Classification device, classification method, program, and learning completion model

Abstract

The invention relates to a classification device, a classification method, a program and a learned model, and provides a classification device which performs classification related to a falling position of a foreign matter relative to a glass plate on the basis of a shape (for example, a concave-convex shape) generated on the surface of the glass plate. A sorting device is provided with a determination unit that inputs image data obtained by capturing an image of a glass plate manufactured into a plate shape in a predetermined flow direction into a learning-completed model for machine learning, and acquires an output result from the learning-completed model. And a determination result relating to the position of the foreign matter falling onto the glass plate in the flow direction on the basis of the shape of the foreign matter on the glass plate.

Description

Technical Field

[0001] The present disclosure relates to a classification device, a classification method, a program, and a learned model. Background Art

[0002] The glass sheet is manufactured by a float process, etc. In the float process, for example, molten glass continuously supplied onto molten tin in a bath is made to flow on the molten tin and formed into a ribbon.

[0003] Patent Document 1 describes an inspection device for detecting defects in optical materials, where the color and shape become irregular (see Patent Document 1).

[0004] Prior art literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Laid-Open No. 2022-56389 Summary of the Invention

[0007] Problems to be solved by the invention

[0008] However, in the conventional technology, there are cases where it is not possible to sufficiently determine the position where foreign matter has fallen onto the glass sheet during the manufacturing process of the glass sheet.

[0009] The present disclosure is completed in consideration of such a situation, and its subject is to provide a classification device, classification method, program and learned model for classifying foreign matter in relation to the falling position of the glass plate based on the shape generated on the surface of the glass plate (for example, concave and convex shapes).

[0010] Technical solutions to problems

[0011] One embodiment of the present disclosure is a classification device comprising a determination unit that inputs image data obtained by photographing a glass plate manufactured in a plate shape in a specified flow direction into a learned model of machine learning, and obtains an output result from the learned model as a determination result based on the shape caused by a foreign object on the glass plate and related to the position of the foreign object falling onto the glass plate in the flow direction.

[0012] One embodiment of the present disclosure is a classification method, in which a classification device inputs image data obtained by photographing a glass plate manufactured into a plate shape in a specified flow direction into a learned model of machine learning, and obtains an output result from the learned model as a judgment result based on the shape caused by the foreign matter on the glass plate and related to the position of the foreign matter falling onto the glass plate in the flow direction.

[0013] One embodiment of the present disclosure is a program for causing a computer to implement the following functions: a function of inputting image data obtained by photographing a glass plate manufactured into a plate shape in a specified flow direction into a learned model of machine learning; and a function of obtaining an output result from the learned model as a judgment result related to the position of the foreign matter falling onto the glass plate in the flow direction based on the shape caused by the foreign matter on the glass plate.

[0014] One method disclosed herein is a machine learning learned model that inputs image data obtained by photographing a glass plate manufactured in a plate shape in a specified flow direction into the machine learning learned model, and outputs judgment result data based on the shape caused by foreign matter on the glass plate and related to the position of the foreign matter falling onto the glass plate in the flow direction.

[0015] Effects of the Invention

[0016] According to the information processing device, classification device, classification method, program and learned model involved in the present disclosure, classification related to the falling position of foreign matter relative to the glass plate is performed based on the shape (e.g., concave and convex shapes) generated on the surface of the glass plate. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a diagram showing a schematic configuration example of a glass manufacturing apparatus according to an embodiment.

[0018] Figure 2 It is a cross-sectional view showing a structural example of a float kiln according to an embodiment.

[0019] Figure 3 It is a diagram showing an example of the arrangement of imaging units according to the embodiment.

[0020] Figure 4 It is a diagram showing a configuration example of a classification device according to an embodiment.

[0021] Figure 5 (A) to (I) schematically illustrate examples of images when a foreign object falls upstream (low viscosity) according to the embodiment.

[0022] Figure 6 (A) to (I) schematically illustrate examples of images when a foreign object falls in the middle (medium viscosity) according to the embodiment.

[0023] Figure 7 (A) to (I) schematically illustrate examples of images when a foreign object falls downstream (high viscosity) according to the embodiment.

[0024] Figure 8A This is a diagram showing an example of the shape of a defect in the height direction when foreign matter has fallen upstream (low viscosity) according to the embodiment.

[0025] Figure 8B This is a diagram showing an example of the shape of a defect in the height direction when a foreign object falls in the middle stream (medium viscosity) according to the embodiment.

[0026] Figure 8C This is a diagram showing an example of the shape of a defect in the height direction when foreign matter falls downstream (high viscosity) according to the embodiment. DETAILED DESCRIPTION

[0027] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings.

[0028] [Glass manufacturing equipment]

[0029] Figure 1 It is a figure which shows the schematic structural example of the glass manufacturing apparatus 1 which concerns on embodiment.

[0030] exist Figure 1 , for convenience of explanation, an XYZ orthogonal coordinate system is shown as a three-dimensional orthogonal coordinate system. The glass manufacturing apparatus 1 includes a melting furnace 11 , a float furnace 12 , an annealing furnace 13 , and an inspection / cutting unit 14 .

