JP3225922U - Glass substrate cutting edge inspection device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a dividing edge of a glass substrate capable of making a judgment with the same accuracy as that of a skilled worker in an inspection device for measuring a rib mark or a penetration depth of a glass divided section by using automatic judgment using image processing. Provide an inspection device. SOLUTION: A table A with suction holding means for receiving a liquid crystal panel a, a camera C arranged outside the peripheral edge of the table A for capturing an image of a cross section of the liquid crystal panel a, and the same cross section from a plurality of directions. It comprises illuminations L1, L2, L3, and L4 that selectively emit light, and a determination device that performs determination using the data from camera C, so that at least one of table A or camera C is driven, The device has a function of comparing the image data obtained by the photometric stereo method with the learned data by deep learning using the glass cross-section image data measured by the camera C, and determining the state of the divided cross section of the liquid crystal panel a. [Selection diagram] Figure 2

Description

  The present invention relates to an apparatus for inspecting a cut edge of a glass substrate in a glass substrate cutting process in manufacturing a liquid crystal panel substrate or the like, and more specifically, a pseudo three-dimensional image emphasizing a shape generated from a photometric stereo method. The present invention relates to an inspection device that can be used with the same accuracy as judgment by a skilled worker.

  When cutting a glass substrate used for a liquid crystal panel substrate or the like with high accuracy, a cutter wheel 21 having a serrated blade portion 21a with a radius of about 2 mm as shown in FIG. 5A is used. As shown in B), the cutter wheel 21 moves in a state of biting into the surface of the glass 30 by about 2 μm, and a repetitive wave shape called a rib mark 32 due to the impact when the blade portion 21 a collides with the glass 30. Cross section occurs.

  Further, FIG. 6 is a view of FIG. 5B viewed from the lateral direction, in which the flatness is relatively high at a high speed from the tip of the crack formed by the cutting edge 21a of the cutter wheel 21 toward the inside of the glass 30. There are cracks. This crack is called penetration 33. Since this infiltration 33 proceeds in almost the same direction as the direction in which the cutter pressure is applied, most of the infiltration 33 is approximately perpendicular to the glass surface.

  The depth of the cutter with the cutting edge 21a in FIG. 6 is 31d, the depth of the rib mark is 32d, the depth of penetration is 33d, and the depth of the region where the glass does not divide (permeation outside region) does not reach. 34d.

  Considering accurate glass processing and transportation, it is necessary to separate the step of scribing and the step of dividing. Therefore, even if the infiltration 33 occurs, it is necessary to leave a portion of the glass 30 which is not broken.

  However, when the glass 30 having a shallow penetration 33 is divided by a load, it may be largely obliquely cracked due to random cracks present inside the glass 30 from the tip of the penetration 33 and non-reproducibility of how the load is applied (in FIG. 7). See the dotted line). This results in chipping and chipping, resulting in a decrease in cutting accuracy.

  Therefore, it is very important to improve the quality of glass cutting by controlling the depth of the infiltration 33 to form the infiltration 33 as deep as possible and reducing the distance the cracks run at random.

  Further, the stable depth of the permeation 33 is related to the depth of the rib 32. Therefore, unless the rib 32 is stably formed due to unevenness due to wear of the cutter wheel 21, the depth of the permeation 33 will not be stable. However, the above problems may occur.

  Conventionally, it is a general method to take out glass from a line and monitor or image the state of the glass cross section with a microscope on a glass piece cut into small pieces. There has been a long-standing need for automation in order not to take out the glass from the glass conveying line (see Patent Documents 1 to 5), and the defect is detected by irradiating the glass end face with light from various angles.

  As another method, the present inventor uses a laser displacement meter as a plane image (bright and dark image data) of an object and a distance to the object (height of the object) as data at the same time. A method has been proposed in which a cutter press-fit portion, a rib mark portion, and a permeation portion are determined by inspecting a glass cross section using a 3D displacement meter that can be taken out (see Patent Document 6).

JP2012-181032A JP, 2018-187170, A International publication WO2012 / 153718 International publication WO2017 / 073628 JP, 2019-137606, A Utility model registration No. 3224054

  The techniques of Patent Documents 1 to 4 each apply light to the glass end face from various angles to image the glass cut surface and detect defects. In addition, the technique of Patent Document 5 is an inspection device that automatically detects the presence or absence of the formation of rib marks and the thickness thereof, but in this case as well, the determination is performed using a normal planar image.

