Inspection device, inspection method, glass plate manufacturing method, and inspection program

Abstract

To achieve both reduction of human cost and maintenance of detection accuracy in detecting the size of a defect generated in a glass plate.SOLUTION: An inspection device (1) includes: an image acquisition section (101) for acquiring an image by capturing a glass plate; and a size detection section (103) for detecting, as the size of a defect, the size of a range in a deduction result obtained by inputting the captured image to a size detection model (112) configured to learn to deduct the position and range of the defect generated in the glass plate.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an inspection device or the like that inspects a glass plate for defects based on an image of the glass plate. [Background technology]

[0002] Conventionally, defect inspection of glass sheets in the manufacture of glass sheets, particularly detection of the size of defects occurring in the glass sheets, has been performed visually, which inevitably results in personnel costs. To address this problem, Patent Document 1 discloses a technology for detecting the size of defects by performing image processing on an image of an edge face of a glass sheet. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2004 / 079352 Summary of the Invention [Problem to be solved by the invention]

[0004] The inventors of the present application have found that the above-described conventional techniques sometimes fail to detect the size of a defect with sufficient accuracy, which is thought to be partly due to the fact that the shading around the defect is often unclear in an image of the glass plate.

[0005] An object of the present invention is to provide an inspection device and the like that can reduce personnel costs while maintaining detection accuracy in detecting the size of defects that occur in glass sheets. [Means for solving the problem]

[0006] In order to solve the above problems, an inspection device according to one aspect of the present invention includes an image acquisition unit that acquires an image of a glass plate, and a size detection unit that inputs the image into a trained model that has learned the position and range of a defect that has occurred in the glass plate, so as to infer the position and range of the defect, and detects the size of the range in the inference result as the size of the defect that has occurred in the imaged glass plate.

[0007] In order to solve the above-mentioned problems, an inspection method according to one aspect of the present invention is an inspection method executed by an inspection device, and includes an image acquisition step of acquiring an image of a glass plate, and a size detection step of inputting the image into a trained model that has learned the position and range of a defect that has occurred in the glass plate, so as to infer the position and range of the defect, and detecting the size of the range in the inference result as the size of the defect that has occurred in the imaged glass plate.

[0008] In order to solve the above-mentioned problems, one aspect of the present invention provides a method for manufacturing a glass plate, the method including: a step of forming a glass raw plate into a glass plate of a predetermined size; and an inspection step of the glass plate, which is carried out by an inspection device; the inspection step including: an image acquisition step of acquiring an image of the glass plate; and a size detection step of inputting the image into a trained model that has learned the position and range of a defect that has occurred in the glass plate, so as to infer the position and range of the defect, and detecting the size of the range in the inference result as the size of the defect that has occurred in the imaged glass plate. [Effects of the Invention]

[0009] According to one aspect of the present invention, in detecting the size of a defect occurring in a glass plate, it is possible to achieve both a reduction in personnel costs and maintenance of detection accuracy. [Brief explanation of the drawings]

[0010] [Figure 1]1 is a block diagram showing an overview of an inspection system according to a first embodiment of the present invention and an example of the configuration of a main part of an inspection device included in the inspection system. [Figure 2] FIG. 10 illustrates an example of inferring the location and extent of a defect using a size detection model. [Figure 3] 3 is a flowchart showing an example of a glass plate manufacturing process involving inspection by the inspection device. [Figure 4] 10 is a flowchart showing an example of an inspection process included in the manufacturing process. [Figure 5] FIG. 10 illustrates an example of defect size detection based on reliability. [Figure 6] 10 is a flowchart showing an example of an inspection process according to the second embodiment of the present invention. [Figure 7] FIG. 10 is a diagram showing an outline of defect size detection performed by a size detection unit according to a third embodiment of the present invention. [Figure 8] 10 is a flowchart showing an example of an inspection process according to the third embodiment of the present invention. [Figure 9] 10 is a flowchart showing an example of an inspection process according to a fourth embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0011] [Embodiment 1] (Inspection system 100) 1 is a block diagram showing an overview of an inspection system 100 according to this embodiment and an example of the configuration of a main part of an inspection device 1 included in the inspection system 100.

[0012] The inspection system 100 is a system for detecting the size of defects occurring on the edge surface of a glass plate. In this embodiment, the glass plate is described as being rectangular, but the shape of the glass plate is not limited to this example. The defects are, for example, cracks or chips occurring on the edge surface of the glass plate. The inspection system 100 includes an inspection device 1, an imaging device 2, an image storage device 3, and a detection result storage device 4.