[0031] In the melting kiln 11, a melting process is performed.

[0032] In the melting step, glass raw materials prepared by mixing various raw materials are melted to obtain molten glass. After the glass raw materials are put into the melting furnace 11, they are melted by heating means such as radiant heat from the flame ejected from the burner or electric melting to form molten glass.

[0033] In the float kiln 12, the forming process is performed.

[0034] In the forming process, the molten glass obtained in the melting process is continuously supplied to the molten tin in the bath, where it flows and forms a glass sheet (the so-called glass ribbon) as sheet glass. The glass sheet is cooled while flowing in a predetermined direction and is pulled up from the molten tin.

[0035] In the annealing furnace 13, an annealing process is performed.

[0036] In the annealing step, the glass sheet obtained in the forming step is annealed inside the annealing furnace 13. Inside the annealing furnace 13, the glass sheet is annealed while being horizontally conveyed on rollers from the entrance to the exit of the annealing furnace 13.

[0037] In the inspection / cutting section 14 , an inspection process and a cutting process are performed.

[0038] In the inspection step, the glass sheet annealed in the annealing step is inspected.

[0039] In the cutting process, the glass sheet, annealed in the annealing process, is cut into predetermined dimensions using a cutting machine. During the cutting process, the widthwise edges of the glass sheet (the so-called ears) are removed. These edges are removed because they are thickened due to surface tension and other factors.

[0040] After the cutting process, the glass sheets are shipped out of the factory.

[0041] [Floating Kiln]

[0042] Figure 2 It is a cross-sectional view showing a structural example of the float kiln 12 according to the embodiment.

[0043] Figure 2 Shown with Figure 1 The same situation applies to the XYZ orthogonal coordinate system.

[0044] Figure 2 The cross section shown is a side cross section of the float kiln 12 when viewed from the negative side of the X-axis to the positive side.

[0045] In this embodiment, for the sake of convenience, the upper surface (the surface on the positive side of the Z axis) of the molten glass G (glass plate) in the float furnace 12 is described as the front surface, and the lower surface (the surface on the negative side of the Z axis) is described as the back surface, but these surfaces may also be referred to by other names.

[0046] The float kiln 12 forms molten glass G by flowing on the molten tin M continuously supplied to the molten tin M in the bath 22. The molten glass G is supplied onto the molten tin M near the inlet 21 of the float kiln 12, cooled while flowing in a predetermined direction, and pulled up from the molten tin M near the outlet 21 of the float kiln 12.

[0047] The float kiln 12 is composed of a bath 22 for containing molten tin M, a side wall 24 provided along the upper edge of the outer periphery of the bath 22, and a ceiling 26 connected to the side wall 24 and covering the upper part of the bath 22. A gas supply path 30 is provided in the ceiling 26 for supplying a reducing gas to a space 28 formed between the bath 22 and the ceiling 26. A heater 32 serving as a heat source is inserted into the gas supply path 30.

[0048] To prevent oxidation of the molten tin M, a gas supply line 30 supplies a reducing gas to the space 28 within the float kiln 12. The reducing gas may contain, for example, 1 to 15% by volume of hydrogen and 85 to 99% by volume of nitrogen. To prevent atmospheric air from entering through gaps between bricks forming the sidewalls 24, the space 28 within the float kiln 12 is set to a pressure higher than atmospheric pressure.

[0049] In order to adjust the temperature distribution within the float furnace 12, a plurality of heaters 32 are provided, for example, at intervals in the flow direction (Y direction) and the width direction (X direction) of the molten glass G. The output of the heaters 32 is controlled so that the temperature of the molten glass G decreases as it moves from the inlet 21 toward the outlet 23 of the float furnace 12. The output of the heaters 32 is controlled so that the thickness of the molten glass G is uniform in the width direction (X direction).

[0050] Bathtub 22 consists of a box-shaped metal shell 34 with an open top, and bottom bricks 36 and side bricks 38 disposed within metal shell 34. Metal shell 34 prevents air from entering bathtub 22 from the sides or below. Multiple bottom bricks 36 are arranged two-dimensionally, with small gaps between them that prevent contact due to thermal expansion. Multiple bottom bricks 36 are surrounded by multiple side bricks 38 arranged in a ring.

[0051] exist Figure 2 Also shown is the upper surface 36a of the bottom brick 36.

[0052] <Example of Configuration of Imaging Unit>

[0053] Figure 3 It is a diagram showing an example of the arrangement of imaging units according to the embodiment.

[0054] exist Figure 3 For the sake of convenience, the Figure 1 The same situation applies to the XY coordinates in the XYZ orthogonal coordinate system.

[0055] Figure 3 The following diagram schematically shows the appearance of the molten glass G when the float furnace 12 is viewed from above (positive side of the Z axis). Figure 3 The glass width region E showing the width of the molten glass G in the float furnace 12 is shown. The glass width region E in the flow direction shows a change in the glass width in the flow direction.

[0056] Tensile forces F1 to F4 directed outward in the width direction (outward on both sides) are applied to the molten glass G. As a result, the width of the glass width region E gradually increases from the upstream toward the downstream of the flow.

