KR102536019B1 - Car tire outfit auto test device throughout ai deep learning method using 2d and 3d images - Google Patents

Abstract

The present invention relates to an apparatus for automatically inspecting the exterior of a car tire through AI deep learning using 2D and 3D images, and more particularly, to inspect the exterior of a car tire by using 2D and 3D cameras simultaneously to obtain images. Acquisition and comparison with the defect database using AI deep learning to find defects quickly, recognizing them as learning data to make inspection more accurate and faster It relates to an automatic tire exterior inspection device.
According to the automatic inspection device for exterior appearance of a car tire through AI deep learning using 2-dimensional and 3-dimensional images formed in a preferred embodiment of the present invention, the inside, bead, sidewall, and tread of a tire are simultaneously photographed and images are acquired, so that the tire It is easy to find each defect that can occur, and the curved part of the tire is divided into upper and lower parts or upper and middle and lower parts, and each of them is photographed. Since the part is imaged using a 2D camera and a 3D camera, many defects that may occur in the tire can be found, and the captured images captured by each camera can be quickly contrasted with the learned combinations through AI learning. Therefore, finding defect data is very fast.

Description

Automotive tire exterior automatic inspection device through AI deep learning using 2D and 3D images

The present invention relates to an apparatus for automatically inspecting the exterior of a car tire through AI deep learning using 2D and 3D images, and more particularly, to inspect the exterior of a car tire by using 2D and 3D cameras simultaneously to obtain images. Acquisition and comparison with the defect database using AI deep learning to find defects quickly, recognizing them as learning data to make inspection more accurate and faster It relates to an automatic tire exterior inspection device.

A tire (English: tire, cultural language: diamond) is a circular frame that is inserted into the wheel tube of a car or bicycle.

As a raw material for the tire, natural rubber is most commonly used as a rubber material, but recently synthetic rubber is also used.

The tire is kneaded with a machine or roll that performs work such as agitation called a bumper mixer, and sulfur as a vulcanizing agent, a vulcanization accelerator, carbon black as a reinforcing agent, and zinc oxide · White clay · Magnesium carbonate, anti-aging agent · coloring agent, and stearic acid · paraffin to facilitate kneading or molding are added to manufacture.

The rubber raw material kneaded as described above comes out as a sheet, and after attaching several layers of tire coats or metal wires for reinforcement to these sheets, it is put into a vulcanizing kiln and heated, and the molding process and the vulcanization process are performed at the same time.

In the curing process, the tire is put into a vulcanizer and subjected to a primary vulcanization process in which heat and pressure are applied for a certain period of time, thereby obtaining a certain elasticity and strength. Since it reaches about 170 to 190 ° C, post-vulcanization is performed by cooling at room temperature while maintaining a constant pressure inside the tire until the internal temperature drops to a certain temperature.

The tire is formed in several layers, a tread (tred), a shoulder (shoulder), and a sidewall (sidewall) are formed in the portion surrounding the outside, each groove is formed on the tread, and an inner liner (internal) is formed on the inside. is formed

Manufactured tires are subjected to various tests, mainly uniformity inspections and appearance inspections, which are left as data.

The uniformity test often uses the run-out method, which rotates the tire in a state where no load is applied to inspect the difference in variation of each material dimension. Comparing the initially measured basic data with the result of the completed tire way. .

A plurality of laser generators are used in the uniformity inspection device of the run-out type, a camera capable of recognizing the laser sensor is formed, and an analysis device for analyzing data collected from each of the cameras is formed.

A method for inspecting the uniformity of a tire using the run-out method has been disclosed in Korean Patent Office Publication No. 2000-0069959 or Patent Registration No. 610681.

In addition, a conventional tire exterior inspection method and exterior inspection auxiliary device are disclosed in Korean Intellectual Property Office Publication Patent No. 2011-0089231.

When the tire is produced, it goes through several processes and includes many defects and defects due to the specificity of the rubber material, and tire manufacturers pay a lot of manpower and money to inspect these defects.

