JP7709028B2 - Tire inspection device and inspection method, identification model generation method and generation device, and program - Google Patents

Description

This disclosure relates to a tire inspection device and inspection method, a method and device for generating an identification model, and a program.

The following Patent Document 1 discloses an apparatus for detecting defects inside a tire using shearography. In shearography, a laser beam is irradiated onto the tire surface, and a speckle image formed by the interference of the light reflected from the tire surface is captured by a camera. Two speckle images captured under two different conditions of temperature and pressure are subjected to image subtraction processing, and the difference between the speckle images is obtained as the measurement result.

For example, in Patent Document 1, a tire is placed in a pressure chamber. When the pressure inside the chamber is reduced, the tire surface deforms. If there are defects such as air bubbles in the rubber part of the tire, the air bubbles expand and the tire surface bulges locally. Image subtraction processing is performed on a speckle image captured before the pressure in the chamber is reduced and a speckle image captured after the pressure is reduced. This results in a measurement image (hereinafter referred to as a "shearography image") that shows the changes in the tire surface caused by the reduced pressure (local bulging). In this shearography image, interference fringes appear in the bulging areas.

Special Publication No. 2009-531690

In shearography images, defects inside a tire appear as stripes. There is a correlation between the size of the stripes (the size of the bulge that appears on the tire surface) and the size of the actual defect. However, the size of the stripes also depends on the tire's rigidity. For example, if the rigidity of the rubber part is high, even if the actual defect (e.g., an air bubble) is large, the localized bulge on the tire surface will be small, and the stripes that appear in the shearography image will also be small. For this reason, it is difficult to accurately determine defects based on the size of the stripes alone.

(1) A discrimination model is stored in the storage means of the tire inspection device proposed in this disclosure. The discrimination model is a model generated by machine learning using a dataset including an image of a defect area, which is an area where defects inside the tire appear in a shearography image, which is an image of the tire acquired by shearography, and tire information related to the material of the tire, as training data. The tire inspection device has an image acquisition means for acquiring a shearography image of the inspection target tire acquired by shearography, a tire information acquisition means for acquiring tire information related to the material of the inspection target tire, and a defect determination means for inputting a part or all of the shearography image of the inspection target tire and the tire information of the inspection target tire into the discrimination model and detecting internal defects of the inspection target tire.

The material of a tire affects its rigidity. In the tire inspection device of (1), tire information related to the material of the tire being inspected is also input into the identification model. This makes it possible to detect with high accuracy the presence or absence of defects in the tire being inspected.

(2) The tire includes a crown portion and a sidewall portion. In the tire inspection device described in (1), the data set for generating the identification model may include part information representing a part of the tire including the defective area in addition to the image of the defective area and the tire information. The defect determination means may input a part of the shearography image of the tire to be inspected, the tire information of the tire to be inspected, and part information representing a part of the tire to be inspected including the part of the shearography image to the identification model.

The crown and sidewall sections have different structures, so the rigidity and the shape of defects that occur may differ between these two sections. With the tire inspection device in (2), the data input to the identification model includes information about the section, which improves the accuracy of defect detection.

(3) The tire inspection device described in (1) may further include a defect area candidate extraction means for extracting an image of a defect area candidate, which is a candidate for an area containing an internal defect of the tire to be inspected, from the shearography image of the tire to be inspected. The defect determination means may input the image of the defect area candidate to the identification model and determine whether or not a defect appears in the defect area candidate. This can further improve the accuracy of defect detection.

(4) In the tire inspection device described in (3), the defect area candidate extraction means may search for a preregistered pattern in the shearography image of the tire to be inspected, and extract an area containing the pattern found by the search as the defect area candidate.

(5) The tire inspection method proposed in this disclosure includes an image acquisition step of acquiring a shearography image, which is an image of a tire to be inspected acquired by shearography, and a tire information acquisition step of acquiring tire information related to the material of the tire to be inspected. The identification model is a model generated by machine learning using a dataset including an image of a defect area, which is an area in the shearography image where a defect inside the tire appears, and tire information related to the tire material, as training data. The tire inspection method includes a defect determination step of inputting a part or all of the shearography image of the tire to be inspected and the tire information of the tire to be inspected into the identification model and detecting internal defects of the tire to be inspected.

The material of the tire affects the tire's rigidity. (5) In the tire inspection method, tire information related to the material of the tire being inspected is also input into the identification model. This makes it possible to detect with high accuracy the presence or absence of defects in the tire being inspected.

(6) The program proposed in this disclosure causes a computer to function as an image acquisition means for acquiring a shearography image, which is an image of a tire to be inspected acquired by shearography, and as a tire information acquisition means for acquiring tire information related to the material of the tire to be inspected. The program also causes the computer to function as a defect determination means. The defect determination means inputs a part or all of the shearography image of the tire to be inspected and the tire information of the tire to be inspected into an identification model generated by machine learning using a dataset including an image of a defect area, which is an area in the shearography image where a defect inside the tire appears, and tire information related to the material of the tire, as training data, and detects internal defects in the tire to be inspected.

