Defect inspection method, defect inspection device, and method for manufacturing transparent plate-shaped body

Abstract

The purpose of the present invention is to provide a defect inspection device, a defect inspection method, and a method for manufacturing a transparent plate-shaped body, whereby it is possible to reduce the amount of computation required for defect inspection. The present invention is provided with a search unit that extracts a first defect candidate for a transparent plate-shaped body from the entire first image data obtained by imaging the plate-shaped body by an imaging device, and calculates a first feature amount of the first defect candidate. And compares the first feature amount with a predetermined threshold value to determine whether the defect is a predetermined defect type.

Description

Technical Field

[0001] This invention relates to a defect inspection method, a defect inspection apparatus, and a method for manufacturing a transparent plate-like material, for detecting defects present in a plate-like material such as a transparent glass substrate. Background Technology

[0002] Currently, glass substrates (hereinafter referred to as glass plates), which are transparent plate-shaped bodies, are used in electronic devices such as flat panel displays. For glass plates used in these devices, it is strongly required that they be thin and have very few or no defects such as bubbles. Therefore, in the inspection process, it is necessary to remove or process portions of the glass plate containing bubbles. Consequently, various devices for inspecting for defects such as bubbles in manufactured glass plates have been proposed.

[0003] Figure 14 This is a diagram illustrating an example of an inspection device (hereinafter referred to as an inspection device) for inspecting defects such as bubbles present in glass plates, as described in the prior art. For example... Figure 14 As shown, in conventional inspection apparatuses, a light source 901 is provided on one side of the transported glass plate 911. The light source 901 projects a linear beam of light, longer than the width of the glass plate 911, towards the glass plate 911 at a predetermined light intensity. Conventional inspection apparatuses use a camera 902 located on the other side of the glass plate 911 to capture a bright-field image of the transmitted light through the glass plate 911, and output the captured image to a processing unit 903. Furthermore, the processing unit 903 of the conventional inspection apparatus extracts the dark areas contained in the bright-field image as defect areas. In addition, conventional inspection apparatuses use a narrow slit extending in the width direction of the glass plate 911 to transmit the light projected by the light source 901 as simulated parallel light. The conventional inspection apparatus captures this simulated parallel light using the camera 902, and the processing unit 903 performs image processing on the captured image, thereby detecting defects in the transported glass plate 911.

[0004] In addition, Patent Document 1 proposes the following scheme: illuminating at an angle nearly perpendicular to the surface of the plate and illuminating at an angle nearly parallel, and detecting defects on the surface and inside of the plate by performing image processing on images obtained using these illuminating devices.

[0005] Figure 15 It is a diagram illustrating the manufacturing process of glass plates using the float glass method.

[0006] In the float glass process, which is used to manufacture sheet glass, such as Figure 15As shown, molten glass is fed from furnace 921, flowing while floating above a floating polishing furnace 922 containing molten metal with a specific gravity greater than glass. The molten glass is then conveyed from the floating polishing furnace 922 to a slow cooling furnace 923, where it is cooled. The cooled glass solidifies into a sheet shape, becoming a glass ribbon 924, which is then cut to produce glass sheets. General inspections, such as for defects, are performed on the glass sheets after cleaning. Inspecting for defects before cleaning has the advantage of reducing the number of sheets deemed defective in subsequent processes, but this is affected by interference from pre-cleaning inspections (dust, dirt, etc.).

[0007] Existing technical documents

[0008] Patent documents

[0009] Patent Document 1: Japanese Patent Application Publication No. 2002-214158 Summary of the Invention

[0010] The problem that the invention aims to solve

[0011] However, in the existing inspection devices and the prior art described in Patent Document 1, when defect inspection is performed before the cleaning process, a large number of defect candidates are detected because a large amount of dust and transport dirt adhere to the glass. The inspection processing speed is not fast enough, and there is a problem that it is difficult to perform inspection in accordance with the transport speed of the glass strip before the cutting process.

[0012] The present invention was made in view of the above-mentioned problems, and provides a defect inspection apparatus, a defect inspection method, and a method for manufacturing a transparent plate-shaped body that can reduce the amount of computation required for defect inspection.

[0013] Technical solutions for solving the problem

[0014] The present invention includes the following [1] to [8].

[0015] [1] A defect inspection method in a defect inspection device, comprising: a first defect candidate extraction step, extracting a first defect candidate for the plate-shaped body from a first image data obtained by taking a picture of the plate-shaped body with transparency from an imaging device; a first feature calculation step, calculating a first feature of the first defect candidate; and a defect category determination step, determining whether the first defect candidate is a predetermined defect category based on the first feature.

[0016] [2] According to the defect inspection method of [1], the defect inspection method further includes the following steps when the first defect candidate is determined not to be the predetermined defect category in the defect category determination step: a step of extracting and generating second image data of a predetermined size containing the pixels of the first defect candidate from the first image data; a second defect candidate extraction step of extracting the second defect candidate from the second image data; a second feature calculation step of calculating the second feature quantity of the second defect candidate; and a step of determining the defect category of the second defect candidate based on the second feature quantity.

[0017] [3] According to the defect inspection method of [1] or [2], wherein the first feature quantity is a feature quantity that can be used to determine at least one of dust and dirt adhering to the surface of the plate-like body.

[0018] [4] The defect inspection method according to any one of [1] to [3], wherein the first defect candidate extraction step searches the first image data for each pixel and extracts the first defect candidate.

[0019] [5] The defect inspection method according to any one of [1] to [4], wherein the plate-shaped body is any one of a glass strip formed by continuously flowing molten glass on a molten metal bath surface, a float glass substrate formed by moving the glass strip along the molten metal bath surface to form a plate, a glass substrate for flat panel displays, a glass plate for building materials, and a glass plate for automobiles.

[0020] [6] A defect inspection device, wherein the defect inspection device includes a search unit that extracts a first defect candidate for the plate-shaped body from a first image data obtained by an imaging device from an image of the plate-shaped body with transparency, calculates a first feature quantity of the first defect candidate, and compares the first feature quantity with a predetermined threshold to determine whether the first defect candidate is a predetermined defect category.

