JP7788904B2 - Meandering monitoring method and system - Google Patents

Description

The present invention relates to a meandering monitoring method and a meandering monitoring system.

When threading strips through a continuous annealing furnace, they may meander (sometimes called "walk") within the furnace. If the amount of meandering (the extent of the meandering) is excessive, it can lead to the strip coming into contact with the furnace wall or breaking. For this reason, it is necessary to suppress the meandering at the latest before the amount of meandering becomes significant.

Generally, using high strip tension can prevent meandering, but it can also cause strip buckling (wrinkles appearing in the strip's widthwise center in the running direction, leading to breakage in severe cases; this is sometimes referred to as heat buckling). For this reason, the strip tension is set to a level that prevents buckling, and when meandering occurs, operations such as reducing the strip running speed to suppress meandering (threading speed reduction operation) or using steering rolls to center the strip on the rolls to suppress meandering (steering roll operation) are sometimes performed. However, because these operations are performed when meandering occurs, i.e., after it has occurred, the amount of meandering may already be significant by the time they are performed, and they cannot fully prevent the strip from contacting the furnace wall or breaking.

In addition, while it is desirable for the strip running through the continuous annealing furnace to be dead flat (precisely flat), in reality, C-warp and unevenness (described below) may occur.

Patent Documents 1 and 2 disclose technology for detecting heat buckles based on image signals obtained by irradiating a strip with a linear laser beam in the width direction and capturing an image of the irradiated area. Patent Documents 1 and 2 mention C-warp of the strip as a noise component in detecting heat buckles. Patent Document 3 discloses technology for correcting predicted meandering of a strip in a continuous annealing furnace using data that combines a meandering calibration amount based on the amount of meandering of the strip obtained by capturing an image of the strip running near the steering roll in the continuous annealing furnace, and a meandering calibration amount based on the analysis results of the shape of the strip after cold rolling before the continuous annealing process.

Japanese Patent Application Publication No. 63-062825 Japanese Unexamined Patent Publication No. 63-274806 Japanese Patent Application Publication No. 5-017831

However, Patent Documents 1 and 2 do not describe the occurrence of meandering or the prevention of meandering, and the techniques disclosed therein naturally cannot predict the occurrence of meandering in a continuous annealing furnace. The technique of Patent Document 3 predicts the occurrence of meandering as a prerequisite for correcting meandering. However, since the shape of the strip after cold rolling is flattened (heat flattened) in the continuous annealing furnace and the degree of flattening is not uniform (because flattening sometimes progresses and sometimes does not), the technique of Patent Document 3 cannot accurately predict the occurrence of meandering. In other words, the conventional techniques listed in the patent documents have a problem in that they cannot accurately predict the occurrence of meandering in a continuous annealing furnace. Note that the techniques of Patent Documents 1 and 2 use linear laser light, making it difficult to detect irregularities (described below) that may occur in the strip. Furthermore, the technique of Patent Document 3 directly detects the amount of meandering of the strip in a continuous annealing furnace, but does not mention irregularities (described below) that may occur in the strip in a continuous annealing furnace.
An object of the present invention is to provide a technique for solving the above-mentioned problems, that is, a technique for predicting the occurrence of meandering of a strip in a continuous annealing furnace simply and with higher accuracy.

One aspect of the present invention is a meandering monitoring method for monitoring the occurrence of meandering in a strip traveling within a continuous annealing furnace, comprising: an imaging step of imaging an area of the strip in the width direction thereof that is equal to or greater than the width of the strip; a light emission step of illuminating the strip with diffused light so that the imaging step can capture a highlight portion that extends in the width direction of the strip as a reflection shape of light reflected from the surface of the strip; and a meandering monitoring step of monitoring the occurrence of meandering in the strip using the image captured in the imaging step, wherein the meandering monitoring step predicts the occurrence of meandering in the strip based on the output results of a trained model trained using the image and training data on the presence or absence of meandering in the strip.

Another aspect of the present invention is a meandering monitoring system that monitors the occurrence of meandering in a strip traveling within a continuous annealing furnace, comprising: an imaging means that captures an image of an area of the strip in the width direction that is equal to or greater than the width of the strip; a light-emitting means that illuminates the strip with diffused light so that the imaging means can capture a highlight portion that extends in the width direction of the strip as a reflection shape of light reflected from the surface of the strip; and a meandering monitoring means that monitors the occurrence of meandering in the strip using the image captured by the imaging means, wherein the meandering monitoring means predicts the occurrence of meandering in the strip based on the determination results of a trained model that has trained using the image and the presence or absence of meandering in the strip as training data.

The present invention makes it possible to predict the occurrence of strip meandering in a continuous annealing furnace easily and with greater accuracy.

1 is a diagram illustrating an example of the configuration of a meandering monitoring system 1. FIG. 10 is an example of an image. FIG. 1 is a diagram illustrating a trained model 300. FIG. 2 is a diagram illustrating an example of the configuration of a model generating device 30. 4 is a flowchart showing an example of the operation of the meandering monitoring device 40. 10A and 10B are diagrams illustrating an example of an evaluation result of the meandering monitoring system 1. 1 is a diagram for confirming the features of the meandering monitoring system 1. FIG.

Embodiments of the present invention will be described in detail below with reference to the drawings. Figure 1 is a diagram showing an example configuration of a meandering monitoring system 1. Figure 2 shows an example image. Specifically, each of Figures 2(A) to 2(C) is a portion of an image captured by the imaging device 10 (an image after cropping by the meandering monitoring device 40). Figure 3 is a diagram explaining a trained model 300. The meandering monitoring system 1 is a system that monitors the occurrence of meandering in a strip traveling within a heat treatment furnace (continuous annealing furnace, for example, a vertical continuous annealing furnace) in a line where strips (steel plates) are welded and continuously processed.

The width of the strip traveling through the continuous annealing furnace is, for example, 800 mm to 1200 mm. The thickness of the strip traveling through the continuous annealing furnace is, for example, 0.50 mm or less (the lower limit is approximately 0.10 mm). The traveling speed of the strip is, for example, 200 (m/min) to 500 (m/min). Note that the strip may have been cold-rolled, for example, at a cold rolling reduction ([cold-rolled base sheet thickness - steel sheet thickness] / cold-rolled base sheet thickness) of 80% or more in a process prior to entering the continuous annealing furnace, and then cleaned with an alkaline cleaning solution or the like. The strip may also be tinplate.