(7) The method for generating an identification model proposed in this disclosure includes an image acquisition step of acquiring a shearography image, which is an image of a tire acquired by shearography; a defect area extraction step of extracting an image of a defect area, which is an area including internal defects of the tire, from the shearography image; a tire information acquisition step of acquiring tire information related to the material of the tire; and a model generation step of generating an identification model for detecting internal defects of an inspected tire using a dataset including the image of the defect area and the tire information as training data. In the method of (7), tire information related to the material of the tire is also used to generate the identification model. Therefore, inspection using this identification model makes it possible to detect the presence or absence of defects in an inspected tire with high accuracy.

(8) The tire includes a crown portion and a sidewall portion. The data set may include, in addition to the image of the defective area and the tire information, part information representing the part of the tire including the defective area. Since the crown portion and the sidewall portion have different structures, the two parts may have different rigidities and may have different types of defects. In the method of (8), the part information representing the part of the tire including the defective area is also used to generate an identification model. Therefore, inspection using this identification model can further improve the accuracy of defect detection.

(9) The device for generating an identification model proposed in the present disclosure includes an image acquisition means for acquiring a shearography image of a tire acquired by shearography, a defect area extraction means for extracting an image of a defect area, which is an area including internal defects of the tire, from the shearography image acquired in the image acquisition step, a tire information acquisition means for acquiring tire information related to the material of the tire, and a model generation means for generating an identification model for detecting internal defects of a tire to be inspected, using a dataset including the image of the defect area and the tire information as training data.

(10) The program proposed in this disclosure causes a computer to function as an image acquisition means for acquiring a shearography image of a tire acquired by shearography, a defect area extraction means for extracting an image of a defect area, which is an area including internal defects of the tire, from the shearography image, a tire information acquisition means for acquiring tire information related to the material of the tire, and a model generation means for generating an identification model for detecting internal defects of a tire to be inspected, using a dataset including the image of the defect area and the tire information as training data.

1 is a block diagram showing hardware of a tire inspection device proposed in the present disclosure. FIG. 1 is a schematic diagram of a shearography device. FIG. 2 is a diagram showing an example of an image obtained by a shearography device. FIG. 1 is a cross-sectional view illustrating an example of a tire structure. 2 is a block diagram showing functions of a control unit of the tire inspection device. FIG. 11 is a diagram for explaining a process executed by a control unit (defect area extraction unit); FIG. FIG. 11 is a diagram illustrating an example of a process executed by a control unit to generate an identification model. An example of a process executed by the control unit to detect a defect in a tire will be described.

The following describes the tire inspection device and inspection method proposed in this disclosure, as well as the method and device for generating an identification model.

Figure 1 is a block diagram showing the hardware of a tire inspection device 10 proposed in this disclosure. As shown in Figure 1, the tire inspection device 10 has a control unit 11, a display unit 13, an input unit 14, a shearography device 15, and a pressure reducer 17. The elements of the tire inspection device 10, such as the control unit 11, function not only as the tire inspection device 10, but also as a device for generating an identification model used by the tire inspection device 10.

As shown in FIG. 1, the shearography device 15 has a laser irradiation unit 15A, an interferometer 15B, and a light receiving unit 15C. FIG. 2 and FIG. 3 are diagrams for explaining the shearography method. FIG. 2 shows an outline of the shearography device, and FIG. 3 shows an example of an image obtained by the shearography device.

The laser irradiation unit 15A includes a laser diode and irradiates the surface S of the tire 90 with laser light. The irradiated laser light is reflected by the surface of the tire 90 and enters the interferometer 15B as shown in FIG. 2. For example, a Michelson interferometer can be used as the interferometer 15B. In FIG. 2, reference numeral 15a denotes a half mirror, and reference numerals 15b and 15c denote mirrors. The interferometer 15B may be a type of interferometer different from the Michelson interferometer. Reference numeral 15C denotes a light receiving unit. The light receiving unit 15C includes a CCD or a CMOS image sensor. The image captured by the light receiving unit 15C is supplied to the control unit 11.

The laser light reflected by the surface S of the tire 90 and incident on the interferometer 15B passes through two optical paths L1 and L2, and then merges again to reach the image surface (light receiving unit 15C). In the example of FIG. 2, optical path L1 is the path of light reflected by the half mirror 15a and reflected by one mirror 15b, and optical path L2 is the path of light transmitted through the half mirror 15a and reflected by the other mirror 15c. The laser light passing through the two optical paths L1 and L2 interferes at the light receiving unit 15C (image surface). The surface S of the tire 90 is microscopically rough, and the optical path difference generated at the interferometer 15B changes irregularly due to the unevenness of the surface S of the tire 90. Therefore, a speckle image with a spotted pattern is captured at the light receiving unit 15C.