[0021] [7] According to the defect inspection apparatus of [6], the defect inspection apparatus further includes an identification unit that calculates a second feature quantity and determines a defect category. When the search unit determines that the first defect candidate is not the predetermined defect category, it extracts and generates second image data of a predetermined size containing pixels of the first defect candidate from the first image data. The search unit extracts a second defect candidate from the second image data. The identification unit calculates the second feature quantity of the second defect candidate. The identification unit determines the defect category of the second defect candidate based on the second feature quantity.

[0022] [8] A method for manufacturing a transparent plate-shaped body, comprising: a first defect candidate extraction step, extracting a first defect candidate for the plate-shaped body from a first image data obtained by taking a picture of the transparent plate-shaped body from an imaging device; a first feature calculation step, calculating a first feature of the first defect candidate; and a defect category determination step, determining whether the first defect candidate is a predetermined defect category based on the first feature.

[0023] Invention Effects

[0024] According to the present invention, a first defect candidate for the plate-like body is extracted from first image data obtained by an imaging device from an image of the plate-like body. A first feature quantity of the extracted first defect candidate is calculated, and the calculated first feature quantity is compared with a predetermined threshold to determine whether it belongs to a predetermined defect category. If it is determined to belong to a predetermined defect category, the extracted first defect candidate is classified as a predetermined category. As a result, image feature quantities extracted from the first image data can be used to determine inspection interferences such as dirt and dust, and non-defects can be excluded from the defect candidates, thus reducing the computational load required for defect inspection. Attached Figure Description

[0025] Figure 1 This is a diagram illustrating the general structure of the defect inspection device 10 according to the first embodiment.

[0026] Figure 2 This is a diagram illustrating the general structure of the processing apparatus 16 according to the first embodiment.

[0027] Figure 3 This is a diagram illustrating the steps of the defect candidate search processing for the entire image according to the first embodiment.

[0028] Figure 4 This is a diagram illustrating the image data being searched according to the first embodiment.

[0029] Figure 5 This is a diagram illustrating the steps of the defect image generation process according to the first embodiment.

[0030] Figure 6 This is a diagram illustrating an example of defect image data involved in the first embodiment.

[0031] Figure 7 This is a diagram illustrating the steps of the first defect identification process involved in the first embodiment.

[0032] Figure 8 This is a diagram illustrating the steps of the second defect identification process involved in the first embodiment.

[0033] Figure 9 This diagram illustrates the steps involved in the good or bad determination process according to the first embodiment.

[0034] Figure 10 This is a diagram illustrating an example of image data of the determination area that is determined to be good or bad according to the first embodiment.

[0035] Figure 11 This is a diagram illustrating the determination result when using the defect inspection device 10 according to the first embodiment.

[0036] Figure 12 This is a diagram illustrating the schematic structure of the processing apparatus 16a according to the second embodiment.

[0037] Figure 13 This is a diagram illustrating an example of cutting a glass plate while avoiding undesirable areas, as described in the second embodiment.

[0038] Figure 14 This is a diagram illustrating an example of an inspection device for defects such as bubbles present in a glass plate, as described in the prior art.

[0039] Figure 15 It is a diagram illustrating the manufacturing process of float glass sheets. Detailed Implementation

[0040] [First Implementation Method]

[0041] Hereinafter, embodiments of the present invention will be described using the accompanying drawings. Figure 1 This is a diagram illustrating the general structure of the defect inspection device 10 according to this embodiment.

[0042] like Figure 1 As shown, the defect inspection device 10 consists of a first defect inspection device 12, a second defect inspection device 14, a processing device 16, and a display 32.

[0043] exist Figure 1 In this context, the direction of flow of the glass plate G, which is a transparent plate-like body, is defined as the x-direction, the width direction of the glass plate G is defined as the z-direction, and the direction perpendicular to both the x-direction and the z-direction is defined as the y-direction.

[0044] The first defect inspection device 12 and the second defect inspection device 14 are sequentially arranged from the upstream side along the conveying path of the glass plate G. The glass plate G is a long strip of sheet material of a specified thickness taken out from the melting furnace and conveyed along the x-axis direction on a plurality of drive rollers 18 provided on the conveying path. In this embodiment, the glass plate G is either a glass ribbon formed by continuously flowing molten glass onto the surface of a molten metal bath, or float glass formed by moving the glass ribbon along the surface of the molten metal bath.

[0045] The first defect inspection device 12 has a first linear light source 20 on the lower surface of the glass plate G, and a first camera 22 and a light path shielding component 24 on the upper surface of the glass plate G.

[0046] The first defect inspection device 12 is located at the upstream end of the conveying side of the defect inspection device 10 and is used to inspect defects (also called flaws) on the glass plate G. Furthermore, the first defect inspection device 12 is suitable for detecting minute defects such as bubbles present in the glass plate G.

[0047] The first linear light source 20 is a light source that emits simulated parallel light, such as an LED (light-emitting diode) light source. The emission outlet of the first linear light source 20 extends linearly along the z-direction of the glass plate G. The emission outlet of the first linear light source 20 is located at a position separated from the surface of the glass plate G, for example, 100 to 900 mm, and the width of the light source (emission outlet) along the transport direction is set to, for example, 1 to 20 mm. Furthermore, considering that high positional accuracy is not required, it is preferable that the first linear light source 20 is located separated from the surface of the glass plate G. Additionally, there are no particular limitations on the type of light in the LED light source; white is preferred, but red, blue, green, etc., are also acceptable.

[0048] For example, an LED light source includes: a light source that emits light; a lens that makes the emitted light approximately parallel; a diffuser that makes the light intensity approximately uniform; and a slit plate (not shown) that contracts the emitted light. Thus, the first linear light source 20 emits approximately parallel light with approximately uniform light intensity.

