Filament abnormality detector and abnormality detection program

Abstract

To provide a technique to detect a yarn route abnormality of filaments with respect to a guide.SOLUTION: A filament abnormality detector detects abnormality of a spun yarn take-up device for taking up a plurality of filaments spun and discharged from a spinning device. The spun yarn take-up device has a guide for guiding a plurality of filaments sent from upstream side to downstream side so as to come closer to each other. The filament abnormality detector has a control device. The control device executes the processing for: acquiring a filament image for estimation that captures a plurality of filaments from a camera disposed so as to include the plurality of filaments passing through the guide at least within an imagine range; determining whether or not a yarn route abnormality occurs based on an estimation model subjected to machine learning in order to detect the yarn route abnormality caused by deviation of at least one of the plurality of filaments from the guide and the filament image for estimation; and outputting determination result of the processing to determine.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present disclosure relates to a filament abnormality detection device and an abnormality detection program.

Background Art

[0002] Japanese Patent Application Laid-Open No. 2023-128942 (Patent Document 1) discloses "a detection system capable of detecting yarn vibration of a yarn-like body with high accuracy in a spinning process in yarn manufacturing."

[0003] The detection system causes an imaging unit to image a yarn-like body extruded from a spinneret, and acquires a plurality of input images that capture the yarn-like body. Then, the detection system calculates the degree of variation in gradation values for pixels at the same position among the plurality of input images, and detects the yarn vibration of the yarn-like body based on the degree of variation.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] A spinning take-up device takes up a plurality of filaments spun from a spinning device, and manufactures a yarn by twisting the plurality of filaments together. The plurality of filaments are drawn toward each other by a guide in the process of being sent through the spinning take-up device. At this time, at least one of the plurality of filaments may come off the guide. Whether such a yarn path abnormality has occurred is often confirmed visually by an operator. However, the filaments are very thin, and it is very difficult to confirm them visually.

[0006] In view of the above points, a technique for detecting a yarn path abnormality of a filament with respect to a guide is desired.

[0007] Note that the detection system disclosed in Patent Document 1 is aimed at detecting the filament sway of the filament, and is not aimed at detecting the filament path abnormality of the filament with respect to the guide.

Means for Solving the Problems

[0008] In an example of the present disclosure, a filament abnormality detection device capable of detecting an abnormality with respect to a spinning take-up device that takes up a plurality of filaments spun from a spinning device is provided. The spinning take-up device includes a guide for guiding the plurality of filaments sent from the upstream side to the downstream side so as to bring them closer to each other. The filament abnormality detection device includes a control device. The control device executes a process of acquiring an estimated filament image that captures the plurality of filaments from a camera arranged so as to include at least the imaging range of the plurality of filaments passing through the guide, an estimation model learned by machine learning for detecting a path abnormality in which at least one of the plurality of filaments has deviated from the guide, a process of determining whether or not the path abnormality has occurred based on the estimated filament image and the estimation model, and a process of outputting a determination result in the determination process.

[0009] In the filament abnormality detection device, an estimation model learned by machine learning for detecting a path abnormality in which the filament has deviated from the guide is used. The filament abnormality detection device can detect a path abnormality by inputting the estimated filament image to the estimation model.

[0010] In an example of the present disclosure, the estimation model is an autoencoder learned to restore the filament image after compressing a normal filament image in which no path abnormality has occurred. In the determination process, it is determined whether or not the path abnormality has occurred based on the similarity between the estimated filament image and the restored filament image obtained by inputting the estimated filament image to the autoencoder.

[0011] In the above filament abnormality detection device, an autoencoder generated from a normal filament image is used. That is, abnormal filament images are not required during machine learning.

[0012] In an example of the present disclosure, in the above determination process, based on the comparison result between the above similarity and a predetermined threshold value, it is determined whether or not the above thread path abnormality has occurred.

[0013] Thereby, the tension abnormality detection device can detect a thread path abnormality according to the threshold value.

[0014] In an example of the present disclosure, the above estimation model is generated by machine learning a plurality of learning data. Each of the plurality of learning data associates a label indicating whether or not the above thread path abnormality has occurred with a learning filament image that captures a plurality of filaments passing through a guide. In the above determination process, based on the output result obtained from the above estimation model by inputting the above estimation filament image into the above estimation model, it is determined whether or not the above thread path abnormality has occurred.

[0015] In the above filament abnormality detection device, an estimation model learned using an abnormal filament image in which a thread path abnormality has occurred is used. Thereby, the detection accuracy of the thread path abnormality is improved.

[0016] In an example of the present disclosure, the above determination process includes a process of performing preprocessing on the above estimation filament image and a process of inputting the estimation filament image after the above preprocessing into the above estimation model. The above preprocessing includes a process of extracting edges from the above estimation filament image.

[0017] Thereby, the filament abnormality detection device can more accurately capture the filaments shown in the estimation filament image. As a result, the detection accuracy of the thread path abnormality is improved.

[0018] In an example of the present disclosure, the spinning take-up device further includes a light source arranged to include at least the plurality of filaments passing upstream of the guide within an illumination range.

[0019] As a result, the luminance difference between the filaments and the background becomes clear, and the filament abnormality detection device can more accurately capture the filaments appearing in the estimated filament image. Consequently, the detection accuracy of the thread path abnormality is improved.

[0020] In an example of the present disclosure, the guide includes a main body having a surface with which the plurality of filaments sent in the gravitational direction from the spinning device come into contact, a discharge port formed on the surface for discharging an oil agent, and two thread guiding members provided on the surface so as to be located on both sides of the discharge port in the horizontal direction, for guiding the plurality of filaments sent in the gravitational direction toward the center side of the discharge port in the horizontal direction. The camera is arranged such that the imaging range of the camera includes at least the plurality of filaments passing upstream of the surface. The light source is arranged such that the illumination range of the light source includes at least the plurality of filaments passing upstream of the surface.

[0021] As a result, the filament abnormality detection device can detect a thread path abnormality in a guide having an oil supply function. Also, since the imaging range of the camera includes the upstream side of the surface and the illumination range of the light source includes the upstream side of the surface, the filament abnormality detection device can more reliably capture the thread path abnormality of the filaments. Consequently, the detection accuracy of the thread path abnormality is improved.

[0022] In an example of the present disclosure, the optical axis of the light source is inclined with respect to the optical axis of the camera.

[0023] As a result, the amount of light reflected by the camera can be reduced. Consequently, the occurrence of white spots in the filament image can be prevented.

[0024] In one example of the present disclosure, the light source is provided on the front side of the guide in the direction of viewing the guide from the camera.

[0025] As a result, the amount of light directly incident on the camera from the light source is reduced. As a result, the occurrence of white spots in the filament image can be prevented.

[0026] In one example of the present disclosure, the spinning take-up device further includes an anti-reflection member for preventing reflection of light irradiated from the light source. The anti-reflection member is provided on the back side of the guide in the direction of viewing the guide from the camera.

[0027] As a result, the amount of light reflected by the camera can be further reduced. As a result, the occurrence of white spots in the filament image can be more reliably prevented.

[0028] In one example of the present disclosure, the anti-reflection member is black.

[0029] As a result, the amount of light reflected by the camera can be further reduced. As a result, the occurrence of white spots in the filament image can be more reliably prevented.

[0030] In one example of the present disclosure, the light source is spot illumination. The illumination range of the spot illumination intersects the optical axis of the camera.

[0031] As a result, it is possible to suppress light from irradiating objects other than the filament. Therefore, the amount of light reflected by the camera can be further reduced. As a result, the occurrence of white spots in the filament image can be more reliably prevented.

[0032] In another example of the present disclosure, there is provided an abnormality detection program executed by a filament information processing device communicable with a spinning take-up device that takes up a plurality of filaments spun from a spinning device. The spinning take-up device includes a guide for guiding the plurality of filaments so as to bring them closer to each other. The abnormality detection program causes the filament information processing device to acquire an estimated filament image capturing the plurality of filaments from a camera arranged so as to include at least the plurality of filaments passing through the guide within a shooting range, input the estimated filament image to an estimation model learned by machine learning to estimate a yarn path abnormality in which at least one of the plurality of filaments has deviated from the guide, determine whether or not the yarn path abnormality has occurred based on the output result of the estimation model, and output the determination result in the determining step.

[0033] In the abnormality detection program, an estimation model learned by machine learning is used to detect a yarn path abnormality in which a filament has deviated from the guide. The abnormality detection program can detect a yarn path abnormality by inputting the estimated filament image to the estimation model.

[0034] In an example of the present disclosure, the estimation model is an autoencoder learned by machine learning to restore a filament image after compressing a normal filament image in which no yarn path abnormality has occurred. In the determining step, it is determined whether or not the yarn path abnormality has occurred based on the degree of similarity between the estimated filament image and the restored filament image obtained by inputting the estimated filament image to the autoencoder.

[0035] In the abnormality detection program, an autoencoder generated from a normal filament image is used. That is, abnormal filament images are not required during machine learning.