The tire 90 is placed in a chamber. For example, a speckle image of the surface S of the tire 90 is captured while the chamber is kept at atmospheric pressure. Then, the pressure reducer 17 reduces the pressure in the chamber. If a defect such as an air bubble (solid line P1 in FIG. 2) occurs inside the tire 90, the air bubble P1 expands due to the reduced pressure in the chamber (see dashed line P1 in FIG. 2), and the surface S of the tire 90 locally bulges (see dashed line S in FIG. 2). Then, after such bulging occurs, a speckle image of the surface S is acquired. Then, a speckle pattern appears in this speckle image as well as before the pressure reduction. The brightness of the speckles appearing in the speckle image after the pressure reduction is the same as that of the speckle image before the pressure reduction at a position where the change in the optical path length caused by the local bulging of the surface S is an integer multiple of the wavelength of the laser light. On the other hand, at a position where the change in the optical path length is a half-integer multiple of the wavelength of the laser light, the brightness of the speckles after the pressure reduction is reversed from the brightness before the pressure reduction.

Therefore, when the difference between the speckle images before and after the reduction in pressure (the difference between the pixel values of each pixel) is calculated and an image is created from the difference, the bulge Sd on the surface S of the tire 90 appears, as shown in FIG. 3. In this specification, the image based on the difference between the speckle images is referred to as a "shearography image."

As shown in FIG. 4, the tire 90 has a crown portion 91 and left and right sidewall portions 93. A rubber portion 90a is formed in the crown portion 91 and the sidewall portion 93. A tread is formed on the surface of the rubber portion 90a (the surface of the crown portion 91). The tire 90 has a belt 94, a carcass 95, a finishing member 96, and the like inside the rubber portion 90a. The carcass 95 is formed from the left sidewall portion 93 to the right sidewall portion 93 of the tire 90. The belt 94 is disposed inside the crown portion 91 and covers the center portion of the carcass 95. The finishing member 96 is disposed on the edge portion of the sidewall portion 93 and covers the bead 97. The belt 94, the carcass 95, and the finishing member 96 are formed of metal. The belt 94 and the center portion of the carcass 95 are located in the crown portion 91. The carcass 95 and finishing member 96 are located in the sidewall portion 93.

The tire 90 is supported by a support device (not shown) that rotates it in the circumferential direction so that the shearography device 15 can obtain a shearography image of the entire tire 90. The support device rotates the tire 90 at a predetermined speed when generating a shearography image (teacher data) for machine learning and when inspecting the tire 90. The light receiving unit 15C continuously captures speckle images of the surface of the tire 90 at a frequency according to the rotation speed of the tire 90 and outputs image data for one circumference of the tire 90. In addition, the shearography device 15 may be movable relative to the tire 90 in a direction along the axis of the tire 90 so that shearography images of the crown portion 91 and sidewall portion 93 of the tire 90 can be obtained.

The control unit 11 (see FIG. 2) has a calculation device such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The control unit 11 also has a memory unit 12. The memory unit 12 has a RAM (Random access memory) and a ROM (Read only memory). The memory unit 12 may include a storage device that is both readable and writable, such as an SSD (Solid State Drive) or an HDD (Hard Disk Drive). The memory unit 12 stores programs executed by a calculation device such as a CPU, image data captured by the shearography device 15, and a discrimination model generated by processing described below.

The display unit 13 is a display device such as a liquid crystal display, and displays various images according to the instructions of the control unit 11.

The input unit 14 is a user interface such as a keyboard and mouse, which accepts user input operations and inputs signals indicating the contents of the input to the control unit 11.

[Processing performed in the control unit]
The functions of the control unit 11 will be described below. Fig. 5 is a block diagram showing the functions of the control unit 11. The control unit 11 has, as its functions, a shearography image acquisition unit 11a, a tire information acquisition unit 11b, a teacher data generation unit 11e, a model generation unit 11j, a defect area candidate extraction unit 11u, and a defect determination unit 11v. These functions are realized by the control unit 11 executing a program stored in the storage unit 12. The control unit 11 also has an identification model M. The identification model M is stored in the storage unit 12.

The shearography image acquisition unit 11a, the tire information acquisition unit 11b, the teacher data generation unit 11e, and the model generation unit 11j generate (learn) the identification model M. In addition, the shearography image acquisition unit 11a, the tire information acquisition unit 11b, the defect area candidate extraction unit 11u, and the defect determination unit 11v perform an inspection of the inside of the tire 90 using the identification model M.

The control unit 11 may be composed of multiple personal computers. Alternatively, the control unit 11 may be composed of one or multiple personal computers and one or multiple server computers. In this case, some of the functions of the control unit 11 (e.g., the defect area candidate extraction unit 11u and the defect determination unit 11v) may be executed by a personal computer, and other functions of the control unit 11 (e.g., the model generation unit 11j) may be executed by another personal computer or server computer.

[Shearography image acquisition section]
The shearography image acquisition unit 11a acquires a shearography image ( FIG. 3 ) calculated from a speckle image captured by the light receiving unit 15C. The shearography image acquisition unit 11a acquires shearography images for each of the crown portion 91 and the sidewall portion 93. Furthermore, when learning (generating) the identification model M, the shearography image acquisition unit 11a acquires shearography images for multiple tires 90 with different material compositions. This makes it possible to generate an identification model M that includes the material composition of the tire, in other words, tire rigidity, as an element for defect judgment.

