Machine learning device, image forming device, and machine learning method

Abstract

To form an image at an appropriate position.SOLUTION: A machine learning device comprises: a storage unit that stores first data including pitch related information related to pitches of a plurality of print ranges provided on a medium conveyed in a predetermined conveyance direction along the conveyance direction in an image forming apparatus; and a learning model creation unit that performs machine learning processing by using the first data and second data for correcting the pitches of the print ranges, thereby creating a learning model to which the first data is input and from which the second data is output. Of the pitch related information included in the first data, a part of pitch related information is a differential value between information related to the pitch of a first print range and information related to the pitch of a second print range provided downstream in the conveyance direction of the first print range.SELECTED DRAWING: Figure 9

Description

[Technical Field]

[0001] The present invention relates to a machine learning device and a machine learning method for learning the position of a print area on a medium on which a plurality of print areas are provided, and an image forming device for forming an image in the print area using a learning model obtained by the machine learning device and the machine learning method. Place Regarding. [Background technology]

[0002] Some image forming devices are capable of forming images on so-called label roll paper, which is a roll of paper on which multiple labels are arranged. For example, Patent Document 1 discloses an image forming device that adjusts the image formation position on each label. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2018-103426 Summary of the Invention [Problem to be solved by the invention]

[0004] In this way, in an image forming device that forms an image on label roll paper, it is desirable to form the image in accordance with the position of the printing range that is inside the label, and it is expected that the image will be formed in the appropriate position.

[0005] The present invention has been made in consideration of the above points, and provides a machine learning device, an image forming device, and and Machine Learning Methods The law This is what I am trying to propose. [Means for solving the problem]

[0006] The machine learning device of the present invention includes a storage unit that stores first data including pitch-related information related to the pitch of a plurality of printing ranges provided along a predetermined transport direction on a medium transported in the image forming device, and a learning model generation unit that performs machine learning processing using the first data and second data for correcting the pitch of the printing ranges to generate a learning model to which the first data is input and the second data is output, and among the pitch-related information included in the first data, some of the pitch-related information is a difference value between information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction. The second print range is the print range immediately before the first print range. do. The machine learning device of the present invention also includes a memory unit that stores first data including pitch-related information related to the pitch of multiple printing ranges provided along a predetermined transport direction on a medium transported in the image forming device in the transport direction, and a learning model generation unit that performs machine learning processing using the first data and second data for correcting the pitch of the printing ranges to generate a learning model in which the first data is input and the second data is output, and some of the pitch-related information included in the first data is a difference value between information related to the pitch of the first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction, and the first data includes, as the pitch-related information, an estimated pitch difference value that is the difference between the estimated value of the pitch of the first printing range and the estimated value of the pitch of the second printing range. Furthermore, the machine learning device of the present invention includes a memory unit that stores first data including pitch-related information related to the pitch of multiple printing ranges provided along a predetermined transport direction on a medium transported in the image forming device in the transport direction, and a learning model generation unit that performs machine learning processing using the first data and second data for correcting the pitch of the printing ranges to generate a learning model to which the first data is input and the second data is output, wherein some of the pitch-related information included in the first data is a difference value between information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction, and the first data includes, as the pitch-related information, a pitch displacement average difference value that is the difference between the average value of the pitch displacements of multiple printing ranges including a third printing range that is a predetermined number before the first printing range and the average value of the pitch displacements of multiple printing ranges that is a fourth printing range that is the predetermined number before the second printing range, and the pitch displacement is the difference between the detection value of the pitch of the printing range detected using the sensor and the preset setting value of the pitch of the printing range.Furthermore, the machine learning device of the present invention comprises a memory unit that stores first data including pitch-related information related to the pitch of multiple printing ranges provided along a predetermined transport direction on a medium transported in the image forming device in the transport direction, and a learning model generation unit that performs machine learning processing using the first data and second data for correcting the pitch of the printing ranges to generate a learning model to which the first data is input and the second data is output, wherein some of the pitch-related information included in the first data is a differential value between information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction, and the second data is a differential value between the pitch of the first printing range and an estimated value of the pitch of the first printing range.

[0007] The image forming apparatus of the present invention includes a transport unit that transports a medium in a predetermined transport direction, an image forming unit that forms images in a plurality of printing ranges provided on the medium along the transport direction, a storage unit that stores a learning model that receives first data including pitch-related information related to the pitch of the printing ranges and outputs second data for correcting the pitch of the printing ranges, a calculation unit that uses the learning model to generate the second data based on the first data, and an adjustment unit that adjusts the image formation position by the image forming unit in the printing ranges based on the calculation result of the calculation unit, and some of the pitch-related information included in the first data is a difference value between information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction. The first data includes, as the pitch-related information, an estimated pitch difference value that is the difference between the estimated value of the pitch in the first printing range and the estimated value of the pitch in the second printing range. do. The image forming apparatus of the present invention includes a transport unit that transports a medium in a predetermined transport direction, an image forming unit that forms images in a plurality of printing ranges provided on the medium along the transport direction, a storage unit that stores a learning model that receives first data including pitch-related information related to the pitch of the printing ranges and outputs second data for correcting the pitch of the printing ranges, a calculation unit that uses the learning model to generate the second data based on the first data, and an adjustment unit that adjusts the image formation position by the image forming unit in the printing ranges based on the calculation results of the calculation unit, and the pitch-related information included in the first data includes information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction. and a sensor for detecting the pitch of the printing range, wherein the first data includes, as the pitch-related information, a pitch displacement average difference value which is the difference between the average value of pitch displacements of a plurality of printing ranges including a third printing range that is a predetermined number before the first printing range and the average value of pitch displacements of a plurality of printing ranges including a fourth printing range that is the predetermined number before the second printing range, and the pitch displacement is the difference between the detected value of the pitch of the printing range detected using the sensor and a preset setting value of the pitch of the printing range, and the first data includes, as the pitch-related information, a detected pitch difference value which is the difference between the detected value of each pitch of the plurality of printing ranges including the third printing range and the detected value of each pitch of the plurality of printing ranges including the fourth printing range.

[0008] The machine learning method of the present invention includes storing first data in a memory unit, the first data including pitch-related information related to the pitch of multiple printing ranges provided along a predetermined transport direction on a medium transported in the image forming device in the transport direction, and performing machine learning processing using the first data and second data for correcting the pitch of the printing ranges, thereby generating a learning model in which the first data is input and the second data is output, wherein some of the pitch-related information included in the first data is a differential value between information related to the pitch of the first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction. [Effects of the Invention]

[0010] According to the machine learning device and the machine learning method of the present invention, an image can be formed at an appropriate position. Place This allows the image to be formed at an appropriate position. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is an explanatory diagram illustrating an example of a configuration of an image forming apparatus according to an embodiment of the present invention. [Figure 2] FIG. 2 is an explanatory diagram illustrating an example of the configuration of the recording medium shown in FIG. [Figure 3] FIG. 2 is an explanatory diagram illustrating an example of the configuration of the image forming unit illustrated in FIG. [Figure 4] 2 is a block diagram illustrating an example of a control system of the image forming apparatus illustrated in FIG. 1. FIG. [Figure 5] 5 is a block diagram illustrating an example of a block related to a label pitch estimation process in the image forming apparatus shown in FIG. 4. FIG. [Figure 6] FIG. 10 is an explanatory diagram illustrating an example of a label pitch estimation process. [Figure 7] 6 is a block diagram illustrating an example of the configuration of a preprocessing unit illustrated in FIG. 5. [Figure 8] 5 is a block diagram illustrating an example of the configuration of a machine learning device that generates the correction value generation model shown in FIG. 4. [Figure 9] 9 is a block diagram illustrating an example of a block related to machine learning processing in the machine learning device shown in FIG. 8. FIG. [Figure 10] 10 is an explanatory diagram illustrating an example of the configuration of a neural network in the learning model generation unit shown in FIG. 9. FIG. [Figure 11] 6 is a flowchart illustrating an example of an operation of the image forming apparatus illustrated in FIG. 5. [Figure 12] 10 is a graph showing the transition of the image writing start position between different image forming apparatuses and explanatory variables input to a correction value generation model. [Figure 13] 10 is a graph showing that the image forming apparatus according to the embodiment of the present invention can accurately estimate the image writing start position. [Figure 14] 9 is a flowchart illustrating an example of an operation of the machine learning device shown in FIG. 8. [Figure 15] FIG. 10 is a block diagram illustrating an example of a control system of an image forming apparatus according to another embodiment. [Figure 16] FIG. 16 is a block diagram relating to a label pitch estimation process in the image forming apparatus shown in FIG. [Figure 17] FIG. 14 is a block diagram illustrating an example of the configuration of a machine learning device that generates the label pitch generation model shown in FIG. 13. [Figure 18] 18 is a block diagram illustrating an example of a block related to machine learning processing in the machine learning device shown in FIG. 17. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[0013] [1. Image forming device] 1 shows an example of the configuration of an image forming apparatus 1 according to an embodiment of the present invention. The image forming apparatus 1 is configured to function as a printer that forms an image on a recording medium 9, such as label roll paper, using an electrophotographic method.