  By the way, the problem with automation is that the flat glass surface has the property of specular reflection, and if the angle of the cut surface is a little bad, the light incident as coaxial incident does not go to the camera at all, and the contrast is very high. This is because there are cases where it is low. Even in that case, humans can judge even with low contrast.

  Originally, the purpose of all glass breaks is to cut at right angles, so if the rest is broken in the same direction as penetration, the boundary of penetration cannot be seen. At that time, since the texture of the glass surface in the portion before the penetration is slightly rough, it cannot be recognized by the conventional image processing. However, a skilled worker can recognize such a subtle difference.

  In a good glass cross section, the rib marks have a very high repeatability and image processing is easy, but if the glass powder sticks to the cutter or cutter that has been used up, the rib marks will be scattered and disturbed. In such a case, it becomes a very difficult case as image processing to determine which is the proper rib mark generated by the cutter.

  In contrast to the methods of Patent Documents 1 to 5, in the method of Patent Document 6 previously proposed by the present inventor, the stereoscopic data of the glass cross section is added and captured as a three-dimensional data image, and the segmentation function of deep learning is obtained. By using the cutter, the cutter press-fitting portion, the rib mark portion, and the penetrating portion could be accurately determined as well as the determination by an expert.

  However, a specific device called a 3D displacement meter, which is capable of taking out the light and dark image data of an object and the distance to the object at the same time by a laser displacement meter, is essential, and the inventor of the present invention uses the device. The present invention was devised after earnest research on a means capable of making a judgment equivalent to that of a skilled person from image data of light and dark of an object.

  The purpose of this invention is to solve the above-mentioned problems, and even with an inspection apparatus that uses automatic determination using image processing consisting of light and dark data by a normal camera, with the same accuracy as a skilled worker, It is an object of the present invention to provide an apparatus for inspecting a cutting edge of a glass substrate, which is capable of measuring a rib mark and a penetration depth of a glass cross section.

  That is, in order to solve the above-mentioned problems, the present invention relates to a table provided on an upper surface thereof to receive a lower surface of a glass substrate, a suction holding means provided to hold the load receiving glass substrate on the table, and A camera C arranged outside the edge of the table so as to capture an image of the cross section of the glass substrate, an illumination for selectively irradiating the cross section of the glass substrate with light from a plurality of directions, and the camera. It is configured to drive at least one of the table or the camera along the edge of the table by advancing / retreating means, and the determination device is Deep learning learns the texture data of the cross-section obtained by using the photometric stereo method for multiple images illuminated with light from different directions on the cross-section. Is compared cutting edge inspection apparatus for a glass substrate in which a function determining the state of the cutting surface of the glass substrate with the observed data

  The photometric stereo method is a method of capturing the brightness data of a subject captured by a camera, capturing a plurality of images while changing the light source direction with respect to the subject, and estimating the unevenness of the surface of the subject from the difference in brightness. Pseudo-texture information of the glass cross section is obtained from this data. That is, by converting the position of illumination light emission based on the photometric stereo method, it is possible to create a pseudo three-dimensional image in which shadows are emphasized on unevenness by using "waviness" generated by switching between multiple images. Texture can be enhanced using various image processing systems.

  The texture image itself with the enhanced cross-section obtained by this photometric stereo method is not the precise three-dimensional data using the laser but the pseudo three-dimensional data. The data is sufficient to determine the press-fitted portion, rib mark portion, and permeation portion.

  However, although this texture difference is at a level that can be easily judged by a skilled person, it is difficult to automatically recognize it by the current image processing. Therefore, in the present invention, it was decided to use deep learning on an image having an emphasized texture to make a region determination close to a person.

  Although these inspections may be performed on the entire glass cut surface, the purpose can be sufficiently achieved even by measuring the range in which the cutter makes one round. Therefore, the effect of the inspection can be expected even when measuring several points.