The shearography image acquisition unit 11a may acquire a shearography image generated by another image processing device and stored in the memory unit 12. Alternatively, the shearography image acquisition unit 11a may acquire the above-mentioned two speckle images (speckle images before and after the reduction in pressure in the chamber) from the light receiving unit 15C, and generate a shearography image from the difference between the speckle images.

When inspecting the tire 90, the shearography image acquisition unit 11a acquires a shearography image of one circumference of the tire 90. On the other hand, when learning (generating) the identification model M, the shearography image acquisition unit 11a acquires a shearography image of the tire 90 for learning (an image for generating teacher data). In this case, the shearography image does not necessarily have to be one circumference of the tire 90.

[Tire information acquisition unit]
The tire information acquisition unit 11b acquires tire information related to the material of the tire 90. More specifically, the tire information is information that identifies the material of the rubber portion 90a of the tire 90. The tire information may be information that directly identifies the material of the rubber portion 90a of the tire 90. For example, the tire information may be the mass ratio of natural rubber and synthetic rubber, or the mass ratio of one or more compounding agents (carbon, silica, oil). The tire information may also be information that indirectly identifies the material of the rubber portion 90a. For example, when there is a one-to-one correspondence between a model number that identifies the type of tire 90 and the material of the rubber portion 90a, the model number may be used as the tire information.

[Process for generating a discrimination model]
A process for generating (learning) the discrimination model M using a shearography image will be described. A teacher data generating unit 11e and a model generating unit 11j shown in Fig. 5 execute the process for generating the discrimination model M. As shown in Fig. 5, the teacher data generating unit 11e has a defect area extracting unit 11f. This defect area extracting unit 11f has a candidate extracting unit 11g and a judgment result receiving unit 11h.

The candidate extraction unit 11g extracts areas (Im2, see FIG. 6) where there may be a defect in the rubber part 90a from the learning shearography image (Im1, see FIG. 6). Defects in the rubber part 90a include, for example, air bubbles P1 (see FIG. 2) in the rubber part 90a, separation (peeling of the rubber part 90a from the belt 94, etc.), and the inclusion of foreign matter in the rubber part 90a. Hereinafter, these defects are referred to as "rubber part defects." Furthermore, areas where there may be a rubber part defect are referred to as "defect area candidates."

The process by the candidate extraction unit 11g is executed, for example, as follows. As described with reference to FIG. 3, a pattern Sd corresponding to the shape and size of the defect in the rubber part 90a appears in the shearography image. The candidate extraction unit 11g searches the shearography image for each of a plurality of patterns (patterns showing defects in the rubber part) that are preregistered in the memory unit 12. For this search, for example, a similarity evaluation (matching using feature vectors) or pattern matching can be used.

As a method of evaluating the similarity, the candidate extraction unit 11g may, for example, calculate a feature vector (feature amount) for a portion of an area (referred to as an evaluation target area) in the shearography image, and calculate the similarity between the feature vector and a predetermined feature vector. Here, the feature vector may be, for example, the spatial frequency of a pattern. The candidate extraction unit 11g may scan the evaluation target area across the entire shearography image, and evaluate the similarity (similarity of spatial frequency) for each evaluation target area. Then, the candidate extraction unit 11g may extract evaluation target areas with a similarity greater than a threshold value as defect area candidates.

Furthermore, the candidate extraction unit 11g may search the entire shearography image for multiple patterns (patterns showing defects in the rubber part) that are pre-registered in the memory unit 12 as pattern matching. At this time, the Euclidean distance may be used as a value representing the similarity with the patterns registered in the memory unit 12. Then, when a pattern that matches a registered pattern is found in the shearography image, the circumscribing rectangle of the found pattern may be extracted as a candidate defect area.

As yet another method, the candidate extraction unit 11g may perform a two-dimensional Fourier transform on the entire shearography image. Then, frequency bands different from the frequencies corresponding to the multiple patterns (patterns showing rubber defects) pre-registered in the memory unit 12 may be cut from the shearography image. This will reveal areas where rubber defects may be present. The circumscribing rectangle of those areas may be extracted as a candidate defect area.

The judgment result receiving unit 11h displays images of the defect area candidates extracted by the candidate extraction unit 11g on the display unit 13. The judgment result receiving unit 11h then receives the worker's judgment result as to whether or not the displayed image actually represents a rubber part defect. The worker can input the judgment result through the input unit 14. When multiple defect area candidates are stored in the memory unit 12, the judgment result receiving unit 11h may display the multiple defect area candidates in sequence on the display unit 13 and receive the worker's judgment result for each defect area candidate.

The worker determines whether the extracted defective area candidate represents a rubber defect while taking into consideration the material composition of the learning tire 90 in addition to the size of the defective area candidate. For example, if the material of the tire 90 contains a large amount of compounding agent that increases the hardness of the rubber portion 90a, even if the size of the defective area candidate is small, the defect that actually occurs in the rubber portion 90a is likely to be large. In such a case, the worker may determine that this defective area candidate represents a rubber defect. Conversely, if the amount of compounding agent that increases the hardness of the rubber portion 90a is small, even if the size of the defective area candidate is large, the defect that actually occurs in the rubber portion 90a is likely to be small. In such a case, the worker may determine that this defective area candidate does not represent a rubber defect.