[0014] FIG. 2 shows an example configuration of a recording medium 9. The recording medium 9 has multiple labels L and a backing sheet 9B. The labels L are temporarily attached to the backing sheet 9B and can be peeled off and attached to various objects. The multiple labels L are arranged side by side on the label surface 9C of the recording medium 9 at a pitch (hereinafter referred to as the label pitch) LP in the longitudinal direction of the recording medium 9 (the conveying direction F1 of the recording medium 9). Specifically, on the recording medium 9, labels L having a label length LL are arranged side by side at a label interval LS. In this example, the label pitch LP is defined based on the leading edge of the label L (the right end of the label L in FIG. 2) when the image forming apparatus 1 pulls out the recording medium 9 from the roll and conveys it in the conveying direction F1 (to the right in FIG. 2).

[0015] The image forming apparatus 1 forms images on these labels L. That is, the printing range (LL in FIG. 2) of the recording medium 9 is the label L that can be peeled off from the backing paper 9B. When forming an image on the label L, the image forming apparatus 1 estimates the position of the label L (called the label LB to be printed (see FIG. 6)) that is positioned K labels (for example, 7 labels) upstream of the reference label LA in the conveying direction F1 based on the label pitch LP of the most recently detected label L (called the reference label LA (see FIG. 6)) and the label pitches LP of multiple labels L that are positioned downstream of the reference label LA in the conveying direction F1 and that have been detected in the past, as will be described later, and forms an image in accordance with the position of the label LB to be printed.

[0016] That is, while it is desirable to arrange the labels L at equal intervals on the recording medium 9, for example, the position of the labels L may be shifted depending on the manufacturing method, for example. One manufacturing method is the so-called flat pressure method. In this flat pressure method, the labels L are formed by sequentially pressing a flat die having blades on its surface that correspond to the contours of the multiple labels L against the recording medium 9 on which the labels L have not yet been formed. In this case, multiple labels L are formed as a set, so, for example, the variation in the label pitch LP within this set is small, and the variation in the label pitch LP between sets is greater than the variation in the label pitch LP within the set. As such, when the label pitch LP of the recording medium 9 varies, it is difficult to form an image at the appropriate position on the label LB to be printed.

[0017] Furthermore, for example, there may be cases where the conveyance speed of the recording medium 9 fluctuates over time in the image forming apparatus 1. In this case as well, it is difficult to form an image at an appropriate position on the label LB to be printed.

[0018] The image forming apparatus 1 estimates the position of the label LB to be printed based on the label pitch LP of the most recently detected label L (reference label LA) and the label pitches LP of multiple labels L that are located downstream of the reference label LA in the conveying direction F1 and that have been detected in the past. This makes it possible for the image forming apparatus 1 to reduce the impact on the image formation position on the label LB to be printed of fluctuations in the manufacturing method of the recording medium 9 and the conveying speed of the recording medium 9, and to form an image in an appropriate position on the label LB to be printed.

[0019] 1, the image forming apparatus 1 includes an image forming unit 10, an LED (Light Emitting Diode) head 14, a primary transfer roller 21, an intermediate transfer belt 22, a drive roller 23, an idle roller 24, a tension roller 25, a backup roller 26, and a secondary transfer roller 27. These components constitute an image forming section in the image forming apparatus 1 that forms an image on a recording medium 9.

[0020] The image forming unit 10 is configured to form a toner image. Note that while the image forming apparatus 1 shown in FIG. 1 has a monochrome printing configuration having one set of image forming unit 10 and primary transfer roller 21, the image forming apparatus 1 may also have a color printing configuration by providing multiple sets of image forming units 10 and primary transfer rollers 21. In this case, for example, four image forming units 10 that form black, cyan, magenta, and yellow toner images, respectively, and four corresponding primary transfer rollers 21 may be arranged along the conveyance direction F2 of the intermediate transfer belt 22.

[0021] 3 is a simplified diagram of an example configuration of the image forming unit 10. The image forming unit 10 includes a photoconductor 11, a cleaning blade 12, a charging roller 13, a developing roller 15, a developing blade 16, a supply roller 17, and a toner storage unit 18.

[0022] The photoreceptor 11 is configured to carry an electrostatic latent image on its surface (surface layer portion). The photoreceptor 11 rotates counterclockwise in FIG. 3 by power transmitted from a photoreceptor motor (not shown). The photoreceptor 11 is charged by the charging roller 13 and exposed by the LED head 14. As a result, an electrostatic latent image is formed on the surface of the photoreceptor 11. Then, as toner is supplied by the developing roller 15, a toner image corresponding to the electrostatic latent image is formed (developed) on the photoreceptor 11.

[0023] The cleaning blade 12 is configured to scrape off and clean the toner remaining on the surface (surface layer portion) of the photoreceptor 11. The cleaning blade 12 is positioned so that its tip abuts against the surface of the photoreceptor 11. The cleaning blade 12 scrapes off, for example, toner that remains on the surface of the photoreceptor 11 without being transferred. The scraped off toner is stored in a waste toner box (not shown).

[0024] The charging roller 13 is configured to charge the surface (surface portion) of the photoreceptor 11 approximately uniformly. The charging roller 13 is arranged so as to be in contact with the surface (circumferential surface) of the photoreceptor 11 and to be pressed against the photoreceptor 11 with a predetermined pressure. The charging roller 13 rotates clockwise in FIG. 3 in accordance with the rotation of the photoreceptor 11. A charging voltage is applied to the charging roller 13 by an image formation control unit 53 (described later).

[0025] The LED head 14 is configured to irradiate light onto the photoconductor 11. The LED head 14 has, for example, a plurality of light-emitting diodes arranged side by side in the main scanning line direction (the depth direction in FIG. 3), and uses these light-emitting diodes to irradiate light onto the photoconductor 11 in dot units. As a result, an electrostatic latent image is formed on the surface of the photoconductor 11.

[0026] The developing roller 15 is configured to carry toner on its surface. The developing roller 15 is disposed so as to be in contact with the surface (circumferential surface) of the photosensitive member 11 and is disposed so as to be pressed against the photosensitive member 11 with a predetermined pressure. The developing roller 15 rotates clockwise in FIG. 3 by power transmitted from a photosensitive member motor (not shown). A developing voltage is applied to the developing roller 15 by an image formation control unit 53 (described later).

[0027] Developing blade 16 is configured to contact the surface of developing roller 15 to form a layer of toner (toner layer) on the surface of developing roller 15 and to regulate (control, adjust) the thickness of the toner layer. Developing blade 16 can be, for example, a plate-like elastic member made of stainless steel or the like bent into an L-shape. Developing blade 16 is positioned so that the bent portion contacts the surface of developing roller 15 and is pressed against developing roller 15 with a predetermined pressure.

[0028] The supply roller 17 is configured to supply the toner contained in the toner container 18 to the developing roller 15. The supply roller 17 is disposed so as to be in contact with the surface (circumferential surface) of the developing roller 15 and is disposed so as to be pressed against the developing roller 15 with a predetermined pressure. The supply roller 17 rotates clockwise in FIG. 3 by power transmitted from a photosensitive motor (not shown). As a result, in the image forming unit 10, friction occurs between the surface of the supply roller 17 and the surface of the developing roller 15. As a result, in the image forming unit 10, the toner is charged by so-called frictional charging. A supply voltage is applied to the supply roller 17 by an image formation control unit 53 (described later).

[0029] The toner storage section 18 is configured to store toner. Specifically, the toner storage section 18 of the image forming unit 10 stores, for example, black toner.

[0030] With this configuration, in the image forming unit 10, the photoconductor 11 is charged by the charging roller 13 and exposed by the LED head 14. As a result, an electrostatic latent image is formed on the surface of the photoconductor 11. Furthermore, the toner contained in the toner container 18 is charged by the supply roller 17 and the developing roller 15 and supplied to the photoconductor 11. As a result, a toner image corresponding to the electrostatic latent image is formed (developed) on the photoconductor 11.

[0031] Returning to FIG. 1, primary transfer roller 21 is configured to electrostatically transfer the toner images formed by image forming units 10 onto the transfer surface of intermediate transfer belt 22. Primary transfer roller 21 is disposed opposite photoconductor 11 of image forming unit 10 across intermediate transfer belt 22. Primary transfer roller 21 is disposed so as to be pressed against photoconductor 11 with a predetermined pressure. A primary transfer voltage is applied to primary transfer roller 21 by image formation control unit 53 (described later). As a result, in image forming apparatus 1, toner images formed by image forming units 10 are transferred (primary transfer) onto the transfer surface of intermediate transfer belt 22.

[0032] The intermediate transfer belt 22 is a circular elastic belt, and is configured to be stretched (suspended) by a drive roller 23, an idle roller 24, a tension roller 25, and a backup roller 26. The intermediate transfer belt 22 is circulated and transported in a transport direction F2 in accordance with the rotation of the drive roller 23. At that time, the intermediate transfer belt 22 passes between the photosensitive member 11 of the image forming unit 10 and the primary transfer roller 21. With this configuration, the intermediate transfer belt 22 supplies the toner image transferred onto the transfer surface by the primary transfer to a secondary transfer section 28 composed of the backup roller 26 and the secondary transfer roller 27.

[0033] The drive roller 23 is configured to circulate and transport the intermediate transfer belt 22. Specifically, the drive roller 23 is disposed upstream of the image forming unit 10 in the transport direction F2, and rotates clockwise in FIG. 1 by power transmitted from a belt motor (not shown). As a result, the drive roller 23 circulates and transports the intermediate transfer belt 22 in the transport direction F2.

