Product inspection system, product inspection method, learning device, and recognition device

Abstract

To inspect products using a film and suppress the occurrence of defective products.SOLUTION: A product inspection system is a product inspection system that inspects a product in which a transparent film is conveyed to a cylindrical member and the conveyed film is wound around the cylindrical member. The product inspection system includes: an imaging unit that captures an image of the product; a measuring unit that measures temperature that affects appearance of the product; and an inspection unit that causes a prediction model to be machine learned using an appearance image of the product manufactured using the film conveyed to the cylindrical member and information on the temperature that affects the appearance of the product as teacher data, inputs the image captured by the imaging unit into the prediction model to obtain inspection information about the appearance of the product, and inspects the appearance of the product based on the inspection information and the temperature measured by the measuring unit.SELECTED DRAWING: Figure 4

Description

Technical Field

[0001] The present invention relates to a product inspection system, a product inspection method, a learning device, a recognition device, a product manufacturing system, and a product manufacturing method.

Background Art

[0002] Conventionally, a technique for inspecting defects in a film to be conveyed is known. As a technique of this kind, there is, for example, a defect inspection system described in Patent Document 1. This defect inspection system identifies the position of a defect in a film using an image obtained by imaging the film, and identifies the type of the defect whose position has been identified.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, although the above-described defect inspection system can identify the type of a defect in a single film, it cannot inspect a product using the film. Further, the above-described defect inspection system cannot suppress the occurrence of defective products.

[0005] One object of the present invention is to provide a product inspection system, a product inspection method, a learning device, a recognition device, a product manufacturing system, and a product manufacturing method that can inspect a product using a film and suppress the occurrence of defective products.

Means for Solving the Problems

[0006] A product inspection system according to one aspect of the present invention is a product inspection system that conveys a transparent film to a cylindrical member and inspects a product in which the conveyed film is wound around the cylindrical member, the system including an imaging unit that images the product, a measurement unit that measures the temperature near the cylindrical member, and a prediction model is machine-learned using, as teacher data, an appearance image of the appearance including wrinkles of the product, the temperature near the cylindrical member, and information regarding the impression of the appearance including wrinkles of the product. The image captured by the imaging unit is input to the prediction model Input , an inspection unit that stores information regarding the impression of the appearance including wrinkles of the product output from the prediction model and the temperature near the cylindrical member output from the prediction model. The information regarding the impression of the product is inspection result indicating good or bad by visual inspection of a manufacturing worker, questionnaire result of an impression when a customer first opens the product, or information based on past complaints. It is a product inspection system.

[0007] A product inspection method according to one aspect of the present invention is a product inspection method that conveys a transparent film to a cylindrical member and inspects a product in which the conveyed film is wound around the cylindrical member, the method including a step of imaging the product, a step of measuring the temperature near the cylindrical member, and the appearance image of the product Appearance including wrinkles , the temperature near the cylindrical member, and information regarding the impression of the product Of the appearance including wrinkles are used as teacher data to machine-learn a prediction model. An image captured is input to the prediction model, and information regarding the impression Of the appearance including wrinkles of the product output from the prediction model , and save the temperature near the cylindrical member output from the prediction model is included. The information regarding the impression of the product is inspection result indicating good or bad by visual inspection of a manufacturing worker, questionnaire result of an impression when a customer first opens the product, or information based on past complaints. It is a product inspection method.

[0008] A learning device according to an aspect of the present invention acquires learning data including an appearance image of a product with wrinkles, which is obtained by transporting a transparent film to a cylindrical member and winding the transported film around the cylindrical member, the temperature near the cylindrical member, and information related to the impression of the appearance of the product including the wrinkles. The learning device further includes a learning data acquisition unit and a learning processing unit that learns a prediction model. The learning processing unit learns the processing parameters of the prediction model so that the prediction model outputs information related to the impression of the appearance of the product including the wrinkles and the temperature near the cylindrical member from an image captured by an imaging unit. The information related to the impression of the product is inspection result indicating good or bad by visual inspection of a manufacturing worker, questionnaire result of the impression when a customer first opens the product, or information based on past complaints. Input A learning device, comprising: a learning processing unit that learns a prediction model, the learning processing unit learning the processing parameters of the prediction model so that the prediction model outputs information related to the impression of the appearance of the product including the wrinkles and the temperature near the cylindrical member from an image captured by an imaging unit; wherein the information related to the impression of the product is inspection result indicating good or bad by visual inspection of a manufacturing worker, questionnaire result of the impression when a customer first opens the product, or information based on past complaints.

[0009] A recognition device according to an aspect of the present invention includes a captured image acquisition unit that acquires a captured image of a film being transported to a cylindrical member in a manufacturing process of a product in which the transported film is wound around the cylindrical member, a measurement unit that measures the temperature near the cylindrical member, and an inspection unit that inputs the image captured by the captured image acquisition unit into a prediction model that is machine-learned using, as teacher data, the appearance image of the product, the temperature near the cylindrical member, and information related to the impression of the product, and outputs information related to the impression output from the prediction model. The information related to the impression of the product is inspection result indicating good or bad by visual inspection of a manufacturing worker, questionnaire result of the impression when a customer first opens the product, or information based on past complaints. Appearance including wrinkles The appearance image of the product, the temperature near the cylindrical member, and information related to the impression of the product Of the appearance including wrinkles A recognition device, comprising: an inspection unit that inputs the image captured by the captured image acquisition unit into a prediction model that is machine-learned using, as teacher data, the appearance image of the product, the temperature near the cylindrical member, and information related to the impression of the product, and outputs information related to the impression output from the prediction model; wherein the information related to the impression of the product is inspection result indicating good or bad by visual inspection of a manufacturing worker, questionnaire result of the impression when a customer first opens the product, or information based on past complaints. Of the appearance including wrinkles of the product Information related to the impression , and save the temperature near the cylindrical member output from the prediction model A recognition device, comprising: an inspection unit that inputs the image captured by the captured image acquisition unit into a prediction model that is machine-learned using, as teacher data, the appearance image of the product, the temperature near the cylindrical member, and information related to the impression of the product, and outputs information related to the impression output from the prediction model; wherein the information related to the impression of the product is inspection result indicating good or bad by visual inspection of a manufacturing worker, questionnaire result of the impression when a customer first opens the product, or information based on past complaints.

Advantages of the Invention

[0012] According to the present invention, it is possible to inspect a product using a film and suppress the continuous occurrence of defective products.

Brief Description of the Drawings

[0013]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Modes for Carrying Out the Invention

[0014] Hereinafter, with reference to the drawings, embodiments of a product inspection system, a product inspection method, a learning apparatus, a recognition apparatus, a product manufacturing system, and a product manufacturing method will be described.

