Inspection method, inspection device, and glass tube manufacturing method

Abstract

To provide a technique for increasing an information amount of image data.SOLUTION: A glass article inspection method includes: a first acquisition step for acquiring first image data which expresses a first image obtained by photographing a glass article to be irradiated with a first light source; a second acquisition step for acquiring second image data which expresses a second image obtained by photographing the glass article to be irradiated with a second light source different from the first light source; and a detection step for detecting that the glass article is not normal with the use of composite image data to be obtained by using the acquired first image data and second image data. A composite pixel among a plurality of pixels included in the composite image data comprises: a first pixel value of a first pixel among the pixels included in the first image data; and a second pixel value of a second pixel among the pixels included in the second image data. A position of the first pixel in the first image and a position of the second pixel in the second image mutually overlap.SELECTED DRAWING: Figure 9

Description

Technical Field

[0001] The technology disclosed in this specification relates to inspecting that an object is abnormal.

Background Art

[0002] Patent Document 1 discloses an inspection apparatus that inputs an image of a target product into a machine learning model and outputs an image showing an estimation result of the likelihood of a defect in the target product.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] When inspecting from an image of an object that the object is abnormal, the higher the amount of information contained in the acquired image data, the higher the accuracy of the inspection.

[0005] This specification provides a technology capable of increasing the amount of information in image data.

Means for Solving the Problems

[0006] A first aspect of the technology disclosed in this specification relates to a method for inspecting a glass article. The inspection method includes a first acquisition step of acquiring first image data representing a first image of the glass article imaged while being irradiated by a first light source, a second acquisition step of acquiring second image data representing a second image of the glass article imaged while being irradiated by a second light source different from the first light source, and a detection step of detecting that the glass article is abnormal using composite image data obtained using the acquired first image data and the second image data. Among a plurality of pixels included in the composite image data, a composite pixel includes a first pixel value of a first pixel among a plurality of pixels included in the first image data and a second pixel value of a second pixel among a plurality of pixels included in the second image data, and a position of the first pixel in the first image and a position of the second pixel in the second image overlap each other.

[0007] According to the above configuration, composite image data is obtained using image data representing two types of images acquired by imaging a glass article irradiated with a plurality of different types of light sources. The composite image data includes information included in the two types of image data. Thereby, the amount of information of the image data used for detecting that the glass article is abnormal can be increased. As a result, the accuracy of the inspection using the composite image data can be improved.

[0008] In a second aspect, in the above first aspect, in the detection step, it may be detected that the glass article is abnormal using the composite image data by a learning model generated by machine learning using learning data.

[0009] According to the above configuration, the amount of information of the image data used in the learning model can be increased. Thereby, the accuracy of the inspection by the learning model can be improved.

[0010] In the third aspect, in the second aspect described above, the learning model is generated using composite learning image data obtained using first learning image data representing a first learning image obtained by imaging a specific glass article irradiated with the first light source and second learning image data representing a second learning image obtained by imaging the specific glass article irradiated with the second light source. Among the plurality of pixels included in the composite learning image data, the composite learning pixels include a first learning pixel value of a first learning pixel among the plurality of pixels included in the first learning image data and a second learning pixel value of a second learning pixel among the plurality of pixels included in the second learning image data. The position of the first learning pixel in the first learning image and the position of the second learning pixel in the second learning image may overlap each other.

[0011] According to the above configuration, in machine learning for generating a learning model, image data with a large amount of information can be used as learning data. Thereby, the accuracy of inspection by the learning model can be improved.

[0012] In the fourth aspect, in the third aspect described above, the specific glass article may be a normal glass article.

[0013] According to the above aspect, it is not necessary to deliberately prepare a non-normal glass article.

[0014] In the fifth aspect, in any one of the first to fourth aspects described above, the first light source is one of backlight illumination, coaxial illumination, and ring illumination, and the second light source may be one of backlight illumination, coaxial illumination, and ring illumination different from the first light source.

[0015] According to the above aspect, different light sources can be used to generate image data representing images in which different parts of the glass article are easily visible. Thereby, the amount of information included in the composite image data can be increased.

[0016] In the sixth aspect, in the above-described first to fourth aspects, the glass article may be a glass tube.

[0017] According to the above aspect, image data representing an image in which different portions of the glass tube are easily visible can be generated by different light sources. Thereby, the amount of information included in the composite image data can be increased.

[0018] In the seventh aspect, in the above-described sixth aspect, the first light source is one of backlight illumination, coaxial illumination, and ring illumination, the second light source is one of backlight illumination, coaxial illumination, and ring illumination different from the first light source, and in the image data obtained by imaging the glass tube irradiated by the backlight illumination, the pixel value of the pixel representing the outer peripheral portion or the inner peripheral portion of the end face of the glass tube is larger than the pixel value of the pixel representing the other portion of the glass tube. In the image data obtained by imaging the glass tube irradiated by the coaxial illumination, the pixel value of the pixel representing the flat portion of the end face of the glass tube is larger than the pixel value of the pixel representing the other portion of the glass tube. In the image data obtained by imaging the glass tube irradiated by the ring illumination, the pixel value of the pixel representing the R-shaped portion formed between the flat portion of the end face of the glass tube and the outer peripheral portion or the inner peripheral portion of the end face may be larger than the pixel value of the pixel representing the other portion of the glass tube.