Meandering monitoring system 1 predicts the occurrence of meandering in the strip. Specifically, meandering monitoring system 1 predicts the occurrence of meandering of an amount equal to or greater than a danger level in the strip. A danger level of meandering may be, for example, the amount of meandering that would cause the strip to come into contact with the furnace wall. For example, if the amount of meandering that would cause the strip to come into contact with the furnace wall is 200 mm, a danger level of meandering may be 50 mm. In other words, meandering monitoring system 1 predicts an increase in the amount of meandering (to 50 mm or more) before the strip begins to meander (when the amount of meandering is 0 mm) or when the amount of meandering is small (when the amount of meandering is less than approximately 50 mm). When meandering monitoring system 1 predicts the occurrence of meandering in the strip (specifically, the occurrence of meandering of an amount equal to or greater than a danger level; the same applies below), it outputs an alarm.

As shown in FIG. 1, the meandering monitoring system 1 includes an imaging device 10, a light source 20, a meandering monitoring device 40, and a display device 50. For ease of explanation, FIG. 1 also shows a trained model 300 (described below) stored in the meandering monitoring device 40 (trained model storage unit 149).

The imaging device 10 is an in-furnace camera that captures images of the strip inside a continuous annealing furnace. The imaging device 10 captures an area of the strip that is equal to or greater than the width of the strip in the width direction (for example, if the width is 1000 mm, then 1000 mm or greater). In other words, the imaging device 10 is installed at a position/angle that allows it to capture an area of the strip that is equal to or greater than the width of the strip in the width direction.

The imaging device 10 may also be configured to capture an image of an area that is 0.4 times the width of the strip in the strip threading direction (traveling direction) (for example, 400 mm or more when the strip width is 1000 mm). In other words, the imaging device 10 may be installed at a position/angle that allows it to capture an image of an area that is equal to or greater than the width of the strip in the strip width direction and 0.4 times the width of the strip in the strip threading direction.

The imaging device 10 may capture images of the strip on the roll (the strip wound around the roll), or the strip between the rolls (the strip between the upper and lower rolls in a vertical continuous annealing furnace). Imaging the strip between the rolls is preferable because it allows for more pronounced imaging of any irregularities (uneven portions) that may occur in the strip.

The imaging method using the imaging device 10 may involve, for example, capturing video at a predetermined frame rate (e.g., 1 frame per 1/30 seconds). The imaging device 10 transmits (outputs) the captured images to the meandering monitoring device 40 via a wired or wireless connection. For example, the imaging device 10 transmits the captured images to the meandering monitoring device 40 every frame (1/30 seconds) (in real time).

The imaging device 10 may be a dedicated device for transmitting captured images to the meandering monitoring device 40, or it may be a device that transmits captured images to both the meandering monitoring device 40 and another device (for example, an in-furnace video display device (in-furnace video monitor) not shown). That is, the imaging device 10 may transmit captured images exclusively to the meandering monitoring device 40, or it may be a device that transmits captured images to both the meandering monitoring device 40 and another device. In the latter case, for example, a distributor (not shown) or the like may be used to transmit captured images to both the meandering monitoring device 40 and another device, or both the meandering monitoring device 40 and another device may be specified as destinations (destinations) for the captured images in the imaging device 10, so that captured images are transmitted to both the meandering monitoring device 40 and another device.

The light source 20 is a diffuse light source (excluding directional laser light, e.g., linear laser light) such as an LED or incandescent lamp. The light source 20 illuminates (irradiates) the strip with diffused light. The light source 20 illuminates the strip with diffused light so that the light reflected from the strip surface can be captured by the imaging device 10; specifically, so that the shape of the light reflected from the strip surface (reflection shape) can be captured by the imaging device 10 as a highlight portion that extends across the width of the strip.

It is known that strips passing through continuous annealing furnaces have wavy shapes (strips exhibiting wavy, uneven shapes) such as C-warp, edge waves, and center roll. The uneven shapes (e.g., symbols B and C in FIG. 2(B)) constituting the wavy shapes can occur at various locations (locations where convex portions or concave portions occur), such as the center in the sheet width direction, one end in the sheet width direction, or the other end in the sheet width direction. Furthermore, the uneven shapes may have, for example, a steep or gentle slope of the slope of the convex portions or concave portions, a wide or narrow region where the convex portions or concave portions are formed, or one or more concave portions or convex portions occurring in the sheet width direction. In other words, the uneven shapes of strips have various characteristics, such as the location where they occur, the slope, the region where they are formed, and the number of concave portions or convex portions occurring.
Regarding the occurrence of meandering, for example, there are cases where meandering is observed when an uneven shape is present at the end of the plate in the plate width direction, and cases where meandering is observed when an uneven shape is present in the center of the plate in the plate width direction. The inventors of the present application have found that it is difficult to attribute this to a specific causal relationship, and have considered that the various characteristics described above interact with each other to determine whether or not meandering occurs.
Therefore, the inventors of the present application considered using a linear laser light as described in Patent Documents 1 and 2 to grasp the unevenness of the wavy shape of the strip in the plate width direction and detect the occurrence of meandering, but found that there were problems with practical use due to non-detection, false detection, and over-detection occurring.

Buckling generally appears wavy in the width direction of the plate (the surface moves up and down in the width direction) and may involve plastic deformation, but does not appear wavy in the running direction (the surface moves up and down in the running direction). In contrast, unevenness appears wavy in both the width and running directions, but does not generally involve plastic deformation.

The reflection shape that appears in the captured image captured by the imaging device 10 changes depending on the uneven shape of the strip, even if the position/angle of the light source 20 is constant (even if the light source 20 is fixed). In other words, the meandering monitoring device 40 can recognize various characteristics of the uneven shape of the strip (such as the steepness of the above-mentioned concave and convex parts) from the reflection shape in the captured image.
The inventors of the present application have found a relationship between the occurrence of irregularities on the surface of the strip and the occurrence of meandering of the strip, and have focused on the details of the irregular shape of the strip in monitoring (predicting) the occurrence of meandering of the strip.

The image shown in Figure 2(A) is an image taken when the surface of the strip is flat (without irregularities). The image shown in Figure 2(A) captures a highlight portion (designated A in the figure) that extends across the width of the strip as a reflection from the strip surface. In other words, when the surface of the strip is flat, the light source 20 is installed at a position/angle that allows the imaging device 10 to capture a highlight portion that extends across the width of the strip as a reflection from the strip surface.

The image shown in Figure 2(B) shows an image of a strip with unevenness on its surface, resulting in a strip with an undulating edge shape. In the image shown in Figure 2(B), the symbol B indicates the reflection shape caused by the convex portions extending in the width direction and the running direction, and the symbol C indicates the reflection shape caused by the concave portions extending in the width direction and the running direction.

The image shown in Figure 2(C) is an image of a strip with unevenness on its surface. While the image shown in Figure 2(B) captures the reflection shape caused by simple convex and concave portions, the image shown in Figure 2(C) captures the reflection shape caused by complex convex and concave portions. In the image shown in Figure 2(C), symbol D indicates the reflection shape caused by convex portions extending in the strip width direction and strip running direction, and symbol E indicates the reflection shape caused by concave portions occurring near the tops of the convex portions.