[0013] The inspection device 1 detects the size of defects occurring on the edge surface of a glass plate. Defect size detection on the edge surface of a glass plate has traditionally been performed visually. However, accurate size detection using image processing can be difficult due to the lack of clarity in the shading around the defect in the captured image. As described in detail below, the inspection device 1 detects the size of the defect by inputting the captured image into a trained model trained to infer the location and range of a defect occurring on the edge surface. In other words, since no human is involved in detecting the size of defects on the edge surface, the labor costs associated with such detection can be reduced. Furthermore, by training the trained model using a large number of images of the edge surface with unclear shading around the defect as training data, it becomes possible to accurately detect defects even from images of the edge surface with unclear shading around the defect. Therefore, it is possible to achieve both reduced labor costs and maintained detection accuracy when detecting the size of defects occurring on the edge surface of a glass plate.

[0014] The imaging device 2 is a device that captures images of the edge surfaces of a glass plate. Here, the edge surfaces refer to the side surfaces of the glass plate when the two widest surfaces are the top and bottom surfaces, respectively, and can also be called the outer periphery. Hereinafter, the four edge surfaces of the glass plate may be referred to as the X1 edge surface, the X2 edge surface, the Y1 edge surface, and the Y2 edge surface. Note that the X1 edge surface and the X2 edge surface are parallel to each other, and the Y1 edge surface and the Y2 edge surface are parallel to each other.

[0015] As an example, the imaging device 2 is disposed at a position along the conveying path of a conveying device (not shown) that conveys the glass sheet, and continuously captures images of one edge of the glass sheet at predetermined time intervals while it is being conveyed by the conveying device. In this way, by capturing images of one edge multiple times, multiple captured images of one edge divided into multiple parts can be obtained, and the resolution of the edge in each captured image can be made high enough for inspection. Of course, if an image of sufficient resolution can be obtained by capturing an image of the entire edge in one capture, it is not necessary to capture the image only once.

[0016] 1 shows only one imaging device 2, an imaging device 2 may be provided for each edge surface of the glass plate. For example, two imaging devices 2 may be arranged facing each other across the conveying path to simultaneously capture images of parallel edge surfaces (for example, the X1 and X2 edge surfaces, or the Y1 and Y2 edge surfaces). Furthermore, if the conveying path branches, an imaging device 2 may be provided at each branch.

[0017] Although not shown in the figure, the imaging device 2 is equipped with an illumination device and an information processing device, so that the illumination device can irradiate light onto the edge surface of the glass plate while the imaging device 2 can capture an image of the edge surface.

[0018] The information processing device is a device that adds additional information to a captured image and stores the image in the image storage device 3. The additional information includes glass identification information that indicates the glass plate shown in the captured image and glass position information that indicates which part of the glass plate is shown in the captured image.

[0019] The glass plate identification information may be, for example, an identification number assigned to each glass plate. The glass plate position information may be any information indicating which part of the glass plate it is. For example, if the conveying speed of the glass plate is constant and the number of times images are taken per edge surface is also constant, the glass plate position information may be information indicating which image was taken.

[0020] The image storage device 3 is a storage device that stores the captured images captured by the imaging device 2. The detection result storage device 4 is a storage device that stores the detection results detected by the inspection device 1, i.e., information indicating the size of the defects. Note that multiple image storage devices 3 and multiple detection result storage devices 4 may be provided. For example, an image storage device 3 and a detection result storage device 4 may be provided for each end face. It is also possible to omit the image storage device 3 and the detection result storage device 4. In this case, the imaging device 2 may transmit the captured images to the inspection device 1, and the inspection device 1 may store the detection results in the storage unit 11.

[0021] (Inspection device 1) As shown in Fig. 1, the inspection device 1 includes a control unit 10, a storage unit 11, and a communication unit 12. The control unit 10 controls all the units of the inspection device 1. The storage unit 11 stores various data used by the inspection device 1. The communication unit 12 enables the inspection device 1 to communicate with other devices. Typical examples of such other devices are an image storage unit 3 and a detection result storage unit 4.

[0022] 1, the control unit 10 includes an image acquisition unit 101, a defect determination unit 102, and a size detection unit 103. The storage unit 11 stores a defect determination model 111 and a size detection model 112.

[0023] The image acquisition unit 101 acquires a captured image of a glass plate. In this embodiment, the captured image acquired by the image acquisition unit 101 is an image of an end face, as described above. As an example, the image acquisition unit 101 receives the captured image from the image storage device 3 via the communication unit 12.

