Abnormal inspection system, abnormal inspection method, and storage medium

Abstract

An abnormality inspection system according to an embodiment includes: an image acquisition unit (first acquisition unit) that acquires shooting data of a cam unit; a first inspection unit that inspects whether there is a suspicion of abnormality in the cam unit by inputting the shooting data of the cam unit into a first learning model; and a second inspection unit that inspects whether there is an abnormality in the cam unit by inputting the shooting data of the cam unit determined by the first inspection unit to have a suspicion of abnormality into a second learning model.

Description

Technical Field

[0001] The present invention relates to an abnormality inspection system, an abnormality inspection method, and a storage medium Background Art

[0002] Industrial components need to be appropriately inspected to prevent low-quality components from being shipped out during their production processes. For example, Japanese Unexamined Patent Application Publication No. 2020-149578 describes a technique for determining the specification classification of components using a learning model learned by machine learning Summary of the Invention

[0003] An inspection system for inspecting components can determine abnormalities on the surface of a component by photographing the component and analyzing its image. Here, in the case where the cross-section of the component is a perfect circle, since the positional relationship between a part of a certain surface of the component, the camera for photographing the component, and the illumination for irradiating the surface remains unchanged, the position irradiated by light on the image also remains unchanged. Therefore, the inspection system can determine whether there is an abnormality in that part by analyzing a plurality of images obtained by photographing each part of the component by AI (Artificial Intelligence) or the like

[0004] However, for components such as cams whose cross-sections are not perfect circles (that is, components having parts where the distance from the surface to the rotation axis is not uniform but different), since the positional relationship between the camera for photographing the component and the illumination for irradiating the surface and the component changes, the position irradiated by light on the image changes. Therefore, the appearance of the surface in the image captured by the camera changes irregularly. Therefore, the above method cannot be used, and the inspection system needs to analyze the captured images one by one. In such a case, the inspection system may misjudge a part of the surface of the cam that is not actually abnormal as abnormal

[0005] The present invention is proposed to solve such problems, and an object thereof is to provide an abnormality inspection system, an abnormality inspection method, and a storage medium that can accurately detect abnormalities in a cam portion

[0006] An abnormality inspection system according to an exemplary embodiment of the present invention includes: a first acquisition unit that acquires shooting data of a cam portion; a first inspection unit that checks whether there is a suspicion of abnormality in the cam portion by inputting the shooting data of the cam portion into a first learning model; and a second inspection unit that checks whether there is an abnormality in the cam portion by inputting the shooting data of the cam portion determined by the first inspection unit to have a suspicion of abnormality into a second learning model different from the first learning model. Since the abnormality inspection system uses two different learning models to determine the abnormality of the cam portion, it is possible to improve the accuracy of abnormality inspection and accurately detect the abnormality of the cam portion

[0007] In addition, in the above abnormal inspection system, it is also possible that the first learning model is a model that outputs whether there is an appearance abnormality in the cam portion. When the output result indicates that there is an appearance abnormality in the cam portion, the first inspection unit determines that there is a suspicion of abnormality in the cam portion. The second learning model is a model that outputs the category of the appearance abnormality of the cam portion. The second inspection unit determines whether there is an abnormality in the cam portion based on the category of the appearance abnormality of the cam portion indicated by the output result. Since the second inspection unit does not perform the determination process when there is no appearance abnormality in the cam portion, the inspection based on the second inspection unit is only performed when necessary. Therefore, the abnormal inspection system can perform the inspection efficiently.

[0008] In addition, in the above abnormal inspection system, it is also possible that when the output result of the second learning model indicates that the appearance abnormality of the cam portion is at least one pattern among droplets, marks of a deburring brush, inspection marks, or grindstone patterns, the second inspection unit determines that there is no abnormality in the cam portion. Thus, the abnormal inspection system does not determine a mere appearance abnormality as a final abnormality when there is no substantial quality abnormality in the component, so over-detection can be suppressed.

[0009] In addition, in the above abnormal inspection system, it is also possible that the second inspection unit cuts out the block determined by the first inspection unit to have a suspicion of abnormality from the captured data and inputs the captured data of the cut-out block into the second learning model. Thus, since unnecessary blocks are excluded from the determination target in the second inspection unit, the influence of noise in the inspection by the second inspection unit can be reduced, and the inspection accuracy can be improved.

[0010] In addition, the above abnormal inspection system may also include: a second acquisition unit that acquires captured data of a journal portion provided on the same component as the cam portion; and a third inspection unit that checks whether there is an abnormality in the journal portion by inputting the captured data of the journal portion into a third learning model different from the first learning model and the second learning model. The journal portion is only inspected by the third inspection unit. Thus, in the inspection of the component, the number of inspections of the journal portion that is not the cam portion is not multiple times but only once, so the total amount of computational processing required in the inspection of the component can be reduced.

[0011] In addition, in the above abnormal inspection system, it is also possible that the first acquisition unit acquires captured data of the cam portion taken in a state where light is irradiated on the cam portion by the first illumination, and the second acquisition unit acquires captured data of the journal portion taken in a state where light is irradiated on the journal portion by a second illumination different from the first illumination. Since the cam portion and the journal portion are irradiated with light by illuminations adapted to their respective shapes, the first acquisition unit and the second acquisition unit can acquire captured data of a quality suitable for inspection. Therefore, the abnormal inspection system can detect abnormalities in the cam portion more accurately.

[0012] In addition, in the above abnormal inspection system, it is also possible that the first acquisition unit acquires the photographed data of the cam part in a state where the cam part is rotated by a rotation mechanism that axially supports and rotates the shaft of the cam part. Thereby, the abnormal inspection system can efficiently acquire images of different parts of the cam part, and thus can shorten the time required for the entire inspection.