[0057] As an example, an example in which the width gradually increases is shown, but the arrangement from upstream to downstream of the flow is important, and the shape is not limited to this.

[0058] Figure 3 The defect inspection unit 311 is schematically shown (in Figure 1 The middle is the inspection part of the inspection / cutting part 14).

[0059] It should be noted that a defect may also be referred to as a shortcoming, for example.

[0060] In the present embodiment, during the manufacturing process of the glass plate, foreign matter dropped from above adheres to the surface (upper surface) of the glass plate, and the shape of the defect including the position and surrounding area of the foreign matter changes over time.

[0061] exist Figure 3 In the example of FIG. 3 , the inspection unit 311 includes three imaging units C1 to C3 arranged in the width direction.

[0062] The imaging unit C1 captures an image of the inspection region R1. The imaging unit C2 captures an image of the inspection region R2. The imaging unit C3 captures an image of the inspection region R3. Here, the inspection regions R1, R2, and R3 are arranged in the width direction corresponding to the imaging units C1 to C3.

[0063] exist Figure 3 In the example shown in FIG, a defective portion H12 exists in the inspection region R1 and a defective portion H11 exists in the inspection region R3. Here, the defective portion H11 corresponds to the defective portion H1 and the defective portion H1a, and the defective portion H12 corresponds to the defective portion H2 and the defective portion H2a.

[0064] In addition, Figure 3 In the example shown in FIG, there is no defective portion in the inspection region R2.

[0065] In this way, each of the imaging units C1 to C3 captures an image of a partial region of the glass sheet. If the region includes a defect, the image includes a portion (defective portion) where the defect is reflected.

[0066] Here, in this embodiment, three imaging units C1 to C3 are shown, but the number of imaging units may be arbitrary.

[0067] In addition, the inspection areas R1 to R3 arranged in the width direction are Figure 3 In order to make it easier to observe the inspection areas R1 to R3, they are shown at positions separated from each other, but they may be adjacent to each other, that is, they may be inspection areas that can be combined to inspect the entire width direction of the molten glass G. Or, as Figure 3 As shown, multiple examination regions separated from one another can also be used.

[0068] In addition, Figure 3 In FIG, each imaging unit C1 to C3 is simply shown, but each imaging unit C1 to C3 may be a unit in which a plurality of cameras or the like are combined.

[0069] <Division of flow direction>

[0070] exist Figure 3 , as three regions in the flow direction, upstream P1, midstream P2, and downstream P3 are shown.

[0071] Such division of flow directions may be used to estimate the location of defect generation, or may not be used.

[0072] It should be noted that in Figure 3 In the example shown, the flow direction is divided into three areas, but the number of divisions and the size of each area can also be arbitrary.

[0073] exist Figure 3 In the example, with respect to the direction of the Y-axis corresponding to the flow direction of the float kiln 12, the temperature of the molten glass G is higher and the viscosity of the molten glass G is lower on the upstream side (negative side of the Y-axis) of the flow of the float kiln 12 than on the downstream side (positive side of the Y-axis).

[0074] In other words, on the downstream side of the flow of the float furnace 12 , the temperature of the molten glass G is lower and the viscosity of the molten glass G is higher than on the upstream side.

[0075] [Classification device]

[0076] Figure 4 It is a diagram showing a configuration example of the classification device 111 according to the embodiment.

[0077] The classification device 111 is constituted by, for example, a computer.

[0078] The classification device 111 includes an input unit 131 , an output unit 132 , a communication unit 133 , a storage unit 134 , and a control unit 135 .

[0079] The output unit 132 includes a display unit 151 .

[0080] The control unit 135 includes a learning control unit 191 and a determination unit 192 .

[0081] The input unit 131 has a function of inputting information according to an operation by a user (eg, a person) and a function of inputting information from an external device.

[0082] The output unit 132 has a function of outputting information to the outside. For example, the display unit 151 has a function of displaying and outputting information of a display object on a screen.

[0083] The communication unit 133 has a function of communicating with an external device.

[0084] The storage unit 134 stores various information. In this embodiment, the storage unit 134 stores a learning model 171, teacher data 172, and the like.

[0085] The learning model 171 may be set in advance in the storage unit 134 , or may be set externally at an arbitrary timing.

[0086] The teacher data 172 may be stored in advance in the storage unit 134 or may be set externally at an arbitrary timing.

[0087] Here, as the learning model 171 , any model can be used, for example, a neural network model can be used.

[0088] The learning model 171 extracts, for example, features of the input image data (image features) and outputs an estimation result (determination result) based on the features. In this embodiment, the features are information indicating the position of a foreign object falling on the glass sheet in the flow direction.

[0089] It should be noted that in this embodiment, the function of a feature extraction unit that extracts feature values of image data and the function of an estimation unit that outputs an estimation result (judgment result) related to the position of the foreign matter falling on the glass plate in the flow direction based on the extracted feature values are integrated into the learning model 171.

[0090] As another example, the feature extraction unit and the estimation unit may be implemented using different learning models. In this case, image data is input to the learning model that functions as the feature extraction unit, and the output from this learning model (feature extraction result) is input to the learning model that functions as the estimation unit. The output from this learning model is then used as the estimation result (determination result).