[0024] The defect determination unit 102 determines whether or not there is a defect on the end face shown in the captured image. Specifically, the defect determination unit 102 inputs the captured image acquired from the image acquisition unit 101 to a defect determination model 111, and determines whether or not there is a defect based on the inference result output from the defect determination model 111.

[0025] Here, the defect determination model 111 will be described. The defect determination model 111 is a trained model that has been trained to infer the presence or absence of a defect in a captured image of an edge surface of a glass plate. Such a defect determination model 111 can be constructed by machine learning using a large number of captured images, the presence or absence of which are known, as training data. The machine learning algorithm is not particularly limited as long as it can generate the defect determination model 111 that can classify captured images into two categories: those with a defect and those without a defect. For example, a deep learning convolutional neural network or the like, which has high image classification accuracy, is suitable, but is not limited to this example.

[0026] The defect determination unit 102 may also determine the type of defect. In this case, training data may be prepared for each type of defect, and machine learning may be performed using the training data. Examples of the types of defects include the cracks and chips described above.

[0027] The size detection unit 103 detects the size of the defect range in the inference result obtained by inputting the captured image to the size detection model 112 as the size of the defect occurring in the captured glass plate.

[0028] Here, the size detection model 112 will be described. The size detection model 112 is a trained model that has learned the position and range of a defect that has occurred on a glass plate so as to infer the position and range of the defect. Such a size detection model 112 can be constructed by machine learning using training data in which the position and range of the defect are associated as ground truth data with a captured image that shows the defect. The position and range of the defect may be represented, for example, by a rectangle that surrounds the defect. In this case, the position of the rectangle indicates the position of the defect, and the width and height of the rectangle indicate the range, i.e., the size, of the defect.

[0029] As the training data, it is preferable to use a large number of captured images in which the shading around the defect is unclear. This makes it possible to accurately detect the size of the defect even from captured images in which the shading around the defect is unclear. The machine learning algorithm, like the defect determination model 111, is not particularly limited.

[0030] Furthermore, when the defect determination unit 102 determines the type of defect, a size detection model 112 may be prepared in advance for each defect type, and size detection may be performed using the size detection model 112 corresponding to the type determined by the defect determination unit 102. This improves the accuracy of size detection. For example, a size detection model 112 for a crack defect may be constructed using a captured image of an edge surface of a glass plate where a crack has occurred as training data, and a size detection model 112 for a chip defect may be constructed using a captured image of an edge surface of a glass plate where a chip has occurred as training data. In this case, for a captured image in which the defect determination unit 102 has determined that a crack defect has occurred, size detection may be performed using the size detection model 112 for the crack defect. On the other hand, for a captured image in which the defect determination unit 102 has determined that a chip defect has occurred, size detection may be performed using the size detection model 112 for the chip defect.

[0031] (Example of inference) 2 is a diagram showing an example of inferring the position and range of a defect using size detection model 112. In this example, the image of the edge surface of a glass plate captured by imaging device 2 is captured image 21. The black portion extending horizontally in the center of captured image 21 is the edge surface of the glass plate, and the white portion near the center of the edge surface of the glass plate is the defect. As described above, captured image 21 is stored in image storage device 3.

[0032] The image acquisition unit 101 of the inspection device 1 acquires the captured image 21 from the image storage device 3. When the captured image 21 is acquired, the defect determination unit 102 first inputs the captured image 21 to the defect determination model 111 and determines whether or not there is a defect based on the inference result output from the defect determination model 111. As described above, a defect is captured in the captured image 21, so the defect determination unit 102 determines that there is a defect. Note that, as described above, the defect determination unit 102 may also determine the type of defect, but here it only determines whether or not there is a defect.

[0033] For a captured image 21 that has been determined to have a defect by the defect determination unit 102, the size is detected by the size detection unit 103. That is, the size detection unit 103 inputs the captured image 21 to a size detection model 112 and obtains an inference result output by the size detection model 112. This inference result indicates the position and range of the defect in the captured image 21.

[0034] The size detection unit 103 stores the range indicated by the above inference result as size information indicating the size of the defect in the detection result storage device 4. For example, if the range is a rectangle, the size detection unit 103 may store representative coordinates indicating the position of the rectangle in the captured image 21 (for example, the coordinates of the upper left corner of the rectangle) and size information indicating the width and height of the rectangle.

[0035] 2, the inference result of the size detection model 112 is shown as a rectangle 41. As shown, the width and height of the rectangle 41 are equal to the width and height of the defect, which indicates that an accurate inference was made.

[0036] Such a rectangle is also called an annotation. By displaying the annotation, the user of the inspection device 1 can visually confirm the inference result. By using the size detection model 112 constructed by machine learning using captured images that show defects as training data, it becomes possible to accurately detect such defective parts.

