Anomaly detection apparatus and anomaly detection method

Abstract

Provided are: a camera (81, 82) that photographs a rolled material (1) during rolling of the rolled material (1) as a detection target of a rolling abnormality; a brightness reference value setting unit (91) that extracts the rolled material (1) from at least one image photographed by the camera (81, 82), calculates a brightness reference value that is determined to be normal for the rolled material (1) based on brightness data of pixels within a range of the rolled material (1); and an image processing unit (92) that extracts the rolled material (1) from one or more images among one or more photographed images, and detects a rolling abnormality of a surface based on a brightness difference between the brightness data within a range of the rolled material (1) and the brightness reference value. Thus, an abnormality detection device and an abnormality detection method that can improve the accuracy of rolling abnormality detection of a plate compared to the past are provided.

Description

Technical Field

[0001] This invention relates to an anomaly detection device and an anomaly detection method. Background Technology

[0002] In a production line, as an example of a surface inspection device that accurately determines the texture signal of the defect-free section and sets the judgment threshold accurately and efficiently, Patent Document 1 describes a surface inspection device for detecting defects on the surface of a traveling metal strip. The threshold calculation unit includes: a concentration histogram calculation unit that calculates the concentration histogram of each unit area from the image signal; a standard deviation calculation unit that calculates the peak position of the concentration histogram, assumes a normal distribution function with the peak position as the average value, and calculates the standard deviation value of the assumed normal distribution function from the concentration histogram degree near the peak position; and a threshold setting unit that sets the judgment threshold corresponding to the standard deviation value, and calculates the judgment threshold in a way that the standard deviation value does not contain defect signal components other than texture signal components as much as possible.

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: JP 2010-85096 Summary of the Invention

[0006] In the case of hot rolling mills, as the rolling of thin plates / high loads increases, the rolling plate extrusion phenomenon (referred to as rolling anomaly) occurs more frequently when the plate tail end passes through after the back tension is released, which is accompanied by rapid swaying of the plate. This results in damage to the roll surface, an increase in the number of roll replacements, and may lead to a decrease in product yield.

[0007] Currently, while interventions such as bending / leveling based on operator visual inspection have avoided the aforementioned problems, they largely depend on the operator's skill. By using computer image processing based on the monitoring image information observed by the operator, intervention operations can be quickly initiated as soon as rolling plate extrusion (rolling anomaly) is detected, increasing the probability of preventing subsequent rolling anomalies or equipment failures.

[0008] The rolling extrusion phenomenon refers to the rolling of metal strips in a folded state as they pass through a rolling mill.

[0009] If the folded part of the plate is rolled in this state, the processing heat in this part becomes higher than that in the others, producing a brightness that is brighter than that of the normal part of the plate surface, and rolling abnormality detection accompanied by brightness difference can be performed.

[0010] Therefore, in hot rolling, in order to detect rolling abnormalities such as extrusion of the rolled plate, a camera is used to monitor the image of the plate surface. When the brightness of the plate surface exceeds a certain threshold and the area increases, it can be detected as an abnormality of the plate.

[0011] The threshold varies depending on the type or material of the plate. For example, Patent Document 1 discloses a method for measuring the texture signal of each unit area of ​​the surface of a metal strip and calculating the standard deviation value excluding the data of the defective part as the texture signal level. Thus, the judgment threshold can be set accurately and efficiently for all types of metal strips.

[0012] In the technology described in Patent Document 1 above, the judgment threshold used is fixed. However, according to the inventors' research, even if the variety or material is the same, the brightness detected by the camera can differ due to factors such as the intensity of the lighting or the amount of water vapor in the surrounding environment. Therefore, since Patent Document 1 did not take this into account, there is room for improvement.

[0013] The purpose of this invention is to provide an anomaly detection device and anomaly detection method that can improve the accuracy of rolling anomaly detection compared with the past.

[0014] This invention includes multiple means to solve the aforementioned problems, but one example is provided: an anomaly detection device for detecting rolling anomalies on the surface of a metal strip rolled by a rolling mill. This anomaly detection device is characterized by comprising: a camera that captures images of the metal strip during the rolling process, which is the object of the rolling anomaly detection; a brightness reference value setting unit that extracts the metal strip from at least one image captured by the camera and calculates a brightness reference value for determining that the strip has no anomalies based on brightness data of pixels within a range of the metal strip; and an image processing unit that uses one or more of the captured images as comparison images, extracts the metal strip from each comparison image, and processes the image based on the brightness data extracted from the comparison images. The rolling abnormality is detected by the brightness difference between the brightness data of pixels within the strip area and the brightness reference value. The brightness reference value setting unit divides the brightness data into three components: R value, G value, and B value. It calculates the value of two or more of these three components as the brightness reference value. The image processing unit divides the brightness data of pixels within the strip area extracted from the comparison image into R value, G value, and B value. It calculates the brightness difference by subtracting the reference R value from the R value, the brightness difference by subtracting the reference G value from the G value, and the brightness difference by subtracting the reference B value from the B value. The rolling abnormality is detected based on the relationship between the brightness differences of the two components in each brightness difference.

[0015] Invention Effects

[0016] According to the present invention, the accuracy of rolling anomaly detection in plates can be improved compared to the past. Other issues, configurations, and effects will become clear from the following description of embodiments. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the configuration of a rolling mill having an anomaly detection device according to an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the configuration of an anomaly detection device according to an embodiment of the present invention.

[0019] Figure 3 This is a diagram illustrating the situation in an anomaly detection apparatus according to an embodiment, where the range of the rolled material is extracted from an image.

[0020] Figure 4 This is an example of a graph showing the luminance (RGB) histogram of pixels within a plate range of an image in an anomaly detection apparatus according to an embodiment.

[0021] Figure 5 This is a diagram illustrating an example of a method for calculating a brightness reference value in an anomaly detection device according to an embodiment.

[0022] Figure 6 This is a diagram illustrating an example of a method for calculating a brightness reference value in an anomaly detection device according to an embodiment.

[0023] Figure 7 This is a diagram illustrating another example of a method for calculating a brightness reference value in an anomaly detection device according to an embodiment.

[0024] Figure 8 This is a diagram illustrating an example of the abnormal region extraction and determination process in the abnormality detection device of an embodiment.

