Computer program, flat knitting machine, and information processing method

Abstract

Provided are a computer program, a flat knitting machine, and an information processing method for outputting information about the position and / or defect of a knitting member determined to be defective.　The computer program causes a computer to execute processing for acquiring an image of a knitting member of a flat knitting machine having a plurality of knitting members arranged in a row, determining the quality of the knitting member on the basis of the acquired image of the knitting member, and outputting information about the position and / or defect of the knitting member determined to be defective.

Description

Computer program, flat knitting machine, and information processing method

[0001] The present disclosure relates to a computer program, a flat knitting machine, and an information processing method.

[0002] A flat knitting machine has a large number of knitting needles arranged in rows on at least two needle beds facing each other in the front and rear, and forms a knitted fabric by selectively driving the knitting needles with a carriage that moves over the needle beds and hooking the supplied knitting yarn. Bent or damaged knitting needles can cause knitting defects.

[0003] Patent Document 1 discloses a method for optically monitoring the quality of needles of a knitting machine by photographing the needles with a camera.

[0004] Special Publication No. 7-33998

[0005] Although the method of Patent Document 1 can detect defects such as gross damage and minute damage to needles, it does not provide information about the location of defective needles or about the defects.

[0006] The present disclosure has been made in consideration of the above circumstances, and aims to provide a computer program, a flat knitting machine, and an information processing method that output at least one of the position of a knitting member determined to be defective and information related to the defect.

[0007] The computer program of this embodiment causes a computer to execute a process of acquiring images of knitting members of a flat knitting machine having multiple knitting members arranged in a row, determining whether the knitting members are good or bad based on the acquired images of the knitting members, and outputting at least one of the position of knitting members determined to be defective and information regarding the defect.

[0008] The flat knitting machine of this embodiment comprises an acquisition unit that acquires images of knitting members of a flat knitting machine having a plurality of knitting members arranged in a row, a judgment unit that judges whether the knitting members are good or bad based on the acquired images of the knitting members, and an output unit that outputs at least one of the position of the knitting members judged to be defective and information regarding the defect.

[0009] The information processing method of this embodiment acquires images of knitting members of a flat knitting machine having multiple knitting members arranged in a row, determines whether the knitting members are good or bad based on the acquired images of the knitting members, and outputs at least one of the position of the knitting members determined to be defective and information regarding the defect.

[0010] In this embodiment, the quality of a knitting member is determined based on an acquired image of the knitting member. The knitting member is, for example, a knitting needle, but may also include loop pressers and transfer jacks. The knitting needle includes a hook at the tip of the needle, a latch, a slider, and a needle shaft. The quality determination can be performed using, for example, the similarity (e.g., cosine similarity) between the acquired image of the knitting member and an image of a normal knitting member, a learning model generated by machine learning, or the like. If the knitting member is determined to be defective, at least one of the location of the knitting member determined to be defective and information related to the defect is identified and output. Information related to the defect includes, for example, the accuracy (probability) of the defect and the cause of the defect. According to this embodiment, at least one of the location of the knitting member determined to be defective and information related to the defect can be provided.

[0011] The computer program of this embodiment causes a computer to execute a process of outputting a countermeasure for a knitting element determined to be defective, based on a database that associates defects in knitting elements with countermeasures.

[0012] A defect in a knitting member is, for example, the cause of the defect, and includes bending or loss of the knitting member. For example, in the case of a hook, this includes bending, breaking, bowing (displacement of the hook tip), etc. Countermeasures include adjusting the knitting method (for example, separate knitting, changing the stitch size), adjusting the output of the servo motor for driving the needle bed, adjusting the pull-down force of the knitted fabric, etc. According to this embodiment, it is possible to provide the user with an appropriate countermeasure for a defective knitting member.

[0013] The computer program of this embodiment acquires historical information including the knitting program used for knitting and knitting machine adjustment data, and determines the factors that impose load on the knitting member determined to be defective based on information on the knitting member determined to be defective and the acquired historical information, and causes the computer to execute a process to output the determination result.

[0014] Information on knitting members determined to be defective includes, for example, the position of the defective knitting member and the cause of the defect. The knitting program contains instructions that instruct which knitting members (e.g., knitting needles, etc.) to move in each knitting course, and how to move the yarn feeder and carriage. A course refers to one movement of the carriage during knitting. The knitting machine adjustment data includes, for example, settings for a pull-down device that pulls down the knitted fabric (e.g., which area of ​​the knitted fabric to pull down, for how long, and with what force), knitting speed settings (carriage movement speed), and stitch size settings (e.g., the amount by which the knitting needles holding the knitting yarn are lowered by a cam to achieve the target stitch size). The history information is the knitting program and flatbed knitting machine adjustment data used for such knitting, collected for each knitting. Specifically, the history information is a knitting program and flatbed knitting machine adjustment data that, when a knitting member is determined to be defective, are used to look up what knitting the knitting member has been used for in the past.

[0015] The connection of knitted stitches can be emulated based on the knitting program. The connection of knitted stitches can be expressed two-dimensionally, for example, with the position of the knitting needles on the horizontal axis and the course (time) on the vertical axis, or it can be expressed three-dimensionally by adding the relationship between the front and rear needle beds. It is possible to monitor the state of all knitting members in each course based on the knitting program and knitting machine adjustment data. Since it is known how the knitting members on each course corresponding to the position of a knitting member determined to be defective will behave and what state they are in terms of knitting, it is possible to determine the factors that apply load to the knitting members.

[0016] In this way, according to this embodiment, by linking defects in knitting elements (defective areas and positions of knitting elements) with the knitting program and knitting machine adjustment data, it is possible to determine the factors that place load on the knitting elements, and to prevent overload from being placed on the knitting elements.

[0017] In this embodiment, when an image of a trained component that is determined to be normal is input, the computer program inputs the acquired image of the trained component into a first learning model that has been trained to reconstruct the image of the trained component, outputs a reconstructed image, and executes a process to compare the acquired image of the trained component with the output reconstructed image to determine whether the trained component is good or bad.

[0018] The first learning model can be, for example, an autoencoder. The first learning model is generated (trained) so that when an image of a normal knitting member is input, it reconstructs the image. The acquired image of the knitting member is input to the first learning model. The input image of the knitting member is compared with the reconstructed image output by the first learning model. The comparison can be performed by calculating the difference between the input image of the knitting member and the reconstructed image. If the difference is small, the knitting member can be determined to be normal, and if the difference is large or small, the knitting member can be determined to be defective. According to this embodiment, even if it is difficult to collect images of defective knitting members as training data, the quality of the knitting member can be determined by using images of normal knitting members as training data.

[0019] The computer program of this embodiment causes a computer to execute a process of determining whether a knitting member is good or bad based on the similarity between the acquired image of the knitting member and an image of a normal knitting member.

[0020] The similarity can be determined, for example, by cosine similarity. Cosine similarity is the cosine value of the angle between two vectors and can be calculated by dividing the dot product of the two vectors by the magnitude of the two vectors. The pixel values ​​of each pixel in the image of the knitting element can be vectorized. Cosine similarity is normalized to a range of -1 to 1. If the cosine similarity is 1 or close to 1, the knitting element can be determined to be normal. If the cosine similarity is -1 or close to -1, the knitting element can be determined to be defective. In other words, if the cosine similarity is below a predetermined threshold, the knitting element can be determined to be defective. According to this embodiment, there is no need to prepare a huge number of images of good and defective products, reducing the effort required for advance preparation and enabling high-precision identification processing.