[0037] (Inspection during the glass plate manufacturing process) Inspection by the inspection device 1 may be performed as part of a manufacturing process of a glass sheet. Here, an example of a manufacturing process of a glass sheet that includes inspection by the inspection device 1 will be described with reference to Fig. 3. Fig. 3 is a flowchart showing an example of a manufacturing process of a glass sheet that includes inspection by the inspection device 1.

[0038] In S101, a glass blank is formed into a glass plate of a predetermined size. The glass blank is a glass plate larger than the glass plate to be produced, and is manufactured by a glass blank manufacturing apparatus. In S101, the glass blank is adjusted to a glass plate of the predetermined size by, for example, a cutting device that cuts the glass blank into a predetermined size and a processing device that processes the end faces of the glass plate after cutting.

[0039] In S102, the inspection device 1 performs an inspection process. A glass plate determined to be a non-defective product in this inspection process becomes a finished product. Details of the inspection process will be described below with reference to FIG.

[0040] (Inspection process flow) Fig. 4 is a flowchart showing an example of the inspection process shown in Fig. 3. It is assumed that before the start of the inspection process, the imaging device 2 images the edge surface of the glass plate to be inspected, and the captured image is stored in the image storage device 3. The timing of executing the inspection process shown in Fig. 4 is not particularly limited. For example, the inspection process may be performed every time a new image is captured and stored in the image storage device 3, every time imaging of one edge surface of one glass plate is completed, or after imaging of all edge surfaces of all glass plates to be inspected is completed.

[0041] In S1 (image acquisition step), the image acquisition unit 101 acquires a captured image from the image storage device 3. If multiple captured images are stored in the image storage device 3, it is sufficient to acquire a captured image that has not yet been subjected to the inspection process. Then, the image acquisition unit 101 outputs the acquired captured image to the defect determination unit 102.

[0042] In S2, the defect determination unit 102 inputs the captured image acquired in S1 into the defect determination model 111, and determines whether or not there is a defect based on the inference result output from the defect determination model 111. Here, the defect determination unit 102 also determines the type of defect.

[0043] In S3, the size detection unit 103 determines to use, for size detection, a size detection model 112 corresponding to the type of defect determined in S2 from among a plurality of size detection models previously stored in the storage unit 11. If it is determined in S2 that there is no defect, the processes from S3 onward are not performed, and the inspection process for the captured image acquired in S1 ends.

[0044] In S4 (size detection step), the size detection unit 103 inputs the captured image acquired in S1 to the size detection model 112 determined in S3. Then, the size detection unit 103 detects the size of the range in the inference result output from the size detection model 112 as the size of the defect. For example, if the range is rectangular, the size detection unit 103 detects the width and height of the rectangle as the width and height of the defect. Note that the size detection unit 103 may convert the detected width and height into actual dimensions.

[0045] In S5, the size detection unit 103 stores the size detected in S4 in the detection result storage device 4. Specifically, the size detection unit 103 transmits size information indicating the width and height of the defect to the detection result storage device 4 and stores it therein. This completes the inspection process for the captured image acquired in S1.

[0046] Although not shown in Figure 4, the inspection process of Figure 4 is repeated until inspection of the captured images of all end faces of at least one glass plate to be inspected is completed, and then the process returns to the manufacturing process of Figure 3.

[0047] 4 may also include a step of determining whether a glass plate, for which the presence or absence of defects has been determined on all edge surfaces, is a pass or fail product. A glass plate determined to be a pass or fail product in this step becomes a finished product. The criteria for determining whether a glass plate is a pass or fail product may be determined as appropriate. For example, a glass plate in which a defect of a predetermined size or larger is detected may be determined to be a fail product, and a glass plate in which no defect is detected or in which the detected defect is smaller than the predetermined size may be determined to be a pass or fail product. This determination may be made by the inspection device 1 or by an information processing device other than the inspection device 1.

[0048] (Actions and Effects) As described above, the inspection device 1 according to this embodiment includes an image acquisition unit 101 that acquires a captured image of an edge surface of a glass plate. Furthermore, the inspection device 1 includes a size detection unit 103 that detects the size of a range in an inference result obtained by inputting the captured image into a size detection model 112 as the size of a defect that has occurred in the imaged glass plate. Here, the size detection model 112 is a trained model that has learned the position and range of a defect that has occurred on the edge surface so as to infer the position and range of the defect.