Also, as shown in Figures 2(B) and 2(C), multiple irregularities may be observed in relatively close proximity. Furthermore, there are various variations in the shape of the irregularities, such as the location where they occur, the degree of inclination, the area where they are formed, and the number of occurrences.

Note that the simple unevenness shown in FIG. 2(B) and the complex unevenness shown in FIG. 2(C) may be distinguished, with the former referred to as a simple unevenness (simple uneven portion) and the latter referred to as a (complex unevenness) complex uneven portion. Furthermore, a depression (complex uneven portion) occurring at the top of a protrusion, such as that indicated by symbol E in FIG. 2(C), continuously appears and disappears, resulting in a motion that can be described as "(the strip) flapping" when viewed in a video. The present inventors have discovered that the presence of a complex uneven shape, such as that shown in FIG. 2(C), is closely related to the occurrence of meandering. Therefore, they have noticed that the occurrence of meandering is related not only to the wave shape (unevenness) of the strip in the strip width direction, but also to the size of the area where the wave shape (unevenness) occurs in the strip traveling direction and the state in which the unevenness intersects, i.e., the occurrence of a complex uneven portion. They have noticed that detecting the wave shape in the strip traveling direction also increases the accuracy of meandering detection.
The above is based on the findings of observations of actual situations, where, when complex unevenness (fluttering) occurs, the shape of the unevenness changes, such as the area where unevenness occurs in the strip traveling direction expanding, or the appearance or disappearance of complex shapes such as recesses near the tops of the protrusions, and the knowledge gained about the changes in the shape of the unevenness that are specific to such fluttering cannot be confirmed unless the unevenness in both the strip width direction and the strip traveling direction is understood.
It should be noted that the methods described in Patent Documents 1 to 3 do not evaluate the wavy shape in the entire width direction of the sheet and in the strip running direction in a continuous annealing furnace, that is, they do not evaluate the shape specific to the complex uneven portion, and therefore it was considered that the accuracy of detecting meandering occurrence is low.

The meandering monitoring device 40 monitors (predicts) the occurrence of meandering in the strip traveling through the continuous annealing furnace based on the image captured by the imaging device 10 (specifically, the reflection shape corresponding to the surface shape of the strip). The meandering monitoring device 40 is configured, for example, by a single server or a single personal computer. As shown in FIG. 1, the meandering monitoring device 40 includes an acquisition unit 141, a cutout unit 142, a meandering monitoring unit 143, a graphing unit 147, an output unit 148, and a trained model storage unit 149. The meandering monitoring unit 143 is configured by a probability calculation unit 144, an average value calculation unit 145, and a determination unit 146.

The acquisition unit 141 acquires captured images from the imaging device 10 and outputs (supplies) the acquired captured images to the cropping unit 142. Specifically, the acquisition unit 141 outputs the acquired captured images to the cropping unit 142 each time it acquires an image from the imaging device 10. For example, when acquiring captured images in frame units (every 1/30 seconds), the acquisition unit 141 outputs the acquired captured images to the cropping unit 142 every 1/30 seconds. Note that, while it has been explained that the imaging device 10 transmits captured images to the meandering monitoring device 40 via a wired or wireless connection, the acquisition unit 141 acquires captured images in a reception mode (wired or wireless) that corresponds to the transmission mode (wired or wireless) of the imaging device 10.

The cropping unit 142 acquires a captured image from the acquisition unit 141, crops out a still image from the acquired captured image, trims the cropped still image, and outputs the cropped still image (such as the image shown in Figures 2(A) to 2(C)) to the meandering monitoring unit 143. Each time the cropping unit 142 acquires a captured image from the acquisition unit 141 (for example, every 1/30 seconds), the cropping unit 142 crops out a still image, trims the cropped still image, and outputs the cropped still image to the meandering monitoring unit 143. Note that the cropping unit 142 may crop out a still image to a trimming size and output the cropped still image to the meandering monitoring unit 143 each time it acquires a captured image from the acquisition unit 141.

The trained model storage unit 149 stores the trained model 300. The trained model storage unit 149 may be a hard disk or SSD (Solid State Drive), etc. The trained model 300 is generated by the model generation device 30 (described below) and stored in the trained model storage unit 149 via communication or via a storage medium (CD-ROM, USB memory, etc.). Note that a trained model represented by parameters (trained parameters) optimized through training (learning) is referred to as a trained model. A trained model is generated by training a training model and optimizing the parameters through training.

The trained model 300 inputs an image captured by the imaging device 10 (specifically, an image processed by the cropping unit 142) and outputs a determination result regarding the occurrence of meandering in the strip. Specifically, the trained model 300 is a model that can output, as a determination result, the probability (probability of meandering occurrence) that meandering will occur in the strip (specifically, meandering with an amount of meandering equal to or greater than the danger level) each time an image is acquired from the cropping unit 142.

The trained model 300 is a model represented using a neural network, for example, as shown in FIG. 3. The trained model 300 may be, for example, a DNN (Deep Neural Network). The trained parameters include, for example, the number of layers of the neural network, the number of neurons in each layer, the connection relationships between neurons, the connection weights (connection loads) between each neuron, and the threshold value of each neuron. The trained model 300 shown in FIG. 3 includes an input layer 301, one or more intermediate layers 302 (hidden layers), and an output layer 303. Each layer 301, 302, 303 has one or more neurons. The number of intermediate layers 302, the number of neurons in each layer, etc. are set by the trained parameters.

The image processed by the cropping unit 142 is input to the input layer 301. The output layer 303 outputs output values indicating the probability that meandering will occur in the strip (0.00 to 1.00) and the probability that meandering will not occur in the strip (0.00 to 1.00). The sum of the probability that meandering will occur in the strip and the probability that meandering will not occur in the strip is 1. Note that a probability of "0.00" is 0%, and a probability of "1.00" is 100%.

As mentioned above, the trained model 300 is generated by the model generation device 30 (described below), which will be described later.

The meandering monitoring unit 143 monitors the occurrence of meandering of the strip based on the captured image captured by the imaging device 10 (specifically, the image after processing by the cropping unit 142) and the trained model 300 stored in the trained model storage unit 149.

The probability calculation unit 144 of the meandering monitoring unit 143 acquires an image from the clipping unit 142, calculates the probability that meandering will occur in the strip using the trained model 300, and outputs the calculated probability to the average calculation unit 145. In other words, the probability calculation unit 144 inputs the image acquired from the clipping unit 142 into the trained model 300, acquires the probability that meandering will occur in the strip as an output from the trained model 300, and outputs the acquired probability to the average calculation unit 145. The probability calculation unit 144 calculates (acquires) the probability each time it acquires an image from the clipping unit 142 (for example, every 1/30 seconds), and outputs it to the average calculation unit 145.