[0021] The computer program of this embodiment causes a computer to execute a process of inputting an acquired image of a knitting member into a second learning model that has been trained to output information regarding the quality of the knitting member when an image of the knitting member is input, and outputting information regarding the quality of the knitting member.

[0022] The information regarding the quality of the knitting member includes, for example, the probability (accuracy) that the knitting member is defective and the probability (accuracy) that the knitting member is normal. For example, if the threshold for determining whether the knitting member is normal is 90%, and the output of the second learning model indicates an 80% probability that the knitting member is normal and a 20% probability that it is defective, the knitting member can be determined to be defective. Furthermore, if the output of the second learning model indicates a 95% probability that the knitting member is normal and a 5% probability that it is defective, the knitting member can be determined to be normal. The threshold may be set to a different value depending on the type of knitting needle (hook, latch, slider, needle stem, etc.), loop presser, transfer jack, etc. According to this embodiment, the accuracy of determining whether the knitting member is good or bad can be improved by appropriately setting the threshold.

[0023] The computer program of this embodiment causes a computer to execute a process in which an acquired image of a knitting component is input into a third learning model that has been trained to output defective parts and causes of defects in the knitting component when an image of the knitting component is input, and the computer then executes a process in which the acquired image of the knitting component is input into a third learning model that has been trained to output defective parts and causes of defects in the knitting component.

[0024] The third learning model can be, for example, YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector), or Faster R-CNN. The third learning model extracts features from the input image of the trained components, represents defective areas of the trained components using bounding boxes, and estimates the probability of the cause of the defect. The cause with the highest probability can be determined to be the cause of the defect.

[0025] According to this embodiment, it is possible to provide the user with information about defective parts of knitted members and the causes of the defects.

[0026] The computer program of this embodiment causes a computer to execute a process of outputting information about defects in knitting members to a monitor.

[0027] According to this embodiment, information regarding defects in knitting members (for example, the probability of the defect, the cause of the defect, etc.) can be provided to the user.

[0028] The computer program of this embodiment causes a computer to execute a process of outputting to a monitor at least one of an image of a knitting member determined to be defective and information for identifying the position of the knitting member determined to be defective.

[0029] According to this embodiment, it is possible to provide the user with at least one of an image of the knitting member determined to be defective and information for identifying the position of the knitting member determined to be defective.

[0030] The computer program of this embodiment causes the computer to execute a process of moving a knitting member determined to be defective to a height different from the height of the knitting member according to the knitting program.

[0031] According to this embodiment, the height of the knitting member determined to be defective is moved to a height different from normal, so that the position of the defective knitting member can be clearly notified to the user.

[0032] The computer program of this embodiment causes the computer to execute a process of outputting the position of defective knitting members and the cause of the defect to the monitor of the customer's flat knitting machine, accepting information on the success or failure of the outputted position of the knitting members and the cause of the defect, and storing the accepted information on success or failure and images of the knitting members determined to be defective as training data.

[0033] According to this embodiment, the positions of defective knitting members and the causes of the defects are provided to the customer, and the customer can judge whether the provided information is correct or not, and can correct the provided information as necessary. Information such as the positions of defective knitting members and the causes of the defects, which has been judged correct or not by the customer, can be collected as training data for a third learning model used in the flat knitting machine operated by the customer.

[0034] The computer program of this embodiment causes a computer to execute a process of identifying the third learning model, which has been re-learned using the training data, as a model for the customer's flat knitting machine.

[0035] According to this embodiment, a third learning model tuned for the customer's flat knitting machine can be used.

[0036] According to the present disclosure, it is possible to provide at least one of the location of a knitting member determined to be defective and information regarding the defect.

[0037] 1 is a diagram showing a front view of the overall configuration of the flat knitting machine of the present embodiment. FIG. 1 is a diagram showing an example of the configuration of a needle bed and a carriage. FIG. 2 is a diagram showing an example of the configuration of a cam mechanism provided on the carriage. FIG. 2 is a diagram showing an example of the configuration of a control unit. FIG. 3 is a diagram showing an example of the configuration of a knitting needle and an image of a defective knitting needle. FIG. 4 is a diagram showing an example of the configuration of a knitting needle and an image of a defective knitting needle. FIG. 5 is a diagram showing an example of the configuration of a knitting needle and an image of a defective knitting needle. FIG. 6 is a diagram showing a first example of determining whether a knitting needle is good or bad. FIG. 7 is a diagram showing a second example of determining whether a knitting needle is good or bad. FIG. 8 is a diagram showing a third example of determining whether a knitting needle is good or bad. FIG. 9 is a diagram showing an example of a method for determining the position of a defective knitting needle and the cause of the defect. FIG. 10 is a diagram showing an example of a method for determining the position of a defective knitting needle and the cause of the defect. FIG. 11 is a diagram showing a first example of a result of determining whether a knitting needle is good or bad. FIG. 12 is a diagram showing a second example of a result of determining whether a knitting needle is good or bad. FIG. 13 is a diagram showing an example of information relating the cause of a defect in a knitting needle to a countermeasure. FIG. 14 is a diagram showing an example of separate knitting. FIG. 15 is a diagram showing an example of separate knitting. FIG. 16 is a diagram showing an example of separate knitting. It is a figure which shows typically the emulation result of the connection of the stitches.It is a figure which shows an example of the factor which gives a load to the knitting needle.It is a figure which shows an example of the processing procedure by the control unit.

[0038] An embodiment of the present disclosure will be described below. Fig. 1 is a diagram showing a front view of the overall configuration of a flatbed knitting machine 100 of this embodiment, Fig. 2 is a diagram showing an example of the configuration of a needle bed 10 and a carriage 20, and Fig. 3 is a diagram showing an example of the configuration of a cam mechanism 21 provided on the carriage 20. In the following description, the directions indicated by arrows U, D, F, B, L, and R in Figs. 1 to 3 are defined as the upward direction, downward direction, forward direction, backward direction, leftward direction, and rightward direction, respectively. In addition, for the sake of convenience, some components are omitted from the drawings as appropriate.

[0039] 1 and 2, the flat knitting machine 100 includes a needle bed 10, a carriage 20, a yarn guide rail 30, a spool stand 40, a servo motor and a control unit (not shown), etc. The flat knitting machine 100 knits a knitted fabric K.

[0040] The needle beds 10 are arranged facing each other in the front and rear with the needle gap S between them. The front and rear needle beds 10 are arranged in an inverted V shape in side view, sloping upward toward the front-rear center (the sides facing each other) (see FIG. 2 ). Each needle bed 10 is provided with a large number of knitting needles 11 arranged along the longitudinal direction (left-right direction) of the needle bed 10. The front and rear needle beds 10 can move relatively left and right when transferring knitted stitches to each other. In this specification, the knitting needles 11 include hooks, latches, sliders, and needle shafts, which are the tips of the needles, as will be described later. Furthermore, while FIGS. 1 and 2 show a configuration in which two needle beds 10 are arranged facing each other in the front and rear, this is not limited thereto. The flatbed knitting machine 100 may also include two upper and two lower needle beds 10, for a total of four needle beds 10. In this embodiment, the knitting member is, for example, the knitting needle 11, but other components include a loop presser and a transfer jack. The loop presser is a component driven by the carriage 20 to press down the yarn between the stitches and is used to prevent the knitted fabric or yarn from lifting. The transfer jack receives the stitch from the knitting needle 11 on the needle bed 10, moves left and right relative to the needle bed 10, and then transfers the stitch back to the knitting needle 11. During the knitting process, defects can occur in the loop presser and transfer jack, just as in the knitting needle 11. For example, if the opposing needle bed becomes smaller than the target racking amount (left and right movement amount) due to increased yarn tension, advancing the loop presser may collide with the opposing knitting needle, causing damage. Furthermore, similar to bending of the knitting needle 11, the transfer jack may bend when the needle bed 10 is moved relative to the needle bed 10 under high yarn tension. In this specification, the knitting needle 11 will be described as an example of the knitting member, but the knitting member is not limited to the knitting needle 11.