[0034] 1 in accordance with the circular transport of the intermediate transfer belt 22. The idle roller 24 is disposed downstream of the image forming unit 10 in the transport direction F2.

[0035] 1 in accordance with the circular transport of the intermediate transfer belt 22. The tension roller 25 is disposed between the drive roller 23 and the backup roller .

[0036] 1 in accordance with the circulating transport of the intermediate transfer belt 22. The backup roller 26 is disposed opposite the secondary transfer roller 27, with the transport path 8 that transports the recording medium 9 and the intermediate transfer belt 22 sandwiched between them. The backup roller 26 and the secondary transfer roller 27 form a secondary transfer unit 28.

[0037] The secondary transfer roller 27 is configured to transfer the toner image on the transfer surface of the intermediate transfer belt 22 onto the transfer surface (label surface 9C) of the recording medium 9. A secondary transfer voltage is applied to the secondary transfer roller 27 by an image formation control unit 53 (described later). As a result, in the image forming apparatus 1, the toner image on the transfer surface of the intermediate transfer belt 22 is transferred (secondary transfer) onto the transfer surface of the recording medium 9.

[0038] The image forming apparatus 1 also has a medium supply roller 31, a first label sensor 32, a second label sensor 33, a cutter 34, a plurality of conveying rollers 35, and a fixing unit 40. These are arranged along a conveying path 8 for the recording medium 9. The recording medium 9 is conveyed along the conveying path 8 by being guided by a medium guide 36 that forms the conveying path 8.

[0039] The medium supply roller 31 is composed of a pair of rollers arranged on either side of the conveying path 8, and is configured to pull out the recording medium 9 from a roll on which the recording medium 9 is wound and convey the recording medium 9 along the conveying path 8.

[0040] The first label sensor 32 is disposed between the medium supply roller 31 and the cutter 34 and is configured to detect the label L on the recording medium 9 supplied by the medium supply roller 31. The first label sensor 32 is configured, for example, using a transmissive optical sensor. Specifically, the first label sensor 32 has a light-emitting unit and a light-receiving unit disposed on either side of the conveyance path 8. The optical path from the light-emitting unit to the light-receiving unit crosses the conveyance path 8 for the recording medium 9. In the first label sensor 32, light of a predetermined intensity emitted from the light-emitting unit is transmitted through the recording medium 9 and received by the light-receiving unit. Therefore, the intensity of the light received by the light-receiving unit varies depending on whether or not the label L is present on the optical path. The first label sensor 32 outputs a detection signal corresponding to the intensity of the light received by the light-receiving unit. The image forming apparatus 1 detects the leading edge of the label L based on the detection signal from the first label sensor 32 and determines the cutting position of the recording medium 9 by the cutter 34 based on the detection result.

[0041] The cutter 34 is configured to cut the recording medium 9. The cutter 34 cuts the recording medium 9 at a cutting position determined by the image forming apparatus 1 based on the detection signal of the first label sensor 32.

[0042] Each of the plurality of conveying rollers 35 is composed of a pair of rollers arranged on either side of the conveying path 8, and is arranged at a predetermined position. The plurality of conveying rollers 35 convey the recording medium 9 along the conveying path 8 in the conveying direction F1.

[0043] The second label sensor 33 is disposed upstream of the secondary transfer unit 28 in the transport direction F1 and in the vicinity of the secondary transfer unit 28. Like the first label sensor 32, the second label sensor 33 is configured to detect the label L on the recording medium 9. That is, the second label sensor 33 is configured using, for example, a transmissive optical sensor. The image forming apparatus 1 detects the leading edge of the label L based on the detection signal of the second label sensor 33. The image forming apparatus 1 then detects the label pitch LP based on the detection result of the leading edge of each label L, and estimates the label pitch LP of the label LB to be printed based on the label pitch LP of the multiple labels L. The image forming apparatus 1 then determines the timing (write timing) at which the LED head 14 (FIG. 3) starts operating to form an image on the label LB to be printed, based on the estimated label pitch LP.

[0044] This allows the image forming apparatus 1 to match the timing at which the label LB to be printed reaches the secondary transfer unit 28 with the timing at which the toner image on the intermediate transfer belt 22 reaches the secondary transfer unit 28. As a result, the secondary transfer unit 28 is able to transfer the toner image to an appropriate position on the label LB to be printed.

[0045] After the toner image is transferred onto the recording medium 9 in the secondary transfer section 28, the recording medium 9 is supplied to the fixing section 40.

[0046] The fixing unit 40 is configured to fix the toner image transferred onto the recording medium 9 to the recording medium 9 by applying heat and pressure to the recording medium 9. The fixing unit 40 has a fixing roller 41 and a pressure roller 42. The fixing roller 41 is configured to include a heater, such as a halogen heater, inside it and to apply heat to the toner on the recording medium 9. The pressure roller 42 is disposed so that a pressure contact portion is formed between it and the fixing roller 41 and is configured to apply pressure to the toner on the recording medium 9. As a result, the toner on the recording medium 9 is heated, melted, and pressurized in the fixing unit 40. As a result, the toner image is fixed onto the recording medium 9.

[0047] Then, the recording medium 9 on which the toner image has been fixed by the fixing unit 40 is discharged from the image forming apparatus 1. The recording medium 9 discharged from the image forming apparatus 1 is wound up, for example, by a rewinder (not shown) using power supplied from a motor (not shown).

[0048] With this configuration, the image forming apparatus 1 is capable of forming images on each label L of the recording medium 9 successively.

[0049] 4 shows an example of the configuration of a control system in the image forming apparatus 1. The image forming apparatus 1 includes a communication unit 51, a display operation unit 52, an image formation control unit 53, a processing unit 60, and a storage unit 70.

[0050] The communication unit 51 is configured to communicate using, for example, a Universal Serial Bus (USB) or a Local Area Network (LAN), and is configured to receive print data including various print setting data and image data sent from, for example, a host computer. The print setting data includes, for example, information about the setting value of the label pitch LP and the number of labels L to be printed.

[0051] The display operation unit 52 is configured to accept user operations and display the operating status of the image forming apparatus 1, and is configured using, for example, a touch panel, various buttons, a liquid crystal display, and various indicators.

[0052] The image formation control unit 53 is configured to control the image forming operation in the image forming apparatus 1 based on instructions from the processing unit 60. Specifically, the image formation control unit 53 controls, for example, the exposure operation of the LED head 14 (FIG. 3). The image formation control unit 53 also controls the operation of various power sources to control the generation operation of charging voltage, developing voltage, supply voltage, primary transfer voltage, and secondary transfer voltage. The image formation control unit 53 also controls the operation of various motors to control the transport operation of the recording medium 9, the formation operation of a toner image in the image forming unit 10, and the transport operation of the intermediate transfer belt 22. The image formation control unit 53 also controls the fixing temperature by controlling the current flowing through the heater of the fixing unit 40.

[0053] The processing unit 60 is configured to control the operation of each block of the image forming apparatus 1, thereby controlling the operation of the image forming apparatus 1. The processing unit 60 is configured using, for example, a processor capable of executing programs, a RAM (Random Access Memory) capable of temporarily storing data, etc. The processing unit 60 has a sensor value acquisition unit 61, a label pitch detection unit 62, a label pitch displacement detection unit 63, a label pitch estimation unit 64, a label pitch correction unit 65, and a write timing calculation unit 66.

[0054] Fig. 5 shows an example of the label pitch detection unit 62, the label pitch variation detection unit 63, the label pitch estimation unit 64, and the label pitch correction unit 65. Fig. 6 shows an example of the label L related to the label pitch LP processed by the label pitch detection unit 62, the label pitch variation detection unit 63, the label pitch estimation unit 64, and the label pitch correction unit 65. In Fig. 6, for example, L i denotes the i-th label L, and L i-N+1 denotes the (i-N+1)th label L, and Li-M denotes the (iM)th label L, and L i+K indicates the (i+K)th label L. These numbers indicate the order in which images are formed in the image forming apparatus 1. The i-th label L i is the reference label LA, which is the most recently detected label L, and the (i+K)th label L i+K is the label LB to be printed. In this embodiment, for the sake of simplicity, it is assumed that the number of the label L matches the page number. In other words, the ith label L i is the label L of the ith page i is.

[0055] The sensor value acquisition unit 61 (FIG. 4) is configured to acquire the sensor value indicated by the detection signal of the second label sensor 33.

[0056] The label pitch detection unit 62 (FIGS. 4 and 5) is configured to detect the label pitch LP of each label L based on the sensor value acquired by the sensor value acquisition unit 61. The label pitch detection unit 62 then updates the label pitch history data 71 stored in the storage unit 70 based on the detected label pitch LP. As shown in FIG. 6, the label pitch history data 71 is data indicating the label pitch LP of N (e.g., seven) labels L including the latest detected label L (reference label LA).

[0057] The label pitch variation detection unit 63 is configured to detect, based on the label pitch LP detected by the label pitch detection unit 62, a label pitch variation LPD that is the difference between the detected value of the label pitch LP and the set value of the label pitch LP included in the print data. The label pitch variation detection unit 63 then updates the variation history data 73 stored in the storage unit 70 based on the detected label pitch variation LPD. As shown in Fig. 6, the variation history data 73 is data that indicates a plurality of label pitch variations LPD of a plurality of labels L, from the label L that is M labels (e.g., 50 labels) before the most recently detected label L (reference label LA) to this reference label LA.