[0025] Figure 9 This is a diagram illustrating an example of the abnormal region extraction and determination process in the abnormality detection device of an embodiment.

[0026] Figure 10 This is a diagram illustrating another example of the abnormal region extraction and determination process in the abnormality detection apparatus of the embodiment.

[0027] Figure 11 This is a diagram illustrating another example of the abnormal region extraction and determination process in the abnormality detection apparatus of the embodiment.

[0028] Figure 12 This is a flowchart illustrating the method for calculating anomaly regions based on reflected light regions in the anomaly detection apparatus of an embodiment.

[0029] Figure 13 This is a flowchart illustrating the method for calculating anomaly regions based on reflected light regions in the anomaly detection apparatus of an embodiment.

[0030] Figure 14 This is a flowchart illustrating the method for calculating anomaly regions based on reflected light regions in the anomaly detection apparatus of an embodiment.

[0031] Figure 15 This is a flowchart of the anomaly determination process in the anomaly detection device of the embodiment. Detailed Implementation

[0032] use Figures 1 to 15 Embodiments of the anomaly detection device and anomaly detection method of the present invention are described below. Furthermore, in the drawings used in this specification, the same or corresponding components are labeled with the same or similar reference numerals, and sometimes repeated descriptions of these components are omitted.

[0033] Furthermore, the metal strip of the object material to be rolled in this invention is a strip of metal material that can be rolled in general, and its type is not particularly limited. In addition to steel plates, non-ferrous materials such as aluminum or copper can also be used.

[0034] Initially, use Figure 1 as well as Figure 2 Describe the overall structure of the rolling equipment, including the anomaly detection device, and the composition of the anomaly detection device. Figure 1 This is a schematic diagram showing the configuration of the anomaly detection device and the rolling equipment having the anomaly detection device according to this embodiment. Figure 2 This is a schematic diagram showing the configuration of an anomaly detection device.

[0035] Figure 1 The rolling mill 100 shown is a finishing mill for rolling the material 1 to be rolled, and includes F1 mill stand 10, F2 mill stand 20, F3 mill stand 30, F4 mill stand 40, F5 mill stand 50, F6 mill stand 60, F7 mill stand 70, cameras 81 and 82, a looper 65 for tension control, an image processing computer 90, a display device 95, etc.

[0036] Among them, the anomaly detection device 101, consisting of cameras 81 and 82, lighting device 67, and image processing computer 90, is used to detect anomalies on the surface of the rolled material 1.

[0037] Furthermore, for the rolling equipment 100, not limited to Figure 1 The configuration shown has seven rolling mill stands, but it is acceptable to have at least one mill stand.

[0038] F1 stand 10, F2 stand 20, F3 stand 30, F4 stand 40, F5 stand 50, F6 stand 60, and F7 stand 70 are rolling mills. Each mill includes an upper workpiece roll and a lower workpiece roll, as well as upper and lower spare rolls supported by contact with these upper and lower workpiece rolls, pressure cylinders 11, 21, 31, 41, 51, 61, and 71 located above the upper spare rolls, and load detectors 12, 22, 32, 42, 52, 62, and 72. Furthermore, the mill can be configured with six stages, including upper and lower intermediate rolls between each upper and lower workpiece roll and each upper and lower spare roll. The roll configuration of the rolling mill is not limited to the described form, as long as it has at least upper and lower workpiece rolls.

[0039] Loop 65 is a roll for controlling linear tension. This roll is configured such that its rotation axis extends in the width direction of the rolled material 1, and the linear tension can be changed by lifting the rolled material 1 vertically upwards or lowering it downwards, respectively, between F1 mill stand 10 and F2 mill stand 20, between F2 mill stand 20 and F3 mill stand 30, between F3 mill stand 30 and F4 mill stand 40, between F4 mill stand 40 and F5 mill stand 50, between F5 mill stand 50 and F6 mill stand 60, and between F6 mill stand 60 and F7 mill stand 70. Furthermore, loop 65 may correspondingly possess the function of a plate shape gauge for detecting the tension distribution in the plate width direction.

[0040] Camera 81 is positioned to capture images of the rolled material 1 between the outlet side of F4 stand 40 and the inlet side of F5 stand 50. For example, it captures images of the rolled material 1 from directly above or diagonally above it at intervals shorter than 0.1 seconds in dynamic image (video) format. The image data captured by camera 81 is transmitted to image processing computer 90 via communication line 85.

[0041] Camera 82 is positioned to capture images of the rolled material 1, including the output side of the F7 machine base 70. Similar to camera 81, it captures images of the rolled material 1 from directly above or diagonally above it at intervals shorter than, for example, less than 0.1 seconds. The image data captured by camera 82 is also transmitted to image processing computer 90 via communication line 85.

[0042] The process of taking pictures using these cameras 81 and 82 includes images of the rolled material 1.

[0043] In this embodiment, the cameras 81 and 82 capture images of the rolled material 1 at different times during the rolling process of a roll that constitutes the object of anomaly detection.

[0044] In addition, it is explained that the camera is located at the exit side of F4 base 40 and the inlet side of F5 base 50, and at the exit side of F7 base 70. However, it is only necessary to have at least one camera. There can be one or more cameras, or they can be located between all bases or at the inlet and outlet sides of the rolling equipment 100.

[0045] The lighting device 67 illuminates the rolled material 1, especially the rolled material 1 within the range captured by the cameras 81 and 82. It can be a common lighting device, such as one appropriately positioned on top of the rolling mill where the rolling equipment 100 is installed, but a dedicated lighting device can also be used.

[0046] The image processing computer 90 is a device consisting of a computer or the like that controls the operation of each machine in the rolling mill 100, and includes a brightness reference value setting unit 91, an image processing unit 92, and a database 93.

[0047] The brightness reference value setting unit 91 comprises the following components: extracting the rolled material 1 from at least one image captured by cameras 81 and 82; and calculating a reference value for the brightness of a plate deemed to have no abnormalities based on the brightness data of pixels within the range of the rolled material 1. This brightness reference value setting unit 91 is the main body executing the brightness reference value setting process. When cameras 81 and 82 capture moving images, the brightness reference value setting unit 91 extracts the image from the moving images.