[0091] The control unit 135 performs various processes and controls.

[0092] The learning control unit 191 uses the teacher data 172 to learn the learning model 171. This learning is machine learning. It should be noted that, in the learning phase, for example, instructions based on user judgment results or the like may also be used for learning.

[0093] The determination unit 192 inputs the image data to be determined into the learned model 171 , and outputs the output from the learned model 171 as a determination result via the output unit 132 .

[0094] For example, the determination unit 192 can directly adopt the output (estimation result) from the completed learning model 171 as the determination result of the determination unit 192, or it can make further judgments (for example, trend judgments, etc.) based on the output (estimation result) from the completed learning model 171, and use the result of the judgment as the determination result of the determination unit 192.

[0095] Here, the image data to be determined may be input from the input unit 131 or may be stored in advance in the storage unit 134 .

[0096] For example, the image data acquired by the inspection unit 311 or image data processed from the image data may be inputted in real time from the input unit 131 , so that the determination unit 192 can always perform determination processing in real time.

[0097] It should be noted that the determination process may be automatically performed in the classification device 111 , for example.

[0098] <Example of an image with unevenness>

[0099] Reference Figure 5 (A)~ Figure 5 (I), Figure 6 (A)~ Figure 6 (I) and Figure 7 (A)~ Figure 7 (I) shows an example of an image of each position where foreign matter falls in the flow direction of the glass plate manufacturing process.

[0100] Figure 5 (A)~ Figure 5 (I) is a diagram schematically showing an example of an image when a foreign object falls upstream (low viscosity) according to the embodiment.

[0101] Figure 6 (A)~ Figure 6 (I) is a diagram schematically showing an example of an image when a foreign object falls in the middle (medium viscosity) according to the embodiment.

[0102] Figure 7 (A)~ Figure 7 (I) is a diagram schematically showing an example of an image when a foreign object falls downstream (high viscosity) according to the embodiment.

[0103] exist Figure 5 (A)~ Figure 5 (I) of FIG. 1 shows nine types of images Q1 to Q9 .

[0104] exist Figure 5 (A)~ Figure 5 As a representative, in (I) Figure 5 (I) shows the Figure 1 The same XYZ orthogonal coordinate system. Figure 5 (A)~ Figure 5 (H), the orientation of the image is also related to Figure 5 The same as (I).

[0105] exist Figure 5 (A)~ Figure 5 In the example of (I), the longitudinal direction in the drawing is the flow direction in the manufacturing process of the glass sheet by the float process.

[0106] These images Q1 to Q9 are images obtained by capturing the same capturing range, but are different in at least one of the capturing conditions and the image processing.

[0107] Image Q1 , image Q2 , and image Q3 are images captured by a transmission optical system camera.

[0108] Image Q4 , image Q5 , and image Q6 are images captured by a reflective optical system camera.

[0109] Image Q7 , image Q8 , and image Q9 are images captured by a diffuse transmission optical system camera.

[0110] In addition, image Q1, image Q4, and image Q7 are directly captured images (original images).

[0111] In addition, the images Q2, Q5, and Q8 are images obtained as a result of image processing for equalizing contrast.

[0112] In addition, the images Q2, Q5, and Q8 are images obtained by performing image processing to enhance contrast.

[0113] Here, as image processing for equalizing contrast, for example, image processing for equalizing histograms can be used.

[0114] The image processing for contrast enhancement is not particularly limited, and for example, a process for enhancing the grayscale values of image data captured by a camera may be performed. As one example, the grayscale value may be converted so that at a predetermined boundary value (predetermined grayscale value), the original grayscale value and the changed grayscale value are consistent, at grayscale values smaller than the boundary value, the changed grayscale value becomes smaller than the original grayscale value, and at grayscale values larger than the boundary value, the changed grayscale value becomes larger than the original grayscale value.

[0115] exist Figure 6 (A)~ Figure 6 (I) of FIG. 1 shows nine images Q11 to Q19. In the middle, the viscosity of the glass plate is medium.

[0116] exist Figure 6 (A)~ Figure 6 As a representative, in (I) Figure 6 (I) shows the Figure 1 The same XYZ orthogonal coordinate system. Figure 6 (A)~ Figure 6 (H), the orientation of the image is also related to Figure 6 The same as (I).

[0117] exist Figure 6 (A)~ Figure 6 In the example of (I), the longitudinal direction in the drawing is the flow direction in the manufacturing process of the glass sheet by the float process.

[0118] These images Q11 to Q19 are images obtained by capturing the same capturing range, but are different in at least one of the capturing conditions and the image processing.

[0119] The types of the images Q11 to Q19 (types of combinations of shooting conditions and image processing) are the same as Figure 5 (A)~ Figure 5 The images Q1 to Q9 shown in (I) are of the same type.

[0120] exist Figure 7 (A)~ Figure 7 (I) of FIG. 1 shows nine types of images Q21 to Q29 .

[0121] exist Figure 7 (A)~ Figure 7 As a representative, in (I) Figure 7 (I) shows the Figure 1 The same XYZ orthogonal coordinate system. Figure 7 (A)~ Figure 7 (H), the orientation of the image is also related to Figure 7 The same as (I).