[0015] The product inspection system of the embodiment inspects products using a film. In the embodiment, the film is a film that deforms with temperature like a thermoplastic resin, for example, a packaging film such as polyvinylidene chloride, vinyl chloride resin, polyethylene, etc. The product is, for example, a wrap film product in which the film is wound around a core member such as a cylindrical shape. The product in the embodiment is one in which the film is wound around a core member such as a cylindrical shape, and does not include a cosmetic case or the like in which the product is packaged. Note that the product is not limited to a wrap film product as long as it is a product in which a film that deforms with temperature is conveyed and wound around a core member, as will be described later.

[0016] [Manufacturing Process of Product] FIG. 1 is a flowchart showing an example of the manufacturing process of a wrap film product in the embodiment. The manufacturing process of the wrap film product includes, for example, a pretreatment process (step S100), a rewinding process (step S102), an inspection / control process (step S104), a sorting process (step S106), and a post-treatment process (step S110).

[0017] The pretreatment process (step S100) includes, for example, a process for manufacturing a wrap film product and a process for inspecting a wrap film product. The process for manufacturing a wrap film product includes a process for manufacturing a wide roll mother roll, a process for installing the wide roll mother roll, and a process for installing a core member. The wide roll mother roll is obtained by winding a film used for a wrap film product around a first cylindrical member. A film for manufacturing a number of wrap film products is wound around the first cylindrical member, for example. The process for inspecting a wrap film product includes a process for setting conditions for detecting wrinkles or the like in the film and a learning process for constructing a recognition engine for inspecting a wrap film product.

[0018] The rewinding process (step S102) is a process of conveying the film in the wide roll mother roll to a core member (second cylindrical member) and rewinding the conveyed film around the core member.

[0019] The inspection and control process (step S104) is a process of automatically inspecting the wrap film product and a process of controlling an air conditioner or the like based on the inspection result. Specifically, the inspection process acquires information on the image of the wrap film product and the temperature that affects the appearance of the product, and inspects the wrap film product using the film based on the image of the film conveyed from the large roll mother roll to the core member. The control process is a process of adjusting the temperature that affects the appearance of the product in order to manufacture a wrap film product as a good product.

[0020] The sorting process (step S106) is a process of sorting the wrap film products based on the result of the inspection process. The sorting process includes, for example, a process of excluding the wrap film products determined to have a defective impression in the inspection process from the post-treatment process.

[0021] The post-treatment process (step S108) is a process after the product manufacturing system described in the embodiment. The post-treatment process includes, for example, a process of packaging the wrap film product, a process of visually inspecting the wrap film product, a learning feedback process including a process of acquiring an image for learning processing, etc. Note that the post-treatment process may include the above-described sorting process. The post-treatment process may include a process of discharging defective products from the manufacturing apparatus for the small roll wrap film products rewound from the large roll mother roll.

[0022] [Product manufacturing apparatus] FIG. 2 is a diagram showing an example of a manufacturing apparatus for manufacturing a wrap film product according to the embodiment. The manufacturing apparatus 100 includes, for example, a conveyance unit 110, a control device 140 and a measurement device 140A, imaging devices 150A, 150B, 150C, and 150D, and an illumination device (light source) 160. When collectively referring to the imaging devices 150A, 150B, 150C, and 150D, they are simply described as "imaging device 150".

[0023] The conveying unit 110 includes, for example, a motor or the like, and rotates the first cylindrical member 124 and the second cylindrical member 134. The second cylindrical member 134 is, for example, a paper tube. By rotating the first cylindrical member 124 and the second cylindrical member 134, the film 122 wound around the thick roll web 120 is conveyed as the film 200 along the conveying direction (Y direction), and the film 200 is wound around the second cylindrical member 134. Thereby, the manufacturing apparatus 100 manufactures the wrap film product 130 in which the film 132 is wound around the second cylindrical member 134.

[0024] Note that the manufacturing apparatus 100 includes a motor or the like as a mechanism for conveying the film 200, but may also include a mechanism such as a roller for adjusting the conveying direction of the film 200, a mechanism for adjusting the tension of the film 200 in the conveying direction and the width direction, and a mechanism for cutting the film 200 to a length required for manufacturing one wrap film product 130. The manufacturing apparatus 100 may include, for example, a jig that is arcuately curved in the X direction as a means for stretching the wrinkles in the Y direction of the film 200. Further, the manufacturing apparatus 100 may include a sensor for detecting the tension in the width direction of the film 200 and a mechanism for controlling the tension in the width direction of the film 200.

[0025] The control device 140 controls each part of the manufacturing device 100. The control device 140 controls, for example, the conveyance unit 110 and the illumination device 160, and acquires image data from the imaging device 150. The imaging devices 150A, 150B, 150C, and 150D respectively image positions on a line parallel to the width direction (X direction) of the wrap film product 130. Thereby, the imaging devices 150A, 150B, 150C, and 150D can image the state extending across both ends in the width direction of the wrap film product 130. The imaging devices 150A, 150B, 150C, and 150D respectively output image data to the control device 140. The illumination device 160 is provided at a position where light 160# is irradiated onto the line when imaging is performed by the imaging device 150. The illumination device 160 is installed such that the traveling direction of the light 160# and the conveyance direction intersect. Note that it is desirable for the illumination device 160 to irradiate the light 160# so that a shadow is imaged due to wrinkles parallel to the conveyance direction.

[0026] The measuring device 140A is a measuring unit that measures the temperature that affects the appearance of the product. The temperature measured by the measuring device 140A is supplied to the control device 140. The control device 140 controls an air conditioner (see FIG. 3) based on the supplied temperature information.

[0027] FIG. 3 is a diagram showing an example of the configuration of the manufacturing device 100 in the embodiment. The manufacturing device 100 includes, inside the device 100A, each part for manufacturing the wrap film product 130 by conveying the film 200 from the large roll stock 120. The manufacturing device 100 supports a plurality of large roll stocks (the large roll stock 120 and the spare large roll stock 120#) on the turret arm 213 of the supply mechanism 210 in a replaceable manner. The film 200 of the large roll stock 120 supported on the turret arm 213 is wound around the take-up drum 230 via the roller mechanism 220. The wound film 200 is supplied to the downstream empty core 231 and rewound onto the core 231.

[0028] The supply position 211 and the standby position 212 of the large roll stock 120 face each other, and a new large roll stock 120 # around which the required amount of film 122 is wound is automatically supplied from upstream to the upstream standby position 212. Below the standby position 212 of the large roll stock 120, conveyors such as a belt conveyor or a roller conveyor for collecting the used large roll stock 120 are installed as required. Although this embodiment has described an example in which the film 122 is automatically supplied, the present invention is not limited thereto, and the present invention is also applicable to a system in which the film 122 is supplied manually.

[0029] The roller mechanism 220 includes a pay-out roller 221 around which the film 200 from the large roll stock 120 positioned and arranged at the supply position 211 is wound, a plurality of guide rollers 222 around which the film 200 from the pay-out roller 221 is wound, an expander roller 223 around which the film 200 from the plurality of guide rollers 222 is wound, and a pinch roller 224 around which the film 200 between the expander roller 223 and the downstream take-up drum 230 is wound.