The crown portion and the sidewall portion have different structures. For example, the belt 94 is located in the crown portion 91, whereas the belt 94 is not located in the sidewall portion 93. Therefore, a defect in which the rubber portion 90a peels off from the belt 94 appears only in the crown portion 91. The worker may determine whether or not the defective area candidate is a rubber portion defect while taking into account the shape and location of the pattern appearing in the defective area candidate. In other words, the worker may determine whether or not the extracted defective area candidate corresponds to a rubber portion defect while taking into account the size of the defective area candidate, the material composition of the learning tire 90, and the location where the defective area candidate was found.

The defect area extraction unit 11f assigns an identification label to each defect area candidate based on the input judgment result. Specifically, when a judgment result indicating that a rubber part defect appears in the defect area candidate is input, the defect area extraction unit 11f assigns the identification label "defect" to the defect area candidate. (Hereinafter, the defect area candidate that is determined to have a rubber part defect is referred to as a "defect area.") Then, the defect area extraction unit 11f treats the image data of the defect area and the tire information acquired by the tire information acquisition unit 11b as one data set, and stores this data set and the identification label (defect) in the storage unit 12 in correspondence with each other. As another example, the defect area extraction unit 11f may treat the image data of the defect area, the tire information, and part information indicating the part where the defect area candidate was found (for example, the crown part 91 or the sidewall part 93) as one data set, and store this data set and the identification label (defect) in correspondence with each other in the storage unit 12.

On the other hand, when a judgment result indicating that the rubber part defect does not appear in the defective area candidate is input, the defective area extraction unit 11f assigns the identification label "normal" to the defective area candidate. (Hereinafter, the defective area candidate that is judged not to have a rubber part defect is referred to as a "normal area.") In this case, too, the defective area extraction unit 11f treats the image data of the defective area candidate (here, a normal area) and the tire information acquired by the tire information acquisition unit 11b as one data set, and stores this data set in the storage unit 12 in correspondence with the identification label (normal). As another example, the defective area extraction unit 11f may treat the image data of the defective area candidate (here, a normal area), the tire information, and part information indicating the part where the defective area candidate was found (for example, the crown part 91 or the sidewall part 93) as one data set, and store this data set in the storage unit 12 in correspondence with the identification label (normal).

The teacher data generating unit 11e may have a normal area extracting unit 11i as shown in FIG. 5. For example, the normal area extracting unit 11i extracts, as a normal area, a part of an area that was not extracted as a defective area candidate in the processing of the candidate extracting unit 11g. The normal area extracting unit 11i assigns an identification label "normal" to an image extracted as a normal area. The normal area extracting unit 11i treats the image data of the normal area and the tire information acquired by the tire information acquiring unit 11b as one data set, and stores this data set in the storage unit 12 in correspondence with the identification label (normal). In another example, the normal area extracting unit 11i may treat the image data of the normal area, the tire information, and part information indicating the part where the defective area candidate was found (for example, the crown part 91 or the sidewall part 93) as one data set, and store this data set in the storage unit 12 in correspondence with the identification label (normal).

The normal area extraction process can be performed, for example, as follows. The normal area extraction unit 11i randomly extracts an image having a predetermined number of pixels and a predetermined aspect ratio from the shearography image. Then, if the randomly extracted image does not have any overlapping portion with the image extracted as the defective area candidate, the normal area extraction unit 11i may determine that this extracted image is a normal area. Here, the predetermined number of pixels and the predetermined aspect ratio may be, for example, the average number of pixels and the average aspect ratio of the defective area candidate.

The teacher data generating unit 11e converts each of the image data of the multiple defective areas obtained by the above-mentioned process into image data having a predetermined number of pixels and a predetermined aspect ratio. (Hereinafter, this number of pixels will be referred to as the "model input pixel number", and this aspect ratio will be referred to as the "model input aspect ratio".) The teacher data generating unit 11e also converts the image data of the multiple normal areas obtained by the above-mentioned process into the model input pixel number and model input aspect ratio. In other words, the teacher data generating unit 11e unifies the dimensions of the teacher data.

The model generation unit 11j inputs a dataset including image data of the defective area as training data into the pre-learning discrimination model. In this way, the model generation unit 11j generates a discrimination model M for detecting defects in the internal structure of the tire 90. In addition to the dataset including image data of the defective area, the model generation unit 11j may also input a dataset including image data of normal areas as training data into the pre-learning discrimination model.

As the discrimination model M, for example, a neural network may be used. As the discrimination model M, a convolutional neural network (CNN) may be used. Alternatively, a support vector machine (SVM), a random forest, or the like may be used as the discrimination model M.

[Processing for tire inspection]
The process for inspecting a tire using a shearography image of the tire to be inspected will now be described. The defect area candidate extraction unit 11u and the defect determination unit 11v execute the process for inspection.

The defect area candidate extraction unit 11u extracts defect area candidates (Im2, see FIG. 6) from the image data of the tire 90 to be inspected, acquired by the shearography image acquisition unit 11a. The processing of the defect area candidate extraction unit 11u may be the same as the processing of the candidate extraction unit 11g of the teacher data generation unit 11e described above. That is, the defect area candidate extraction unit 11u searches the shearography image for each of the multiple patterns registered in advance in the memory unit 12.