[0048] The image processing unit 92 comprises the following components: using one or more captured images as comparison images; extracting the rolled material 1 from each comparison image; and detecting surface rolling abnormalities based on the brightness difference between the brightness data of pixels within the range of the rolled material 1 and the brightness reference value. This image processing unit 92 is the main body executing the image processing process. The image processing unit 92 includes a detection area setting unit 92A, a reflected light area setting unit 92B, and an abnormal area setting unit 92C, etc.

[0049] The detection area setting unit 92A is a portion of the detection area of ​​the rolled material 1 that is set as an abnormal detection object from the comparison image.

[0050] The reflected light area setting unit 92B sets a portion of the reflected light area irradiated by the illumination light from the illumination device 67 on the surface of the rolled material 1 from the comparison image.

[0051] The abnormal region setting unit 92C is a part that detects abnormal parts of the surface from the detection area with a reference value as a threshold and sets the area after removing the reflected light area as the abnormal region.

[0052] The details of the processing of these brightness reference value setting unit 91, image processing unit 92, detection area setting unit 92A, reflected light area setting unit 92B, and abnormal area setting unit 92C will be explained later.

[0053] Furthermore, the comparison images processed in the image processing unit 92 are basically set to the latest images.

[0054] Database 93 is a storage device that stores information about the threshold boundary for distinguishing normal and abnormal values. This information is based on the distribution of the brightness difference of one component relative to the brightness difference of another component in the brightness of pixels within the range of the rolled material 1 in the image of the rolled material 1. This distribution is plotted in a two-dimensional chart composed of the two components. For example, it stores information about the threshold boundary for distinguishing normal and abnormal values. Figure 10 or Figure 11 Information about the boundary lines shown. Appropriately composed of SSDs or HDDs, etc.

[0055] The display device 95 is a display device such as a monitor or an audio device such as an alarm. For example, it is a device used to inform the operator of the occurrence of the board shape abnormality and its countermeasures when the image processing computer 90 determines that the board shape abnormality is abnormal. Therefore, a monitor is mostly used as the display device 95.

[0056] The operator visually checks the display screen of the display device 95 or the individual machine stands and the spaces between machine stands during machine operation to confirm whether there are any rolling abnormalities such as plate extrusion. For example, if the operator confirms that plate extrusion has occurred on the display device 95, and the display shows the operations that should be performed, such as bending correction, leveling correction, roll speed correction, and opening the roll gap of the downstream machine stand, the operator can manually perform the operations according to the instructions, thereby preventing the rolling abnormality from progressing to a further deterioration.

[0057] Furthermore, it can automatically control the operation to avoid rolling abnormalities by displaying the abnormality of the rolled plate extrusion phenomenon to the operator on the display screen of the display device 95 and sending a signal to the mill control device to correct or stop the rolling abnormality. Alternatively, it can automatically control various correction operations or stop the machine operation without displaying it on the display device 95.

[0058] Next, use Figure 3 Next, a specific example of anomaly detection processing for judging anomalies in the shape of the rolled material 1 in this invention will be described. Figure 3 This is a diagram illustrating the extraction of the range of rolled material from an image. Figure 4 This is an example of a graph showing the luminance (RGB) histogram of pixels relative to a plate range of an image. Figures 5-7This diagram illustrates an example of how the brightness reference value for a board deemed to be without abnormalities is calculated. Figures 8-11 This diagram illustrates an example of the process for extracting and determining abnormal regions. Figures 12-14 This is a flowchart illustrating the method for calculating abnormal areas based on the lighting area.

[0059] In this way, even if the type or material of the rolled material 1 is the same, the brightness detected by the camera will be different due to the influence of the lighting brightness during rolling, the amount of water vapor in the surrounding area, and the image processing settings of the camera, so a corresponding countermeasure is required.

[0060] In contrast, the inventors of this invention have conducted intensive research and proposed a concept that changes the threshold for judgment for each rolled coil even if the type or material of the rolled material 1 is the same, and thus completed this invention.

[0061] The following details the processing.

[0062] First, images including the rolled material 1 are acquired using cameras 81 and 82. Preferably, the rolling material 1 is continuously captured as a moving image. The captured images are then output to an image processing computer 90 via communication line 85.

[0063] Regarding the image processing computer 90, in the brightness reference value setting unit 91, such as Figure 3 As shown, the background and the area where the rolled material 1 exists are determined by binarization processing from the image of the rolled material 1 during rolling, and the range 1A of the rolled material 1 is extracted from the brightness area above a certain threshold. Thereafter, in the image processing unit 92, which includes the detection area setting unit 92A, the reflected light area setting unit 92B, and the abnormal area setting unit 92C, abnormal detection processing is performed relative to the extracted range 1A.

[0064] Furthermore, the extracted range 1A may include a region that is slightly inside or slightly outside compared to the actual plate area, and does not necessarily have to be consistent with the rolled material 1. However, since there are cases where the position of the rolled material 1 captured by moving the looper 65 up and down changes, the range 1A must be extracted in order to follow such changes.

[0065] Next, regarding the image processing computer 90, in the brightness reference value setting unit 91, the brightness distribution of pixels within the range 1A of the rolled material 1 is calculated relative to a selected image, and a brightness reference value for a plate determined to be without abnormalities is calculated. Here, the brightness reference value setting unit 91 divides the brightness data into three components: R value, G value, and B value, and calculates three values ​​as brightness reference values: reference R value (R0 value), reference G value (G0 value), and reference B value (B0 value).

[0066] In addition, two values ​​(R0 and G0, R0 and B0, B0 and G0) can also be obtained from the brightness reference value.

[0067] Furthermore, the brightness reference value setting unit 91 can process brightness data using grayscale levels such as 8 bits (256 gray levels) or 16 bits (65536 gray levels). When using grayscale levels, the subsequent processing is essentially the same, so details are omitted.

[0068] As part of the process of determining the three reference values ​​(R0, G0, and B0) as luminance reference values, methods for selecting from one image and methods for selecting from multiple images will be described. First, the method of selecting a luminance reference value from one image will be explained.

[0069] In this case, the brightness reference value setting unit 91 calculates the value relative to an image captured by the cameras 81 and 82 to determine the brightness reference value for the image. Figure 3 The histogram represents the luminance distribution (radius, color, and luminance) of each R, G, and B value, expressed as the number of measurement points for luminance (RGB) across all pixels within the range 1A. Here, R represents red, G represents green, and B represents blue, with RGB values ​​ranging from 0 to 255. Figure 4 This is an example showing a histogram of luminance (RGB) relative to a plate range of an image.

