Inspection device, learning data generation method, learning model generation method, and method for manufacturing object

Abstract

Provided is a technique capable of identifying the type of a defect using a learning model. The inspection device includes: a specifying unit that specifies a defect position of a predetermined defect in a learning object using wide-area learning image data representing the learning object; a microscopic image generation unit that generates microscopic learning image data indicating a predetermined defect by performing microscopic imaging on the identified defect position; and an acquisition unit that acquires a combination of type information indicating a type of a prescribed defect obtained using the generated microlearning image data, and defect image data indicating a prescribed defect included in the wide-area learning image data, the type information indicating the type of the prescribed defect obtained using the microlearning image data that has been generated, and the defect image data indicating the type of the prescribed defect included in the wide-area learning image data. The specifying unit specifies the type of the defect in the object to be inspected using wide-area inspection image data representing the object to be inspected using a learning model generated by machine learning using the combination of the acquired type information and the defect image data as learning data.

Description

Technical Field

[0001] The technology disclosed in this specification relates to abnormalities of an object to be inspected. Background Art

[0002] A technology for detecting defects in products is disclosed in Patent Document 1. In this technology, image data indicating a qualified product or a defective product is used as teacher data, and a learning model is generated by machine learning such as a neural network. Using the learning model, qualified products and defective products are classified based on the image data of the products.

[0003] Prior Art Documents

[0004] Patent Documents

[0005] Patent Document 1: International Publication No. 2019 / 230356 Summary of the Invention

[0006] Problems to be Solved by the Invention

[0007] When the object to be inspected contains a defect, it is sometimes necessary to determine the type of defect indicating which defect it is.

[0008] In this specification, a technology is provided that can determine the type of defect using a learning model.

[0009] Means for Solving the Problems

[0010] The first aspect disclosed in this specification relates to an inspection device. The inspection device includes: a determination unit that uses wide-area learning image data indicating a learning object to determine the defect position of a specified defect in the learning object; a microscopic image generation unit that generates microscopic learning image data indicating the specified defect by performing microscopic photography on the determined defect position; and an acquisition unit that acquires a combination of type information and defect image data, the type information indicating the type of the specified defect obtained by using the generated microscopic learning image data, the defect image data indicating the specified defect included in the wide-area learning image data, the determination unit uses a learning model and uses wide-area inspection image data indicating an object to be inspected to determine the type of defect in the object to be inspected, and the learning model is generated by machine learning using the combination of the acquired type information and the defect image data as learning data.

[0011] According to this structure, by microscopically photographing a specified defect for which the defect position has been determined, microscopic learning image data magnifying the specified defect can be generated. By using the microscopic learning image data, even small defects can be easily classified. By generating a learning model using the combination of the defect type obtained using the microscopic learning image data and the defect image data, even when the size of the defect is small when using the learned learning model, the defect type can be determined.

[0012] In the second mode, on the basis of the above first mode, it may also be that the object to be learned includes a glass plate, and the inspection device further includes an inspection image generation unit that generates the wide-field learning image data. The inspection image generation unit includes: a bright-field wide-field photographing unit that generates bright-field wide-field image data representing the glass plate by photographing first transmitted light that has passed through the glass plate in a field of view including a bright field; and a dark-field wide-field photographing unit that generates dark-field wide-field image data representing the glass plate by photographing second transmitted light that is incident on the glass plate at an angle different from the first transmitted light and has passed through the glass plate in a field of view including a dark field. The wide-field learning image data includes the bright-field wide-field image data and the dark-field wide-field image data.

[0013] Defects such as small surface irregularities, bubbles, and foreign substances may occur on the glass plate. Depending on the type of these defects, the appearance of the image of the defect when photographed in a field of view including a bright field and in a field of view including a dark field is different. According to the above structure, by using the bright-field wide-field image data photographed in a field of view including a bright field and the dark-field wide-field image data photographed in a field of view including a dark field, the type of the defect can be easily determined.

[0014] In the third mode, on the basis of the above first or second mode, it may also be that the object to be learned includes a glass plate, and the microscopic image generation unit includes: a bright-field microscopic photographing unit that generates bright-field microscopic image data representing the glass plate by photographing reflected light reflected by the glass plate in a field of view including a bright field; and a dark-field microscopic photographing unit that generates dark-field microscopic image data representing the glass plate by photographing scattered light scattered by the defect in a field of view including a dark field. The microscopic learning image data includes the bright-field microscopic image data and the dark-field microscopic image data.

[0015] According to the above structure, by using the bright-field microscopic image data photographed in a field of view including a bright field and the dark-field microscopic image data photographed in a field of view including a dark field, the type of the defect can be easily determined.

[0016] In the fourth mode, based on any one of the above first to third modes, it is also possible that the microscopic image generation unit includes: a camera unit that microscopically photographs the specified defect; and a moving unit that moves the camera unit to the defect position. When the learning object includes a plurality of defects, the microscopic learning image data of the specified defects having a number smaller than the number of the plurality of defects is generated, and the plurality of defects include one or more of the specified defects.

[0017] According to the above structure, by moving the camera unit, it is possible to appropriately photograph the specified defect. Furthermore, not all of the plurality of defects included in the object need to be used as learning data. As a result, the time required to generate the microscopic learning image data can be shortened, and the burden of generating the learning data and the learning period in the learning model can be suppressed.

[0018] In the fifth mode, based on the above fourth mode, it is also possible that the microscopic image generation unit generates the microscopic learning image data of the specified defects in ascending order of size from the specified defects.

[0019] It is more difficult to determine the type of a defect with a relatively small size compared to a defect with a relatively large size. According to the above structure, by selectively using the image data representing the defects with a relatively small size as learning data, the accuracy of determining the type of the defects with a relatively small size can be improved.

[0020] In the sixth mode, based on any one of the above first to fifth modes, it is also possible that the learning object and the inspection object include a glass plate for a display device.

[0021] The smaller the size of the pixels displayed by the display device, the smaller the size of the defects allowed in the glass plate for the display device. The above inspection device can be applied to determine the type of small defects in the glass plate for the display device.

[0022] In the seventh mode, based on the above sixth mode, it is also possible that the wide-area inspection image data includes the entire effective surface of the glass plate.

[0023] In the glass plate for the display device, in a subsequent process, it is sometimes used as a product after cutting the peripheral portion. The effective surface refers to the surface used as a product. According to this structure, it is possible to comprehensively determine the defects included in the effective surface of the glass plate.