[0049] The first camera 22 (imaging device) is a line sensor type camera positioned opposite the first linear light source 20, separated by a glass plate G, and directly reads the transmitted light passing through the glass plate G using its light-receiving surface. Multiple first cameras 22 are arranged in the z-direction to capture images of the same position in the transport direction (x-direction). Furthermore, in Figure 1 The diagram schematically illustrates a first camera 22. Multiple first cameras 22 are configured such that their fields of view in the width direction (z-direction) of the glass plate G partially overlap each other, and are arranged such that there are no non-inspection areas in the inspection portion of the glass plate G.

[0050] The first camera 22 is configured such that its light-receiving surface is located at the focusing position of an imaging lens (not shown), for example, 200-400 mm from the surface of the glass plate G. The first camera 22 has an optical system with an imaging lens and an aperture (not shown) with an adjustment opening. Image data obtained from the first camera 22 is sequentially sent to the processing device 16 whenever it is read as a line.

[0051] The light path blocking member 24 is a blade-shaped component that blocks a portion of the light path of transmitted light from the glass plate G at a position on the front surface of the first camera 22. The leading edge of the light path is sharpened in a blade-like manner. The light path blocking member 24 is positioned on the front surface of the optical system (imaging lens) of the first camera 22, for example, at a distance of 1 to 5 mm. A mechanism is provided in the portion holding the light path blocking member 24 that allows the light path blocking member 24 to move in the X direction in a manner that traverses the light path. At this time, when determining the circle of confusion, which represents the field of view observed from the light-receiving surface, formed on a plane orthogonal to the optical axis of the imaging lens of the first camera 22 at the point where the transmitted light of the light path blocking member 24 passes, it is preferable that the area of ​​the circle of confusion blocked by the light path blocking member 24 is equivalent to 30 to 70% of the area of ​​the circle of confusion. If it is less than 30%, it is difficult to produce the bright areas described later in the bright field image; if it exceeds 70%, it is easy to become a dark field image.

[0052] Such a range of light path blocking can be achieved, for example, by adjusting the distance between the light path blocking component 24 and the glass plate G and the aperture value of the first camera 22. This is to efficiently form bright areas by setting the range of light path blocking within this range, close to the dark areas formed by defects such as bubbles present in the bright field image.

[0053] The image obtained by the first defect inspection device 12 is a bright-field image as described above. The defect in the glass plate G appears as a dark area in the image due to diffuse reflection from the defect region. Furthermore, as described above, a blade-shaped light path shielding member 24 is provided on the front surface of the optical system of the first camera 22, blocking a portion of the light path. This light path shielding member 24 forms a bright area that is close to or nearly opposite the dark area in the image. This bright area is produced due to the abnormal refraction of the defect portion and has a higher brightness compared to the background of the bright-field image.

[0054] The second defect inspection device 14 includes a second light source 28 and a second camera 30. The second defect inspection device 14 is a device that illuminates the glass plate G, which is the object of inspection of the first defect inspection device, from one side, and uses the second camera 30 to receive the illumination light reflected from the surface and back of the glass plate G to inspect for defects.

[0055] The second light source 28 is a light source (e.g., an LED light source) that emits approximately parallel light towards the surface of the glass plate G, causing the light to enter from a direction inclined relative to the surface of the glass plate G. For example... Figure 1As shown, the second light source 28 is angled in the y-direction. The type of light used in the LED light source for the second light source 28 is not particularly limited; white is preferred, but red, blue, green, etc., are also acceptable. Specifically, the LED light source includes: a light source that emits light; a lens that makes the emitted light approximately parallel; a diffuser that makes the light intensity approximately uniform; and a slit plate (not shown) that contracts the emitted light. Thus, the second light source 28 emits approximately parallel light with approximately uniform light intensity.

[0056] The second camera 30 (imaging device) is a line sensor type camera that focuses reflected light emitted from the surface of the glass plate G and captures a bright-field reflected image. When viewed from the glass plate G, the second camera 30 is positioned on the same side as the second light source 28.

[0057] The image captured by the second camera 30 includes an image reflected from the back of the glass plate G (the side opposite to the side where the second light source 28 and the second camera 30 are located) illuminated by the second light source 28, and is an image of the dark area in the defective area of ​​the glass plate G.

[0058] The image of the defect comprises two images. The first image is a real image of the defect formed by incident light from a direction inclined relative to the surface of the glass plate G, reflecting off the back of the glass plate G, and passing through the region of the defect in the optical path of the reflected light. The second image is a mirror image of the defect formed by incident light from a direction inclined relative to the surface of the glass plate G, reflecting off the back of the glass plate G after passing through the region of the defect in the optical path within the glass plate G. Image data obtained by the second camera 30 is sent to the processing device 16 whenever it is read as a line.

[0059] The processing device 16 extracts defect candidates (first defect candidates) based on the captured image data (first image data), as described later. Then, it determines the defect category of the pixels of the extracted first defect candidates, and if the pixel is determined to be a non-defect (suspected defect), it excludes the first defect candidate from the defect candidates. The processing device 16 also generates image data (second image data) containing defect candidates that were not excluded from the first defect candidates. The processing device 16 identifies the defect categories in detail based on the generated second image data. Based on the identification results, the processing device 16 determines the good and defective areas of the glass plate and outputs the determination results to the display 32.

[0060] A display 32 is connected to the processing device 16. The display 32 shows the images, defect detection results, identification results, and / or defect location determination results obtained by the first defect inspection device 12 and the second defect inspection device 14.

[0061] Next, a general description of the operation of the processing device 16 in this embodiment will be given.

[0062] When defects such as bubbles exist in the glass plate, image data of a whitish state is obtained. On the other hand, when dust or other contaminants adhere to the glass surface, image data of a black state is obtained. However, even in the case of image data obtained as a black state, small defects such as bubbles may still exist. Especially in the manufacture of high-quality glass plates, it is required to detect such small bubble defects. In addition, when inspected in a cleanroom before cleaning, a large amount of dust and dirt (suspected defects) may be found adhering to the glass plate.