[0070] Next, the brightness reference value setting unit 91... Figure 5 as well as Figure 6 As shown, the value that divides the area represented by the integral value of each distribution into two parts (center position: 50%) is used as the initial brightness reference value for each RGB component of the board judged to be without abnormalities. Alternatively, a value can be selected within an allowable range of 50% to ±20% at the center position as the brightness reference value for each RGB component. Furthermore, the allowable range does not need to be "±20%" and can be appropriately changed.

[0071] The value (position) used to determine the brightness reference value can be specified by the operator before rolling the rolled material 1 of the abnormal detection object, or it can be a preset value of the device, or it can be a value that is appropriately learned by machine learning.

[0072] In addition, instead of using a value that divides the area into two parts, it is also possible to achieve... Figure 7As shown, the brightness (highest frequency value) of the RGB components that has the highest number of measurement points in each distribution is set as the initial brightness reference value for the boards judged to be without abnormalities. Alternatively, a value can be selected from the highest frequency value within a tolerance range of ±30% as the brightness reference value for each RGB component. Furthermore, the tolerance range does not need to be "±30%" and can be appropriately changed.

[0073] In this method of selecting a brightness reference value from an image, the brightness reference value can be updated at any time by directly using the value obtained from the first image during the rolling process of the corresponding roll material, or by re-observing the brightness reference value from a newly captured image. There are no particular limitations.

[0074] Next, a method for selecting a brightness reference value from multiple images will be described. In this method, the brightness reference value setting unit 91 calculates the brightness reference value by averaging the brightness data captured in multiple images. However, it is desirable to add the brightness data captured in the new image to the already calculated brightness data each time a new image is captured to recalculate the brightness reference value.

[0075] First, the brightness reference value setting unit 91 sets the brightness reference value based on two or more images captured by the cameras 81 and 82. Figures 4-7 The method shown calculates, for example, the brightness reference value (R) of each image captured by camera 81. i G i B i For example, brightness reference values ​​R1, G1, and B1 are obtained from the first image, R2, G2, and B2 from the second image, and R3, G3, and B3 from the third image. Furthermore, brightness reference values ​​are obtained for each image of a specific part captured by each camera.

[0076] As for subsequent processing, there are, for example, (i) a method that uses multiple images simultaneously to calculate the arithmetic mean, and (ii) a method that uses a moving average for each of the multiple images. The case of (i) using multiple images simultaneously to calculate the arithmetic mean will be explained first.

[0077] The brightness reference value setting unit 91 calculates the arithmetic mean of the number of images used when setting the brightness reference value. For one image, the brightness reference value R0 = R1; for two images, R0 = (R1 + R2) / 2; and for three images, R0 = (R1 + R2 + R3) / 3. The same method is used for brightness G and brightness B.

[0078] Next, the case of (ii) obtained by processing a moving average for each multiple images will be explained. The brightness reference value setting unit 91 takes a moving average based on the number of images used when setting the brightness reference value. In the case of two images, the brightness reference value R = (R1 + R2) / 2 = R 12 If a new image is added at the next time interval, resulting in three images, the brightness reference value R0 = (R 12 +R3) / 2. The same method is used for brightness G and brightness B.

[0079] Furthermore, in the method of selecting a brightness reference value from multiple images, the update of the brightness reference value can be stopped midway when it is determined to be stable, or it can be continuously updated during the rolling process of a roll material, without any particular limitation.

[0080] Subsequently, the image processing unit 92 divides the brightness data from the image into R values, G values, and B values. It calculates the brightness difference by subtracting a reference R value from the R value, the brightness difference by subtracting a reference G value from the G value, and the brightness difference by subtracting a reference B value from the B value. Here, we simply assume that the brightness difference of brightness data R = R - R0, the brightness difference of brightness data G = G - G0, and the brightness difference of brightness data B = B - B0.

[0081] In addition, when using multiple images to set the baseline value, the latest (most recent) image among the multiple images can be used as the comparison image.

[0082] Subsequently, the image processing unit 92 detects rolling anomalies based on the relationship between the brightness differences of the two components in each brightness difference. Two methods are given as anomaly determination methods: (i) determination based on the relationship between the brightness differences of the two components (using a threshold value with a specific value), and (ii) determination based on the relationship between the brightness differences of the two components (using an arbitrary threshold boundary). Initially, details of determination (i) based on the relationship between the brightness differences of the two components (using a threshold value with a specific value) will be explained.

[0083] In this case, the image processing unit 92 detects an anomaly when the brightness difference between the two components in the R value, G value, and B value both exceed their respective thresholds.

[0084] For example, the image processing unit 92 selects from R value, G value, and B value, such as Figure 8 The calculation shows whether the brightness difference between the B value and the reference B value, and the brightness difference between the G value and the reference G value, are both above the threshold values ​​for their respective brightness differences, and as shown... Figure 9 The calculation shows whether the brightness difference between the B value and the reference B value, and the brightness difference between the R value and the reference R value, are both above the threshold values ​​for each brightness difference. Furthermore, the brightness difference between the two components forming a combination is above the threshold values ​​for each brightness difference (within...). Figure 8 G is in the middle. a And Ba The brightness difference between the two components that are combined with other components is the threshold value (R) for each brightness difference. a And B a In the above cases, the detection indicates a rolling abnormality.

[0085] Furthermore, although not illustrated, the relationship between brightness difference G-G0 and brightness difference R-R0 can be used to determine whether the brightness difference between the two components is within their respective thresholds (G0 and R0). a and R a )above.

[0086] Since water vapor generated during the rolling process becomes noise in image processing, it is preferable to use an environment with low water vapor levels. Figure 8 as well as Figure 9 This determination process is particularly effective. Furthermore, this process is preferably performed together with the process described later for removing the area of ​​reflected light irradiated by the illumination light from the illumination device 67 on the surface of the rolled material.

[0087] Next, the details of the determination based on the relationship between the brightness difference of the two components (using an arbitrary threshold boundary) in (ii) will be explained. This process differs from (i) above in that it is unaffected by the presence or absence of water vapor, making it suitable for execution under rolling conditions with high water vapor levels. Furthermore, it can be performed appropriately regardless of whether the area of ​​reflected light irradiated by the illumination light from the illumination device 67 on the surface of the rolled material is removed.