[0024] This specification also discloses a learning data generation method for learning model generation. The learning data generation method includes: a wide-area learning image acquisition step of acquiring wide-area learning image data representing an object; a learning defect position acquisition step of acquiring the defect position of a specified defect of the object in the object; a microscopic learning image generation step of generating microscopic learning image data representing the specified defect by microscopically photographing the acquired defect position; and a storage step of storing a combination of type information and defect image data, where the type information represents the type of the specified defect obtained by using the generated microscopic learning image data, and the defect image data represents the specified defect included in the wide-area learning image data.

[0025] According to this configuration, by microscopically photographing a specified defect for which the defect position has been determined, microscopic learning image data magnifying the specified defect can be generated. By using the microscopic learning image data, even small defects can be easily determined in terms of their type. A combination of the type of defect obtained using the microscopic learning image data and the defect image data can be generated as learning data. Thus, the type of defect more accurately determined using the microscopic image magnifying the specified defect can be used as learning data. As a result, the determination accuracy of the learning model for the type of defect can be improved.

[0026] This specification also discloses a learning model generation method. Alternatively, the learning model generation method may include: a wide-area learning image acquisition step of acquiring wide-area learning image data representing an object; a learning defect position acquisition step of acquiring the defect position of a specified defect of the object in the object; a microscopic learning image generation step of generating microscopic learning image data representing the specified defect by microscopically photographing the acquired defect position; a storage step of storing a combination of type information and defect image data, where the type information represents the type of the specified defect obtained by using the generated microscopic learning image data, and the defect image data represents the specified defect included in the wide-area learning image data; and a learning model generation step of generating a learning model by machine learning using the stored combination as learning data.

[0027] According to this configuration, a learning model can be generated using highly accurate learning data generated by using microscopic learning image data magnifying a specified defect.

[0028] This specification also discloses a method for manufacturing an object. The method for manufacturing an object includes: a learning defect type determination step of determining the type of a specified defect in the learning object using wide-area learning image data representing the learning object; a position acquisition step of acquiring the defect position of the specified defect in the learning object using the wide-area learning image data; a microscopic learning image generation step of generating microscopic learning image data representing the specified defect by microscopically photographing the acquired defect position; and a storage step of storing a combination of type information and a defect image, where the type information represents the type of the specified defect obtained using the generated microscopic learning image data, and the defect image represents the specified defect included in the wide-area learning image data. The method for manufacturing an object further includes an inspection object defect type determination step in which, using a learning model and wide-area inspection image data representing the inspection object, the type of a defect in the inspection object is determined, and the learning model is generated by machine learning using the combination of the stored type information and the defect image data as learning data.

[0029] According to this structure, the same effect as the above-described inspection device can be achieved.

[0030] Alternatively, the manufacturing method may further include a quality determination step in which, based on defect information, the quality of the object is determined, and the defect information includes at least one of the number of defects included in the object, the size of the defects, and the type of the defects determined in the inspection object defect type determination step.

[0031] According to this structure, the quality of the object can be determined based on defect information corresponding to the required quality of the object.

[0032] Alternatively, the object includes a glass plate for a display device, and the manufacturing method may further include a removal step in which, based on the quality determined in the quality determination step, the objects are classified according to each quality.

[0033] According to this structure, by classifying the glass plates according to each quality, when the glass plates are shipped from the factory, glass plates of a quality consistent with the required quality can be shipped to the destination.

[0034] Alternatively, the manufacturing method may further include a non-conforming product classification step in which non-conforming products determined to be of non-conforming quality in the quality determination step are distinguished from products not determined to be of non-conforming quality.

[0035] According to this structure, it is possible to prevent defective products from being included in the objects produced at the factory. Description of the Drawings

[0036] Figure 1 is a block diagram of the manufacturing unit.

[0037] Figure 2 is a schematic diagram showing an overview of the inspection device, the imaging device, and the transfer device.

[0038] Figure 3 is a schematic diagram for explaining the arrangement of the camera and the lighting of the inspection device.

[0039] Figure 4 is a schematic diagram for explaining the arrangement of the camera and the lighting of the imaging device.

[0040] Figure 5 is a flowchart showing the steps of the manufacturing method.

[0041] Figure 6 is a flowchart showing the steps of the inspection process.

[0042] Figure 7 shows an example of an image representing a defect captured by the inspection device.

[0043] Figure 8 is a diagram showing a list of data transmitted after the defective position acquisition process for learning.

[0044] Figure 9 shows an example of an image representing a defect captured by the imaging device.

[0045] Figure 10 is a diagram showing a list of data transmitted after the microscopic learning image generation process.

[0046] Figure 11 is a diagram showing a list of data transmitted in the storage process.

[0047] Figure 12 is a schematic diagram showing a library of learning data.

[0048] Figure 13 is a flowchart showing the steps of the manufacturing method in other embodiments.

[0049] Description of Reference Numerals:

[0050] 10: Manufacturing unit; 12: Shaping device; 14: Cooling device; 16: Cutting device; 18: Inspection device; 19: Detection device; 20: Microscopic image generation device; 22: Handling device; 30: Control device; 40: Wide-field shooting unit; 44: Bright-field wide-field shooting system; 46: Dark-field wide-field shooting system; 50: Control device; 60: Microscopic shooting unit; 61: Moving unit; 62: Moving unit; 63: Camera unit; 64: Bright-field microscopic shooting system; 65: Dark-field microscopic shooting system; 70: Acquisition terminal; G: Glass plate; G1: Glass plate for learning; G2: Glass plate to be inspected; L1: Transmitted light; L2: Transmitted light; L3: Transmitted light; L4: Reflected light; L5: Scattered light; P1: Wide-field learning image data; P2: Microscopic learning image data; P3: Wide-field inspection image data; P4: Bright-field wide-field image data; P5: Dark-field wide-field image data; P6: Bright-field microscopic image data; P7: Dark-field microscopic image data. Detailed implementation mode

[0051] (Structure of the manufacturing unit)

[0052] As Figure 1 shown, the manufacturing unit 10 manufactures a glass plate G (refer to Figure 2 ). The glass plate G is used, for example, for the cover and substrate of a display device. It should be noted that the use of the glass plate G is not particularly limited. The glass plate G includes a glass plate G1 for learning and a glass plate G2 to be inspected. It should be noted that the glass plate G1 for learning is an example of a "learning object", and the glass plate G2 to be inspected is an example of an "object to be inspected".

[0053] The manufacturing unit 10 includes a shaping device 12, a cooling device 14, a cutting device 16, an inspection device 18, and a handling device 22. The shaping device 12 extends the molten glass downward to form a strip-shaped glass ribbon. The shaping device 12 uses, for example, the overflow down-draw method to form a glass ribbon from the molten glass melted in a melting furnace. It should be noted that the shaping device 12 is not particularly limited, and other down-draw methods such as the slot down-draw method and the redraw method, or the float method can be used to produce the glass ribbon.