[0036] In an example of the present disclosure, the above determination step includes a step of performing preprocessing on the filament image for estimation, and a step of inputting the filament image for estimation after the preprocessing into the estimation model. The preprocessing includes a process of extracting edges from the filament image for estimation.

[0037] As a result, the filament abnormality detection device can more accurately capture the filaments shown in the filament image for estimation. As a result, the detection accuracy of the thread path abnormality is improved.

[0038] The above and other objects, features, aspects and advantages of the present invention will become apparent from the following detailed description of the present invention understood in connection with the accompanying drawings.

Brief Description of the Drawings

[0039]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

Figure 21

Embodiments for Carrying Out the Invention

[0040] Hereinafter, each embodiment according to the present invention will be described with reference to the drawings. In the following description, the same parts and components are denoted by the same reference numerals. Their names and functions are also the same. Therefore, detailed descriptions thereof will not be repeated. In addition, each embodiment and each modification example described below may be selectively combined as appropriate.

[0041] <A. Spinning and Take-up Device 1> First, with reference to FIG. 1, the spinning and take-up device 1, which is a manufacturing device for drawn yarn, will be described. FIG. 1 is a schematic diagram showing an example of the device configuration of the spinning and take-up device 1.

[0042] As shown in Fig. 1, the spinning and winding device 1 respectively draws out a synthetic fiber yarn Y composed of a plurality of filaments F spun from the spinning device 2, and winds them around a plurality of bobbins B to form a plurality of packages P. Hereinafter, the vertical direction, the front-rear direction, and the left-right direction shown in Fig. 1 will be defined as the vertical direction, the front-rear direction, and the left-right direction of the spinning and winding device 1, respectively, for the purpose of explanation.

[0043] Also, hereinafter, the direction in which a plurality of filaments F are sent in the spinning and winding device 1 will be defined as the downstream side, and the direction opposite to the direction in which a plurality of filaments F are sent in the spinning and winding device 1 will be defined as the upstream side for the purpose of explanation.

[0044] The spinning and winding device 1 includes a cooling section 3, an oil supply section 4, a stretching section 5, take-up rollers 6 and 7, an interlacing device 8, a winding device 9, etc. First, in the spinning device 2, a polymer supplied from a polymer supply device (not shown) composed of a gear pump or the like is pushed downward from a plurality of nozzles 2a arranged in the left-right direction (the depth direction of the paper surface in Fig. 1), and each group of filaments F is spun in a state of being arranged in the left-right direction.

[0045] Thereafter, each group of filaments F is sent to the cooling section 3 and the oil supply section 4. The filaments F are gathered into one yarn Y for each group in the oil supply section 4. Thereafter, the plurality of yarns Y travel along a yarn path along the stretching section 5, the take-up roller 6, the interlacing device 8, and the take-up roller 7 in a state of being arranged in the left-right direction. Further, the plurality of yarns Y are distributed in the front-rear direction from the take-up roller 7 and then wound around a plurality of bobbins B in the winding device 9, respectively.

[0046] The cooling unit 3 has a plurality of cylindrical cooling cylinders 10, and each cooling cylinder 10 is disposed below a plurality of spinnerets 2a provided in the spinning device 2. A plurality of filaments F spun from the spinnerets 2a of the spinning device 2 travel through the internal space 10a of each cooling cylinder 10 from above to below along the axial direction of the cooling cylinder 10. A rectifying portion 10b is provided around the internal space 10a, and cooling air supplied from a compressed air supply device (not shown) flows into the internal space 10a while being rectified by the rectifying portion 10b. Mainly, the rectifying portion 10b rectifies the cooling air flowing into the internal space 10a so that the flow rate of the cooling air is substantially uniform in the circumferential direction of the cooling cylinder 10.

[0047] The oil supply unit 4 has a plurality of oil supply guides 11 disposed below each cooling cylinder 10. The oil supply guide 11 bundles a plurality of filaments F spun from the spinneret 2a into one yarn Y and imparts an oil agent to the yarn Y (a plurality of filaments F). The oil supply guide 11 will be described in detail later.

[0048] Also, a camera 30 is provided inside the spinning take-up device 1. The camera 30 is disposed so as to include at least a plurality of filaments F passing through the upstream side of the oil supply guide 11 within the imaging range 30R. That is, the camera 30 may be disposed so as to include the oil supply guide 11 within the imaging range 30R, or may be disposed so as not to include the oil supply guide 11 within the imaging range 30R.

[0049] Also, a light source 40 is provided inside the spinning take-up device 1. The light source 40 is disposed so as to include at least a plurality of filaments F passing through the upstream side of the oil supply guide 11 within the illumination range 40R. That is, the light source 40 may be disposed so as to include the oil supply guide 11 within the illumination range 40R, or may be disposed so as not to include the oil supply guide 11 within the illumination range 40R. The illumination range 40R overlaps with the imaging range 30R at least in part.

[0050] The extension part 5 is arranged below the oil supply part 4. The extension part 5 has a heat preservation box 12 and a plurality of heating rollers (not shown) accommodated in the heat preservation box 12. The extension part 5 extends while heating the plurality of yarns Y respectively by the plurality of heating rollers.

[0051] The plurality of yarns Y extended by the extension part 5 are sent to the winding device 9 by the take-up rollers 6 and 7. The interlacing device 8 is arranged between the take-up roller 6 and the take-up roller 7, and interlaces a plurality of filaments F constituting one yarn Y to impart interlacing.

[0052] The winding device 9 includes a machine base 13, a turret 14, two bobbin holders 15, a support frame 16, a contact roller 17, a traversing device 18, etc. The winding device 9 forms a plurality of packages P by simultaneously winding the plurality of yarns Y sent from the take-up roller 7 around the plurality of bobbins B by rotating the bobbin holders 15.

[0053] The turret 14 is a disc-shaped member and is attached to the machine base 13. The turret 14 is rotationally driven by a motor (not shown). The two bobbin holders 15 are cantilever-supported on the turret 14 in a posture extending in the front-rear direction. A plurality of cylindrical bobbins B are mounted on each bobbin holder 15 in a state of being arranged along its axial direction. By rotating the turret 14, the two bobbin holders 15 can be switched between an upper winding position and a lower retracted position.

[0054] The support frame 16 is a long frame-shaped member extending in the front-rear direction. This support frame 16 is fixed to the machine base 13. A roller support member 19 that is long in the front-rear direction is attached to the lower part of the support frame 16 so as to be vertically movable with respect to the support frame 16. A contact roller 17 extending along the axial direction of the bobbin holder 15 is rotatably supported on the roller support member 19. By bringing this contact roller 17 into contact with the package P being formed and applying a predetermined contact pressure to the package P, the shape of the package P is adjusted.

[0055] The traverse device 18 has a plurality of traverse guides 18a arranged in the front-rear direction. The plurality of traverse guides 18a are driven by a motor (not shown) and reciprocate in the front-rear direction respectively. When the traverse guide 18a reciprocates with the yarn Y hung thereon, the yarn Y is wound around the corresponding bobbin B while being twill-woven back and forth about the fulcrum guide 18b.

[0056] <B. Oil supply guide 11> Next, with reference to FIGS. 2 and 3, the oil supply guide 11 shown in FIG. 1 will be described. FIG. 2 is a view showing the oil supply guide 11 from the front direction. FIG. 3 is a cross-sectional view taken along line III-III shown in FIG. 2.

[0057] As described above, the oil supply guide 11 applies an oil agent to the yarn Y composed of a large number of filaments F spun from the spinning device 2. The oil supply guide 11 is formed of a ceramic material such as alumina or zirconia and has a guide body 20 as shown in FIGS. 2 and 3. The front surface 21 of the guide body 20 extends along the vertical direction. And the yarn Y (a plurality of filaments F) traveling in the direction from above to below (that is, the gravitational direction), which is sent from the cooling unit 3, contacts the surface 21. The above-described camera 30 is arranged such that its imaging range 30R includes a plurality of filaments F passing through the upstream side of the surface 21. Also, the above-described light source 40 is arranged such that its illumination range 40R includes a plurality of filaments F passing through the upstream side of the surface 21.

[0058] Further, the guide body 20 has an oil agent flow path 22. The oil agent flow path 22 is formed inside the oil supply guide 11 and extends in the front-rear direction. The front end of the oil agent flow path 22 is a discharge port 25 formed on the surface 21, and by discharging the oil agent from the discharge port 25, the oil agent is applied to the yarn Y (a plurality of filaments F). The concentration of the oil agent discharged from the discharge port 25 is, for example, about 85%. The concentration of the oil agent refers to the concentration including all active ingredients such as oil components and additives other than water.

[0059] Here, the surface 21 of the guide body 20 has an upper curved surface 26 above the upper end 25a of the discharge port 25 and a lower curved surface 27 below the lower end 25b of the discharge port 25. Both the curved surfaces 26 and 27 are curved so as to be convex to the outside of the guide body 20.

