Ai generator for inspection, ai generation method for inspection, ai inspection system, and ai inspection method

Abstract

To provide an AI construction method which operates in high accuracy in a state that data related to an object for inspection are not sufficiently obtained when constructing AI, and a system etc. mounted with the method.SOLUTION: An AI generator for inspection comprises: a first construction unit which constructs first AI capable of extracting a feature of a product on the basis of image data of normal goods of the product; a second construction unit which constructs second AI capable of falsely generating the image data of abnormal goods of a target product, on the basis of information on abnormality of the product obtained from the image data of abnormal goods of the product, and the image data of normal goods of the target product; and a third construction unit which constructs third AI which can determine whether a plurality of pieces of image data becoming determination objects of the target product are normal / abnormal, by using the first AI and the second AI.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present disclosure relates to an inspection AI generation device, an inspection AI generation method, an AI inspection system, and an AI inspection method. [Background technology]

[0002] In recent years, with the development of manufacturing techniques, the occurrence of defective products has become extremely rare in many industrial products.

[0003] However, even if it is rare, defective products that do occur cannot be allowed to be distributed, so inspection is essential.

[0004] Even in factories where the manufacturing process is highly automated, there are many cases where the inspection process is still carried out by human inspectors.

[0005] It is feared that quality assurance through inspection will become even more difficult in the future due to a shortage of inspectors, and there is a strong demand for the development of a highly accurate inspection system.

[0006] The application of AI technology is advancing in the development of highly accurate inspection systems (e.g., Patent Document 1). [Prior art documents] [Patent documents]

[0007] [Patent Document 1] JP 2020-52044 A Summary of the Invention [Problem to be solved by the invention]

[0008] However, due to the fact that defective products occur very rarely, it is difficult to collect a large amount of data on defective products as training data.

[0009] In recent years, techniques that use unsupervised learning to build AI using only normal products without using data on defective products have become common, but generally speaking, when there is sufficient data, supervised AI tends to be more accurate.

[0010] Up until now, technologies have been developed to generate pseudo-failure images of inspection targets using defective images of the inspection targets. However, for small-lot products of several hundred to several thousand units, it is difficult to collect defective images of the inspection targets in advance, making it difficult to apply existing technologies.

[0011] Therefore, it is an object of the present disclosure to provide technical improvements that solve or mitigate at least some of the problems of the prior art discussed above.

[0012] Specifically, the objective of the present disclosure is to provide a method for building an AI that operates with high accuracy even in situations where sufficient data on the inspection target is not available when building the AI, and a system equipped with the same. [Means for solving the problem]

[0013] The inspection AI generation device of the present invention is characterized by comprising a first construction unit that constructs a first AI capable of extracting characteristics of a product based on image data of a normal product of the product, a second construction unit that constructs a second AI that can pseudo-generate image data of a defective product of a target product based on information about the abnormality of the product obtained from the image data of the defective product and image data of the normal product of the target product, and a third construction unit that uses the first AI and the second AI to construct a third AI that can determine whether multiple image data of the target product are normal / abnormal.

[0014] The second construction unit obtains information on the abnormality in the product by removing the texture information of the product from the image data of the abnormal product, and by adding the texture information of the target product to the abnormality information, it is possible to pseudo-generate image data of the abnormality in the target product.

[0015] The first construction unit may be capable of extracting characteristics of the product based on image data of an abnormal product in addition to image data of a normal product.

[0016] The target product is one type of product among product types, and the occurrence mechanism of the abnormality in the product and the abnormality in the target product can be the same.

[0017] The product and the target product have at least one cut surface, and the second construction unit can generate artificial image data of a defective product of the target product corresponding to the cut surface.

[0018] The method for generating an AI for inspection according to the present disclosure is characterized in that it has one or more computer processors execute a first construction step of constructing a first AI capable of extracting features of a product based on image data of a normal product of the product, a second construction step of constructing a second AI capable of pseudo-generating image data of an abnormal product of a target product based on information on an abnormality of the product obtained from the image data of the abnormal product and image data of the normal product of the target product, and a third construction step of constructing a third AI capable of determining whether a target product is normal or abnormal using the first AI and the second AI.

[0019] The AI ​​inspection system of the present disclosure is an AI inspection system including a second information processing device connectable to a first information processing device, and an inspection AI generation device connectable to the second information processing device, the inspection AI generation device having a first construction unit that constructs a first AI capable of extracting features of a product based on image data of a normal product of the product, a second construction unit that constructs a second AI capable of pseudo-generating image data of an abnormal product of a target product based on information on an abnormality of the product obtained from the image data of the abnormal product and image data of a normal product of the target product, and a third construction unit that constructs a third AI capable of determining normality / abnormality for a plurality of image data to be used to judge the target product using the first AI and the second AI, and the second information processing device has a first data acquisition unit that acquires data necessary for installing and updating the third AI from the inspection AI generation device, a second data acquisition unit that acquires a plurality of image data to be used to judge the target product, and a third construction unit that acquires a plurality of image data to be used to judge the target product using the third AI. 、 The device is characterized by having a judgment unit that judges whether a plurality of image data that are used to judge the target product are normal / abnormal, and an output unit that outputs the result of the judgment by the judgment unit to the first information processing device.