The average value calculation unit 145 of the meandering monitoring unit 143 acquires the probabilities from the probability calculation unit 144, calculates a moving average (moving average value) of the acquired probabilities, and outputs the calculated moving average value to the determination unit 146 and the graphing unit 147. The average value calculation unit 145 calculates the moving average value of the immediately preceding predetermined number of probabilities (e.g., 10 to 12,000) including the currently acquired probability. In other words, the average value calculation unit 145 calculates the moving average value of the probabilities calculated from each of the immediately preceding predetermined number of images (10 to 12,000 images) including the currently acquired image by the probability calculation unit 144. When the probability calculation unit 144 calculates probabilities every 1/30 seconds, the average value calculation unit 145 calculates the moving average value of the probabilities calculated from each of the images (10 to 12,000 images) in the time range of 1/3 second (10 images) to 400 seconds (12,000 images). The average value calculation unit 145 calculates a moving average value each time it obtains a probability from the probability calculation unit 144 (for example, every 1/30 seconds), and outputs the calculated moving average value to the determination unit 146 and graphing unit 147.

The determination unit 146 of the meandering monitoring unit 143 obtains the moving average value from the average value calculation unit 145, compares the obtained moving average value with a predetermined threshold, and monitors the occurrence of meandering of the strip. Specifically, if the moving average value exceeds the predetermined threshold, the determination unit 146 outputs alarm information to the output unit 148 to warn of meandering, and if the moving average value is equal to or less than the predetermined threshold, the determination unit 146 does not output alarm information to the output unit 148. Each time the determination unit 146 obtains the moving average value from the average value calculation unit 145 (for example, every 1/30 seconds), it compares the moving average value with the predetermined threshold, and if the moving average value exceeds the predetermined threshold, it outputs alarm information to the output unit 148.

The control parameters that control the operation of the meandering monitoring unit 143 (specifically, the range (number of probabilities/number of images/time) over which the average value calculation unit 145 calculates the moving average value, and the predetermined threshold value that the determination unit 146 compares with the moving average value) can be set as appropriate. For example, the control parameters may be determined depending on the frequency of false alarms (when an alarm is output but no meandering actually occurred) and the acceptable range for false alarms. As one example, 30 probabilities (i.e., 30 images equivalent to 1 second) may be set as the range over which the average value calculation unit 145 calculates the moving average value, and 0.5 may be set as the predetermined threshold value that the determination unit 146 compares with the moving average value.

Note that if the average value calculation unit 145 sets 30 probabilities as the range for calculating the moving average value, and the determination unit 146 sets 0.5 as the predetermined threshold for comparison with the moving average value, even if the average value calculation unit 145 continues to obtain a probability of "0.00" from the probability calculation unit 144 before a certain timing (timing T) and continues to obtain a probability of "1.00" from the probability calculation unit 144 after timing T, the moving average value calculated by the average value calculation unit 145 will exceed "0.5" when the 16th probability of "1.00" is obtained from timing T. In other words, as is natural given the nature of moving average values, when the determination unit 146 determines that the moving average value exceeds the predetermined threshold, it will not suddenly exceed the predetermined threshold, but will approach the predetermined threshold (or sometimes move away from it) before exceeding the predetermined threshold.

The graphing unit 147 acquires the moving average value from the average value calculation unit 145, generates graphing display information for displaying the acquired moving average value in a graph (e.g., a line graph or a bar graph), and outputs the graphing display information to the output unit 148. The graphing unit 147 outputs the graphing display information to the output unit 148 each time it acquires a moving average value from the average value calculation unit 145 (e.g., every 1/30 seconds).

The output unit 148 acquires graphing display information from the graphing unit 147 and outputs the acquired graphing display information to the display device 50. The output unit 148 outputs the graphing display information to the display device 50 every time it acquires graphing display information from the graphing unit 147 (for example, every 1/30 seconds). Furthermore, when the output unit 148 acquires alarm information from the determination unit 146, it outputs the acquired alarm information to the display device 50.

The display device 50 is an indicator monitor that displays indicators of strip meandering. Specifically, the display device 50 acquires graph display information from the output unit 148 and displays a graph (e.g., a line graph of moving average values, a bar graph of moving average values) based on the acquired graph display information. Furthermore, when alarm information is acquired from the output unit 148, the display device 50 displays an alarm based on the acquired alarm information (e.g., on the same screen as the graph). Note that when alarm information is acquired, the display device 50 may output an alarm as sound instead of or in addition to displaying the alarm.

As described above, the display device 50 always displays the graph (regardless of whether the moving average value is between 0.00 and 1.00), and outputs an alarm if alarm information is acquired. Upon checking the alarm, the operator can consider and implement measures to address the occurrence of meandering (for example, changing the strip threading speed (slowing down) or changing the strip tension (increasing tension)), as necessary.

Figure 4 is a diagram showing an example configuration of the model generation device 30. The model generation device 30 generates a trained model 300 to be used in the meandering monitoring device 40. The model generation device 30 is configured, for example, by a single server or a single personal computer. As shown in Figure 4, the model generation device 30 includes a model generation unit 133 and a storage unit 139. For ease of explanation, Figure 4 also shows a training dataset 250 (described below) and a trained model 300.

The model generation device 30 trains a learning model using the learning dataset 250 and generates a trained model 300. The learning dataset 250 includes input samples and output samples. The input samples are data that are input to the input layer (input layer 301 in Figure 3) when training the learning model. The output samples are data that serve as correct answers (also referred to as training data or correct answer labels) for comparison with output values from the output layer (output layer 303 in Figure 3) when training the learning model.

The input samples in the training dataset 250 are images based on images captured by the imaging device 10 (still images (e.g., cropped images) obtained from images captured by the imaging device 10). Specifically, the input samples are an image based on an image captured before actual meandering (specifically, meandering of an amount equal to or greater than the danger level) occurred (hereinafter referred to as an "image with meandering"), and an image based on an image captured when no meandering occurred (hereinafter referred to as an "image without meandering").

The output samples in the training dataset 250 are information regarding the presence or absence of meandering in the strip (presence or absence of meandering) for each input sample. Specifically, for the input sample "image with meandering," the output sample indicates the presence of meandering (information indicating the presence of meandering), and for the input sample "image without meandering," the output sample indicates the absence of meandering (information indicating the absence of meandering). In other words, the "input sample - output sample" pairs are "image with meandering - presence of meandering" and "image without meandering - absence of meandering."

It is believed that the percentage (possibility) of complex unevenness being captured as a strip reflection shape in an image with meandering is higher than the percentage (possibility) of complex unevenness being captured as a strip reflection shape in an image without meandering.