[0060] Also, when viewed from the left - right direction (in the cross - section of FIG. 3), the upper end 25a of the discharge port 25 and the upper portion 20a of the guide body 20, which is above the upper end 25a of the discharge port 25, do not overlap with the tangent line L1 of the lower curved surface 27 at the position of the lower end 25b of the discharge port 25. Further, when viewed from the left - right direction, the upper end 25a of the discharge port 25 and the upper portion 20a of the guide body 20 also do not overlap with the straight line L2 obtained by rotating the tangent line L1 10° in the clockwise direction (the direction approaching the upper end 25a of the discharge port 25) around the lower end 25b of the discharge port 25.

[0061] The length K between the upper end 25a and the lower end 25b of the discharge port 25 is, for example, about 0.1 [mm] in the direction orthogonal to the tangent line L1, so that the upper end 25a and the lower end 25b of the discharge port 25 have the positional relationship as described above.

[0062] Also, the oil supply guide 11 is arranged such that the tangent line L1 is substantially parallel to the traveling direction of the yarn Y (filament F) sent from the cooling part 3 when viewed from the left - right direction.

[0063] Further, the oil supply guide 11 is configured to guide a plurality of filaments F sent from the upstream side to the downstream side so as to bring them closer to each other. More specifically, two yarn guiding members 23 are arranged on the surface 21 of the guide body 20. The two yarn guiding members 23 are respectively arranged on the right side portion of the surface 21 relative to the discharge port 25 and the left side portion of the surface 21 relative to the discharge port 25. That is, the two yarn guiding members 23 are arranged on the surface 21 so as to be located on both sides of the discharge port 25 in the left-right direction. Further, the two yarn guiding members 23 are inclined and extend in the vertical direction so as to approach the central portion of the discharge port 25 in the left-right direction as they go from above to below. Thereby, the interval between the two yarn guiding members 23 in the left-right direction becomes smaller as they go from above to below. And the plurality of filaments F sent from the cooling unit 3 are gradually converged and gathered into one yarn Y by being guided in a direction approaching the central side of the discharge port 25 in the left-right direction by the two yarn guiding members 23 while passing through the oil supply guide 11.

[0064] <C. Outline> As described above, the plurality of filaments F are attracted to each other by the oil supply guide 11 in the process of being sent inside the spinning take-up device 1. At this time, at least one of the plurality of filaments F may come off the oil supply guide 11. Hereinafter, at least one of the plurality of filaments F coming off the normal yarn path on the oil supply guide 11 is also referred to as "yarn path abnormality".

[0065] FIG. 4 is a diagram showing a specific example of the yarn path abnormality of the filament F. In the example of FIG. 4(A), the filament FA has come off the yarn guiding member 23 of the oil supply guide 11. In the example of FIG. 4(B), the filament FB is broken for some reason and has come off the yarn guiding member 23 of the oil supply guide 11. As another example, there is also a yarn path abnormality in which the filament deviates from the oil supply guide through which it should originally pass and enters the adjacent oil supply guide by mistake.

[0066] When the thread path abnormality as described above occurs, the quality of the produced thread Y deteriorates. Whether or not a thread path abnormality has occurred is often confirmed visually by an operator. However, the filament F is very thin, and it is very difficult to confirm it visually. Therefore, the inventors devised a function for automatically detecting the thread path abnormality of the filament F.

[0067] Referring to FIG. 5, the outline of the thread path abnormality detection function will be described. FIG. 5 is a diagram showing the main configuration of a filament abnormality detection device 50 according to an embodiment.

[0068] As shown in FIG. 5, the filament abnormality detection device 50 includes the above-described camera 30 and an information processing device 100. The camera 30 of the filament abnormality detection device 50 is disposed inside the spinning take-up device 1 so as to include at least a plurality of filaments F passing through the oil supply guide 11 of the spinning take-up device 1 within the imaging range. The thread path abnormality detection function is implemented, for example, in the information processing device 100. The information processing device 100 is configured to be communicable with the spinning take-up device 1. The information processing device 100 may be disposed inside the spinning take-up device 1 or may be disposed outside the spinning take-up device 1.

[0069] The information processing device 100 is, for example, a control unit of the spinning take-up device 1. The control unit controls various drive devices (for example, the winding device 9, etc.) provided in the spinning take-up device 1. As another example, the spinning take-up device 1 may be a server configured to be communicable with the spinning take-up device 1.

[0070] The information processing apparatus 100 includes a control device 101. The control device 101 is constituted by, for example, at least one integrated circuit. The integrated circuit can be constituted by, for example, at least one CPU (Central Processing Unit), at least one GPU (Graphics Processing Unit), at least one ASIC (Application Specific Integrated Circuit), at least one FPGA (Field Programmable Gate Array), or a combination thereof.

[0071] First, the control device 101 sends an imaging instruction to the camera 30 to obtain an image (hereinafter also referred to as a "filament image") that captures a plurality of filaments F. Further, the control device 101 obtains an estimation model 124. The estimation model 124 has been pre-trained by machine learning so as to be able to detect a thread path abnormality. The learning process for generating the estimation model 124 will be described later. Then, the control device 101 determines whether a thread path abnormality has occurred based on the estimation model 124 and the filament image for estimation obtained from the camera 30, and outputs the determination result.

[0072] As described above, the information processing apparatus 100 can detect a thread path abnormality by using the estimation model 124 pre-trained by machine learning to detect a thread path abnormality in which the filament F has come off the oil supply guide 11.

[0073] <D. Hardware Configuration of Information Processing Apparatus 100> Next, with reference to FIG. 6, the hardware configuration of the information processing apparatus 100 shown in FIG. 5 will be described. FIG. 6 is a diagram showing an example of the hardware configuration of the information processing apparatus 100.

[0074] The information processing apparatus 100 includes the above-described control device 101, a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, a communication interface 104, a display interface 105, an input interface 107, and an auxiliary storage device 120. These components are connected to a bus 115.

[0075] The control device 101 controls the operation of the information processing apparatus 100 by executing various programs such as a learning program 126 and an abnormality detection program 128. Based on receiving an execution instruction for various programs, the control device 101 reads out the program to be executed from the auxiliary storage device 120 or the ROM 102 to the RAM 103. The RAM 103 functions as a working memory and temporarily stores various data necessary for program execution.

[0076] The communication interface 104 is an interface for the information processing apparatus 100 to communicate with external devices. The information processing apparatus 100 exchanges data with external devices via the communication interface 104. The external devices include, for example, the above-described camera 30, the above-described light source 40, and the like.

[0077] A display 106 is connected to the display interface 105. The display interface 105 sends an image signal for displaying an image to the display 106 according to a command from the control device 101 or the like. The display 106 is, for example, a liquid crystal display, an organic EL (Electro Luminescence) display, or other displays. Note that the display 106 may be integrally configured with the information processing apparatus 100 or may be configured separately from the information processing apparatus 100.

[0078] An input device 108 is connected to the input interface 107. The input device 108 is, for example, a mouse, a keyboard, a touch panel, or any other device capable of receiving user operations. Note that the input device 108 may be integrally configured with the information processing apparatus 100 or may be configured separately from the information processing apparatus 100.

[0079] The auxiliary storage device 120 is a storage medium such as a hard disk, a flash memory, and an SSD (Solid State Drive), for example. The auxiliary storage device 120 stores, for example, the learning dataset 122, the above-described estimation model 124, the learning program 126, and the anomaly detection program 128. These storage locations are not limited to the auxiliary storage device 120 and may be stored in the storage area of the control device 101 (such as a cache memory), the ROM 102, the RAM 103, an external device (such as a server), or the like.

[0080] The learning program 126 is a program for generating the estimation model 124 using the learning dataset 122. The learning program 126 may be provided not as a single program but incorporated into a part of any program. In this case, the learning process by the learning program 126 is realized in cooperation with any program. Even a program that does not include such a part of the module does not deviate from the gist of the learning program 126 according to the present embodiment. Further, part or all of the functions provided by the learning program 126 may be realized by dedicated hardware. Further, the information processing apparatus 100 may be configured in a form such as a so-called cloud service in which at least one server executes a part of the processing of the learning program 126.

[0081] The abnormality detection program 128 is a program for detecting a thread path abnormality of the filament F using the estimation model 124. The abnormality detection program 128 may be provided incorporated into a part of an arbitrary program instead of as a single program. In this case, the estimation process by the abnormality detection program 128 is realized in cooperation with an arbitrary program. Even a program that does not include such a part of the module does not deviate from the gist of the abnormality detection program 128 according to the present embodiment. Further, part or all of the functions provided by the abnormality detection program 128 may be realized by dedicated hardware. Further, the information processing apparatus 100 may be configured in a form such as a so-called cloud service in which at least one server executes a part of the processing of the abnormality detection program 128.

[0082] <E. Training dataset 122> Next, with reference to FIG. 7, the training dataset 122 used when generating the estimation model 124 shown in FIG. 6 will be described. FIG. 7 is a diagram showing an example of the training dataset 122.