[0019] By using any one of backlight illumination, coaxial illumination, and the ring illumination for the glass tube, image data representing an image in which portions where defects are likely to occur in the manufacture of the glass tube are easily visible can be generated.

[0020] In the eighth aspect, in any one of the above-described first to seventh aspects, an output step of outputting a composite image represented by the composite image data is further provided, and the composite pixel may include the first pixel value assigned the first color and the second pixel value assigned the second color.

[0021] According to the above-described embodiment, the user can acquire an image in which the glass article is represented in a plurality of colors. Thereby, the user can use the image to check the state of the glass article.

[0022] In a ninth embodiment, in the above-described first aspect, a third acquisition step of acquiring third image data representing a third image obtained by imaging the glass article irradiated by a third light source different from the first light source and the second light source is further provided. In the detection step, using the composite image data obtained by further using the acquired third image data, it is detected that the glass article is abnormal. The composite pixel further includes a third pixel value of a third pixel among a plurality of pixels included in the third image data. The position of the first pixel in the first image, the position of the second pixel in the second image, and the position of the third pixel in the third image may overlap with each other.

[0023] According to the above-described embodiment, the amount of information of the image data used for detecting that the glass article is abnormal can be increased. As a result, the accuracy of the inspection using the composite image data can be improved.

[0024] In a tenth embodiment, in the above-described ninth aspect, in the detection step, it may be detected that the glass article is abnormal by using the composite image data by a learning model generated by machine learning using learning data.

[0025] According to the above-described embodiment, the amount of information of the image data used in the learning model can be increased. Thereby, the accuracy of the inspection by the learning model can be improved.

[0026] In the eleventh form, in the above-mentioned tenth aspect, the learning model is generated using composite learning image data obtained using first learning image data representing a first learning image of a specific glass article imaged by the first light source, second learning image data representing a second learning image of the specific glass article irradiated by the second light source, and third learning image data representing a third learning image of the specific glass article irradiated by the third light source. Among the plurality of pixels included in the composite learning image data, the composite learning pixels include a first learning pixel value of a first learning pixel among the plurality of pixels included in the first learning image data, a second learning pixel value of a second learning pixel among the plurality of pixels included in the second learning image data, and a third learning pixel value of a third learning pixel among the plurality of pixels included in the third learning image data. The position of the first learning pixel in the first learning image, the position of the second learning pixel in the second learning image, and the position of the third learning pixel in the third learning image may overlap with each other.

[0027] According to the above aspect, in machine learning for generating a learning model, image data with a large amount of information can be used as learning data. Thereby, the accuracy of inspection by the learning model can be improved.

[0028] In the twelfth form, in any one of the ninth to eleventh forms above, the first light source is one of backlight illumination, coaxial illumination, and ring illumination, the second light source is one of backlight illumination, coaxial illumination, and ring illumination different from the first light source, and the third light source may be one of backlight illumination, coaxial illumination, and ring illumination different from the first light source and the second light source.

[0029] According to the above aspect, image data representing images captured using three types of light sources, namely backlight illumination, coaxial illumination, and ring illumination, can be obtained. Thereby, the amount of information included in the composite image data can be increased.

[0030] In the 13th form, in the 9th to 12th forms described above, an output step of outputting a composite image represented by the composite image data is further provided, and the composite pixel may include the first pixel value to which the first color is assigned, the second pixel value to which the second color is assigned, and the third pixel value to which the third color is assigned.

[0031] According to the above form, the user can acquire a color image. Thereby, the user can check the state of the glass article using the image.

[0032] A 14th aspect of the technology disclosed in this specification relates to an inspection apparatus. The inspection apparatus includes a first imaging unit including a first light source and a first imaging device that images the glass article irradiated by the first light source to generate first image data representing a first image, a second light source different from the first light source, and a second imaging device that images the glass article irradiated by the second light source to generate second image data representing a second image, and a detection unit that detects that the glass article is abnormal using the composite image data obtained using the generated first image data and the second image data. Among the plurality of pixels included in the composite image data, the composite pixel includes the first pixel value of the first pixel among the plurality of pixels included in the first image data and the second pixel value of the second pixel among the plurality of pixels included in the second image data, and the position of the first pixel in the first image and the position of the second pixel in the second image overlap each other.

[0033] According to the above configuration, composite image data is obtained using the image data representing two types of images acquired by imaging a glass article irradiated with two different types of light sources. The composite image data includes the information included in the image data representing the two types of images. Thereby, the amount of information of the image data used for detecting that the glass article is abnormal can be increased. As a result, the accuracy of the inspection using the composite image data can be improved.

[0034] In the 15th form, in the above-described 14th form, the first imaging device and the second imaging device may be the same imaging device.

