the temperature map based on deep learning; and (d) performing a detailed inspection on the corresponding slab determined to be tax-free and cracked.

Further, according to the present invention, in step (a), the metal plate may be a copper plate in contact with the solidification cell.

In addition, according to the present invention, in the step (a), it is possible to measure the temperature of each part of the metal plate in real time by using thermocouples installed in n rows and m columns installed on the rear surface of the metal plate.

In addition, according to the present invention, in the step (b), the temperature map shows the temperature state according to time of the thermocouples for each row by color so that the low temperature section can be displayed in the form of a cold spot (cold spot). It may be a thermal image showing the image.

In addition, according to the present invention, in the step (c), if it is determined as a cold spot more than a reference number of times in the same column of each row thermocouple using a deep learning-based object detection model, it can be determined as a duty-free crack occurrence.

In addition, according to the present invention, the deep learning-based object detection model is a CNN (Convolutional Neural Network) that can operate in a machine learning mode that learns an optimization weight using a loss function and an object detection mode that determines a crack by duty-free may include.

In addition, according to the present invention, the deep learning-based object detection model extracts a feature map from a plurality of temperature maps, including at least any one of a convolution layer, a pooling layer, an activation function, and combinations thereof. extraction unit; a reconstruction unit for reconstructing the defect feature map into a reconstructed image including at least one of a deconvolution layer, an unpooling layer, an activation function, and combinations thereof; and a classifier for classifying the characteristics of the reconstructed image so as to display the position or size of the vertical crack.

In addition, according to the present invention, in the step (d), the detailed examination may be a mill scale examination for removing the oxidized surface of the slab and examining the epidermis.

On the other hand, the continuous casting crack detection system through image learning according to the spirit of the present invention for solving the above problems, a temperature collecting device for collecting the temperature of the metal plate in the mold in the continuous casting process in real time; a temperature map generating device for generating a temperature map of the metal plate in units of slabs; a vertical crack determination device for determining whether a vertical crack occurs in the temperature map based on deep learning; and an information output device for outputting target slab information so as to perform a detailed inspection on the corresponding slab determined to be tax-free cracking.

According to some embodiments of the present invention made as described above, a temperature map according to time of each row thermocouple installed in the copper plate of the mold is formed and deep learning is performed using the deep learning-based object detection model to detect vertical cracks. It is to have the effect of improving the consistency of the cast iron and greatly improving the quality of the cast steel by performing a selective mill scale inspection through this. Of course, the scope of the present invention is not limited by these effects.

1 is a graph conceptually illustrating a process in which a temperature difference due to an air gap is generated in a copper plate in a conventional mold.
FIG. 2 is a graph showing the copper plate temperature according to time measured by the temperature sensor of FIG. 1 .
FIG. 3 is a photograph showing an example of longitudinal cracks that may be generated by the air gap of FIG. 1 .
4 is a conceptual diagram conceptually illustrating a longitudinal crack detection system of continuous casting through image learning according to some embodiments of the present invention.
5 is a cross-sectional view showing an example of a thermocouple of a crack detection system of continuous casting through image learning of FIG. 4 .
6 is a conceptual diagram conceptually illustrating a learning mode of the deep learning-based object detection model of the continuous casting freeway crack detection system through image learning of FIG. 4 .
7 is a formula showing an example of a loss function applied in the learning mode of the deep learning-based object detection model of the continuous casting freeway crack detection system through image learning of FIG. 6 .
8 is a conceptual diagram conceptually illustrating a determination mode of a deep learning-based object detection model of the continuous casting freeway crack detection system through image learning of FIG. 4 .
9 is a flowchart illustrating a method for detecting longitudinal cracks in continuous casting through image learning according to some embodiments of the present invention.

Hereinafter, several preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Examples of the present invention are provided to more completely explain the present invention to those of ordinary skill in the art, and the following examples may be modified in various other forms, and the scope of the present invention is as follows It is not limited to an Example. Rather, these embodiments are provided so as to more fully and complete the present disclosure, and to fully convey the spirit of the present invention to those skilled in the art. In addition, in the drawings, the thickness or size of each layer is exaggerated for convenience and clarity of description.

Although the terms first, second, etc. are used herein to describe various members, parts, regions, layers and/or parts, these members, parts, regions, layers and/or parts are limited by these terms so that they It is self-evident that These terms are used only to distinguish one member, component, region, layer or portion from another region, layer or portion. Accordingly, a first member, component, region, layer or portion discussed below may refer to a second member, component, region, layer or portion without departing from the teachings of the present invention.