[0049] Furthermore, as described above, the inspection method according to this embodiment includes an image acquisition step (S1) for acquiring an image of a glass plate, and a size detection step (S4) for detecting the size of the range in the inference result obtained by inputting the image into a size detection model 112, which is a trained model that has learned the position and range of a defect that has occurred on the glass plate, as the size of the defect that has occurred on the imaged glass plate.

[0050] Furthermore, as described above, the glass plate manufacturing method according to this embodiment is a glass plate manufacturing method including a step (S101) of forming a glass raw plate into a glass plate of a predetermined size, and an inspection step (S102) of the glass plate, which is executed by the inspection device 1. The inspection step includes an image acquisition step (S1) of acquiring an image of the glass plate, and a defect detection step (S4) of inputting the image into a size detection model 112, which is a trained model that has learned the position and range of a defect that has occurred in the glass plate, so as to infer the position and range of the defect, and detecting the size of the range in the inference result as the size of the defect that has occurred in the imaged glass plate.

[0051] With these configurations, since no human beings are involved in detecting the size of defects, the labor costs involved in such detection can be reduced. Furthermore, in training the size detection model 112, by training a large number of images in which the shading around the defects is unclear as training data, it becomes possible to accurately detect defects even from images in which the shading around the defects is unclear. Therefore, in detecting the size of defects that have occurred on glass sheets, it is possible to achieve both a reduction in labor costs and maintenance of detection accuracy.

[0052] [Embodiment 2] Other embodiments of the present invention will be described below. For ease of explanation, the same reference numerals will be used to designate components having the same functions as those described in the above embodiments, and their descriptions will not be repeated. This also applies to the third and subsequent embodiments.

[0053] In this embodiment, an example will be described in which the size of a defect is detected based on the reliability of the inference result, which is output together with the inference result by the size detection model 112. More specifically, the size detection unit 103 according to this embodiment detects the size of a rectangle in an inference result whose reliability is equal to or greater than a predetermined threshold as the size of a defect that has occurred in the imaged glass plate.

[0054] The reliability is a value indicating the likelihood of the inference result, and is, for example, a numerical value between 0 and 1. The reliability according to this embodiment indicates that the higher the numerical value, the higher the possibility that the inference result is actually a defect.

[0055] The reliability may be calculated, for example, by numerically analyzing the pixel values ​​of each pixel that constitutes the captured image. In the captured image, pixel values ​​(shading) vary greatly between areas with defects and areas without defects. Therefore, the reliability may be calculated according to the amount of change in pixel value, such as by setting the reliability low when the amount of change in pixel value is small and setting the reliability high when the amount of change in pixel value is large. In this way, the reliability of the value corresponding to the amount of change in pixel value is calculated.

[0056] (Example of confidence-based size detection) Fig. 5 is a diagram showing an example of defect size detection based on reliability. Captured image 32 shown in Fig. 5 is an image of an edge surface of a glass plate having a defect, on which rectangles 42 to 44 indicating the inference results of size detection model 112 and numerical values ​​52 to 54 indicating the reliability of these inference results are drawn.

[0057] As shown, in this example, the size detection model 112 infers that the areas indicated by rectangles 42-44 are defects, and the confidence levels of these inferences are 0.95, 0.90, and 0.75, respectively, as indicated by numbers 52-54.

[0058] For example, suppose the reliability threshold is set to 0.80. In this case, among rectangles 42 to 44, size detection unit 103 detects the sizes of rectangles 42 and 43, whose reliability is 0.80 or greater, as the size of a defect. On the other hand, size detection unit 103 does not detect the size of rectangle 44, whose reliability is less than 0.80, as the size of a defect.

[0059] Therefore, when the final size detection result by size detection unit 103 is displayed on captured image 32, only rectangle 42 and rectangle 43 are shown, as shown in the lower part of Fig. 5. In this case, size detection unit 103 transmits size information indicating the width and height of rectangle 42 and size information indicating the width and height of rectangle 43 to detection result storage device 4, where they are stored.

[0060] (Inspection process flow) Fig. 6 is a flowchart showing an example of an inspection process according to the present embodiment. In this flowchart, the same processes as those in the flowchart of Fig. 4 are assigned the same numbers as in Fig. 4, and description thereof will not be repeated.

[0061] In S11, the size detection unit 103 inputs the captured image 21 into the size detection model 112 determined in S3, and acquires an inference result of the size of the defect. The inference result includes the position and size of the area inferred by the size detection model 112 to be a defect, as well as the reliability of the inference.

[0062] In S12, the size detection unit 103 detects the size of the range in the inference result output in S11, whose reliability is equal to or greater than a threshold, as the size of the defect.