[0083] The training dataset 122 includes a plurality of training data 123. The number of training data 123 included in the training dataset 122 is arbitrary. As an example, the number of training data 123 is several tens to several hundreds of thousands.

[0084] In each of the training data 123, an image name and a filament image are associated. The image name is an identifier for uniquely identifying the filament image.

[0085] In the present embodiment, the training dataset 122 is composed of normal filament images in which the filament F is not detached from the oil supply guide 11. The normal filament image represents a state in which the filament F is running inside two thread guide members 23 (see FIG. 2).

[0086] The learning dataset 122 may include the filament images acquired from the camera 30 in the above-described spinning take-up device 1, or may include the filament images acquired from the cameras in other spinning take-up devices 1.

[0087] <Functional configuration of the information processing device 100> Next, with reference to FIGS. 8 to 12, the functional configuration of the information processing device 100 will be described. FIG. 8 is a diagram showing an example of the functional configuration of the information processing device 100.

[0088] The information processing device 100 functions as a preprocessing unit 151, a learning unit 152, a preprocessing unit 153, a determination unit 154, and an output unit 155 by executing the above-described learning program 126 or the above-described anomaly detection program 128. As an example, the functions of the preprocessing unit 151 and the learning unit 152 are implemented in the learning program 126, and the functions of the preprocessing unit 153, the determination unit 154, and the output unit 155 are implemented in the above-described anomaly detection program 128.

[0089] Hereinafter, the preprocessing unit 151, the learning unit 152, the preprocessing unit 153, the determination unit 154, and the output unit 155 will be described in order.

[0090] (F1. Preprocessing unit 151) First, with reference to FIGS. 9 and 10, the functions of the preprocessing unit 151 shown in FIG. 8 will be described. FIG. 9 is a diagram for explaining an example of the preprocessing by the preprocessing unit 151. FIG. 10 is a diagram for explaining another example of the preprocessing by the preprocessing unit 151.

[0091] The preprocessing unit 151 performs preprocessing on the filament images defined in the above-described learning data 123 (see FIG. 7). As a result, the filament images are processed into a form suitable for detecting thread path anomalies.

[0092] In the example shown in FIG. 9, the preprocessing unit 151 executes preprocessing for extracting edges from the filament image IM1 for learning. More specifically, the preprocessing unit 151 applies an edge filter (i.e., a differential filter) to the filament image IM1 for learning, and generates a filament image IM2 in which edge portions are emphasized.

[0093] The background portion in the filament image IM1 is more blurred than the fuel supply guide 11 and the filament F. Therefore, the background portion disappears by the edge extraction process, and the contour of the fuel supply guide 11 and the contour of the filament F remain in the filament image IM2. As a result, information unnecessary for detecting the thread path abnormality disappears. As a result, the detection accuracy of the thread path abnormality is improved.

[0094] Preferably, the preprocessing unit 151 adds image information AD1 and AD2 to the filament image IM2 after edge extraction. In the example of FIG. 9, stripe pattern image information AD1 is added to the upper part of the filament image IM2. On the other hand, stripe pattern image information AD2 is added to the lower part of the filament image IM2. By adding the image information AD1 and AD2 to the filament image IM2, it is possible to avoid the learning of the learning unit 152 described later from ending early.

[0095] In the example shown in FIG. 10, the preprocessing unit 151 executes preprocessing for removing the portion where the fuel supply guide 11 appears from the filament image IM1 for learning. The method for removing the portion where the fuel supply guide 11 appears is not particularly limited.

[0096] In one aspect, the preprocessing unit 151 searches for an image region R2 where the fuel supply guide 11 appears by searching for the fuel supply guide 11 from the filament image IM1, and removes the image region R2 from the filament image IM1.

[0097] For the search process of the fuel supply guide 11, various existing image processes are used. As an example, the fuel supply guide 11 in the filament image IM1 is recognized using a learned model. The learned model is generated in advance by a learning process using a learning dataset. The learning dataset includes a plurality of learning images in which the fuel supply guide 11 is shown. Each learning image is associated with a label indicating whether the fuel supply guide 11 is shown or not. The internal parameters of the learned model are optimized in advance by a learning process using such a learning dataset.

[0098] For the learning method for generating the learned model, various machine learning algorithms can be adopted. As an example, as the machine learning algorithm, deep learning, convolutional neural network (CNN), fully convolutional neural network (FCN), support vector machine, etc. are adopted.

[0099] In another aspect, the preprocessing unit 151 specifies a predetermined range in the filament image IM1 as an image region R2 in which the fuel supply guide 11 is shown, and removes the image region R2 from the filament image IM1. Typically, the predetermined range is the lower region in the filament image IM1.

[0100] Thereafter, the preprocessing unit 151 removes the image region R2 from the learning filament image IM1, and generates the remaining image region R1 as the learning filament image IM3. As shown in FIG. 10, in the learning filament image IM3, only the filament F passing through the upstream side of the fuel supply guide 11 is shown, and the fuel supply guide 11 is not shown.

[0101] By using the filament image IM3 from which the oil supply guide 11 has been removed in the learning process described later, the information processing apparatus 100 can detect thread path abnormalities without being affected by the type of the oil supply guide 11. As a result, the detection accuracy of thread path abnormalities is improved. In addition, the designer does not need to collect filament images for learning for all types of oil supply guides 11, and the thread path abnormality detection function can be more easily realized.

[0102] Preferably, the preprocessing unit 151 adds image information AD3 and AD4 to the filament image IM3 after removing the image area R2. In the example of FIG. 10, stripe pattern image information AD3 is added to the upper part of the filament image IM3. On the other hand, stripe pattern image information AD4 is added to the lower part of the filament image IM3. By adding the image information AD3 and AD4 to the filament image IM3, it is possible to avoid the learning of the learning unit 152 described later from ending prematurely.

[0103] Note that the preprocessing by the preprocessing unit 151 does not necessarily have to be executed. For example, when the above-described camera 30 is arranged so as not to include the oil supply guide 11 in the shooting range, only a plurality of filaments F passing through the upstream side of the oil supply guide 11 are captured in the filament image IM for learning. In this case, the preprocessing shown in FIG. 10 does not need to be executed.

[0104] In addition, the preprocessing unit 151 may execute both the preprocessing shown in FIG. 9 and the preprocessing shown in FIG. 10, or may execute only one of the preprocessings.

[0105] Further, the preprocessing unit 151 may perform preprocessing different from the preprocessing shown in FIGS. 9 and 10. As an example, the preprocessing unit 151 may perform a process of cutting out a portion in which the fuel supply guide 11 and the filament F are shown in the filament image IM1. In other words, the preprocessing unit 151 may perform a process of removing portions other than the portions in which the fuel supply guide 11 and the filament F are shown in the filament image IM1. For example, the preprocessing unit 151 removes portions other than the fuel supply guide 11 and the filament F by cutting out a predetermined area within the filament image IM1.

[0106] (F2. Learning unit 152) Next, with reference to FIG. 11, the function of the learning unit 152 shown in FIG. 8 will be described. FIG. 11 is a diagram for explaining an example of the learning process by the learning unit 152.

[0107] The learning unit 152 executes a learning process for generating the autoencoder 124A. The autoencoder 124A is an example of the above-described estimation model 124. The autoencoder 124A is machine-learned to restore the filament image after compressing a normal filament image in which no thread path abnormality has occurred.

[0108] The machine learning algorithm employed in the learning process is not particularly limited, and for example, a neural network such as deep learning may be employed. Hereinafter, the learning process using a neural network will be described.

[0109] As shown in FIG. 11, the autoencoder 124A is composed of an input layer X, an intermediate layer H, and an output layer Y.

[0110] The input layer X is configured to receive the input of the normal filament image after preprocessing by the preprocessing unit 151. The input layer X is composed of, for example, N units x1 to x N (N is a natural number). The number of units constituting the input layer X is the same as the number of dimensions of the input normal filament image.

[0111] As an example, when the number of pixels of the normal filament image is N pixels and each pixel of the normal filament image is directly input to the input layer X, the input layer X is composed of N units. As another example, the feature amount extracted from the normal filament image may be input to the input layer X. In this case, the input layer X is configured such that the number of its units is the same as the number of dimensions of the feature amount. Each unit constituting the input layer X outputs the input data to each unit of the first layer of the intermediate layer H.

[0112] The intermediate layer H is composed of one or more layers. In the example of FIG. 11, the intermediate layer H is composed of L layers (L is a natural number). Each layer of the intermediate layer H includes a plurality of units. The number of units in each layer of the intermediate layer H may be the same or different. In the example of FIG. 11, the first layer of the intermediate layer H is composed of Q units h A1 ~h AQ (Q is a natural number). Also, the last layer of the intermediate layer H is composed of R units h L1 ~h LR (R is a natural number).