[0054] The cooling device 14 anneals the glass ribbon extending from the shaping device 12. The cooling device 14 includes an annealing furnace with a specified temperature gradient set downward in the internal space. The glass ribbon is annealed while being guided by a handling device such as an annealing roller, and the temperature decreases as it moves downward in the internal space of the annealing furnace. Thereby, the strain of the glass ribbon is reduced. The cooling device 14 also includes a cooling chamber for cooling the annealed glass ribbon to near room temperature.

[0055] The cutting device 16 cuts the glass ribbon into a specified length. The cutting device 16 clamps both end portions of the glass ribbon, forms a scoring line using a cutter, applies a bending stress along the scoring line, and thereby cuts (i.e., severs) the glass ribbon along the scoring line. Thus, a rectangular glass plate G of a specified length is obtained from the glass ribbon. It should be noted that the cutting method used in the cutting device 16 is not limited to severing based on bending stress, and for example, it may also be laser severing, laser melting, etc. In addition, the cutting device 16 may further cut both end portions in the width direction of the glass plate G. The thickness of both end portions in the width direction of the glass plate G sometimes becomes relatively larger compared to the central portion in the width direction, and these end portions are called ears.

[0056] The glass plate G cut by the cutting device 16 is held by the transfer device 22. The glass plate G is transferred to the inspection device 18 while being held by the transfer device 22. As Figure 2 shown, the transfer device 22 includes clamping mechanisms 24 and 25 that respectively clamp the upper end edge and the lower end edge of the glass plate G, and a pair of rails 26, 26 that extend in the transfer direction. The clamping mechanism 24 includes a plurality of clamps 24a that clamp the upper end edge of the glass plate G, and a moving body 24b that movably mounts the plurality of clamps 24a on the rail 26. The clamping mechanism 25 includes a plurality of clamps 25a that clamp the lower end edge of the glass plate G, and a moving body 25b that movably mounts the plurality of clamps 24a on the rail 26. The clamping mechanisms 24 and 25 transfer the glass plate G along the rails 26, 26 while clamping the glass plate G. By performing inspection in a state where the upper end edge and the lower end edge of the glass plate G are clamped, the shaking of the glass plate G can be reduced. Thus, the shift of the focus when the inspection device 18 photographs the glass plate G can be reduced, and the type of defect can be correctly determined.

[0057] (Structure of the inspection device)

[0058] The inspection device 18 includes a detection device 19 and a microscopic image generation device 20. The detection device 19 detects defects present in the glass plate G, and determines the defect positions and defect types. The microscopic image generation device 20 selects a specified number (for example, 2 to 3) of defects from the defects detected by the detection device 19, and photographs an image of the selected defects (hereinafter referred to as "specified defects") magnified. Defects include bubbles and foreign substances encapsulated in the glass plate G during the manufacture of the glass plate G, attachments such as dust adhering to the surface of the glass plate G, damage formed on the surface of the glass plate G, and unevenness. It should be noted that the detection device 19 is an example of a "determination unit", and the microscopic image generation device 20 is an example of a "microscopic image generation unit".

[0059] (Structure of the detection device)

[0060] The detection device 19 includes a control device 30 and a wide-area imaging unit 40. The control device 30 includes a control unit 32 and a communication interface (hereinafter referred to as "communication I / F") 36. The control unit 32 controls the detection device 19. The control unit 32 includes a memory composed of a CPU and a non-volatile memory, etc. The CPU executes the processes described below according to the computer program stored in the memory. Note that the wide-area imaging unit 40 is an example of an "inspection image generation unit", and the image data of the image captured by the wide-area imaging unit 40 is an example of "wide-area learning image data P1" and "wide-area inspection image data P3".

[0061] In the memory, in addition to the computer program, a learning program and a learning model 34 are also stored. The learning model 34 is a model (i.e., a mathematical formula) of a multi-layer neural network. The multi-layer neural network is a model of so-called deep learning, such as convolutional type, fully-connected type, etc. The learning model 34 includes the values of the respective weights in the intermediate layer of the neural network, i.e., learning parameters. The multi-layer neural network is a known technology, and detailed description thereof is omitted here. The learning model 34 performs machine learning using learning data that combines the image data representing the defects of the learning glass plate G1 and the types of the defects.

[0062] The communication I / F 36 is an interface for the detection device 19 to perform communication with the microscopic image generation device 20 via a wired or wireless communication network such as a LAN or Wi-Fi (registered trademark).

[0063] The control device 30 controls the wide-area imaging unit 40. The wide-area imaging unit 40 includes a line camera unit 42 that arranges multiple cameras and light sources in the vertical direction. As Figure 3 shown, in the line camera unit 42, a combination of a bright-field wide-area imaging system 44 and a dark-field wide-area imaging system 46 is arranged and configured in multiple numbers in the vertical direction. Figure 3 is a view of a combination of one bright-field wide-area imaging system 44 and one dark-field wide-area imaging system 46 as seen from above. Note that the bright-field wide-area imaging system 44 is an example of a "bright-field wide-area imaging unit", and the dark-field wide-area imaging system 46 is an example of a "dark-field wide-area imaging unit".

[0064] The glass plate G is conveyed by a conveying device 22 in the X direction perpendicular to the vertical direction and parallel to the surface of the glass plate G. The bright-field wide-area imaging system 44 includes a camera 44a, a shielding plate 44b, and a light source 44c. The camera 44a captures the transmitted light L1 that is irradiated from the light source 44c and passes through the glass plate G. The shielding plate 44b blocks a part (e.g., half) of the transmitted light L1 to form a bright part and a dark part in the field of view of the camera 44a. Note that the transmitted light L1 is an example of "first transmitted light".

[0065] The light source 44c is arranged on the surface Ga side of the glass plate G, and the camera 44a is arranged on the surface Gb side of the glass plate G. The optical axis of the light source 44c is the direction in which light is incident perpendicularly to the surface Ga of the glass plate G. The optical axis of the camera 44a is arranged on the straight line of the optical axis of the light source 44c. Thus, the camera 44a is in a state of photographing the transmitted light L1 in bright field without the light shielding plate 44b. The camera 44a shields a part of the transmitted light L1 through the light shielding plate 44b and photographs the transmitted light L1 in semi-bright field.

[0066] The dark field wide-area photographing system 46 includes a camera 46a and light sources 46b and 46c. It should be noted that the light source 46c is arranged overlapping the light source 44c in the vertical direction. The camera 46a photographs the transmitted light L2 that is irradiated from the light source 46c and passes through the glass plate G in bright field. In addition, the camera 46a photographs the transmitted light L3 that is irradiated from the light source 46b and passes through the glass plate G in dark field. It should be noted that the transmitted light L3 is an example of the "second transmitted light".