[0088] In this case, the image processing unit 92 detects anomalies on the plate surface based on the brightness difference of one component, the brightness difference of another component, and an arbitrary threshold boundary.

[0089] For example, the image processing unit 92 processes the R value, G value, and B value as follows: Figure 10 As shown, the detection result for whether something is abnormal is determined based on the brightness difference between the two components that form a combination (the brightness difference between the B value and the reference B value, and the brightness difference between the G value and the reference G value) and the boundary derived from them, and as... Figure 11 As shown, the detection result for whether something is abnormal is determined based on the brightness difference between the two components that are combined (the brightness difference between the B value and the reference B value, and the brightness difference between the R value and the reference R value) and the boundary derived from them. More specifically, it is determined whether the plotted brightness difference (R-R0, G-G0, B-B0) relative to the reference value is located on the abnormal side compared to the boundary line.

[0090] Subsequently, if the image processing unit 92 finds that any detection result is in an abnormal position, it detects an anomaly on the plate surface.

[0091] Furthermore, although the illustration is omitted, the relationship between the brightness difference G-G0 and the brightness difference R-R0 can be used. However, the image processing unit 92 prefers to use the brightness difference of the G value and the B value in the three combinations, as well as the brightness difference of one of them.

[0092] Additionally, this explains the case where two of the three combinations are used, but it's also possible to use only one of the three combinations for judgment. Furthermore, an anomaly can be detected if either of the two combinations is above any threshold boundary. It's also possible to use all three combinations for judgment. When using all three combinations, an anomaly can be detected if one of them is above any threshold boundary, if the two that are in a majority (majority rule) are above any threshold boundary, or if an anomaly is detected only if all three are above any threshold boundary. In these cases, as described above, the brightness difference between the G and B values ​​can be prioritized.

[0093] against Figure 10 or Figure 11 The arbitrary threshold boundary line shown is preferably obtained by collecting multiple normal / abnormal image data in advance, finding an approximate formula that can be separated in advance, and storing it in database 93 in advance.

[0094] For example, the operator of the rolling mill 100 or an employee of the manufacturer of the anomaly detection device can judge and draw the threshold boundary line for separation based on the result image. Alternatively, mathematical methods, such as data clustering (data classification), can be used to classify as normal or abnormal. As one method, a first-order equation can be derived using a binary classification method based on SVM (Support Vector Machine) to maximize the distance (difference) between normal and abnormal data.

[0095] Next, use Figures 12-14 This section describes the details of removing the area of ​​reflected light from the lighting device 67.

[0096] like Figure 12 As shown, the abnormal region becomes the shape of the brightness distribution along the rolling direction (transport direction). Conversely, it actually becomes the area captured in the images taken by cameras 81 and 82 that contains a mixture of this abnormal region and reflected light from the illumination device 67 (in...). Figure 12 An image of the extrusion section and reflective light area 1A1 (width (longitudinal) direction of the middle plate) is shown, and it is desired to remove the reflective light area formed by the lighting device 67 from the extrusion section and reflective light area 1A1.

[0097] Here, since the RGB color of the reflected light area formed by the lighting device 67 is close to white, for example, the RGB color of the extrusion part is close to yellow, after binarization processing using the brightness threshold to convert it into a white or black two-color image, the area that is judged to be white for a long continuous length in the width direction is extracted and removed as the reflected light area formed by the lighting device 67.

[0098] Therefore, in the image processing unit 92, the detection region setting unit 92A, for the latest image in the comparison images, uses the above-mentioned... Figures 8-11 The method identifies the abnormal area as the extrusion section and reflected light region 1A1. Based on this, such as... Figure 13 As shown, using the reflected light area setting unit 92B, a relatively long reflected light area 1A2 in the plate width direction formed by the lighting device 67 is extracted and deleted from the extrusion section and reflected light area 1A1. Figure 14 As shown, only the true abnormal region 1A3 (the region that is relatively long in the transport direction) is extracted. Then, the abnormal region setting unit 92C sets the detected true abnormal region 1A3 as an abnormal region, and the process transitions to the final detection process to determine whether the rolling abnormality is detected.

[0099] Furthermore, it explains the use of Figures 8-11 The sequential detection method involves removing the reflective light region 1A2 formed by the illumination device 67 from the abnormal extrusion area and reflective light region 1A1 on the surface. However, it is also possible to remove the reflective light region 1A2 formed by the illumination device 67 first after the detection plate range 1A, and then use... Figure 8 The processing order of abnormal rolling regions is set.

[0100] Subsequently, the image processing unit 92 preferably removes abnormal areas defined by the area of ​​reflected light formed by the illumination device 67, or utilizes... Figure 8 If the total area of ​​each pixel in the abnormal region of the abnormal part of the surface is detected by methods such as [e.g., the total area value, or the value obtained by dividing the total area value by the area of ​​the entire range 1A of the rolled material 1, is greater than a certain threshold (ε), it is determined that a rolling abnormality has occurred in the rolled material 1.

[0101] Next, refer to Figure 15 This describes an anomaly detection method for detecting anomalies on the surface of the rolled material 1, which is rolled by a rolling mill, according to this embodiment. Figure 15 This is a flowchart of the anomaly determination process in the anomaly detection device of the embodiment.

[0102] First, such as Figure 15 As shown, images are acquired using cameras 81 and 82 (step S101). This step S101 is equivalent to the shooting process, and it is preferable to shoot dynamic images.

[0103] Next, in the brightness reference value setting unit 91 of the image processing computer 90, it is determined whether the rolled material 1 exists in the image captured in step S101 (step S102). If it is determined that the rolled material 1 exists, the process proceeds to step S103; otherwise, if it is determined that it does not exist, the process proceeds to step S111.

[0104] Subsequently, in the brightness reference value setting unit 91, after the setting plate detection area (step S103), the brightness reference value of each object roll is selected (step S104). These steps S103 and S104 correspond to the brightness reference value setting process.