[0020] The inspection AI generation device further has a third data acquisition unit that acquires image data of target products that have been judged to be normal by the judgment unit and / or image data of target products that have been judged to be abnormal by the judgment unit from the second information processing device, and the third construction unit can use the data acquired by the first AI, the second AI, and the third data acquisition unit to reconstruct a third AI capable of determining whether multiple image data of target products are normal / abnormal.

[0021] A plurality of image data to be used for judging the target product can be captured by an imaging device connectable to the first information processing device.

[0022] The third data acquisition unit can acquire image data of target products that have been determined to be normal by the judgment unit and / or image data of target products that have been determined to be abnormal by the judgment unit from a plurality of second information processing devices located at a plurality of locations.

[0023] The AI ​​inspection method disclosed herein is an AI inspection method in an AI inspection system including a second information processing device connectable to a first information processing device and an inspection AI generation device connectable to the second information processing device, and includes a first construction step of constructing a first AI capable of extracting features of a product based on image data of a normal product of the product in one or more computer processors; a second construction step of constructing a second AI capable of pseudo-generating image data of an abnormal product of a target product based on information on an abnormality of the product obtained from the image data of the abnormal product and image data of a normal product of the target product; a third construction step of constructing a third AI capable of determining whether a plurality of image data to be used for judging the target product is normal / abnormal using the first AI and the second AI; a first data acquisition step of acquiring data necessary for installing and updating the third AI from the inspection AI generation device; a second data acquisition step of acquiring multiple image data to be used for judging the target product; 、 A determination step of determining whether a plurality of image data that are used to determine the target product are normal / abnormal, and an output step of outputting the result of the determination in the determination step to the first information processing apparatus can be executed. Effect of the Invention

[0024] The present disclosure provides technical improvements that solve or mitigate at least some of the problems with the prior art discussed above.

[0025] Specifically, according to the present disclosure, it is possible to provide an inspection AI generation device, an inspection AI generation method, an AI inspection system, and an AI inspection method that operate with high accuracy using a supervised learning framework even in situations where sufficient data on the inspection target is not available when building an AI. [Brief description of the drawings]

[0026] [Figure 1] FIG. 1 is a system configuration diagram showing the system configuration of an AI inspection system according to the present disclosure. [Diagram 2] FIG. 1 is a conceptual diagram showing an example application of the AI ​​inspection system according to the present disclosure. [Diagram 3] FIG. 2 is a configuration diagram showing an example of the hardware configuration of an inspection AI generation device in the present disclosure. [Figure 4] FIG. 2 is a configuration diagram showing an example of the functional configuration of an inspection AI generation device in the present disclosure. [Diagram 5] FIG. 2 is a configuration diagram showing an example of a functional configuration of a second information processing device in the present disclosure. [Figure 6] FIG. 11 is an image diagram showing an example of a screen displayed on a first information processing device in the present disclosure. [Figure 7] FIG. 13 is a configuration diagram showing another example of the functional configuration of the inspection AI generation device in the present disclosure. [Figure 8] FIG. 1 is a conceptual diagram showing an example application of the AI ​​inspection system according to the present disclosure. [Figure 9] FIG. 1 is a conceptual diagram showing an example application of the AI ​​inspection system according to the present disclosure. [Figure 10] FIG. 1 is a flow diagram showing an example of the flow of an inspection AI generation method according to the present disclosure. [Figure 11] FIG. 1 is a flow diagram showing an example of the flow of an AI inspection method according to the present disclosure. [Figure 12] FIG. 2 is a circuit configuration diagram showing an example of a circuit configuration for implementing a computer program according to the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0027] <System configuration> As shown as an example in FIG. 1, an AI inspection system 1000 according to the present disclosure may include an inspection AI generation device 100 and a second information processing device 300 connectable to a first information processing device 200.

[0028] FIG. 2 shows an example application of the AI ​​inspection system 1000 according to the present disclosure, and shows an inspection AI generation device 100 as a server device, a first information processing device 200 which is an on-site computer including a display, and a second information processing device 300 which is an on-site computer equipped with AI.

[0029] In the example shown in FIG. 2, the site is a manufacturing factory for the target product, and the first information processing device 200 is placed on each of the three manufacturing lines, and one second information processing device 300 is placed for each of these multiple manufacturing lines, but the configuration is not limited to this.

[0030] The inspection AI generation device 100 and the second information processing device 300, and the first information processing device 200 and the second information processing device 300 can be connected via a network.

[0031] In particular, when the inspection AI generation device 100 is a server device on the cloud, the inspection AI generation device 100 and the second information processing device 300 can be connected to the Internet via a mobile router or the like.

[0032] The first information processing device 200 and the second information processing device 300 may be connected via a LAN, either wired or wirelessly.