[0067] The optical axis of the camera 46a is arranged on the straight line of the optical axis of the light source 46c separated by a beam splitter 48 to be described later so that the transmitted light L2 can be supplemented by the camera 46a. The camera 46a photographs the transmitted light L2 in bright field.

[0068] The light source 46b is arranged on the surface Ga side of the glass plate G. The optical axis of the light source 46b is arranged inclined with respect to the surface Ga of the glass plate G as the incident direction of light. The light source 46b is arranged on both sides of the light source 44c and the light source 46c respectively. The optical axis of the camera 46a is arranged at a position deviated from the straight line of the optical axis of the light source 46b so that the transmitted light L3 does not enter the camera 46a. The camera 46a photographs the transmitted light L3 in dark field. When scattering occurs in the glass plate G due to defects or the like, the transmitted light L3 is received by the camera 46a. It should be noted that for easy understanding, the inclination angle of the transmitted light L3 is exaggerated compared to the actual situation, but usually the transmitted light L3 is incident on the beam splitter 48.

[0069] In the camera 46a, the light after synthesizing the transmitted light L2 and the transmitted light L3 is photographed. That is, according to the camera 46a, an image obtained by synthesizing bright field and dark field is generated. The light sources 44c, 46b, and 46c are arranged adjacent to each other, and the positions where the transmitted lights L1, L2, and L3 pass through the glass plate G are the same. The lighting times of the light sources 44c, 46b, and 46c can be simultaneous or in an order consistent with the photographing time. The lighting times of the light sources 44c, 46b, and 46c are controlled by the control unit 32.

[0070] The beam splitter 48 is arranged on the optical axes of the cameras 44a and 46a. The shutter plate 44b is arranged between the beam splitter 48 and the camera 44a. The beam splitter 48 transmits a specific wavelength and reflects wavelengths other than the specific wavelength. The beam splitter 48 separates the transmitted light that is irradiated from the light sources 44c, 46b, 46c and passes through the glass plate G into a component including the transmitted light L1 and a component including the transmitted lights L2 and L3. Specifically, by using the light sources 44c, 46b, 46c with different wavelengths, for example, LEDs of different colors, the transmitted light L1 of the light source 44c, the transmitted light L2 of the light source 46b, and the transmitted light L3 of the light source 46c are separated by the beam splitter 48. The component including the transmitted light L1 passes through the beam splitter 48 and is captured by the camera 44a, and the component including the transmitted lights L2, L3 is reflected by the beam splitter 48 and is captured by the camera 46a. It should be noted that the light sources 44c, 46b, 46c are not limited to LEDs. For example, they can also be metal halide lamps or laser sources, etc.

[0071] Although not shown in the figure, a plurality of the light sources 44c, 46b, 46c are arranged in correspondence with the plurality of cameras 44a, 46a that constitute the line array camera unit 42. During the period when the glass plate G is transported by the transport device 22, the entire area of the effective surface of the glass plate G is captured. The effective surface of the glass plate G is the surface used when the glass plate G is used for a product, i.e., a display device. For example, in the case where the peripheral portion of the glass plate G is cut after inspection, the cut peripheral portion is not included in the effective surface, and the portion used for the product after cutting is included in the effective surface. It should be noted that the surface to be captured may also include a surface other than the effective surface. It should be noted that the image data of the image captured by the bright-field wide-field imaging system 44 is an example of the "bright-field wide-field image data P4", and the image data of the image captured by the dark-field wide-field imaging system 46 is an example of the "dark-field wide-field image data P5". Therefore, the "wide-field learning image data P1" includes the "bright-field wide-field image data P4" and the "dark-field wide-field image data P5".

[0072] (Structure of the microscopic image generation device)

[0073] The microscopic image generation device 20 includes a control device 50 and a microscopic imaging unit 60. The control device 50 includes a control unit 52 and a communication I / F 56. The control unit 52 controls the microscopic image generation device 20. The control unit 52 includes a memory constituted by a CPU and a non-volatile memory, etc. The CPU executes the processing described later according to the computer program stored in the memory.

[0074] The communication I / F 56 is an interface for the microscopic image generation device 20 to communicate with the detection device 19 via a wired or wireless communication network such as a LAN or Wi-Fi (registered trademark).

[0075] The control device 50 controls the microscopic photographing unit 60. The microscopic photographing unit 60 includes moving units 61 and 62 and a camera unit 63. As Figure 4 shown, the camera unit 63 includes a microscope 66, a bright-field light source 67, and a dark-field light source 68. The magnification of the microscope 66 is higher than that of the cameras 44a and 46a. The bright-field light source 67 is used for the microscope 66 to photograph the glass plate G in bright field. The optical axis of the microscope 66 is set to photograph, using the microscope 66, the reflected light L4 after the light irradiated from the bright-field light source 67 is reflected by the defect D. That is, the microscope 66 and the bright-field light source 67 constitute a bright-field microscopic photographing system 64. The dark-field light source 68 is used for the microscope 66 to photograph the glass plate G in dark field. The optical axis of the microscope 66 is set to photograph, using the microscope 66, the scattered light L5 after the light irradiated from the dark-field light source 68 irradiates the defect D. That is, the microscope 66 and the dark-field light source 68 constitute a dark-field microscopic photographing system 65. In the present embodiment, when photographing the defect using the microscopic photographing unit 60, after photographing using the bright-field microscopic photographing system 64, photographing is performed using the dark-field microscopic photographing system 65, but photographing may also be performed using the bright-field microscopic photographing system 64 after photographing using the dark-field microscopic photographing system 65. It should be noted that the bright-field microscopic photographing system 64 is an example of a "bright-field microscopic photographing unit", and the dark-field microscopic photographing system 65 is an example of a "dark-field microscopic photographing unit". In addition, the image data of the image photographed by the microscopic photographing unit 60 is an example of "microscopic learning image data P2", the image data of the image photographed by the bright-field microscopic photographing system 64 is an example of "bright-field microscopic image data P6", and the image data of the image photographed by the dark-field microscopic photographing system 65 is an example of "dark-field microscopic image data P7". Therefore, "microscopic learning image data P2" includes "bright-field microscopic image data P6" and "dark-field microscopic image data P7".

[0076] Specifically, the microscope 66 can use, for example, a digital microscope with a magnification of 5 times or more and 50 times or less. The magnification of the microscope 66 may also be variable.