[0105] Next, in the image processing unit 92 of the image processing computer 90, the brightness difference between the image and the comparison image captured after selecting the brightness reference value is calculated (step S105), and an anomaly candidate is determined (step S106). At this time, preferably, the image processing unit 92 performs a process to remove the reflected light area formed by the illumination device 67 (step S107). After that, in the image processing unit 92, a final anomaly determination process is performed (step S108) to determine whether a rolling anomaly such as extrusion has occurred (step S109). If an anomaly is determined to have occurred, the process proceeds to step S110; if no anomaly is determined to have occurred, the process proceeds to step S111. These steps S105 to S109 correspond to image processing steps.

[0106] In step S110, when it is determined that an abnormality has occurred, the image processing unit 92 records an abnormality occurrence flag = 1 (step S110). In step S102, when it is determined that there is no rolled material 1, or in step S110, when it is determined that no abnormality has occurred, the image processing unit 92 records an abnormality occurrence flag = 0 (step S111), and processing starts again at the next time interval.

[0107] In the image processing computer 90, when an error generation flag 1 is recorded, the message is displayed to the display device 95. Alternatively, intervention processing can be automatically performed relative to each of the bases 10, 20, 30, 40, 50, 60, and 70.

[0108] The above steps S101 to S111 are performed during the rolling process of a rolled coil, and the process ends at the point when the rolling of a rolled coil is completed.

[0109] Next, the effects of this embodiment will be explained.

[0110] The anomaly detection device described in this embodiment is an apparatus for detecting rolling anomalies on the surface of a rolled material 1 rolled by a rolling mill. The apparatus includes: cameras 81 and 82 that capture images of the rolled material 1 during the rolling process, which is the object of rolling anomaly detection; a brightness reference value setting unit 91 that extracts the rolled material 1 from at least one image captured by the cameras 81 and 82 and calculates a brightness reference value that determines the rolled material 1 to be free of anomalies based on the brightness data of pixels within the range of the rolled material 1; and an image processing unit 92 that uses one or more of the captured images as comparison images, extracts the rolled material 1 from each comparison image, and detects surface rolling anomalies based on the brightness difference between the brightness data within the range of the rolled material 1 and the brightness reference value.

[0111] In this way, in the rolled material 1, which is the object of rolling anomaly detection, a brightness reference value is determined as the reference for each comparison of anomalies. This reference value is used only for comparison images of rolled material 1 relative to the same coil. Therefore, it is possible to detect rolling anomalies on the surface of rolled material 1 by taking into account the inherent external interference factors of the rolled material 1. Compared with the past, the accuracy of rolling anomaly detection of the plate can be improved.

[0112] Furthermore, the brightness reference value setting unit 91 divides the brightness data into three components: R value, G value, and B value. It calculates the value of two or more of these three components as reference values. The image processing unit 92 divides the brightness data of pixels within the range of the rolled material 1 into R value, G value, and B value. It calculates the brightness difference by subtracting the reference R value from the R value, the brightness difference by subtracting the reference G value from the G value, and the brightness difference by subtracting the reference B value from the B value. Based on the relationship between the brightness differences of the two components in each brightness difference, rolling abnormalities are detected. Therefore, abnormalities can be judged in combination, thereby further improving the accuracy of abnormality detection.

[0113] Furthermore, cameras 81 and 82 capture dynamic images, and the brightness reference value setting unit 91 extracts images from the dynamic images. It then performs simple averaging or weighted moving average processing on the brightness data captured in multiple images to calculate the brightness reference value. This allows for the acquisition of a brightness reference value calculated based on brightness data within the range of the rolled material 1 captured in multiple consecutive images. Even if there are instantaneous image fluctuations in a particular image, by referring to the state of other images to reduce anomalies caused by these fluctuations, a normal brightness reference value that is close to what is considered normal for the plate to be free of abnormalities can still be obtained. Therefore, this helps improve detection accuracy.

[0114] In addition, whenever the latest image is captured, the brightness reference value setting unit 91 adds the brightness data captured in the new image to the brightness data that has already been calculated and recalculates the brightness reference value. Thus, even if there are fluctuations in the image, since the brightness reference value is calculated using the brightness data of the previous image, the abnormality caused by the fluctuation can be reduced, and a value close to the normal brightness reference value that is judged to be that the board has no abnormality can be obtained.

[0115] Furthermore, the brightness reference value setting unit 91 divides the brightness data into three components: R value, G value, and B value. It calculates the brightness distribution of each R value, G value, and B value expressed as the number of measurement points relative to the brightness. The initial brightness reference value is set by dividing the area represented by the integral value of each distribution into two parts, or by the brightness with the most measurement points in each distribution. A specific brightness range including the initial brightness reference value is further set. The R value, G value, and B value are selected from the set brightness range and set as the respective brightness reference values. Thus, as long as the brightness data is divided into half the area represented by the integral value of the brightness data distribution of the R component, G component, and B component, or the brightness periphery with the most drawing points, it is sufficient to reflect the brightness data of the majority of the normal surface area in the detection range. Therefore, the influence of the brightness data based on the relatively narrow part of the reflected light area captured on the board surface by the illumination device 67 can be eliminated, and the brightness (reference value) of the normal surface area of ​​the board can be calculated with high accuracy.

[0116] In addition, the unit also includes an illumination device 67 that illuminates the rolled material 1. The image processing unit 92 sets the detection area of ​​the rolled material 1, which is the object of abnormality detection, from the latest image. It sets the reflected light area irradiated by the illumination light from the illumination device 67 on the surface of the rolled material 1 from the latest image. It resets the detection area from the detection area after removing the reflected light area. Based on the relationship between the brightness data of the pixels in the reset detection area and the brightness difference of the brightness reference value, it sets the abnormal area and detects rolling abnormalities. As a result, it is possible to extract the true surface abnormalities after removing the reflected light area of ​​the illumination light from the illumination device 67, and further improve the accuracy of abnormality detection.

[0117] Furthermore, the brightness difference is obtained by subtracting the brightness reference value from the brightness data. The image processing unit 92 detects rolling abnormalities when the brightness difference between the two components in the R value, G value, and B value is above the respective brightness difference threshold. Therefore, the abnormality is detected when the brightness difference between the two components is above the threshold, thus reducing false detections and further improving the accuracy of abnormality detection.