[0035] According to the above form, in order to acquire the image data of two types of light sources, it is not necessary to prepare a plurality of imaging devices.

[0036] In the 16th form, in the above-described 15th aspect, a third light source different from the first light source and the second light source, and a third imaging device that generates third image data representing a third image by imaging the glass article irradiated by the third light source are provided. A third imaging unit is further provided. The detection unit detects that the glass article is abnormal by using the composite image data obtained by using the generated first image data, the second image data, and the third image data. The composite pixel further includes a third pixel value of a third pixel among a plurality of pixels included in the third image data. The position of the first pixel in the first image, the position of the second pixel in the second image, and the position of the third pixel in the third image may overlap each other.

[0037] According to the above form, the amount of information of the image data used for detecting that the glass article is abnormal can be increased. As a result, the accuracy of the inspection using the composite image data can be improved.

[0038] In the 17th form, in the above-described 16th form, the first imaging device, the second imaging device, and the third imaging device may be the same imaging device.

[0039] According to the above form, in order to acquire the image data of three types of light sources, it is not necessary to prepare a plurality of imaging devices.

[0040] The 18th aspect of the technology disclosed in this specification relates to a method for manufacturing a glass tube. The method for manufacturing a glass tube includes a forming step of forming a tube member longer than the glass tube from molten glass, a cutting step of forming a tube intermediate material by cutting the tube member, a mouth-burning step of creating a glass tube by mouth-burning the ends of the tube intermediate material, and a detecting step of detecting abnormal glass tubes by the inspection method according to claim 6.

[0041] According to the above aspect, it becomes possible to appropriately remove fine cracks at the ends of the glass tube formed in the cutting step and not completely removed in the mouth-burning step, and glass tubes having mouth-burning defects (such as shape) that may occur in the mouth-burning step, improving the yield rate of glass tube products.

Brief Description of the Drawings

[0042]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Embodiments for Carrying Out the Invention

[0043] (Configuration of the inspection device: FIGS. 1, 2, and 3) The inspection device 10 identifies a glass tube G having a defect before the manufactured glass tube G is shipped. The glass tubes G conveyed from the manufacturing device are stored in a storage container (not shown) with a plurality of glass tubes G adjacent to each other. In FIGS. 1 to 3, only a plurality of glass tubes G arranged near the imaging device 12 are shown. The bottom plate 14 of the storage container is configured in a mesh shape and has translucency. Note that the bottom plate 14 does not necessarily have to be configured in a mesh shape as long as it has translucency.

[0044] The inspection device 10 includes an imaging unit 11, a beam splitter 32 (see FIG. 2), and a control device 50. The imaging unit 11 includes an imaging device 12, and light sources 20, 30 (see FIG. 2), and 40 (see FIG. 3). The imaging device 12 has a camera that generates image data representing an image of a subject. The lens of the imaging device 12 is arranged facing downward. The optical axis of the imaging device 12 is arranged parallel to the vertical direction.

[0045] Below the imaging device 12, the light sources 20, 30, and 40 are arranged. In FIGS. 1 to 3, for ease of understanding, the light sources 20, 30, and 40 are individually shown. However, in the inspection device 10, the light sources 20, 30, and 40 are fixed below the imaging device 12. According to the imaging timing, the lighting and extinguishing of each of the light sources 20, 30, and 40 are switched. Each of the light sources 20, 30, and 40 has, for example, a plurality of LEDs. At least one of the light sources 20, 30, and 40 may be a metal halide lamp, a laser light source, or the like.

[0046] The light source 20 is arranged below the glass tube G arranged below the imaging device 12, sandwiching the glass tube G. The optical axis of the light source 20 is arranged parallel to the optical axis of the imaging device 12. The light source 20 irradiates light from the opposite side of the glass tube G to the imaging device 12 toward the imaging device 12. The imaging device 12 images the transmitted light L1 from the light source 20 that has passed through the glass tube G. The light source 20 is backlight illumination.

[0047] As shown in FIG. 2, the light source 30 is arranged between the imaging device 12 and the glass tube G in the vertical direction. The optical axis of the light source 30 is arranged in the vertical direction with respect to the optical axis of the imaging device 12. The light source 30 irradiates light toward the space between the imaging device 12 and the glass tube G. A beam splitter 32 is arranged on the optical axis of the light source 30. The beam splitter 32 is arranged sandwiched between the imaging device 12 and the glass tube G. The light irradiated from the light source 30 is reflected by the beam splitter 32 and irradiated onto the glass tube G parallel to the optical axis of the imaging device 12 from directly above in the vertical direction. The imaging device 12 images the reflected light L2 that is reflected by the glass tube G and transmitted through the beam splitter 32. The light source 30 is coaxial illumination.