[0030] The pay-out roller 221 is swingably disposed between the large roll stock 120 at the supply position 211 and the plurality of guide rollers 222 downstream in the upper direction. The pay-out roller 221 is in sliding contact with the lower part downstream of the large roll stock 120 at the supply position 211 and functions to pay out and supply the film 200 to the plurality of guide rollers 222. The plurality of guide rollers 222 are interposed between the pay-out roller 221 upstream in the lower direction and the expander roller 223 downstream, and are pivotally supported at a predetermined interval in the vertical direction so as to be rotatable freely.

[0031] Each guide roller 222 is configured such that a support shaft is inserted through the center of a cylindrical roller portion that is in sliding contact with the film 200, for example. Both end portions of each guide roller 222 protruding from the roller portion of the support shaft are rotatably supported via bearings. Each guide roller 222 is located downstream of the pay-out roller 221 and changes the supply direction of the film 200 or removes wrinkles and slack of the film 200.

[0032] The expander roller 223 has, for example, a support shaft inserted through a rubber roller portion, and both end portions protruding from the roller portion of the support shaft are rotatably supported via bearings. It is positioned curved downstream of the lowermost guide roller 222 to remove wrinkles and slack in the film 200. Further, the pinch roller 224 has, for example, a support shaft inserted through a rubber cylindrical roller portion, and both end portions protruding from the roller portion of the support shaft are rotatably supported via bearings. It is positioned downstream of the expander roller 223 to supply the film 200 to the take-up drum 230 at a constant speed.

[0033] The take-up drum 230 has a paper core 231 formed in a cylindrical shape sequentially supplied, rotates in sliding contact with the core 231, and functions to wind the film 200 supplied from the pinch roller 224 around the outer peripheral surface of the core 231 for a predetermined length.

[0034] In the above configuration, when manufacturing the wrap film product 130, the manufacturing apparatus 100 is started and the wide roll mother roll 120 at the supply position 111 is rotated in the forward direction. Then, the film 122 wound around the wide roll mother roll 120 is sequentially fed and supplied to the unwind roller 221, the plurality of guide rollers 222, the expander roller 223, the pinch roller 224, and the take-up drum 230, and is wound around the outer peripheral surface of the core 231 on the take-up drum 230 for a predetermined length.

[0035] In the above configuration, the manufacturing apparatus 100 includes an in-apparatus temperature sensor 310 and an out-of-apparatus temperature sensor 320 as a measuring device 140A, an air-cooling fan 300 as an in-apparatus air-conditioning device, and an out-of-apparatus air-conditioning device 322 in order to suppress appearance wrinkles in the wrap film product 130. The manufacturing apparatus 100 supplies the in-apparatus temperature detected by the in-apparatus temperature sensor 310 and the out-of-apparatus temperature detected by the out-of-apparatus temperature sensor 320 to the control device 140.

[0036] In order to measure the temperature of the film 200 being conveyed, the in-device temperature sensor 310 preferably detects the temperature near the core 231 that is likely to affect the quality of the wrap film product 130. The wrap film product 130 preferably detects the temperature near the core 231 in order to detect the ambient temperature inside the device 100A. Also, the in-device temperature sensor 310 may detect the temperature of a guide roller 222 or the like that serves as a conveyance path for the film 200, and it may be near a roller such as the guide roller 222 that does not affect the replacement of the wide roll mother roll 120 or the cutting operation of the film 200. Furthermore, the in-device temperature sensor 310 may measure the temperature of the film 200 immediately after being conveyed and wound around the second cylindrical member 134. Note that the surface of the second cylindrical member 134 is high-quality paper with low gloss compared to a metal roll with a metal coating, and the temperature measurement can be stably performed compared to metals such as the guide roller 222.

[0037] Furthermore, the air-cooling fan 300 and the out-of-device air conditioner 322 control the temperature and flow rate of the air flow according to a control signal from the control device 140. Also, a medium passage 222A is formed inside the support shaft of each guide roller 222. A temperature-adjusted medium is supplied to the medium passage 222A from a pump (not shown), whereby the surface temperature of each guide roller 222 is adjusted.

[0038] An air-cooling fan 300 is provided below the take-up drum 230. The air-cooling fan 300 supplies an air flow from below upward to the take-up drum 230 and the core 231 according to the control of the control device 140. The air-cooling fan 300 has a damper (not shown) driven by the control of the control device 140 to control the supply and stop of the air flow. The air-cooling fan 300 only needs to be able to supply an air flow to locations that affect the temperature of the film 200, such as not only the take-up drum 230 and the core 231, but also, for example, the large roll stock 120, the guide roller 222, the roller mechanism 220, and the film 200 itself being conveyed. Since, for example, wrinkles in the film 200 start from the position where the film 200 is pinched (contacted) by the take-up drum 230, it is desirable to supply an air flow to this position. In addition, in order to suppress blowing dust onto the core 231 together with the air flow from the air-cooling fan 300, it is necessary to take measures such as blowing the air flow at a location avoiding the core 231 or blowing the air flow via a dust filter.

[0039] In order to suppress wrinkles in the wrapped film product 130 caused by the temperature of the film 200, it is desirable that the temperature measurement position of the film 200 be a position close to the core 231. Also, the temperature measurement position of the film 200 may be between rollers such as the guide roller 222 or on the roll, at a location where the film 200 is a single layer.

[0040] Furthermore, in order to estimate the temperature of the film 200, the temperature of either of the two large roll stocks 120, 120# may be measured. For example, table data or calculation formulas representing the correlation between the temperatures of the large roll stocks 120, 120# and the temperature inside the apparatus 100A may be prepared, and the temperature inside the apparatus 100A may be estimated based on the temperatures of the large roll stocks 120, 120#. Table data or calculation formulas representing the correlation between the temperature inside the apparatus 100A and the temperatures of the core 231 and the take-up drum 230 may be set, and the temperatures of the core 231 and the take-up drum 230 may be estimated based on the temperature inside the apparatus 100A.

[0041] Furthermore, the in-device temperature sensor 310 may detect the temperature at an arbitrary position inside the device 100A, and estimate the temperatures at other positions inside the device 100A in consideration of the thermal conductivity of the film 200 and heat transfer due to the operation of the manufacturing device 100. For example, the temperature of the guide roller 222 may be detected, and the temperature of the core 231 may be estimated in consideration of heat transfer from the guide roller 222 to the core 231. Thereby, the temperature inside the device 100A can be detected from a plurality of positions inside the device 100A.

[0042] Furthermore, in the manufacturing device 100, it is desirable to use a sensor that detects the temperature on the long surface in the width direction of the film 200 to uniformly control the temperature in the width direction of the film 200. Thereby, the temperature can be controlled so as to suppress wrinkles generated throughout the width direction of the wrap film product 130. Also, the in-device temperature sensor 310 may detect, for example, the temperature of one of the guide rollers 222 itself and the temperature of the film 200 on the other guide roller 222. Furthermore, the in-device temperature sensor 310 may measure the change in the temperature of the film 200 by detecting the temperatures at a plurality of locations. Furthermore, by using one temperature sensor for the film 200 rather than using a plurality of temperature sensors in the width direction of the film 200, the number of components and costs can be reduced.