  As described above, according to the device for inspecting a broken edge of a glass substrate of the present invention, it is not possible to judge a rib mark only by plane image data as in the conventional case or a precise 3D image using a laser displacement meter. Instead, the light and dark data of the cross-sections illuminated by light sources in multiple directions are captured as a pseudo three-dimensional data image by the photometric stereo method, and by using the deep learning segmentation function, the cutter press-fitting part, rib mark It is possible to accurately determine the part and the penetrating part as in the case of human identification.

  In addition, conventionally, the life of the glass cutting cutter is judged based on the estimated traveling distance, and it is impossible to use it to the limit of the cutter life at all times, so it was originally necessary to replace the cutter that can still be used. By devising the device, it is possible to know the life of the blade of the cutter by accurately judging each area of the press-fitting part, the rib mark part and the penetrating part, and it is possible to issue a cutter replacement instruction just before the condition really deteriorates, operating cost In addition to contributing to the reduction of waste, it is also possible to develop it as an instruction system for an automatic cutter exchange system.

  Furthermore, if the cutter becomes poorly cut, the penetration will naturally become shallow, so it was necessary for humans to pull out the glass, discard the glass and cut it in order to judge the penetration. With this device, it is possible to inspect the glass without discarding it, and if it is determined that the permeation area is insufficient, it is possible to protect the quality of the product for a long time by automatically increasing the pressurization amount, and it is possible to extend the life of the cutter. In addition, these can be fully automated.

It is a partially notched side view which shows embodiment of this invention. It is a partially notched enlarged plan view of the above. It is an inspection flowchart figure of a glass cross section. 3A and 3B show a normal captured image and an enhanced image by a photometric stereo method. (A) is a front view of a cutter wheel, (B) is an enlarged cross-sectional front view of a cut surface at the time of cutting glass. It is a side view of glass at the time of scribing with a cutter wheel. It is a side view of glass after scribing.

An embodiment of the present invention will be described below with reference to FIGS. 1 and 2 of the accompanying drawings.
1A and 2A is a table for receiving the liquid crystal panel a on its upper surface.
The received liquid crystal panel a is held on the table A.

  In this holding method, in the case of the drawing, a hollow table A is used to suck the inside of the table A through a hose (not shown) connected to the suction port 1 and is provided on the top wall 2 of the table A. By sucking the table A through the innumerable small holes 3, the liquid crystal panel a received on the top wall 2 is sucked and held, but it is not limited and may be held by other methods. .

Further, outside one edge of the table A, a camera C that captures an image of a divided cross section of the edge of the liquid crystal panel a is arranged on a fixed type seat material 12, and the camera C on the seat material 12 also has the same structure. Four lights L ₁ , L ₂ , L ₃ and L ₄ are arranged on both sides.

  Further, either the table A side or the camera C side is moved forward and backward by the traveling means D along the side edge (edge for capturing an image) of the liquid crystal panel a. The base 4 on which A is installed is caused to travel in the front-back direction in FIG. 2 by the travel means D, and the camera C side is the fixed side.

  In the illustrated case, the traveling means D includes two left and right stationary guide rails 6 arranged side by side below the seat plate 5 provided below the base 4, and the guide rails 6 are provided with a square bottom surface of the seat plate 5. A slider 7 provided at the end of the seat 7 is slidably engaged, and one end of a male screw 9 whose both ends are rotatably supported by bearings 8 is reversibly driven by a motor 10 and supported on the lower surface of the seat plate 5. By screwing the female screw 11 into the male screw 9, the table A is moved forward and backward together with the base 4 in the front-rear direction of FIG. 2, but not limited thereto. It may be movable.

  Further, the traveling means D is configured to drive the male screw 9 into which the female screw 11 is screwed, but of course, it is not limited to this system, and for example, a linear motor system can be adopted.

  With the above configuration, the camera C can capture the image of the cross section of the liquid crystal panel a on the table A, and the traveling unit D causes the camera C to cover the whole cross section of the liquid crystal panel a. You can capture an image of the situation.

It should be noted that the imaging of the cross-section is performed by the camera C by irradiating light from only one of the illuminations L ₁ , L ₂ , L ₃ , and L ₄ (for example, L ₁ ) at the imaging position. Then, the camera C irradiates light from only another illumination (for example, L ₂ ) at the same position to capture an image, and further illuminates the illuminations L ₃ and L ₄ in this order while sequentially illuminating the camera. Imaging by C is performed.