[0033] <Hardware configuration> Here, the hardware configuration of the inspection AI generation device 100 will be described with reference to Fig. 3. The inspection AI generation device 100 includes a processor 101, a memory 102, a storage 103, an input / output interface (input / output I / F) 104, and a communication interface (communication I / F) 105. The components are connected to each other via a bus B.

[0034] The inspection AI generation device 100 can realize the functions and methods described in this embodiment through cooperation between a processor 101, a memory 102, a storage 103, an input / output I / F 104, and a communication I / F 105.

[0035] The processor 101 executes functions and / or methods realized by codes or instructions included in a program stored in the storage 103. The processor 101 includes, for example, a central processing unit (CPU), a micro processing unit (MPU), a graphics processing unit (GPU), a microprocessor, a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc., and may realize each process disclosed in each embodiment by a logic circuit (hardware) or a dedicated circuit formed in an integrated circuit (an integrated circuit (IC) chip, a large scale integration (LSI)), etc.). These circuits may be realized by one or more integrated circuits, and a plurality of processes shown in each embodiment may be realized by one integrated circuit. An LSI may also be called a VLSI, a super LSI, an ultra LSI, etc., depending on the degree of integration.

[0036] The memory 102 temporarily stores the programs loaded from the storage 103 and provides a working area for the processor 101. The memory 102 also temporarily stores various data generated while the processor 101 is executing the programs. The memory 102 includes, for example, a random access memory (RAM) and a read only memory (ROM).

[0037] The storage 103 stores programs. The storage 103 includes, for example, a hard disk drive (HDD), a solid state drive (SSD), a flash memory, and the like.

[0038] The communication I / F 105 is implemented as hardware such as a network adapter, communication software, or a combination of these, and transmits and receives various data via a network. The communication may be performed either wired or wirelessly, and any communication protocol may be used as long as the communication between the devices can be performed. The communication I / F 105 communicates with other devices via the network. The communication I / F 105 transmits various data to other devices according to instructions from the processor 101. The communication I / F 105 also receives various data transmitted from other devices and transmits the data to the processor 101.

[0039] The input / output I / F 104 includes an input device for inputting various operations to the inspection AI generation device 100, and an output device for outputting the processing results processed by the inspection AI generation device 100. The input / output I / F 104 may be an integrated input device and output device, or may be separate input device and output device.

[0040] The input device is realized by any one or a combination of all kinds of devices that can receive input from a user and transmit information related to the input to the processor 101. The input device includes, for example, a touch panel, a touch display, hardware keys such as a keyboard, a pointing device such as a mouse, a camera (operation input via an image), and a microphone (operation input by voice).

[0041] Moreover, the first information processing device 200 and the second information processing device 300 in the present disclosure can also be configured with the same hardware configuration as that in FIG. 3, unless otherwise noted.

[0042] <Embodiment> Next, various functions that can be executed by the inspection AI generation device 100 according to an embodiment of the present disclosure will be described with reference to the drawings.

[0043] The inspection AI generation device 100 in the present disclosure is characterized by having a first construction unit 110, a second construction unit 120, and a third construction unit 130, as shown as an example in FIG.

[0044] The first construction unit 110 has a function of constructing a first AI capable of extracting the characteristics of a product based on image data of the product when it is in a normal condition.

[0045] "Product" refers to a higher-level concept than "target product." For example, if the target product to be inspected is "automobile of model A," the product means "automobile" without any model restrictions. Or, if the target product to be inspected is "metal pipe A," the product means "metal pipe" without any restrictions on the material or size of the pipe, or "pipes in general" without any restrictions on metal pipes.

[0046] Additionally, the "product" in this embodiment may include metal parts (screws, pipes, etc.), furniture, molds, fabrics (bags, etc.), food (bar-type snacks with different flavors, etc.), books, etc. Furthermore, the "product" may include agricultural products such as fruits (fruits that have similar bruising patterns, such as peaches and plums, etc.).

[0047] The manufactured "products" include "normal products" that can be distributed and "abnormal products" that cannot be distributed. In the invention of this disclosure, a "normal product" refers to a product that has no abnormality in its appearance or the degree of abnormality is less than a threshold. Similarly, a "abnormal product" refers to a product that has an abnormality in its appearance or the degree of abnormality is equal to or greater than a threshold.

[0048] Furthermore, the "image data of a normal product" need only include the part that is the subject of inspection, and does not necessarily have to include the entire product.

[0049] Furthermore, the "feature of the product" can refer to the feature amount of the image, specifically, a feature vector.

[0050] It should be noted that the image data may include data other than image data of a normal product (i.e., image data of a defective product).

[0051] The second construction unit 120 has the function of constructing a second AI that can artificially generate image data of an abnormal product of the target product based on information about the abnormality of the product obtained from image data of the abnormal product and image data of a normal product of the target product.

[0052] The "image data of the defective product" and the "image data of the normal target product" need only include the part that is the subject of inspection, and do not necessarily need to include the entire product.

[0053] As an example, "information about an abnormality" refers to data in which the texture information of the product has been removed from the image data of the abnormal product.