[0118] Furthermore, the image processing unit 92 detects rolling abnormalities when the brightness difference between two components that form a combination of R, G, and B values ​​is above their respective thresholds, and when the brightness difference between two components that form other combinations is above their respective brightness difference thresholds. Thus, by using two sets of combinations of two components, when both sides detect abnormalities, it determines that the surface is abnormal. Therefore, it becomes less likely to misdetect normal parts as abnormal parts, and the accuracy of abnormality detection can be further improved.

[0119] It also includes a database 93, which plots multiple distributions of the brightness difference of one component relative to the brightness difference of another component in the R, G, and B values ​​of the image of the rolled material 1 captured in the image. A threshold boundary for distinguishing normal and abnormal values ​​is pre-determined. The image processing unit 92 detects rolling anomalies based on the brightness difference of one component, the brightness difference of the other component, and the threshold boundary. If brightness difference data are plotted for each combination of R and G values, G and B values, and R and B values, it can be observed that the data groups representing normal and abnormal data tend to separate along a certain line. Therefore, by using this line as the threshold boundary for the brightness difference of each component, the accuracy of anomaly detection can be further improved.

[0120] Furthermore, the image processing unit 92 detects rolling abnormalities when the detection result obtained from the brightness difference of two components that form a combination of R, G, and B values ​​and the threshold boundary obtained from them is abnormal, and when the detection result obtained from the brightness difference of two components that form other combinations and the threshold boundary obtained from them is abnormal. Thus, by using two sets of combinations of two components, when both sides detect abnormalities, it determines that it is a surface abnormality. Therefore, it becomes less likely to mistakenly detect normal parts as abnormal parts, and the accuracy of abnormality detection can be further improved.

[0121] Furthermore, the image processing unit 92 uses the brightness difference between the G and B values. Since the rolled material 1 has a light reddish tone, there are cases where the brightness of the normal and abnormal areas hardly changes if the R value is used, making it unsuitable for judgment. Therefore, by using the G and B values, anomaly detection can be performed without considering the background color of the rolled material 1.

[0122] <Other>

[0123] Furthermore, the present invention is not limited to the above embodiments and can be modified and applied in various ways. The above embodiments have been described in detail for ease of understanding of the present invention, but are not limited to all the described configurations.

[0124] Explanation of reference numerals in the attached figures

[0125] 1. Rolled material (metal strip)

[0126] 1A range of the rolled material extracted

[0127] 1A1 extrusion section and reflective light area

[0128] 1A2 reflected light area

[0129] 1A3 Abnormal Area

[0130] 10F1 base

[0131] 11, 21, 31, 41, 51, 61, 71 pressing cylinders

[0132] 12, 22, 32, 42, 52, 62, 72 load detectors

[0133] 20F2 frame

[0134] 30F3 frame

[0135] 40F4 frame

[0136] 50F5 frame

[0137] 60F6 frame

[0138] 61 Pressing cylinder

[0139] 65 Loop for tension control

[0140] 67 lighting fixtures

[0141] 70F7 frame

[0142] 71 Pressing cylinder

[0143] Cameras 81 and 82

[0144] 85 communication line

[0145] 90 Image Processing Computers

[0146] 91 Brightness reference value setting unit

[0147] 92 Image Processing Department

[0148] 92A Detection Area Setting Department

[0149] 92B Reflected Light Area Setting Unit

[0150] 92C Abnormal Area Setting Department

[0151] 93 Database

[0152] 95 display devices

[0153] 100 Rolling Equipment

[0154] 101 Anomaly Detection Device.

Claims

1. An anomaly detection device for detecting rolling anomalies on the surface of a metal strip rolled by a rolling mill, the anomaly detection device being characterized by comprising: A camera that captures images of the metal strip during the rolling process, which is the object of the rolling anomaly detection. The brightness reference value setting unit extracts the metal strip from at least one image captured by the camera and calculates a brightness reference value that is determined to be normal based on the brightness data of the pixels within the range of the metal strip. The image processing unit takes one or more of the captured images as comparison images, extracts the metal strip from each comparison image, and detects the rolling abnormality based on the brightness difference between the brightness data of the pixels within the range of the metal strip extracted from the comparison images and the brightness reference value. as well as database, The brightness reference value setting unit divides the brightness data into three components: R value, G value, and B value, and calculates the value of two or more of these three components—reference R value, reference G value, and reference B value—as the brightness reference value. The image processing unit extracts the brightness data of pixels within the range of the metal strip plate from the comparison image and divides them into R values, G values, and B values. It then calculates the brightness difference by subtracting the reference R value from the R value, the brightness difference by subtracting the reference G value from the G value, and the brightness difference by subtracting the reference B value from the B value. The database plots multiple distributions of the brightness difference of one component (R, G, and B values) relative to the brightness difference of another component in the captured image of the metal strip in a two-dimensional chart composed of two components, and pre-determines a threshold boundary to distinguish between normal and outlier values. The image processing unit detects the rolling abnormality based on the brightness difference of the one component, the brightness difference of the other component, and the threshold boundary.

2. The anomaly detection device according to claim 1, characterized in that, The camera captures dynamic images. The brightness reference value setting unit extracts the image from the dynamic image and performs simple averaging or weighted moving average processing on the brightness data captured in multiple images to obtain the brightness reference value.

3. The anomaly detection device according to claim 2, characterized in that, Whenever a new image is captured, the brightness reference value setting unit adds the brightness data captured in the new image to the brightness data that has already been calculated, and recalculates the brightness reference value.

4. The anomaly detection device according to any one of claims 1 to 3, characterized in that, The brightness reference value setting unit divides the brightness data into three components: R value, G value, and B value. Determine the luminance distribution of each of the R, G, and B values, expressed in terms of the number of measurement points for luminance. The initial brightness reference value is set by taking the brightness of each RGB component that divides the area represented by the integral value of each distribution into two parts, or the brightness of the component with the most measurement points in each distribution. A specific brightness range including the initial brightness reference value is further set, and the R value, G value and B value are selected from the set brightness range and set as the brightness reference value for each component.

5. The anomaly detection device according to any one of claims 1 to 3, characterized in that, It also includes an illumination device for illuminating the metal strip. The image processing unit sets the detection area of ​​the metal strip that is the object of abnormal detection according to the latest image, sets the reflected light area of ​​the illumination light from the illumination device on the surface of the metal strip according to the latest image, resets the area after removing the reflected light area from the detection area as the detection area, sets the abnormal area based on the relationship between the brightness data of the pixels in the newly obtained detection area and the brightness difference of the brightness reference value, and detects the rolling abnormality.