[0113] Each unit constituting each layer of the intermediate layer H is connected to each unit of the previous layer and each unit of the next layer. Each unit of each layer receives the output value from each unit of the previous layer, multiplies each output value by a weight, accumulates the multiplication results, adds (or subtracts) a predetermined bias to the accumulated result, inputs the addition result (or subtraction result) to a predetermined function (for example, a sigmoid function), and outputs the output value of the function to each unit of the next layer.

[0114] In the autoencoder 124A, the number of units constituting each layer of the intermediate layer H is less than the number of units constituting the input layer X. As a result, the dimension of the normal filament image is compressed in the process of being transmitted from the input layer X to the intermediate layer H.

[0115] The output layer Y is configured to restore the normal filament image compressed by the intermediate layer H. More specifically, the output layer Y is composed of the same number of units as the input layer X. As an example, when the input layer X is composed of N units, the output layer Y is composed of N units. In the example of FIG. 11, the output layer Y is composed of N units y1 to y N In the following, the units y1 to y N are also referred to as unit y.

[0116] Each of the units y is connected to each unit h L1 ~h LR of the final layer of the intermediate layer H. Each of the units y receives the output value from each unit of the final layer of the intermediate layer H, multiplies each output value by a weight, accumulates the multiplication results, adds (or subtracts) a predetermined bias to the accumulated result, inputs the addition result (or subtraction result) into a predetermined function (for example, a sigmoid function), and outputs the output result of the function as an output value.

[0117] Next, the update process of the internal parameters of the autoencoder 124A by the learning unit 152 will be described.

[0118] The learning unit 152 inputs the pixels P(1) to P(N) of the first normal filament image into the autoencoder 124A. Thereby, the autoencoder 124A compresses the first normal filament image. Then, the autoencoder 124A restores the compressed filament image so as to approach the input normal filament image. The number of dimensions of the restored filament image is the same as the number of dimensions of the input normal filament image. That is, the restored filament image is composed of pixels P'(1) to P'(N). Next, the learning unit 152 calculates the error "Z" between the input filament image and the restored filament image. As an example, the error "Z" is calculated based on the following formula (1).

[0119] Z = {(P(1) - P'(1)) 2 + ··· + (P(N) - P'(N)) 2} / N ··· (1) Next, the learning unit 152 updates the internal parameters (e.g., weights and biases) of the autoencoder 124A so that the error "Z" becomes smaller. The update of the internal parameters is realized, for example, by the error backpropagation method.

[0120] The learning unit 152 repeatedly performs the update process of the internal parameters of the autoencoder 124A for each normal filament image to be learned. As a result, the autoencoder 124A is learned to restore the normal filament image after compressing the normal filament image.

[0121] That is, when a normal filament image is input, the autoencoder 124A outputs a filament image similar to the normal filament image, and when an abnormal filament image is input, the autoencoder 124A outputs a filament image different from the abnormal filament image. In other words, the autoencoder 124A functions like a kind of filter that passes normal filament images while not passing abnormal filament images.

[0122] (F3. Preprocessing unit 153) Next, the function of the preprocessing unit 153 shown in FIG. 8 will be described.

[0123] The preprocessing unit 153 performs preprocessing on the estimated filament image acquired from the above-described camera 30 (see FIG. 1). Typically, the preprocessing unit 153 executes the same preprocessing as the above-described preprocessing unit 151.

[0124] As an example, the preprocessing unit 153 executes a process of extracting edges from the estimated filament image. Since the edge extraction process has been described in FIG. 9, the description thereof will not be repeated.

[0125] As another example, the preprocessing unit 153 executes a process of removing the portion where the fuel supply guide 11 appears from the estimated filament image. Since the removal process has been described in FIG. 10, the description thereof will not be repeated.

[0126] Note that the preprocessing by the preprocessing unit 153 does not necessarily have to be executed. For example, when the above-described camera 30 is arranged so as not to include the fuel supply guide 11 in the imaging range, only a plurality of filaments F passing through the upstream side of the fuel supply guide 11 are imaged in the estimated filament image. In this case, the above removal process does not need to be executed.

[0127] Also, the preprocessing unit 153 may execute both the above edge extraction process and the above removal process, or may execute only one of the preprocessing.

[0128] Also, the preprocessing unit 153 may execute preprocessing different from the above edge extraction process and the above removal process. As an example, the preprocessing unit 153 may perform a process of cutting out a portion where the fuel supply guide 11 and the filament F are imaged on the filament image IM1. In other words, the preprocessing unit 153 may perform a process of removing portions other than the portion where the fuel supply guide 11 and the filament F are imaged on the filament image IM1. For example, the preprocessing unit 153 removes portions other than the fuel supply guide 11 and the filament F by cutting out a predetermined area within the filament image IM1.

[0129] (F4. Judgment Unit 154) Next, with reference to FIG. 12, the function of the judgment unit 154 shown in FIG. 8 will be described. FIG. 12 is a diagram for explaining an example of the judgment process by the judgment unit 154.

[0130] The judgment unit 154 judges whether or not a thread path abnormality of the filament F has occurred based on the similarity between the estimated filament image after the preprocessing by the above-described preprocessing unit 153 and the restored filament image obtained by inputting the estimated filament image into the autoencoder 124A.

[0131] More specifically, first, the determination unit 154 acquires the autoencoder 124A and inputs the estimated filament image after preprocessing by the preprocessing unit 153 into the autoencoder 124A. The acquisition source of the autoencoder 124A may be a storage device within the information processing apparatus 100 or an external device. When a normal filament image is input, the autoencoder 124A outputs a filament image similar to the normal filament image, and when an abnormal filament image is input, the autoencoder 124A outputs a filament image different from the abnormal filament image.

[0132] Thereafter, the determination unit 154 calculates the similarity between the estimated filament image and the restored filament image. For calculating the similarity, any algorithm can be adopted for the calculation method of the similarity. As an example, algorithms for calculating the similarity may include Mean Squared Error (MSE), Sum of Squared Difference (SSD), Sum of Absolute Difference (SAD), Normalized Cross-Correlation (NCC), or Zero-mean Normalized Cross-Correlation (ZNCC).

[0133] The determination unit 154 determines whether or not a thread path abnormality has occurred based on the comparison result between the calculated similarity and a predetermined threshold value. Here, the magnitude relationship of the calculated similarity may vary depending on the algorithm adopted. That is, the value of the calculated similarity may be smaller as the estimated filament image and the restored filament image are more similar to each other, or may be larger as the estimated filament image and the restored filament image are more similar to each other.

[0134] As an example, an algorithm is adopted in which the smaller the similarity between the estimated filament image and the restored filament image, the more similar they are to each other. In this case, when the calculated similarity is equal to or greater than a predetermined threshold value, the determination unit 154 determines that a thread path abnormality has occurred. On the other hand, when the calculated similarity is smaller than the predetermined threshold value, the determination unit 154 determines that no thread path abnormality has occurred.

[0135] As another example, an algorithm is adopted in which the greater the similarity between the estimated filament image and the restored filament image, the more similar they are to each other. In this case, when the calculated similarity is equal to or less than a predetermined threshold value, the determination unit 154 determines that a thread path abnormality has occurred. On the other hand, when the calculated similarity is greater than the predetermined threshold value, the determination unit 154 determines that no thread path abnormality has occurred.

[0136] The above threshold value may be arbitrarily set by the installer or user of the information processing apparatus 100. Preferably, the above threshold value is set in advance. Since the detection accuracy of the thread path abnormality depends on the setting of the threshold value, the detection accuracy of the thread path abnormality is improved by setting the threshold value in advance.

[0137] As described above, in the present embodiment, the determination unit 154 can detect a thread path abnormality by the autoencoder 124A learned using the normal filament image. That is, in the present embodiment, the designer can realize the detection function of the thread path abnormality without collecting various learning data representing the thread path abnormality.

[0138] (F5. Output unit 155) Next, the function of the output unit 155 shown in FIG. 8 will be described.

[0139] The output unit 155 outputs a control command corresponding to the determination result of the determination unit 154 described above to a predetermined output destination.

[0140] In one aspect, the control command from the output unit 155 is output to the above-mentioned display 106 (see FIG. 6). As a result, the display 106 displays a warning indicating that a yarn path abnormality has occurred. Preferably, an image of the abnormal filament is also displayed on the display 106.

[0141] In another aspect, the control command from the output unit 155 is output to a notification lamp (not shown) provided in the yarn take-up device 1. As an example, when the determination result by the determination unit 154 indicates normality, the output unit 155 lights up the notification lamp in a specific color (for example, green). On the other hand, when the determination result by the determination unit 154 indicates a yarn path abnormality, the output unit 155 lights up the notification lamp in a color (for example, red) different from that used in normal times.

[0142] In still another aspect, the control command from the output unit 155 is output to an alarm buzzer (not shown) provided in the yarn take-up device 1. As an example, when the determination result by the determination unit 154 indicates normality, the output unit 155 does not sound the alarm buzzer. On the other hand, when the determination result by the determination unit 154 indicates a yarn path abnormality, the output unit 155 sounds the alarm buzzer in a predetermined manner.