[0013] The abnormal inspection method according to an exemplary embodiment of the present invention is executed by an abnormal inspection system, and includes: an acquisition step of acquiring photographed data of a cam part; a first inspection step of inspecting whether there is a suspicion of abnormality in the cam part by inputting the photographed data of the cam part into a first learning model; and a second inspection step of inspecting whether there is an abnormality in the cam part by inputting the photographed data of the cam part determined to have a suspicion of abnormality into a second learning model different from the first learning model. The abnormal inspection system uses two different learning models to determine the abnormality of the cam part, so that the accuracy of the abnormal inspection can be improved, and the abnormality of the cam part can be accurately detected.

[0014] A storage medium according to an exemplary embodiment of the present invention stores a program. The program causes a computer to execute the following steps: an acquisition step of acquiring photographed data of a cam part; a first inspection step of inspecting whether there is a suspicion of abnormality in the cam part by inputting the photographed data of the cam part into a first learning model; and a second inspection step of inspecting whether there is an abnormality in the cam part by inputting the photographed data of the cam part determined to have a suspicion of abnormality into a second learning model different from the first learning model. Through this program, the computer can use two different learning models to determine the abnormality of the cam part, so that the accuracy of the abnormal inspection can be improved, and the abnormality of the cam part can be accurately detected.

[0015] Through the present invention, it is possible to provide an abnormal inspection system, an abnormal inspection method, and a storage medium that can accurately detect the abnormality of a cam part. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The features, advantages, and technical and industrial significance of the exemplary embodiments of the present invention will be described with reference to the accompanying drawings, in which the same reference numerals denote the same components, wherein:

[0017] Figure 1 is a block diagram showing an example of the abnormal inspection system according to Embodiment 1.

[0018] Figure 2A is a detailed view showing an example of the inspection device of the abnormal inspection system according to Embodiment 1.

[0019] Figure 2B is a view showing an example of a component to be inspected according to Embodiment 1.

[0020] Figure 2C It is a schematic diagram showing an example of the positional relationship among the cam portion, the illumination for cam portion inspection, and the camera for cam portion inspection according to Embodiment 1.

[0021] Figure 3A It is a block diagram showing an example of the control portion according to Embodiment 1.

[0022] Figure 3B It is a schematic diagram showing the relationship between the inspection portion and the learning model according to Embodiment 1.

[0023] Figure 3C It is a schematic diagram showing an example of image cutting according to Embodiment 1.

[0024] Figure 4A It is a flowchart showing an example of the processing when the abnormality inspection system according to Embodiment 1 performs the inspection of the cam portion.

[0025] Figure 4B It is a flowchart showing an example of the processing when the abnormality inspection system according to Embodiment 1 performs the inspection of the cam portion.

[0026] Figure 4C It is a flowchart showing an example of the processing when the abnormality inspection system according to Embodiment 1 performs the inspection of the joint portion. Detailed Embodiment

[0027] Embodiment 1

[0028] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0029] <Abnormality Inspection System>

[0030] Figure 1 It is a diagram for explaining the abnormality inspection system according to the embodiment. As Figure 1 shown, the abnormality inspection system S1 according to the present embodiment includes an inspection device 10 and an information processing device 20. The abnormality inspection system S1 according to the present embodiment causes the camera of the inspection device 10 to photograph the surface of the inspection target component by the user operating the information processing device 20, and causes the information processing device 20 to inspect whether there is an abnormality on the surface. In this example, the inspection target component is a camshaft. Hereinafter, the details of the inspection device 10 and the information processing device 20 will be described.

[0031] The inspection device 10 includes an illumination 11 for cam portion inspection, a camera 12 for cam portion inspection, an illumination 13 for journal portion inspection, a camera 14 for journal portion inspection, and a rotation motor 15.

[0032] Figure 2AFIG. 0 is a detailed view showing an example of the inspection apparatus 10, and each component of the inspection apparatus 10 will be described using this figure. As the illumination 11 for cam portion inspection, an illumination 11A for cam portion inspection is provided on the front side of the component W to be inspected, and an illumination 11B for cam portion inspection is provided on the rear side thereof. When the information processing apparatus 20 inspects the cam portion of the component W to be inspected, the respective illuminations are turned on. As the cameras 12 for cam portion inspection, cameras 12A - 12D for cam portion inspection are provided from the front side to the rear side of the component W to be inspected. Each camera 12 for cam portion inspection, under the control of the information processing apparatus 20, continuously takes pictures of the area of the cam portion irradiated with light by the illumination 11 for cam portion inspection a plurality of times, thereby taking picture data (images) of the cam portion.

[0033] The illumination 13 for journal portion inspection is turned on when inspecting the journal portion of the component W to be inspected under the control of the information processing apparatus 20. As the cameras 14 for journal portion inspection, cameras 14A - 14D for journal portion inspection are provided from the front side to the rear side of the component W to be inspected. Each camera 14 for journal portion inspection, under the control of the information processing apparatus 20, continuously takes pictures of the area of the journal portion irradiated with light by the illumination 13 for journal portion inspection a plurality of times, thereby taking picture data of the journal portion. In addition, the cameras 12 for cam portion inspection and the cameras 14 for journal portion inspection are arranged as so-called area cameras in which the imaging elements of the cameras are planar (i.e., having a plurality of imaging elements in the horizontal and vertical directions).

[0034] Figure 2B FIG. 7 is an enlarged view of the camshaft as the component W to be inspected. This camshaft has eight cam portions C1 - C8 and four journal portions J1 - J4. The illumination 11 for cam portion inspection irradiates light onto the cam portions C1 - C8, and the cameras 12 for cam portion inspection take pictures of the irradiated area. The illumination 13 for journal portion inspection irradiates light onto the journal portions J1 - J4, and the cameras 14 for journal portion inspection take pictures of the irradiated area.