[0054] "Texture information" can be information that represents the appearance of a product, such as its color, pattern, and texture. For example, in the case of a metal product such as a screw, the roughness or glossiness corresponds to the texture information. In addition, in the case of a bar-type snack, the black color or degree of burntness of a chocolate-flavored snack corresponds to the texture information. In addition, in the case of a book, the texture of the paper on the cut surface or the pattern created by the ink corresponds to the texture information.

[0055] Then, as one example, the second construction unit 120 can generate pseudo image data of the abnormal part of the target product by adding texture information of the target product to the information of the abnormality.

[0056] Furthermore, the more image data of abnormal products that are provided, and the more image data of normal products of the target product that are provided, the more accurate the second AI can be constructed.

[0057] The third construction unit 130 has a function of using the first AI and the second AI to construct a third AI capable of determining normality / abnormality for multiple image data to be used to judge the target product.

[0058] The judgment process is executed by the second information processing device 300, but the third construction unit 130 can reconstruct the third AI by acquiring image data of the target product that is the subject of the judgment and learning from it. Details will be described later.

[0059] According to the above configuration, it is possible to provide a technical improvement that solves or alleviates at least some of the problems of the conventional technology described above.

[0060] Specifically, with the above configuration, it is possible to provide an inspection AI generation device that operates with high accuracy even in situations where sufficient data on the inspection object is not available when building the AI.

[0061] The inspection AI generation device 100 in the present disclosure is characterized by having a construction unit.

[0062] The construction unit uses image data of normal products of the same type as the target product as learning data to extract and learn the features of the product, and artificially generates image data of the abnormal product of the target product by adding texture information of the target product to information on the abnormality of the product, which is data obtained by removing the texture information of the product from the image data of the abnormal product, and constructs an AI that learns information about the abnormality of the target product by using the image data of the abnormal product of the target product as learning data.

[0063] The AI ​​has the ability to determine whether a target product is normal or abnormal when multiple image data of the target product are input.

[0064] According to the above configuration, it is possible to provide a technical improvement that solves or alleviates at least some of the problems of the conventional technology described above.

[0065] Specifically, with the above configuration, it is possible to provide an inspection AI generation device that operates with high accuracy even in situations where sufficient data on the inspection object is not available when building the AI.

[0066] As described above, the second construction unit 120 obtains information on the abnormality of the product by removing the texture information of the product from the image data of the abnormal product, and by adding the texture information of the target product to the abnormality information, it is possible to pseudo-generate image data of the abnormality of the target product.

[0067] With this configuration, it is possible to provide a method for building an AI that operates with high accuracy even in situations where sufficient data on the inspection target is not available when building the AI.

[0068] Above AI In addition to image data of normal products, image data of abnormal products Using as training data , the features of the product Learn It is possible.

[0069] With this configuration, it is possible to provide an inspection AI generation device that uses a supervised learning framework to operate with high accuracy even in situations where sufficient data on the inspection target is not available when building the AI.

[0070] As described above, the target product is one type of product among various types of products, and the occurrence mechanism of the abnormality in the product and the abnormality in the target product can be the same.

[0071] As an example, if the product and the target product are books, the abnormality is something that can be seen from the outside, such as a crease, a stain, etc. The mechanism by which these abnormalities occur is generally the same for any type of book.

[0072] Similarly, when the product and the target product are the above-mentioned metal parts, furniture, molds, fabrics, food, fruits, etc., the abnormality is something that can be confirmed by appearance, such as scratches, stains, etc. The above occurrence mechanisms are generally the same for the same product.

[0073] The product and the target product may have at least one cut surface.

[0074] As examples of "products," screws, molds, and materials have been given, but these may include a cutting process in the manufacturing process and have at least one cut surface.

[0075] The above-mentioned externally identifiable abnormalities can be found by inspecting the cut surfaces.

[0076] At this time, Above AI It is possible to generate pseudo image data of defective products of the target product corresponding to the cut surface.

[0077] The AI ​​inspection system 1000 in the present disclosure is characterized by including the above-mentioned inspection AI generation device 100 and the above-mentioned second information processing device 300, as shown as an example in FIG. 1.

[0078] The second information processing device 300 can be connected to the first information processing device 200 .

[0079] In addition, as another embodiment, the AI ​​inspection system 1000 in the present disclosure may be equipped with a first information processing device 200.

[0080] The inspection AI generation device 100 can be connected to a second information processing device 300 via a network.

[0081] The inspection AI generation device 100 has a first construction unit 110, a second construction unit 120, and a third construction unit 130. The description of each component is as described above.

[0082] As shown in FIG. 5 as an example, the second information processing device 300 includes a first data acquisition section 310, a second data acquisition section 320, a determination section 330, and an output section 340.

[0083] The first data acquisition unit 310 acquires data required for installing and updating the third AI from the inspection AI generation device 100.

[0084] The necessary data may be, for example, pre-processing parameters, an AI model, AI-learned parameters, and post-processing parameters.

[0085] The second data acquisition unit 320 acquires a plurality of image data that are used to judge the target product.