[0143] In the above description, an example has been described in which the output result from the output unit 155 is output to a device within the yarn take-up device 1, but the output result from the output unit 155 may be sent to an external device different from the yarn take-up device 1. As an example, the output result may be sent to a pre-registered communication terminal. This allows a person in charge or a manager to recognize that a yarn path abnormality has occurred within the yarn take-up device 1.

[0144] <G.配置パターン> As described above, the camera 30 is disposed inside the yarn take-up device 1 so as to be able to photograph the plurality of filaments F passing through the oil supply guide 11. In addition, the light source 40 is disposed inside the yarn take-up device 1 so as to be able to illuminate the plurality of filaments F passing through the oil supply guide 11.

[0145] At this time, when the light source 40 irradiates the filament F, the luminance difference between the filament F and the background becomes distinct, making it easier to remove noise such as the background in the edge detection processing in the above-described preprocessing units 151 and 153. As a result, the detection accuracy of the thread path abnormality of the filament F is improved.

[0146] Preferably, the camera 30 and the light source 40 are arranged such that the optical axis of the light source 40 is inclined with respect to the optical axis of the camera 30. Thereby, the amount of light reflected by the camera 30 can be reduced. As a result, the occurrence of white spots in the filament image can be prevented.

[0147] More preferably, the light source 40 is a spot illumination capable of locally illuminating a specific location. The irradiation angle of the light of the spot illumination is, for example, within 45°. Preferably, the irradiation angle of the light of the spot illumination is, for example, within 30°. The spot illumination is arranged such that its illumination range 40R includes at least a plurality of filaments passing through the oil supply guide 11. That is, the spot illumination is arranged so as not to irradiate objects other than the filaments as much as possible. Also, the spot illumination is arranged such that its illumination range 40R intersects the optical axis of the camera 30.

[0148] Thereby, it is possible to suppress the irradiation of light to objects other than the filaments. Therefore, the amount of light reflected by the camera 30 can be further reduced. As a result, the occurrence of white spots in the filament image can be more reliably prevented.

[0149] Hereinafter, with reference to FIGS. 13 to 15, specific examples will be described regarding the arrangement pattern of the camera 30 and the light source 40 with respect to the oil supply guide 11.

[0150] (G1. Specific Example 1) FIG. 13 is a diagram for explaining an example of the arrangement pattern of the camera 30 and the light source 40. In FIG. 13, the positional relationship among the oil supply guide 11, the camera 30, and the light source 40 is shown from the right direction.

[0151] In this example, the light source 40 is arranged so as to overlap with the camera 30 when viewed from above or below. Also, the light source 40 is arranged below the camera 30. This makes it possible to reduce the amount of light reflected to the camera 30. As a result, it is possible to prevent overexposure in the filament image.

[0152] 13, the optical axis 40AX of the light source 40 intersects with the optical axis 30AX of the camera 30. The angle θ formed by the optical axes 30AX and 40AX is greater than 0° and less than 90°. The angle θ may be greater than 30°, greater than 45°, or greater than 60°.

[0153] Furthermore, the light source 40 is provided closer to the camera 30 than the fuel filler guide 11 (i.e., in the forward direction) in the direction in which the fuel filler guide 11 is viewed from the camera 30 (i.e., in the rearward direction). As a result, the amount of light that directly enters the camera 30 from the light source 40 is reduced. This makes it possible to prevent overexposure in the filament image.

[0154] Preferably, an anti-reflection member 60 for preventing reflection of light emitted from the light source 40 is provided inside the yarn take-up device 1. The anti-reflection member 60 is provided further back (i.e., rearward) than the oil supply guide 11 in the direction in which the oil supply guide 11 is viewed from the camera 30. In this case, the mechanisms are arranged in the order of "camera 30 (light source 40) → oil supply guide 11 → anti-reflection member 60" when viewed from the front side. As a result, the anti-reflection member 60 prevents light emitted from the light source 40 from being reflected by the camera 30. As a result, it is possible to more reliably prevent overexposure in the filament image.

[0155] More preferably, the anti-reflection member 60 is black. This makes the reflectance of black lower than the reflectance of other colors. Therefore, the black anti-reflection member 60 can more reliably prevent light emitted from the light source 40 from being reflected back at the camera 30.

[0156] (G2. Example 2) Fig. 14 is a diagram illustrating another example of the arrangement pattern of the camera 30 and the light source 40. In Fig. 14, the positional relationship between the fuel filler guide 11, the camera 30, and the light source 40 is shown from the right.

[0157] The light source 40 is arranged so as to overlap with the camera 30 when viewed from above or below. In this example, the light source 40 is arranged above the camera 30. This reduces the amount of light reflected to the camera 30. As a result, it is possible to prevent overexposure in the filament image.

[0158] 14, the optical axis 40AX of the light source 40 intersects with the optical axis 30AX of the camera 30. The angle θ formed by the optical axes 30AX and 40AX is greater than 0° and less than 90°. The angle θ may be greater than 30°, greater than 45°, or greater than 60°.

[0159] Furthermore, the light source 40 is provided closer to the camera 30 than the fuel filler guide 11 (i.e., in the forward direction) in the direction in which the fuel filler guide 11 is viewed from the camera 30 (i.e., in the rearward direction). As a result, the amount of light that directly enters the camera 30 from the light source 40 is reduced. This makes it possible to prevent overexposure in the filament image.

[0160] Preferably, an anti-reflection member 60 for preventing reflection of light emitted from the light source 40 is provided inside the yarn take-up device 1. The anti-reflection member 60 is as described above, and therefore, description thereof will not be repeated.

[0161] Although the above description has been given of an example in which one light source 40 is arranged, the number of light sources 40 may be two or more. As an example, the first light source 40 may be arranged above the camera 30, and the second light source 40 may be arranged below the camera 30.

[0162] (G3. Example 3) FIG. 15 is a diagram for explaining yet another example of the arrangement pattern of the camera 30 and the light source 40. In FIG. 15, the positional relationship among the fuel supply guide 11, the camera 30, and the light source 40 is shown from above.

[0163] In this example, the light source 40 is arranged on the right side of the camera 30. Thereby, the amount of light reflected by the camera 30 can be reduced. As a result, the occurrence of white spots in the filament image can be prevented.

[0164] In the example of FIG. 15, the optical axis 40AX of the light source 40 intersects with the optical axis 30AX of the camera 30. The angle θ formed by the optical axes 30AX and 40AX is greater than 0° and less than 90°. The angle θ may be 30° or more, may be 45° or more, or may be 60° or more.

[0165] Also, the light source 40 is provided on the front side (i.e., the forward direction) of the fuel supply guide 11 in the direction in which the fuel supply guide 11 is viewed from the camera 30 (i.e., the backward direction). As a result, the amount of light directly incident from the light source 40 to the camera 30 is reduced. Thereby, the occurrence of white spots in the filament image can be prevented.

[0166] Preferably, an anti-reflection member 60 for preventing reflection of the light irradiated from the light source 40 is provided inside the spinning take-up device 1. Since the anti-reflection member 60 is as described above, the description thereof will not be repeated.

[0167] Note that, in the above description, the example in which the light source 40 is arranged on the right side of the camera 30 has been described, but the light source 40 may be arranged on the left side of the camera 30.

[0168] Also, in the above description, the example in which one light source 40 is arranged has been described, but the number of light sources 40 may be two or more. As an example, the first light source 40 may be arranged on the right side of the camera 30, and the second light source 40 may be arranged on the left side of the camera 30.

[0169] <H.学習処理に係るフローチャート> Next, a flowchart relating to the learning process performed by the information processing device 100 will be described with reference to Fig. 16. Fig. 16 is a flowchart showing the flow of the learning process.

[0170] The control device 101 of the information processing device 100 executes the above-described learning program 126 (see FIG. 6) to perform the various processes shown in FIG. 16. In another aspect, some or all of the processes shown in FIG. 16 may be performed by circuit elements or other hardware.

[0171] In step S110, the control device 101 functions as the above-described preprocessing unit 151 (see FIG. 8) and performs predetermined preprocessing on the filament image as the learning data 123. The preprocessing is as described above, and therefore the description thereof will not be repeated.

[0172] In step S112, the control device 101 inputs the filament image after preprocessing in step S110 to the above-mentioned autoencoder 124A.

[0173] In step S114, the control device 101 functions as the learning unit 152 (see FIG. 8) described above and calculates the error between the filament image input to the autoencoder 124A and the restored filament image output from the autoencoder 124A. Thereafter, the control device 101 updates the internal parameters of the autoencoder 124A so that the error becomes smaller than the current error. The parameters are updated, for example, by the backpropagation method. The learning process in step S114 is as described above, and therefore will not be described again.