[0035] Figure 2CIt is a schematic diagram showing an example of the positional relationship among the cam portion C1, the illumination 11A for cam portion inspection, and the camera 12A for cam portion inspection. The front portions of the illumination and the camera of the illumination 11A for cam portion inspection and the camera 12A for cam portion inspection are arranged on the same line, and the cam portion C1 is arranged on its extension line. That is, the illumination 11A for cam portion inspection and the camera 12A for cam portion inspection are arranged so as to have the same attitude (in-phase) with respect to the cam portion C1. The illumination 11A for cam portion inspection irradiates light on the cam portion C1. The camera 12A for cam portion inspection has an angle θ as the viewing angle for photographing the rotation direction of the cam portion, and photographs the photographing area RC of the cam portion. Not only for the cam portion C1, but also for other cam portions, the illumination 11A for cam portion inspection and the camera 12A for cam portion inspection are arranged in the same positional relationship, and can similarly photograph a prescribed photographing area of the cam portion. In addition, the illumination 11B for cam portion inspection and the camera 12B for cam portion inspection are also arranged in the same positional relationship with respect to each cam as the illumination 11A for cam portion inspection and the camera 12A for cam portion inspection, and can similarly photograph a prescribed photographing area of the cam portion.

[0036] Return to Figure 2A , and continue the description. The rotation motor 15, together with the fixture FC provided on the front side of the inspection object part W and the fixture RC provided on the rear side, constitutes a rotation mechanism for axially supporting and rotating the axis of the inspection object part W. After the user makes the inspection object part W rotatable by fixing both ends of the inspection object part W with the fixture FC and the fixture RC, the rotation motor 15 rotates the inspection object part W under the control of the information processing device 20.

[0037] Specifically, when performing the inspection of the cam portion of the inspection object part W, the information processing device 20 rotates the rotation motor 15, and during the period when the inspection object part W rotates one week with the illumination 11 for cam portion inspection lit, controls the camera 12 for cam portion inspection to continuously photograph the cam portion multiple times. Thus, the camera 12 for cam portion inspection photographs different areas of the cam portion at each photographing.

[0038] After photographing the cam portion as shown above, the information processing device 20 turns off the illumination 11 for cam portion inspection, and instead turns on the illumination 13 for journal portion inspection. Then, the information processing device 20 rotates the rotation motor 15, and during the period when the inspection object part W rotates one week, controls the camera 14 for journal portion inspection to continuously photograph the journal portion multiple times. Thus, the camera 14 for journal portion inspection photographs different areas of the journal portion at each photographing.

[0039] The information processing device 20 determines the shooting interval and the number of shots (shooting time) of the cam part inspection camera 12, and the rotation speed of the rotation motor 15 in such a way that the cam part inspection camera 12 performs multiple shots to capture the entire surface of the cam part of the inspection target part W. In addition, the information processing device 20 determines the shooting interval and the number of shots (shooting time) of the journal part inspection camera 14, and the rotation speed of the rotation motor 15 in such a way that the journal part inspection camera 14 performs multiple shots to capture the entire surface of the journal part of the inspection target part W. The information processing device 20 acquires the images of the cam part and the journal part obtained in the above manner, and determines whether there are abnormalities in each image by performing the processing described later.

[0040] In addition, in the images continuously captured by the cam part inspection camera 12, there may or may not be an overlapping part between the area of the cam part captured in one image and the area of the cam part captured in the images before and after it. The same can be said for the shooting of the journal part by the journal part inspection camera 14.

[0041] Next, the information processing device 20 will be described. The information processing device 20 includes a DB (Database) 21, a display panel 22, an input unit 23, and a control unit 24.

[0042] Three learning models required for inspection are stored in the DB 21. The learning model is an AI (Artificial Intelligence) model obtained by previously learning images as training data through machine learning such as deep learning. The details of this learning model will be described later. In addition, all the thresholds used in the determination of the inspection unit described later are stored in the DB 21.

[0043] The DB 21 is composed of a storage device such as a flash memory, a memory card, an HDD (Hard Disk Drive), an optical disk drive, etc., but the type of the storage device is not limited to this. In addition, the DB 21 can also be provided outside the information processing device 20. In this case, the information processing device 20 can be connected to the DB 21 via an information transceiver (not shown) to obtain the data stored in the DB 21.

[0044] The display panel 22 is an interface for enabling the user to view the determination result of whether there is an abnormality in the information processing device 20. The input unit 23 is an interface for the user to input instructions related to the start of the inspection and the setting of the inspection to the information processing device 20.

[0045] The control unit 24 controls the cam portion inspection illumination 11, the cam portion inspection camera 12, the journal portion inspection illumination 13, the journal portion inspection camera 14, and the rotation motor 15 as described above, and photographs the cam portion and the journal portion of the component W to be inspected. Then, the control unit 24 acquires the images respectively photographed by the cam portion inspection camera 12 and the journal portion inspection camera 14, and performs inspections as described below.

[0046] Figure 3A FIG. is a block diagram for explaining the configuration of the control unit 24. The control unit 24 includes a memory 241, an I / O (Input / Output) unit 242, and an information processing unit 243. Hereinafter, each unit of the control unit 24 will be described.

[0047] The memory 241 is composed of a volatile memory, a non-volatile memory, or a combination thereof. The memory 241 is not limited to one, and a plurality of them may be provided. In addition, the volatile memory may be, for example, a RAM (Random Access Memory) such as a DRAM (Dynamic Random Access Memory) or an SRAM (Static Random Access Memory). The non-volatile memory may be, for example, a PROM (Programmable ROM), an EPROM (Erasable Programmable Read Only Memory), or a flash memory.