[0086] Such a plurality of image data can be acquired from the first information processing device 100. The first information processing device 100 acquires image data from an imaging device 400, the details of which will be described later.

[0087] The second information processing device 300 may further include an image processing unit that performs preprocessing on the data acquired by the second data acquisition unit 320 , and the preprocessed image may be passed to the determination unit 330 .

[0088] The judgment unit 330 judges whether a plurality of image data items that are to be judged as to the target product are normal / abnormal using a third AI.

[0089] The output section 340 outputs the result of the determination made by the determination section 330 to the first information processing apparatus 200 .

[0090] Fig. 6 is an example of a screen displayed on the display of the first information processing device 200. As shown as an example in Fig. 6, the screen 20 displays a field 21 for displaying information showing the appearance of the photographed product and abnormal locations, a field 22 for displaying information on the timing of occurrence of the abnormal image (time, number of products produced, etc.), a field 23 for displaying information on the line (number of products produced up to now), and a field 24 for setting parameters such as pre-processing, post-processing, and alarm volume.

[0091] 6 is just an example, and the layout positions and sizes of these fields are not particularly limited. Also, the screen 20 can be displayed on the second information processing device 300 or another display installed on the line.

[0092] According to the above configuration, it is possible to provide a technical improvement that solves or alleviates at least some of the problems of the conventional technology described above.

[0093] Specifically, with the above configuration, it is possible to provide an AI inspection system that operates with high accuracy even in situations where sufficient data on the inspection target is not available when building the AI.

[0094] The inspection AI generation device 100 in the present disclosure is assumed to have a construction unit. The description of this component is as described above.

[0095] In order to realize AI functions in the second information processing device, the second information processing device may have a first data acquisition unit that acquires data including pre-processing parameters, an AI model, AI learned parameters, and post-processing parameters from an inspection AI generation device, a second data acquisition unit that acquires multiple image data that are to be used to judge the target product from the first information processing device, a judgment unit that judges whether the multiple image data that are to be used to judge the target product are normal / abnormal using AI realized using the data acquired in the first data acquisition step, and an output unit that outputs the result of the judgment by the judgment unit to the first information processing device.

[0096] The inspection AI generation device may further include a data acquisition unit 140, as shown in FIG.

[0097] The data acquisition unit 140 acquires, from the second information processing device 300, image data of a target product that has been determined to be normal by the judgment unit 330 and / or image data of a target product that has been determined to be abnormal by the judgment unit 330.

[0098] At this time, Above AI teeth, moreover, Acquired by the data acquisition department The image data of the target product determined to be normal and / or the image data of the target product determined to be abnormal are used as learning data to perform additional learning, and the first data acquisition unit acquires data including pre-processing parameters, AI models, AI learned parameters, and post-processing parameters from the inspection AI generation device to realize the latest AI functions configured by the construction unit. This shall be done.

[0099] As shown in FIG. 2, a plurality of image data to be used for determining the target product may be captured by an imaging device 400 connectable to the first information processing device 200.

[0100] As shown in FIG. 8, the data acquisition unit 140 can acquire image data of target products that have been determined to be normal by the judgment unit 330 and / or image data of target products that have been determined to be abnormal by the judgment unit 330 from multiple second information processing devices 300 located at multiple locations.

[0101] According to this configuration, the latest AI is always installed in the second information processing device 300 at each base from the server. The second information processing device 300 at each base always uploads captured images to the cloud. This allows the AI ​​generation device 100 to learn periodically using data from multiple bases and improve accuracy.

[0102] The result of the determination by the determining unit 330 can be output to a speaker and a display as shown in Fig. 9. By checking these, an operator can remove defective products from the line.

[0103] Although the above example shows an example in which the worker manually removes the debris, it is also possible to configure the system so that this task is performed automatically.

[0104] As explained above, the AI ​​construction method disclosed in this disclosure consists of the following three parts. 1) Building an AI or AI module capable of extracting product features using previously manufactured good products (or both good and defective products) 2) Building an AI or AI module that can generate artificial defects for inspection based on past defects 3) A learning framework that uses 1) and 2) above to quickly build an inspection AI from a few normal pieces of data to be inspected.

[0105] The inspection AI system also consists of the following three parts: 1) Department for obtaining samples of normal products to be inspected 2)AI construction department 3) Inspection Results Transmission Department

[0106] According to the above configuration, it is possible to provide a technical improvement that solves or alleviates at least some of the problems of the conventional technology described above.

[0107] Specifically, with the above configuration, it is possible to provide an inspection AI generation device and an inspection AI system that can operate with high accuracy even in situations where sufficient data on the inspection object is not available when building the AI.

[0108] Next, an embodiment of the inspection AI generation method according to the present disclosure will be described with reference to the drawings.

[0109] The inspection AI generation method disclosed herein is characterized in that, as an example, as shown in FIG. 10, one or more computer processors execute a first construction step S110, a second construction step S120, and a third construction step S130.

[0110] In the first construction step S110, a first AI capable of extracting the characteristics of a product based on image data of a normal product is constructed.