[0058] Furthermore, the label pitch displacement detection unit 63 calculates a label pitch displacement average value LPDA, which is the average value of the label pitch displacements LPD of N (e.g., seven) labels L including the latest detected label L (reference label LA), based on the displacement history data 73, and stores this calculation result in the storage unit 70 as displacement average data 74. Furthermore, the label pitch displacement detection unit 63 calculates, based on the displacement history data 73, the difference between the label pitch displacement LPD of the latest detected label L (reference label LA) and the label pitch displacement LPD of the label L that is M (e.g., 50) labels before the reference label LA, and stores this calculation result in the storage unit 70 as displacement fluctuation data 75.

[0059] The label pitch estimation unit 64 is configured to calculate an estimated label pitch LP0 by estimating the label pitch LP of the labels LB to be printed based on the label pitch history data 71 stored in the storage unit 70, and to store this estimated label pitch LP0 in the storage unit 70 as label pitch estimation data 72. For example, if the recording medium 9 is manufactured using a flat pressing method, periodicity may occur in the label pitch LP. The label pitch estimation unit 64 is configured to estimate the label pitch LP of the labels LB to be printed by analyzing this periodicity of the label pitch LP.

[0060] The label pitch correction unit 65 is configured to correct the estimated label pitch LP0 calculated by the label pitch estimation unit 64. As shown in FIG. 5, the label pitch correction unit 65 has a preprocessing unit 65A, a correction value generation unit 65B, and a correction processing unit 65C.

[0061] As shown in detail in FIG. 7, the pre-processing unit 65A associates the estimated label pitch LP0 indicated by the label pitch history data 71 with the page number n of the current (i.e., latest) label LB to be printed (i.e., the estimated label pitch LP0 nThe pre-processing unit 65A also associates the label pitch displacement average value LPDA indicated by the displacement average data 74 with the page number n of the current label LB to be printed (i.e., the label pitch displacement average value LPDA n ) and stores it in the storage unit 70 as pre-processing history data 76. In this way, the pre-processing unit 65A calculates the current and past estimated label pitches LP0 (LP0 n , LP0 n-1 , LP0 n-2 , ...) and the current and past calculated label pitch displacement average value LPDA (LPDA n , L.P.D.A. n-1 , L.P.D.A. n-2 , ...) are stored in the storage unit 70 as pre-processing history data 76. The pre-processing history data 76 includes at least the estimated label pitch LP0 of the current label LB to be printed. n and the estimated label pitch LP0 of the label L one page before (i.e., one before) the current label LB to be printed. n-1 and the current label pitch displacement average value LPDA n and the average label pitch displacement value LPDA of the previous page (i.e., the previous page) n-1 It is sufficient if the following is included.

[0062] The pre-processing unit 65A calculates, from the pre-processing history data 76, the estimated label pitch LP0 of the label L one page before the current label LB to be printed (i.e., the label L adjacent to the current label LB to be printed downstream in the conveyance direction). n-1 Get the current estimated label pitch LP0 n Estimated label pitch LP0 from the previous page n-1 The value obtained by subtracting the above (this is called an estimated LP difference value) is input to the correction value generating unit 65B.

[0063] The pre-processing unit 65A also estimates the label pitch LP of the label L one page before the current label LB from the pre-processing history data 76. n-1Get the current label pitch displacement average value LPDA n Label pitch displacement average value LPDA from one page before n-1 The value obtained by subtracting the current label pitch displacement average value LPDA (this value is called the LP displacement average difference value) is input to the correction value generation unit 65B. n is the average value of the label pitch displacement LPD of N labels L including the reference label LA K labels before the label LB to be printed, and the average label pitch displacement LPDA n-1 is the average value of the label pitch displacements LPD of N labels L including the label L one page before the reference label LA. The estimated LP difference value and the LP displacement average difference value calculated by the preprocessing unit 65A may be stored in the storage unit 70, and the correction value generation unit 65B may acquire these from the storage unit 70.

[0064] The correction value generation unit 65B (FIG. 5) is configured to use a correction value generation model M obtained by machine learning processing to generate data indicating a correction value ΔLP for the estimated label pitch LP0 indicated by the label pitch estimation data 72, based on the label pitch history data 71 and displacement fluctuation data 75 as pitch-related information stored in the storage unit 70, and the estimated LP difference value and LP displacement average difference value calculated by the pre-processing unit 65A. The correction processing unit 65C is configured to generate an estimated label pitch LP1 by correcting the estimated label pitch LP0 based on this correction value ΔLP. Specifically, as shown in FIG. 6, the correction processing unit 65C is configured to generate the estimated label pitch LP1 by adding the estimated label pitch LP0 and the correction value ΔLP.

[0065] In this way, in the image forming device 1, the label pitch estimation unit 64 generates an estimated label pitch LP0 by estimating the label pitch LP of the label LB to be printed, and the label pitch correction unit 65 corrects this estimated label pitch LP0 to generate an estimated label pitch LP1.

[0066] The write timing calculation unit 66 (FIG. 4) is configured to calculate the timing (write timing) at which the LED head 14 (FIG. 3) starts operating to form an image on the label LB to be printed, based on the estimated label pitch LP1 generated by the label pitch correction unit 65.

[0067] The storage unit 70 (FIGS. 4 and 5) is configured using, for example, a volatile memory, a non-volatile memory, a hard disk drive, etc., and is configured to store various programs and various setting data. The storage unit 70 stores label pitch history data 71, label pitch estimation data 72, displacement history data 73, displacement average data 74, displacement fluctuation data 75, preprocessing history data 76, and a correction value generation model M.

[0068] As described above, the label pitch history data 71 is data indicating the label pitches LP of N (for example, seven) labels L including the latest detected label L (reference label LA).

[0069] As described above, the label pitch estimation data 72 is data indicating the label pitch LP (estimated label pitch LP0) of the label LB to be printed, estimated by the label pitch estimation unit 64 based on the label pitches LP of N (e.g., seven) labels L included in the label pitch history data 71.

[0070] As described above, the displacement history data 73 is data indicating multiple label pitch displacements LPD of multiple labels L from the label L that is M (e.g., 50) before the latest detected label L (reference label LA) to this reference label LA.

[0071] As described above, the displacement average data 74 is data indicating the average value (label pitch displacement average value LPDA) of the label pitch displacements LPD of N (e.g., seven) labels L including the latest detected label L (reference label LA).

[0072] As described above, the displacement fluctuation data 75 is data that indicates the difference between the label pitch displacement LPD of the latest detected label L (reference label LA) and the label pitch displacement LPD of the label L that is M labels (50 labels in this example) before this reference label LA.

[0073] As described above, the pre-processing history data 76 includes the estimated label pitch LP0 (LP0 n , LP0 n-1 , LP0 n-2 , ...) and the current and past calculated label pitch displacement average value LPDA (LPDA n , L.P.D.A. n-1 , L.P.D.A. n-2 , ...).

[0074] The correction value generation model M is a learning model generated in advance by performing machine learning processing, and is stored in the storage unit 70 of the image forming apparatus 1. The correction value generation model M receives label pitch history data 71, displacement fluctuation data 75, an estimated LP difference value, and an LP displacement average difference value as input, and outputs data indicating a correction value ΔLP for the estimated value of the label pitch LP (estimated label pitch LP0) indicated by the label pitch estimation data 72.

[0075] [2. Machine Learning Device] FIG. 8 shows an example configuration of a machine learning device 200 that generates a correction value generation model M. The machine learning device 200 is, for example, a personal computer. Note that the machine learning device 200 is not limited to this, and instead, for example, the processing unit 60 in the image forming apparatus 1 may perform the machine learning processing. A display 201, a keyboard 202, and a mouse 203 are connected to the machine learning device 200. The display 201 is configured to display an image based on an image signal supplied from the machine learning device 200. The keyboard 202 and mouse 203 are used by the user to input information.

[0076] The machine learning device 200 includes a processing unit 210, a storage unit 220, a memory 230, an interface 240, a display interface 250, and a communication unit 260. The processing unit 210, the storage unit 220, the memory 230, the interface 240, the display interface 250, and the communication unit 260 are connected to a bus 290.

[0077] The processing unit 210 is configured using, for example, a CPU (Central Processing Unit). The processing unit 210 has a dataset acquisition unit 211, a teacher data generation unit 212, and a learning model generation unit 213.

[0078] FIG. 9 shows an example of the data set acquisition unit 211, the teacher data generation unit 212, and the learning model generation unit 213.

[0079] The dataset acquisition unit 211 is configured to acquire a dataset DS, which is a set of data required for machine learning processing. The dataset acquisition unit 211 then stores the acquired dataset DS in the storage unit 220. The dataset DS includes label pitch history data 221, label pitch estimation data 222, estimated LP difference value data 223, LP displacement average difference value data 224, displacement fluctuation data 225, and label pitch data 226.

[0080] The label pitch history data 221, like the label pitch history data 71 (FIGS. 4 and 5), is data indicating the label pitches LP of N (for example, seven) labels L including a certain label L (reference label LA).