[0063] Therefore, the processing apparatus 16 of this embodiment excludes candidate defects caused by dirt and dust from the image data during the initial processing stage, reducing the processing load in subsequent processing stages. Furthermore, through subsequent processing, the processing apparatus 16 of this embodiment identifies whether small, black-state defects, white-displaying defects, and dust and dirt missed in the initial processing stage are defects or not. Then, if a defect is identified, the processing apparatus 16 of this embodiment identifies the defect category. Finally, based on the identification result, the processing apparatus 16 of this embodiment detects the portion identified as defective as a defective area.

[0064] Next, use Figure 2 The processing device 16 will be described. Figure 2 This is a diagram illustrating the schematic structure of the processing apparatus 16 involved in this embodiment.

[0065] like Figure 2 As shown, the processing device 16 is connected to a shooting device (a first camera 22 and a second camera 30) and a display 32.

[0066] The processing device 16 includes an image acquisition unit 100, a search unit 101, a storage unit 102, a recognition unit 103, a storage unit 104, a determination unit 105, and a database unit 106.

[0067] The image acquisition unit 100 acquires first image data captured by the imaging device (first camera 22 and second camera 30). If the acquired image data is an analog signal, it converts it into a digital signal and outputs the converted first image data a to the search unit 101. If the image data captured by the imaging device is a digital signal, the acquired image data is directly output to the search unit 101 as first image data a.

[0068] The search unit 101 scans the entire first image data a output from the image acquisition unit 100 for each pixel. The search unit 101 compares the brightness of each scanned pixel with a predetermined first threshold. The search unit 101 extracts pixels with brightness higher than the predetermined first threshold or pixels with brightness lower than the predetermined first threshold as first defect candidates, and stores the coordinate values ​​of the extracted first defect candidate pixels in the storage unit 102.

[0069] Search unit 101 calculates a first feature value of a portion (first defect candidate) including the pixels of the extracted first defect candidate. Search unit 101 compares the calculated first feature value with a predetermined first determination criterion to classify the defect category. If the defect is determined to be non-defect (suspected defect), search unit 101 excludes the first defect candidate from the defect candidates. Based on the classification result, search unit 101 outputs information representing the defect category, the coordinate values ​​of the pixels of the first defect candidate, and the first feature value of the first defect candidate to determination unit 105. Figure 2 (c). In addition, there may be multiple first feature quantities calculated. In this case, the search unit 101 compares the multiple calculated first feature quantities with a predetermined first determination condition for each feature quantity to classify the defect category.

[0070] Search unit 101 generates second image data containing pixels of first defect candidates that were not excluded from the defect candidates. Furthermore, search unit 101 scans the generated second image data for each pixel. Search unit 101 compares the brightness of the scanned pixels with a predetermined second threshold. Search unit 101 extracts pixels with brightness higher than the predetermined second threshold or pixels with brightness lower than the predetermined second threshold as second defect candidates, and stores the coordinate values ​​of the extracted second defect candidate pixels in storage unit 102. Search unit 101 outputs the second image data and the scan results to recognition unit 103. Figure 2 (b).

[0071] The storage unit 102 stores in association the coordinate values ​​of the first and second defect candidates, the first feature value of the first defect candidate, information indicating the defect category of the first defect candidate, and information indicating whether to exclude the coordinates of the first defect candidate from the defects. Additionally, the storage unit 102 stores first image data.

[0072] The recognition unit 103 uses the second image data output by the search unit 101 and the scanning results to calculate a second feature quantity. The second feature quantity calculated here is the shape (pattern), brightness, area, etc. of the defect.

[0073] Based on the calculated second feature quantity, the identification unit 103 identifies and classifies the defect categories in detail. The identification unit 103 writes the second image data containing second defect candidates classified as defects into the database unit 106 and updates it. Figure 2 (e). Additionally, based on the identification result, the identification unit 103 outputs information indicating the defect category, representing a detailed classification related to the second defect candidate, to the determination unit 105. Figure 2 d).

[0074] Storage unit 104 stores the second feature quantity in association with the second image data.

[0075] The determination unit 105 uses information representing the defect category of the first defect candidate output by the search unit 101, the coordinate values ​​of the pixels of the first and second defect candidates, and information representing the detailed classification of the defect category related to the second defect candidate output by the recognition unit 103 to determine good and defective areas in the determination area of ​​the glass plate. Based on the determination result, the determination unit 105 writes and updates the information representing the defect category, the coordinate values ​​of the defect, and the feature values ​​of the defect (first and second feature values) in association with the second image data of the defect area stored in the database unit 106. Figure 2 f).

[0076] The database unit 106 establishes and saves information representing the defect category, the coordinate value of the defect, the feature quantity of the defect (second feature quantity), and the second image data of the defect area in a related manner.

[0077] Next, use Figure 3 and Figure 4 The defect candidate search processing performed by the search unit 101 on the entire image will be described. Figure 3 This diagram illustrates the steps of the defect candidate search processing for the entire image involved in this embodiment. Figure 4 This is a diagram illustrating the image data being searched in this embodiment.

[0078] exist Figure 4In this image, the flow direction of the glass (also called the travel direction) is defined as the x-direction, and the width direction of the glass is defined as the z-direction. Therefore, the glass flows in the positive x-direction as indicated by arrow g21. Label g1 represents the first image data a (converted to a digital signal if the acquired image data is an analog signal) captured by the imaging device (first camera 22, second camera 30), for example, pixel A2 in the x-direction and pixel A1 in the z-direction. Label g2 represents a pixel. Labels g11 to g18 represent defect candidate pixels (hereinafter also referred to as "defect pixels"). Label g10-1 represents the second image data including defect pixels g11 and g12. Label g10-2 represents the second image data including defect pixel g13. Label g10-3 represents the second image data containing defect pixels g14 to g18.

[0079] (Step S1) The search unit 101 uses the first image data a (which is converted into a digital signal if the acquired image data is an analog signal) captured by the imaging device (first camera 22, second camera 30) to scan the entire image data g1 for each pixel (g2). After step S1 is completed, step S2 is performed.

