TWM603618U - Cutting edge inspection device for glass substrates - Google Patents

Abstract

A device for inspecting the dividing edge of a glass substrate is provided, comprising: a table A, which carries a liquid crystal panel a and has a holding and sucking means; a camera C, is arranged on the outside of the edge of the table A and shoots an image of the dividing surface of the liquid crystal panel a; Devices L ₁ , L ₂ , L ₃ , and L ₄ selectively irradiate light from multiple directions on the same dividing plane; and a determination device 40, a signal connected to camera C to use data from camera C for determination; table A or camera C At least one of them will move; the judging device has the following function: use the image data of the glass profile measured by the camera C, obtain the image data by the optical stereo method, and compare it with the data learned by the deep learning to determine the liquid crystal panel a The split surface situation. This creation uses image processing to automatically determine the ribbing or penetration depth of the glass dividing surface, and uses this automatic determination to determine the same accuracy as a skilled operator.

Description

Partition edge inspection device of glass substrate

This creation is about the glass substrate cutting edge inspection device in the manufacturing of liquid crystal panel substrates, etc., and the glass substrate cutting process. Specifically, it relates to an inspection device that uses a simulated three-dimensional image that emphasizes the shape generated by the optical stereo method, and can perform judgments with the same accuracy as a skilled operator.

As shown in FIG. 5(A), when cutting a glass substrate used for a liquid crystal panel substrate or the like with high precision, a cutting wheel 21 having an uneven blade portion 21a with a radius of about 2 mm is used. As shown in Fig. 5(B), when the cutting wheel 21 moves while cutting into the surface of the glass 30 about 2 μm, and the blade 21a collides with the glass 30, a weight called rib mark 32 will be generated due to the impact. Covering wave shape profile.

Next, please refer to Fig. 6, which is a side view of Fig. 5(B) viewed from the lateral direction. Specifically, from the tip of the crack formed by the tip 21a of the cutting wheel 21, it forms a flat surface at high speed toward the inside of the glass 30 Cracks with higher degrees of severity. Such a rupture is called soaking 33, which advances in almost the same direction as the direction in which the cutting machine applies pressure, so it is almost at a right angle to the glass surface.

And in Figure 6, the cutting depth of the cutting edge 21a is 31d, the rib depth is 32d, the penetration depth is 33d, and the glass is not divided and penetrates the unreached area (the penetration outer area) depth is 34d.

Considering the processing and transportation of high-precision glass, it is necessary to separate the cutting and dividing steps. Therefore, even if the penetration 33 occurs, the glass 30 needs to have an unbroken part.

However, if the glass 30 is divided by a load while the penetration 33 is relatively shallow, micro-cracks randomly present from the tip of the penetration 33 toward the inside of the glass 30 and the non-reproducibility of the load press method may occur. Large oblique rupture (please refer to the dotted line in Figure 7). This will form chipping or gaps and reduce the accuracy of segmentation.

Therefore, it is very important to control the depth of penetration 33 to form as deep penetration 33 as possible, and to reduce the growth distance of random cracks, which is very important for improving the quality of glass cutting.

In addition, the stable depth of the penetration 33 is related to the depth of the ribs 32. Therefore, inconsistency occurs in the abrasion of the cutting wheel 21, and the ribs 32 cannot be formed stably. At this time, the depth of the penetration 33 is also not stable, and there is a concern that the above-mentioned problems may occur.

A common method used in the past is to take out the glass from the wire and use a microscope to monitor and photograph the state of the glass section of the small slice of glass. In order not to take it out from the glass conveying line, there has been a long-standing demand for automation (see Patent Documents 1 to 5). These are to project light on the glass end surface through various angles to detect defects.

The creator of this case has proposed other methods: use a 3D displacement meter to check the glass section, which uses a laser displacement meter to image the object plane (shading image data) and the distance to the object (object height) At the same time, take it out as data. By this, the press-fitting part, the rib part and the soaking part of a cutter were judged (please refer to patent document 6).

[Prior Technical Literature]

[Patent Literature]

Patent Document 1: Japanese Patent Application Publication No. 2012-181032; Patent Document 2: Japanese Patent Application Publication No. 2018-187170; Patent Document 3: International Publication WO2012/153718; Patent Document 4: International Publication WO2017/073628; Patent Document 5: Japanese Patent Application Laid-Open No. 2019-137606; Patent Document 6: Japanese Utility Model Registration No. 3224054. .