Defect types that occur after conventional tire production include Air type, Crack type, Cord detachment & defect type, Uncured type, Overcured type, Thickness defect type, Non-uniform or deformed type, Color abnormal type, Joint abnormal type, Cutting Type, mold abnormal type, foreign matter type, and general type in which these appear in combination.

These types of bonding occur respectively in the bead, internal, external, sidewall, sidewall, bead, and tread for each part.

Defects must be found considering all of the above types by part and each defect type, and it takes a lot of effort and time to find these defects.

However, the conventional tire exterior inspection apparatus and tire exterior inspection method have the following problems.

(1) When photographing the exterior of a tire, it is difficult to find many tire defects because the surface to be photographed is small.

(2) Since each part of the tire is photographed and analyzed as a whole, defects cannot be easily found due to the sidewall and bead connection or distortion of the inner curved surface.

(3) A 2D line camera, an RGB camera, or a 3D camera is used as a camera for imaging, but it is difficult to find many defects that may occur in a tire because certain parts are imaged with only one type of camera.

(4) It takes a lot of time to find defect data in a captured image, so the processing speed is remarkably low.

In order to overcome the above problems, the present invention is a bed in which a height adjustment sphere is formed at the bottom in a rectangular shape;

a circular test bed that is rotatable and transportable up and down in the center of the bed;

A first pillar is formed at the upper edge of the bed to be spaced apart from the test bed, and a first column is formed that is formed horizontally with the first pillar and is transportable up and down. a sidewall imaging unit having an arm formed and at least two cameras for taking images in a vertical direction at an end of the first arm;

A second column is formed at the upper edge of the bed to be spaced apart from the test bed, and a second column that can be transported vertically along the second column is formed. an internal imaging unit having an arm and having a plurality of cameras formed at an end of a vertical portion of the second arm;

A third column is formed at the upper edge of the bed to be spaced apart from the test bed, and a third column is formed horizontally to be transportable up and down on one side of the third column, and the third column is at the end of the third column. a tread imaging unit having a third arm formed to be slidable and a plurality of cameras formed at an end of the third arm;

It is fixed to the top of the bed, and it is characterized in that it consists of a main processing unit in which a processing device capable of acquiring and analyzing the captured image while controlling the position of each camera is formed.

According to the automatic inspection device for external appearance of automobile tires through AI deep learning using 2-dimensional and 3-dimensional images formed in a preferred embodiment of the present invention, the following effects occur.

(1) Since the inside, bead, sidewall, and tread of the tire are simultaneously photographed and images are acquired, it is easy to find each possible defect in the tire.

(2) The curved part of the tire is divided into upper, lower, upper, middle and lower parts, and each is photographed, and the part where the distortion occurs can be found with a 2D camera.

(3) Since each part of the tire is imaged using a 2D line camera and a 3D camera or a 2D RGB camera and a 3D camera, many defects that may occur in the tire can be found.

(4) The captured images captured by each camera can be quickly compared with the combinations learned through AI learning, so finding defect data is very fast, and defects that can occur by manufacturer can be quickly found using the deep learning method. Processing speed is very high.

1 is a conceptual diagram showing the name of each part of a conventional tire.
2 is a conceptual perspective view of an automatic vehicle tire appearance inspection device through AI deep learning using 2-dimensional and 3-dimensional images formed in a preferred embodiment of the present invention.
3 is a side conceptual diagram of an automatic vehicle tire appearance inspection device through AI deep learning using 2-dimensional and 3-dimensional images formed in a preferred embodiment of the present invention.
4 is a plan conceptual view of an automatic vehicle tire exterior inspection device through AI deep learning using 2-dimensional and 3-dimensional images formed in a preferred embodiment of the present invention.
5 to 7 show sidewalls, treads and internal parts of defects detected by an automatic automotive tire exterior inspection device through AI deep learning using 2D and 3D images formed in a preferred embodiment of the present invention. picture.