[0048] This memory 241 is used to store one or more commands. Here, the one or more commands are stored in the memory 241 as a group of software modules. The information processing unit 243 can perform the following processing by reading and executing one or more commands from the memory 241.

[0049] The I / O unit 242 is a hardware interface that performs input / output of information with the outside of the control unit 24. In the present embodiment, the control unit 24 is connected to the cam portion inspection illumination 11, the cam portion inspection camera 12, the journal portion inspection illumination 13, the journal portion inspection camera 14, and the rotation motor 15, and appropriately performs input / output of information with them via the I / O unit 242.

[0050] The information processing unit 243 is composed of an arbitrary processor or the like for analyzing images. In this example, as the processor, the information processing unit 243 includes a GPU (Graphics Processing Unit) useful for image processing. However, as the processor, the information processing unit 243 may also have a CPU (Central Processing Unit), MPU (MicroProcessing Unit), FPGA (Field-Programmable Gate Array), DSP (Digital Signal Processor), or ASIC (Application Specific Integrated Circuit). In addition, the above-mentioned memory 241 may include not only the memory provided outside the information processing unit 243 but also the memory built in the information processing unit 243.

[0051] The information processing unit 243 realizes the functions of the image acquisition unit 244, the first inspection unit 245, the second inspection unit 246, the third inspection unit 247, etc. by reading and executing software (computer program) from the memory 241. The image acquisition unit 244 acquires the captured images from the cam unit inspection camera 12 and the journal inspection camera 14 via the I / O unit 242. The acquired image of the cam unit is output to the first inspection unit 245, and the image of the journal is output to the third inspection unit 247.

[0052] Figure 3B Shows the first inspection unit 245 - the third inspection unit 247, and three learning models stored in the DB 21 and respectively accessed by the first inspection unit 245 - the third inspection unit 247 for inspection. Next, refer to Figure 3B for a detailed description of each inspection unit.

[0053] The first inspection unit 245 inputs the image (first captured data) of the cam unit captured by the cam unit inspection camera 12 into the first learning model M1, and determines whether there is a suspicion of abnormality in the cam unit according to the result output from the first learning model M1.

[0054] Here, the first learning model M1 is a model obtained by learning using an image of the cam portion as training data. In this model, as the output result when an image is input, a calculated value related to whether there is an abnormality in the appearance of the cam portion is used as the output result. As a detailed example, the first learning model M1 calculates a first determination value of the input image by performing semantic segmentation processing on the input image and outputs it. This first determination value represents the normality of the cam portion reflected in the image. The larger the value, the cleaner the surface of the cam portion in the image; the smaller the value, the more a pattern similar to a defect such as a scratch or a shrinkage cavity is reflected on its surface. The first inspection unit 245 compares the calculated first determination value with a threshold TH1 stored in the DB 21, and determines an image with a first determination value less than or equal to the threshold TH1 as an image in which the cam portion has an appearance abnormality. On the other hand, the first inspection unit 245 determines an image with a first determination value greater than the threshold TH1 as an image in which the cam portion has no appearance abnormality.

[0055] In addition, the first learning model M1 can also divide the image into multiple blocks and calculate the first determination value for each block. In this case, the first inspection unit 245 can determine the image as an image in which the cam portion has an appearance abnormality when the first determination value of at least one block of the image is less than or equal to the threshold TH1. Or, the first inspection unit 245 can also determine the image as an image in which the cam portion has an appearance abnormality when the first determination values of more than one threshold number of blocks in the image are less than or equal to the threshold TH1.

[0056] An appearance abnormality means that the surface of the cam portion is not in a clean state. For example, it means that a pattern such as a line or a circle can be visually recognized on the surface, or there are parts with different brightness on the surface (for example, there are parts darker than the surroundings). When there is a pattern such as a circle on the surface of the cam portion, or when there is a dark part on the surface, there may be shrinkage cavities on the surface. When there is a line pattern on the surface, there may be scratches on the surface. Defects and shrinkage cavities mean defects (substantive abnormalities) of the component W to be inspected. However, even if there is a pattern such as a circle on the surface of the cam portion, this pattern may originate from droplets (such as cleaning liquid) on the surface. In addition, even if there is a line pattern on the surface, this pattern may be the trace of a deburring brush, the inspection trace of the component W to be inspected, or the grindstone pattern generated during grindstone processing. These patterns are naturally generated through the processing in the manufacturing stage of the component W to be inspected and are not defects of the component W to be inspected. Therefore, there may be a problem of over-detection as follows: the abnormality inspection system determines an image with such a pattern that has an appearance abnormality but is not a substantive abnormality (hereinafter also referred to as a false abnormality) as an abnormal image.

[0057] In the present invention, when an abnormality in appearance is determined in the image for the cam portion, the first inspection unit 245 determines that there is a suspicion of an abnormality in the cam portion. The image determined to have a suspicion of an abnormality becomes the object of re-inspection by the second inspection unit 246. When it is determined through this re-inspection that a defect is reflected in the image, first, it is determined that the inspection target component W is abnormal. In addition, the images determined not to have a suspicion of an abnormality are excluded from the inspection target of the second inspection unit 246.

[0058] The second inspection unit 246 inputs all the images of the images (second captured data) of the cam portion determined to have a suspicion of an abnormality to the second learning model M2. In addition, when the first learning model M1 is a model that divides an image into a plurality of blocks and calculates a first determination value for each block, the second inspection unit 246 may also cut out the blocks of the image whose first determination value is equal to or less than the threshold value TH1, and input only the images of the cut-out blocks to the second learning model M2.