[0081] The label pitch estimation data 222, like the label pitch estimation data 72 (FIGS. 4 and 5), is data indicating the label pitch LP (estimated label pitch LP0) of a label L that is positioned K labels (e.g., 7 labels) upstream of the reference label LA in the conveying direction F1, estimated based on the label pitches LP of N labels (e.g., 7 labels) included in the label pitch history data 221.

[0082] The estimated LP difference value data 223 is data indicating the estimated LP difference value calculated by the pre-processing unit 65A (FIG. 7). That is, the estimated LP difference value data 223 is the estimated label pitch LP0 (LP0 n ) to the estimated label pitch LP0 (LP0) one page before the estimated label pitch LP0 n-1 ) is the data showing the value obtained by subtracting

[0083] The LP displacement average difference value data 224 is data indicating the LP displacement average difference value calculated by the pre-processing unit 65A (FIG. 7). That is, the LP displacement average difference value data 224 is n The average label pitch displacement LPDA calculated when n From the estimated label pitch LP0 of the previous page n-1 The average label pitch displacement LPDA calculated when n-1 This is data showing the value obtained by subtracting .

[0084] Similar to the displacement fluctuation data 75 (Figures 4 and 5), the displacement fluctuation data 225 is data indicating the difference between the label pitch displacement LPD of a certain label L (reference label LA) and the label pitch displacement LPD of the label L that is M labels (e.g., 50 labels) before this reference label LA.

[0085] The label pitch data 226 is data indicating the label pitch LP of the label L arranged K labels (for example, 7 labels) upstream of the reference label LA in the conveying direction F1.

[0086] The dataset acquisition unit 211 can acquire a dataset DS including the label pitch history data 221, label pitch estimation data 222, estimated LP difference value data 223, LP displacement average difference value data 224, displacement fluctuation data 225, and label pitch data 226 from, for example, the image forming apparatus 1. That is, the image forming apparatus 1 can acquire multiple datasets DS by, for example, sequentially detecting labels L while conveying the recording medium 9 without forming an image on the recording medium 9. Specifically, each time the image forming apparatus 1 sequentially detects a label L, it sets the most recently detected label L as a reference label LA and acquires the dataset DS. The dataset acquisition unit 211 acquires multiple datasets DS obtained by the image forming apparatus 1 in this manner. The dataset acquisition unit 211 then stores the acquired multiple datasets DS in the storage unit 220.

[0087] The teacher data generation unit 212 is configured to generate a correction value ΔLP as teacher data DT based on the difference between the label pitch LP indicated by the label pitch data 226 and the estimated label pitch LP0 indicated by the label pitch estimation data 222, both of which are included in the data set DS stored in the storage unit 220. Specifically, the teacher data generation unit 212 is configured to generate the correction value ΔLP by subtracting the estimated label pitch LP0 indicated by the label pitch estimation data 222 from the label pitch LP indicated by the label pitch data 226.

[0088] The learning model generation unit 213 is configured to generate a correction value generation model M by performing machine learning processing based on the label pitch history data 221, estimated LP difference value data 223, LP displacement average difference value data 224, displacement fluctuation data 225 stored in the storage unit 220, and the teacher data DT (correction value ΔLP) generated by the teacher data generation unit 212. In this example, the learning model generation unit 213 generates the correction value generation model M by performing so-called supervised learning using a neural network model. Then, the learning model generation unit 213 stores the generated correction value generation model M in the storage unit 220.

[0089] FIG. 10 shows an example of a neural network model for supervised learning performed in the learning model generation unit 213. The neural network NN in the neural network model includes k neurons x (neurons x1 to x2) in an input layer LX. k ) and m neurons y1 (neurons y11 to y1 m ) and n neurons y2 (neurons y21 to y2 n ) and one neuron z (neuron z1) in the output layer LZ. In this example, two hidden layers are provided, but this is not limited to this, and three or more hidden layers may be provided, or one hidden layer may be provided.

[0090] Nodes connecting neurons are provided between the input layer LX and the first hidden layer LY1, between the first hidden layer LY1 and the second hidden layer LY2, and between the second hidden layer LY2 and the output layer LZ, and each node is assigned a weight wj (j is a natural number).

[0091] Below, we will explain the process of generating a correction value generation model M by performing machine learning processing based on the label pitch history data 221, estimated LP difference value data 223, LP displacement average difference value data 224, displacement fluctuation data 225 stored in the memory unit 220, and the correction value ΔLP supplied by the teacher data generation unit 212.

[0092] The learning model generation unit 213 learns the correlation between the label pitch history data 221, estimated LP difference value data 223, LP displacement average difference value data 224, and displacement fluctuation data 225 stored in the storage unit 220 and the correction value ΔLP supplied by the training data generation unit 212. Specifically, the learning model generation unit 213 calculates the value of a neuron z in the output layer LZ by associating the label pitch history data 221, estimated LP difference value data 223, LP displacement average difference value data 224, and displacement fluctuation data 225 with multiple neurons x in the input layer LX. First, the learning model generation unit 213 calculates the values ​​of m neurons y1 in the first hidden layer LY1 based on the values ​​of k neurons x in the input layer LX. Specifically, the learning model generation unit 213 calculates the value of each neuron y1 in the first hidden layer LY1 by performing weighted addition using weights Wi associated with each node based on the values ​​of k neurons x in the input layer LX connected to this neuron y1. Similarly, the learning model generation unit 213 calculates the values ​​of n neurons y2 in the second intermediate layer LY2 based on the values ​​of m neurons y1 in the first intermediate layer LY1, and calculates the value of neuron z in the output layer LZ based on the values ​​of n neurons y2 in the second intermediate layer LY2.

[0093] The learning model generation unit 213 then compares the calculated value of neuron z1 in the output layer LZ with the value of data t1 included in the training data DT to determine an error. Here, the value of neuron z1 is a correction value ΔLP calculated based on label pitch history data 221, estimated LP difference value data 223, LP displacement average difference value data 224, and displacement fluctuation data 225, and the value of data t1 is the correction value ΔLP in the training data DT. The learning model generation unit 213 then iteratively adjusts the weight wi associated with each node (backpropagation) to reduce the determined error.

[0094] Then, when the above-described series of steps are repeated a predetermined number of times, or until a predetermined condition, such as the above-described error being smaller than a permissible value, is satisfied, the learning model generation unit 213 terminates the learning and stores the neural network model in the storage unit 220 as the correction value generation model M. In this way, the learning model generation unit 213 generates the correction value generation model M including information on all weights wj associated with each of the nodes of the neural network model.

[0095] The storage unit 220 ( FIG. 8 ) is configured using, for example, a hard disk drive (HDD) or a solid state drive (SSD), and is configured to store various data used in the machine learning device 200. The storage unit 220 stores a dataset DS and a correction value generation model M. The dataset DS includes label pitch history data 221, label pitch estimation data 222, estimated LP difference value data 223, LP displacement average difference value data 224, displacement variation data 225, and label pitch data 226, and is acquired by the dataset acquisition unit 211 and stored in the storage unit 220. The correction value generation model M is generated by the learning model generation unit 213 and stored in the storage unit 220.

[0096] Memory 230 is configured to temporarily store data when processing unit 210 performs processing. Interface 240 is an interface for connecting external devices to machine learning device 200, and is connected to, for example, keyboard 202 and mouse 203. Display interface 250 is an interface for connecting display 201 to machine learning device 200, and is configured to supply image signals to display 201. Communication unit 260 is configured to perform communication using, for example, a LAN (Local Area Network).

[0097] [3. Operation] Next, the operations of the image forming device 1 and the machine learning device 200 according to this embodiment will be described.

[0098] [3-1. Overview of overall operation] First, an overview of the overall operation of image forming apparatus 1 will be described with reference to Figures 1, 3, and 4. In image forming apparatus 1 (Figure 1), medium supply roller 31 pulls out recording medium 9 from a roll on which the recording medium 9 is wound, and conveys the recording medium 9 along conveyance path 8. Conveyance roller 35 conveys recording medium 9 along conveyance path 8. Second label sensor 33 detects label L of recording medium 9 conveyed along conveyance path 8.

[0099] The sensor value acquisition unit 61 of the processing unit 60 acquires the sensor value indicated by the detection signal of the second label sensor 33. The label pitch detection unit 62 detects the label pitch LP of each label L based on the sensor value acquired by the sensor value acquisition unit 61. Then, the label pitch detection unit 62 updates the label pitch history data 71 stored in the storage unit 70 based on the detected label pitch LP.

[0100] Based on the label pitch LP detected by the label pitch detection unit 62, the label pitch variation detection unit 63 detects a label pitch variation LPD, which is the difference between the detected value of the label pitch LP and the set value of the label pitch LP included in the print data. The label pitch variation detection unit 63 then updates the variation history data 73 stored in the storage unit 70 based on the detected label pitch variation LPD. The label pitch variation detection unit 63 also calculates a label pitch variation average value LPDA, which is the average value of the label pitch variations LPD of N labels L (seven in this example) including the most recently detected label L (reference label LA), based on the variation history data 73, and stores this calculation result in the storage unit 70 as variation average data 74. The label pitch variation detection unit 63 also calculates, based on the variation history data 73, the difference between the label pitch variation LPD of the most recently detected label L (reference label LA) and the label pitch variation LPD of the label L that is M labels (50 labels in this example) before the reference label LA, and stores this calculation result in the storage unit 70 as variation fluctuation data 75.