[0111] Such a first construction step S110 can be executed in the above-mentioned first construction unit 110. Details of the first construction unit 110 are as described above.

[0112] In the second construction step S120, a second AI is constructed that can artificially generate image data of an abnormal product of the target product based on information about the abnormality of the product obtained from image data of the abnormal product and image data of a normal product of the target product.

[0113] Such a second construction step S120 can be executed by the above-mentioned second construction unit 120. Details of the second construction unit 120 are as described above.

[0114] In a third construction step S130, the first AI and the second AI are used to construct a third AI capable of determining normality / abnormality for multiple image data to be used to judge the target product.

[0115] Such a third construction step S130 can be executed by the above-mentioned third construction unit 130. Details of the third construction unit 130 are as described above.

[0116] According to the above configuration, it is possible to provide a technical improvement that solves or alleviates at least some of the problems of the conventional technology described above.

[0117] Specifically, with the above configuration, it is possible to provide a method for generating inspection AI that operates with high accuracy even in situations where sufficient data on the inspection object is not available when building the AI.

[0118] Alternatively, the inspection AI generation method disclosed herein can cause one or more computer processors to execute the construction step.

[0119] In the construction step, image data of normal products of the same variety as the target product is used as learning data to extract and learn the features of the product, and image data of the abnormal product of the target product is artificially generated by adding texture information of the target product to information on the abnormality of the product, which is data obtained by removing the texture information of the product from the image data of the abnormal product, and an AI is constructed that learns information on the abnormality of the target product by using the image data of the abnormal product of the target product as learning data.

[0120] In addition, the AI ​​has the ability to determine whether a target product is normal or abnormal when multiple image data of the target product are input.

[0121] According to the above configuration, it is possible to provide a technical improvement that solves or alleviates at least some of the problems of the conventional technology described above.

[0122] Specifically, with the above configuration, it is possible to provide a method for generating inspection AI that operates with high accuracy even in situations where sufficient data on the inspection object is not available when building the AI.

[0123] Next, an embodiment of the AI ​​inspection method according to the present disclosure will be described with reference to the drawings.

[0124] The AI ​​inspection method disclosed herein is an AI inspection method in an AI inspection system 1000 that includes a second information processing device 300 connectable to a first information processing device 200 as shown in FIG. 2, and an inspection AI generation device 100 connectable to the second information processing device, and is characterized by executing a first construction step S110, a second construction step S120, a third construction step S130, a first data acquisition step S140, a second data acquisition step S150, a judgment step S160, and an output step S170, as shown in FIG. 11 as an example.

[0125] In the first construction step S110, a first AI capable of extracting the characteristics of a product based on image data of a normal product is constructed.

[0126] Such a first construction step S110 can be executed in the above-mentioned first construction unit 110. Details of the first construction unit 110 are as described above.

[0127] In the second construction step S120, a second AI is constructed that can artificially generate image data of an abnormal product of the target product based on information about the abnormality of the product obtained from image data of the abnormal product and image data of a normal product of the target product.

[0128] Such a second construction step S120 can be executed by the above-mentioned second construction unit 120. Details of the second construction unit 120 are as described above.

[0129] In a third construction step S130, the first AI and the second AI are used to construct a third AI capable of determining normality / abnormality for a plurality of image data to be used to judge the target product.

[0130] Such a third construction step S130 can be executed by the above-mentioned third construction unit 130. Details of the third construction unit 130 are as described above.

[0131] In the first data acquisition step S140, data required for installing and updating the third AI is acquired from the inspection AI generation device 100.

[0132] The first data acquisition step S140 can be executed by the above-mentioned first data acquisition unit 310. Details of the first data acquisition unit 310 are as described above.

[0133] In the second data acquisition step S150, a plurality of image data items that are to be used to judge the target product are acquired.

[0134] The second data acquisition step S150 can be executed by the above-mentioned second data acquisition unit 320. Details of the second data acquisition unit 320 are as described above.

[0135] In the judgment step S160, a third AI is used to judge whether a plurality of image data items that are to be judged for the target product are normal / abnormal.

[0136] The determination step S160 can be executed by the above-mentioned determination unit 330. Details of the determination unit 330 are as described above.

[0137] In the output step S170, the result of the determination in the determination step S160 is output to the first information processing apparatus 200.

[0138] The output step S170 can be performed in the above-mentioned output unit 340. Details of the output unit 340 are as described above.

[0139] According to the above configuration, it is possible to provide a technical improvement that solves or alleviates at least some of the problems of the conventional technology described above.

[0140] Specifically, with the above configuration, it is possible to provide an AI inspection method that operates with high accuracy even in situations where sufficient data on the inspection target is not available when building the AI.

[0141] Alternatively, the AI ​​inspection method of the present disclosure is an AI inspection method in an AI inspection system 1000 that includes a second information processing device 300 connectable to a first information processing device 200 as shown in FIG. 2, and an inspection AI generation device 100 connectable to the second information processing device, and is characterized in that it causes one or more computer processors provided in the inspection AI generation device to execute a construction step, and causes one or more computer processors provided in the second information processing device to execute a first data acquisition step, a second data acquisition step, a judgment step, and an output step.