The defect determination unit 11v inputs the image data of the defective area candidate extracted by the defective area candidate extraction unit 11u to the discrimination model M, and detects defects inside the tire 90 to be inspected. For example, the defect determination unit 11v inputs the image data of the defective area candidate to the discrimination model M, and classifies the defective area candidate into one of a plurality of classes. Here, the plurality of classes are, for example, a class in which a rubber part defect appears in the defective area candidate (a class to which the discrimination label "defect" is assigned), or a class in which a rubber part defect does not appear in the defective area candidate (a class to which the discrimination label "normal" is assigned). This process by the defect determination unit 11v makes it possible to detect defects in the internal structure of the tire 90 to be inspected. The output of the defect determination unit 11v is, for example, the probability that the defective area candidate falls into each class. Alternatively, the output of the defect determination unit 11v may be the discrimination label of the class to which the defective area candidate has the highest probability of falling.

The defect determination unit 11v may display the determination result (return value from the identification model M) on the display unit 13 together with the image of the defect area candidate input to the identification model M and part information indicating the part where the defect area candidate was found (crown part 91 or sidewall part 93) or store them in the memory unit 12. If multiple defect area candidates are found for one tire 90, the defect determination unit 11v may input all of the multiple defect area candidates in order to the identification model M and display each of the outputs on the display unit 13 or store them in the memory unit 12. If there is a candidate classified as a "defect" class among the multiple defect area candidates, the defect determination unit 11v may determine that the tire 90 is defective.

Alternatively, the identification model M may be generated to determine that even minute defects (defects that should be tolerated) are "defective." In this case, if the number of such minute defects exceeds a threshold value set for each part of the tire 90 (crown portion 91 and sidewall portion 93), that part may be determined to be defective. For example, if the number of defective areas found in the shearography image of the crown portion 91 is greater than the threshold value, the defect determination unit 11v may determine that the crown portion 91 is defective. Conversely, if the number of defective areas found in the shearography image of the sidewall portion 93 is greater than the threshold value, the defect determination unit 11v may determine that the sidewall portion 93 is defective.

[Process flow for generating a classification model]
Fig. 7 is a diagram showing an example of processing executed by the control unit 11 to generate an identification model M. The control unit 11 executes the processing shown in Fig. 7 for a plurality of tires 90 having different material compositions, and generates one identification model M using data obtained from these plurality of tires 90.

The control unit 11 acquires a shearography image and acquires tire information that identifies the tire material (S101). This process is executed by the shearography image acquisition unit 11a and tire information acquisition unit 11b described above.

Next, the teacher data generating unit 11e generates teacher data using the shearography image and tire information. Specifically, the candidate extracting unit 11g extracts areas where rubber defects may occur (defective area candidates) from the shearography image acquired in S101 (S102). The judgment result receiving unit 11h displays the extracted defective area candidates on the display unit 13. Then, the judgment result receiving unit 11h receives the worker's judgment result as to whether the displayed defective area candidates contain rubber defects, and assigns an identification label (defective or normal) according to the judgment result to the defective area candidates (S103).

The defective area extraction unit 11f associates a data set including the extracted image data with an identification label (normal or defective) and stores them in the storage unit 12 (S104). More specifically, for image data (defective area) to which the identification label "defect" has been assigned, the defective area extraction unit 11f stores the image data, the tire information acquired in S101, and part information indicating the part where the image data area was found as one data set, and stores the data set in the storage unit 12 in association with the identification label (defect). For image data (normal area) to which the identification label "normal" has been assigned, the defective area extraction unit 11f stores the image data, the tire information acquired in S101, and part information indicating the part where the image data area was found as one data set, and stores the data set in the storage unit 12 in association with the identification label (normal).

The normal area extraction unit 11i extracts, as a normal area, a portion of the area that was not extracted (selected) as a defective area candidate in S102 from the shearography image (S105). The normal area extraction unit 11i combines the image data of the normal area, the tire information, and the part information into one data set, and stores this data set in the storage unit 12 in association with the identification label (normal) (S106).

Next, the teacher data generating unit 11e determines whether a predetermined number of data sets have been prepared for each of the multiple classes by the processes of S104 and S106 (S107). Here, the multiple classes are, for example, a class that includes a rubber part defect (a class that has been assigned the identification label "defect") and a class that does not include a rubber part defect (a class that has been assigned the identification label "normal"). The predetermined number is a number that is deemed necessary to generate the identification model M. If the number of data sets for each part (crown part 91 and sidewall part 93) has not reached the predetermined number, the teacher data generating unit 11e returns to the process of S102 and executes the subsequent processes.

The teacher data generating unit 11e converts the number of pixels and the aspect ratio of each image data (defective areas and normal areas) stored in the memory unit 12 into the model input number of pixels and the model input aspect ratio described above, respectively (S108).

The model generation unit 11j inputs the data sets stored in the memory unit 12 in S104 and S106 to the discrimination model M, which has not yet finished learning, and generates (trains) the discrimination model M for detecting rubber part defects (S109). The above is an example of the process executed by the control unit 11 to generate the discrimination model M.