[0041] A pair of carriages 20 are arranged in front and behind so as to face each other from above the front and rear needle beds 10. The carriages 20 can be reciprocated along the longitudinal direction of the needle beds 10 by a servo motor (not shown).

[0042] Of the front and rear needle beds 10, the front needle bed 10 is also referred to as the front needle bed 10F, and the rear needle bed 10 is also referred to as the rear needle bed 10B. Also, of the front and rear carriages 20, the front carriage 20 is also referred to as the front carriage 20F, and the rear carriage 20 is also referred to as the rear carriage 20B.

[0043] A yarn carrier 31 that supplies the knitting yarn Y is supported on the yarn guide rail 30 so as to be movable.

[0044] As shown in Fig. 2, the carriage 20 is provided with a camera 22 and an optical element 23 (for example, a prism or a lens). The camera 22 photographs the knitting needle 11 when the knitting needle 11 is moved to a predetermined position by the cam mechanism 21 and when no knitting yarn is engaged with the knitting needle 11. The optical element 23 guides light from the knitting needle 11 to the camera 22 by refracting or reflecting the light. In addition, an illumination unit 24 is provided at a required position on the carriage 20. The illumination unit 24 includes, for example, a required number of LED elements, and the LED elements used for each needle bed 10 are separated, allowing the illumination conditions (exposure time, current of the LED elements, etc.) to be optimized.

[0045] 3, the front carriage 20F is provided with three cam mechanisms 21 for moving the knitting needles 11 forward and backward. Specifically, a first stitch transfer cam mechanism 21A, a knit cam mechanism 21B, and a second stitch transfer cam mechanism 21C are arranged along the movement direction of the carriage 20 (left and right direction).

[0046] Each cam mechanism 21 can move the knitting needle 11 forward and backward by guiding the butt of the knitting needle 11 selected based on the knitting program along the forward and backward trajectory L. This makes it possible to form a stitch using the knitting yarn Y and to transfer the stitch. The butt is a member that is operated by the cam mechanism 21 when the knitting needle 11 is advanced to the needle gap S.

[0047] As shown by the arrow in Figure 3, when the carriage 20 moves to the right, the knit cam mechanism 21B becomes the leading system and forms a stitch. In this case, the first stitch transfer cam mechanism 21A becomes the trailing system and transfers the stitch formed by the knit cam mechanism 21B. In this case, the second stitch transfer cam mechanism 21C does not transfer a stitch. The same applies when the carriage 20 moves to the left.

[0048] The above-described operation of the cam mechanism 21 is an example, and for example, three cam mechanisms 21 may be used to sequentially perform transfer, stitch formation, and transfer.

[0049] Although the front carriage 20F has been described with reference to FIG. 3, the rear carriage 20B also operates in the same manner.

[0050] 4 is a diagram showing an example of the configuration of the control unit 50. The control unit 50 includes a control unit 51 that controls the entire control unit 50, a communication unit 52, a memory 53, an interface unit 54, a determination unit 55, and a storage unit 56. A database 71, a monitor 72, and a servo motor 73 are connected to the control unit 50.

[0051] The control unit 51 is configured by incorporating a required number of CPUs, MPUs, GPUs, etc. The control unit 51 may also be configured by combining a DSP, FPGA, etc.

[0052] The communication unit 52 includes a communication module and has a function of communicating with an external device (not shown).

[0053] The memory 53 can be configured with a semiconductor memory such as an SRAM, a DRAM, a ROM, or a flash memory.

[0054] The interface unit 54 has an interface function with the monitor 72 and the servo motor 73 .

[0055] The control unit 51 can control the operation of the servo motor 73 via the interface unit 54. The control unit 51 can move the carriage 20 as desired by controlling the operation of the servo motor 73. The control unit 51 can also detect the position of the carriage 20 based on the number of rotations of the servo motor 73. The control unit 51 can also control the operations of the cam mechanism 21, the camera 22, and the lighting unit 24 via the interface unit 54.

[0056] The control unit 51 controls the operation of the servo motor 73, the camera 22, and the lighting unit 24, so that it can photograph the knitting needles 11 and acquire images of the knitting needles 11 between the time when one knitting fabric is finished and the time when the next knitting fabric is started, immediately before the knitting machine is turned off after the required number of garments are knitted, or when the machine operation is temporarily stopped for oiling or cleaning. That is, the control unit 51 functions as an acquisition unit that acquires images of the knitting needles 11. In the case of knitting that places a large burden on the knitting needles 11, the predetermined number may be small, and in the case of knitting that places a small burden on the knitting needles 11, the predetermined number may be large. When photographing the knitting needles 11, it is sufficient to displace the knitting needles 11 by the cam mechanism 21 into the advanced state and to set the knitting yarn in a state where it is not engaged with the knitting needles 11.

[0057] Of the images captured by the camera 22, only images of parts that may be defective may be cut out, and the cut-out images may be output to the control unit 50. This makes it possible to prevent a decrease in the accuracy of determining whether the knitting needles 11 are good or bad due to the influence of yarn debris (fluff, fiber waste, etc.).

[0058] The storage unit 56 can be configured, for example, with a hard disk or a semiconductor memory, and can store a computer program (program product) 60, a knitting program 61, knitting machine adjustment data 62, a first learning model 63, a second learning model 64, a third learning model 65, a template image 66, and required information. The first learning model 63, the second learning model 64, and the third learning model 65 include a model before learning, a model after learning, and a model after relearning.

[0059] The computer program 60 is loaded into the memory 53 and executed by the control unit 51. The control unit 51 can execute the processing defined by the computer program 60. In other words, the processing by the control unit 51 is also the processing by the computer program 60.

[0060] The computer program 60 may be downloaded from an external device via the communication unit 52 and stored in the storage unit 56. Alternatively, the computer program 60 may be recorded on a recording medium M (e.g., an optically readable disk storage medium such as a CD-ROM) and read by a recording medium reading unit (not shown) and stored in the storage unit 56. The computer program 60 may be deployed to be executed on a single computer, or on multiple computers located at one site, or distributed across multiple sites and interconnected by a communications network.

[0061] The determination unit 55, the knitting program 61, the knitting machine adjustment data 62, the first learning model 63, the second learning model 64, the third learning model 65, and the template image 66 will be described later. Note that instead of storing the first learning model 63, the second learning model 64, and the third learning model 65 in the control unit 50, the first learning model 63, the second learning model 64, and the third learning model 65 may be provided on a cloud server, and processing using the first learning model 63, the second learning model 64, and the third learning model 65 may be performed on the server, and the processing results may be transmitted to the control unit 50.