[0142] In the construction step, image data of a normal product of a product of the same type as the target product is used as learning data to extract and learn the features of the product, and By adding texture information of the target product to information on abnormalities in the product, which is data obtained by removing the texture information of the product from the image data of the abnormal product, image data of the abnormal product is artificially generated, and an AI is constructed that learns information on abnormalities in the target product by using the image data of the abnormal product as learning data.

[0143] In addition, the AI ​​has the ability to determine whether a target product is normal or abnormal when multiple image data of the target product are input.

[0144] In the first data acquisition step, in order to realize AI functions in the second information processing device, data including pre-processing parameters, an AI model, AI learned parameters, and post-processing parameters are acquired from an inspection AI generation device.

[0145] In the second data acquisition step, a plurality of image data items to be used for judging the target product are acquired from the first information processing device.

[0146] In the judgment step, an AI realized using the data acquired in the first data acquisition step is used to judge whether multiple image data of the target product are normal or abnormal.

[0147] In the output step, the result of the determination in the determining step is output to the first information processing device.

[0148] According to the above configuration, it is possible to provide a technical improvement that solves or alleviates at least some of the problems of the conventional technology described above.

[0149] Specifically, with the above configuration, it is possible to provide an AI inspection method that operates with high accuracy even in situations where sufficient data on the inspection target is not available when building the AI.

[0150] Finally, an embodiment of a computer program according to the present disclosure will be described with reference to the drawings.

[0151] A computer program in the present disclosure is characterized in that it causes one or more computer processors to implement a first construction function, a second construction function, and a third construction function.

[0152] The first construction function constructs a first AI capable of extracting characteristics of a product based on image data of a normal product of the product.

[0153] The second construction function constructs a second AI that can artificially generate image data of an abnormal product of the target product based on information about the abnormality of the product obtained from image data of the abnormal product and image data of a normal product of the target product.

[0154] The third construction function uses the first AI and the second AI to construct a third AI capable of determining whether multiple image data that are to be judged for the target product are normal / abnormal.

[0155] The above functions can be realized by the first configuration circuit 1110, the second configuration circuit 1120, and the third configuration circuit 1130 shown in Fig. 12. The first configuration circuit 1110, the second configuration circuit 1120, and the third configuration circuit 1130 are realized by the above-mentioned first configuration unit 110, the second configuration unit 120, and the third configuration unit 130, respectively. Details of each unit are as described above.

[0156] According to the above configuration, it is possible to provide a technical improvement that solves or alleviates at least some of the problems of the conventional technology described above.

[0157] Specifically, with the above configuration, it is possible to provide a computer program for an inspection AI generation device that operates with high accuracy even in situations where sufficient data on the inspection object is not available when building an AI.

[0158] Alternatively, the computer program for an inspection AI generation device in the present disclosure can cause one or more computer processors to realize a construction function.

[0159] The construction function uses image data of normal products of the same type as the target product as learning data to extract and learn the features of the product, and artificially generates image data of the abnormal product of the target product by adding texture information of the target product to information on the abnormality of the product, which is data obtained by removing the texture information of the product from the image data of the abnormal product, and builds an AI that learns information about the abnormality of the target product by using the image data of the abnormal product of the target product as learning data.

[0160] In addition, the AI ​​has the ability to determine whether a target product is normal or abnormal when multiple image data of the target product are input.

[0161] According to the above configuration, it is possible to provide a technical improvement that solves or alleviates at least some of the problems of the conventional technology described above.

[0162] Specifically, with the above configuration, it is possible to provide a computer program for an inspection AI generation device that operates with high accuracy even in situations where sufficient data on the inspection object is not available when building an AI.

[0163] In addition, an information processing device such as a computer or a mobile phone can be suitably used to function as the server device or terminal device according to the above-described embodiment. Such an information processing device can be realized by storing a program describing the processing contents for realizing each function of the server device or terminal device according to the embodiment in a storage unit of the information processing device, and reading and executing the program by a CPU of the information processing device.

[0164] Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope and spirit of the invention, and are included in the scope of the invention and its equivalents described in the claims.

[0165] The method described in the embodiment can be stored as a program that can be executed by a calculator (computer) on a recording medium such as a magnetic disk (floppy disk, hard disk, etc.), an optical disk (CD-ROM, DVD, MO, etc.), or a semiconductor memory (ROM, RAM, flash memory, etc.), and can also be distributed by transmitting it via a communication medium. The program stored on the medium side also includes a setting program that configures software means (including not only execution programs but also tables and data structures) that the computer executes. The computer that realizes this device reads the program recorded on the recording medium, and in some cases, configures the software means by the setting program, and executes the above-mentioned processing by controlling the operation of the software means. The recording medium referred to in this specification is not limited to a storage medium for distribution, but also includes a storage medium such as a magnetic disk or semiconductor memory provided inside the computer or in a device connected via a network. The storage unit may function as, for example, a main storage device, an auxiliary storage device, or a cache memory. [Explanation of symbols]