[0063] (Actions and Effects) As described above, in the inspection device 1 according to this embodiment, the size detection model 112 further outputs the reliability of the inference result. Furthermore, the size detection unit 103 detects the size of the defect range in the inference result whose reliability is equal to or greater than a predetermined threshold as the size of the defect that has occurred in the imaged glass plate.

[0064] According to this, the size of the range in the highly reliable inference result is detected as the size of the defect, so that even if the position and range are obtained as the inference result for a part that is not a defect, the possibility of erroneously detecting the size of the range as the size of a defect can be reduced.

[0065] In particular, variations in the conveyance direction of the glass plate, etc., can change the way light hits the plate when it is imaged, and color unevenness can occur even in areas without defects. The size detection model 112 may mistakenly infer that such color unevenness occurs as a defect, but by setting an appropriate threshold, it becomes possible to distinguish between color unevenness that does not affect the quality of the product and true defects, and perform appropriate detection.

[0066] [Embodiment 3] When multiple inference results are obtained, the size detection unit 103 according to this embodiment detects the size of the area that includes multiple ranges in the multiple inference results as the size of the defect that has occurred on the imaged glass plate.

[0067] Fig. 7 is a diagram showing an outline of defect size detection performed by size detection unit 103 of this embodiment. Captured image 33 shown in Fig. 7 is a captured image of an edge surface of a glass plate including a defective portion, on which rectangles 45 to 48 indicating the inference results of size detection model 112 are drawn.

[0068] When multiple defect areas are detected in a single captured image, the size detection unit 103 of this embodiment determines the size of the smallest area that encompasses those areas. For example, the size detection unit 103 may identify the topmost side of the rectangles 45 to 48 and the bottommost side of the rectangles 45 to 48, and use the distance between those sides as the height of the area to be determined. Alternatively, the size detection unit 103 may identify the leftmost side of the rectangles 45 to 48 and the rightmost side of the rectangles 45 to 48, and use the distance between those sides as the width of the area to be determined.

[0069] Through this processing, size detection unit 103 can identify inclusion area 61 that encompasses rectangles 45 to 48. Size detection unit 103 then transmits size information indicating the width and height of inclusion area 61 to detection result storage device 4, where it is stored.

[0070] 7, inclusion area 61 that includes mutually adjacent rectangles 45 to 48 is identified, but the present invention is not limited to this. In other words, size detection unit 103 may detect the size of an area that includes multiple ranges that are not mutually adjacent as the size of the defect.

[0071] This configuration is effective, for example, when only the two ends of an actual defect are inferred as defects and the center portion of the defect is not inferred as a defect. In this case, the size detection unit 103 identifies an inclusion area that includes the range of the detected two ends, and can detect the size of the inclusion area that has a small error from the actual size of the defect as the size of the defect.

[0072] (Inspection process flow) Fig. 8 is a flowchart showing an example of an inspection process according to this embodiment. In this flowchart, the same processes as those in the flowchart of Fig. 4 are assigned the same numbers as those in Fig. 4. Furthermore, the same processes as those in the flowchart of Fig. 6 are assigned the same numbers as those in Fig. 6. Description of the same processes will not be repeated.

[0073] In S21, the size detection unit 103 determines whether the inference result acquired in S11 is multiple or not. If it is determined that there is multiple (YES in S21), the inspection process proceeds to S22. On the other hand, if it is determined that there is not multiple (NO in S21), the inspection process proceeds to S23.

[0074] In S22, size detection unit 103 calculates the size of an inclusion area that includes the ranges indicated by the inference result obtained in S11, and detects this size as the size of the defect. The method for calculating the size of the inclusion area is as described with reference to FIG. 7.

[0075] In S23, the size detection unit 103 detects the size of the range in the acquired inference result as the size of the defect. That is, the size detection unit 103 detects the width and height indicated in one of the inference results acquired in S11 as the size of the defect.

[0076] (Actions and Effects) As described above, in the inspection device 1 according to this embodiment, when multiple inference results are obtained, the size detection unit 103 detects the size of the inclusion area that encompasses the multiple ranges in the multiple inference results as the size of the defect that has occurred on the imaged glass plate.

[0077] With this configuration, the size of the inclusion area is detected as the size of the defect, so even if multiple ranges are obtained that have a large size difference from the actual defect, the multiple ranges can be corrected to an inclusion area that has a small size difference from the actual defect.As a result, even if multiple ranges are obtained, the defect size detected using size detection model 112 can be detected with a smaller difference from the actual size.