(Specific examples of the training data set 250 (input samples, output samples))
1,000 images (images with meandering) taken before an actual meandering (specifically, a meandering with a meandering amount equal to or greater than a danger level) occurred (at the time of the meandering occurrence) may be used as input samples, and information indicating the occurrence of meandering (information indicating the occurrence of meandering) may be attached to each of the input samples. For example, 1,000 images (for example, similar images may be aggregated to increase the variety and the 1,000 images may be selected) from 3,600 images (images with meandering) taken during a two-minute (120-second) period from a point in time (time Ta1) six minutes before a certain point in time (time Ta3) when meandering (e.g., a meandering amount of 50 mm or more) occurred to a point in time (time Ta2) four minutes before time Ta3 may be selected as input samples, and information indicating the occurrence of meandering may be attached to each of the output samples.

1,000 images (images without meandering) taken when no meandering occurs (when no meandering occurs) may be used as input samples, with "no meandering" (information indicating no meandering) attached to each output sample. For example, 1,000 images may be extracted from 3,600 images (images without meandering) taken in the first two minutes (two minutes from time Tb1) of any six-minute period (from time Tb1 to time Tb2) when no meandering occurs (for example, similar images may be aggregated to extract 1,000 images to increase variation) and used as input samples, with "no meandering" attached to each output sample.

Note that (N x 1000) images with meandering may be prepared as input samples from N (N is 2 or more) instances of meandering. When (N x 1000) images with meandering are prepared as input samples (images with meandering), it is preferable to prepare the same number of images, i.e., (N x 1000) images without meandering, as input samples.

The storage unit 139 stores the training dataset 250. The storage unit 139 may be a hard disk or a solid state drive (SSD). The training dataset 250 is stored in the storage unit 139 via communication or via a storage medium (CD-ROM, USB memory, etc.). The storage unit 139 also stores the training model and parameters.

The model generation unit 133 trains the learning model using the learning dataset 250 stored in the memory unit 139, and generates the trained model 300. For example, when the model generation unit 133 receives a model generation instruction operation (when the model generation instruction operation is received via an operation unit not shown, or when a model generation instruction command is received via a communication unit not shown), it trains the learning model using the learning dataset 250 stored in the memory unit 139, and generates the trained model 300. In other words, the trained model 300 is generated in the memory unit 139 by the model generation unit 133.

For example, the model generation unit 133 inputs each input sample to the input layer 301, updates the parameters of the learning model so as to reduce the difference between the output value obtained from the output layer 303 and each output sample (correct label), and generates the learned model 300. As an example, the model generation unit 133 may generate the learned model 300 using backpropagation, as shown in (1) to (5) below. Note that the learning model used by the model generation unit 133 may employ feedback alignment and target propagation in addition to backpropagation.

(1) The model generation unit 133 inputs input samples (image 1, image 2, image 3, ...) of the learning dataset 250 stored in the storage unit 139 to the input layer 301, and performs calculation processing in the forward propagation direction of the learning model to obtain output values (output value of image 1, output value of image 2, output value of image 3, ...) from the output layer 303. The output values from the output layer 303 are the probability that meandering will occur in the strip and the probability that meandering will not occur in the strip.
(2) For each input sample (image 1, image 2, image 3, ...), the model generation unit 133 calculates the error between the output value from the output layer 303 and the output sample, for example, by backpropagation. The output sample indicates whether meandering has occurred (information indicating the presence of meandering) or not (information indicating the absence of meandering). For example, if a certain image is an image with meandering, and the image is input to the input layer 301 and the output layer 303 outputs a probability of "1.00" that meandering will occur in the strip (and a probability of "0.00" that meandering will not occur in the strip), the error associated with the image is zero. Also, if a certain image is an image without meandering, and the image is input to the input layer 301 and the output layer 303 outputs a probability of "0.00" that meandering will occur in the strip (and a probability of "1.00" that meandering will not occur in the strip), the error associated with the image is zero.
(3) The model generation unit 133 determines whether the calculated error is within a predetermined value.
(4) If the calculated error is within a predetermined value, the model generation unit 133 determines that various parameters such as the connection weights between neurons (such as the connection weights between neurons and the thresholds of each neuron) have been optimized, and ends learning (training). In other words, the generation of the trained model 300 by the model generation unit 133 is completed (the trained model in which the various parameters are trained parameters is referred to as the trained model 300).
(5) If the calculated error is not within a predetermined value, the model generation unit 133 updates various parameters based on the calculated errors. Thereafter, the model generation unit 133 inputs the input sample to the input layer 301 again for the learning model whose various parameters have been updated, and repeatedly updates various parameters until the error falls within a predetermined value.

The trained model 300 generated in the memory unit 139 by the model generation unit 133 is stored in the meandering monitoring device 40 (trained model memory unit 149) via communication or via a storage medium.

Figure 5 is a flowchart showing an example of the operation of the meandering monitoring device 40. The flowchart in Figure 5 starts repeatedly (for example, every 1/30 seconds). The range over which the average value calculation unit 145 calculates the moving average value is assumed to be 30 probabilities. The predetermined threshold value that the determination unit 146 compares with the moving average value is assumed to be 0.5. The learned model storage unit 149 is assumed to store a learned model.

The acquisition unit 141 acquires a captured image (video) captured by the imaging device 10 (step S1). The acquisition unit 141 outputs the captured image to the cropping unit 142. Then, the process proceeds to step S2.

The cropping unit 142 acquires the captured image from the acquisition unit 141, crops a still image from the captured image, and trims it (step S2). The cropping unit 142 outputs the cropped still image to the probability calculation unit 144. The process then proceeds to step S3.

The probability calculation unit 144 acquires the image from the cropping unit 142 and uses the trained model 300 to calculate the probability that meandering will occur in the strip (meandering occurrence probability) (step S3). The probability calculation unit 144 outputs the calculated probability to the average value calculation unit 145. The process then proceeds to step S4.

The average value calculation unit 145 acquires the probabilities from the probability calculation unit 144 and calculates the moving average (moving average value) of the acquired probabilities (step S4). Specifically, the average value calculation unit 145 calculates the moving average value of the most recently acquired 30 probabilities. The average value calculation unit 145 outputs the calculated moving average value to the determination unit 146 and the graphing unit 147. Next, the process proceeds to step S5.

The graphing unit 147 acquires the moving average value from the average value calculation unit 145 and generates graph display information for the acquired moving average value (step S5). The graphing unit 147 outputs the generated graph display information to the output unit 148. Then, the process proceeds to step S6.

The output unit 148 acquires the graphing display information from the graphing unit 147 and outputs the acquired graphing display information to the display device 50 (step S5). The process then proceeds to step S7. Having acquired the graphing display information from the output unit 148, the display device 50 displays a graph based on the acquired graphing display information.

The determination unit 146 acquires the moving average value from the average value calculation unit 145 and determines whether the acquired moving average value exceeds a predetermined threshold value (0.5) (step S7). If the moving average value exceeds the predetermined threshold value (0.5) (step S7: YES), the process proceeds to step S8. If the moving average value does not exceed the predetermined threshold value (0.5) (step S7: NO), the process ends (returns to step S1).