[0080] (Step S2) After scanning one pixel, the search unit 101 compares the brightness of the scanned pixel with a predetermined first threshold. After step S2, the process proceeds to step S3.

[0081] (Step S3) Based on the comparison results, if the brightness of the scanned pixel is greater than or less than a predetermined threshold, the search unit 101 sets that pixel as a first defect candidate pixel (in addition, the threshold for the former and the threshold for the latter may be different). The search unit 101 stores the coordinate values ​​of the first defect candidate pixels in the storage unit 102 in a table form, for example. After step S3 is completed, step S4 is performed.

[0082] (Step S4) Based on the coordinate values ​​stored in the storage unit 102, the search unit 101 determines a set of pixels that are close to the first defect candidate as a first defect candidate. The set of pixels that are close to the first defect candidate is, for example, in... Figure 4 The middle pixels are g11 and g12.

[0083] Next, the search unit 101 calculates image feature quantities (first feature quantities) of the region (first defect candidate) that serves as the first defect candidate. Image feature quantities include, for example, brightness, area of ​​the region, brightness integral, signal level, brightness and darkness information, width and length, transmission length, aspect ratio, density, etc.

[0084] Furthermore, the luminance integral refers to the integral value of the luminance of each pixel included in the first defect candidate. After step S4, proceed to step S5.

[0085] (Step S5) The search unit 101 requests defect category discrimination condition information (first discrimination condition) from the determination unit 105 and obtains the category discrimination condition information. Furthermore, the category discrimination condition information is used to determine the category of a defect, including the feature quantities of each defect category and the defect classification processing steps. The search unit 101 compares the category discrimination condition information with the first feature quantity of the portion that was selected as the first defect candidate in step S4, and determines the defect category of the portion that was selected as the first defect candidate. Defect categories include bubbles, foreign objects, dust, dirt, etc. The category discrimination condition information includes, for example, size, brightness, the size of the circumscribed circle, and information indicating the density of defects. Additionally, the category discrimination condition information sets size, brightness, the size of the circumscribed circle, and information indicating the density of defects for the bright areas, dark areas, and the overall image data, respectively. Furthermore, for... Figure 1 Each of the shown shooting devices (first camera 22, second camera 30) has these category discrimination condition information set.

[0086] After step S5 is completed, proceed to step S6.

[0087] (Step S6) If the search unit 101 determines that the first defect candidate belongs to a predetermined defect category, it excludes the first defect candidate belonging to the predetermined defect category from the defect candidate list. The predetermined defect category is, for example, non-defects such as dust or dirt adhering to the surface of the glass plate. Dust and dirt adhering to the surface of the glass plate can be removed by cleaning in a subsequent process, and therefore are excluded from the defect candidate list.

[0088] Next, the search unit 101 adds exclusion information to the coordinate values ​​of the pixels included in the first defect candidates that are determined to be dust or dirt, stored in the storage unit 102. The search unit 101 performs the processing of steps S5 and S6 on all the first defect candidates.

[0089] Next, based on the discrimination result, the search unit 101 outputs various information c (information representing the defect category, the coordinate value of the defect, and the calculated feature quantity) to the determination unit 105.

[0090] Next, use Figure 5 and Figure 6 The defect image generation process performed by the search unit 101 will be explained. Figure 5 This diagram illustrates the steps involved in the defect image generation and processing described in this embodiment. Figure 6 This is a diagram illustrating an example of defective image data (second image data) involved in this embodiment. Figure 6 In this context, the direction of glass flow is defined as the x-direction, and the width direction of the glass is defined as the z-direction. Figure 6 In the image data g10-1, there is a sequence of images containing g10-1.Figure 4 The image data g1 contains defective pixels g11 and g12, and the second image data g10-1, for example, in the x-direction A. 11 Pixel, z-direction is A 12 Pixel.

[0091] (Step S11) The search unit 101 sequentially refers to the list of coordinate values ​​of the first defect candidates stored in the storage unit 102. In this case, the search unit 101 does not refer to the coordinate values ​​for which exclusion information was added in step S6. After step S11 is completed, the process proceeds to step S12.

[0092] (Step S12) as Figure 6 As shown, the search unit 101 generates a region (A) including defective pixels g11 and g12 from the overall first image data g1. 11 Pixels × A 12 The second image data g10-1 (pixels). In this case, such as Figure 4 As shown, the second image data g10-1 is generated with the first defect candidate pixel positioned at the center of the generated second image data. However, in the case of image data such as image data g10-2, where there are only two pixels in the negative z-direction relative to the defect pixel g13, i.e., when it is difficult to generate second image data centered on the defect pixel, second image data is generated where the defect pixel is as close to the center as possible. After step S12, the process proceeds to step S13.

[0093] (Step S13) The search unit 101 repeatedly performs steps S11 and S12 on the list of coordinate values ​​of all first defect candidates without attached exclusion information stored in the storage unit 102 to generate second image data.

[0094] Next, the search unit 101 stores the generated second image data in the storage unit 102.

[0095] This concludes the defect image generation process.

[0096] Next, use Figure 7 The first defect recognition process for image data of defect candidates performed by the search unit 101 and the recognition unit 103 will be described. Figure 7 This is a diagram illustrating the steps of the first defect identification process involved in this embodiment.

[0097] (Step S21) The search unit 101 sequentially selects one image data from the second image data stored in the storage unit 102 and scans it pixel by pixel. After step S21 is completed, the process proceeds to step S22. The difference between step S1 and S21 is that in step S1, the entire first image data is scanned, but in step S21, the second image data is scanned.

[0098] (Step S22) After scanning each pixel, the search unit 101 compares the brightness of the scanned pixel with a predetermined second threshold. The second threshold used in step S22 may be the same as or different from the first threshold used in step S2. After step S22, the process proceeds to step S23.