The technologies of the aforementioned Patent Documents 1 to 4 project light on the glass end surface from various angles and photograph the cut surface of the glass to detect the defective state; the technology of Patent Document 5 provides an inspection device that automatically detects the presence or absence of ribs and thickness , But this is also judged by the general plane image.

However, the problem with automation is that the flat glass surface has specular reflection properties. Even if it is a small amount, if the angle of the cut surface is poor, the irradiated light as coaxial falling will not go to the camera at all, which causes the contrast to be very low sometimes. If it is a human judgment, such a situation can be judged even in low contrast.

Originally, the purpose of glass division is to cut all of them at right angles, so if the residue in the same direction as the penetration is divided, the boundary of the penetration cannot be seen. At this time, the texture of the glass surface from the penetrating tip portion is only slightly rough, and the degree of difference is not large, so such a difference cannot be confirmed by conventional image processing. However, a skilled operator can recognize such subtle differences.

The ribs in a good glass section have high repeatability and easy image processing. However, when the cutting machine or cutting machine used for a long time is stained with glass powder, the ribs will become irregular and chaotic. In this case, it is very difficult for image processing to determine where the regular ribs produced by the cutting machine are.

Compared with the methods in Patent Documents 1 to 5, the method in Patent Document 6 proposed by the creator of this case is to obtain three-dimensional glass profile data and use them as a three-dimensional data image. The partition function allows the cutting machine press-in part, rib part and soaked part to be judged with the same degree of accuracy as a skilled person.

However, the above method requires a specific device such as a 3D displacement meter, which can simultaneously extract the light and dark image data of the object and the distance to the object by the laser displacement meter. The creator of this case devotes himself to researching and developing a method to make the same judgment as the skilled person based on the light and dark image data of the object without using the device, thus completing the creation.

In summary, the purpose of this creation is to solve the above-mentioned problems and provide a glass substrate dividing edge inspection device. Even an inspection device that uses image processing of light and dark data obtained from a general camera to perform automatic determination can achieve the same accuracy as a skilled operator to measure the ribs or penetration depth of the glass dividing surface.

This creation provides a device for inspecting the dividing edge of a glass substrate, which includes: a table on which the bottom of the glass substrate is carried; a holding suction means, configured to hold the supporting glass substrate on the table; and a camera, arranged on the outside of the edge of the table , And shoot the image of the split surface of the glass substrate; the lighting device selectively irradiates light on the split surface of the glass substrate from a plurality of directions; and the determination device, the signal is connected to the camera C to use the data from the camera to determine; where, the table Or at least one of the cameras is moved along the edge of the table by means of forward and retreat movement; the judging device has the following function: irradiating light on the dividing surface from different directions to obtain a complex image, and using the optical stereo method to obtain the complex image Split the surface texture data and compare it with the data that has been learned by deep learning to determine the split surface of the glass substrate.

The so-called optical stereo method is to obtain multiple images while changing the light source direction relative to the subject while changing the light source direction of the subject taken by the camera, and estimate the unevenness on the surface of the subject based on the brightness difference. The simulated texture information of the glass dividing surface is obtained from the aforementioned data. in other words, According to the optical three-dimensional method, the light-emitting position of the lighting device is changed to switch the multiple images, and the generated "moire" is used to create a simulated three-dimensional image that emphasizes the shadow with respect to the unevenness. Through such an image processing system, texture is emphasized.

The texture image that emphasizes the split surface obtained by the aforementioned optical stereo method is not a precise 3D data using a laser, but a simulated 3D data. However, it is still possible to fully determine the condition of the glass dividing surface, that is, to determine the cutting machine press-in part, rib part and soaking part.

Although the aforementioned poor texture is easy to judge for a skilled operator, it is still difficult to automatically recognize in current image processing. Therefore, this creation uses deep learning for images that have an emphasis on texture, and makes judgments in areas similar to humans.

In addition, these inspections can be performed with respect to all glass cutting surfaces, but the measurement within 1 week of the cutting machine can fully achieve the purpose. Therefore, even for point measurement of several points, the inspection effect can be expected.

Based on the above, the cutting edge inspection device of the glass substrate according to the present invention is not used to judge the ribs only by the plane image data, or judged by the precise 3D image using the laser displacement meter. Instead, the cut surface shading data is obtained by illuminating light from multiple directions, using optical stereo method to obtain simulated three-dimensional data images, and using the segmentation function of deep learning. As a result, the high-precision judgment of the cutting machine press-in part, rib part and impregnated part is achieved in the same way as human judgment.