[0043] [Product Inspection System] FIG. 4 is a block diagram showing an example of a product inspection system according to an embodiment. The product inspection system 400 includes, for example, a control device 140, a product inspection device 410, and a terminal device 500. The control device 140, the product inspection device 410, and the terminal device 500 are connected to, for example, a communication network. Each device connected to the communication network includes a communication interface (not shown) such as a NIC (Network Interface Card) or a wireless communication module. The communication network includes, for example, the Internet, a WAN (Wide Area Network), a LAN (Local Area Network), a cellular network, and the like.

[0044] The control device 140 includes a control unit 142 and a notification unit 144. Functional units such as the control unit 142 and the notification unit 144 are realized, for example, by a processor such as a CPU (Central Processing Unit) executing a program stored in a program memory. Also, some or all of these functional units may be realized by hardware such as an LSI (Large Scale Integration), an ASIC (Application Specific Integrated Circuit), or an FPGA (Field-Programmable Gate Array), or may be realized by the cooperation of software and hardware. The control unit 142 controls each part of the manufacturing device 100 and transmits the captured image acquired from the imaging device 150 to the product inspection device 410. The control unit 142 receives the inspection result. The notification unit 144 includes a display or the like and performs display or the like based on the inspection result.

[0045] The terminal device 500 is a portable terminal device equipped with a camera device such as a smartphone or a tablet terminal. It is desirable for the terminal device 500 to have illumination (not shown) that assists imaging by the camera device. The terminal device 500 images the wrap film product 130 based on a user's operation. The user is, for example, a worker who inspects the wrap film product 130 visually or the like. The user inspects the state of the wrap film product 130. The terminal device 500 acquires a captured image of the wrap film product 130 as information regarding the external impression of the film 132 in the wrap film product 130. The captured image is, for example, a correct image of the wrap film product 130 that gives the impression of being a non-wrinkled good product, but is not limited to this, and may be a negative example image including wrinkles in the wrap film product 130. The terminal device 500 transmits the positive example image to the product inspection device 410. Note that the terminal device 500 may add information indicating that "the impression is good or bad", "the wrinkles do not disappear due to aging after product manufacturing", or "the wrinkles are continuous in a direction intersecting the film conveyance direction" to the image and transmit it based on the user's operation.

[0046] The product inspection device 410 includes, for example, a recognition device 412, a learning device 414, and a learning image database 416. Functional units such as the recognition device 412 and the learning device 414 are realized by, for example, a processor such as a CPU executing a program stored in a program memory. The learning image database 416 includes, for example, a storage device such as an HDD device and a database management device.

[0047] FIG. 5 is a block diagram showing an example of the learning device 414 in the embodiment. The learning device 414 includes, for example, a learning data acquisition unit 420, a learning processing unit 424, a convolutional neural network (CNN) 422, and a learning result storage unit 426. The learning data acquisition unit 420 acquires, from the terminal device 500, as learning data, a learning image, an inspection result, an internal temperature of the device, an external temperature of the device, and information regarding the impression of the wrap film product 130. Note that the learning data may be read as a set of teacher data.

[0048] FIG. 6 is a diagram showing the relationship between the input variable and the output variable in the convolutional neural network 422. The learning image corresponds to the input variable in the convolutional neural network 422, and the inspection result, the internal temperature of the device, the external temperature of the device, and the information regarding the impression of the wrap film product 130 correspond to the output variables in the convolutional neural network 422. The output variable corresponds to the inspection information that the recognition device 320 should output. Note that the impression of the wrap film product 130 may be information based on, for example, an inspection result by visual inspection of a manufacturing worker, a questionnaire result of an impression when a customer first opens the product, and past complaint (claim) cases.

[0049] When the learning processing unit 424 inputs the learning images among the learning data into the convolutional neural network 422, it learns the processing parameters of the convolutional neural network 422 so as to output the output variable among the learning data from the convolutional neural network 422. Specifically, the learning processing unit 424 recursively calculates (updates) the processing parameters using the learning data so that the difference between the output variable corresponding to the input variable shown in FIG. 6 and the output of the convolutional neural network 422 becomes small. The learning processing unit 424 performs, for example, deep learning in order to obtain the processing parameters. Deep learning is machine learning using a multi-layer structure, particularly a neural network having three or more layers. In the embodiment, a convolutional neural network 422 is used as the neural network having a multi-layer structure. The learning processing unit 424 stores the updated processing parameters in the learning result storage unit 426.

[0050] FIG. 7 is a block diagram showing an example of the recognition device 412 in the embodiment. The recognition device 412 includes a photographed image acquisition unit 430, a convolutional neural network 432, a recognition processing unit 434, and a recognition result storage unit 436. The photographed image acquisition unit 430 acquires a photographed image from the control device 140. The convolutional neural network 432 is a prediction model in which the processing parameters stored in the learning result storage unit 426 are set. The convolutional neural network 432 inputs the photographed image and outputs a recognition result such as an inspection result. The recognition processing unit 434 stores the recognition result output from the convolutional neural network 432 in the recognition result storage unit 436. The recognition result includes information corresponding to the learning data (set of teacher data) shown in FIG. 6. That is, the recognition result includes an inspection result, an appropriate temperature inside the device, an appropriate temperature outside the device, and information regarding the impression of the wrap film product 130.

[0051] FIG. 8 is a diagram showing an example of recognition results in the embodiment. The recognition processing unit 412 stores the captured image ID, the recognition result as the output of the convolutional neural network 432, and the product ID in association with each other. The product ID is identification information of the wrap film product 130, and is, for example, the manufacturing number or lot number of the wrap film product 130. The recognition processing unit 434 acquires, for example, the captured image ID corresponding to the captured image from the control device 140 and the product ID related to the captured image. The recognition processing unit 434 associates the captured image ID and the product ID with the recognition result output from the convolutional neural network 432 and stores the result in the recognition result storage unit 436. Note that the product ID may be information recognized based on an image captured by the imaging device 150 of a character string printed on the second cylindrical member 134, for example.

[0052] The recognition device 412 outputs the information stored in the recognition result storage unit 436 to the control device 140. The notification unit 144 notifies, for example, the recognition result corresponding to the product ID. Thereby, the control device 140 can cause the sorting process and the post-processing process based on the recognition result to be performed for each wrap film product 130 corresponding to the product ID.