[0059] Figure 3C FIG. is an example of a diagram showing the process of the second inspection unit 246 cutting out an image. In the original image IM captured by the cam portion inspection camera 12, since there is a pattern DA with a suspicion of an abnormality, the first inspection unit 245 determines that the first determination value of the block DA of the image IM is equal to or less than the threshold value TH1. At this time, the second inspection unit 246 may cut out the block DA from the image IM and input only the image of the block DA to the second learning model M2.

[0060] The second learning model M2 outputs a calculation result based on the image input as described above. The second inspection unit 246 determines whether there is an abnormality in the cam portion based on the output result from the second learning model M2.

[0061] Here, the second learning model M2 is an AI model of a different type from the first learning model M1. The second learning model M2 is a model obtained by learning using the image of the cam portion as training data. In this model, as the output result when an image is input, a calculated value related to the category of the abnormality on the appearance of the cam portion is used as the output result. Specifically, the second learning model M2 calculates the second determination value of the input image by performing classification processing on the input image and outputs it. This second determination value is the coincidence rate between the input image and the pattern of the false abnormality described above, and refers to the cosine value (cosθ) when calculating the inner product of the input image and the pattern of the false abnormality. The second learning model M2 calculates the number of second determination values corresponding to the number of types of the modeled false abnormality patterns. The larger this second determination value (closer to 1), the more similar the pattern on the surface of the cam portion in the image is to the pattern of the false abnormality (for example, at least one of the patterns of droplets, marks of a deburring brush, inspection marks, or grindstone patterns). There is no substantial abnormality. On the other hand, the smaller the second determination value, the less similar the pattern on the surface of the cam portion is to the pattern of the false abnormality. Therefore, the possibility that this pattern is a defect such as a scratch or a blowhole is considered to be higher.

[0062] The second inspection unit 246 compares the calculated second determination value with the threshold TH2 stored in the DB 21. In addition, the threshold TH2 is also set to the number corresponding to the number of types of the modeled false abnormality patterns. The second inspection unit 246 determines an image with an abnormality in the cam portion as an image in which the second determination value is equal to or less than the threshold TH2 for all types of the false abnormality patterns. On the other hand, the second inspection unit 246 determines an image without an abnormality in the cam portion as an image in which the second determination value is larger than the threshold TH2 for at least one type of the false abnormality pattern.

[0063] In addition, when the second inspection unit 246 inputs only a block image obtained by cutting out a part of the captured original image into the second learning model M2, the second learning model M2 performs the above classification processing on the block, calculates the second determination value of the block, and outputs it.

[0064] The first inspection unit 245 and the second inspection unit 246 perform the above processing on each image captured by each cam unit inspection camera 12. Then, when there is no image determined by the second inspection unit 246 to have an abnormality in the cam unit, the second inspection unit 246 determines that the inspected object, i.e., the cam unit, has no abnormality. On the other hand, when there is even one image determined to have an abnormality in the cam unit, the second inspection unit 246 determines that the inspected object, i.e., the cam unit, has an abnormality. However, the second inspection unit 246 may also determine that the inspected object, i.e., the cam unit, has an abnormality when the number of images determined to have an abnormality in the cam unit exceeds a specified threshold greater than 1. When the second inspection unit 246 determines that the inspected object, i.e., the cam unit, has an abnormality, it determines that the inspected object component W has an abnormality.

[0065] The second inspection unit 246 can display the above determination result on the display panel 22 of the information processing device 20. In addition, for the cam unit determined to have an abnormality, the second inspection unit 246 can also identify it by determining the cam unit inspection camera 12 that captured the cam unit and its shooting direction, and display the identification result on the display panel 22.

[0066] In addition, the third inspection unit 247 inputs the image of the journal part (the third shooting data) captured by the journal part inspection camera 14 into the third learning model M3, and determines whether there is an abnormality in the journal part according to the result output from the third learning model M3. For example, the third inspection unit 247 connects multiple images of the same part of the journal part captured in all the images of the journal part captured in the order of the shooting time to generate one image. Then, one image is generated for each different part in a manner that covers the surface of the journal part, and this image is input into the third learning model M3.

[0067] Here, the third learning model M3 is an AI model different from the first learning model M1 and the second learning model M2. The third learning model M3 is a model obtained by learning using the image of the journal part as training data. In this model, as the output result when the image is input, a calculated value related to whether there is an abnormality in the journal part is used as the output result. Specifically, the third learning model M3 calculates a third determination value of the image by performing a segmentation process on the input image and outputs it. This third determination value represents the normality of the journal part reflected in the image. The larger the value, the cleaner the surface of the journal part in the image; the smaller the value, the more a pattern similar to defects such as scars or pores is reflected on its surface. The third inspection unit 247 compares the calculated third determination value with the threshold TH3 stored in the DB 21, and determines an image with an abnormality in the journal part as an image whose third determination value is below the threshold TH3. On the other hand, the third inspection unit 247 determines an image with a third determination value greater than the threshold TH3 as an image without an abnormality in the journal part.

[0068] The third inspection unit 247 can perform the above-described image determination process on each part image of the surface of the journal part captured by each cam part inspection camera 12. Then, when there is no image determined by the third inspection unit 247 to have an abnormality in the journal part, the third inspection unit 247 determines that the inspected object, i.e., the journal part, has no abnormality. On the other hand, when there is even one image determined to have an abnormality in the journal part, the third inspection unit 247 determines that the inspected object, i.e., the journal part, has an abnormality. However, the third inspection unit 247 can also determine that the inspected object, i.e., the journal part, has an abnormality when the number of images determined to have an abnormality in the journal part is greater than a threshold value of 1. When the third inspection unit 247 determines that the inspected object, i.e., the journal part, has an abnormality, it determines that the inspected object component W has an abnormality.