Also, relative terms such as "above" or "above" and "below" or "below" may be used herein to describe the relationship of certain elements to other elements as illustrated in the drawings. It may be understood that relative terms are intended to include other orientations of the element in addition to the orientation depicted in the drawings. For example, if an element is turned over in the figures, elements depicted as being on the face above the other elements will have orientation on the face below the other elements. Thus, the term “top” by way of example may include both “bottom” and “top” directions depending on the particular orientation of the drawing. If the element is oriented in a different orientation (rotated 90 degrees relative to the other orientation), the relative descriptions used herein may be interpreted accordingly.

The terminology used herein is used to describe specific embodiments, not to limit the present invention. As used herein, the singular forms may include the plural forms unless the context clearly dictates otherwise. Also, as used herein, “comprise” and/or “comprising” refers to the presence of the recited shapes, numbers, steps, actions, members, elements, and/or groups of those specified. and does not exclude the presence or addition of one or more other shapes, numbers, movements, members, elements and/or groups.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described with reference to the drawings schematically illustrating ideal embodiments of the present invention. In the drawings, variations of the illustrated shape can be expected, for example depending on manufacturing technology and/or tolerances. Accordingly, embodiments of the inventive concept should not be construed as limited to the specific shape of the region shown in the present specification, but should include, for example, changes in shape caused by manufacturing.

4 is a conceptual diagram conceptually illustrating a longitudinal crack detection system 1000 of continuous casting through image learning according to some embodiments of the present invention.

As shown in Figure 4, the continuous casting apparatus of the present invention is a ladle (10) and a tundish (Tundish, 20), a mold (Mold, 30), a segment in which the support rolls 60 are installed, the spray means 65 ), a pinch roll, and a cutter, and the like.

For example, the ladle 10 may be provided as a pair, and may alternately receive the molten steel M and alternately supply it to the tundish 20 . The molten steel M accommodated in the ladle 10 flows to the tundish 20 through a shroud nozzle 15 extending toward the tundish 20 . The shroud nozzle 15 may extend to be submerged in the molten steel M in the tundish 20 so that the molten steel M is not oxidized or nitridized by exposure to air.

For reference, a case in which the molten steel M is exposed to the air due to breakage of the shroud nozzle 15 or the like may be referred to as open casting.

Also, for example, the tundish 20 receives the molten steel M from the ladle 10 and supplies the molten steel M to the mold 30 . The tundish 20 may control the supply speed of the molten steel M, distribute the molten steel M to the mold 30, store the molten steel M, and separate slag and non-metallic inclusions.

Here, the molten steel M accommodated in the tundish 20 may flow between the molds 30 through a submerged entry nozzle (SEN) 25 extending into the mold 30 . The submerged nozzle 25 may be disposed in the center of the mold 30 so that the flow of the molten steel M discharged from both outlets of the submerged nozzle 25 may be symmetrical.

In addition, for example, the start, stop, and discharge speed of the discharging of the molten steel M through the submerged nozzle 25 is determined by a stopper 21 installed in the tundish 20 corresponding to the submerged nozzle 25 . can be decided. Here, the stopper 21 may be vertically moved along the same line as the submerged nozzle 25 to open and close the inlet of the submerged nozzle 25 .

In addition, here, the control of the flow of the molten steel M through the submerged nozzle 25 may use a slide gate method, which is different from the above-described stopper method. The slide gate can control the discharge flow rate of the molten steel M through the immersion nozzle 25 while the plate slides in the horizontal direction in the tundish 20 .

Also, for example, the mold 30 may include a copper plate that is typically a water-cooled copper material, and the molten steel M is first cooled. The mold 30 may form a hollow part in which the molten steel M is accommodated in the form in which a pair of structurally facing surfaces are opened.

For example, the degree of solidification of the molten steel M varies depending on the carbon content according to the steel type, the type of powder, the casting speed, and the like. The molten steel M discharged to the mold 30 may be solidified from a portion in contact with the wall surface of the mold 30 . This is because heat is lost by the mold 30 in which the periphery rather than the center of the molten steel M is water-cooled. As such, according to the method in which the peripheral portion of the molten steel M is solidified first, the rear portion along the casting direction of the strand is the unsolidified molten steel 82 wrapped in the solidified shell 81 in which the molten steel M is solidified. will take shape

In addition, the mold 30 may reciprocate in order to prevent the molten steel from adhering to the wall surface of the mold. A lubricant may be used to reduce friction between the mold 30 and the strands during reciprocating motion and to prevent burning. Here, as the lubricant, there are rapeseed oil to be sprayed and mold powder added to the surface of the molten steel in the mold 30 .