[0053] The control unit 142 calculates the temperature difference inside the device between the temperature inside the device acquired from the in-device temperature sensor 310 and the appropriate temperature inside the device acquired from the recognition device 412, and outputs a temperature control signal for controlling the air-cooling fan 300 and / or the out-of-device air conditioner 322 so as to reduce the temperature difference inside the device. Similarly, the control unit 142 calculates the temperature difference outside the device between the temperature outside the device acquired from the out-of-device temperature sensor 320 and the appropriate temperature outside the device acquired from the recognition device 412, and outputs a temperature control signal for controlling the air-cooling fan 300 and / or the out-of-device air conditioner 322 so as to reduce the temperature difference outside the device. Further, the notification unit 144 may prompt the administrator of the manufacturing device 100 to operate the damper and the duct by displaying display information that prompts reducing the temperature difference inside the device and / or the temperature difference outside the device. Thereby, for example, problems such as components contained in the film 200 bleeding out due to the temperature of the film 200 rising too much and winding around the guide roller 222 etc. on the pass line of the film 200, or the winding state of the film end being disturbed as a problem with the wrap film product 130 can be suppressed.

[0054] Also, although it is desirable for the manufacturing device 100 to adjust the temperatures of both the film 200 conveyed from the large roll raw material 120 and the large roll raw material 120, by adjusting the temperature of the conveyed film 200 rather than adjusting the temperature of the large roll raw material 120, problems with the film 200 can be effectively suppressed. The reason for this is that when rewinding the film 200, it comes into contact with a plurality of rolls inside the manufacturing device 100, and the temperature of the large roll raw material 120 changes due to heat conduction. Therefore, even if the temperature of the large roll raw material 120 is precisely temperature-controlled, it has a large impact on the temperature inside the device 100A.

[0055] Furthermore, it is desirable for the manufacturing apparatus 100 to adjust the temperature of the spare wide roll stock 120# in addition to the wide roll stock 120 from which the film 200 being rewound is fed out. For this purpose, the manufacturing apparatus 100 adjusts the temperature of the wide roll stock 120# and the placement location of the spare wide roll stock 120# based on the temperature detected by the in-apparatus temperature sensor 310 or the out-of-apparatus temperature sensor 320, so as to stabilize the temperature of the spare wide roll stock 120#. Thereby, the manufacturing apparatus 100 can suppress the frequency of temperature adjustment of the interior 100A of the apparatus when the temperature of the wide roll stock 120 is not appropriate, and can also be expected to reduce the wear of the sensor switch and prevent failures by reducing the number of times the damper opens and closes.

[0056] [Convolutional Neural Network] Hereinafter, an example of the convolutional neural network 422 and the convolutional neural network 432 in the embodiment will be described. In this description, the convolutional neural network 422 and the convolutional neural network 432 are collectively referred to as the "convolutional neural network".

[0057] As described above, the learning processing unit 424 sets the pixel value of the learning image as an input variable input to the input layer of the convolutional neural network 422 for learning, and the inspection information as an output variable output from the output layer. The learning processing unit 424 performs machine learning using the learning dataset of the learning image and the inspection information. The recognition processing unit 434 inputs the pixel value of the captured image to the input layer of the learned convolutional neural network 323 and acquires the inspection information from the output layer.

[0058] FIG. 9 is a diagram for explaining a process of performing recognition processing using a convolutional neural network in an embodiment. The convolutional neural network includes, for example, layer L0, layer L1, layer L2, layer Li, and layer LI. Layer L0 is an input layer, layers L1 to Li are intermediate layers or hidden layers, and layer LI is also called an output layer. In the convolutional neural network, an input image is input to the input layer L0. The input image is represented by a pixel matrix D11 having the vertical position and the horizontal position of the input image as matrix positions. Each element of the pixel matrix D11 includes an R (red) sub-pixel value, a G (green) sub-pixel value, and a B (blue) sub-pixel value as sub-pixel values of a pixel corresponding to the matrix position. The first intermediate layer L1 is a layer in which a convolution process (also called a filter process) and a pooling process are performed.

[0059] (Convolution) An example of the convolution process of the intermediate layer L1 will be described. The convolution process is a process of applying a filter to an original image and outputting a feature map. Specifically, the input pixel values are divided into an R sub-pixel matrix D121, a B sub-pixel matrix D122, and a G sub-pixel matrix D123, respectively. Each sub-pixel matrix D121, D122, D123 (each also referred to as "sub-pixel matrix D12") has the elements of each element of the sub-matrix and the s-row t-column convolution matrix CM1 (also called a kernel) multiplied and added for each s-row t-column sub-matrix, whereby a first pixel value is calculated. The first pixel values calculated for each sub-pixel matrix D12 are multiplied and added with weight coefficients, respectively, whereby a second pixel value is calculated. The second pixel value is set as each element of the convolution image matrix D131 as a matrix element corresponding to the position of the sub-matrix. By shifting the position of the sub-matrix for each element (sub-pixel) in each sub-pixel matrix D12, the second pixel value at each position is calculated, and all matrix elements of the convolution image matrix D131 are calculated.

[0060] FIG. 9 is an example of a convolutional neural network using, for example, a 3-row 3-column convolution matrix CM (convolution matrix) 1. The convolution pixel value D1311 has its first pixel value calculated for a 3-row 3-column sub-matrix from the second row to the fourth row and from the second column to the fourth column of each sub-pixel matrix D12. By calculating and adding weight coefficients to the first pixel values of each of the sub-pixel matrices D121, D122, and D123, a second pixel value is calculated as the matrix element at the second row and second column of the convolution image matrix D131. Similarly, a second pixel value of the matrix element at the third row and second column of the convolution image matrix D131 is calculated from a sub-matrix from the third row to the fifth row and from the second column to the fourth column. Also similarly, convolution image matrices D132, ··· are calculated using other weighting coefficients or other convolution matrices.

[0061] (Pooling) An example of the pooling process for the intermediate layer L1 will be described. The pooling process is a process of reducing an image while retaining the features of the image. Specifically, for each region PM (pooling matrix) at the u-th row and v-th column in the convolution image matrix D131, a representative value of the matrix elements within the region is calculated. The representative value is, for example, the maximum value of the matrix elements within the region. The representative value is set as each element of the CNN image matrix D141 as the matrix element corresponding to the position of the region PM. By shifting the regions in the convolution image matrix D131 for each region PM, representative values at each position are calculated, and all elements of the CNN image matrix D141 are calculated.

[0062] FIG. 9 is an example of a convolutional neural network using, for example, a 2-row 2-column region PM. For a 2-row 2-column region PM from the third row to the fourth row and from the third column to the fourth column of the convolution image matrix D131, the maximum value within the region PM is calculated as the representative value. This representative value is set as the matrix element at the second row and second column of the CNN image matrix D141. Similarly, a representative value of the matrix element at the third row and second column of the CNN image matrix D141 is calculated from a sub-matrix from the fifth row to the sixth row and from the second column to the fourth column. Also similarly, CNN image matrices D142, ··· are calculated from the convolution image matrices D132, ···.

[0063] Each matrix element (N elements) of the CNN image matrices D141, D142, ··· is arranged in a predetermined order to generate a vector X. The vector X includes N elements x n (n = 1, 2, 3, ··· N). The vector X corresponds to the vector u (0) .