[0062] Next, the images of the knitting needles 11 and the defective knitting needles 11 will be described.

[0063] 5A to 5D are diagrams showing an example of the configuration of a knitting needle 11 and an image of a defective knitting needle 11. FIG. 5A shows the configuration of the tip of a latch needle 1 serving as the knitting needle 11. A hook 2 is provided at the tip of the latch needle 1, and a latch 3 opens and closes the hook 2. The latch 3 swings around an axis 4. The latch 3 can swing between a state in which the hook 2 is closed and a state in which the hook 2 is fully opened. When the latch needle 1, which has the needle loop of the stitch locked within the hook 2, moves to the right in FIG. 5A, if the sinker loop of the stitch is pressed down by a movable sinker or the like and the old loop is pressed down by a stitch presser or the like, the knitting yarn of the needle loop moves to the left relative to the latch needle 1. If the latch needle 1 continues to move to the right, the knitting yarn of the needle loop opens the latch 3 and moves toward the needle stem 5. A shoulder 6 with an increased step is provided on the needle stem 5 side, and even if the latch needle 1 continues to move rightward during transfer, the knitting yarn of the needle loop remains at the shoulder 6 and does not move to the base 7 side of the latch needle 1.

[0064] Figure 5B shows the main parts of a compound needle 12 serving as a knitting needle 11. The compound needle 12 includes a needle body 13 and a slider 16. The slider 16 includes two blades 14 and a base 15. The needle body 13 has a hook 13a at the tip of the needle shaft, which is provided with a slider groove 13c. The blades 14 are stacked in the width direction, with at least their lower portions accommodated in the slider groove 13c of the needle body 13. A tang 14a is provided at the front end of the blade 14. The blade 14 and base 15 function as the slider 16 when combined. Figure 5B shows the state in which the hook opening of the hook 13a is closed by the tang 14a. When the needle body 13 advances relatively to the slider 16, the hook opening of the hook 13a opens.

[0065] In this specification, the knitting needle 11 includes the hook 2, 13a, the latch 3, the slider 16 (particularly the blade 14), the needle stem 5, and the like.

[0066] FIG. 5C shows an example of an image of a defective hook. Causes of defective hooks include, for example, a broken (damaged) hook, a bent hook, and a bowed (reverse bowed) hook. A broken hook is a state in which the tip of the hook is missing. A bent hook is a state in which a portion of the hook is curved or bent. A bowed hook is a state in which the tip of the hook is closer to the needle stem than a good hook, and a reverse bowed hook is a state in which the tip of the hook is on the opposite side of the needle stem than a good hook.

[0067] Fig. 5D shows an example of an image of a defective slider. In Fig. 5D, one of the two sliders is partially missing. Note that the image of the defective knitting needle 11 is not limited to the example in Fig. 5.

[0068] Next, a method for determining whether the knitting needle 11 is good or bad will be described.

[0069] Fig. 6 is a diagram showing a first example of quality determination of the knitting needle 11. The control unit 51 can determine whether the acquired knitting needle 11 is quality based on the similarity between the acquired image of the knitting needle 11 (also referred to as "photographed image") and an image of a normal knitting needle 11 (also referred to as "template image"). As shown in Fig. 6, the size (resolution) of the photographed image and the template image is N dots x M dots, and the pixel value of pixel (i, j) of the photographed image is a i, j and the pixel value of pixel (i, j) of the template image is b i, j The captured image can be an image including the entire knitting needle 11. Note that areas that are not subject to quality determination can be deleted in advance from the captured image of the knitting needle 11.

[0070] The similarity can be determined, for example, by cosine similarity. Cosine similarity is the cosine value of the angle between two vectors and can be calculated by dividing the dot product of the two vectors by the magnitude of the two vectors. The pixel values ​​of each pixel in the image can be vectorized. Cosine similarity is normalized to a range of -1 to 1. If the cosine similarity is 1 or close to 1, the knitting needle 11 can be determined to be normal. If the cosine similarity is -1 or close to -1, the knitting needle 11 can be determined to be defective. In other words, if the cosine similarity is below a predetermined threshold, the knitting needle 11 can be determined to be defective. This embodiment eliminates the need to prepare a huge number of images of non-defective and defective products, reducing the effort required for advance preparation and enabling highly accurate identification processing. Note that the similarity is not limited to cosine similarity; other methods, such as Euclidean distance, may also be used.

[0071] Furthermore, the judgment of pass / fail may be made not only by judging whether the knitting needle 11 is normal (good) or defective, but also by judging the deterioration level of the knitting needle 11. The deterioration level can be classified into several levels, such as "considerably deteriorated," "slightly deteriorated," "not very deteriorated," and "not deteriorated." The level may be judged depending on the value of the cosine similarity within the range of -1 to 1.

[0072] An ID is assigned to each photographed image of all knitting needles 11 to be judged in association with the position of the knitting needle 11, and the position of the defective knitting needle 11 can be identified by the ID of the photographed image judged to be defective.

[0073] The template image 66 can be stored in advance in the storage unit 56 for each type of knitting needle 11 (hook, latch, slider, needle stem, etc.). Furthermore, the operating time of the flatbed knitting machine 100 may be divided into multiple sections, and a template image 66 may be prepared for each section. Since the condition of the knitting needles 11 gradually deteriorates as the operating time of the flatbed knitting machine 100 increases, the accuracy of the pass / fail judgment can be improved by using a template image 66 that reflects the deteriorated state.

[0074] Furthermore, a plurality of template images 66 may be prepared corresponding to the length of time the knitting needles 11 have been used. The length of time the knitting needles 11 have been used since the start of use of a new knitting needle 11 in the flatbed knitting machine 11 or the start of use after the knitting needles 11 have been replaced is stored for each knitting needle 11. When the control unit 51 determines whether the knitting needles 11 are good or bad, it can identify a template image 66 corresponding to the use time from among the plurality of template images 66 according to the use time of the knitting needle 11 at the time of determination, and use the identified template image 66 to perform the quality determination.

[0075] 7 is a diagram showing a second example of determining whether the knitting needle 11 is good or bad. When the control unit 51 receives an image of the knitting needle 11 that is determined to be normal, the control unit 51 inputs the acquired image of the knitting needle 11 to a first learning model 63 that has been trained to reconstruct an image of the knitting needle 11, outputs the reconstructed image, and can determine whether the knitting needle 11 is good or bad by comparing the acquired image of the knitting needle 11 with the reconstructed image output by the first learning model 63. In other words, the control unit 51 functions as a determination unit that determines whether the knitting needle 11 is good or bad using the first learning model 63.

[0076] The first learning model 63 can be, for example, an autoencoder and includes an encoder 631 and a decoder 632. The encoder 631 extracts a feature vector from an input image, and the decoder 632 restores (reconstructs) the original image based on the extracted feature vector. Note that the first learning model 63 is not limited to an autoencoder, and other models such as a Conditional GAN ​​may also be used.