[0077] As Figure 2As shown, the moving unit 61 includes a pair of guide rails and an actuator that move the camera unit 63 in a direction parallel to the X direction. The guide rails of the moving unit 61 extend parallel to the X direction outside the conveying device 22. The actuator moves the moving unit 62 along the guide rails. The moving unit 62 includes: guide rails that extend perpendicular to the pair of guide rails of the moving unit 61 and span the pair of guide rails of the moving unit 61; and an actuator that is mounted on the camera unit 63 and moves the camera unit 63 along the guide rails. The moving units 61 and 62 are controlled by the control device 50. The actuators of the moving unit 61 and the moving unit 62 may include, for example, a servo motor and a ball screw, or may include a linear motor. Alternatively, a timing belt, a chain, etc. may be used instead of the ball screw. The moving speed of the camera unit 63 based on each of the moving units 61 and 62 is, for example, 500 mm / second or more and 2000 mm / second or less.

[0078] An acquisition terminal 70 is communicably connected to the inspection device 18. The acquisition terminal 70 is a desktop or laptop computer, a tablet terminal, a mobile terminal, etc. The acquisition terminal 70 includes a control unit 72, an operation unit 74, a display unit 75, and a communication I / F 76. The control unit 72 includes a memory composed of a CPU and a non-volatile memory, etc. The CPU executes the processes described below according to the computer programs stored in the memory. The control unit 72 is communicably connected to the operation unit 74, the display unit 75, and the communication I / F 76 through wirings (not shown). The operation unit 74 receives operations performed by the user. The operation unit 74 includes a keyboard and a mouse. The display unit 75 displays an image in a manner visually recognizable by the user. The display unit 75 includes a display device such as a liquid crystal display or an organic EL display. It should be noted that the operation unit 74 and the display unit 75 may also be integrally formed like a touch panel, for example. The communication I / F 76 is configured in the same manner as the communication I / Fs 36 and 56 and is used for the acquisition terminal 70 to communicate with the detection device 19 and the microscopic image generation device 20 respectively. It should be noted that the acquisition terminal 70 is an example of an "acquisition unit".

[0079] (Method for manufacturing a glass plate)

[0080] Refer to Figure 5 , and the manufacturing method of the glass plate G will be described. First, in S12, a forming process is performed in the forming device 12. Thus, a glass ribbon is formed from molten glass. Next, in S14, a cooling process is performed in the cooling device 14. Thus, the glass ribbon is annealed while moving in the annealing furnace, and then cooled while moving in the cooling chamber. In S16, a cutting process is performed in the cutting device 16. Thus, the glass plate G is produced from the glass ribbon.

[0081] Next, in S18, an inspection process is performed in the inspection device 18. At the end of the inspection process, in S20, an unloading process is performed in the transfer device 22. In the unloading process, the glass plate G is unloaded from the manufacturing unit 10 by the transfer device 22 and placed on a tray or the like, thereby completing the manufacturing method.

[0082] It should be noted that, for the purpose of improving productivity, multiple display devices are sometimes cut out from a single glass plate G. Therefore, the size of the glass plate G in the present embodiment is preferably 1000 mm × 1000 mm or more and 4000 mm × 4000 mm or less. The lower limit value of the size of the glass plate G is more preferably 1500 mm × 1500 mm or more, and further preferably 2000 mm × 2000 mm or more. The upper limit value of the size of the glass plate G is more preferably 3500 mm × 3500 mm or less, and further preferably 3000 mm × 3000 mm or less.

[0083] (Inspection process)

[0084] Refer to Figure 6 , and the inspection process of S18 will be described. In the inspection process, in S32, the glass plate G is transferred from the cutting device 16 to the inspection device 19 by the transfer device 22. Hereinafter, the inspection device 19 and the microscopic image generation device 20 will be mainly described, but the processes performed by the inspection device 19 and the microscopic image generation device 20 are the processes performed by each part of the inspection device 19 under the control of the control unit 32 of the inspection device 19 and the processes performed by each part of the microscopic image generation device 20 under the control of the control unit 52 of the microscopic image generation device 20.

[0085] In S34, the inspection device 19 uses the line array camera unit 42 to photograph the glass plate G transferred by the transfer device 22 (wide-area learning image acquisition process). The line array camera unit 42 photographs the glass plate G in bright field through the bright field wide-area photographing system 44, and photographs in a state where the bright field and the dark field are combined through the dark field wide-area photographing system 46. As Figure 7As shown, for example, in the case where the defect is a bubble, the defect images representing the defect are shown differently in the bright-field wide-area image data P4 and the dark-field wide-area image data P5. In S35, the detection device 19 determines whether the entire surface of the glass plate G has been photographed. Specifically, the glass plate G is continuously conveyed by the conveying device 22. The line array camera unit 42 photographs the entire glass plate G by photographing a part of the glass plate G facing the line array camera unit 42 multiple times. When the detection device 19 has performed a prescribed number of photographings on one glass plate G, it determines that the entire surface of the glass plate G has been photographed (S35: Yes), and proceeds to S36. On the other hand, when the detection device 19 has not performed a prescribed number of photographings on one glass plate G, it determines that the entire surface of the glass plate G has not been photographed (S35: No), and returns to S34. Thus, the entire surface of the glass plate G is photographed. Note that in S35, it is also possible to determine whether the entire surface of the glass plate G has been photographed by detecting the position of the glass plate G using a sensor (not shown).

[0086] The detection device 19 stores the bright-field wide-area image data P4 and the dark-field wide-area image data P5 in the memory of the control unit 32. The detection device 19 also stores in combination the position information indicating the position of the images represented by the bright-field wide-area image data P4 and the dark-field wide-area image data P5 in the glass plate G with the bright-field wide-area image data P4 and the dark-field wide-area image data P5. The position information is determined by the positions of the clamping mechanisms 24 and 25 in the X direction and the positions of the cameras 44a and 46a in the vertical direction.

[0087] In S36, the detection device 19 uses the bright-field wide-area image data P4 and the dark-field wide-area image data P5 to determine whether a defect is detected in the images represented by the bright-field wide-area image data P4 and the dark-field wide-area image data P5. As Figure 7 shown, when there is a defect, in the images represented by the bright-field wide-area image data P4 and the dark-field wide-area image data P5, the part representing the defect appears darker than other parts. Thus, when the image contains a part that appears darker, i.e., has a lower brightness, it is determined that a defect has been detected. When it is determined that no defect has been detected (S36: No), the inspection process ends. On the other hand, when it is determined that a defect has been detected (S36: Yes), the process proceeds to S38. Note that the data including the defect in the bright-field wide-area image data P4 and the dark-field wide-area image data P5 is an example of "defect image data".

[0088] In S38, the detection device 19 determines the defect position (learning defect position acquisition process) indicating the position of the detected defect. Specifically, using the information saved in S34, the position of the region represented by the image in the glass plate G is determined. Next, the position of the defect within the image is determined. Thereby, the position of the defect in the glass plate G is determined. Next, in S40, the detection device 19 determines the type of the defect (learning defect type determination process and inspection target defect type determination process). Specifically, according to the learning model 34, the bright-field wide-field image data P4 and the dark-field wide-field image data P5 are used to determine the type of the defect detected in S36.