[0064] The intermediate layer Li represents the i-th (i = 2, ···) intermediate layer. The intermediate layer Li includes the vector u (i) . From each node of the intermediate layer Li, a vector z (i) is output. The vector z (i) has a value obtained by inputting the vector u (i) into the function f(u (i) ) where the function f is an activation function. The vector u (i) has a value obtained by adding the value obtained by multiplying the vector z (i-1) output from the nodes of the (i - 1)-th intermediate layer by the weight matrix W (i) and the vector b (i) . The vector b (i) is a bias.

[0065] The output layer LI includes the vector z (I-1) . The output of the output layer LI is M y m (m = 1, 2, ··· M). That is, the output layer LI outputs a vector Y (y1, y2, y3, ··· y m ) having M y M as elements. Thus, when pixel values of an image are input as input variables, the convolutional neural network outputs a vector Y as an output variable. The vector Y in the embodiment represents inspection information.

[0066] FIG. 10 is a diagram for explaining the learning process using the convolutional neural network in the embodiment. Let the vector output from the first intermediate layer L1 be vector [X] with respect to the pixel values of the images in the learning dataset. Let the vector representing the definite class in the learning dataset be vector [Y]. The definite class is the output coefficient (inspection information such as inspection results, number of wrinkles, density of wrinkles, visual impression, etc.) shown in FIG. 6 in the embodiment.

[0067] Weight matrix W (i) is set with an initial value. When vector [X] is input to the second intermediate layer L2 based on the input image being input to the input layer L0, vector Y(X) based on vector [X] is output from the output layer LI. The error E between vector Y(X) and vector [Y] is calculated using a loss function. The gradient ΔE of the i-th layer i is calculated using the output z from each layer i and the error signal δ. i The error signal δ i is calculated using the error signal δ. i-1 Note that the process of calculating the error signal on the input layer side from the error signal on the output layer side from the output layer LI towards the input layer L0 is also called backpropagation. The weight matrix W (i) is updated based on the gradient ΔE. i Similarly, in the first intermediate layer L1, the convolution matrix CM or the weight coefficient is also updated.

[0068] [Setting of Prediction Model] The learning processing unit 424 sets processing parameters such as the number of layers, the number of nodes in each layer, the connection method of nodes between layers, activation functions, error functions, and gradient descent algorithms, pooling regions, kernels, weight coefficients, and weight matrices for the convolutional neural network. For example, the learning processing unit 424 sets the number of layers to 3 (I = 3). The learning processing unit 424 sets the number of nodes in each layer (also referred to as the "number of nodes") to 800 for the number of elements of the vector X (number of nodes N), 500 for the number of nodes in the second intermediate layer (i = 2), and 10 for the output layer (i = 3). However, the embodiment is not limited to this, and the total number may be 4 or more layers, and different values may be set for the number of nodes.

[0069] The learning processing unit 424 creates 20 5x5 convolutional matrices CM and sets a 2x2 region PM. However, the embodiment is not limited to this, and different numbers of matrices or different numbers of convolutional matrices CM may be set. Also, a region PM with a different number of matrices may be set. The learning processing unit 424 may perform more convolutional processing or pooling processing.

[0070] The learning processing unit 424 sets full connection for the connection of each layer of the convolutional neural network. However, the embodiment is not limited to this, and the connection of some or all layers may be set to non-full connection. The learning processing unit 424 sets the sigmoid function as the activation function for all layers. However, the embodiment is not limited to this, and the activation function for each layer may be other activation functions such as step functions, linear combinations, softsign, softplus, ramp functions, clipped power functions, polynomials, absolute values, radial basis functions, wavelets, maxout, etc. Also, the activation function of a certain layer may be of a different type from other layers.

[0071] The learning processing unit 424 sets the squared loss (mean squared error) as the error function. However, the present invention is not limited to this, and the error function may be cross entropy, τ-quantile loss, Huber loss, ε-insensitive loss (ε tolerance error function). Further, the learning processing unit 424 sets SGD (stochastic gradient descent) as an algorithm (gradient descent algorithm) for calculating the gradient. However, the present invention is not limited to this, and algorithms such as Momentum (inertial term) SDG, AdaGrad, RMSprop, AdaDelta, Adam (Adaptive moment estimation), etc. may be used for the gradient descent algorithm.

[0072] The learning processing unit 424 may set not only a convolutional neural network (CNN), but also other neural networks such as a perceptron neural network, a recurrent neural network (RNN), a residual network (ResNet). Further, the learning processing unit 424 may set, in part or in whole, a supervised learning prediction model such as a decision tree, a regression tree, a random forest, a gradient boosting tree, linear regression, logistic regression, or an SVM (support vector machine).

[0073] [Effects of the Embodiment] As described above, according to the product inspection system 400 of the embodiment, it is a product inspection system that conveys the transparent film 200 to the core 231 and inspects the wrap film product 130 in which the conveyed film 200 is wound around the core 231. An imaging device 150 that images the wrap film product 130, a measuring device 140A (an in-device temperature sensor 310 and an out-of-device temperature sensor 320) that measures the temperature that affects the appearance of the wrap film product 130, and a wrap film product 130 manufactured using the film 200 conveyed to the core 231. The convolutional neural network 422 is machine-learned using the appearance image and the temperature information that affects the production of the film 200 as teacher data. By inputting the temperature measured by the measuring device 140A into the convolutional neural network 422, inspection information regarding the appearance of the wrap film product 130 is obtained, and based on the inspection information and the temperature measured by the measuring device 140A, a product inspection device 410 that inspects the appearance of the wrap film product 130 is provided. The product inspection system 400 can be realized. For example, when an image is input to the convolutional neural network 432, the product inspection system 400 can inspect the wrap film product 130 based on the appropriate temperature output from the convolutional neural network 432 and the temperature actually measured by the measuring device 140A. As a result, according to the product inspection system 400 of the embodiment, the wrap film product 130 can be inspected using the image captured in the manufacturing process. Further, according to the product inspection system 400, the inspection accuracy is increased by repeatedly machine-learning the convolutional neural network 422 using the appearance image of the wrap film product 130 and the temperature information that affects the production of the film 200 as teacher data, and the continuous occurrence of defective wrap film products 130 can be suppressed. The product inspection system 400 can suppress the continuous occurrence of defective wrap film products 130, for example, by operating so as to bring the temperature actually measured by the measuring device 140A closer to an appropriate temperature.

[0074] In the product inspection system 400 of the embodiment, as the temperature affecting the production of the film 200, the temperature of the large roll mother roll 120 or the film 200 may be measured. However, when actually measuring the temperature of the conveyed film 200, since the film 200 is a colorless and transparent thin film with a thickness of less than the order of 10 microns and is conveyed at high speed, it may be difficult to accurately measure the temperature even using a non-contact sensor such as a laser. In contrast, the product inspection system 400 desirably measures the temperature inside the apparatus 100A or the temperature outside the manufacturing apparatus 100 as the temperature affecting the production of the film 200 in addition to or instead of the conveyed film. In order to estimate the accurate temperature of the film 200, it is desirable to measure the temperature inside the apparatus 100A rather than the temperature outside the manufacturing apparatus 100.