The processing performed by the teacher data generating unit 11e is not limited to the example shown in FIG. 7. For example, the control unit 11 does not need to extract the normal area in S105, or store the data set including the image data of the normal area in the storage unit 12 in S106. The control unit 11 may store only the data set including the image data of the defective area in the storage unit 12 in S104, and input only the data set including the image data of the defective area to the identification model in S109 to generate (learn) the identification model. In this case, a neural network or a convolutional neural network may be used as the identification model M. In the tire inspection process, the probability that the data set input to the identification model M corresponds to a "defect" may be output.

[Tire inspection process flow]
Next, an example of a process executed by the control unit 11 to detect defects in a rubber portion of a tire will be described with reference to FIG.

The control unit 11 acquires a shearography image and acquires tire information that identifies the tire material (S201). This process is executed by the shearography image acquisition unit 11a and tire information acquisition unit 11b described above.

The defect area candidate extraction unit 11u extracts areas where a rubber defect may occur (defect area candidate) from the shearography image acquired in S201 (S202). In S202, the defect area candidate extraction unit 11u extracts all defect area candidates from the shearography image. In addition, the defect area candidate extraction unit 11u converts the pixel count and aspect ratio of the extracted defect area candidate into a model input pixel count and model input aspect ratio (S203).

The defect determination unit 11v inputs the image data of the defect area candidate, the tire information acquired in S101, and the part information indicating the part where the defect area candidate was found into the identification model as one data set (S204). Then, the defect determination unit 11v classifies the defect area candidate into a plurality of classes (S205). The plurality of classes are, for example, a class including a rubber part defect (a class to which the identification label "defect" is assigned) or a class not including a rubber part defect (a class to which the identification label "normal" is assigned). The defect determination unit 11v determines whether or not there are any defect area candidates that have not yet been classified (S206). If there are any defect area candidates that have not yet been classified, the defect determination unit 11v returns to S204 and performs the subsequent processing on the unclassified defect area candidates.

If it is determined in S206 that classification of all defect area candidates extracted from the shearography image has been completed, the control unit 11 ends the process. In other words, the control unit 11 ends the inspection of the internal structure of the tire 90.

[summary]
The memory unit 12 of the tire inspection device 10 stores an identification model M. The identification model M is a model generated by machine learning using a data set including an image of a defective area in a shearography image and tire information related to the material of the tire as teacher data. The tire inspection device 10 has a shearography image acquisition unit 11a that acquires a shearography image of the inspection target tire 90, a tire information acquisition unit 11b that acquires tire information related to the material of the inspection target tire 90, and a defect determination unit 11b that inputs image data of a defective area candidate in the shearography image of the inspection target tire 90 and the tire information of the inspection target tire 90 to the identification model M and detects defects inside (rubber part 90a) of the inspection target tire 90. The material of the tire affects the rigidity of the tire. In the tire inspection device 10 described above, the tire information related to the material of the inspection target tire 90 is also input to the identification model M, so that the presence or absence of a defect in the inspection target tire 90 can be detected with high accuracy.

The identification model M is generated by machine learning using a data set including an image of the defective area, tire information, and part information representing the part of the tire including the defective area as training data. The defect determination unit 11v inputs image data of the defective area candidate extracted from the shearography image of the tire 90 to be inspected, the tire information of the tire 90 to be inspected, and part information representing the part of the tire 90 to be inspected including the defective area candidate to the identification model M. Since the crown and sidewall parts of the tire have different structures, the rigidity and the shape of the defect that occurs may be different between these two parts. In inspection using the tire inspection device 10, the data input to the identification model M includes part information, which improves the accuracy of defect detection.

In the method for generating the discrimination model M proposed in this disclosure, a shearography image of the tire is acquired, an image of a defective area inside the tire is extracted from the shearography image, tire information related to the tire material is acquired, and a discrimination model is generated using a dataset including the image of the defective area and the tire information as training data. In this method, tire information related to the tire material is also used to generate the discrimination model. Therefore, inspection using this discrimination model makes it possible to detect the presence or absence of defects in the tire to be inspected with high accuracy. Furthermore, in the method for generating the discrimination model M, the dataset input to the discrimination model includes part information representing the part of the tire including the defective area in addition to the image of the defective area and tire information. Inspection using this discrimination model M can further improve the accuracy of defect detection.

[others]
The tire inspection device 10 proposed in this disclosure is not limited to the above-mentioned tire inspection device.

For example, the model generation unit 11j may separately generate an identification model for detecting rubber defects in the crown portion 91 and an identification model for detecting rubber defects in the sidewall portion 93. In this case, in the tire inspection process, the defect area candidates extracted from the shearography image of the crown portion 91 may be input to the identification model for the crown portion 91. In addition, the defect area candidates extracted from the shearography image of the sidewall portion 93 may be input to the identification model for the sidewall portion 93.

In yet another example, the control unit 11 (teacher data generation unit 11e) may not have a normal region extraction unit 11i.