For example, the mold powder is added to the molten steel in the mold 30 to become slag, and not only lubricates the mold 30 and the strands, but also prevents oxidation and nitridation of the molten steel in the mold 30, heat retention, and non-metals floating on the surface of the molten metal. It can also perform the function of absorption of inclusions.

In addition, for example, the strand is supported so that the solidification angle is not changed by the support roll 60, and when the tip 83 of the strand is pulled by the pinch roll, the unsolidified molten steel 82 together with the solidified shell 81. move in the casting direction. The unsolidified molten steel 82 may be cooled by the spray means 65 for spraying cooling water in the process of moving above. This causes the thickness occupied by the unsolidified molten steel 82 in the strands to gradually decrease. When the strand reaches a point (85), the strand is filled with the solidifying shell (81) to its entire thickness. The solidified strand may be cut to a predetermined size at the cutting point 91 and divided into slabs such as slabs (S).

However, the continuous casting apparatus according to some embodiments of the present invention is not necessarily limited to the drawings, and a wide variety of shapes and types of continuous casting equipment may all be applied.

Here, for example, the longitudinal crack detection system 1000 of continuous casting through image learning of the continuous casting apparatus according to some embodiments of the present invention, the temperature of the metal plate 31 in the mold 30 in the continuous casting process A temperature collection device 100 that collects in real time, a temperature map generation device 200 that generates temperature maps M1, M2, and M3 of the metal plate 31 in units of slabs, and the temperature based on deep learning The object to be able to perform a detailed inspection on the longitudinal crack determination device 300 that determines whether the crack occurs in the duty free way of the maps M1, M2, and M3 and the corresponding slab S that is determined as the occurrence of the crack in the duty free way The slab (S) may include an information output device 400 for outputting information.

Therefore, the longitudinal crack detection system 1000 of continuous casting through image learning according to some embodiments of the present invention collects the temperature of the metal plate 31 in the mold 30 in the continuous casting process in real time, and the slab (S) generates temperature maps (M1) (M2) (M3) of the metal plate 31 in units, and determines whether or not cracks occur in the temperature maps (M1) (M2) (M3) based on deep learning And, it is possible to perform a series of processes of performing a detailed inspection of the slab (S) determined to be tax-free crack generation.

5 is a cross-sectional view showing an example of the thermocouple (G) of the longitudinal crack detection system 1000 of continuous casting through image learning of FIG.

As shown in FIG. 5 , the metal plate 31 may be a copper plate made of copper in contact with the solidification cell 81 .

In addition, as shown in FIG. 5 , the thermocouples G are installed in n rows and m columns installed on the rear surface of the metal plate 31 to measure the temperature of each part of the metal plate 31 in real time.

Therefore, using the thermocouples G, the temperature maps M1, M2, and M3 show the thermocouples (M1), M2, and M3 for each row so that the low temperature section can be displayed in the form of a cold spot (CS). It is possible to collect a thermal image representing the temperature state over time of G) in color.

6 is a conceptual diagram conceptually showing the learning mode of the deep learning-based object detection model of the continuous casting freeway crack detection system 1000 through image learning of FIG. 4, and FIG. 7 is continuous casting through image learning of FIG. It is a formula showing an example of a loss function applied in the learning mode of the deep learning-based object detection model of the free longitudinal crack detection system 1000, and FIG. 8 is a continuous casting crack detection system through image learning of FIG. 1000) is a conceptual diagram conceptually showing the determination mode of the deep learning-based object detection model.

As shown in FIGS. 6 to 8, the deep learning-based object detection model is a CNN that can operate in a machine learning mode that learns an optimization weight using a loss function and an object detection mode that determines a crack by duty-free operation ( Convolutional Neural Network).

As shown in FIGS. 6 and 7 , in the case of the machine learning mode, a loss function that minimizes the deviation of the result value by repeatedly changing and applying the weight according to the learning target using the CNN may be applied. Here, this loss function can use the temperature map set for training to specify the coordinates and size of the crack part of the image of the temperature map in advance, and iteratively learns these data to learn discriminant power with high consistency according to the determined weight value. have.