[0075] Note that by continuously measuring the temperature of the film 200 over time, the variation in the thickness of the film 200 may be estimated based on the temporal change in the temperature of the film 200, and wrinkles and impressions occurring in the wrap film product 130 due to the variation in thickness may be estimated. In this case, the learning device 414 inputs the temperature difference and the variation in thickness into the convolutional neural network 422 in addition to the image and impression of the wrap film product 130 for learning, and inputs the image and the temperature difference into the convolutional neural network 432 at the time of inspection, so that the convolutional neural network 432 can output the variation in thickness.

[0076] Defects occurring when winding the film 200 around the core 231 differ depending on the width dimension of the wrap film product 130. For this reason, it is desirable for the product inspection system 400 to control the temperature according to the width of the wrap film product 130. For example, the product inspection system 400 controls so that the temperature of the film 200, the temperature inside the apparatus 100A, and / or the temperature outside the apparatus 100A becomes higher as the width of the wrap film product 130 is larger.

[0077] According to the product inspection system 400 of the embodiment, the convolutional neural network 422 is trained with the image captured by the imaging device 150 and the external impression of the film 200 in the wrap film product 130 as inspection information. When an image captured by the imaging device 150 is input, the convolutional neural network 422 is machine-learned to output information regarding the external impression of the film in the wrap film product 130 as inspection information. According to the product inspection system 400 of the embodiment, just by imaging the film 200, it is possible to inspect the external impression of the wrap film product 130 without performing inspections such as visual inspection.

[0078] According to the product inspection system 400 of the embodiment, the convolutional neural network 422 is trained with the image of a good product among the images captured by the imaging device 150 and the temperature information corresponding to the good wrap film product 130. When an image captured by the imaging device 150 is input, the convolutional neural network 432 outputs temperature control information based on the temperature information corresponding to the good wrap film product 130. Thereby, according to the product inspection system 400 of the embodiment, just by imaging the film 200, it is possible to approach a temperature at which wrinkles do not occur in the film 200.

[0079] According to the product inspection system 400 of the embodiment, there is provided a product inspection method for inspecting a wrap film product 130 in which a transparent film 200 is conveyed to a core 231 and the conveyed film 200 is wound around the core 231. The method includes: imaging the wrap film product 130; measuring the temperature that affects the production of the film 200; using the film 200 conveyed to the core 231 to machine-learn a convolutional neural network 422 with an appearance image of the wrap film product 130 and temperature information that affects the appearance of the wrap film product 130 as teacher data; inputting the imaged image into the convolutional neural network 432 to obtain inspection information regarding the appearance of the wrap film product 130, and inspecting the appearance of the wrap film product 130 based on the inspection information and the measured temperature. According to the product inspection system 400 of the embodiment, it is possible to inspect the wrap film product 130 using the film 200 in the manufacturing process and suppress the occurrence of defective wrap film products 130.

[0080] According to the embodiment, there is provided a learning data acquisition unit 420 that acquires learning data including a learning image of a wrap film product 130 in which a transparent film 200 is conveyed to a core 231 and the conveyed film 200 is wound around the core 231, and temperature information that affects the appearance of the wrap film product 130, and a learning processing unit 424 that learns a convolutional neural network 422. When the learning image is input into the convolutional neural network 422, the learning processing unit 424 learns the processing parameters of the convolutional neural network 422 so as to output inspection information regarding the appearance of the wrap film product 130 from the convolutional neural network 422. According to the embodiment, it is possible to construct a recognition engine for inspecting the wrap film product 130 using the film 200 in the manufacturing process.

[0081] According to an embodiment, in the manufacturing process of the wrap film product 130 in which the transparent film 200 is conveyed to the core 231 and the conveyed film 200 is wound around the core 231, a captured image acquisition unit 430 that acquires a captured image of the film 200 conveyed to the core 231, a measuring device 140A that measures the temperature affecting the appearance of the wrap film product 130, and the image of the film 200 conveyed to the core 231 captured in the past and the temperature information affecting the appearance of the wrap film product 130 are input into the convolutional neural network 432 that has been machine-learned using the teacher data, and inspection information regarding the appearance of the wrap film product 130 is acquired. Based on the inspection information and the measured temperature, a product inspection device 410 that inspects the appearance of the wrap film product 130 is provided, and a recognition device can be realized. According to the embodiment, the wrap film product 130 can be inspected using the film 200 in the manufacturing process, and the occurrence of defective wrap film products 130 can be suppressed.

[0082] According to an embodiment, there is provided a product manufacturing system (manufacturing apparatus 100) for manufacturing a wrap film product 130 by conveying a transparent film 200 to a core 231 and winding the conveyed film 231 around the core 231. The product manufacturing system includes an imaging device 150 for imaging the wrap film product 130, a measuring device 140A for measuring the temperature that affects the appearance of the wrap film product 130, an air conditioning unit (air cooling fan 300, external air conditioning device 322) for adjusting the temperature that affects the appearance of the wrap film product 130, a product inspection device 410 for obtaining inspection information regarding the appearance of the wrap film product 130 by inputting the imaged appearance image of the wrap film product 130 and the image of a good product among the imaged images and the temperature information corresponding to the good product as teacher data to a convolutional neural network 422 and inspecting the appearance of the wrap film product 130 based on the inspection information and the measured temperature. The convolutional neural network 432 outputs temperature control information based on the temperature information corresponding to a good product when an imaged image is input, and the air conditioning unit adjusts the temperature that affects the appearance of the wrap film product 130 based on the control information, thereby realizing the product manufacturing system. According to the embodiment, for example, by controlling the temperature that affects the appearance of the wrap film product 130 so as to suppress wrinkles in the wrap film product 130, the occurrence of defective wrap film products 130 can be suppressed.

[0083] According to an embodiment, there is provided a product manufacturing method for manufacturing a wrap film product 130 by transporting a transparent film 200 to a core 231 and winding the transported film 200 around the core 231, the method including: imaging the wrap film product 130; measuring a temperature that affects the appearance of the wrap film product 130; adjusting the temperature that affects the appearance of the wrap film product 130; using the film 200 transported to the core 231 to train a convolutional neural network 422 with an appearance image of the wrap film product 130, an image of a good product among the captured images, and temperature information corresponding to the good product as teacher data; inputting the captured image into a convolutional neural network 432 to obtain inspection information regarding the appearance of the wrap film product 130, and inspecting the appearance of the wrap film product 130 based on the inspection information and the measured temperature; and the convolutional neural network 432 outputs temperature control information based on the temperature information corresponding to a good wrap film product 130 when the captured image is input, and adjusts the temperature that affects the appearance of the wrap film product 130 based on the control information. According to the embodiment, for example, by controlling the temperature that affects the appearance of the wrap film product 130 so as to suppress wrinkles in the wrap film product 130, the occurrence of defective wrap film products 130 can be suppressed.