Prior to describing specific embodiments of the present invention, the drawings of the present specification are somewhat exaggerated or simplified in order to clarify the description of the size or shape of the components shown in the drawings in order to more clearly describe the present invention. can be displayed

In addition, in order to clearly describe the present invention, descriptions of parts not related to the technical spirit of the present invention are omitted, and the same or similar components are described with the same reference numerals throughout the present specification.

Since the terms and symbols defined in the present invention are terms arbitrarily defined or selectively used by users, operators, and creators, these terms have meanings consistent with the technical spirit of the present invention based on the contents throughout the present specification. It should be interpreted as a concept and should not be limited to the meaning of the term itself.

The present invention has a bed 100 in which a height adjustment sphere 110 is formed at the bottom in a rectangular shape; a circular test bed 200 that is rotatable and transportable up and down in the center of the bed 100;

A first pillar 310 is formed at the upper edge of the bed 100 to be spaced apart from the test bed 200, and is formed horizontally with the first pillar 310 to be vertically transportable. A first column ( 320) is formed, a slidable first arm 330 is formed on the first column 320, and at least two cameras for capturing images in a vertical direction are formed at the end of the first arm 330 Sidewall an imaging unit 300;

A second pillar 410 is formed at the upper edge of the bed 100 to be spaced apart from the test bed 200, and a second column 420 that can be moved up and down along the second pillar 410 is formed, The second column 420 has a slidable 'L'-shaped second arm 430, and an internal imaging unit in which a plurality of cameras are formed at the end of the vertical portion of the second arm 430. (400) and;

A third column 510 is formed at the upper edge of the bed 100 to be spaced apart from the test bed 200, and a third column 520 is horizontally movable on one side of the third column 510 so as to be able to move up and down. is formed, a third arm 530 is formed at the end of the third column 520 so as to be slidable to the third column 520, and a plurality of cameras are formed at the end of the third arm 530 a tread imaging unit 500;

It is fixed to the top of the bed 100 and consists of a main processing unit 600 in which a processing device capable of acquiring and analyzing images while controlling the position of each camera is formed.

At the lower end of the bed 100, a height adjusting sphere 110 capable of adjusting the height is formed, and is formed to supply power.

A test bed 200 is formed in which a motor 220 is formed at a lower end of the upper center of the bed 100 and a circular table 210 formed to be rotatable by an axis of the motor 220 is formed.

A plurality of stoppers 212 are formed on the periphery of the circular table 210 to fix the position of the tire 10 being loaded.

The circular table 210 has a surface treated so as not to reflect light, and is composed of a lighter color than the tire 10.

The circular table 210 is formed to rotate in a certain direction by a motor 220, and the motor 220 uses a timing motor 220 to capture the sidewall imager 300, the internal imager 400, and the It is formed to precisely match the timing recognized by all the cameras of the tread imaging unit 500 .

In the sidewall imaging unit 300, a first pillar 310 is formed, a first column 320 is formed horizontally on the first pillar 310, and the first column 320 slides horizontally. A first arm 330 is formed.

At the end of the first arm 330, a side wall upper line scanner 340 and a side wall upper 3D camera 345 are formed so that the upper part (solder) of the side wall can be photographed in a vertical downward direction, and side by side A lower side wall line scanner 350 and a lower side wall 3D camera 355 are formed to install the lower part, respectively.

The upper sidewall rice scanner and the lower sidewall line scanner 350 are two-dimensional line cameras, and are configured to capture images of the upper sidewall of the rotating tire 10 .

The sidewall upper 3D camera 345 and the lower sidewall 3D camera 355 are 3D cameras and use a camera that measures depth using a laser.

The internal imaging unit 400 has a second column 410 formed, a second column 420 formed horizontally on the second column 410, and a sliding 'a' on the second column 420. A second arm 430 having a 'shape' is formed.

The horizontal part of the second arm 430 is formed to slide horizontally on the second column 420, and at least one internal 3D camera 440 is installed at the end of the vertical part of the second arm 430. It is formed to image the inside of the tire 10 while forming a certain angle.

It is appropriate that three internal 3D cameras 440 are formed, and they are formed to take images by dividing the internal half of the tire 10 into upper/middle/lower parts.