However, the loss function is not limited to the drawings, and various types of loss functions that are easy to detect vertical cracks may be applied.

More specifically, for example, as shown in FIG. 6 , the deep learning-based object detection model is at least one of a convolution layer 311 , a pooling layer 312 , an activation function 313 , and combinations thereof. A feature extraction unit 310 that extracts a feature map from a plurality of the temperature maps M1, M2, and M3 including one or more, and at least a deconvolution layer 321, an unpooling layer 322, and activation A function 323 and any one or more of combinations thereof, the restoration unit 320 for restoring the defect feature map to a restored image, and the feature of the restored image so as to display the position or size of the vertical crack. A classifier 330 for classifying may be included.

More specifically, for example, as shown in FIGS. 6 to 8 , the convolution layer 311 (convolution layer) of the feature extraction unit 310 convolves an input value with a filter to form a smaller size. A layer capable of extracting a feature map of , the pooling layer 312 is a layer capable of subsampling data as much as a pooling kernel size in the feature map into data of a smaller size, and the activation function 313 is It may be a function that additionally activates data after a convolution operation.

Also, for example, the deconvolution layer 321 of the reconstruction unit 320 is a layer capable of extracting a feature map of a larger size by deconvolving an input value with a filter, and the unpooling layer 322 ) is a layer capable of upsampling data corresponding to the size of the pooling kernel in the feature map to data of a larger size, and the activation function 323 may be a function that additionally activates data after a convolution operation.

For example, as shown in FIGS. 6 and 8 , the convolutional layer 311 of the feature extraction unit 310 may include a first convolutional layer and a second convolutional layer. In this case, the second convolutional layer exists behind the first convolutional layer. That is, the second convolutional layer may use the output of the first convolutional layer.

More specifically, the first convolution layer may extract a feature map based on a convolution filter kernel having a size smaller than the size of the input original image. For example, the convolutional filter kernel is a matrix having a size smaller than the original size. In this case, the first convolutional layer may extract the feature map by multiplying the original image while moving the convolutional filter kernel. In this case, after the convolution operation is completed, the activation function 313 may be additionally passed through the activation function 313, for example, ReLU, for data activation.

After passing through the convolution layer, it may pass through the pooling layer 312 in order to reduce the amount of data and computation.

In this case, the pooling layer 312 may subsample data corresponding to the size of a pooling kernel (eg, n x m) in the feature map into data having a smaller size.

In this way, the CNN of the feature extraction unit 310 may perform an operation by repeatedly passing through the plurality of convolutional layers 311 and pooling layers 312 .

Also, for example, the deconvolution layer 321 of the reconstruction unit 320 may include a first deconvolution layer and a second deconvolution layer. In this case, the second deconvolution layer exists behind the first deconvolution layer. That is, the second deconvolution layer may use the output of the first deconvolution layer.

More specifically, the first deconvolution layer may extract a feature map having a size wider than that of the input feature map. In this case, the first deconvolution layer may extract the feature map by multiplying the original image while moving the deconvolution filter kernel. In this case, after the deconvolution operation is completed, the activation function 323 may be additionally passed through the activation function 323, for example, ReLU, for data activation.

After passing through the deconvolution layer, it may pass through the unpooling layer 322 in order to increase the amount of data and computation.

In this case, the unpooling layer 322 may upsample data corresponding to the size of the unpooling kernel in the feature map into data of a larger size.

In this way, the CNN of the reconstruction unit 320 may perform an operation by repeatedly passing through the plurality of deconvolution layers 321 and unpooling layers 322 .

Therefore, by repeatedly learning the above-described reconstructed images and the interpolated images using the feature extraction unit 310, the restoration unit 320, and the classifier 330, the defect of the product is finally classified with high accuracy. and can be determined with precision.

9 is a flowchart illustrating a method for detecting longitudinal cracks in continuous casting through image learning according to some embodiments of the present invention.

4 to 9, the method for detecting longitudinal cracks in continuous casting through image learning according to some embodiments of the present invention, (a) a metal plate 31 in the mold 30 in the continuous casting process (b) generating a temperature map (M1) (M2) (M3) of the metal plate 31 in units of slabs (S), (c) based on deep learning, It may include the steps of determining whether cracks occur in the temperature map (M1) (M2) (M3) and (d) performing a detailed inspection on the corresponding slab (S) determined as the occurrence of cracks in the duty-free direction. .