[0048] As shown in FIG. 3, the light source 40 is arranged sandwiched between the imaging device 12 and the glass tube G in the vertical direction. The light source 40 has a cylindrical shape coaxial with the optical axis of the imaging device 12 between the imaging device 12 and the glass tube G. The light source 40 is a dome light. The light source 40 is arranged in a circumferential direction on the inner peripheral surface of a frustum of a cone shape that gradually expands downward. The optical axis of the light source 40 is arranged in a direction in which light is incident obliquely with respect to the glass tube G. The imaging device 12 images the reflected light L3 reflected by the glass tube G. The light source 40 is ring illumination.

[0049] The control device 50 controls the light sources 20, 30, 40 and the imaging device 12. The control device 50 includes a control unit 52, a storage unit 54, and an output unit 56. The control unit 52 controls the light sources 20, 30, 40 and the imaging device 12. The control unit 52 switches the lighting and extinguishing of each of the light sources 20, 30, 40 according to the imaging timing. The control unit 52 includes a CPU. The CPU executes various processes according to a computer program stored in the storage unit 54. The storage unit 54 includes a memory constituted by a non-volatile memory or the like. The storage unit 54 stores a computer program. Further, the storage unit 54 stores data such as image data generated during the execution of the process by the control unit 52 and during the process.

[0050] The storage unit 54 stores a learning model 60, model parameters 62, and a learning program 64 in addition to a computer program. The learning model 60 is a model (i.e., a mathematical formula) of a multi-layer neural network. The multi-layer neural network is a so-called deep learning model and can be, for example, a convolutional type, a fully connected type, or the like. The model parameters 62 are the values of each weight in the intermediate layer of the learning model 60. The multi-layer neural network is a known technique, and thus detailed description thereof is omitted here. The learning program 64 updates the model parameters 62 by performing machine learning using learning data.

[0051] (Configuration of the glass tube G) The inspection device 10 images the upper end of the glass tube G. As shown in FIG. 4, the upper end of the glass tube G includes a flat portion 102 disposed at the central portion in the thickness direction, R-shaped portions 108 and 110 disposed on the inner peripheral side and the outer peripheral side of the flat portion 102, respectively, an inner peripheral portion 104 located at the inner peripheral end, and an outer peripheral portion 106 located at the outer peripheral end. Each of the flat portion 102, the R-shaped portions 108 and 110, the inner peripheral portion 104, and the outer peripheral portion 106 extends once in the circumferential direction of the glass tube G. The flat portion 102 is formed substantially perpendicular to the central axis of the glass tube G. The R-shaped portions 108 and 110 are disposed adjacent to the flat portion 102. The R-shaped portions 108 and 110 are formed by a process of heating the end portion of the glass tube G during the manufacturing of the glass tube G, so-called necking. In the glass tube G, fine cracks formed during manufacturing such as cutting are easily removed by the necking. The inner peripheral portion 104 is located at the upper end edge of the inner peripheral surface of the glass tube G. The outer peripheral portion 106 is located at the upper end edge of the outer peripheral surface of the glass tube G.

[0052] (Inspection method) Next, with reference to FIG. 5, a inspection method executed by the inspection apparatus 10 will be described. The inspection method is executed by the control unit 52. The inspection apparatus 10 is executed for each of a plurality of glass tubes G stored in a storage container. In S12, as shown in FIG. 2, the control unit 52 causes the imaging device 12 to image the glass tube G irradiated by backlight illumination using the light source 20. The imaging device 12 supplies the control unit 52 with image data generated by imaging the glass tube G. In a modification, the control unit 52 may generate the image data using an electrical signal acquired from the light receiving element of the imaging device 12. The same applies to the processes of S14 and S16 below. The image data acquired in S12 represents an image composed of the number of pixels of x pixels × y pixels.

[0053] In FIG. 6, an image captured by backlight illumination in S12 is schematically shown. Note that in FIGS. 6 to 8, for ease of viewing, the inner peripheral portion 104 and the outer peripheral portion 106 are shown thicker than the actual dimensions with respect to the flat portion 102 and the R-shaped portions 108 and 110. Also, in S12, one or more glass tubes G and the mesh of the bottom plate 14 located around the glass tube G to be imaged are imaged. In FIGS. 6 to 8, only the glass tube G to be imaged is shown. In backlight illumination, an image is generated in which the inner peripheral portion 104 and the outer peripheral portion  106 have high (i.e., white) luminance and the flat portion 102 and the R-shaped portions 108 and 110 have low (i.e., black) luminance. The portions colored gray in FIG. 6, i.e., the flat portion 102 and the R-shaped portions 108 and 110, actually appear with a greater contrast (i.e., blacker) than the inner peripheral portion 104 and the outer peripheral portion 106.

[0054] As shown in FIG. 9, in the image IM of the glass tube G, x×y pixels are arranged side by side. The pixel values X1, ···, X9, ··· of each pixel included in the image IM1 captured by backlight illumination include values representing the luminance at each pixel. The pixel values of the pixels representing the inner peripheral portion 104 and the outer peripheral portion 106 are larger than the pixel values of the pixels representing the flat portion 102 and the R-shaped portions 108 and 110.