A 2D camera 450 is formed to be spaced apart from the internal 3D cameras 440, and the 2D camera 450 is formed in color.

In the tread imaging unit 500, a third pillar 510 is formed, a third column 520 is formed horizontally on one side of the third pillar 510 so as to be able to move vertically, and the third column ( A third arm 530 is formed to slide on the 520 .

A tread line scanner and a tread 3D camera 550 are respectively formed at the end of the third arm 530 to capture images of the tread of the tire 10 .

The tread line scanner captures a two-dimensional (2D) image of the tread portion of the tire 10 using a line camera.

The tread 3D camera 550 is a 3D camera that measures depth using a laser and captures an image to create a three-dimensional image.

The main processing unit 600 is formed with a plurality of data stores, a control device, and an arithmetic device capable of analyzing and learning defects through AI deep learning learning.

In the data storage, learning data and test data are stored, and they are learned. In the learning data, good product learning data and bad product learning data are stored, and in the test data, real good test data, real bad test data, and virtual bad test data are stored. Saved.

Here, the terms of data refer to defective data, which is data secured from the defective tire 10, good data, which is data secured from the non-defective tire 10, and defective data, which is data obtained by detecting only the defective area from the defective data. It can be defined as non-defective product detection data, which is data obtained by detecting only a certain area in the detection data and non-defective product data.

The May processing unit performs semantic segmentation on the captured image to recognize defects in the image of the tire 10 captured by each camera and recognizes the defect.

The method of detecting a defect by the main processing unit 600 includes an image acquisition step of acquiring a captured image; a decoding step of encoding the image obtained in the image acquisition step by performing atrous convolution using a multi-layer artificial neural network, and then decoding the image into a low-level image by using a decoder; a hierarchical classification encoding step of classifying the image by layer and performing convolution and encoding while extending and convolving the image in the image acquisition step; an image determination step of encoding the best defect image by overlapping the hierarchical classification encoding; It consists of a defect separation step of detecting and separating the defect part by deep learning the low-level video in the decoding step together with the defect image encoded in the image determination step.

The algorithm used in the deep learning uses ResNet DUC + HDC, DeepLab V3 +, RefineNet, HRNet or EfficientPS, and the separated defect image is used as defect detection data of the learning algorithm, as well as to distinguish between defective products and good products.

In the main processing unit 600, the AI deep learning algorithm comprises a dataset construction step of constructing a set using defect detection data; a learning step of comparing defective products with non-defective product detection data after the set constructing step and learning to distinguish defective products from non-defective products; a neural network model determination step of determining a neural network model through neural network classification in the learning step; a test set constructing step of constructing a test set with the good data and bad data; an analysis step of separating defects from the image using a defect detection method for the test set; an application step of applying the neural network algorithm determined in the neural network model determination step after the analysis step; After the applying step, it is composed of a determining step of outputting whether the tire 10 of the test set is a good product or a defective product.

5 to 7 are photographs showing defects separated by an automatic exterior inspection device for automobile tires 10 through AI deep learning using 2-dimensional and 3-dimensional images formed in a preferred embodiment of the present invention, according to the photographs A defect that cannot be detected by a 3D camera can be detected by a 2D camera, and a defect that a 2D camera cannot detect can be detected by a 3D camera, and in some cases, both types can be detected.

5 is an image of a state in which the sidewall of the tire 10 is photographed and subjected to semantic segmentation, and this image is the final image given by the contrast between good data and bad data and weight values by the neural network module. distinguish between defective and good products.

FIG. 6 shows an image of the tread portion of the tire 10, and FIG. 7 is an image obtained by taking an image of the internal (inside) of the tire 10 and performing semantic segmentation.

As described above, 2D and 3D cameras are built in various ways to take images, and even if you have a large amount of data, you can easily identify defective parts through AI deep learning, so you can quickly distinguish defective products from good products. effect appears.