Here, in the step (c), if the number of cold spots (CS) in the same column of each row and thermocouple is determined more than the reference number by using a deep learning-based object detection model, it can be determined as a duty-free crack occurrence.

For example, if it is determined as a cold spot three or more times in the same column of each row and thermocouple by the deep learning-based object detection model, it is finally determined as a duty-free crack occurrence, and a follow-up, detailed inspection can be performed.

That is, in the step (d), the detailed examination may be a mill scale examination for removing the oxidized surface of the slab (S) and examining the epidermis.

Therefore, the temperature maps M1, M2, and M3 according to time of each row thermocouple G installed in the copper plate 31 of the mold 30 are formed, and deep learning is performed using the deep learning-based object detection model. By learning, it is possible to improve the consistency of crack detection by duty-free, and through this, the quality of cast steel can be greatly improved by performing selective mill scale inspection.

Although the present invention has been described with reference to the embodiment shown in the drawings, which is merely exemplary, those skilled in the art will understand that various modifications and equivalent other embodiments are possible therefrom. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the appended claims.

10: ladle
20: tundish
30: mold
60: support roll
65: spray means
M: molten steel
15: shroud nozzle
21: stopper
25: immersion nozzle
81: solidified shell
82: unsolidified molten steel
83: tip
91: cut point
S: Slav
31: metal plate
M1, M2, M3: temperature map
G: thermocouple
CS: Cold Spot
100: temperature collecting device
200: temperature map generating device
300: tax-free crack determination device
310: feature extraction unit
320: restoration unit
330: classifier
400: information output device
1000: duty free crack detection system

Claims (9)

(a) collecting the temperature of the metal plate in the mold in real time in the continuous casting process;
(b) generating a temperature map of the metal plate in units of slabs;
(c) determining whether a crack occurs in the temperature map based on deep learning; and
(d) performing a detailed inspection on the slab determined to be tax-free crack generation;
Containing, a method for detecting longitudinal cracks in continuous casting through image learning. The method of claim 1,
In the step (a), the metal plate is a copper plate in contact with the solidification cell, a method for detecting longitudinal cracks in continuous casting through image learning. The method of claim 1,
In the step (a), the temperature of each part of the metal plate is measured in real time by using thermocouples installed in n rows and m columns installed on the back side of the metal plate, a continuous casting crack detection method through image learning. 4. The method of claim 3,
In step (b), the temperature map is a thermal image image indicating the temperature state over time of the thermocouples for each row in color so that the low temperature section can be displayed in the form of a cold spot. Crack detection method in duty-free continuous casting through learning. 5. The method of claim 4,
In step (c), if the number of cold spots or more is determined as a cold spot in the same column of each row and thermocouple using a deep learning-based object detection model in step (c), it is determined that a crack occurs in a duty-free manner. detection method. The method of claim 1,
The deep learning-based object detection model includes a CNN (Convolutional Neural Network) that can operate in a machine learning mode that learns an optimization weight using a loss function and an object detection mode that determines cracks in duty-free image learning Crack detection method by duty-free continuous casting through 7. The method of claim 6,
The deep learning-based object detection model is
a feature extraction unit for extracting a feature map from the plurality of temperature maps including at least any one of a convolution layer, a pooling layer, an activation function, and combinations thereof;
a reconstruction unit for reconstructing the defect feature map into a reconstructed image including at least one of a deconvolution layer, an unpooling layer, an activation function, and combinations thereof; and
a classifier for classifying features of the reconstructed image so as to indicate the position or size of the vertical crack;
Containing, a method for detecting longitudinal cracks in continuous casting through image learning. The method of claim 1,
In the step (d), the detailed inspection is a mill scale inspection that removes the oxidized surface of the slab and inspects the epidermis, a method for detecting cracks in continuous casting through image learning. a temperature collecting device that collects the temperature of the metal plate in the mold in real time in the continuous casting process;
a temperature map generating device for generating a temperature map of the metal plate in units of slabs;
a vertical crack determination device for determining whether a vertical crack occurs in the temperature map based on deep learning; and
an information output device for outputting target slab information so as to perform a detailed inspection on the corresponding slab determined to be tax-free cracking;
Including, longitudinal crack detection system of continuous casting through image learning.

Images