[0089] It should be noted that in S40, the detection device 19 can also perform image processing using the bright-field wide-field image data P4 and the dark-field wide-field image data P5 to generate one or more processed image data in a manner that facilitates the discrimination of defects. For example, filtering processes such as contour enhancement and noise removal can also be performed.

[0090] Next, in S42, as Figure 8 shown, the detection device 19 causes the control unit 32 to save the combination of the defect position determined in S38, the type of the defect determined in S40, the bright-field wide-field image data P4 and the dark-field wide-field image data P5 acquired in S34. It should be noted that when one or more processed image data are generated by image processing in S40, the combination including the processed image data is also caused to be saved by the control unit 32.

[0091] Next, in S44, the detection device 19 sends the combination saved in S42 to the microscopic image generation device 20 and the acquisition terminal 70. It should be noted that when multiple defects are detected in one glass plate G, the detection device 19 sends the saved combination for each of the multiple defects to the microscopic image generation device 20 and the acquisition terminal 70. When the control unit 72 of the acquisition terminal 70 receives the combination, it can cause the display unit 75 to display the images represented by the bright-field wide-field image data and the dark-field wide-field image data, the defect position, and the type of the defect. Thereby, the user can grasp the position and type of the defects existing in the glass plate G.

[0092] In S46, the microscopic image generation device 20 determines the defects to be photographed in the microscopic image generation device 20 from the combinations received in S44. Specifically, when the number of received combinations is equal to or less than a specified number (for example, three), the microscopic image generation device 20 determines all the defects as the defects to be photographed. On the other hand, when the number of received combinations is greater than the specified number, the microscopic image generation device 20 determines the specified number of defects as the defects to be photographed from the received combinations in ascending order of the defect size.

[0093] If the entire glass plate G is photographed by the inspection device 19, the glass plate G is transported from the inspection device 19 to the microscopic image generation device 20 by the transport device 22. In S48, the microscopic image generation device 20 drives the moving units 61 and 62 to move the camera unit 63 to the positions of the defects determined in S46. Next, in S50, the microscopic image generation device 20 photographs the defects through the camera unit 63 (microscopic learning image generation process). In S50, the microscopic image generation device 20 photographs the number of defects determined in S46.

[0094] In the microscopic image generation device 20, a microscope 66 is used to photograph the magnified defects. As Figure 9 shown, in the microscopic image generation device 20, an image of the bright-field defect D is photographed in the bright-field microscopic photographing system, and an image of the dark-field defect D is photographed in the dark-field microscopic photographing system. Hereinafter, the image photographed by the microscopic image generation device 20 is referred to as a microscopic image, and the image data representing the microscopic image is referred to as microscopic image data. When the defect D is a foreign object, it becomes a black image in the bright-field wide-field image data P4 and a white image in the dark-field wide-field image data P5, but sometimes the image itself cannot be obtained. When the defect D is a bubble, the bright-field wide-field image data P4 and the dark-field wide-field image data P5 become an approximately oval shape. When the defect D is an attachment such as dust, it becomes, for example, a black undulating shape in the bright-field wide-field image data P4 and a white undulating shape in the dark-field wide-field image data P5.

[0095] Next, in S52, as Figure 10 shown, the microscopic image generation device 20 appends the image data representing the image photographed in S50 to the combination determined in S46, sends it to the acquisition terminal 70, and ends the process. That is, in S52, the wide-field learning image data P1 (bright-field wide-field image data P4 and dark-field wide-field image data P5), the microscopic learning image data P2 (bright-field microscopic image data P6 and dark-field microscopic image data P7), the defect positions, and the types of defects are sent to the acquisition terminal 70.

[0096] The user determines the type of a defect using the combination received in the acquisition terminal 70, namely the wide-field learning image data P1, the microscopic learning image data P2, the defect position, and the type of the defect, that is, the wide-field learning image data P1 and the microscopic learning image data P2. With this configuration, the user can use the magnified microscopic learning image data P2 to determine the type of a small defect. Thus, for a small defect, the type of the defect can be accurately determined.

[0097] As Figure 11 shown, the user saves the determined type of the defect and the wide-field learning image data P1 as learning data in the acquisition terminal 70. As a result, as Figure 12 shown, for multiple defects, a learning database (saving process) is saved that contains multiple learning data formed by combining multiple image data with the types of the defects.

[0098] When generating the learning database, the user sends the generated learning database to the inspection device 19. When receiving the learning database, the inspection device 19 updates the learning model using the learning data contained in the learning database. Thus, the determination accuracy of the type of a defect based on the learning model can be improved. In addition, since the microscopic learning image data P2 of the defect is generated by the manufacturing unit 10, there is no need to separately perform a process for generating the microscopic learning image data P2.

[0099] According to the inspection method performed in the inspection process of the present embodiment, the microscopic learning image data P2 of the defect of the learning object detected by the inspection device 19 can be captured by the microscopic image generation device 20. Thus, even for a defect with a small size, the user can easily determine the type of the defect by confirming the microscopic learning image data P2. By generating a learning model using the combination of the type of the defect and the image data as learning data, even when the size of the defect of the inspection object is small, the type of the defect can be determined using the learned learning model 34. It should be noted that the image data of the image captured by the microscopic image generation device 20 for the learning glass plate G1 (learning object) is the "wide-field learning image data P1", and the image data of the image captured by the microscopic image generation device 20 for the inspection object glass plate G2 (inspection object) is the "wide-field inspection image data P3". In addition, the process of determining the type of the defect included in the learning glass plate G1 (learning object) is the "learning defect type determination process", and the process of determining the type of the defect included in the inspection object glass plate G2 (inspection object) is the "inspection object defect type determination process".

[0100] Multiple defects such as small unevenness, air bubbles, and foreign substances are generated on the surface of the glass plate G. Depending on the type and state of the defects, there are defects that are clearly captured when photographed in bright field and defects that are clearly captured when photographed in dark field. In the detection device 19, by using the image data taken in bright field and the image data synthesized from dark field and bright field, the type of defect can be easily determined in the learning model. In the microscopic image generation device 20, by using the bright-field microscopic image data P6 taken in bright field and the dark-field microscopic image data P7 taken in dark field, the user can easily determine the type of defect.