[0077] That is, the first learning model 63 is generated (trained) so that when an image of a normal knitting needle 11 is input, the first learning model 63 reconstructs the image. As shown in FIG. 7 , the acquired image (photographed image) of the knitting needle 11 is input to the first learning model 63. The input image of the knitting needle 11 is compared with the reconstructed image output by the first learning model 63. The comparison can be performed by calculating the difference between the input image of the knitting needle 11 and the reconstructed image. The difference can be determined by calculating the difference in pixel value for each corresponding pixel between the image of the knitting needle 11 and the reconstructed image, and then summing the calculated differences for each pixel. If the difference is small, the knitting needle 11 can be determined to be normal, and if the difference is large or small, the knitting needle 11 can be determined to be defective. According to this embodiment, even if it is difficult to collect images of defective knitting needles 11 as training data, the quality of the knitting needle 11 can be determined by using images of normal knitting needles 11 as training data. The difference may also be calculated as follows. That is, the difference in pixel value for each corresponding pixel between the image of the knitting needle 11 and the reconstructed image may be compared with a predetermined threshold, and the number of pixels having a difference equal to or greater than the threshold may be determined as the difference.

[0078] The image of the knitting needle 11 input to the first learning model 63 may be an image including one entire knitting needle 11, or may be a divided image obtained by dividing the image into parts of the knitting needle 11 (such as the hook, slider, and latch). Areas not subject to pass / fail judgment can be deleted in advance from the image of the knitting needle 11. When an image including the entire knitting needle 11 is used, the pass / fail judgment for each area can be made by calculating the difference for each of multiple areas of the entire image, and the location of the abnormality in the knitting needle 11 can be determined by identifying the defective area. In the case of a divided image, the difference between the divided image and the reconstructed image corresponding to the divided image can be calculated, and the pass / fail judgment for each divided image can be made.

[0079] Furthermore, the quality determination may be made by determining whether the knitting needle 11 is normal (good) or defective, or by determining the deterioration level of the knitting needle 11. The deterioration level can be classified into several levels, such as "considerably deteriorated," "slightly deteriorated," "not very deteriorated," and "not deteriorated." The level can be determined according to the value of the difference. This makes it possible to display, like a heat map, which parts of the knitting needle 11 have deteriorated and which parts have not.

[0080] The first learning model 63 can be generated (trained) as follows: Images of normal knitting needles 11 are collected as training data. In this case, training data for images of normal knitting needles 11 may be collected by changing the photographing conditions, the distance to the knitting needles 11, the position, etc. Based on the collected training data, the images of normal knitting needles 11 are input to the first learning model 63, and the parameters of the first learning model 63 are adjusted so that the reconstructed images output by the first learning model 63 approach the input images of the knitting needles 11.

[0081] 8 is a diagram showing a third example of determining whether a knitting needle 11 is good or bad. The control unit 51 can input the acquired image of the knitting needle 11 (photographed image) to a second learning model 64 that has been trained to output information regarding the quality of the knitting needle 11 when an image of the knitting needle 11 is input, and output information regarding the quality of the knitting needle 11. In other words, the control unit 51 has a function as an output unit that outputs information regarding the quality of the knitting needle 11 using the second learning model 64. The image of the knitting needle 11 input to the second learning model 64 may be an image including the entire knitting needle 11, or may be a divided image obtained by dividing the image into parts of the knitting needle 11 (hook, slider, latch, etc.). Note that areas that are not subject to quality determination can be deleted in advance from the photographed image of the knitting needle 11.

[0082] The second learning model 64 may use, for example, a convolutional neural network (CNN). The information regarding the quality of the knitting needle 11 includes, for example, the probability (accuracy) that the knitting needle 11 is defective and the probability (accuracy) that the knitting needle 11 is normal. For example, if the threshold for determining that the knitting needle 11 is normal is 90%, and the output of the second learning model 64 indicates that the probability that the knitting needle 11 is normal is 80% and the probability that the knitting needle 11 is defective is 20%, the knitting needle 11 can be determined to be defective. Furthermore, if the output of the second learning model 64 indicates that the probability that the knitting needle 11 is normal is 95% and the probability that the knitting needle 11 is defective is 5%, the knitting needle 11 can be determined to be normal. Note that the threshold may be set to a different value depending on the type of knitting needle 11 (hook, latch, slider, needle stem, etc.). According to this embodiment, the accuracy of determining the quality of the knitting needle 11 can be improved by appropriately setting the threshold.

[0083] Furthermore, the judgment of pass / fail may be made not only by judging whether the knitting needle 11 is normal (good) or defective, but also by judging the deterioration level of the knitting needle 11. The deterioration level can be classified into several levels, such as "considerably deteriorated," "slightly deteriorated," "not very deteriorated," and "not deteriorated." The level may be judged according to the value of the probability of being defective.

[0084] The second learning model 64 can be generated (trained) as follows. Images of normal knitting needles 11, images of defective knitting needles 11, a "normal" label associated with the image of the normal knitting needles 11, and a "defective" label associated with the image of the defective knitting needles 11 are collected as training data. Based on the collected training data, the image of the normal knitting needle 11 is input to the second learning model 64, and the parameters of the second learning model 64 are adjusted so that the label output by the second learning model 64 approaches the "normal" label. Furthermore, based on the collected training data, an image of a defective knitting needle 11 is input to the second learning model 64, and the parameters of the second learning model 64 are adjusted so that the label output by the second learning model 64 approaches the "defective" label. In addition, by using training data that associates images of defective knitting needles 11 with "defective" labels that indicate the cause of the defect in the knitting needles 11 (for example, bent, bowed or broken hooks, bent or missing sliders, bent or missing latches, etc.), the second learning model 64 can be configured to output the cause of the defect in the knitting needles 11 when an image of the knitting needle 11 is input.

[0085] Next, a method for determining the defective portion and the cause of the defect of the knitting needle 11 determined to be defective will be described.

[0086] 9A and 9B are diagrams showing an example of a method for determining the defective portion and cause of the defect of a defective knitting needle 11. The control unit 51 can input the acquired image of the knitting needle 11 into a third learning model 65 that has been trained to output the defective portion and cause of the defect of the knitting needle 11 when an image of the knitting needle 11 is input, and output the defective portion and cause of the defect of the knitting needle 11. In other words, the control unit 51 functions as an output unit that outputs the defective portion and cause of the defect of the knitting needle 11 using the third learning model 65. The image of the knitting needle 11 input into the third learning model 65 may be an image including the entire knitting needle 11, or may be a divided image obtained by dividing the image into parts of the knitting needle 11 (hook, slider, latch, etc.). Note that areas not subject to pass / fail judgment can be deleted in advance from the image of the knitting needle 11.

[0087] The third learning model 65 may use, for example, YOLO (You Only Look Once), SSD (Single Shot Multibox Detector), or Faster R-CNN. As shown in FIG. 9A , the third learning model 65 includes an input layer 651, a convolutional layer / pooling layer 652, and an output layer 653. The third learning model 65 extracts features from an input image (captured image) of the knitting needle 11. As shown in FIG. 9B , the third learning model 65 represents, for example, a defective portion of the knitting needle 11 using a bounding box and estimates the probability of the cause of the defect. The cause with the highest probability can be determined to be the cause of the defect (for example, a bent, bowed, or broken hook, a bent or missing slider, a bent or missing latch, etc.). Note that the third learning model 65 is not limited to YOLO.

[0088] According to the above-described configuration, the defective part of the defective knitting needle 11 and the cause of the defect can be provided to the user.