[0078] Furthermore, even if multiple defects are detected on a glass plate, if the detected defects are all small in size, the inspection device 1 may ultimately determine that the glass plate is a non-defective product. This determination result is valid if the defects are actually small in size, but if a large defect actually exists and only a portion of it is detected as a defect, this determination result is an erroneous determination. According to the configuration of this embodiment, the size of an inclusion area that encompasses the range of multiple defects is detected, thereby reducing the possibility of such an erroneous determination.

[0079] (Variation) This embodiment can be combined with embodiment 2. Specifically, when there are multiple inference results whose reliability is equal to or greater than a predetermined threshold, the size detection unit 103 may identify an inclusion area that includes the range of the inference results, and detect the size of the inclusion area as the size of the defect that has occurred in the imaged glass plate.

[0080] [Embodiment 4] When the size of a detected defect is outside the normal range, the size detection unit 103 according to this embodiment performs numerical analysis of the pixel values ​​of each pixel constituting the captured image to re-detect the size of the defect. The normal range may be determined in advance based on, for example, the size of a glass plate or the size of a general defect. The size of a defect outside the normal range may be, for example, a size that satisfies at least one of the following two conditions: (1) the width of the rectangle resulting from the inference is outside a predetermined first numerical range, and (2) the height of the rectangle resulting from the inference is outside a predetermined second numerical range.

[0081] Like the size detection unit 103 in each of the above-described embodiments, the size detection unit 103 in this embodiment also detects the size based on the inference result output by the size detection model 112. The size detection unit 103 in this embodiment differs from the size detection unit 103 in each of the above-described embodiments in that it determines whether the detected size is within a normal range.

[0082] If the detected size is outside the normal range, the size detection unit 103 according to this embodiment performs numerical analysis on the pixel values ​​of each pixel constituting the captured image to re-detect the size of the defect. The size detection unit 103 according to this embodiment differs from the size detection unit 103 of each of the above-described embodiments in this respect as well. The processing block that performs the numerical analysis may be a processing block separate from the size detection unit 103.

[0083] The method for detecting the size of a defect by numerical analysis is not particularly limited, and various methods can be applied. For example, as shown in Figure 2, in the captured image, pixel values ​​change significantly at the boundary between the area where the edge of the glass plate is captured and the background area. Therefore, the size detection unit 103 may first extract the area where the edge of the glass plate is captured based on this change in pixel values.

[0084] 2 and other figures, the pixel values ​​of the defect-containing portion of the edge of the glass plate captured in the captured image are different from those of the defect-free portion. Therefore, if the region containing the defect-containing portion of the edge of the glass plate extracted as described above includes a region with pixel values ​​specific to the defect-containing portion, the size detection unit 103 can detect the width and height of that region as the defect size. The size detection unit 103 then transmits size information indicating the detected width and height to the detection result storage device 4 for storage.

[0085] (Inspection process flow) Fig. 9 is a flowchart showing an example of an inspection process according to the present embodiment. In this flowchart, the same processes as those in the flowchart of Fig. 4 are assigned the same numbers as in Fig. 4, and description thereof will not be repeated.

[0086] In S31, the size detection unit 103 determines whether the size of the defect detected in S4 is within the normal range. If it is determined to be within the normal range (YES in S31), the inspection process proceeds to S5. If it is determined to be not within the normal range, i.e., out of the normal range (NO in S31), the inspection process proceeds to S32.

[0087] In S32, the size detection unit 103 discards the size detection result based on the inference result output by the size detection model 112. In S33, the size detection unit 103 performs numerical analysis of the pixel values ​​of each pixel constituting the captured image to redetect the size of the defect. After this, the inspection process proceeds to S5.

[0088] In S5 after transition from S33, the size detection unit 103 may store information indicating that re-detection has been performed in association with the size information in the detection result storage device 4. This allows the user of the inspection system 100 to identify the captured image for which re-detection has been performed.

[0089] Also, S32 may be omitted. In this case, in S5, the size detection unit 103 may store in the detection result storage device 4 both the size information indicating the detection result of S4 and the size information indicating the detection result of S33.

[0090] (Actions and Effects) As described above, in the inspection device 1 according to this embodiment, if the size of a detected defect is outside the normal range, the size detection unit 103 performs numerical analysis of the pixel values ​​of each pixel that constitutes the captured image to redetect the size of the defect.

[0091] With this configuration, for defects whose sizes are outside the normal range, the size is re-detected by numerically analyzing the pixel values ​​of each pixel that makes up the captured image. In other words, the size is re-detected using a method different from the defect size detection using the trained model, so the defect can be corrected to an appropriate size.