[0055] In S14, as shown in FIG. 3, the control unit 52 switches the light source 20 to the light source 30. Using the light source 30, the control unit 52 causes the imaging device 12 to image the glass tube G irradiated by coaxial illumination. The imaging device 12 supplies the control unit 52 with image data generated by imaging the glass tube G. The image data acquired in S14 is composed of the number of pixels of x pixels × y pixels, similar to the image data acquired in S12, and represents an image having the same resolution as the image represented by the image data acquired in S12.

[0056] In FIG. 7, an image captured by coaxial illumination in S14 is schematically shown. In coaxial illumination, an image is generated in which the luminance of the flat portion 102 is high (i.e., white), and the luminances of the inner peripheral portion 104, the outer peripheral portion 106, and the R-shaped portions 108, 110 are low (i.e., black). The portions colored in gray in FIG. 7, i.e., the inner peripheral portion 104, the outer peripheral portion 106, and the R-shaped portions 108, 110, actually appear with a greater contrast (i.e., blacker) than the flat portion 102.

[0057] As shown in FIG. 9, the pixel values Y1, ···, Y9, ··· of each pixel included in the image IM2 captured by coaxial illumination include values representing the luminance at each pixel. The pixel value of the pixel representing the flat portion 102 is greater than the pixel values of the pixels representing the inner peripheral portion 104, the outer peripheral portion 106, and the R-shaped portions 108, 110.

[0058] In S16, as shown in FIG. 4, the control unit 52 switches the light source 30 to the light source 40. Using the light source 40, the control unit 52 causes the imaging device 12 to image the glass tube G irradiated by ring illumination. The imaging device 12 supplies the control unit 52 with image data generated by imaging the glass tube G. The image data acquired in S16 is composed of the number of pixels of x pixels × y pixels, similar to the image data acquired in S12, and represents an image having the same resolution as the image represented by the image data acquired in S12.

[0059] In FIG. 8, an image captured by ring illumination in S16 is schematically shown. In ring illumination, an image is generated in which the luminances of the R-shaped portions 108 and 110 are high (i.e., white), and the luminances of the flat portion 102, the inner peripheral portion 104, and the outer peripheral portion 106 are low (i.e., black). The portions colored gray in FIG. 8, i.e., the flat portion 102, the inner peripheral portion 104, and the outer peripheral portion 106, actually appear with a greater contrast (i.e., blacker) than the R-shaped portions 108 and 110.

[0060] As shown in FIG. 9, the pixel values Z1, ···, Z9, ··· of each pixel included in the image IM3 captured by ring illumination include values representing the luminance at each pixel. The pixel values of the pixels representing the R-shaped portions 108 and 110 are greater than the pixel values of the pixels representing the flat portion 102, the inner peripheral portion 104, and the outer peripheral portion 106.

[0061] Next, in S18, the image data acquired in S12, S14, and S16 is combined to generate composite image data. As shown in FIG. 9, the image data acquired in S12, S14, and S16 each represent an image IM having the same number of pixels and the same resolution. Since the illumination is different in S12, S14, and S16, the pixel values of each pixel included in the image IM are different. Specifically, for each of the arbitrary pixels P1 to P9 included in the image IM, as shown in the image IM1 represented by the image data acquired in S12, it includes the pixel values X1 to X9, and as shown in the image IM2 represented by the image data acquired in S14, it includes the pixel values Y1 to Y9, and as shown in the image IM3 represented by the image data acquired in S16, it includes the pixel values Z1 to Z9. Each of the pixels P1 of the images IM1, IM2, and IM3 represents the same region. The same applies to the pixels P2 to P9 of the images IM1, IM2, and IM3.

[0062] The control unit 52 generates a composite pixel P1 having pixel values (X1, Y1, Z1) using the pixel value of the pixel P1 included in the image represented by the image data acquired in S12, S14, and S16. Similarly for pixels P2 to P9, composite pixels having the pixel values of the images IM1, IM2, and IM3 are generated. In this way, the control unit 52 generates, for each pixel of the image IM, a composite pixel having the pixel value of each pixel of the image represented by the image data acquired in S12, S14, and S16, thereby generating composite image data representing an image CIM having the same number of pixels and resolution as the image represented by the image data acquired in S12, S14, and S16.

[0063] In S20, the control unit 52 determines whether the imaged glass tube G is normal. Specifically, the control unit 52 inputs the composite image data generated in S18 to the learning model 60 to determine whether the glass tube G contains defects. Defects in the glass tube G include poor firing, adhesion of glass powder, inclusion of foreign matter, chipping, cracks, and the like.

[0064] The learning program 64 generates a learning model 60 by deep learning using, as learning data, composite image data obtained by imaging a glass tube G without defects with the imaging device 12. By using deep learning, the three pixel values of each pixel of the composite image data can be processed simultaneously. In the learning program 64, the composite image data of an image representing a glass tube G having a defect is not used. Therefore, it is not necessary to prepare a glass tube G having a defect deliberately. Also, for the generation of the learning model 60, the image data generated from S12 to S16 is used. As a result, it is not necessary to image the glass tube G to generate the learning data. Note that the control unit 52 may further update the learning model 60 using the composite image data obtained by the inspection method. In the generated learning model 60, a non-defective product learning score (i.e., a threshold for pass / fail determination) is calculated. In S20, in the learning model 60, an anomaly score is calculated based on the degree of similarity between the image represented by the composite image data generated in S18 and the image of the glass tube G without the learned defect. The control unit 52 determines that it is abnormal when the anomaly score is equal to or higher than the non-defective product learning score, and determines that it is normal when the anomaly score is less than the non-defective product learning score.