[0174] In step S120, the control device 101 determines whether to end the learning process. As an example, the control device 101 determines to end the learning process when the estimation accuracy using the test data exceeds a desired accuracy. Alternatively, the control device 101 determines to end the learning process when the number of updates of the internal parameters of the autoencoder 124A exceeds a predetermined number.

[0175] If it is determined that the learning process is to be ended (YES in step S120), the control device 101 ends the process shown in Fig. 16. Otherwise (NO in step S120), the control device 101 returns the control to step S112.

[0176] <I.異常検出処理に係るフローチャート> Next, a flowchart of the abnormality detection process performed by the information processing device 100 will be described with reference to Fig. 17. Fig. 17 is a flowchart showing the flow of the abnormality detection process.

[0177] The control device 101 of the information processing device 100 executes the above-described abnormality detection program 128 (see FIG. 6) to perform various processes shown in FIG. 17. In another aspect, some or all of the processes shown in FIG. 17 may be performed by circuit elements or other hardware.

[0178] In step S210, the control device 101 acquires a filament image for estimation from the camera 30 described above.

[0179] In step S212, the control device 101 functions as the above-described preprocessing unit 153 (see FIG. 8) and performs predetermined preprocessing on the filament image acquired in step S210. The preprocessing is as described above, and therefore the description thereof will not be repeated.

[0180] In step S214, the control device 101 inputs the filament image after the preprocessing in step S212 to the trained autoencoder 124A.

[0181] In step S216, the control device 101 functions as the above-described determination unit 154 (see FIG. 8), and calculates the degree of similarity between the filament image input to the autoencoder 124A and the restored filament image output from the autoencoder 124A.

[0182] In step S230, the control device 101 functions as the above-described determination unit 154, and determines whether the degree of similarity calculated in step S216 satisfies the abnormal condition. The abnormal condition is satisfied when the filament image and the restored filament image are not similar. As an example, when an algorithm is adopted in which the degree of similarity decreases as the estimated filament image and the restored filament image become more similar to each other, the above abnormal condition is satisfied when the degree of similarity exceeds a predetermined threshold.

[0183] When the control device 101 determines that the degree of similarity calculated in step S216 satisfies the abnormal condition (YES in step S230), the control is switched to step S232. Otherwise (NO in step S230), the control device 101 switches the control to step S234.

[0184] In step S232, the control device 101 functions as the above-described output unit 155 (see FIG. 8), and outputs a determination result indicating the occurrence of a thread path abnormality. Since the output process is as described above, the description thereof will not be repeated.

[0185] In step S234, the control device 101 functions as the above-described output unit 155, and outputs a determination result indicating normality. Since the output process is as described above, the description thereof will not be repeated.

[0186] Preferably, the abnormality detection process shown in FIG. 17 is periodically executed when the spinning take-up device 1 is manufacturing a thread.

[0187] <J. First Modified Example> Next, a first modified example of the above embodiment will be described with reference to Fig. 18. Fig. 18 is a diagram showing an example of the device configuration of a filament abnormality detection system 500 in this modified example.

[0188] 8, the learning function and the anomaly detection function are implemented in the same information processing device 100. However, the learning function and the anomaly detection function do not necessarily have to be implemented in the same information processing device 100. As an example, the learning function and the anomaly detection function may be implemented in different information processing devices 100.

[0189] As shown in Fig. 18, the filament abnormality detection system 500 includes one or more yarn take-up devices 1 and one or more information processing devices 100. In the example of Fig. 18, the filament abnormality detection system 500 is configured with three yarn take-up devices 1A to 1C and two information processing devices 100A and 100B.

[0190] The information processing device 100A collects the above-mentioned learning data 123 (see FIG. 7) from the yarn take-up devices 1 (for example, the yarn take-up devices 1A and 1B) connected to the network NW. Next, the learning unit 152 of the information processing device 100A executes a learning process using the learning data 123 that has been preprocessed by the preprocessing unit 151, and generates the above-mentioned estimation model 124. The generated estimation model 124 is transmitted to the information processing device 100B.

[0191] The information processing device 100B acquires an estimation filament image from the spinning take-up device 1C. Next, the judgment unit 154 of the information processing device 100B inputs the estimation filament image preprocessed by the preprocessing unit 153 to the autoencoder 124A. Next, the judgment unit 154 of the information processing device 100B judges whether or not a filament thread path error has occurred based on the similarity between the estimation filament image and the restored filament image obtained from the autoencoder 124A. The judgment process is as described above, so its description will not be repeated. Thereafter, the output unit 155 of the information processing device 100B outputs the judgment result to the spinning take-up device 1C.

[0192] <K. Second Modification Example> In the above, the estimation model 124 as the autoencoder 124A has been described, but the estimation model 124 is not limited to the autoencoder 124A.

[0193] Hereinafter, with reference to FIGS. 19 to 21, an example of detecting a thread path abnormality using an estimation model 124 other than the autoencoder 124A will be described.

[0194] (K1. Training Dataset 122) First, with reference to FIG. 19, the training dataset 122 used to generate an estimation model 124 other than the autoencoder 124A will be described. FIG. 19 is a diagram showing the training dataset 122 according to this modification example.

[0195] The training dataset 122 includes a plurality of training data 123. The number of training data 123 included in the training dataset 122 is arbitrary. As an example, the number of training data 123 is several tens to several hundreds of thousands.

[0196] The training data 123 includes training filament images that depict a plurality of filaments passing through the oil supply guide 11. In this modification example, a label indicating whether or not a thread path abnormality has occurred in the filament is further associated with each of the training filament images. In other respects, it is the same as the training data 123 shown in FIG. 7 above.

[0197] As an example, the labels defined in the training dataset 122 include "thread path abnormality" indicating that the filament has deviated from the normal thread path on the oil supply guide 11, and "normal" indicating that the filament is passing through the normal thread path on the oil supply guide 11. Each label may be distinguished by a combination of numerical values or by a combination of character strings.

[0198] (K2. Learning Unit 152) Next, a modified example of the learning unit 152 shown in Fig. 8 will be described with reference to Fig. 20. Fig. 20 is a diagram for explaining an example of the learning process performed by the learning unit 152.

[0199] The learning unit 152 generates an estimation model 124B according to this modification by a learning process using the learning dataset 122 shown in Fig. 19. The machine learning algorithm used in the learning process is not particularly limited, and may be, for example, deep learning, a convolutional neural network (CNN), a full-layer convolutional neural network (FCN), R-CNN (Regions with Convolutional Neural Network), Fast R-CNN, Faster R-CNN, YOLO (You Only Look Once), a support vector machine, or the like. The learning process using deep learning will be described below.

[0200] As shown in FIG. 20, the estimation model 124B is configured with an input layer X, an intermediate layer H, and an output layer Y.

[0201] The input layer X is configured to receive as input a filament image as learning data 123. The input layer X includes, for example, N units x1 to x N (N is a natural number). The number of units that make up the input layer X is the same as the number of dimensions of the input information.

[0202] As an example, if the number of pixels in the filament image for learning is N pixels and each pixel of the filament image is input directly to the input layer X, the input layer X will be composed of N units. As another example, feature extraction processing (preprocessing) may be performed on the filament image. In this case, the extracted feature quantities are input to the input layer X. The input layer X is configured so that the number of units is the same as the number of dimensions of the feature quantities after feature extraction. Each unit constituting the input layer X outputs the input data to each unit in the first layer of the intermediate layer H.

[0203] The intermediate layer H is composed of one or more layers. In the example of FIG. 20, the intermediate layer H is composed of L layers (L is a natural number). Each layer of the intermediate layer H includes a plurality of units. The number of units in each layer of the intermediate layer H may be the same or different. In the example of FIG. 20, the first layer of the intermediate layer H is composed of Q units h A1 ~h AQ (Q is a natural number). Also, the last layer of the intermediate layer H is composed of R units h L1 ~h LR (R is a natural number).

[0204] Each unit constituting each layer of the intermediate layer H is connected to each unit of the previous layer and each unit of the next layer. Each unit of each layer receives the output values from each unit of the previous layer, multiplies each output value by a weight, accumulates the multiplication results, adds (or subtracts) a predetermined bias to the accumulated result, inputs the addition result (or subtraction result) into a predetermined function (for example, a sigmoid function), and outputs the output value of the function to each unit of the next layer.

[0205] The output layer Y outputs an estimation result according to the input filament image. The output layer Y is composed of, for example, units y1, y2.

[0206] Each of the units y1, y2 is connected to the units h L1 ~h LR of the last layer of the intermediate layer H. Each of the units y1, y2 receives the output values from each unit of the last layer of the intermediate layer H, multiplies each output value by a weight, accumulates the multiplication results, adds (or subtracts) a predetermined bias to the accumulated result, inputs the addition result (or subtraction result) into a predetermined function (for example, a sigmoid function), and outputs the output result of the function as an output value.