According to the automatic inspection device for exterior appearance of a car tire through AI deep learning using 2-dimensional and 3-dimensional images formed in a preferred embodiment of the present invention, the inside, bead, sidewall, and tread of a tire are simultaneously photographed and images are acquired, so that the tire It is easy to find each defect that can occur, and the curved part of the tire is divided into upper and lower parts or upper and middle and lower parts, and each of them is photographed. Since the part is imaged using a 2D line camera and a 3D camera or a 2D RGB camera and a 3D camera, many defects that may occur in tires can be found, and the captured images captured by each camera can be used for AI learning Since it can be quickly contrasted with learned combinations, it is very fast to find defect data, and it is possible to quickly find defects that can occur by manufacturer using a deep learning method, so the processing speed is very high.

Although the present invention has been described with reference to the preferred embodiments with reference to the accompanying drawings, it is clear that various modifications are possible to those skilled in the art from this description without departing from the scope of the present invention covered by the claims to be described later.

10 : tire
100: bed 110: height adjustment
200: test bed 210: circular table
212: stopper 220: motor
300: side wall imaging unit 310: first pillar
320: first column 330: first arm
340: Sidewall upper line scanner 345: Sidewall upper 3D camera
350: Sidewall lower line scanner 355: Sidewall lower 3D camera
400: internal imaging unit 410: second pillar
420; Second column 430: second arm
440: Internal 3D camera 450: 2D camera
500: tread imaging unit 510: third pillar
520: third column 530: third arm
540: tread line scanner 550: tread 3D camera
600: main processing unit

Claims (5)

delete A bed 100 in which a height adjustment sphere 110 is formed at the bottom in a rectangular shape; A circular test bed 200 that is rotatable while being transportable up and down in the center of the bed 100; and a sidewall imaging unit 300 having at least two cameras formed thereon; an internal imaging unit 400 in which a plurality of cameras are formed to be spaced apart from the sidewall imaging unit 300; a tread imaging unit 500 in which a plurality of cameras are formed to be spaced apart from the internal imaging unit 400; A general two-dimensional and three-dimensional image consisting of a main processing unit 600 fixed to the top of the bed 100 and having a processing device capable of acquiring and analyzing images captured while controlling the position of each camera. In the used automobile tire exterior automatic inspection device,
The sidewall imaging unit 300 is formed at the end of the first arm 330, and the first arm 330 is spaced apart from the test bed 200 by placing a first pillar on the upper edge of the bed 100 ( 310) is formed, and a first column 320 formed horizontally with the first pillar 310 to be transportable up and down is formed, and is formed to be slidable to the first column 320,
The internal imaging unit 400 is formed at the end of the second arm 430, and the second arm 430 is spaced apart from the test bed 200 by placing a second pillar 410 on the upper edge of the bed 100. ) is formed, and a second column 420 that can be transported vertically along the second pillar 410 is formed, and is formed in an 'L' shape that can slide on the second column 420,
The tread imaging unit 500 is formed at the end of the third arm 530, and the third arm 530 has a third pillar 510 at the upper edge of the bed 100 so as to be spaced apart from the test bed 200. Is formed, a third column 520 is formed horizontally to be transportable up and down on one side of the third column 510, and is formed to be slidable at the end of the third column 520,

The main processing unit 600 is formed with a plurality of data stores, a control device, and an arithmetic device capable of analyzing and learning defects through AI deep learning learning.
The method of detecting a defect by the main processing unit 600 includes an image acquisition step of acquiring a captured image;
a decoding step of encoding the image obtained in the image acquisition step by performing atrous convolution using a multi-layer artificial neural network and then decoding the image into a low-level image by using a decoder;
a hierarchical classification encoding step of classifying the image by layer and performing convolution and encoding while extending and convoluting the image in the image acquisition step;
an image determination step of encoding the best defect image by overlapping the hierarchical classification encoding;
AI deep learning using two-dimensional and three-dimensional images, characterized in that it consists of a defect separation step of detecting and separating the defect part by deep learning the low-level image in the decoding step together with the defect image encoded in the image confirmation step. An automatic inspection device for automobile tire appearance through running. delete delete delete

Images