[0122] exist Figure 7 (A)~ Figure 7 In the example of (I), the longitudinal direction in the drawing is the flow direction in the manufacturing process of the glass sheet by the float process.

[0123] These images Q21 to Q29 are images obtained by capturing the same capturing range, but are different in at least one of the capturing conditions and the image processing.

[0124] The types of the images Q21 to Q29 (types of combinations of shooting conditions and image processing) are the same as Figure 5 (A)~ Figure 5 The images Q1 to Q9 shown in (I) are of the same type.

[0125] In the machine learning and determination of the present embodiment, for example, one type of image may be used as the input image, or an arbitrary number (number of types) of two or more types of images may be used.

[0126] When a single type of image is used as the input image, for example, any of the nine types of images shown as examples may be used. Note that types of images not shown in this example may also be used as the input image.

[0127] When two or more images are used as input images, for example, any two or more of the nine types of images shown can be used. It should be noted that images of types not shown in this example may also be used as part or all of the two or more input images.

[0128] Here, in machine learning and determination, for example, it is considered that the greater the number of types of input images, the higher the determination accuracy (classification accuracy).

[0129] <Example of the Shape of the Defect in the Height Direction>

[0130] Reference Figure 8A 、 Figure 8B and Figure 8C , showing an example of the shape of the defect in the height direction.

[0131] Figure 8A This is a diagram showing an example of the shape of a defect in the height direction when foreign matter has fallen upstream (low viscosity) according to the embodiment.

[0132] Figure 8B This is a diagram showing an example of the shape of a defect in the height direction when a foreign object falls in the middle stream (medium viscosity) according to the embodiment.

[0133] Figure 8C This is a diagram showing an example of the shape of a defect in the height direction when foreign matter falls downstream (high viscosity) according to the embodiment.

[0134] Here, these shapes represent, for example, the images captured by the camera (in the Figure 3 In the example of , it is a shape reflected in the image captured by any one of the imaging devices constituting the imaging units C1 to C3.

[0135] exist Figure 8A 、 Figure 8B and Figure 8C In the figure, we show Figure 1 The same XYZ orthogonal coordinate system.

[0136] exist Figure 8A 、 Figure 8B and Figure 8C In each of the graphs , the horizontal axis represents the flow direction in the manufacturing process of the glass sheet, and the vertical axis represents the height position of the upper surface of the glass sheet for one defect and its surroundings.

[0137] exist Figure 8A 、 Figure 8B and Figure 8C In the example of FIG, a case is shown where a foreign object falls from above a glass plate onto the upper surface of the glass plate, and the glass plate reacts with the foreign object to generate a defective portion.

[0138] exist Figure 8A 1011 shows the shape characteristics of the defect in the height direction relative to the position in the flow direction.

[0139] like Figure 8A As shown, when a foreign object falls upstream (the glass plate has low viscosity), the periphery of the defect becomes a concave shape in the captured image captured by the camera device arranged downstream, that is, there are portions whose height direction positions become lower on the upstream and downstream sides of the periphery of the defect.

[0140] exist Figure 8B 1021 shows the shape characteristics of the defect in the height direction relative to the position in the flow direction.

[0141] like Figure 8B As shown, when a foreign object falls in the midstream (the glass plate has a medium degree of viscosity), in the captured image captured by the camera device arranged downstream, the periphery of the defect becomes a raised shape, that is, there are portions whose height direction positions become higher on the upstream and downstream sides of the periphery of the defect.

[0142] exist Figure 8C 1031 shows the shape characteristics of the defect in the height direction relative to the position in the flow direction.

[0143] like Figure 8C As shown, when a foreign object falls downstream (the glass plate has high viscosity), the image captured by the camera located downstream thereof shows a shape in which a depression in the defective portion (the depression in the center of the defective portion) and a raised portion around it do not exist.

[0144] <First Example of Learning and Judgment>

[0145] As input data of the learning model 171 (teacher data 172 , determination target data), data of an image of a glass plate including foreign matter is used.

[0146] Then, the output data of the learning model 171 is used as data of a determination result regarding the position in the flow direction where the foreign matter has fallen onto the glass sheet.

[0147] Here, as a judgment result related to the position in the flow direction, for example, it can be one or more of information determining the position (itself) and information determining an area (divided area) including the position among multiple areas (divided areas) in the flow direction.

[0148] As such an area, for example, Figure 3 The areas shown are upstream P1, midstream P2, and downstream P3.

[0149] As a specific example, the determination unit 192 can use the learned model (the learned model 171) to determine the index of the position in the flow direction where the foreign matter falls based on the image obtained by shooting (scanning) information related to the position in the flow direction where the foreign matter falls.

[0150] For example, the determination unit 192 may determine an index of the position in the flow direction where the foreign matter has fallen for a large number of images that change over time in the same inspection areas R1 to R3 and perform trend statistics on the determination results, thereby more accurately determining (determining) the index.

[0151] Here, in this example, in machine learning, for example, in the case of a shape having characteristics such as no depression at the location of a defect caused by a foreign object (for example, the location of the center of the defect) and no bulge around the location of the defect, the result of the judgment that the location where the foreign object fell is downstream (for the sake of convenience, referred to as judgment A1) can also be used as teacher data 172.