[0099] (Step S23) Based on the comparison results, if the brightness of the scanned pixel is greater than or less than a predetermined threshold, the search unit 101 sets it as a second defect candidate (in addition, the threshold for the former and the threshold for the latter may be different). The search unit 101 repeatedly compares the pixels contained in the second image data with the second threshold.

[0100] Next, the search unit 101 repeatedly performs steps S21 to S23 on all image data output by the search unit 101. The search unit 101 outputs second image data b containing the second defect candidate to the recognition unit 103. After step S23 is completed, step S24 is performed.

[0101] (Step S24) The recognition unit 103 calculates various image feature quantities (second feature quantities) on the second image data b output by the search unit 101. For example, the recognition unit 103 calculates the shape of the defect (circle, line, etc.), which region of the image data the high-brightness part belongs to (upward or downward in the y-axis direction from the center), the area of ​​the high-brightness range, whether it is continuous, etc. as image feature quantities, and stores the calculated second feature quantities in association with the second image data in the storage unit 104.

[0102] This concludes the first defect identification process.

[0103] As described above, the search unit 101 extracts a first defect candidate (the portion containing the pixels of the first defect candidate) from the captured first image data as a whole, and calculates a first feature quantity of the extracted first defect candidate. The search unit 101 classifies the calculated first feature quantity into a defect category using predetermined first determination conditions. If the defect candidate is a non-defect (suspected defect) based on dust or dirt, and is set as an exclusion object, the search unit 101 excludes the first defect candidate from the defect candidates. Based on the classification result, the search unit 101 outputs information indicating the defect category associated with the first defect candidate, the coordinate values ​​of the pixels of the defect candidate, and the first feature quantity of the defect portion to the determination unit 105. Furthermore, there may be multiple calculated first feature quantities. In this case, the search unit 101 uses predetermined first determination conditions to classify the multiple calculated first feature quantities into a defect category for each feature quantity.

[0104] The search unit 101 generates second image data, which includes pixels that are defect candidates that were not excluded from the first defect candidates. Furthermore, the search unit 101 extracts second defect candidates from the generated second image data and outputs second image data b, including the second defect candidates, to the recognition unit 103.

[0105] The recognition unit 103 calculates various image feature quantities (second feature quantities) on the second image data b output by the search unit 101.

[0106] Next, use Figure 8 The second defect recognition process performed by the recognition unit 103 on the second image data of the second defect candidate will be described. Figure 8 This is a diagram illustrating the steps of the second defect identification process involved in this embodiment.

[0107] (Step S31) The recognition unit 103 sequentially reads out the second feature quantity of the second image data containing the second defect candidate stored in the storage unit 104. After step S31 is completed, the process proceeds to step S32.

[0108] (Step S32) The identification unit 103 identifies and classifies the defect category of the second defect candidate in detail through the defect classification process described later.

[0109] After classification, the identification unit 103 outputs information d representing the defect category of the second defect candidate to the determination unit 105 based on the classification result. Then, the identification unit 103 writes the second image data e containing the classified second defect candidate to the database unit 106 to update the database.

[0110] The identification unit 103 repeatedly performs steps S31 and S32 to classify defects based on the second feature values ​​of all second image data containing second defect candidates stored in the storage unit 104. Through this classification process, the identification unit 103 classifies defect candidates that could not be classified by the search unit 101 as defects caused by dust, foreign objects, etc., determines whether they are non-defects or defects based on the classified defect category, and excludes them from the list. Furthermore, the same process is performed on defect candidates classified by the search unit 101, resulting in a reclassification process.

[0111] This concludes the second defect identification process.

[0112] As described above, the recognition unit 103 calculates a second feature value for the second image data including second defect candidates generated by the search unit 101. Furthermore, based on the calculated second feature value, the recognition unit 103 determines in detail the defect category of the second defect candidate.

[0113] Here, the defect classification process performed by the identification unit 103 will be explained.

[0114] The defect classification processing in the identification unit 103 determines whether there are any areas (brighter areas) in the second image data of the second defect candidate that are whiter than the background. If the identification unit 103 determines that there are areas whiter than the background, it determines that the second image data of the second defect candidate contains a defect. The defect category is, for example, an air bubble or a foreign object. On the other hand, if the identification unit 103 determines that there are no areas whiter than the background, it determines that the second image data of the second defect candidate does not contain a defect, that is, the defect candidate is a non-defect.

[0115] Next, use Figure 9 and Figure 10 The good or bad determination process performed by the determination unit 105 is explained.

[0116] Figure 9 This diagram illustrates the steps involved in the good or bad determination process of this embodiment. Figure 10 This is a diagram illustrating an example of image data of the determination area that is judged as good or bad according to this embodiment.

[0117] exist Figure 10 In this context, the width direction of the glass plate G is defined as the z-direction, and the flow direction of the glass plate G is defined as the x-direction. Figure 10 In the diagram, g201 indicates the area to be judged, g202 indicates the part judged as defective, and g203 indicates the area judged as good.

[0118] (Step S201) The determination unit 105 uses the information c (information representing the defect category, coordinate value, and first feature quantity of the first defect candidate) output by the search unit 101 and the information d (information representing the defect category of the second defect candidate) output by the recognition unit 103 to determine whether each first defect candidate is good or bad for the determination region (usually the entire region of the first image data a). In addition, the good or bad determination comparison threshold for each first defect candidate can be arbitrarily changed.

[0119] (Step S202) The determination unit 105 determines the areas where no first defect candidate was detected and the areas where the first defect candidate was detected but were determined to be dirt or dust attached to the surface of the glass plate G (areas that do not contain defects) as good product areas.

[0120] (Step S203) If the first defect candidate is determined to be a bubble or foreign object generated inside or on the surface of the glass plate G, and if it is determined to be the size of a defective product, the determination unit 105 determines the area containing the first defect candidate (the area containing the defect) as a defective area.

[0121] The result, such as Figure 10As shown, the determination unit 105 determines the area g202 containing defects as a defective area and the area g203 as a good area.