6. The anomaly detection device according to claim 1, characterized in that, The image processing unit detects a rolling abnormality when the detection result obtained from the brightness difference of two components that form a combination of R, G, and B values ​​and the threshold boundary obtained from them is abnormal, and when the detection result obtained from the brightness difference of two components that form other combinations and the threshold boundary obtained from them is abnormal.

7. The anomaly detection device according to claim 1, characterized in that, The image processing unit uses the brightness difference of the G value and the brightness difference of the B value.

8. An anomaly detection method for detecting rolling anomalies on the surface of a metal strip rolled by a rolling mill, the anomaly detection method being characterized by comprising: The filming process involves using a camera to film the metal strip during its rolling process, which is the object of anomaly detection. The brightness reference value setting process involves extracting the metal strip from at least one image captured in the shooting process and calculating the brightness reference value based on the brightness data of the pixels within the range of the metal strip. as well as The image processing step involves using one or more of the captured images as comparison images, extracting the metal strip from each comparison image, and detecting the rolling abnormality based on the brightness difference between the brightness data of pixels within the range of the metal strip extracted from the comparison images and the brightness reference value. In the brightness reference value setting process, the brightness data is divided into three components: R value, G value, and B value. These components are used as the brightness reference value to calculate the reference R value, reference G value, and reference B value. In the image processing step, the brightness data of pixels within the range of the metal strip is divided into R values, G values, and B values. The brightness difference is calculated by subtracting the reference R value from the R value, the brightness difference is calculated by subtracting the reference G value from the G value, and the brightness difference is calculated by subtracting the reference B value from the B value. Based on the brightness difference of one component, the brightness difference of another component, and a threshold boundary for distinguishing normal and abnormal values, the rolling abnormality is detected. The threshold boundary is pre-determined by the database based on multiple distributions of the brightness difference of one component relative to the brightness difference of the other component in the R, G, and B values ​​of the image of the metal strip, plotted in a two-dimensional chart composed of the two components.

9. The anomaly detection method according to claim 8, characterized in that, During the shooting process, dynamic images are captured. In the brightness reference value setting process, the image is extracted from the dynamic image, and the brightness data captured in multiple images are processed by simple averaging or weighted moving average to obtain the brightness reference value.

10. The anomaly detection method according to claim 9, characterized in that, In the brightness reference value setting process, whenever a new image is captured, the brightness data captured in the new image is added to the brightness data that has already been calculated, and the brightness reference value is recalculated.

11. An anomaly detection device for detecting rolling anomalies on the surface of a metal strip rolled by a rolling mill, the anomaly detection device being characterized by comprising: A camera that captures images of the metal strip during the rolling process, which is the object of the rolling anomaly detection. The brightness reference value setting unit extracts the metal strip from at least one image captured by the camera and calculates a brightness reference value that is determined to be normal based on the brightness data of the pixels within the range of the metal strip. The image processing unit takes one or more of the captured images as comparison images, extracts the metal strip from each comparison image, and detects the rolling abnormality based on the brightness difference between the brightness data of the pixels within the range of the metal strip extracted from the comparison images and the brightness reference value. An illumination device for illuminating the metal strip; as well as database, The brightness reference value setting unit divides the brightness data into three components—R value, G value, and B value—based on the latest image, and calculates values ​​for two or more of these three components—reference R value, reference G value, and reference B value—as the brightness reference value. The image processing unit extracts the brightness data of pixels within the range of the metal strip from the latest image in the comparison images and divides it into R values, G values, and B values. It then calculates the brightness difference by subtracting the reference R value from the R value, the brightness difference by subtracting the reference G value from the G value, and the brightness difference by subtracting the reference B value from the B value. The database plots multiple distributions of the brightness difference of one component (R, G, and B values) relative to the brightness difference of another component in the captured image of the metal strip in a two-dimensional chart composed of two components, and pre-determines a threshold boundary to distinguish between normal and outlier values. The image processing unit sets a detection area for the metal strip that is an abnormal detection object based on the brightness difference of one component, the brightness difference of the other component, and the threshold boundary. It sets a reflected light area on the surface of the metal strip by the illumination light from the illumination device according to the latest image. It resets the area after removing the reflected light area from the detection area as the detection area. It sets an abnormal area based on the relationship between the brightness data of the pixels in the newly obtained detection area and the brightness difference of the brightness reference value, and detects the rolling abnormality.

12. An anomaly detection method for detecting rolling anomalies on the surface of a metal strip rolled by a rolling mill, the anomaly detection method being characterized by comprising: The filming process involves using a camera to film the metal strip during its rolling process, which is the object of anomaly detection. The brightness reference value setting process involves extracting the metal strip from at least one image captured in the shooting process and calculating the brightness reference value based on the brightness data of the pixels within the range of the metal strip. as well as The image processing step involves using one or more of the captured images as comparison images, extracting the metal strip from each comparison image, and detecting the rolling abnormality based on the brightness difference between the brightness data of pixels within the range of the metal strip extracted from the comparison images and the brightness reference value. In the brightness reference value setting process, the brightness data is divided into three components—R value, G value, and B value—based on the latest image, and these components are used as the brightness reference value to calculate the reference R value, reference G value, and reference B value. In the image processing step, the brightness data of pixels within the range of the metal strip in the latest image are divided into R values, G values, and B values. The brightness difference is calculated by subtracting the reference R value from the R value, the reference G value from the G value, and the reference B value from the B value. Based on the brightness difference of one component, the brightness difference of another component, and a threshold boundary for distinguishing normal and abnormal values, a detection area for the metal strip that is the object of anomaly detection is set. According to the latest image, a reflected light area is set on the surface of the metal strip from the illumination light from the illumination device illuminating the metal strip. The area after removing the reflected light area from the detection area is redefined as the detection area. An abnormal area is set based on the relationship between the brightness data of pixels in the newly obtained detection area and the brightness difference of the brightness reference value, and the rolling abnormality is detected. The threshold boundary is pre-determined by the database based on multiple distributions of the brightness difference of one component relative to the brightness difference of the other component in the R, G, and B values ​​of the image of the metal strip, plotted in a two-dimensional chart composed of the two components.

Images