[0152] In addition, in this example, in machine learning, for example, in the case of a shape having a characteristic such as a depression at the location of a defect caused by a foreign object (for example, the location of the center of the defect) and a bulge around the location of the defect, the result of the judgment that the location where the foreign object fell is midstream (for the sake of convenience, referred to as judgment A2) can also be used as teacher data 172.

[0153] In addition, in this example, in machine learning, for example, in the case of a shape having a characteristic such as a depression at the location of a defect caused by a foreign object (for example, the location of the center of the defect), but no bulge around the location of the defect, the result of the judgment that the location where the foreign object fell is upstream (for the sake of convenience, referred to as judgment A3) can also be used as teacher data 172.

[0154] In addition, for example, the result of judging that any one of a plurality of discrete positions and continuous positions in the flow direction is the position where the foreign object has fallen based on one or more of the degree of appearance of the feature of the shape focused on in the above-mentioned judgment A1, the degree of appearance of the feature of the shape focused on in the above-mentioned judgment A2, and the degree of appearance of the feature of the shape focused on in the above-mentioned judgment A3 can be used as teacher data 172.

[0155] In the learned model (learned model 171), for example, as a feature quantity, one or more of the depression (concave portion) at the location of the foreign body, the depression (concave portion) around the location of the foreign body, and the protrusion (convex portion) around the location of the foreign body can also be used.

[0156] <Second Example of Learning and Judgment>

[0157] Data on an image of a glass sheet containing foreign matter and supplementary data (auxiliary data) may be used as input data (teacher data 172, determination target data) to the learning model 171. Furthermore, the output data from the learning model 171 is used as data for the determination result regarding the position in the flow direction of a foreign matter dropped onto the glass sheet.

[0158] Here, as auxiliary data, data representing various characteristics can be used.

[0159] <Example of Determination Result Regarding the Position of Foreign Matter Falling onto the Glass Sheet in the Flow Direction>

[0160] In this embodiment, the judgment result related to the position of the foreign matter falling onto the glass plate in the flow direction is an indicator of the position, for example, it can be a value representing the position itself, or it can be a value representing the area including the position, or it can be information used to derive such a value.

[0161] Here, as the information on the position in the flow direction of the foreign matter dropped on the glass sheet, for example, information on each foreign matter may be used, or information on a collection of a plurality of foreign matters may be used.

[0162] For example, a configuration may be adopted in which, when a unit area to be determined (determination unit area) contains multiple foreign objects, a determination result related to the position in the flow direction of the foreign objects falling onto the glass sheet is obtained for the set of multiple foreign objects. As a specific example, image data obtained by capturing the unit area to be determined (determination unit area) can be used as image data used during machine learning and determination (image data input to the learning model 171).

[0163] <Example of Imaging Device>

[0164] In this embodiment, as the imaging device (composed of Figure 3 The imaging devices of the imaging units C1 to C3 shown in the figure are preferably a transmission camera (transmission optical system camera). However, the imaging device is not limited to this, and other devices such as a reflection optical system camera or a diffuse transmission camera may also be used.

[0165] Furthermore, in this embodiment, it is particularly preferable to use the result of performing image processing to equalize contrast on the image captured by the transmission camera. However, other image processing may be performed as the image processing, or image processing is not necessarily required.

[0166] It should be noted that the location where the shooting device is set ( Figure 3 The positions of the imaging units C1 to C3 shown in FIG. 1 are not necessarily limited to Figure 3 For example, it can also be set at other positions (other parts) in the manufacturing process of the glass plate.

[0167] As described above, in the classification device 111 according to the present embodiment, classification is performed based on the shape generated on the surface of the glass sheet and on the position where the foreign matter has fallen with respect to the glass sheet.

[0168] The shape is not particularly limited, and for example, one or more of a part or all of the shape of the concave portion, a part or all of the shape of the convex portion, and other shapes may be used.

[0169] It should be noted that the classification device 111 may be, for example, an offline device, or an online device (eg, a cloud server or a local server, etc.).

[0170] In addition, in this embodiment, although the case where the float process is used as a method for manufacturing a plate-shaped glass plate is described, a fusion process may be used as another example.

[0171] As a structural example, in the classification device 111, the judgment unit 192 inputs image data obtained by photographing a glass plate manufactured into a plate shape in a specified flow direction into a learned model of machine learning, and obtains an output result from the learned model as a judgment result based on the shape caused by the foreign matter on the glass plate and related to the position of the foreign matter falling onto the glass plate in the flow direction.

[0172] Therefore, in the classification device 111, classification is performed based on the shape (for example, unevenness or the like) generated on the surface of the glass plate, in relation to the falling position of the foreign matter with respect to the glass plate.

[0173] As one configuration example, in the classification device 111 , in the learned model, one or more of a depression at the location of the foreign matter, a depression around the location of the foreign matter, and a protrusion around the location of the foreign matter are used as feature quantities.

[0174] Therefore, in the classification device 111, the classification accuracy is improved.