[0101] The label pitch estimation unit 64 calculates an estimated label pitch LP0 by estimating the label pitch LP of the label LB to be printed based on the label pitch history data 71 stored in the memory unit 70, and stores this estimated label pitch LP0 in the memory unit 70 as label pitch estimation data 72.

[0102] The label pitch correction unit 65 is configured to correct the estimated label pitch LP0 calculated by the label pitch estimation unit 64. That is, a pre-processing unit 65A of the label pitch correction unit 65 corrects the estimated label pitch LP0 of the latest label LB to be printed. n Estimated label pitch LP0 from the previous page n-1 The pre-processing unit 65A calculates an estimated LP difference value, which is the value obtained by subtracting the latest label pitch displacement average value LPDA n From the estimated label pitch LP0 of the previous page n Average label pitch displacement when LPDA is estimated n-1 The LP displacement average difference value is calculated by subtracting

[0103] The correction value generation unit 65B uses a correction value generation model M obtained by machine learning processing to generate a correction value ΔLP for the estimated label pitch LP0 indicated by the label pitch estimation data 72, based on the label pitch history data 71 and displacement fluctuation data 75 stored in the storage unit 70, and the estimated LP difference value and LP displacement average difference value calculated by the preprocessing unit 65A. The correction processing unit 65C of the label pitch correction unit 65 generates an estimated label pitch LP1 by correcting the estimated label pitch LP0 based on this correction value ΔLP.

[0104] The write timing calculation unit 66 calculates the timing (write timing) at which the LED head 14 starts operating to form an image on the label LB to be printed, based on the estimated label pitch LP1 generated by the label pitch correction unit 65.

[0105] The image formation control unit 53 starts the operation of the LED head 14 of the image forming unit 10 at a timing corresponding to the write timing calculated by the write timing calculation unit 66. This causes the image forming unit 10 to form a toner image. The primary transfer roller 21 then transfers (primary transfer) the toner image on the photoreceptor 11 of the image forming unit 10 onto the transfer surface of the intermediate transfer belt 22. The toner image on the intermediate transfer belt 22 reaches the secondary transfer unit 28 as the intermediate transfer belt 22 is circulated and conveyed. The secondary transfer unit 28 transfers (secondary transfer) the toner image on the transfer surface of the intermediate transfer belt 22 onto the transfer surface (label surface 9C) of the recording medium 9. The fixing unit 40 fixes the toner image onto the recording medium 9. The recording medium 9 with the toner image fixed by the fixing unit 40 is then discharged from the image forming apparatus 1.

[0106] Next, an overview of the overall operation of the machine learning device 200 will be described with reference to FIGS. 8 and 9. The dataset acquisition unit 211 acquires a dataset DS (label pitch history data 221, label pitch estimation data 222, estimated LP difference value data 223, LP displacement average difference value data 224, displacement variation data 225, and label pitch data 226) which is a set of data required for machine learning processing. The teacher data generation unit 212 generates a correction value ΔLP as teacher data DT based on the difference between the label pitch LP indicated by the label pitch data 226 and the estimated label pitch LP0 indicated by the label pitch estimation data 222, both of which are included in the dataset DS. The learning model generation unit 213 generates a correction value generation model M by performing machine learning processing based on the label pitch history data 221, estimated LP difference value data 223, LP displacement average difference value data 224, displacement variation data 225 stored in the storage unit 220, and the teacher data DT (correction value ΔLP) generated by the teacher data generation unit 212.

[0107] [3-2. Detailed operation of image forming device] The image forming apparatus 1 estimates the position of the label L (label LB to be printed) that is positioned K labels (for example, 7 labels) upstream of the reference label LA in the conveying direction F1 based on the label pitch LP of the most recently detected label L (reference label LA) and the label pitches LP of multiple labels L that are positioned downstream of the reference label LA in the conveying direction F1 and that have been detected in the past, and forms an image in accordance with the position of the label LB to be printed. The following describes the detailed operation of the image forming apparatus 1.

[0108] 11 is a flowchart showing an example of the operation of the image forming apparatus 1. Every time the sensor value acquisition unit 61 acquires a sensor value indicated by the detection signal of the second label sensor 33, the processing unit 60 performs the following process.

[0109] First, the label pitch detection unit 62 checks whether the label pitch LP has been detected (step S101). If the label pitch detection unit 62 has not detected the label pitch LP ("NO" in step S101), this step S101 is repeated until the label pitch LP is detected.

[0110] In step S101, if the label pitch detection unit 62 detects the label pitch LP ("YES" in step S101), the label pitch detection unit 62 updates the label pitch history data 71 stored in the memory unit 70 based on the detected label pitch LP (step S102).

[0111] Next, based on the detected label pitch LP, the label pitch variation detection unit 63 detects the label pitch variation LPD, which is the difference between the detected value of the label pitch LP and the set value of the label pitch LP included in the print data (step S103).

[0112] Next, the label pitch variation detection unit 63 updates the variation history data 73 stored in the storage unit 70 based on the detected label pitch variation LPD (step S104).

[0113] Next, the label pitch variation detection unit 63 generates variation average data 74 by calculating the average value of the label pitch variations LPD of N (e.g., seven) labels L including the latest detected label L (reference label LA) based on the variation history data 73 (step S105). Then, the label pitch variation detection unit 63 stores the generated variation average data 74 in the storage unit 70.

[0114] Next, the label pitch variation detection unit 63 generates variation variation data 75 by calculating the difference between the label pitch variation LPD of the latest detected label L (reference label LA) and the label pitch variation LPD of the label L M labels (50 labels in this example) before this reference label LA based on the variation history data 73 (step S106). Then, the label pitch variation detection unit 63 stores the generated variation variation data 75 in the storage unit 70.

[0115] Next, the label pitch estimation unit 64 generates label pitch estimation data 72 by estimating the label pitch LP (estimated label pitch LP0) of the label LB to be printed based on the label pitch history data 71 (step S107).The label pitch estimation unit 64 then stores the generated label pitch estimation data 72 in the storage unit 70.

[0116] Next, the pre-processing unit 65A of the label pitch correction unit 65 associates the estimated label pitch LP0 indicated by the label pitch estimation data 72 with the page number n of the label LB to be printed (i.e., the estimated label pitch LP0 n The pre-processing unit 65A also associates the label pitch displacement average value LPDA indicated by the displacement average data 74 with the page number n of the label LB to be printed (i.e., the label pitch displacement average value LPDA n The preprocessing unit 65A then calculates the estimated label pitch LP0 of the previous page from the preprocessing history data 76. n-1 Get the latest estimated label pitch LP0 n Estimated label pitch LP0 from the previous pagen-1 The pre-processing unit 65A also calculates an estimated LP difference value by subtracting the average label pitch displacement value LPDA of the previous page from the pre-processing history data 76. n-1 Get the latest label pitch displacement average value LPDA n Label pitch displacement average value LPDA from one page before n-1 The LP displacement average difference value is calculated by subtracting (step S108).

[0117] Next, the correction value generation unit 65B of the label pitch correction unit 65 uses the correction value generation model M stored in the memory unit 70 to generate data indicating a correction value ΔLP for the estimated label pitch LP0 indicated by the label pitch estimation data 72, based on the label pitch history data 71 and displacement fluctuation data 75 stored in the memory unit 70, and the estimated LP difference value and LP displacement average difference value calculated by the pre-processing unit 65A (step S109).

[0118] Next, the correction processing unit 65C of the label pitch correction unit 65 generates an estimated label pitch LP1 by correcting the estimated label pitch LP0 based on the correction value ΔLP (step S110).

[0119] Then, the write timing calculation unit 66 calculates the write timing for forming an image on the label LB to be printed based on the estimated label pitch LP1 (step S111). This is the end of this flow.

[0120] [3-3. Reasons for using differential values] Here, we will explain why, when correcting the estimated label pitch LP0 using the correction value generation model M obtained by machine learning, we use the difference values ​​from the values ​​of the previous page (estimated LP difference value and LP displacement average difference value) as explanatory variables instead of using the estimated label pitch LP0 and the label pitch displacement average value LPDA as they are.

[0121] As described in the applicant's Patent Application No. 2020-160691, even when the estimated label pitch LP0 and the label pitch displacement average value LPDA are used as is, the position of the label LB to be printed (i.e., the position where the image starts to be written) can be estimated with higher accuracy compared to when the estimated label pitch LP0 is not corrected (i.e., when learning is not applied).

[0122] However, if the estimated label pitch LP0 and the label pitch displacement average value LPDA are used as they are, there is a problem in that the accuracy of estimating the position of the label LB to be printed (i.e., the position where the image starts to be written) may decrease if the image forming device 1 that acquired the dataset DS required for machine learning (i.e., the image forming device 1 used for machine learning) and the image forming device 1 that actually corrects the estimated label pitch LP0 using the correction value generation model M obtained by machine learning are different devices.

[0123] In addition, when the image forming device 1 used for machine learning and the image forming device 1 that actually performs correction using the correction value generation model M obtained by machine learning are separate devices, for example, the image forming device 1 used for machine learning is owned by the manufacturer, and the image forming device 1 that actually performs correction using the correction value generation model M obtained by machine learning is owned by the user.