[0166] 1000 AI inspection system according to the present disclosure 100 AI generation device for inspection 110 First Construction Department 120 Second Construction Department 130 Third Construction Department 140 Data Acquisition Section 200 First information processing device 300 Second information processing device 400 Imaging device

Claims

1. A first construction unit that constructs a first AI capable of extracting features of a product based on image data of a normal product of the product; A second construction unit that constructs a second AI capable of pseudo-generating image data of the abnormal product of the target product based on information on the abnormality of the product obtained from the image data of the abnormal product of the product and image data of a normal product of the target product; A third construction unit that constructs a third AI capable of determining whether a plurality of image data that are used to judge the target product are normal / abnormal using the first AI and the second AI; An inspection AI generation device equipped with the above.

2. The inspection AI generation device according to claim 1, characterized in that the second construction unit acquires information on the abnormality of the product by removing texture information of the product from image data of the abnormality of the product, and adds texture information of the target product to the abnormality information, thereby pseudo-generating image data of the abnormality of the target product.

3. The inspection AI generation device according to claim 1, characterized in that the first construction unit is capable of extracting characteristics of the product based on image data of defective products in addition to image data of normal products of the product.

4. The target product is one of the product varieties, The inspection AI generation device according to claim 1, characterized in that the mechanism by which the abnormality in the product and the abnormality in the target product occur is the same.

5. The inspection AI generation device described in claim 1, characterized in that the product and the target product have at least one cut surface, and the second construction unit artificially generates image data of an abnormality in the target product corresponding to the cut surface.

6. one or more computer processors, A first construction step of constructing a first AI capable of extracting features of a product based on image data of a normal product of the product; A second construction step of constructing a second AI capable of pseudo-generating image data of the abnormal product of the target product based on information on the abnormality of the product obtained from the image data of the abnormal product of the product and image data of a normal product of the target product; A third construction step of constructing a third AI capable of determining normality / abnormality for a plurality of image data to be subjected to judgment of the target product using the first AI and the second AI; A method for generating AI for inspection.

7. a second information processing device connectable to the first information processing device; An inspection AI generation device connectable to the second information processing device; An AI inspection system comprising: The inspection AI generation device includes: A first construction unit that constructs a first AI capable of extracting features of a product based on image data of a normal product of the product; A second construction unit that constructs a second AI capable of pseudo-generating image data of the abnormal product of the target product based on information on the abnormality of the product obtained from the image data of the abnormal product of the product and image data of a normal product of the target product; a third construction unit that constructs a third AI capable of determining whether a plurality of image data that are to be used to judge the target product are normal / abnormal, using the first AI and the second AI; having The second information processing device, A first data acquisition unit that acquires data necessary for installing and updating the third AI from the inspection AI generation device; A second data acquisition unit that acquires a plurality of image data that are used to judge the target product; A determination unit that determines whether a plurality of image data items that are to be used to determine the target product are normal or abnormal using the third AI; an output unit that outputs a result of the determination by the determination unit to the first information processing device; An AI inspection system.

8. The inspection AI generation device further includes: a third data acquisition unit that acquires, from the second information processing device, image data of the target product that has been determined to be normal by the determination unit and / or image data of the target product that has been determined to be abnormal by the determination unit; The AI ​​inspection system described in claim 7, characterized in that the third construction unit uses the data acquired by the first AI, the second AI, and the third data acquisition unit to reconstruct a third AI capable of determining whether multiple image data that are the subject of judgment of the target product are normal / abnormal.

9. The AI ​​inspection system according to claim 7, characterized in that a plurality of image data to be used to judge the target product are captured by an imaging device connectable to the first information processing device.

10. The AI ​​inspection system described in claim 8, characterized in that the third data acquisition unit acquires image data of the target products that have been determined to be normal by the judgment unit and / or image data of the target products that have been determined to be abnormal by the judgment unit from a plurality of the second information processing devices arranged at a plurality of locations.

11. a second information processing device connectable to the first information processing device; An inspection AI generation device connectable to the second information processing device; An AI inspection method in an AI inspection system comprising: one or more computer processors, A first construction step of constructing a first AI capable of extracting features of a product based on image data of a normal product of the product; A second construction step of constructing a second AI capable of pseudo-generating image data of the abnormal product of the target product based on information on the abnormality of the product obtained from the image data of the abnormal product of the product and image data of a normal product of the target product; A third construction step of constructing a third AI capable of determining normality / abnormality for a plurality of image data to be subjected to judgment of the target product using the first AI and the second AI; A first data acquisition step of acquiring data necessary for installing and updating the third AI from the inspection AI generation device; a second data acquisition step of acquiring a plurality of image data that are used to judge the target product; A determination step of determining whether a plurality of image data to be used for determining the target product is normal / abnormal using the third AI; an output step of outputting a result of the determination in the determination step to the first information processing device; An AI inspection method that performs the following.

Images