[0065] In S20, further, the control unit 52 determines whether the imaged glass tube G is normal or not using the image data generated in S12, S14, and S16. Specifically, in the storage unit 54, for each of the image captured by backlight illumination, the image captured by coaxial illumination, and the image captured by ring illumination, an algorithm representing a rule for determining whether the glass tube G is abnormal or not, that is, abnormality detection using a rule-based method, is used to determine whether the glass tube G is normal or not. For example, in the image represented by the image data generated in S14, when the luminance of the pixel representing the flat portion 102 is equal to or lower than a predetermined threshold value (that is, when it is captured darkly), a rule is set to determine that it is not normal. In S20, when it is determined that it is not normal in at least one of the learning model 60 and the abnormality detection using a rule-based method, it is determined that the glass tube G is not normal. Note that in a modified example, the control unit 52 may not perform a normality determination using the image data generated in S12, S14, and S16 other than the composite image data. <The original text of is not provided, so it cannot be translated.> <The original text of

[0066] is not provided, so it cannot be translated.> <The original text of is not provided, so it cannot be translated.> The control unit 52 generates and updates the learning model 60 using the composite image data generated in S18 as learning data. Thereby, in order to prepare the learning data, it is not necessary to separately prepare a learning glass tube and perform imaging. <The original text of is not provided, so it cannot be translated.> <The original text of

[0067] is not provided, so it cannot be translated.> <The original text of is not provided, so it cannot be translated.> In S22, the control unit 52 determines whether an output instruction for displaying the composite image CIM is acquired from the user or not. The user can input an output instruction (for example, an operation of a display button) for causing the output unit 56 to display the composite image CIM by operating an operation unit (not shown) of the inspection apparatus 10. When the output instruction is input, the output instruction is acquired (YES in S22), and the process proceeds to S24. When the output instruction is not input, the output instruction is not acquired (NO in S22), and the inspection method ends. <The original text of is not provided, so it cannot be translated.> <The original text of

[0068] is not provided, so it cannot be translated.> <The original text of is not provided, so it cannot be translated.>In S24, the composite image CIM can be displayed on the output unit 56. The control unit 52 defines the value of each pixel (X1 to X9 in FIG. 9) included in the image data generated in S12 as the value of the first color (for example, R) among R (red), G (green), and B (blue). Similarly, the control unit 52 defines the value of each pixel (Y1 to Y9 in FIG. 9) included in the image data generated in S14 as the value of the second color (for example, G) among R (red), G (green), and B (blue), and defines the value of each pixel (Z1 to Z9 in FIG. 9) included in the image data generated in S16 as the value of the third color (for example, B) among R (red), G (green), and B (blue). As a result, an image represented in RGB color is displayed on the output unit 56. As a result, the user can easily check the state of the glass tube G by checking the displayed color image.

[0069] (Method for manufacturing a glass tube) Next, with reference to FIG. 10, a method for manufacturing the glass tube G using the above inspection method will be described. In the manufacturing method, in S32, a forming process is executed. In the forming process, a long glass tube member is formed from molten glass. In the forming process, for example, the Danner method is used to form the tube member. In the Danner method, a cylindrical sleeve is rotated and molten glass is supplied to the sleeve surface. The molten glass is formed into a tubular shape while wrapping around the sleeve. A long tube member is formed by pulling the tubular molten glass from the sleeve. Note that in the forming process, a method other than the Danner method, such as the redraw method or the down-draw method, may be used to form the tube member.

[0070] In S34, a cutting process for cutting the tube member that has been formed in S32 is executed. In the cutting process, using a cutting device such as a cutter, the tube member is cut to the length of the glass tube G. As a result, a tubular intermediate tube member having a length substantially equal to that of the glass tube G is formed. Next, in S36, an end-burning process for end-burning each of the axial ends of the intermediate tube member formed in S34 is executed. In the end-burning process, by heating each of the axial ends of the intermediate tube member, fine cracks formed in the cutting process are removed. Note that in the end-burning process, one end of the intermediate tube member may be heated. As a result, on the end face of the glass tube G, R-shaped portions 108 and 110 are formed on both radial sides of the flat portion 102.

[0071] In S38, a detection process using the inspection method shown in FIG. 5 is executed by housing a plurality of glass tubes G that have been end-burned in S36 in a container. The glass tubes G determined to be abnormal in the detection process are removed from the container, and the normal glass tubes G are conveyed as products.