[0084] Note that the program of the product inspection system 400 according to an aspect of the present invention may be a program (a program that causes a computer to function) that controls one or more processors such as a CPU so as to realize the functions shown in the above-described embodiments and modifications related to an aspect of the present invention. Information handled by each of these devices is temporarily stored in a RAM during its processing, and then stored in various storages such as a flash memory, an SSD, or an HDD, and may be read by the CPU and corrected / written as necessary.

[0085] Note that, in each of the above-described embodiments and modified examples, part or all of the learning device or recognition device included in the product inspection system 400 may be realized by one or a plurality of computers equipped with processors. In that case, a program for realizing this control function may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be read into a computer system and executed to realize it. Further, the "computer-readable recording medium" refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, etc., and a storage device such as a hard disk built into a computer system. Furthermore, the "computer-readable recording medium" refers to something that holds a program dynamically for a short time, like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line, and also includes something that holds a program for a certain time, like a volatile memory inside a computer system that serves as a server or a client in that case. Also, the above program may be for realizing a part of the aforementioned functions, and may further be realizable in combination with a program already recorded in the computer system for realizing the aforementioned functions.

[0086] Further, in each of the above-described embodiments and modified examples, part or all of the learning device or recognition device included in the product inspection system 400 may typically be realized as an LSI, which is an integrated circuit, or as a chip set. Also, the method of integrating into an integrated circuit is not limited to an LSI and may be realized by a dedicated circuit and / or a general-purpose processor. Also, when a technology for integrating into an integrated circuit that replaces an LSI appears due to the progress of semiconductor technology, it is also possible to use an integrated circuit based on that technology.

[0087] As described above in detail with reference to the drawings for each embodiment and modification as one aspect of the present invention, the specific configuration is not limited to each embodiment and modification, and also includes design changes and the like within the scope not departing from the gist of the present invention. Also, one aspect of the present invention can be variously modified within the scope shown in the claims, and embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention. Further, a configuration in which elements described in the above embodiments and modifications and having the same effects are substituted for each other is also included.

[0088] For example, one aspect of the present invention may be realized by combining some or all of the above embodiments.

[0089] As described above, the embodiments for carrying out the present invention have been described using the embodiments, but the present invention is not limited to such embodiments, and various modifications and substitutions can be made without departing from the gist of the present invention.

Explanation of Reference Numerals

[0090] 100 Manufacturing apparatus, 110 Conveyor unit, 120 Raw web, 122 Film, 124 First cylindrical member, 130 Wrap film product, 132 Film, 134 Second cylindrical member, 140 Control device, 140A Measuring device, 142 Control unit, 144 Notification unit, 150 Imaging device, 160 Lighting device, 200 Film, 210 Feeding mechanism, 211 Feeding position, 212 Standby position, 213 Turret arm, 220 Roller mechanism, 221 Pay-out roller, 222 Guide roller, 222A Media passage, 223 Expander roller, 224 Pinch roller, 230 Take-up drum, 231 Core, 231 Film, 300 Air-cooling fan, 310 In-device temperature sensor, 320 Recognition device, 320 Out-of-device temperature sensor, 322 Out-of-device air conditioner, 323 Neural network, 400 Product inspection system, 410 Product inspection device, 412 Recognition device, 412 Recognition processing unit, 414 Learning device, 416 Learning image database, 420 Learning data acquisition unit, 422 Neural network, 424 Learning processing unit, 426 Learning result storage unit, 430 Captured image acquisition unit, 432 Neural network, 434 Recognition processing unit, 436 Recognition result storage unit, 500 Terminal device

Claims

1. A product inspection system for inspecting a product in which a transparent film is conveyed to a cylindrical member and the conveyed film is wound around the cylindrical member, comprising: an imaging unit that images the product; a measuring unit that measures the temperature near the cylindrical member; a prediction model is machine-learned using, as teacher data, an appearance image of the appearance including wrinkles of the product, the temperature near the cylindrical member, and information regarding the impression of the appearance including wrinkles of the product, an image captured by the imaging unit is input to the prediction model, and information regarding the impression of the appearance including wrinkles of the product output from the prediction model and the temperature near the cylindrical member output from the prediction model are stored; and an inspection unit, The product inspection system, wherein the information regarding the impression of the product is inspection result indicating good or bad by visual inspection of a manufacturing worker, questionnaire result of an impression when a customer first opens the product, or information based on past complaints.

2. The product inspection system according to claim 1, wherein the film contains a thermoplastic resin.

3. A product inspection method for inspecting a product in which a transparent film is conveyed to a cylindrical member and the conveyed film is wound around the cylindrical member, comprising: imaging the product; measuring the temperature near the cylindrical member; machine-learning a prediction model using, as teacher data, an appearance image of the appearance including wrinkles of the product, the temperature near the cylindrical member, and information regarding the impression of the appearance including wrinkles of the product; inputting the captured image into the prediction model, and storing information regarding the impression of the appearance including wrinkles of the product output from the prediction model and the temperature near the cylindrical member output from the prediction model; and The product inspection method, wherein the information regarding the impression of the product is inspection result indicating good or bad by visual inspection of a manufacturing worker, questionnaire result of an impression when a customer first opens the product, or information based on past complaints.

4. A learning data acquisition unit that acquires learning data including an appearance image of the appearance including wrinkles of a product in which a transparent film is conveyed to a cylindrical member and the conveyed film is wound around the cylindrical member, the temperature near the cylindrical member, and information regarding the impression of the appearance including wrinkles of the product A learning processing unit that learns a prediction model, which inputs an image captured by an imaging unit into the prediction model and learns the processing parameters of the prediction model so as to output information regarding the appearance impression including wrinkles of the product and the temperature near the cylindrical member. The learning device, wherein the information regarding the impression of the product is inspection result indicating good or bad by visual inspection of a manufacturing worker, questionnaire result of the impression when a customer first opens the product, or information based on past complaints.

5. In a manufacturing process of a product in which a transparent film is conveyed to a cylindrical member and the conveyed film is wound around the cylindrical member, an imaging image acquisition unit that acquires a captured image of the film conveyed to the cylindrical member. A measurement unit that measures the temperature near the cylindrical member. An inspection unit that inputs the image captured by the imaging image acquisition unit into a prediction model machine-learned using, as teacher data, an appearance image of the appearance including wrinkles of the product, the temperature near the cylindrical member, and information regarding the impression of the appearance including wrinkles of the product, and stores the information regarding the impression of the appearance including wrinkles of the product output from the prediction model and the temperature near the cylindrical member output from the prediction model. The recognition device, wherein the information regarding the impression of the product is inspection result indicating good or bad by visual inspection of a manufacturing worker, questionnaire result of the impression when a customer first opens the product, or information based on past complaints.

Images