[0089] The third learning model 65 can be generated (trained) as follows. Training data that associates images of defective knitting needles 11, defective locations on the defective knitting needles 11, and "cause" labels indicating the causes of the defects is collected in advance. Based on the collected training data, images of the defective knitting needles 11 are input to the third learning model 65, and the parameters of the third learning model 65 are adjusted so that the defective locations and labels output by the third learning model 65 approach the defective locations and "cause" labels of the defective knitting needles 11 in the training data. Furthermore, training data including images of normal knitting needles 11 is collected in advance, and images of the normal knitting needles 11 are input to the third learning model 65, and the parameters of the third learning model 65 are adjusted so that the third learning model 65 does not output defective locations.

[0090] Next, a method for providing the user with the judgment result of whether the knitting needle 11 is good or bad will be described.

[0091] FIG. 10 is a diagram showing a first example of the quality determination result for the knitting needles 11. The quality determination result screen 200 can be displayed on the monitor 72. Images of the knitting needles 11 and quality determination, etc., are automatically performed between the end of knitting a knitted fabric and the start of knitting the next fabric, immediately before the knitting machine is turned off after the required number of garments are knitted, or when the machine operation is temporarily stopped for lubrication or cleaning, and the quality determination result screen 200 is displayed on the monitor 72. That is, the control unit 51 functions as an output unit that outputs at least one of the positions of knitting needles 11 determined to be defective and information related to the defects. The quality determination result screen 200 displays the detection results. The detection results include, for example, the total number of defective needles among all knitting needles 11 determined to be defective, the number of defective needles by bed (front upper, front lower, back upper, back lower) in a model equipped with four needle beds, the positions of the defective needles, and the deterioration level of the defective needles. The positions of the defective needles by bed are displayed by selecting the bed selection tab 204. In the example of Figure 10, the front upper bed is selected, and the number of the defective needle (needle number) is displayed. The deterioration level of the defective needle can be classified into four levels, for example, from level 1 to 4, where level 1 is "not deteriorated," level 2 is "not deteriorated much," level 3 is "slightly deteriorated," and level 4 is "severely deteriorated." Level 4 is a deterioration level that requires replacement, and level 3 can be a deterioration level that requires attention. Although not shown, the accuracy (probability) of the judgment of each defective needle may also be displayed. For example, the defect probability of the XXth knitting needle may be 00%. Furthermore, the position of the defective needle can include information on which needle bed the needle belongs to in addition to information on the number of the defective needle.

[0092] The quality judgment result screen 200 displays a deterioration status field 206. The deterioration status field 206 displays a selection field for selecting a defective needle, and the deterioration level of the selected defective needle at the time of the previous judgment and the deterioration level at the time of the current judgment. This allows the user to grasp the degree of deterioration at each quality judgment.

[0093] The pass / fail judgment result screen 200 can display images of normal needles (good products) and images of needles judged to be defective so that they can be compared. Also, the pass / fail judgment result screen 200 displays a remedy column 207, which can, for example, urge the user to replace a defective knitting needle 11 that needs to be replaced. Note that the remedy column 207 may display a remedy for a knitting needle 11 judged to be defective based on a database 71 that associates the cause of the defect of the knitting needle 11 with a remedy.

[0094] The threshold is a threshold for determining whether the knitting needle 11 is good or bad, and a threshold can be set for each of the hook, slider, and latch. When the cosine similarity illustrated in FIG. 6 is used to determine whether the knitting needle 11 is good or bad, the determination is made based on whether the cosine similarity is equal to or greater than a threshold. When the first learning model 63 illustrated in FIG. 7 is used to determine whether the knitting needle 11 is good or bad, the determination is made based on whether the difference is equal to or greater than a threshold. When the second learning model 64 illustrated in FIG. 8 is used to determine whether the knitting needle 11 is good or bad, the determination is made based on whether the probability is equal to or greater than a threshold. The "re-determine" icon 202 is an icon for confirming the input threshold.

[0095] By operating the "start" icon 201, the quality of the knitting needles 11 can be manually determined. The "marking" icon 203 is an icon for performing dummy knitting and changing the height of needles determined to be defective. Dummy knitting is knitting in which the knitting needles 11 are raised and lowered without resting yarn from the yarn feeder. The knitting needles 11 are photographed during the dummy knitting process to determine their quality. Images of needles determined to be defective are displayed, allowing the user to determine their quality. The user can check the images and operate the confirmation icon 205 of the needle (needle number) determined to be defective. After checking all suspicious needle images, the user can operate the "marking" icon 203 to move the carriage and displace the knitting needles 11 determined to be defective to a height different from the normal height. Note that the configuration is not limited to manually displacing the height of knitting needles 11 determined to be defective to a height different from the normal height based on the operation of the "marking" icon 203, and the configuration may also be such that the height of knitting needles 11 determined to be defective is automatically displaced to a height different from the normal height.

[0096] As described above, the control unit 51 can output information relating to defects in the knitting needles 11 to the monitor 72. This makes it possible to provide the user with information relating to defects in the knitting needles 11 (for example, the accuracy (probability) of the defect, the cause of the defect, etc.).

[0097] Furthermore, the control unit 51 can output at least one of an image of the knitting needle 11 determined to be defective and information for specifying the position of the knitting needle 11 determined to be defective to the monitor 72. This makes it possible to provide the user with at least one of an image of the knitting needle 11 determined to be defective and information for specifying the position of the knitting needle 11 determined to be defective.

[0098] Furthermore, the control unit 51 can displace (move) the knitting needles 11 determined to be defective to a height different from the height of the knitting needles 11 according to the knitting program. This makes it possible to clearly notify the user of the defective knitting needles 11 among the knitting needles 11 arranged in a row on the needle bed 10.

[0099] FIG. 11 is a diagram showing a second example of the quality judgment result for the knitting needles 11. The detection results on the quality judgment result screen 210 include the position of the defective needle for each bed (upper front, lower front, upper back, lower back), the cause of the defect, etc. In the example of FIG. 11, the upper front bed is selected, and the number of the defective needle (needle number) and the cause of the defect of the defective needle are displayed. For example, it can be seen that the cause of the defect of the XXth needle is a bent hook, and the cause of the defect of the XOth needle is a missing slider. The user can display an image of the defective needle by operating the selection icon 215 for the knitting needle 11 that has been judged to be defective.

[0100] The pass / fail judgment result screen 210 can be displayed using the results of processing using the third learning model 65 illustrated in FIG.

[0101] On the pass / fail judgment result screen 210, the user (the customer using the flat knitting machine 100) can input whether the detection result is correct or not in response to the detection result by the control unit 50. As shown in Fig. 11 , the user can correct the detection result by inputting or selecting whether the position of the defective knitting needle 11 is correct or not and whether the cause of the defect is correct or not in the pass / fail reception field, and operating the "correction" icon 211.

[0102] As described above, the control unit 51 outputs the position of the defective knitting needle 11 and the cause of the defect to the monitor 72 of the customer's flat knitting machine, receives information on the success or failure of the output position of the knitting needle 11 and the cause of the defect, and associates the received information on success or failure with an image of the knitting needle 11 determined to be defective, and stores the result as training data in the memory unit 56 or the database 71.

[0103] This allows the position of the defective knitting needle 11 and the cause of the defect to be provided to the customer, allowing the customer to judge whether the provided information is correct or not, and allows the provided information to be corrected as necessary. Information such as the position of the defective knitting needle 11 and the cause of the defect, which has been judged correct or not by the customer, can be collected as training data for the third learning model 65 used in the flat knitting machine operated by the customer.