[0101] The microscopic image generation device 20 selects the defects to be photographed from among the multiple defects included in the glass plate G. As a result, not all the defects included in the glass plate G need to be used as learning data. Thus, the time required to generate the microscopic learning image data can be shortened, and the burden of generating the learning data and the learning period in the learning model 34 can be suppressed. In addition, since defects with relatively small sizes are selected, image data representing defects with relatively small sizes can be used as learning data. For defects with relatively large sizes, the type of defect can be relatively easily determined from the images taken by the detection device 19. On the other hand, it is difficult to determine the type of defect with relatively small sizes from the images taken by the detection device 19. In the present embodiment, by collecting learning data for defects with small sizes, the determination accuracy of the type of defect using the learning model 34 can be improved.

[0102] The smaller the size of the pixels displayed by the display device, the smaller the size of the defects allowed for the glass plate G used for the display device. By improving the determination accuracy of the type of defect of the learning model 34, the inspection device 18 can appropriately determine the type of defect of the glass plate G used for the display device.

[0103] In addition, in the present embodiment, the size of the glass plate G is, for example, 1000 mm × 1000 mm or more. In such a large glass plate G, when photographing all the defects included in the glass plate G using the microscopic image generation device 20, the inspection time significantly increases, deteriorating the production efficiency of the glass plate G. By selecting defects with relatively small sizes and photographing them using the microscopic image generation device 20, the determination accuracy of the type of defect using the learning model 34 can be improved without deteriorating the production efficiency of the glass plate G.

[0104] In another embodiment, compared with the above embodiment, the manufacturing method of the glass plate is different. Specifically, as Figure 13 shown, the processes of S12 to S18 are performed in the same manner as in the above embodiment. In the inspection process of S18, in addition to the type of defect included in the glass plate G, the number and size of the defects included in the glass plate G are also determined.

[0105] Specifically, in S40 of the inspection process, the detection device 19 also determines the number of defects detected in S36 and the size of each defect. For example, the detection device 19 can also count the number of defects included in the bright-field wide-field image data P4 and the dark-field wide-field image data P5. Additionally, for example, the detection device 19 can count the number of times the type of defect has been determined as the number of defects. The detection device 19 can also determine the size of the defect by measuring the size of the defects included in the bright-field wide-field image data P4 and the dark-field wide-field image data P5. Additionally, the detection device 19 can classify the defects into multiple grades (for example, 10 Hereinafter, 10 to 50 , 50 and above). The detection device 19 stores the type, number, and size of the determined defects.

[0106] Next, at the end of the inspection process in S18, in S19, a quality determination process is executed. In the quality determination process, the quality of the glass plate G is determined. More specifically, in the quality determination process, based on the defect information including the type, number, and size of the defects determined in the inspection process of S18, the quality of the glass plate G is determined. The glass plate G is classified into multiple grades according to its quality. The multiple grades are three grades or more including one grade of non-conforming quality and two or more grades of good quality.

[0107] For example, in the control unit that controls the transfer device 22, non-conforming product thresholds are stored in advance for the number and size of the defects respectively. When at least one of the number and size of the defects determined in the inspection process of S18 by the control unit of the transfer device 22 exceeds the stored non-conforming product threshold, the glass plate G is determined to be of non-conforming quality. Additionally, when the glass plate G contains a specific type of defect, the control unit of the transfer device 22 determines the glass plate G to be of non-conforming quality. When neither the number nor the size of the defects determined in the inspection process of S18 by the control unit of the transfer device 22 exceeds the stored non-conforming product threshold and the glass plate G does not contain a specific type of defect, the control unit of the transfer device 22 determines the glass plate G to be of good quality.

[0108] In the control unit that controls the transfer device 22, grade thresholds are also stored in advance for the number of defects and the size of the defects, respectively. The control unit of the transfer device 22 differentiates grades based on the situation where both the number of defects and the size of the defects determined in the inspection process of S18 do not exceed the stored grade thresholds and the situation where at least one of the determined number of defects and the size of the defects exceeds the stored grade thresholds. It should be noted that grades can also be differentiated according to the type of defect. It should be noted that the quality determination process can also be executed by other structures of the manufacturing unit 10, for example. It should be noted that the quality determination process can also determine the quality using at least one of the number of defects, the size of the defects, and the type of defect.

[0109] At the end of the quality determination process, the carry-out process of S20 is executed. In the carry-out process, the glass plate G is carried out of the manufacturing unit 10 by the transfer device 22 and classified into trays of each grade according to the quality grade determined in the quality determination process of S19. In addition, in S21, a non-conforming product classification process is executed in parallel with the carry-out process of S20. In the non-conforming product classification process, the glass plate G determined to be of non-conforming quality in the quality determination process of S19 is carried by the transfer device 22 to a waste tray or the like. Thus, the glass plate G determined to be of non-conforming quality in the quality determination process of S19 is discarded. The manufacturing unit 10 can also further execute a waste process that discards the glass plate G transported to a waste tray or the like.

[0110] In this embodiment, by classifying the glass plates G according to each quality, it is not necessary to confirm the quality when the glass plates G are shipped out. The glass plates G of a quality consistent with the quality required for the display device can be shipped to the destination.

[0111] By executing the non-conforming product classification process, it is possible to prevent non-conforming products from being included in the multiple glass plates G carried out.

[0112] The above has described specific examples of the technology disclosed in this specification, but these are merely examples and do not limit the technical solution. The technology described in the technical solution includes the technology obtained by various deformations and changes to the above-described specific examples. For example, the following deformation examples can also be adopted.

[0113] (Modification Example 1) The bright-field wide-field imaging system 44 generates image data by performing imaging in the bright field. However, the bright-field wide-field imaging system 44 can also generate image data by synthesizing the bright field and the dark field. The dark-field wide-field imaging system 46 can also generate image data by performing imaging in the dark field. This also applies to the microscopic image generation device 20.

[0114] (Modification Example 2) The acquisition terminal 70 may also be included in the inspection device 18. For example, the acquisition terminal 70 may be integrally formed with the microscopic image generation device 20.

[0115] (Modification Example 3) The inspection device 18 may not include a photographing system. For example, there may be other photographing devices independent of the inspection device 18, and the inspection device 18 acquires photographed image data from the other photographing devices via wireless communication such as wireless LAN.

[0116] (Modification Example 4) There may also be other detection devices that use the microscopic learning image data P2 to determine the type of defect. That is, instead of the user using the microscopic learning image data P2 to determine the type of defect, other detection devices equipped with a prescribed learning model determine the type of defect according to the prescribed learning model.

[0117] (Modification Example 5) The transfer device 22 may not include the clamping mechanism 25 that clamps the lower end edge of the glass plate G. That is, the glass plate G may be transferred in a state where only the upper end edge is clamped by the clamping mechanism 24.