[0124] In this way, if the image forming apparatus 1 used for machine learning and the image forming apparatus 1 that actually performs correction using the correction value generation model M obtained by machine learning are different apparatuses, even if these image forming apparatuses 1 are the same model, there will be differences in the size of the label L to be printed and in the conveying characteristics (deterioration of the conveying roller 35, etc., manufacturing errors of the image forming apparatus 1) between these image forming apparatuses 1. As a result, the accuracy of estimating the position of the label LB to be printed (i.e., the position where the image starts to be written) will decrease on the image forming apparatus 1 side that actually performs correction using the correction value generation model M obtained by machine learning.

[0125] 12A, it is assumed that the image forming apparatus 1 (herein referred to as apparatus A) used for machine learning and the image forming apparatus 1 (herein referred to as apparatus B) that actually performs correction using the correction value generation model M obtained by machine learning show almost the same transitions in the image writing position when several tens of pages are printed. It is assumed that learning has not been applied to apparatus A and apparatus B.

[0126] In this way, even though the transition of the image writing position is almost the same between device A and device B, as shown in Figure 12(B), there are cases where the values ​​used as explanatory variables for machine learning (e.g., estimated label pitch LP0) are significantly different between device A and device B. This is because there are differences in the size of the label L to be printed and differences in the conveying characteristics between device A and device B.

[0127] In other words, if there is a difference in the size of the label L to be printed or a difference in the conveying characteristics between device A and device B, the explanatory variables of the machine learning (such as the estimated label pitch LP0) will be different even if the objective variable of the machine learning (the image writing position) is almost the same.

[0128] In such a case, if the estimated label pitch LP0 is corrected on the device B side using the correction value generation model M obtained by machine learning using the explanatory variables obtained on the device A side, as described above, the accuracy of estimating the image start position on the device B side will decrease. Specifically, as shown by the solid line (corrected) in Figure 13(A), if the estimated label pitch LP0 is corrected on the device B side using the correction value generation model M obtained by machine learning using the explanatory variables obtained on the device A side, the image start position will deviate from the ideal value indicated by the thick dotted line, and the accuracy of estimating the image start position will decrease.

[0129] On the other hand, as can be seen from Fig. 12(B), the transitions of the explanatory variables on the device A side (for example, estimated label pitch LP0) are almost the same as the transitions of the explanatory variables on the device B side. For this reason, when the difference between the explanatory variables on the device A side and the value one page before is taken, and similarly the difference between the explanatory variables on the device B side and the value one page before is taken, the difference between the explanatory variables on the device A side and the difference between the explanatory variables on the device B side are almost the same, as shown in Fig. 12(C).

[0130] In other words, if the changes in the image writing position are roughly the same between device A and device B, the changes in the difference in the explanatory variables will be roughly the same even if there are differences in the size of the label L to be printed or in the conveying characteristics.

[0131] For this reason, in this embodiment, when correcting the estimated label pitch LP0 using the correction value generation model M obtained by machine learning, the estimated label pitch LP0 and the label pitch displacement average value LPDA are not used as explanatory variables, but rather the difference values ​​from the values ​​of the previous page (estimated LP difference value and LP displacement average difference value) are used.

[0132] By doing this, in this embodiment, as shown by the dashed line (corrected (difference value)) in Figure 13 (A), when the estimated label pitch LP0 is corrected on the device B side using the correction value generation model M obtained by machine learning using the explanatory variables obtained on the device A side, the image writing position approximates the ideal value, and the image writing position can be estimated with high accuracy.

[0133] Here, Fig. 13(B) shows the variability of the image start position measured when learning is not applied, when the estimated label pitch LP0 and the label pitch displacement average value LPDA are used as explanatory variables (corrected), and when the estimated LP difference value and the LP displacement average difference value are used as explanatory variables (corrected (difference value)). Note that the average 2σ on the vertical axis of Fig. 13(B) is an index showing the variability of the image start position, and the smaller this value, the more stable the image start position.

[0134] As can be seen from Figure 13(B), in the image forming device 1 of this embodiment, the variation in the image writing position is smaller compared to when learning is not applied, and the variation in the image writing position is also smaller compared to when the estimated label pitch LP0 and the label pitch displacement average value LPDA are used as explanatory variables as is (with correction).

[0135] In this way, in the image forming device 1, instead of using the estimated label pitch LP0 and the label pitch displacement average value LPDA as explanatory variables for machine learning, the difference value from the value of the previous page (estimated LP difference value and LP displacement average difference value) is used, thereby making it possible to accurately estimate the position of the label LB to be printed, and as a result, to form an image at an appropriate position on the label LB to be printed.

[0136] In this embodiment, of the label pitch history data 71, estimated label pitch LP0, average label pitch variation LPDA, and variation fluctuation data 75, the estimated label pitch LP0 and average label pitch variation LPDA are calculated as differences from the values ​​of one page earlier. Meanwhile, the variation fluctuation data 75 is essentially the difference between the detected label pitch LP and the set value of the label pitch LP included in the print data, and since calculating the difference with the value of one page earlier is of little benefit, it is used as is. Meanwhile, the label pitch history data 71 is data indicating N label pitches LP, and although the difference with the value of one page earlier is not calculated to reduce the processing load, the difference with the label pitch LP of one page earlier may be calculated for each of the N label pitches LP.

[0137] In this case, the difference between the detected value of the label pitch LP of each of the N labels L including the reference label LA and the detected value of the label pitch LP of each of the N labels L including the label L one page before the reference label LA is taken, and this is used as the detected LP difference value as the detected pitch difference value, and is used as an explanatory variable instead of the label pitch history data 71.

[0138] [3-4. Detailed operation of machine learning device] 14 is a flowchart illustrating an example of an operation of the machine learning device 200. First, the processing unit 210 prepares an initial model (correction value generation model M0) of the correction value generation model M to be used in the machine learning process (step S201). In the neural network NN of this correction value generation model M0, the weights Wj are set to predetermined initial values.

[0139] Next, the dataset acquisition unit 211 acquires a dataset DS including the label pitch history data 221, the label pitch estimation data 222, the estimated LP difference value data 223, the LP displacement average difference value data 224, the displacement fluctuation data 225, and the label pitch data 226 (step S202). Then, the dataset acquisition unit 211 stores the acquired dataset DS in the storage unit 220.

[0140] Next, the teacher data generation unit 212 generates a correction value ΔLP as teacher data DT based on the difference between the label pitch LP indicated by the label pitch data 226 included in the data set DS and the estimated label pitch LP0 indicated by the label pitch estimation data 222 (step S203).

[0141] Next, the learning model generation unit 213 inputs the label pitch history data 221, the estimated LP difference value data 223, the LP displacement average difference value data 224, and the displacement fluctuation data 225 included in the data set DS to the input layer LX of the correction value generation model M0 (step S204). As a result, data indicating the correction value ΔLP is output from the output layer LZ of the correction value generation model M0.

[0142] Then, the learning model generation unit 213 performs machine learning processing based on the data output from the output layer LZ of the correction value generation model M0 and the correction value ΔLP (teacher data DT) generated in step S203 (step S205). Specifically, the learning model generation unit 213 compares the data output from the output layer LZ with the correction value ΔLP included in the teacher data DT to detect an error between them, and adjusts the weight Wj in the correction value generation model M0 so that data that reduces this error is output from the output layer LZ.

[0143] Next, the learning model generation unit 213 checks whether the machine learning process has ended (step S206). Specifically, the learning model generation unit 213 determines that the machine learning process has ended if it determines that further machine learning process is not required. If the machine learning process has not ended ("NO" in step S206), the process returns to step S202, and the processes of steps S202 to S206 are repeated until the machine learning process ends. By repeating this process in this manner, the accuracy of the correction value generation model M0 improves.

[0144] Then, in step S206, if the machine learning process is completed ("YES" in step S206), the learning model generation unit 213 stores the correction value generation model M0 used in this machine learning process in the storage unit 220 as the learned correction value generation model M (step S207). This is the end of this flow.

[0145] [4. Summary and Effects] As described above, the image forming apparatus 1 of this embodiment includes a medium supply roller 31 and a conveying roller 35 as a conveying section that conveys the recording medium 9 as a medium in a predetermined conveying direction, an image forming unit 10, a primary transfer roller 21, an intermediate transfer belt 22, and a secondary transfer section 28 as an image forming section that forms an image on labels L as printing areas provided on the recording medium 9 along the conveying direction, and a second label sensor 33 that is provided upstream of the image forming section in the conveying direction and is a sensor that detects the multiple labels L.

[0146] Furthermore, the image forming apparatus 1 includes a memory unit 70 that stores a correction value generation model M as a learning model that receives first data (i.e., explanatory variables) including label pitch history data 71 as pitch-related information related to the label pitch LP of the label L, displacement fluctuation data 75, an estimated LP difference value, and an LP displacement average difference value, and outputs a correction value ΔLP as second data for correcting the label pitch LP.

[0147] Furthermore, the image forming apparatus 1 includes a label pitch correction unit 65 as a calculation unit that generates a correction value ΔLP based on explanatory variables as first data using a correction value generation model M, and a write timing calculation unit 66 as an adjustment unit that adjusts the image formation position by the image forming unit on the label L (label LB to be printed) based on the calculation result of the label pitch correction unit 65.