[0122] Next, the determination unit 105 writes the defect information (defect type, coordinate value, first and second feature quantities, etc.) of the area determined to be a defective area into the database unit 106 and updates it.

[0123] The above concludes the determination of whether the assessment is satisfactory or not.

[0124] Figure 11 This is a diagram illustrating the determination result when using the defect inspection device 10 according to this embodiment.

[0125] exist Figure 11 In this diagram, row r1 represents the processing result of the defect candidate search process (steps S1 to S6), row r2 represents the processing result of the defect image generation process (steps S11 to S13), row r3 represents the processing result of the first defect identification process (steps S21 to S24), and row r4 represents the processing result of the second defect identification process (steps S31 to S32). Additionally, column c1 represents the processing result when the defect is dirt, column c2 represents the processing result when the defect is dust, column c3 represents the processing result when the defect is a bubble, and column c4 represents the processing result when the defect is a foreign object.

[0126] In this embodiment, the search unit 101 determines the defect category of the defect candidate (first defect candidate) extracted from the entire scanned (searched) image data. As in case 2 of columns c1 to c4, the defect category of the first defect candidate can be determined in the defect candidate search process of row r1. If it is determined to be a predetermined defect category, no further determination processing is required for the first defect candidate, thereby reducing the amount of computation.

[0127] In addition, as in Case 1 of columns c1 to c4, if the defect category cannot be determined in the defect candidate search process of row r1, the defect inspection device 10 according to this embodiment can determine the defect category through the second defect identification process of row r4 after the defect image generation process of row r2 and the first defect identification process of row r3.

[0128] In this way, by scanning the entire first image data, the exploration unit 101 can eliminate dirt and dust that are not defects (suspected defects) from the defect candidates in the pixels that are first defect candidates, thereby greatly reducing the processing burden of the identification unit 103 and the determination unit 105.

[0129] As described above, in the defect inspection apparatus 10 of this embodiment, the search unit 101, which performs initial stage processing, extracts first defect candidates from the first image data. Then, the search unit 101 calculates the feature quantity of the extracted first defect candidates and determines the defect category based on the calculation result. For first defect candidates that can be determined at this stage, if they are set as exclusion targets, second image data is not generated, that is, second image data is not sent to the recognition unit 103. Especially when inspection is carried out before cutting, before cleaning, or outside a cleanroom, in the case of prior art defect inspection apparatuses, a large number of first defect candidates are extracted because a large amount of dirt adheres to the glass plate. On the other hand, in the defect inspection apparatus 10 of this embodiment, the search unit 101 can extract and exclude these multiple non-defect (suspected defect) defect candidates such as dirt and dust adhering to the surface of the glass plate during the initial stage processing. As a result, the processing apparatus 16 of this embodiment can reduce the amount of data output to the recognition unit 103, and thus reduce the amount of computation performed by the recognition unit 103.

[0130] Furthermore, in the defect inspection apparatus 10 according to this embodiment, the search unit 101 extracts second image data containing pixels of first defect candidates that were not identified in the initial stage processing. Then, the recognition unit 103 calculates a second feature value based on the second image data, identifies the defect category in detail, and performs classification. That is, for second defect candidates that were not identified by the search unit 101, the recognition unit 103 also performs classification, determining whether they are defects or not. Then, the determination unit 105 determines the location of pixels identified as defects by the search unit 101 and the recognition unit 103 as a defective area.

[0131] [Second Implementation]

[0132] Figure 12 This is a diagram illustrating the schematic structure of the processing device 16a included in the defect inspection device 10a according to the second embodiment.

[0133] like Figure 12 As shown, the processing device 16a is connected to a shooting device (first camera 22, second camera 30) and a glass plate cutting control device 200.

[0134] The processing device 16a includes an image acquisition unit 100, a search unit 101, a storage unit 102, a recognition unit 103, a storage unit 104, a determination unit 105a, a database unit 106, and a size determination unit 107. The difference from the processing device 16 in the first embodiment is the determination unit 105a and the size determination unit 107.

[0135] Similar to the first embodiment, the determination unit 105a determines defective areas and good areas within the determination area. Based on the determination result, the determination unit 105a outputs the coordinate values ​​of the defective areas to the size determination unit 107.

[0136] The size determination unit 107 is based on the coordinate values ​​of the defective area output by the determination unit 105a, such as... Figure 13 As shown, the instruction to cut the glass plate G by avoiding the defective area is output to the glass plate cutting control device 200.

[0137] Figure 13 This diagram illustrates an example of cutting a glass plate while avoiding undesirable areas, as described in this embodiment.

[0138] exist Figure 13 In this context, the direction of flow in the glass plate G is defined as the x-direction, and the width direction of the glass plate G is defined as the z-direction. Figure 13 In the diagram, the designation g301 (g301-1, g301-31, g301-32, and g301-51) indicates a defective area containing defects, and the designation g311 indicates a good area (g311-12, g311-21, g311-22, g311-41, g311-42, and g311-51) and a defective area (g311-11, g311-3, and g311-52).

[0139] In addition, Figure 13 In the diagram, L1 to L5 indicate the cut-out positions of the glass plate G in the x-direction. Furthermore, the widths of positions L1, L2, L4, and L5 in the x-axis direction are equal, while the width of position L3 in the x-axis direction is, for example, shorter than the width of position L1 in the x-axis direction.

[0140] like Figure 13 As shown, at position L1, the size determination unit 107 determines that the glass plate g311-11 contains a defective area g301-1. Next, at position L1, the size determination unit 107 determines that the glass plate g311-12 can be cut out by avoiding the defective area g301-1, and generates an instruction to cut out the glass plate g311-12.

[0141] At position L2, the size determination unit 107 determines that there is no defective area and generates an instruction to cut out glass plates g311-21 and g311-22.

[0142] At position L3, the size determination unit 107 determines that it is impossible to cut the glass plate without avoiding the defective areas g301-31 and g301-32. Therefore, the size determination unit 107 generates an instruction not to cut out the area containing the defective areas g301-31 and g301-32.