[0118] (Modification Example 6) The microscopic photographing unit 60 includes a microscope 66, a bright-field light source 67, and a dark-field light source 68 on the surface Gb side of the glass plate G, and photographs the reflected light L4 and the scattered light L5 of the defect D. However, a microscope 66 may be provided on the surface Gb side of the glass plate G, and a bright-field light source 67 and a dark-field light source 68 may be provided on the surface Ga side, and the scattered light and the transmitted light of the defect D are photographed. In this case, in addition to the moving units 61 and 62 for moving the microscope 66, there are also moving units for moving the bright-field light source 67 and the dark-field light source 68, and the microscope 66, the bright-field light source 67, and the dark-field light source 68 are moved synchronously.

[0119] (Modification Example 7) In the quality determination step S19, the quality of the glass plate G may be determined based on defect information that includes, in addition to the type, number, and size of the defects determined in the inspection step of S19, the position of the defects. For example, as a post-process of the quality determination step S19, in the case of having a step of cutting out a second glass plate with a smaller size from the glass plate G according to a prescribed cutting pattern, when there is a cutting pattern such that at least one of the number and size of the defects included in the glass plate G is greater than the nonconforming product threshold, but the number and size of the defects included in the second glass plate are below the nonconforming product threshold, the glass plate G may be determined to be of good quality.

[0120] The technical elements described in this specification or the drawings exhibit technical utility either individually or in various combinations, and are not limited to the combinations described in the technical solution at the time of application. Additionally, the technologies exemplified in this specification or the drawings achieve multiple purposes simultaneously, and the achievement of any one of these purposes inherently has technical utility.

Claims

1. An inspection device, wherein: The inspection device comprises: a determination unit that determines a defect position of a predetermined defect in the learning object using wide-area learning image data representing the learning object; a microscopic image generating unit for generating microscopic learning image data representing the predetermined defect by microscopically photographing the determined defect position; as well as an acquisition unit that acquires a combination of type information and defect image data, wherein the type information indicates the type of the predetermined defect obtained using the generated microscopic learning image data, and the defect image data indicates the predetermined defect included in the wide-area learning image data, The determination unit determines the type of defect in the inspection object using wide-area inspection image data representing the inspection object using a learning model generated by machine learning using the combination of the acquired type information and the defect image data as learning data.

2. The inspection device according to claim 1, wherein: The learning object includes a glass plate, The inspection device further includes an inspection image generating unit that generates the wide area learning image data. The inspection image generating unit comprises: a bright field wide area imaging unit for generating bright field wide area image data representing the glass plate by imaging the first transmitted light that has passed through the glass plate in a field of view including the bright field; as well as a dark field wide area imaging unit for generating dark field wide area image data representing the glass plate by imaging, in a field of view including the dark field, second transmitted light incident on the glass plate at an angle different from that of the first transmitted light and transmitted through the glass plate, The wide-area learning image data includes the bright-field wide-area image data and the dark-field wide-area image data.

3. The inspection device according to claim 1 or 2, wherein: The learning object includes a glass plate, The microscopic image generating unit comprises: a bright field microscopic imaging unit for generating bright field microscopic image data representing the glass plate by imaging reflected light reflected by the glass plate in a field of view including the bright field; as well as a dark field microscopic imaging unit for generating dark field microscopic image data representing the glass plate by imaging scattered light scattered by the defect in a field of view including the dark field, The microscopic learning image data includes the bright field microscopic image data and the dark field microscopic image data.

4. The inspection device according to claim 1 or 2, wherein: The microscopic image generating unit comprises: a camera unit for taking a microscopic photograph of the predetermined defect; and a moving unit that moves the camera unit toward the defect position, When the learning object includes a plurality of defects, the microscopic learning image data of the predetermined defects whose number is smaller than the number of the plurality of defects is generated, and the plurality of defects include one or more of the predetermined defects.

5. The inspection device according to claim 4, wherein: The microscopic image generating unit generates the microscopic learning image data of the predetermined defects in ascending order of size from the predetermined defects.

6. The inspection device according to claim 1 or 2, wherein: The learning object includes a glass plate for a display device.

7. The inspection device according to claim 6, wherein: The wide-area inspection image data includes the entire effective surface of the glass plate.

8. A method for generating learning data, wherein: The learning data generation method comprises: a wide area learning image acquisition step of acquiring wide area learning image data representing an object; a defect position acquisition step for learning, acquiring a defect position of a predetermined defect of the object in the object; a microscopic learning image generating step of generating microscopic learning image data representing the predetermined defect by microscopically photographing the acquired defect position; and A storing step of storing a combination of type information indicating the type of the predetermined defect obtained using the generated microscopic learning image data and defect image data indicating the predetermined defect included in the wide area learning image data.

9. A method for generating a learning model, wherein: The learning model generation method comprises: a wide area learning image acquisition step of acquiring wide area learning image data representing an object; a defect position acquisition step for learning, acquiring a defect position of a predetermined defect of the object in the object; a microscopic learning image generating step of generating microscopic learning image data representing the predetermined defect by microscopically photographing the acquired defect position; a storing step of storing a combination of type information indicating the type of the predetermined defect obtained using the generated microscopic learning image data and defect image data indicating the predetermined defect included in the wide area learning image data; and The learning model generating step generates a learning model by performing machine learning using the stored combination as learning data.

10. A method for manufacturing an object, wherein: The method for manufacturing the object comprises: a defect type determination step for learning, using wide-area learning image data representing a learning object to determine a type of a predetermined defect in the learning object; a defect position acquisition step for learning, using the wide-area learning image data to acquire a defect position of the predetermined defect in the learning object; a microscopic learning image generating step of generating microscopic learning image data representing the predetermined defect by microscopically photographing the acquired defect position; and a storing step of storing a combination of type information and defect image data, wherein the type information indicates the type of the predetermined defect obtained using the generated microscopic learning image data, and the defect image data indicates the predetermined defect included in the wide-area learning image data; The manufacturing method of the object also includes a step of determining the type of defect of the inspection object. In the step of determining the type of defect of the inspection object, the type of defect in the inspection object is determined by using a learning model and wide-area inspection image data representing the inspection object. The learning model is generated by machine learning by using the combination of the saved type information and the defect image data as learning data.

11. The method for manufacturing an object according to claim 10, wherein: The manufacturing method of the object also includes a quality judgment step, in which the quality of the object is judged based on defect information, and the defect information includes the number of defects contained in the object, the size of the defects, and at least one of the types of defects determined in the inspection object defect type determination step.

12. The method for manufacturing an object according to claim 11, wherein: The object includes a glass plate for a display device, The object manufacturing method further includes a carrying-out step of classifying the object for each quality based on the quality determined in the quality determination step.

13. The method for producing an object according to claim 11 or 12, wherein: The method for manufacturing an object further includes a defective product classification step of distinguishing the object determined as defective in the quality determination step from the object not determined as defective.

Images