[0072] (Effect) In the inspection device 10, three pieces of image data are generated by imaging one glass tube G using three different types of illumination. In the composite image data obtained by combining the three pieces of image data, each pixel includes the pixel values included in each pixel of the three pieces of image data. The pixel value of each pixel in the composite image data has three values. Thereby, the amount of information in the image data used when inspecting whether the glass tube G is normal can be increased. Thereby, the accuracy of the inspection can be improved.

[0073] The imaging unit 11 includes a backlight illumination light source 20, a coaxial illumination light source 30, and a ring illumination light source 40. In the inspection method, the glass tube G irradiated by each of the backlight illumination, coaxial illumination, and ring illumination is imaged. Depending on various illuminations, image data with different brightness levels of the flat portion 102, inner peripheral portion 104, outer peripheral portion 106, and R-shaped portions 108, 110 at the edge of the glass tube G can be generated. Depending on the type and location of defects such as mouth burning defects, adhesion of glass powder, foreign matter inclusion, chipping, and cracks, it may be difficult to detect them using an image captured with one type of illumination. In the inspection apparatus 10, by using a plurality of types of illumination, it is possible to make it easier to detect defects that are difficult to detect with one type of illumination using other types.

[0074] Also, by generating composite image data, it is possible to make defects that are difficult to detect using only an image obtained from one type of illumination stand out.

[0075] In the inspection apparatus 10, an inspection is performed to detect a non-normal glass tube G using the learning model 60. In anomaly detection using the learning model 60, the calculation accuracy of the anomaly score can be improved by increasing the amount of information in the image data.

[0076] The imaging unit 11 includes one imaging device 12 for the three types of light sources 20, 30, and 40. This eliminates the need to increase the number of imaging devices 12.

[0077] Each of the light sources 20, 30, and 40 is an example of each of the "first light source", "second light source", and "third light source". Each of the image data generated by S12, S14, and S16 is an example of each of the "first image data", "second image data", and "third image data".

[0078] As described above, specific examples of the technology disclosed in this specification have been explained, but these are merely examples and do not limit the scope of the claims. The technology described in the claims includes various modifications and changes of the specific examples illustrated above. For example, the following modification examples may be adopted.

[0079] (1) The inspection method and inspection apparatus 10 of this embodiment can be used for inspecting glass articles such as glass plates in addition to the glass tube G.

[0080] (2) The inspection apparatus 10 may not include any one of the light sources 20, 30, and 40. In this case, in the inspection method, it may be determined whether it is normal using the image data generated by imaging using two types of light sources.

[0081] (3) The inspection apparatus 10 may not include the imaging unit 11. In this case, the inspection apparatus 10 may acquire image data from an external device.

[0082] The technical elements described in this specification or the drawings exhibit technical utility alone or in various combinations, and are not limited to the combinations described in the claims at the time of filing. In addition, the technology exemplified in this specification or the drawings achieves a plurality of purposes simultaneously, and achieving one of those purposes itself has technical utility.

Description of Reference Numerals

[0083] 10: Inspection apparatus, 11: Imaging unit, 12: Imaging device, 20, 30, 40: Light sources, 50: Control device, 52: Control unit, 54: Storage unit, 56: Output unit, 60: Learning model, 62: Model parameters, 64: Learning program, 102: Flat part, 104: Inner peripheral part, 106: Outer peripheral part, 108, 110: R-shaped parts, G: Glass tube

Claims

1. A method for inspecting a glass article, comprising: a first acquisition step of acquiring first image data representing a first image obtained by imaging the glass article irradiated by a first light source; a second acquisition step of acquiring second image data representing a second image obtained by imaging the glass article irradiated by a second light source different from the first light source; a detection step of detecting that the glass article is abnormal using composite image data obtained by using the acquired first image data and the second image data, wherein a composite pixel among a plurality of pixels included in the composite image data includes a first pixel value of a first pixel among a plurality of pixels included in the first image data and a second pixel value of a second pixel among a plurality of pixels included in the second image data, and a position of the first pixel in the first image and a position of the second pixel in the second image overlap each other.

2. The inspection method according to claim 1, wherein in the detection step, it is detected that the glass article is abnormal using the composite image data by a learning model generated by machine learning using learning data.

3. The learning model is generated using composite learning image data obtained by using first learning image data representing a first learning image obtained by imaging a specific glass article irradiated by the first light source and second learning image data representing a second learning image obtained by imaging the specific glass article irradiated by the second light source, wherein a composite learning pixel among a plurality of pixels included in the composite learning image data includes a first learning pixel value of a first learning pixel among a plurality of pixels included in the first learning image data and a second learning pixel value of a second learning pixel among a plurality of pixels included in the second learning image data, and a position of the first learning pixel in the first learning image and a position of the second learning pixel in the second learning image overlap each other.

4. The inspection method according to claim 3, wherein the specific glass article is a normal glass article.

5. The first light source is one of backlight illumination, coaxial illumination, and ring illumination, and the second light source is one of backlight illumination, coaxial illumination, and ring illumination different from the first light source.