[0143] At position L4, the size determination unit 107 determines that there is no defective area and generates instructions to cut out glass plates g311-41 and g311-42.

[0144] At position L5, the size determination unit 107 determines that the glass plate g311-52 contains a defective area g301-51. Next, at position L5, the size determination unit 107 determines that the glass plate g311-51 can be cut out by avoiding the defective area g301-51, and generates an instruction to cut out the glass plate g311-51.

[0145] As described above, similarly to the first embodiment, the determination unit 105a detects defective areas before cutting. Then, the size determination unit 107 generates an instruction to efficiently cut the glass plate G while avoiding defective areas, based on the determination result of the determination unit 105a. As a result, the glass plate cutting control device 200 is able to efficiently cut the glass plate while avoiding defective areas.

[0146] In addition, the glass plate cutting control device 200 can also be used for... Figure 13 The glass plate containing defective areas, such as g311-11, as described in the text, is cut to a smaller size than glass plate g311-12, avoiding the defective area g301-1.

[0147] Furthermore, the defect inspection devices 10 and 10a of the first and second embodiments can be used, for example, in the manufacturing processes of glass plates for FPDs (flat panel displays) such as liquid crystal displays, plasma displays, and organic EL (electroluminescent) displays, as well as glass for building materials and automotive applications. In this case, such as Figure 13 As explained, defective areas can be detected before cutting, and glass sheets can be manufactured efficiently by avoiding detected defective areas.

[0148] In addition, the implementation method can also be used to implement the implementation method. Figure 1 and Figure 2 Defect inspection device 10 or Figure 12 The program for the function of the defect inspection device 10a is recorded on a computer-readable recording medium, which is then read and executed by the computer system to perform processing at each stage. Furthermore, the term "computer system" as used here includes hardware such as the operating system and peripheral devices.

[0149] In addition, if the system is a WWW system, then "computer system" also includes the homepage providing environment (or display environment).

[0150] Furthermore, "computer-readable recording media" refers to removable media such as floppy disks, optical disks, ROM (Read Only Memory), and CD-ROMs, as well as storage devices such as USB storage devices connected via USB (Universal Serial Bus) I / F (interface), and hard disks built into computer systems. Moreover, "computer-readable recording media" also includes recording media that retain programs for a certain period of time, such as volatile memory within a computer system acting as a server or client. Additionally, the aforementioned programs can be used to implement a portion of the functions described above, and can also achieve the aforementioned functions through combination with programs already recorded in the computer system.

[0151] Although the present invention has been described in detail with reference to specific embodiments, those skilled in the art can make various changes and modifications without departing from the spirit and scope of the invention.

[0152] This application is based on Japanese Patent Application No. 2024-124689, filed on July 31, 2024, and Japanese Patent Application No. 2025-121592, filed on July 18, 2025, the contents of which are incorporated herein by reference.

[0153] Marker description

[0154] 10···Defect inspection device, 12···First defect inspection device, 14···Second defect inspection device, 16···Processing device, 20···First linear light source, 22···First camera, 24···Light path shielding component, 28···Second light source, 30···Second camera, 32···Display, 100···Image acquisition unit, 101···Search unit, 102, 104···Storage unit, 103···Identification unit, 105···Determination unit, 106···Database unit, 107···Size determination unit.

Claims

1. A defect inspection method in a defect inspection device, wherein, include: The first defect candidate extraction step involves extracting a first defect candidate for the plate-like body from the first image data obtained by the imaging device from the first image data of the plate-like body with transparency. The first feature calculation step calculates the first feature of the first defect candidate. as well as The defect category determination process determines whether the first defect candidate belongs to a predetermined defect category based on the first feature quantity.

2. The defect inspection method according to claim 1, wherein, The defect inspection method further includes the following steps when the first defect candidate is determined not to be a predetermined defect category in the defect category determination step: The process of extracting second image data of a predetermined size from the first image data and generating second image data containing pixels of the first defect candidate; The second defect candidate extraction process for extracting second defect candidates from the second image data; The second feature quantity calculation step for calculating the second feature quantity of the second defect candidate; as well as The process of determining the defect category of the second defect candidate based on the second feature quantity.

3. The defect inspection method according to claim 1, wherein, The first feature quantity is a feature quantity that can be used to determine at least one of the dust and dirt adhering to the surface of the plate-like body.

4. The defect inspection method according to claim 1, wherein, The first defect candidate extraction process searches the first image data for each pixel and extracts the first defect candidate.

5. The defect inspection method according to any one of claims 1 to 4, wherein, The plate-like body is any one of the following: a glass strip formed by continuously flowing molten glass on a molten metal bath surface; a float glass substrate formed by moving the glass strip along the molten metal bath surface to form a plate; a glass substrate for flat panel displays; a glass plate for building materials; and a glass plate for automobiles.

6. A defect inspection device, wherein, The defect inspection device includes a search unit that extracts a first defect candidate for the plate-shaped body from the first image data obtained by the imaging device from the first image data of the plate-shaped body with transparency, calculates a first feature value of the first defect candidate, and compares the first feature value with a predetermined threshold to determine whether the first defect candidate is a predetermined defect category.

7. The defect inspection device according to claim 6, wherein, The defect inspection device also includes an identification unit, which calculates a second characteristic quantity and determines the defect category. If the search unit determines that the first defect candidate is not the predetermined defect category, it extracts and generates second image data of a predetermined size containing the pixels of the first defect candidate from the first image data. The search unit extracts second defect candidates from the second image data. The identification unit calculates the second feature quantity of the second defect candidate, and The identification unit determines the defect category of the second defect candidate based on the second feature quantity.

8. A method for manufacturing a transparent plate-like body, wherein, include: The first defect candidate extraction step involves extracting a first defect candidate for the plate-like body from the first image data obtained by the imaging device from the first image data of the plate-like body with transparency. The first feature calculation step calculates the first feature of the first defect candidate. as well as The defect category determination process determines whether the first defect candidate belongs to a predetermined defect category based on the first feature quantity.

Images