The output unit 148 acquires the alarm information from the determination unit 146 and outputs the acquired alarm information to the display device 50 (step S8). This flowchart then ends (returns to step S1). Note that, having acquired the alarm information from the output unit 148, the display device 50 outputs (displays, outputs audio) an alarm based on the acquired alarm information.

Figure 6 is a diagram illustrating an example of the evaluation results of the meandering monitoring system 1. Specifically, Figure 6 shows the evaluation results regarding whether or not an alarm is output when meandering occurs during actual operation, divided by imaging condition (by imaging range in the strip running direction of the imaging device 10). The imaging conditions are the following four (imaging condition a to imaging condition d).

(Image capture condition a)
The imaging range of the imaging device 10 is set so that the width direction of the plate is equal to or greater than the plate width and the width in the plate passing direction is 0.6 times the plate width (for example, 600 mm when the plate width is 1000 mm) (imaging condition a).

(Image capture condition b)
The imaging range of the imaging device 10 is set so that the width direction is equal to or greater than the width of the plate and the width in the passing direction is 0.4 times the width of the plate (for example, 400 mm when the plate width is 1000 mm) (imaging condition b).

(Image capture condition c)
The imaging range of the imaging device 10 is set so that the width direction of the plate is equal to or greater than the plate width and the running direction of the plate is 0.2 times the plate width (for example, 200 mm when the plate width is 1000 mm) (imaging condition c). Note that the effective imaging range in the running direction may be set to 0.2 times the plate width by, for example, applying a mask in front of the lens of the imaging device 10.

(Image capture condition d)
The imaging range of the imaging device 10 is set so that the width direction of the sheet is equal to or greater than the sheet width and the width direction of the sheet is 0.02 times the sheet width (for example, 20 mm when the sheet width is 1000 mm) (imaging condition d). Note that the effective imaging range in the sheet passing direction may be set to 0.02 times the sheet width by, for example, applying a mask in front of the lens of the imaging device 10.

The model generation device 30 generates a trained model 300 for each imaging condition. For example, under imaging condition a, the model generation device 30 generates a trained model 300 (referred to as trained model 300a) using captured images acquired from the imaging device 10 set to imaging condition a, i.e., captured images with an imaging range equal to or greater than the plate width in the plate width direction and 0.6 times the plate width in the plate threading direction (using 1,000 images with meandering and 1,000 images without meandering, as shown in the "Specific Example" above). Under imaging condition b, the model generation device 30 generates a trained model 300 (referred to as trained model 300b) using captured images acquired from the imaging device 10 set to imaging condition b, i.e., captured images with an imaging range equal to or greater than the plate width in the plate width direction and 0.4 times the plate width in the plate threading direction (using 1,000 images with meandering and 1,000 images without meandering, as shown in the "Specific Example" above). Under imaging condition c, the model generation device 30 generates a trained model 300 (referred to as trained model 300c) using captured images acquired from the imaging device 10 set under imaging condition c, i.e., captured images with an imaging range equal to or greater than the plate width in the plate width direction and 0.2 times the plate width in the plate threading direction (using 1,000 images with meandering and 1,000 images without meandering, as shown in the "Specific Example" above). Under imaging condition d, the model generation device 30 generates a trained model 300 (referred to as trained model 300d) using captured images acquired from the imaging device 10 set under imaging condition d, i.e., captured images with an imaging range equal to or greater than the plate width in the plate width direction and 0.02 times the plate width in the plate threading direction (using 1,000 images with meandering and 1,000 images without meandering, as shown in the "Specific Example" above).

In the table in Figure 6, the leftmost column shows the imaging conditions. The second column from the left shows the imaging range of the imaging device 10 (specifically, how many times the strip width in the strip threading direction; note that in the strip width direction, it is equal to or greater than the strip width for all imaging conditions). The third column from the left shows the alarm occurrence rate (alarm display rate on the display device 50) at the time when meandering occurs (the time when the operator visually confirms the occurrence of meandering). The rightmost column shows the alarm occurrence rate three minutes before meandering occurs (three minutes before the operator visually confirms the occurrence of meandering).

(Evaluation of imaging condition a)
The meandering monitoring device 40 checked whether or not an alarm was issued when meandering occurred and three minutes before the occurrence of meandering (by comparing the moving average value of the meandering occurrence probability with a predetermined threshold) based on the output result (probability of meandering occurrence) obtained by inputting images captured by the imaging device 10 set as imaging condition a (images captured in an imaging range equal to or greater than the plate width in the plate width direction and 0.6 times the plate width in the plate threading direction) into the trained model 300a. As a result, as shown in Figure 6, an alarm was output 100% of the time when meandering occurred, and an alarm was output 83% of the time three minutes before meandering occurred.

Note that, common to all imaging conditions (imaging conditions a to d), the average value calculation unit 145 calculated the moving average value of the 30 meandering occurrence probabilities output from the probability calculation unit 144, and the determination unit 146 compared the moving average value output from the average value calculation unit 145 with 0.5 (a predetermined threshold value).

(Evaluation of imaging condition b)
The meandering monitoring device 40 checked whether or not an alarm was issued when meandering occurred and three minutes before the occurrence of meandering (by comparing the moving average value of the meandering occurrence probability with a predetermined threshold) based on the output result (probability of meandering occurrence) obtained by inputting images captured by the imaging device 10 set as imaging condition b (images captured in an imaging range equal to or greater than the plate width in the plate width direction and 0.4 times the plate width in the plate threading direction) into the trained model 300b. As a result, as shown in Figure 6, an alarm was output 100% of the time when meandering occurred, and an alarm was output 83% of the time three minutes before meandering occurred.

(Evaluation of imaging condition c)
The meandering monitoring device 40 checked whether or not an alarm was issued when meandering occurred and three minutes before the occurrence of meandering (by comparing the moving average value of the meandering occurrence probability with a predetermined threshold) based on the output result (probability of meandering occurrence) obtained by inputting images captured by the imaging device 10 set as imaging condition c (images captured in an imaging range equal to or greater than the plate width in the plate width direction and 0.2 times the plate width in the plate threading direction) into the trained model 300c. As a result, as shown in Figure 6, an alarm was output 60% of the time when meandering occurred, and an alarm was output 20% of the time three minutes before meandering occurred.

(Evaluation of imaging condition d)
In the case of imaging condition d, the imaging range in the strip passing direction was narrow, at 0.02 times the strip width, so the reflected light of the optical annular and semi-optical annular shapes described below could not be confirmed, and the highlight area in the strip width direction only appeared as a dotted line, and a valid training data set 250 (teacher data, etc.) was not obtained (and therefore a valid trained model 300d was not generated), making evaluation impossible. Note that it is thought that similar results would be obtained if laser light were used.