In the prior art, a cutting machine for glass cutting uses a predetermined moving distance to judge the life. In terms of application, the cutting machine cannot be used to the end of its life frequently, so it will exchange for a cutting machine that can still be used. However, this creation can know the service life of the cutting machine blade by accurately judging the areas of the press-in part, the rib part and the soaking part, so that it can be used to indicate before the use condition deteriorates. Change the cutting machine to reduce operating costs. This can also further develop the indicator system for the automatic cutting machine exchange system.

In addition, when the cutting of the cutting machine becomes poor, the natural soaking will also become shallow. Therefore, in order to determine the penetration, the glass needs to be taken out, discarded and cut, and this needs to be done manually. However, with the device created by this invention, inspection can be performed without scrapping the glass. In addition, if it is determined that the saturated area is insufficient, the applied pressure is automatically increased, thereby maintaining product quality for a long period of time, extending the life of the cutting machine, and achieving full automation.

a: LCD panel

A: Table

C: Camera

L1, L2, L3, L4: lighting device

1: Suction port

2: top wall

3: small hole

4: base

5: Seat plate

6: Rail

7: Slide

8: Bearing

9: male screw

10: Motor

11: Female screw

12: Seat material

21: Cutting wheel

21a: Tip

30: glass

31d: Tool tip depth

32: rib

32d: rib depth

33: Soak

33d: penetration depth

34d: Depth of penetration into outer domain

40: judging device

Figure 1 shows the side view of the implementation of this creation, partly omitted.

Figure 2 shows a partially enlarged plan view showing the implementation of this creation.

Figure 3 is the inspection flow chart of the glass dividing surface.

Fig. 4 shows a normal shot image and an enhanced image obtained by the optical stereo method.

Fig. 5(A) is a front view of the cutting wheel, and Fig. 5(B) is an enlarged sectional front view of the cutting surface during glass cutting.

Figure 6 is a side view showing the glass when cut with a cutting wheel.

Figure 7 is a side view showing the cut glass.

First of all, according to Figure 1 and Figure 2, the implementation of this creation will be explained. Figures 1 and 2 show the table A on which the liquid crystal panel a is mounted. The liquid crystal panel a mentioned above is held on the table A.

As shown in the figure, the above-mentioned holding method is to use a hollow table A, suck the table A through a hose (not shown) connected to the suction port 1, and through the top wall 2 installed on the table A Countless small Hole 3 and suck table A. Thereby, the liquid crystal panel a carried on the top wall 2 is maintained and sucked. But this creation is not limited to this, it can also be maintained in other ways.

As shown in Figures 1 and 2, on the outside of an edge of the table A, a camera C and a determination device 40 for signal connection to the camera C are arranged on the fixed seat material 12, and the camera C photographs the edge of the liquid crystal panel a Split surface image. On both sides of the camera C on the base material 12, four illumination devices L ₁ , L ₂ , L ₃ , and L ₄ are arranged.

The outside of the edge of one of the table A or the camera C is arranged to move forward and backward by the moving means D along the edge of the liquid crystal panel a (the edge where the image is taken). As shown in Fig. 2, the base 4 on which the table A is installed is moved in the front-rear direction by the moving means D, and the camera C side is the fixed side.

As shown in the figure, the aforementioned moving means D is specifically as follows: two fixed guide rails 6 on the left and right are arranged in parallel on the lower side of the seat plate 5 arranged under the base 4, and four fixed guide rails 6 are slidably engaged and arranged under the seat plate 5. The corner slide 7 is freely rotated by bearings (not shown) at both ends, so that one end of the male screw 9 of the bearing is reversibly driven by a motor (not shown). Next, the female screw 11 supported on the lower surface of the seat plate 5 is screwed into the male screw 9. As a result, as shown in FIG. 2, the table A and the base 4 are moved forward and backward in the forward and backward directions. However, this creation is not limited to this, and the table A can be fixed and the camera C can be moved (forward and backward).

Furthermore, the moving means D is performed by driving the male screw 9 screwed to the female screw 11. But this creation is not limited to this, for example: linear motors can be used.

Through the aforementioned configuration, the image of the split surface of the liquid crystal panel a on the table 1 is captured by the camera C, and at the same time, the image of the split surface of the liquid crystal panel a is captured by the camera C through the action of the moving means D. .