[0175] As an example of a configuration, the classification device 111 includes an image capturing device (in Figure 2 In the example of FIG. 1 , the imaging device constituting the imaging units C1 to C3 is a transmission camera.

[0176] Therefore, the classification device 111 can easily recognize the shape of foreign matter in the image data, thereby further improving the classification accuracy.

[0177] As one configuration example, in the classification device 111, as a result of the determination, information specifying one of a plurality of areas in the flow direction is obtained as the position in the flow direction.

[0178] Therefore, the classification device 111 performs classification using regions such as upstream, midstream, and downstream.

[0179] As one configuration example, in the classification device 111 , the image data is data of an image obtained by capturing a portion of the glass plate, and includes a portion where foreign matter is reflected.

[0180] Therefore, the classification device 111 classifies each portion of the glass plate where the foreign matter exists based on the shape (eg, unevenness) generated on the surface of the glass plate, in relation to the falling position of the foreign matter relative to the glass plate.

[0181] As one configuration example, in the classification device 111 , the learning control unit 191 generates a learned model using teacher data.

[0182] Therefore, the classification device 111 generates a learned model and uses the learned model to classify the foreign matter according to the falling position of the foreign matter on the glass sheet based on the shape (eg, unevenness) generated on the surface of the glass sheet.

[0183] It should be noted that in this embodiment, the classification device 111 is shown to have both a learning control unit 191 and a determination unit 192, but as another example, the learning control unit 191 may be not provided in the classification device 111 but in another device, and the classification device 111 utilizes the learned model obtained through the other device.

[0184] It should be noted that the program for realizing the functions of any structural part in any of the devices described above can also be recorded on a computer-readable recording medium, and the computer system can read and execute the program. It should be noted that the "computer system" mentioned here includes hardware such as an operating system or peripheral devices. In addition, the so-called "computer-readable recording medium" refers to portable media such as floppy disks, optical magnetic disks, ROMs, CD (Compact Disc: compressed disk)-ROM (Read Only Memory: read-only memory), and storage devices such as hard disks built into the computer system. Moreover, the so-called "computer-readable recording medium" also includes a recording medium that keeps the program for a certain period of time, such as a volatile memory inside a computer system that becomes a server or client when the program is sent via a network such as the Internet or a communication line such as a telephone line. The volatile memory can be, for example, RAM (Random Access Memory). The recording medium can also be, for example, a non-temporary recording medium.

[0185] Furthermore, the program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by transmission waves in the transmission medium. Here, the "transmission medium" that transmits the program refers to a medium that has the function of transmitting information, such as a network such as the Internet or a communication line such as a telephone line.

[0186] In addition, the above program may be a program for implementing a portion of the above functions. Furthermore, the above program may be a program that can implement the above functions by combining it with a program already recorded in the computer system, i.e., a so-called differential file. A differential file may also be called a differential program.

[0187] In addition, the functions of any structural parts in any of the devices described above can also be implemented by a processor. For example, each process in the embodiment can also be implemented by a processor that acts based on information such as a program and a computer-readable recording medium that stores information such as a program. Here, the processor can, for example, implement the functions of each part through separate hardware, or it can also implement the functions of each part through integrated hardware. For example, the processor may include hardware that includes at least one of a circuit for processing digital signals and a circuit for processing analog signals. For example, the processor can be composed of one or more circuit devices mounted on a circuit substrate, or one or both of one or more circuit elements. As a circuit device, an IC (Integrated Circuit) or the like can be used, and as a circuit element, a resistor or a capacitor or the like can be used.

[0188] Here, the processor may be, for example, a CPU. However, the processor is not limited to the CPU, and various processors such as a GPU (Graphics Processing Unit) or a DSP (Digital Signal Processor) may also be used. In addition, the processor may be, for example, a hardware circuit based on an ASIC (Application Specific Integrated Circuit). In addition, the processor may be composed of, for example, a plurality of CPUs, or may be composed of a plurality of ASIC-based hardware circuits. In addition, the processor may be composed of, for example, a combination of a plurality of CPUs and a plurality of ASIC-based hardware circuits. In addition, the processor may also include, for example, one or more of an amplifier circuit or a filter circuit that processes analog signals.

[0189] While the embodiments of the present disclosure have been described in detail above with reference to the drawings, the specific configuration is not limited to the embodiments and includes designs and the like that do not depart from the scope of the present disclosure.

[0190] [Note]

[0191] (Configuration Example 1) to (Configuration Example 9) are shown.

[0192] (Structural Example 1)

[0193] A classification device comprising:

[0194] A determination unit that inputs image data obtained by photographing a glass plate manufactured in a plate shape in a specified flow direction into a learned model of machine learning, and obtains an output result from the learned model as a determination result related to the position of the foreign matter falling on the glass plate in the flow direction based on the shape caused by the foreign matter on the glass plate.

[0195] (Structural Example 2)

[0196] According to the classification device described in (Structural Example 1), in which, in the learned model, one or more of the depression at the position of the foreign matter, the depression around the position of the foreign matter, and the protrusion around the position of the foreign matter is used as a feature value.

[0197] (Structural Example 3)

[0198] The classification device according to (Structural Example 1) or (Structural Example 2), wherein

[0199] The classification device includes a photographing device for photographing an image of the image data.