According to Figure 6, if the imaging range of the imaging device 10 is equal to or greater than the width of the plate in the plate width direction and equal to or greater than 0.4 times the width of the plate in the plate threading direction, it is possible to predict, for example, more than 80% of meandering that will occur after three minutes. Also, according to Figure 6, prediction performance remains the same whether the plate threading direction is 0.4 or 0.6 times the width of the plate.

Figure 7 is a diagram that clearly explains the features of the meandering monitoring system 1. Figure 7(A) shows a schematic diagram of unevenness occurring on the strip. In Figure 7(A), D shows a schematic diagram of unevenness occurring on the strip (a convex portion of a simple uneven portion).

Figures 7(B) and 7(C) schematically show the vertical position (direction perpendicular to the strip surface) of the strip surface where irregularities have occurred. Figure 7(B) shows the vertical position of the strip surface along the dashed-dotted line L1-R1 in the width direction shown in Figure 7(A). Figure 7(C) shows the vertical position of the strip surface along the diagonal dashed-dotted line L2-R2 in Figure 7(A). In Figures 7(B) and 7(C), d indicates the portion that crosses the irregularity D. In other words, Figure 7(B) is a schematic diagram of a cross section (cross section along dashed-dotted line L1-R1) of the strip including the irregularity D, and Figure 7(C) is a schematic diagram of a diagonal cross section (cross section along dashed-dotted line L2-R2) of the strip including the irregularity D.

Figure 7(D) shows a schematic diagram of wrinkles (buckling) occurring in a strip. In Figure 7(D), S indicates a wrinkle (an upward wrinkle) that has occurred in the strip. The dashed line within S indicates the end of the wrinkle, and the solid line indicates the top. In Figure 7(D), a cross section of the strip in the width direction including wrinkle S (cross section taken along dash-dotted line L1-R1) is shown approximately as in Figure 7(B). Also, in Figure 7(D), a diagonal cross section of the strip including wrinkle S (cross section taken along dash-dotted line L2-R2) is shown approximately as in Figure 7(C).

In other words, the surface of the strip may be uneven or wrinkled, but when the surface of the strip is viewed as a "line," whether in the width direction (dotted line L1-R1) or in a diagonal direction (dotted line L2-R2), it is difficult to distinguish whether the surface of the strip is uneven or wrinkled.

Figures 7(E) and 7(F) show the imaging range of the imaging device 10. Figure 7(E) shows the imaging range P of the imaging device 10 for the strip shown in Figure 7(A) (a strip with irregularities). Figure 7(F) shows the imaging range P of the imaging device 10 for the strip shown in Figure 7(C) (a strip with wrinkles). When the surface of the strip is captured as a "plane," it is possible to distinguish whether the surface of the strip is irregular or wrinkled.

In other words, while it is difficult to recognize unevenness when using laser light or the like, the meandering monitoring system 1 uses a light source 20 (diffuse light source) and an imaging device 10 (camera), making it easy to recognize unevenness. The meandering monitoring system 1 predicts the occurrence of meandering based on the output results (probability of meandering occurrence) of the trained model 300, which was trained using images with meandering, which are considered to have a relatively high probability of having unevenness captured as a reflection shape of the strip, and images without meandering, which are considered to have a relatively low probability of having unevenness captured as a reflection shape of the strip.

In this embodiment, if the strip surface is uneven, it is possible to capture characteristics such as highlight areas being interrupted halfway across the width of the strip (reflecting in a direction that cannot be captured), or taking on an annular or semi-annular shape depending on the condition of the curved surface of the unevenness, or the brightness of the reflected light gradually decreasing from the peak to the base of the curve. The size of the uneven area can be observed as the distance at which the size and brightness of the annular or semi-annular shape gradually decrease. In other words, if unevenness occurs in the strip (even if complex unevenness occurs in the strip, causing it to flutter), it is easy to detect that unevenness has occurred in the strip.

While this embodiment uses a light source 20, if the strip surface is as smooth as a mirror, it would be difficult to capture highlights. However, because the cold rolling process creates microscopic irregularities on the strip surface (irregularities expressed in terms of surface roughness or coarseness, which are different from the irregularities focused on in this embodiment), highlights can be captured (Figure 2). This embodiment uses a diffused light source (a light source with directional laser light, e.g., excluding linear laser light). However, if linear laser light is used instead of a diffused light source, even if irregularities exist on the strip, the high directivity prevents the appearance of optical annular or semi-optical annular features, and it is difficult to capture changes (decreases) in brightness corresponding to the steepness of the irregularities. Even if the width of the linear laser light irradiated in the strip width direction (the width of the light in the direction of strip travel) is increased, there is a limit to how wide it can be, resulting in the same results as above.

It is also possible to connect (combine) the "lines" created by the laser light to obtain "surface" information similar to the imaging range of the imaging device 10 (for example, 0.4 times the width of the strip in the strip running direction or more). However, considering the running speed of the strip (200 m/min or more) and the processing load and requirements for composition (real-time alarm notification), it is not easy to obtain "surface" information from the laser light similar to the imaging range of the imaging device 10.

Furthermore, as described above, the imaging device 10 of the meandering monitoring system 1 may be a device that transmits captured images to the meandering monitoring device 40 or to other devices (e.g., a furnace video monitor), and a camera already installed for other purposes (such as transmitting captured images to other devices) may be used as the imaging device 10. In other words, the meandering monitoring system 1 can monitor the occurrence of strip meandering without introducing new equipment such as laser light.

In this embodiment, an example has been described in which the meandering monitoring system 1 predicts (monitors) the occurrence of meandering on a strip with a meander amount above the danger level. However, the meandering monitoring system 1 may also predict the occurrence of meandering on a strip with a meander amount above the danger level. Furthermore, the meandering monitoring system 1 may predict the occurrence of meandering on a strip with a meander amount above a predetermined standard (or a meander amount exceeding the predetermined standard). The predetermined standard meander amount is a meander amount smaller than the danger level meander amount (i.e., zero meander amount (no meandering) < predetermined standard meander amount < danger level meander amount). Specifically, a training dataset corresponding to the meander amount to be predicted (meander amount above the danger level, meander amount above the predetermined standard, meander amount exceeding the predetermined standard) is prepared, trained data corresponding to the meander amount to be predicted is generated, and the trained model can be used to predict the occurrence of meandering of the desired meander amount.

The meandering monitoring system 1 according to this embodiment has been described above. The meandering monitoring system 1 makes it possible to predict the occurrence of meandering of strips in a continuous annealing furnace easily and with greater accuracy.