Regarding the split surface image, first use only one of the lighting devices L ₁ , L ₂ , L ₃ , and L ₄ at the shooting position (for example, L1), illuminate the light and shoot with the camera C; then, use only the same position Other lighting devices (for example, L ₂ ) irradiate light and shoot with camera C; next, while irradiating light with lighting devices L ₃ and L ₄ in sequence, shoot with camera C.

In the foregoing embodiment, although a total of four lighting devices L ₁ , L ₂ , L ₃ , and L _{4 are} used, as long as there are at least two lighting devices, the optical stereo method can be implemented. Therefore, the number of lighting devices is two or more, and is not particularly limited. And the irradiation sequence is not limited.

As described above, at the inspection position, through the optical stereo method of the camera C and the lighting devices L ₁ , L ₂ , L ₃ , and L ₄ to photograph the glass dividing surface, and then move the table A by the moving means D while photographing The other parts of the glass dividing surface of the liquid crystal panel a are repeatedly photographed up to a predetermined number.

As shown in Fig. 3, it is a flowchart of the inspection of the glass split surface, and serves as an explanation of the procedure for inspecting the glass split surface from the data obtained by the camera C.

Obtain the three-dimensional emphasized image data (light and dark image data) of the glass dividing surface, which is obtained from the image of the optical three-dimensional method, and input the data into the learned network input layer in the judging device 40.

Figure 4 (A) and (B) are images of the glass dividing surface of the liquid crystal panel a. Specifically, Fig. 4(A) is a normal shot image, and Fig. 4(B) is an image that emphasizes the uneven surface obtained by the optical stereo method. Compared with the normal image in Figure 4(A), the image obtained from the optical stereo method in Figure 4(B) shows that the rib area or the saturated area is clearly distinguishable by the naked eye. In addition, it can also be distinguished outside the permeated area, which can be fully judged by a skilled operator.

However, for the automatic judgment of the device, it is difficult to achieve the same fully automatic judgment field as human. Therefore, the determination through deep learning will be performed in the determination device 40 below.

The judging device 40 accumulates data that has been learned by deep learning, and automatically compares and reviews with the data, and detects the rib area by the segmentation function of the learned network. If the evaluation value at this time exceeds the threshold, it is "rib detection OK"; if it does not exceed the threshold, it is "rib detection NG".

Then, through the segmentation function of the learned network, the penetration of the external domain is detected. It is easier to judge the soaked outer domain than the soaked domain. If the depth of penetration of the outer domain 34d is known, the depth 33d of penetration 33 can be determined. If the evaluation value at this time exceeds the threshold, it will be "saturation detection OK"; if it does not exceed the threshold, it will be "saturation detection NG".

When the aforementioned judgment is "rib detection OK" or "saturation detection OK", the individual rib depth 32d and the penetration depth 33d are output, and the inspection ends.

With the output of the cutting machine depth 31d, rib depth 32d, and penetration depth 33d, the cutting wheel can be exchanged manually or automatically, or the pressure or speed during cutting can be adjusted.

The above describes the implementation type of this creation. However, this creation is not limited to the above-mentioned implementation type. The glass segmentation edge inspection device can be appropriately changed within the scope of the creation; the judgment method using the deep learning in the judgment device can also be within the scope of the creation. Implement appropriate changes.

a: LCD panel

A: Table

C: Camera

L1, L2, L3, L4: lighting device

2: top wall

3: small hole

6: Rail

8: Bearing

9: male screw

10: Motor

12: Seat material

40: judging device

Claims (1)

A device for inspecting the dividing edge of a glass substrate, comprising: a table on which the lower surface of the glass substrate is carried; a holding suction means configured to hold the glass substrate on the table; and a camera, which is arranged outside the edge of the table , And shoot the image of the split surface of the glass substrate; an illumination device that selectively irradiates light on the split surface of the glass substrate from a plurality of directions; and a determination device that signals the camera to connect to the camera to use the data from the camera to determine; wherein, At least one of the table or the camera is moved along the edge of the table by means of moving forward and backward; the judging device has the following function: irradiating light on the dividing surface from different directions to obtain a plurality of images, and the plurality of images The image uses the optical stereo method to obtain the split surface texture data, and compares it with the data that has been learned by deep learning to determine the split surface of the glass substrate.

Images