[0200] The photographing device is a transmission camera.

[0201] (Structural Example 4)

[0202] The classification device according to any one of (Structural Example 1) to (Structural Example 3), wherein

[0203] The determination result acquires information specifying one of a plurality of areas in the flow direction as the position in the flow direction.

[0204] (Structural Example 5)

[0205] The classification device according to any one of (Structural Example 1) to (Structural Example 4), wherein

[0206] The image data is data of an image obtained by photographing a portion of the glass plate, and includes a portion where the foreign matter is reflected.

[0207] (Structural Example 6)

[0208] The classification device according to any one of (Structural Example 1) to (Structural Example 5), wherein

[0209] The classification device includes a learning control unit that generates the learned model using teacher data.

[0210] (Structural Example 7)

[0211] It is also possible to provide a method (classification method) of processing performed by the classification device.

[0212] A classification method,

[0213] The classification device inputs image data obtained by photographing a glass plate manufactured into a plate shape in a specified flow direction into a learned model of machine learning, and obtains an output result from the learned model as a judgment result based on the shape caused by the foreign matter on the glass plate and related to the position of the foreign matter falling onto the glass plate in the flow direction.

[0214] (Structural Example 8)

[0215] It is also possible to provide a program executed by a computer (computer program).

[0216] A program that enables a computer to:

[0217] A function of inputting image data obtained by photographing a glass plate manufactured in a plate shape in a predetermined flow direction into a machine learning model; and

[0218] A function of acquiring an output result from the learned model as a determination result related to a position in the flow direction of the glass sheet where the foreign matter falls, based on a shape caused by the foreign matter on the glass sheet.

[0219] (Structural Example 9)

[0220] It can also provide a learning model for machine learning.

[0221] A machine learning learned model inputs image data obtained by photographing a glass plate manufactured into a plate shape in a specified flow direction into the machine learning learned model, and outputs judgment result data related to the position of the foreign matter falling onto the glass plate in the flow direction based on the shape caused by the foreign matter on the glass plate.

[0222] Description of labels

[0223] 1...Glassmaking device, 11...Melting furnace, 12...Float kiln, 13...Annealing furnace, 14...Inspection / cutting unit, 21...Inlet, 22...Bath, 23...Outlet, 24...Side wall, 26...Ceiling, 28...Space, 30...Gas supply path, 32...Heater, 34...Metal casing, 36...Bottom brick, 36a...Upper surface, 38...Side brick, 111...Sorting device, 131...Input unit, 132...Output unit, 133...Communication unit, 134...Storage unit, 135...Control unit, 151...Display unit, 171 …learning model, 172…teacher data, 191…learning control unit, 192…judgment unit, 311…inspection unit, 1011, 1021, 1031…characteristics, C1~C3…shooting unit, E…glass width area, F1~F4…tensile force, G…molten glass, H1, H1a, H2, H2a, H11, H12…defective part, M…molten tin, P1…upstream, P2…midstream, P3…downstream, Q1~Q9, Q11~Q19, Q21~Q29…image, R1~R3…inspection area.

Claims

1. A classification device comprising: A determination unit that inputs image data obtained by photographing a glass plate manufactured in a plate shape in a specified flow direction into a learned model of machine learning, and obtains an output result from the learned model as a determination result related to the position of the foreign matter falling on the glass plate in the flow direction based on the shape caused by the foreign matter on the glass plate.

2. The classification device according to claim 1, wherein: In the learned model, one or more of a depression at the location of the foreign object, a depression around the location of the foreign object, and a protrusion around the location of the foreign object is used as a feature amount.

3. The classification device according to claim 1 or 2, wherein: The classification device includes a photographing device for photographing an image of the image data. The photographing device is a transmission camera.

4. The classification device according to claim 1 or 2, wherein: The determination result acquires information specifying one of a plurality of areas in the flow direction as the position in the flow direction.

5. The classification device according to claim 1 or 2, wherein: The image data is data of an image obtained by photographing a portion of the glass plate, and includes a portion where the foreign matter is reflected.

6. The classification device according to claim 1 or 2, wherein: The classification device includes a learning control unit that generates the learned model using teacher data.

7. A classification method, The classification device inputs image data obtained by photographing a glass plate manufactured into a plate shape in a specified flow direction into a learned model of machine learning, and obtains an output result from the learned model as a judgment result based on the shape caused by the foreign matter on the glass plate and related to the position of the foreign matter falling onto the glass plate in the flow direction.

8. A program for causing a computer to implement the following functions: A function of inputting image data obtained by photographing a glass plate manufactured in a plate shape in a predetermined flow direction into a machine learning model; and An output result from the learned model is obtained as a determination result related to a position in the flow direction where the foreign matter falls onto the glass sheet, based on a shape of the foreign matter on the glass sheet.

9. A machine learning learning model that inputs image data obtained by photographing a glass plate manufactured in a plate shape in a specified flow direction into the machine learning learning model, and outputs judgment result data based on the shape caused by foreign matter on the glass plate and related to the position of the foreign matter falling onto the glass plate in the flow direction.

Images