The embodiments of the present invention have been described above, but the embodiments of the present invention include at least the following configurations.
(1) A meandering monitoring method (a meandering monitoring method using a meandering monitoring system 1) for monitoring the occurrence of meandering of a strip traveling in a continuous annealing furnace, the meandering monitoring method comprising: an imaging step of imaging (using an imaging device 10) an area of the strip in the width direction of the strip that is equal to or greater than the width of the strip; a light emitting step of illuminating the strip with diffused light (diffused light by a light source 20) so that a highlight portion extending in the width direction of the strip can be imaged in the imaging step as a reflection shape of light reflected from the surface of the strip; and a meandering monitoring step of monitoring (using a meandering monitoring device 40) the occurrence of meandering of the strip using the image captured in the imaging step, wherein the meandering monitoring step predicts the occurrence of meandering of the strip based on an output result (probability of meandering occurrence) of a trained model (trained model 300) that has been trained using the image and the presence or absence of meandering of the strip, which is training data.
(2) A meandering monitoring system (meandering monitoring system 1) that monitors the occurrence of meandering of a strip traveling in a continuous annealing furnace, comprising: an imaging means (imaging device 10) that images an area of the strip in the width direction of the strip that is equal to or greater than the width of the strip; a light emitting means (light source 20) that illuminates the strip with diffused light so that the imaging means can image a highlight portion that extends in the width direction of the strip as a reflection shape of light reflected from the surface of the strip; and a meandering monitoring means (meandering monitoring device 40) that monitors the occurrence of meandering of the strip using the image captured by the imaging means, wherein the meandering monitoring system predicts the occurrence of meandering of the strip based on the judgment results of a trained model (trained model 300) that has learned the image captured and the presence or absence of meandering of the strip as training data.

In addition, in (1) and (2) above, "a trained model trained using the captured image and the training data of whether or not the strip meanders" may be rephrased as "a trained model that inputs the captured image and outputs a determination result regarding the occurrence of meandering in the strip." Since a trained model is a type of algorithm, it may also be rephrased as "an algorithm that is generated using the imaging results of the imaging step (the imaging means) including the reflection shape of the reflected light and the presence or absence of meandering in the strip as training data, and determines whether or not meandering in the strip occurs."

The above describes an embodiment, but the above embodiment is merely an example, and the specific configuration is not limited to the above embodiment, and includes designs that do not deviate from the gist of the invention.

For example, in the above embodiment, processing was performed every frame (1/30 seconds), but processing may also be performed every several frames.

Furthermore, in the above embodiment, an example was described in which each meandering monitoring device 40 was configured by one device (one server or one personal computer), but the meandering monitoring device 40 may be configured by two or more devices. Further, in the above embodiment, an example was described in which each model generation device 30 was configured by one device (one server or one personal computer), but the model generation device 30 may be configured by two or more devices.

In addition, in the above embodiment, an example was described in which the model generation device 30 and the meandering monitoring device 40 were configured as different devices, but the model generation device 30 and the meandering monitoring device 40 may be configured as the same device. In other words, the meandering monitoring device 40 may also have the functions of the model generation device 30 (or the model generation device 30 may also have the functions of the meandering monitoring device 40).

In addition, in the above embodiment, an example was described in which the meandering monitoring device 40 and the display device 50 were configured as different devices, but the meandering monitoring device 40 and the display device 50 may also be configured as the same device. In other words, the meandering monitoring device 40 may also have the functions of the display device 50.

The programs for implementing the meandering monitoring system 1, model generation device 30, and meandering monitoring device 40 described above may be recorded on a computer-readable recording medium and loaded into a computer system for execution. The term "computer system" as used herein includes hardware such as an OS and peripheral devices. Furthermore, "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into a computer system. Furthermore, "computer-readable recording medium" also includes devices that retain a program for a certain period of time, such as volatile memory (RAM) within a computer system that acts as a server or client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. The program may also be transmitted from a computer system storing the program in a storage device to another computer system via a transmission medium or by transmission waves within the transmission medium. Here, the "transmission medium" that transmits the program refers to a medium capable of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may also be used to implement some of the aforementioned functions. Furthermore, the above functions may be realized in combination with a program already recorded in the computer system, a so-called differential file (differential program).

REFERENCE SIGNS 1 meandering monitoring system 10 imaging device 20 light source 30 model generation device 133 model generation unit 139 storage unit 40 meandering monitoring device 141 acquisition unit 142 extraction unit 143 meandering monitoring unit 144 probability calculation unit 145 average value calculation unit 146 determination unit 147 graphing unit 148 output unit 149 trained model storage unit 50 display device 250 training dataset 300 trained model

Claims (4)

A meandering monitoring method for monitoring occurrence of meandering of a strip traveling in a continuous annealing furnace, comprising:
an imaging step of imaging an area equal to or greater than the width of the strip in a width direction of the strip;
a light emitting step of illuminating the strip with diffused light so that a highlight portion extending in the width direction of the strip can be captured by the capturing step as a reflection shape of light reflected from the surface of the strip;
a meandering monitoring step of monitoring occurrence of meandering of the strip using the captured image captured by the imaging step,
The meandering monitoring step includes:
a trained model trained using a training dataset including a plurality of images captured in advance in the continuous annealing furnace where the imaging step is performed , and a correct answer label indicating whether or not each of the plurality of images represents an actual occurrence of meandering, is trained using a training dataset including the images captured in the imaging step, and the trained model is trained using the training dataset, the correct answer label indicating whether or not each of the plurality of images represents an actual occurrence of meandering, and the trained model is trained using the trained dataset including the images captured in the imaging step, and the correct answer label indicating whether or not each of the plurality of images represents an actual occurrence of meandering, is trained using the trained model, The imaging step includes:
The meandering monitoring method according to claim 1, wherein an image of an area of at least 0.4 times the width of the strip in the strip running direction is captured. The trained model is
outputting the meandering occurrence probability of the strip;
The meandering monitoring step includes:
The meandering monitoring method according to claim 1 or 2, characterized in that the captured images are sequentially input into the trained model, a moving average value of the meandering occurrence probability sequentially output from the trained model is sequentially calculated, the moving average value is compared with a predetermined threshold value, and the occurrence of meandering of the strip is predicted. A meandering monitoring system for monitoring the occurrence of meandering of a strip traveling in a continuous annealing furnace, comprising:
an imaging means for imaging an area of the strip that is equal to or greater than the width of the strip in the width direction of the strip;
a light emitting means for illuminating the strip with diffused light so that the imaging means can capture an image of a highlight portion that spreads in the width direction of the strip as a reflection shape of light reflected from the surface of the strip;
a meandering monitoring means for monitoring occurrence of meandering of the strip using the image captured by the imaging means,
The meandering monitoring means
A meandering monitoring system characterized by acquiring information indicating the probability of meandering of the strip based on a determination result obtained by inputting images captured by the imaging means into a trained model trained using a training dataset including a plurality of images captured in advance in the continuous annealing furnace where imaging is performed by the imaging means, and correct answer labels indicating whether or not meandering actually occurred for each of the plurality of images, and predicting the occurrence of the meandering .