[0092] [Modification] In the above-described embodiments, examples of detecting the size of a defect occurring on an edge surface of a glass sheet have been described, but the size of a defect occurring in a portion other than the edge surface of the glass sheet may also be detected. Furthermore, in the above-described embodiments, examples of detecting the size of a defect occurring in a glass sheet cut from a glass blank have been described, but the glass sheet to be detected is not limited to one cut from a glass blank. For example, it is also possible to detect the size of a defect occurring in a glass sheet portion of a product that includes a glass sheet.

[0093] In addition, in each of the above-described embodiments, an example has been shown in which one inspection device is used to determine the presence or absence (and type) of a defect and to detect its size, but these processes may be performed by separate devices. In other words, the inspection method described in each of the above-described embodiments may be performed by one inspection device or by multiple inspection devices.

[0094] Furthermore, in the above-described embodiments, examples have been described in which the defect determination model 111 and the size detection model 112 are separate models, but it is also possible to use a single model that has learned the presence or absence of a defect and the location and range of the defect. In this case, when a captured image is input to the single model, an inference result indicating the presence or absence of a defect is output. Furthermore, if a defect is present, an inference result indicating the location and range of the defect is also output. It is also possible to have this single model infer the type of defect.

[0095] [Software implementation example] The functions of the inspection device 1 (hereinafter referred to as the "device") can be realized by a program (inspection program) that causes a computer to function as the device, and that causes a computer to function as each control block of the device (particularly each part included in the control unit 10).

[0096] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The control device and storage device execute the program, thereby realizing the functions described in each of the above embodiments.

[0097] The program may be non-transitory and may be recorded on one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0098] Furthermore, some or all of the functions of the control blocks can be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the control blocks is formed is also included in the scope of the present invention. In addition, the functions of the control blocks can also be realized by, for example, a quantum computer.

[0099] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention. [Explanation of symbols]

[0100] 1. Inspection equipment 101 Image acquisition unit 103 Size detection unit 112 Size Detection Model

Claims

1. an image acquisition unit that acquires an image of the glass plate; a size detection unit that inputs the captured image into a trained model that has learned the position and range of a defect occurring in the glass plate, and detects the size of the range in an inference result obtained by inputting the captured image as the size of the defect occurring in the imaged glass plate; When multiple inference results are obtained, the size detection unit detects the size of an area that includes multiple ranges in the multiple inference results as the size of a defect that has occurred in the imaged glass plate.

2. The inspection device described in Claim 1, wherein when multiple inference results are obtained, the size detection unit detects the size of an area in the multiple inference results that includes multiple ranges that are not in contact with each other as the size of a defect that has occurred in the imaged glass plate.

3. The trained model further outputs a confidence level for the inference result, The inspection device according to claim 1 or 2, wherein the size detection unit detects the size of the range in the inference result in which the reliability is equal to or greater than a predetermined threshold as the size of the defect occurring in the imaged glass plate.

4. 4. The inspection device according to claim 1, wherein, when the size of the detected defect is outside a normal range, the size detection unit redetects the size of the defect by performing a numerical analysis of pixel values ​​of each pixel constituting the captured image.

5. The inspection device according to claim 1 , wherein the image acquisition unit acquires the captured image of an edge surface of the glass plate.

6. 1. An inspection method performed by an inspection device, comprising: an image acquisition step of acquiring an image of the glass plate; a size detection step of detecting the size of the range in an inference result obtained by inputting the captured image into a trained model that has learned the position and range of a defect occurring in the glass plate so as to infer the position and range of the defect, as the size of the defect occurring in the imaged glass plate; In the size detection step, if multiple inference results are obtained, the size of the area that includes multiple ranges in the multiple inference results is detected as the size of the defect that has occurred in the imaged glass plate.

7. A method for manufacturing a glass plate, comprising: a step of forming a glass raw plate into a glass plate of a predetermined size; and an inspection step of inspecting the glass plate, the inspection step being carried out by an inspection device, The inspection step includes: an image acquisition step of acquiring an image of the glass plate; a size detection step of detecting the size of the range in an inference result obtained by inputting the captured image into a trained model that has learned the position and range of a defect occurring in the glass plate so as to infer the position and range of the defect, as the size of the defect occurring in the imaged glass plate; A method for manufacturing a glass plate, wherein, if multiple inference results are obtained in the size detection step, the size of an area that includes multiple ranges in the multiple inference results is detected as the size of a defect that has occurred in the imaged glass plate.

8. 2. An inspection program for causing a computer to function as the inspection apparatus according to claim 1, the inspection program causing the computer to function as the image acquisition unit and the size detection unit.

Images