[0069] The third inspection unit 247 can display the above determination results on the display panel 22 of the information processing device 20. In addition, for the journal part determined to have an abnormality, the third inspection unit 247 can also discriminate by determining the journal part inspection camera 14 that captured the journal part and its shooting direction, and display the discrimination result on the display panel 22.

[0070] Figure 4A 、 4B is a flowchart showing an example of the process of the abnormality inspection system S1 performing the inspection of the cam part. Hereinafter, with reference to Figure 4A 、 4B this process will be described. In addition, the details of each process are as described above, and the description will be appropriately omitted.

[0071] First, the user operates the input unit 23, whereby the information processing device 20 causes the inspection device 10 to perform an inspection of the cam portion of the component W to be inspected. By performing this inspection, the image acquisition unit 244 reads all the images captured by the cam portion inspection camera 12 (step S11).

[0072] The first inspection unit 245 inputs one of the read images into the first learning model M1. Then, it determines whether the first determination value calculated by the first learning model M1 is less than or equal to the threshold value TH1 (step S12).

[0073] When the first determination value is less than or equal to the threshold value TH1 (Yes in step S12), the second inspection unit 246 cuts out a block of the image for which the first determination value is less than or equal to the threshold value TH1 (step S13). Then, the second inspection unit 246 inputs the cut-out block image into the second learning model M2. The second inspection unit 246 determines whether the second determination values calculated by the second learning model M2 are all less than or equal to the threshold value TH2 for all types of false anomaly patterns (step S14).

[0074] When the second determination values are all less than or equal to the threshold value TH2 for all types of false anomaly patterns (Yes in step S14), the second inspection unit 246 determines, based on the determination result of this image, that there is an anomaly in the cam portion, and the anomaly inspection system S1 ends the inspection process (step S15).

[0075] On the other hand, in step S12, when the first determination value is greater than the threshold value TH1 (No in step S12), the first inspection unit 245 determines that the image of the object to be inspected is normal (no anomaly) (step S16). Also, in step S14, when the second determination value is greater than the threshold value TH2 for at least one type of false anomaly pattern (No in step S14), the second inspection unit 246 also determines that the image of the object to be inspected is normal (step S16).

[0076] Then, the first inspection unit 245 determines whether the above-described determination has ended for all the images read in step S11 (step S17). When there is an image for which the determination has not ended (No in step S17), the third inspection unit 247 returns to step S12 and performs processing on the undetermined image. On the other hand, when the determination for all the images has ended (Yes in step S17), the first inspection unit 245 determines that the cam portion is normal, and the anomaly inspection system S1 ends the inspection process (step S18).

[0077] In addition, in the process shown above, it is described that the information processing device 20 performs inspections based on the first inspection unit 245 and the second inspection unit 246 on one image, and sequentially performs such inspections on each image. However, the information processing device 20 may also perform inspections based on the first inspection unit 245 on multiple images (for example, all the read images) in a lump, and perform inspections based on the second inspection unit 246 on the images whose first determination value is below the threshold TH1 among them.

[0078] Figure 4C is a flowchart showing an example of the process in which the abnormality inspection system S1 performs an inspection of the joint part. Hereinafter, with reference to Figure 4C this process will be described. In addition, the detailed situation of each process is as described above, and the description will be appropriately omitted.

[0079] First, the user operates the input unit 23, whereby the information processing device 20 causes the inspection device 10 to perform an inspection of the shaft neck part of the inspection object part W. By performing this inspection, the image acquisition unit 244 reads all the images taken by the shaft neck inspection camera 14 (step S21).

[0080] The third inspection unit 247 uses all the read images of the shaft neck to generate one image for each part of the shaft neck surface. The detailed situation is as described above. Then, the third inspection unit 247 inputs this image into the third learning model M3 and determines whether the third determination value calculated by the third learning model M3 is below the threshold TH3 (step S22).

[0081] When the third determination value is below the threshold TH3 ( "Yes" in step S22), the third inspection unit 247 determines that there is an abnormality in the shaft neck based on the determination result of this image, and the abnormality inspection system S1 ends the inspection (step S23).

[0082] On the other hand, when the third determination value is greater than the threshold TH3 ( "No" in step S22), the third inspection unit 247 determines that the image of the inspection object is normal (step S24).

[0083] Then, the third inspection unit 247 determines whether the above - shown determination has ended for all the images related to different parts of the shaft neck surface (step S25). When there are images for which the determination has not ended ( "No" in step S25), the third inspection unit 247 returns to step S22 and performs processing on the undetermined images. On the other hand, when the determination for all the images has ended ( "Yes" in step S25), the third inspection unit 247 determines that the shaft neck is normal, and the abnormality inspection system S1 ends the inspection process (step S26).

[0084] As described above, in the abnormality inspection system S1, the first inspection unit 245 checks whether there is a suspicion of abnormality in the cam part by inputting the photographed data of the cam part into the first learning model M1. Then, the second inspection unit 246 checks whether there is an abnormality in the cam part by inputting the photographed data of the cam part determined to have a suspicion of abnormality into the second learning model M2.

[0085] In the case of inspecting a special part such as a cam part whose shape is not a perfect circle, since the method such as the inspection of the journal part described above cannot be applied, the abnormality inspection system needs to inspect a large number of photographed images of the cam part one by one. The applicant newly found the following problem: in such a case, if the inspection system uses one AI model to perform the determination of the cam part, since the AI model is learned in a way that suppresses the omission of defects such as air holes or scratches, the over-detection rate in the determination becomes high. The reason is that the threshold value used for abnormality determination in the AI model is learned in a way that suppresses the omission of detection, and as a result, for patterns such as droplets and grinding marks (patterns of false abnormalities) that have a shape or brightness similar to the pattern of defects, the AI model also determines them as abnormal.