[0207] The number of units constituting the output layer Y is determined according to the number of types of labels defined in the learning data 123. As an example, if the learning data 123 contains two types of labels, "yarn path abnormality" and "normal," the number of units constituting the output layer Y will be two, units y1 and y2. In this case, unit y1 outputs a score "sa" indicating the possibility that a yarn path abnormality has occurred. Unit y2 outputs a score "sb" indicating the possibility that the filament yarn path is normal.

[0208] Next, the update process of the internal parameters of the estimation model 124B by the learning unit 152 will be described.

[0209] First, the learning unit 152 inputs the filament image defined in the first training data 123 into the estimation model 124B. Next, the learning unit 152 compares the estimation results “sa”, “sb” output from the estimation model 124B with the correct scores “sa′”, “sb′” corresponding to the labels associated with the first training data 123.

[0210] As an example, if the label associated with the learning data 123 is "yarn path abnormality", the correct score is (sa', sb') = (1, 0). On the other hand, if the label associated with the learning data 123 is "normal", the correct score is (sa', sb') = (0, 1).

[0211] The learning unit 152 calculates the error "Z" between the output results "sa", "sb" of the estimation model 124B and the correct scores "sa'", "sb'". The error "Z" is calculated, for example, based on the following equation (2).

[0212] Z={(sa-sa') 2 +(sb-sb') 2} / 2···(2) Next, the learning unit 152 updates various parameters (for example, weights and biases) included in the estimation model 124B so as to reduce the error "Z." The parameter update is realized, for example, by the backpropagation algorithm.

[0213] The learning unit 152 repeatedly performs the update process of the internal parameters of the estimation model 124B for each learning data 123 included in the learning dataset 122. As a result, the estimation model 124B will output accurate estimation results as learning progresses.

[0214] Note that the learning unit 152 does not necessarily need to use all the learning data 123 included in the learning dataset 122 for the learning process, and the estimation model 124B may be generated using some of the learning data 123 included in the learning dataset 122. The remaining learning data 123 is used, for example, for evaluating the estimation model 124B.

[0215] (K3. Judgment unit 154) Next, with reference to FIG. 21, a modified example of the judgment unit 154 shown in FIG. 8 will be described. FIG. 21 is a diagram for explaining an example of the judgment process by the judgment unit 154.

[0216] First, the judgment unit 154 acquires the estimation model 124B generated by the learning unit 152 from the storage destination. The acquisition destination of the estimation model 124B may be the auxiliary storage device 120 described above or an external device.

[0217] Next, the judgment unit 154 causes the fuel supply guide 11 through which the filament passes to be photographed by the above-described camera 30, and acquires an estimation filament image from the camera 30. Next, the judgment unit 154 determines whether or not a filament path abnormality has occurred based on the output result obtained from the estimation model 124B by inputting the estimation filament image into the estimation model 124B.

[0218] The estimation model 124B includes, for example, a score "sa" indicating the possibility that the filament path is abnormal and a score "sb" indicating the possibility that the filament path is normal. In this case, when the score "sa" exceeds the first threshold value and the score "sb" is below the second threshold value, the determination unit 154 determines that a filament path abnormality has occurred. Otherwise, the determination unit 154 determines that no filament path abnormality has occurred.

[0219] The first threshold value and the second threshold value may be set in advance or may be arbitrarily set by the user. Also, the first threshold value and the second threshold value may be the same or different.

[0220] Note that in the above description, the explanation was made on the premise that the estimation model 124B outputs two scores "sa" and "sb". However, the estimation model 124B may be configured to output only the score "sa" indicating the possibility of occurrence of a filament path abnormality. In this case, when the score "sa" exceeds the first threshold value, the determination unit 154 determines that a filament path abnormality has occurred. On the other hand, when the score "sa" is less than or equal to the first threshold value, the determination unit 154 determines that no filament path abnormality has occurred.

[0221] It should be considered that all the embodiments disclosed this time are illustrative in all respects and not restrictive. The scope of the present invention is shown not by the above description but by the claims, and it is intended that all modifications within the meaning and scope equivalent to the claims are included.

Explanation of Signs

[0222] 1 Spinning take-up device 30 Camera 40 Light source 60 Anti-reflection member 100 Information processing device 101 Control device 123 Learning data 124 Estimation model 124A Autoencoder 124B Presumption Model

Claims

1. A filament abnormality detection device capable of detecting an abnormality in a spinning take-up device that takes up a plurality of filaments spun from a spinning device, The spinning take-up device includes a guide for guiding the plurality of filaments sent from the upstream side to the downstream side so as to bring them closer to each other, The filament abnormality detection device includes a control device, The control device, A process of acquiring an estimated filament image that captures the plurality of filaments from a camera arranged so as to include at least the imaging range of the plurality of filaments passing through the guide, Based on an estimation model learned by machine learning for detecting a thread path abnormality in which at least one of the plurality of filaments has deviated from the guide, and the estimated filament image, a process of determining whether or not the thread path abnormality has occurred, A filament abnormality detection device that executes a process of outputting a determination result in the determination process.

2. The estimation model is an autoencoder learned by machine learning to restore the filament image after compressing a normal filament image in which no thread path abnormality has occurred, In the determination process, based on the similarity between the estimated filament image and the restored filament image obtained by inputting the estimated filament image into the autoencoder, it is determined whether or not the thread path abnormality has occurred. The filament abnormality detection device according to claim 1.

3. In the determination process, based on the comparison result between the similarity and a predetermined threshold value, it is determined whether or not the thread path abnormality has occurred. The filament abnormality detection device according to claim 2.

4. The estimation model is generated by machine learning a plurality of learning data, Each of the plurality of learning data is associated with a label indicating whether or not the thread path abnormality has occurred with respect to a learning filament image that captures a plurality of filaments passing through the guide, In the determination process, based on the output result obtained from the estimation model by inputting the estimated filament image into the estimation model, it is determined whether or not the thread path abnormality has occurred. The filament abnormality detection device according to claim 1.

5. The determination process includes, A process of performing preprocessing on the estimated filament image, A process of inputting the preprocessed estimated filament image into the estimation model. The filament abnormality detection device according to any one of claims 1 to 4, wherein the preprocessing includes a process of extracting edges from the filament image for estimation.

6. The filament abnormality detection device according to any one of claims 1 to 5, wherein the spinning take-up device further includes a light source arranged so as to include at least the plurality of filaments passing through the upstream side of the guide in an illumination range.

7. The guide has a main body having a surface with which the plurality of filaments sent in the direction of gravity from the spinning device come into contact, a discharge port formed on the surface for discharging an oil agent, and includes two yarn guiding members provided on the surface so as to be located on both sides of the discharge port in the horizontal direction, for guiding the plurality of filaments sent in the direction of gravity to the center side of the discharge port in the horizontal direction. The camera is arranged so that the imaging range of the camera includes at least the plurality of filaments passing through the upstream side of the surface. The filament abnormality detection device according to claim 6, wherein the light source is arranged so that the illumination range of the light source includes at least the plurality of filaments passing through the upstream side of the surface.

8. The filament abnormality detection device according to claim 6 or 7, wherein the optical axis of the light source is inclined with respect to the optical axis of the camera.

9. The filament abnormality detection device according to any one of claims 6 to 8, wherein the light source is provided on the front side of the guide in the direction of viewing the guide from the camera.

10. The spinning take-up device further includes an anti-reflection member for preventing reflection of light irradiated from the light source. The filament abnormality detection device according to any one of claims 6 to 9, wherein the anti-reflection member is provided on the back side of the guide in the direction of viewing the guide from the camera.

11. The filament abnormality detection device according to claim 10, wherein the anti-reflection member is black.

12. The light source is spot illumination. The filament abnormality detection device according to any one of claims 6 to 11, wherein the illumination range of the spot illumination intersects the optical axis of the camera.

13. An abnormality detection program capable of detecting an abnormality in a spinning take-up device that takes up a plurality of filaments spun from a spinning device. The spinning take-up device includes a guide for guiding the plurality of filaments so as to bring them closer to each other. The abnormality detection program causes a filament abnormality detection device to acquire an estimated filament image capturing the plurality of filaments from a camera arranged so as to include at least the imaging range of the plurality of filaments passing through the guide; input the estimated filament image into an estimation model that has been machine-learned to estimate a thread path abnormality in which at least one of the plurality of filaments has deviated from the guide; determine whether or not the thread path abnormality has occurred based on the output result of the estimation model; An abnormality detection program that executes a step of outputting the determination result in the determining step. **Claim 14** The estimation model is an autoencoder that has been machine-learned to restore the filament image after compressing a normal filament image in which no thread path abnormality has occurred. In the determining step, it is determined whether or not the thread path abnormality has occurred based on the degree of similarity between the estimated filament image and the restored filament image obtained by inputting the estimated filament image into the autoencoder. The abnormality detection program according to claim 13. **Claim 15** The determining step includes a step of performing preprocessing on the estimated filament image; and a step of inputting the preprocessed estimated filament image into the estimation model. The preprocessing includes a process of extracting edges from the estimated filament image. The abnormality detection program according to claim 13 or 14.

Images