[0104] The control unit 51 can identify the third learning model 65 that has been re-trained using the training data for each customer collected as described above, as the model for that customer's flat knitting machine. The identified third learning model 65 can replace the third learning model 65 already installed in that customer's flat knitting machine. This makes it possible to use a customer-specific third learning model 65 that has been tuned for the customer's flat knitting machine, even if it is the same third learning model 65.

[0105] The database 71 stores information that associates the causes of defects in the knitting needles 11 with countermeasures.

[0106] Fig. 12 is a diagram showing an example of information relating the cause of a defect in the knitting needle 11 to a countermeasure. As shown in Fig. 12, to address the bending of the hook, separate knitting can basically be performed, but it is also possible to use countermeasures such as changing the stitch size (yarn length). For example, by knitting with a larger stitch size, it is possible to prevent the hook from bending due to a certain degree of pulling.

[0107] Figures 13A to 13F are diagrams showing an example of separate knitting. Separate knitting will be explained with reference to Figure 13. Figures 13A to 13C show an example without separate knitting, and Figures 13D to 13F show an example with separate knitting. As shown in Figures 13A and 13B, when looking at loop numbers #2 and #3, #2 and #3 are directly connected by one needle pitch. As shown in Figure 13C, in the process of knitting a cable pattern in which the positions of pairs #1, #2 and #3, #4 are interchanged, #2 and #3 are pulled diagonally by three needle pitches. This may cause the hooks to bend.

[0108] Figures 13D to 13F show the case of separate knitting in which one course of separate knitting is inserted. Focusing on loop numbers #2, #3, and #5 as shown in Figures 13D and 13E, #2 and #5 are directly connected at three diagonal pitches, and #3 is directly connected to #2A and #4 formed in the knitting course immediately before #2. As shown in Figure 13F, in the process of knitting the knitting yarn, the front and rear needle beds move relative to each other by the distance of three needles, causing #2 and #3 to be pulled apart diagonally, but because it is #2A that is directly connected to #3, the burden on the needles is alleviated, and it is possible to prevent or suppress the occurrence of bent hooks.

[0109] Returning to Fig. 12, the output of the servo motor 73 for driving the needle bed can be adjusted to prevent the hook from bowing. One possible cause of the hook bowing is that when racking (relative movement of the needle bed) is performed with an extreme load on the knitting needle 11, the position of the needle bed deviates from the target position due to the pull of the knitted fabric, causing the hook to collide with the yarn feeder that is stopped when the needle advances, or with the opposing knitting needle 11 when transferring the stitch. To avoid this, the output of the servo motor 73 is adjusted so that it does not succumb to the lateral pull (external force) of the knitting fabric.

[0110] In response to damage to the slider or latch, the pulling-down force of the knitted fabric can be adjusted. If the adjustment of the device that pulls down the knitted fabric is not appropriate, an excessive load will be applied to the knitting needle 11, causing damage to the slider or latch. Note that the information relating the cause of the defect in the knitting needle 11 and the countermeasures is only an example and is not limited to the example in Figure 12.

[0111] As described above, the control unit 51 can output a countermeasure for a knitting needle 11 that has been determined to be defective, based on the database 71 that associates defects (causes of defects) of the knitting needles 11 with countermeasures. The knitting needles 11 that have been determined to be defective and the countermeasures can be output (displayed) on the monitor 72. This makes it possible to provide the user with an appropriate countermeasure for the defective knitting needles 11.

[0112] Next, conditions that may impose a load on the knitting needles 11 that have been determined to be defective are extracted and instructed to the user, and it is possible to leave it up to the user to decide how to deal with them (processing). This point will be described below.

[0113] The information on the knitting needle 11 determined to be defective includes, for example, the position of the defective knitting needle 11 and the cause of the defect. The knitting program 61 contains instructions for instructing which knitting needle 11 to move in each knitting course, and how to move the yarn feeder and carriage. A course refers to one movement of the carriage during knitting. The knitting machine adjustment data 62 includes, for example, settings for a pull-down device that pulls down the knitted fabric (e.g., which area of ​​the knitted fabric to pull down, for how long, and with what force), knitting speed settings (carriage movement speed), and stitch value settings (such as the amount by which the knitting needle 11 that holds the knitting yarn is lowered by a cam to achieve the target stitch size). The history information can be stored in a database 71, and is the knitting program and flatbed knitting machine adjustment data used during the knitting operation collected for each knitting operation.

[0114] FIG. 14 is a diagram showing a schematic diagram of the emulation result of the stitch connection. The aforementioned history information can be used for the emulation. In FIG. 14, the horizontal axis indicates the needle number (position) of the knitting needle 11, and the vertical axis indicates the course (time). Course 1 represents the bottom side (lowest end) of the garment product, and increases in course number lead to the top side (top end). The emulation result shown in FIG. 14 emulates the stitch connection based on the knitting program 61, and it is possible to monitor the state of the knitting needle 11 in each course based on the knitting program 61 and the knitting machine adjustment data 62. Specifically, it is possible to determine the degree of pulling force (burden) applied to the knitting needle 11 depending on the number of yarns held by the knitting needle 11, the needle pitch between the stitches, the size of the stitches, the movement of the knitting needle 11, and the like. A stitch whose pulling force applied to the knitting needle 11 exceeds a predetermined threshold can be determined in advance as a knitting that places a burden on the needle.

[0115] A program capable of tracing the above-mentioned determination method in reverse is prepared. As shown in Fig. 14, when a needle with a certain number is determined to be a defective needle, by inputting the defective needle number into the above-mentioned program, it is possible to identify knitting that places a load on the needle in each course corresponding to the defective needle number. In other words, it is possible to determine how the knitting needle 11 behaves in each course corresponding to the number of the knitting needle 11 determined to be defective, and what the knitting state is, such as the number of yarns held by the knitting needle 11, the needle pitch between stitches, and the size of the stitches, and therefore it is possible to determine the cause of the load on the knitting needle 11. In the example of Fig. 14, it is identified that knitting places a load on the needle in two courses.

[0116] 15 is a diagram showing an example of factors that impose a load on the knitting needle 11. The factors that impose a load on the knitting needle 11 include, for example, racking, stitch size, multiple hooks, a special knitting method, the friction coefficient of the yarn, and the pull-down force of the knitted fabric.

[0117] Regarding racking, the relative movement of the needle beds pulls the yarn passing back and forth. Regarding stitch size, when forming a small stitch size, the sliding resistance increases when the needle passes through the small stitch loop when advancing the needle. Regarding the pulling force of the knitted fabric, if the pulling force is inappropriate, an excessive load is placed on the knitting needle 11. Note that the factors that impose a load on the knitting needle 11 are just examples and are not limited to the example in FIG. 15 .

[0118] The determination unit 55 (control unit 51) acquires history information including the knitting program used for knitting and knitting machine adjustment data, and determines the factors that apply load to the knitting needle 11 that has been determined to be defective based on the information on the knitting needle 11 that has been determined to be defective and the acquired history information, and outputs the determination result to the monitor 72. The information on the knitting needle 11 that has been determined to be defective includes, for example, the position (needle number) of the knitting needle 11.

[0119] Specifically, based on the acquired history information, the control unit 51 specifies the correspondence between the knitting patterns that place a load on the knitting needles 11 in each course of knitting and the knitting needles 11. For the knitting needles 11 that have been determined to be defective, the control unit 51 uses the specified correspondence to specify the knitting patterns that place a load on the defective knitting needles 11, and can determine the factors that impose a load on the knitting needles 11 based on the specified knitting patterns.