6. The inspection method according to any one of claims 1 to 4, wherein the glass article is a glass tube.

7. The first light source is one of backlight illumination, coaxial illumination, and ring illumination, the second light source is one of backlight illumination, coaxial illumination, and ring illumination different from the first light source, in the image data obtained by imaging the glass tube irradiated by the backlight illumination, the pixel value of the pixel representing the outer peripheral portion or the inner peripheral portion of the end face of the glass tube is larger than the pixel value of the pixel representing other portions of the glass tube, in the image data obtained by imaging the glass tube irradiated by the coaxial illumination, the pixel value of the pixel representing the flat portion of the end face of the glass tube is larger than the pixel value of the pixel representing other portions of the glass tube, in the image data obtained by imaging the glass tube irradiated by the ring illumination, the pixel value of the pixel representing the R-shaped portion formed between the flat portion of the end face of the glass tube and the outer peripheral portion or the inner peripheral portion of the end face is larger than the pixel value of the pixel representing other portions of the glass tube, the inspection method according to claim 6.

8. further comprising an output step of outputting a composite image represented by the composite image data, the composite pixel includes the first pixel value assigned with the first color and the second pixel value assigned with the second color, the inspection method according to any one of claims 1 to 4.

9. further comprising a third acquisition step of acquiring third image data representing a third image obtained by imaging the glass article irradiated by a third light source different from the first light source and the second light source, in the detection step, using the composite image data obtained by further using the acquired third image data, it is detected that the glass article is abnormal, the composite pixel further includes a third pixel value of a third pixel among a plurality of pixels included in the third image data, the position of the first pixel in the first image, the position of the second pixel in the second image, and the position of the third pixel in the third image overlap with each other, the inspection method according to claim 1.

10. in the detection step, using the composite image data, it is detected that the glass article is abnormal by a learning model generated by machine learning using learning data, the inspection method according to claim 9.

11. The learning model is generated using composite learning image data obtained using first learning image data representing a first learning image in which a specific glass article irradiated by the first light source is imaged, second learning image data representing a second learning image in which the specific glass article irradiated by the second light source is imaged, and third learning image data representing a third learning image in which the specific glass article irradiated by the third light source is imaged. Among the plurality of pixels included in the composite learning image data, the composite learning pixels include a first learning pixel value of a first learning pixel among the plurality of pixels included in the first learning image data, a second learning pixel value of a second learning pixel among the plurality of pixels included in the second learning image data, and a third learning pixel value of a third learning pixel among the plurality of pixels included in the third learning image data. The method of inspection according to claim 10, wherein the position of the first learning pixel in the first learning image, the position of the second learning pixel in the second learning image, and the position of the third learning pixel in the third learning image overlap with each other.

12. The first light source is one of backlight illumination, coaxial illumination, and ring illumination. The second light source is one of backlight illumination, coaxial illumination, and ring illumination that is different from the first light source. The inspection method according to any one of claims 9 to 11, wherein the third light source is one of backlight illumination, coaxial illumination, and ring illumination that is different from the first light source and the second light source.

13. The inspection method further includes an output step of outputting a composite image represented by the composite image data. The inspection method according to any one of claims 9 to 11, wherein the composite pixel includes the first pixel value assigned a first color, the second pixel value assigned a second color, and the third pixel value assigned a third color.

14. An inspection apparatus for a glass article, comprising: a first imaging unit including a first light source and a first imaging device that images the glass article irradiated by the first light source to generate first image data representing a first image; a second imaging unit including a second light source different from the first light source and a second imaging device that images the glass article irradiated by the second light source to generate second image data representing a second image. A detection unit that detects that the glass article is abnormal using the composite image data obtained by using the generated first image data and the second image data. Among the plurality of pixels included in the composite image data, the composite pixel includes the first pixel value of the first pixel among the plurality of pixels included in the first image data and the second pixel value of the second pixel among the plurality of pixels included in the second image data. An inspection apparatus, wherein the position of the first pixel in the first image and the position of the second pixel in the second image overlap each other.

15. The inspection apparatus according to claim 14, wherein the first imaging device and the second imaging device are the same imaging device.

16. Further comprising a third imaging unit including a third light source different from the first light source and the second light source, and a third imaging device that generates third image data representing a third image by imaging the glass article irradiated by the third light source. The detection unit detects that the glass article is abnormal using the composite image data obtained by using the generated first image data, the second image data, and the third image data. The composite pixel further includes the third pixel value of the third pixel among the plurality of pixels included in the third image data. The inspection apparatus according to claim 14 or 15, wherein the position of the first pixel in the first image, the position of the second pixel in the second image, and the position of the third pixel in the third image overlap each other.

17. The inspection apparatus according to claim 16, wherein the first imaging device, the second imaging device, and the third imaging device are the same imaging device.

18. A method for manufacturing a glass tube, comprising: A forming step of forming a tube member longer than the glass tube from molten glass; A cutting step of forming a tube intermediate member by cutting the tube member; A mouth-blowing step of creating a glass tube by mouth-blowing the end of the tube intermediate member; A detection step of detecting an abnormal glass tube by the inspection method according to claim 6.

Images