[0086] On the other hand, the applicant also newly found that if the AI model is learned in a way that suppresses the over-detection rate, although the AI model can more accurately identify the patterns of false abnormalities, a new problem of an increased omission rate of abnormalities will occur. This is because there may be a situation where the AI model misjudges the pattern of the defect of the cam part as the pattern of a false abnormality, or in the case where the pattern of the defect and the pattern of the false abnormality are simultaneously reflected in one image, the AI model misjudges the image as non-abnormal.

[0087] The abnormality inspection system S1 according to the present invention uses two different learning models to determine the abnormality of the cam part. Therefore, even in a situation where the patterns reflected in the image to be determined are diverse, the accuracy of the abnormality inspection can be improved. Therefore, the abnormality inspection system S1 can accurately detect the abnormality of the cam part. In addition, since the accuracy of the abnormality inspection can be improved through software processing, relatively inexpensive components such as area array cameras can be used as the photographing unit for photographing the inspection target part.

[0088] In addition, the second learning model M2 can also be a model with higher accuracy than the first learning model M1. In this case, by using the first learning model M1, the number of images input to the second learning model M2 that requires a long processing time can be suppressed, so the time required for overall calculation can be reduced.

[0089] In addition, the first learning model M1 is a model that outputs whether there is an appearance abnormality in the cam portion. The first inspection unit 245 can also determine that there is a suspicion of an abnormality in the cam portion when its output result indicates an appearance abnormality in the cam portion. Moreover, the second learning model M2 is a model that outputs the category of the appearance abnormality in the cam portion. The second inspection unit 246 can also determine whether there is an abnormality in the cam portion based on the category of the appearance abnormality in the cam portion indicated by its output result. The second inspection unit 246 does not perform the determination process when there is no appearance abnormality in the cam portion, so its inspection is only performed when necessary. Therefore, the abnormality inspection system S1 can perform the inspection efficiently.

[0090] In addition, when the output result of the second learning model M2 indicates that the appearance abnormality in the cam portion is at least one pattern among droplets, marks of a deburring brush, inspection marks, or grindstone patterns, the second inspection unit 246 can also determine that there is no abnormality in the cam portion. Thus, the abnormality inspection system S1 does not determine an abnormality that is actually not an abnormality in the substantial quality of the component but only an appearance abnormality as a final abnormality, so over-detection can be suppressed.

[0091] In addition, the second inspection unit 246 can also cut out the block determined by the first inspection unit 245 to have a suspicion of an abnormality from the captured data and input the captured data of the cut-out block into the second learning model M2. Thus, since unnecessary blocks are excluded from the determination target in the second inspection unit, the number and types of noises that may occur in the inspection of the second inspection unit can be reduced, and the inspection accuracy can be improved.

[0092] In addition, in the inspection of the object inspection component W, the shaft neck portion that is not the cam portion can also be inspected once by the third inspection unit 247 instead of multiple times. As described above, for a component such as the cam portion, in the inspection using a method of segmentation and classification employing one AI model, the accuracy of the determination is limited. However, since the shape of the shaft neck portion is a perfect circle, such a problem does not occur. Therefore, by setting the number of inspections to once, the total amount of computational processing required for the inspection of the entire inspection target component can be reduced, and an efficient inspection can be performed. In addition, the inspection parameters in the third learning model M3 can also be set according to values such as the target inspection time and pass rate.

[0093] In addition, the image acquisition unit 244 (the first and second acquisition units) may also acquire the captured data of the cam portion captured in a state where light is irradiated onto the cam portion by the cam portion inspection illumination 11 (the first illumination), and acquire the captured data of the journal portion captured in a state where light is irradiated onto the journal portion by the journal portion inspection illumination 13 (the second illumination). Since the cam portion and the journal portion are irradiated with light by illuminations that match their respective shapes, the image acquisition unit 244 can acquire captured data of a quality suitable for inspection. Therefore, the abnormality inspection system S1 can more accurately detect abnormalities in the cam portion.

[0094] In addition, the image acquisition unit 244 may also acquire the captured data of the cam portion captured in a state where the cam portion is rotatably supported by the rotation motor 15 and rotated. Thereby, the abnormality inspection system S1 can efficiently acquire images of different portions of the cam portion, and thus can shorten the time required for the entire inspection.

[0095] Furthermore, the present invention is not limited to the above-described embodiments, and can be appropriately modified without departing from the gist.

[0096] For example, the component to be inspected may not be a camshaft, but another type of component having a non-circular cam portion. Even in such a case, the abnormality inspection system S1 can perform the inspection described in the embodiments on the cam portion of the component.

[0097] In the embodiment, the case where the cam portion is inspected in two stages based on the first inspection unit 245 and the second inspection unit 246 has been described, but as the inspection of the cam portion, an inspection in three or more stages may also be performed. For example, an inspection not based on the second inspection unit 246 may be further performed either before or after the inspection based on the first inspection unit 245.

[0098] In the inspection, not the entire surface of the cam portion may be captured and inspected, but only a part of its surface may be captured and become the inspection object. The same applies to the journal portion.

[0099] In the abnormality inspection system S1, when it is determined whether there is an abnormality in the cam portion of the inspection target component W as described in the embodiment, the user can also input to the information processing device 20 whether there is an actual abnormality in the cam portion or the type of pattern of the cam portion reflected in the image (for example, the type of pattern showing a defect or the type of pattern showing a false abnormality). Thus, the first inspection unit 245 and the second inspection unit 246 can correct at least one of the first learning model M1 and the second learning model M2 based on the data fed back in this way. Similarly, when it is determined whether there is an abnormality in the journal portion of the inspection target component W, the user can also input to the information processing device 20 whether there is an actual abnormality in the journal portion or the type of pattern of the journal portion reflected in the image. Thus, the third inspection unit 247 can correct the third learning model M3 based on the data fed back in this way.