[0120] In this way, by linking the knitting program 61 and the knitting machine adjustment data 62 with respect to defects in the knitting needles 11 (position of the knitting needles 11), it is possible to determine the cause of the load on the knitting needles 11. By informing the user of the determination results, the user can take the necessary steps to prevent the knitting needles 11 from being overloaded.

[0121] Fig. 16 is a diagram showing an example of a processing procedure by the control unit 50. The control unit 51 acquires an image of the knitting needle 11 (S11), and judges whether the knitting needle 11 is good or bad based on the acquired image (S12). The judgment of good or bad can be performed using, for example, the methods exemplified in Figs. 6 to 8. The control unit 51 judges whether there is a defective knitting needle 11 (S13), and if there is no defective knitting needle 11 (NO in S13), performs the processing of step S22 described below.

[0122] If there is a defective knitting needle 11 (YES in S13), the control unit 51 determines the position of the defective knitting needle 11 and the cause of the defect (S14). The processing of step S14 can use, for example, the method exemplified in Fig. 9. The control unit 51 specifies a countermeasure for the defective knitting needle 11 based on a database 71 that associates defects in the knitting needles 11 with countermeasures (S15).

[0123] The control unit 51 determines the cause of applying a load to the defective knitting needle 11 based on the knitting program and the knitting machine adjustment data (S16). The processing of step S16 can use, for example, the methods exemplified in Figures 14 and 15. The control unit 51 outputs an image of the defective knitting needle 11, its position, the cause of the defect, and a remedy (S17), and outputs the cause of applying a load to the knitting needle 11 (S18).

[0124] The control unit 51 determines whether or not a request for the determination result has been received (S19), and if a request for the determination result has been received (YES in S19), the request is received (S20), and the corrected determination result is stored in the storage unit 56 (S21). If a request for the determination result has not been received (NO in S19), the control unit 51 performs the process of step S22 described above.

[0125] The control unit 51 determines whether or not to end the process (S22), and if not (NO in S22), continues knitting (S23), and determines whether or not it is time to photograph the knitting needles 11 (S24). The photographing timing can be, for example, the time when a predetermined number of garment products have been produced, but is not limited to this.

[0126] If it is not time to photograph (NO in S24), the control unit 51 continues the processing of step S24. If it is time to photograph (YES in S24), the control unit 51 photographs the knitting needle 11 and performs the processing from step S11 onwards. If it is time to end the processing (YES in S22), the control unit 51 ends the processing.

[0127] As described above, the control unit 51 acquires images of the knitting needles 11 of the flat knitting machine 100 having a plurality of knitting needles 11, judges whether the knitting needles 11 are good or bad based on the acquired images of the knitting needles 11, and outputs at least one of the position of the knitting needles 11 judged to be defective and information related to the defect. The information related to the defect includes, for example, the certainty (probability) of the defect, the cause of the defect, etc. According to the present embodiment, at least one of the position of the knitting needles 11 judged to be defective and information related to the defect can be provided.

[0128] In this embodiment, as a means for notifying the position of a knitting needle 11 determined to be defective, for example, a method of indicating the defective knitting needle 11 with a laser pointer, or a method of marking the defective knitting needle 11 with a required color in an identifiable manner may be used.

[0129] This embodiment can also be applied to determining the quality of knitting members other than the knitting needles 11. The knitting members include, for example, members such as loop pressers or transfer jacks that are arranged in a row in large numbers on the needle bed and are configured to be able to advance and retreat with respect to the needle gap.

[0130] REFERENCE SIGNS LIST 10 Needle bed 11 Knitting needle 20 Carriage 21 Cam mechanism 22 Camera 23 Optical element 24 Illumination unit 50 Control unit 51 Control unit 52 Communication unit 53 Memory 54 Interface unit 55 Determination unit 56 Storage unit 60 Computer program 61 Knitting program 62 Knitting machine adjustment data 63 First learning model 64 Second learning model 65 Third learning model 66 Template image 71 Database 72 Monitor 73 Servo motor 100 Flat knitting machine

Claims

1. Acquire images of knitting members of a flat knitting machine having a plurality of knitting members arranged in a row; determining whether the knitting member is good or bad based on the acquired image of the knitting member; At least one of the position of the knitting member determined to be defective and information regarding the defect is output. A computer program that causes a computer to perform a process.

2. Based on a database that associates defects in knitting components with countermeasures, a countermeasure for knitting components that are determined to be defective is output.

2. The computer program according to claim 1, which causes a computer to execute a process.

3. Acquire history information including a knitting program used for knitting and knitting machine adjustment data; determining a factor that applies a load to the knitting member determined to be defective based on information about the knitting member determined to be defective and the acquired history information; Output the judgment result 2. The computer program according to claim 1, which causes a computer to execute a process.

4. When an image of a knitting member determined to be normal is input, the acquired image of the knitting member is input to a first learning model that has been trained to reconstruct an image of the knitting member, and a reconstructed image is output; The acquired image of the knitting member is compared with the output reconstructed image to determine whether the knitting member is good or bad.

4. The computer program according to claim 1, which causes a computer to execute a process.

5. The quality of the knitting member is determined based on the similarity between the acquired image of the knitting member and an image of a normal knitting member.

4. The computer program according to claim 1, which causes a computer to execute a process.

6. When an image of a knitting member is input, the acquired image of the knitting member is input to a second learning model that has been trained to output information regarding the quality of the knitting member, and information regarding the quality of the knitting member is output.

4. The computer program according to claim 1, which causes a computer to execute a process.

7. The acquired image of the knitting member is input to a third learning model that has been trained to output the defective part and the cause of the defect when an image of the knitting member is input, and the defective part and the cause of the defect of the knitting member are output.

4. The computer program according to claim 1, which causes a computer to execute a process.

8. Outputs information about defects in train components to the monitor 4. The computer program according to claim 1, which causes a computer to execute a process.

9. At least one of an image of the knitting member determined to be defective and information for identifying the position of the knitting member determined to be defective is output to the monitor.

4. The computer program according to claim 1, which causes a computer to execute a process.

10. The knitting member determined to be defective is moved to a height different from the height of the knitting member according to the knitting program.

4. The computer program according to claim 1, which causes a computer to execute a process.

11. The position of the defective knitting member and the cause of the defect are output to the monitor of the customer's flat knitting machine, Accept the output information on the position of the knitting member and the cause of the defect, The received information on success or failure and images of the knitting components determined to be defective are stored as training data.

8. A computer program according to claim 7, which causes a computer to execute a process.

12. The third learning model retrained using the training data is identified for the customer's flat knitting machine.

12. A computer program product according to claim 11, which causes a computer to carry out a process.

13. An acquisition unit that acquires images of knitting members of a flat knitting machine having a plurality of knitting members arranged in a row; a determination unit that determines whether the knitting member is good or bad based on the acquired image of the knitting member; an output unit that outputs at least one of the position of the knitting member determined to be defective and information regarding the defect; Equipped with Flat knitting machine.

14. Acquire images of knitting members of a flat knitting machine having a plurality of knitting members arranged in a row; determining whether the knitting member is good or bad based on the acquired image of the knitting member; At least one of the position of the knitting member determined to be defective and information regarding the defect is output. Information processing methods.