Although the four illuminations L ₁ , L ₂ , L ₃ , and L ₄ are used in the above-described embodiment, the photometric stereo method can be performed if the number of illuminations is two or more. Is not particularly limited to two or more, and the irradiation order is not limited.

As described above, the glass cross section is photographed by the photometric stereo method using the camera C and the illuminations L ₁ , L ₂ , L ₃ , and L ₄ at the inspection position, and then the table A is moved by the traveling means D while the liquid crystal is being moved. The other part of the glass cross section of the panel a is also photographed, and the photographing is repeated up to a prescribed number.

  Next, the procedure for inspecting the glass cross section based on the data obtained by the camera C will be described based on the flowchart for inspecting the glass cross section in FIG.

  The stereo-emphasized image data (light and dark image data) of the glass cross section obtained by the imaging of the photometric stereo method is acquired, and this data is input to the learned network input layer in the determination device.

  4A and 4B are images showing a state of a glass cross section of the liquid crystal panel a, in which FIG. 4A is a normal captured image, and FIG. 4B is an image in which the uneven surface obtained by the photometric stereo method is emphasized. In the image (B) obtained by the photometric stereo method, the rib area and the penetration area can be clearly seen compared to the normal image (A), and the outside of the penetration area can be recognized, so that a skilled worker can If so, you can make a sufficient judgment.

  However, in the automatic determination by the apparatus, it is difficult to determine the area completely automatically like a person. Therefore, hereinafter, the determination by deep learning will be performed in the determination apparatus.

  Data that has been learned by deep learning is stored in the determination device, and comparison and examination with this data are automatically performed, and the rib mark area is detected by the segmentation function of the learned network. If the evaluation value at this time exceeds the threshold value, "rib mark detection OK" is set, and if the evaluation value does not exceed the threshold value, "rib mark detection NG" is set.

  Next, the extra-penetration region is detected by the segmentation function of the learned network. The permeation outside region is easier to determine than the permeation region, and if the permeation outside region depth 34d is known, the depth 33d of the permeation 33 can be found. If the evaluation value at this time exceeds the threshold value, "permeation detection OK" is set, and if the evaluation value does not exceed the threshold value, "permeation detection NG" is set.

  In the case of "rib mark detection OK" and "penetration detection OK", respectively, the rib depth 32d and the penetration depth 33d are output, and the inspection ends.

  By using the output cutter depth 31d, rib depth 32d, and penetration depth 33d, the cutter wheel can be replaced or the pressure and speed at the time of cutting can be adjusted appropriately or automatically thereafter. Become.

  Although the embodiment of the present invention has been described above, the present invention is not limited to this embodiment, and the glass cutting edge inspection device can be appropriately modified and implemented within the scope of the purpose of the present invention. However, the determination method using deep learning in the determination device can be appropriately changed and implemented within the scope of the object of the present invention.

a Liquid crystal panel A Table C Camera L ₁ , L ₂ , L ₃ , L ₄ Illumination 1 Suction port 2 Top wall 3 Small hole 4 Base 5 Seat plate 6 Guide rail 7 Slider 8 Bearing 9 Male screw 10 Motor 11 Female screw 12 seat Material 21 Cutter Wheel 21a Blade Edge 30 Glass 31d Blade Edge Depth 32 Rib 32d Rib Depth 33 Penetration 33d Penetration Depth 34d Outside Penetration Area Depth

Claims (1)

  A table provided on the upper surface so as to receive the lower surface of the glass substrate, a suction holding means provided on the table so as to hold the received glass substrate, and a table provided on the outer side of the edge of the table for separating the glass substrate. It consists of a camera that is provided to capture the image of the cross section, an illumination that selectively irradiates the cross section of the glass substrate with light from a plurality of directions, and a determination device that performs determination using the data from the camera. , At least one of the table and the camera is made to travel along the edge of the table by means of advancing / retreating travel means, and the determination device is provided with a plurality of imaging images irradiated with light from different directions with respect to the division plane. The cross-sectional texture of the glass substrate is compared with the data of the texture of the cross-section obtained by using the photometric stereo method and the data already learned by deep learning. Cutting edge inspection apparatus for a glass substrate and having a function of determining.

Images