[0100] As described above, one or more processors included in the abnormality inspection system in the above-described embodiment execute the following one or more programs: which include a command group for causing a computer to execute the algorithms described with reference to the accompanying drawings. Through this processing, the processing described in each embodiment can be realized.

[0101] The program includes the following command group (or software code): which, when read into a computer, causes the computer to execute one or more functions described in the embodiment. The program can also be stored in a non-temporary computer-readable medium or a tangible storage medium. By way of example and not limitation, computer-readable media or tangible storage media include random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD) or other memory technologies, CD-ROM, digital versatile disk (DVD), Blu-ray (registered trademark) disc, or other optical disc storage, cassette tape, magnetic tape, magnetic disk storage or other magnetic storage devices. The program can also be transmitted on a temporary computer-readable medium or a communication medium. By way of example and not limitation, temporary computer-readable media or communication media include electrical, optical, acoustic or other forms of propagated signals. Additionally, the program can, for example, also take the form of an application program.

[0102] As described above, the present invention has been described with reference to the embodiments, but the present invention is not limited to the above. Specifically, the configuration of the present invention can be variously modified within the scope of the present invention that can be understood by those skilled in the art.

Claims

1. An abnormality inspection system, comprising: A first acquisition unit that acquires captured data of a cam part; A first inspection unit that checks whether there is a suspicion of abnormality in the cam part by inputting the captured data of the cam part into a first learning model; and A second inspection unit that checks whether there is an abnormality in the cam part by inputting the captured data of the cam part determined by the first inspection unit to have a suspicion of abnormality into a second learning model different from the first learning model, The first learning model is a model that outputs whether there is an appearance abnormality in the cam part. When the output result indicates that there is an appearance abnormality in the cam part, the first inspection unit determines that there is a suspicion of abnormality in the cam part. The second learning model is a model that calculates and outputs the coincidence rate between the captured data of the cam part and a pattern of false abnormality. The false abnormality is a non-substantive appearance abnormality that naturally occurs through processing in the manufacturing stage of the cam part. The second inspection unit determines whether there is an abnormality in the cam part by comparing the coincidence rate with a preset threshold.

2. The abnormality inspection system according to claim 1, wherein When the output result of the second learning model indicates that the appearance abnormality in the cam part is at least one of a pattern such as a droplet, a trace of a deburring brush, an inspection trace, or a grindstone pattern, the second inspection unit determines that there is no abnormality in the cam part.

3. The abnormality inspection system according to claim 1, wherein The second inspection unit cuts out a block determined by the first inspection unit to have a suspicion of abnormality from the captured data, and inputs the captured data of the cut-out block into the second learning model.

4. The abnormality inspection system according to claim 1, wherein It further comprises: A second acquisition unit that acquires captured data of a journal part provided on the same component as the cam part; and A third inspection unit that checks whether there is an abnormality in the journal part by inputting the captured data of the journal part into a third learning model different from the first learning model and the second learning model, The journal part is only inspected by the third inspection unit.

5. The abnormality inspection system according to claim 4, wherein The first acquisition unit acquires the captured data of the cam part in a state where light is irradiated on the cam part by a first illumination. The second acquisition unit acquires the captured data of the journal part in a state where light is irradiated on the journal part by a second illumination different from the first illumination.

6. The abnormality inspection system according to any one of claims 1 to 5, wherein The first acquisition unit acquires the captured data of the cam part in a state where the cam part is rotated by a rotation mechanism that axially supports and rotates the shaft of the cam part.

7. An abnormality inspection method, which is executed by an abnormality inspection system, comprising: An acquisition step of acquiring captured data of a cam part; A first inspection step of checking whether there is a suspicion of abnormality in the cam part by inputting the captured data of the cam part into a first learning model; And A second inspection step of inspecting whether there is an abnormality in the cam part by inputting the photographed data of the cam part suspected of having an abnormality into a second learning model different from the first learning model. The first learning model is a model that outputs whether there is an appearance abnormality in the cam part. In the first inspection step, when the output result indicates that there is an appearance abnormality in the cam part, it is determined that the cam part is suspected of having an abnormality. The second learning model is a model that calculates and outputs the coincidence rate between the photographed data of the cam part and a pattern of false abnormality. The false abnormality is a non-substantive appearance abnormality that naturally occurs through the processing in the manufacturing stage of the cam part. In the second inspection step, it is determined whether there is an abnormality in the cam part by comparing the coincidence rate with a preset threshold.

8. A storage medium storing a program that causes a computer to execute the following steps: An acquisition step of acquiring photographed data of a cam part; A first inspection step of inspecting whether the cam part is suspected of having an abnormality by inputting the photographed data of the cam part into a first learning model; and A second inspection step of inspecting whether there is an abnormality in the cam part by inputting the photographed data of the cam part determined to be suspected of having an abnormality into a second learning model different from the first learning model. The first learning model is a model that outputs whether there is an appearance abnormality in the cam part. In the first inspection step, when the output result indicates that there is an appearance abnormality in the cam part, it is determined that the cam part is suspected of having an abnormality. The second learning model is a model that calculates and outputs the coincidence rate between the photographed data of the cam part and a pattern of false abnormality. The false abnormality is a non-substantive appearance abnormality that naturally occurs through the processing in the manufacturing stage of the cam part. In the second inspection step, it is determined whether there is an abnormality in the cam part by comparing the coincidence rate with a preset threshold.

Images