In the above explanation, a part of the shearography image (a candidate defect area) is extracted, and the identification model M is generated using this candidate defect area. Also, when inspecting tires, a part of the shearography image (a candidate defect area) is extracted, and this candidate defect area is input to the identification model M to detect tire defects. Alternatively, such candidate defect areas may not be extracted, and the entire shearography image may be input to the pre-learning identification model M to generate the identification model M. Also, when inspecting tires, the entire shearography image may be input to the identification model M to determine tire defects.

10: Tire inspection device, 11: Control unit, 11a: Shearography image acquisition unit, 11b: Tire information acquisition unit, 11e: Teacher data generation unit, 11f: Defect area extraction unit, 11g: Candidate extraction unit, 11h: Judgment result reception unit, 11i: Normal area extraction unit, 11j: Model generation unit, 11u: Defect area candidate extraction unit, 11v: Defect judgment unit, 12: Memory unit, 13: Display unit, 14: Input unit, 15: Shearography device, 15A: Laser irradiation unit, 15B: Interferometer, 15C: Light receiving unit, 15a: Half mirror, 15b, 15c: Mirror, 17: Pressure reducer, 90: Tire, 90a: Rubber part, 91: Crown part, 93: Sidewall part, 94: Belt, 95: Carcass, 96: Finishing member, 97: Bead, M: Identification model.

Claims (10)

A storage means for storing an identification model generated by machine learning using a dataset including an image of a tire obtained by shearography, the image being an image of a tire, an image of a defect area being an area where a defect inside the tire is shown in the shearography image, and tire information related to the material of the tire, as training data;
an image acquisition means for acquiring a shearography image of the tire to be inspected by shearography;
a tire information acquiring means for acquiring tire information related to a material of the inspection target tire;
a defect determination means for inputting a part or all of the shearography image of the tire to be inspected and the tire information of the tire to be inspected into the identification model and detecting internal defects of the tire to be inspected. The tire portion includes a crown portion and a sidewall portion,
The data set for generating the discrimination model includes, in addition to the image of the defect area and the tire information, part information representing a part of the tire including the defect area;
2. The tire inspection device according to claim 1, wherein the defect determination means inputs a portion of the shearography image of the inspection target tire, the tire information of the inspection target tire, and part information representing a part of the inspection target tire including the portion of the shearography image into the identification model. The present invention further includes a defect area candidate extraction means for extracting an image of a defect area candidate, which is a candidate of an area including an internal defect of the inspection target tire, from the shearography image of the inspection target tire,
The tire inspection device according to claim 1 , wherein the defect determination means inputs an image of the defect area candidate to the discrimination model and determines whether or not a defect appears in the defect area candidate. The tire inspection device according to claim 3, wherein the defect area candidate extraction means searches for a preregistered pattern in the shearography image of the tire to be inspected, and extracts an area containing the pattern found by the search as the defect area candidate. an image acquisition step of acquiring a shearography image, which is an image of the tire to be inspected acquired by shearography;
a tire information acquisition step of acquiring tire information related to a material of the inspection target tire;
a defect determination step of inputting a part or all of the shearography image of the tire to be inspected and the tire information of the tire to be inspected into an identification model generated by machine learning using as training data a dataset including an image of a defect area, which is an area in the shearography image where defects inside the tire are revealed, and tire information related to the tire material, and detecting internal defects of the tire to be inspected. an image acquisition means for acquiring a shearography image, which is an image of the tire to be inspected acquired by shearography;
a tire information acquisition means for acquiring tire information related to the material of the tire to be inspected; and a defect determination means for inputting a part or all of the shearography image of the tire to be inspected and the tire information of the tire to be inspected into an identification model generated by machine learning using as training data a dataset including an image of a defect area, which is an area in the shearography image where defects inside the tire are revealed, and tire information related to the tire material, and detecting internal defects of the tire to be inspected. an image acquisition step of acquiring a shearography image, which is an image of the tire acquired by shearography;
a defect area extraction step of extracting an image of a defect area, which is an area including a defect inside the tire, from the shearography image;
a tire information acquisition step of acquiring tire information related to a material of the tire;
and a model generation step of generating an identification model for detecting internal defects in an inspection target tire, using a dataset including the image of the defect area and the tire information as training data. The tire includes a crown portion and a sidewall portion.
The method for generating an identification model according to claim 7 , wherein the dataset includes part information representing a part of the tire including the defect area in addition to the image of the defect area and the tire information. an image acquisition means for acquiring a shearography image of the tire by shearography;
a defect area extraction means for extracting an image of a defect area, which is an area including a defect inside the tire, from the shearography image acquired by the image acquisition means ;
a tire information acquiring means for acquiring tire information related to a material of the tire;
and a model generation means for generating an identification model for detecting internal defects in an inspection target tire, using a dataset including the image of the defect area and the tire information as training data. an image acquisition means for acquiring a shearography image of the tire by shearography;
a defect area extraction means for extracting an image of a defect area, which is an area including a defect inside the tire, from the shearography image;
A program that causes a computer to function as: a tire information acquisition means that acquires tire information related to the tire material; and a model generation means that uses a dataset including the image of the defect area and the tire information as training data to generate an identification model for detecting internal defects in an inspection target tire.