[0148] Furthermore, among the pitch-related information included in the first data, part of the pitch-related information, that is, the estimated LP difference value and the LP displacement average difference value, is information related to the label pitch LP of the current label LB to be printed as the first printing range (i.e., the estimated label pitch LP0 n , Label pitch displacement average value LPDA n ), and information related to the label pitch LP of the label L as the second printing range provided downstream in the conveying direction from the current label LB to be printed (i.e., the label L one page before the label LB to be printed) (i.e., the estimated label pitch LP0 n-1 , Label pitch displacement average value LPDA n-1 ) and the difference value.

[0149] That is, the first data includes, as pitch-related information, an estimated label pitch LP0 as an estimated value of the pitch of the label LB to be printed. n and the estimated label pitch LP0 as an estimate of the pitch of label L one page before the label LB to be printed. n―1 The estimated LP difference value is included as an estimated pitch difference value, which is the difference between the

[0150] The first data also includes, as pitch-related information, a label pitch displacement average value LPDA, which is the average value of the label pitch displacements LPD of N labels L including the reference label LA as the third printing range K labels before the label LB to be printed. n and the label pitch displacement average value LPDA, which is the average value of the label pitch displacements LPD of N labels L including the label L (the label L one page before the reference label LA) as the fourth printing range one page before the label L K before the label LB to be printed. n―1 The LP displacement average difference value is included as the pitch displacement average difference value, which is the difference between the

[0151] In this way, in the image forming apparatus 1, some of the pitch-related information used as explanatory variables is taken as the difference value between the information related to the label pitch LP of the current label LB to be printed and the information related to the label pitch LP of the label L arranged downstream in the conveying direction from the current label LB to be printed. This makes it possible to estimate the position of the label LB to be printed (i.e., the position where the image starts to be written) with greater accuracy than when the difference is not taken, and as a result, the image can be formed at an appropriate position on the label LB to be printed.

[0152] In other words, in this embodiment, even when actual correction is performed using the correction value generation model M obtained by machine learning in an image forming device 1 that is a device different from the image forming device 1 used for machine learning, the start position of the image can be estimated with high accuracy, and the image can be formed in an appropriate position on the label LB to be printed.

[0153] The machine learning device 200 of this embodiment also includes a memory unit 220 that stores first data (i.e., explanatory variables) including label pitch history data 71, displacement fluctuation data 75, estimated LP difference values, and LP displacement average difference values ​​as pitch-related information related to the label pitch LP of labels L as printing areas provided in multiple directions along the conveying direction on a recording medium 9 as a medium conveyed in a predetermined conveying direction in the image forming device 1.

[0154] Furthermore, the machine learning device 200 includes a learning model generation unit 213 that performs machine learning processing using explanatory variables as first data and correction values ​​ΔLP as second data for correcting the label pitch LP, thereby generating a correction value generation model M as a learning model that receives explanatory variables as first data and outputs correction values ​​ΔLP as second data.

[0155] Furthermore, among the pitch-related information included in the first data, the estimated LP difference value and the LP displacement average difference value, which are part of the pitch-related information, are information related to the label pitch LP of the label L (the label L K before the reference label LA) as the first printing range (i.e., the estimated label pitch LP0 n , Label pitch displacement average value LPDA n ), and information related to the label pitch LP of the label L as the second printing range provided downstream in the conveying direction from the label L as the first printing range (i.e., the label L one page before the label L K pages before the reference label LA) (i.e., the estimated label pitch LP0 n-1 , Label pitch displacement average value LPDA n-1 ) and the difference value.

[0156] In this way, in the machine learning device 200, some of the pitch-related information used as explanatory variables is taken as the difference value between the information related to the label pitch LP of the label L K labels before the reference label LA and the information related to the label pitch LP of the label L arranged downstream in the conveying direction from the reference label LA. This makes it possible to estimate the position of the label LB to be printed (i.e., the position where the image starts to be written) more accurately than when the difference is not taken, and as a result, the image can be formed at an appropriate position on the label LB to be printed.

[0157] 5. Other Embodiments [5-1. Another embodiment 1] In the above-described embodiment, the label pitch estimation unit 64 generates the estimated label pitch LP0, the label pitch correction unit 65 generates a correction value ΔLP for the estimated label pitch LP0 using a learning model (correction value generation model M), and corrects the estimated label pitch LP0 based on this correction value ΔLP to generate the estimated label pitch LP1, but this is not limited to this. Alternatively, for example, the label pitch correction unit may generate the estimated label pitch LP1 using a learning model. Below, the image forming apparatus 1A and the machine learning device 200A according to this modification will be described in detail.

[0158] Fig. 15 shows an example of a control system in image forming apparatus 1A. Image forming apparatus 1A includes a processing unit 60A and a storage unit 70A. Processing unit 60A has a label pitch correction unit 67. Fig. 16 shows an example of a label pitch detection unit 62, a label pitch variation detection unit 63, a label pitch estimation unit 64, a label pitch correction unit 67, and a storage unit 70A.

[0159] The label pitch correction unit 67 serving as a calculation unit is configured to correct the estimated label pitch LP0 calculated by the label pitch estimation unit 64. Specifically, the label pitch correction unit 67 has a preprocessing unit 67A and a label pitch correction processing unit 67B, and the label pitch correction processing unit 67B uses a label pitch generation model MA serving as a learning model obtained by machine learning processing to correct the estimated label pitch LP0 indicated by the label pitch estimation data 72 based on the label pitch history data 71 and displacement fluctuation data 75 stored in the storage unit 70A, and the estimated LP difference value and LP displacement average difference value calculated by the preprocessing unit 67A, thereby generating an estimated label pitch LP1.

[0160] The storage unit 70A stores a label pitch generation model MA. The label pitch generation model MA receives label pitch history data 71, an estimated LP difference value, an LP displacement average difference value, and displacement fluctuation data 75, and outputs data indicating an estimated label pitch LP1.

[0161] Fig. 17 shows an example configuration of a machine learning device 200A that generates a label pitch generation model MA. The machine learning device 200A has a processing unit 210A and a storage unit 220A. The processing unit 210A has a dataset acquisition unit 211 and a learning model generation unit 213A. Fig. 18 shows an example of the dataset acquisition unit 211, the learning model generation unit 213A, and the storage unit 220A.

[0162] The learning model generation unit 213A is configured to generate a label pitch generation model MA by performing machine learning processing based on the label pitch history data 221, label pitch estimation data 222, estimated LP difference value data 223, LP displacement average difference value data 224, displacement fluctuation data 225, and label pitch data 226 stored in the storage unit 220A. In this example, the learning model generation unit 213A generates the label pitch generation model MA by performing so-called supervised learning using a neural network model. The learning model generation unit 213A then stores the generated label pitch generation model MA in the storage unit 220A.

[0163] The storage unit 220A stores the data set DS and the label pitch generation model MA. The label pitch generation model MA is generated by the learning model generation unit 213A and stored in the storage unit 220A.

[0164] [5-2. Other embodiment 2] Furthermore, in the above-described embodiment, the label pitch history data 71, the estimated LP difference value, the LP displacement average difference value, and the displacement fluctuation data 75 are used as explanatory variables for machine learning, and the estimated LP difference value and the LP displacement average difference value are each set to a difference value from the value of one page before. However, the present invention is not limited to this, and explanatory variables other than these four explanatory variables may be used, as long as at least one of the estimated LP difference value and the LP displacement average difference value is included in the explanatory variables.

[0165] Furthermore, in the above-described embodiment, the latest estimated label pitch LP0 nEstimated label pitch LP0 from the previous page n-1 The value obtained by subtracting the value from the estimated LP difference value is used, but this is not limited to this. n Alternatively, the estimated LP difference value may be calculated by subtracting the estimated label pitch LP0 from the average label pitch LPDA of the most recent page. n Label pitch displacement average value LPDA from one page before n-1 The value obtained by subtracting the LP displacement average difference value is used, but this is not limited to this. The latest label pitch displacement average value LPDA n Alternatively, the value obtained by subtracting the average label pitch displacement LPDA of two or more pages before from the above may be used as the LP displacement average difference value.

[0166] [5-3. Other embodiment 3] Furthermore, in the above-described embodiment, the present invention is applied to an image forming apparatus 1 that forms an image on a recording medium 9 (i.e., label paper) provided with a plurality of labels L. However, the present invention is not limited to this, and may be applied to an image forming apparatus that forms an image on a medium different from the recording medium 9. For example, the present invention may be applied to an image forming apparatus that forms an image on a medium in which a print range is set by a black mark, a notch, or the like.

[0167] [5-4. Other embodiment 4] Furthermore, the present invention is not limited to the above-described embodiments and other embodiments, and the scope of application of the present invention extends to embodiments in which the above-described embodiments and other embodiments are combined in part or in whole, or to embodiments in which some parts are different.

[0168] For example, in the above-described embodiment, the toner image formed by the image forming unit 10 is first transferred to the intermediate transfer belt 22, and then the toner image transferred to the intermediate transfer belt 22 is transferred to the recording medium 9. However, this is not limited to this, and instead, for example, the toner image formed by the image forming unit 10 may be directly transferred to the recording medium 9. [Industrial Applicability]

[0169] The present invention can be widely used in printers that form images in a printing area on labels and the like. [Explanation of symbols]

[0170] 1, 1A... image forming apparatus, 8... conveying path, 9... recording medium, 9B... mount, 9C... label surface, 10... image forming unit, 21... primary transfer roller, 22... intermediate transfer belt, 28... secondary transfer unit, 31... medium supply roller, 32... first label sensor, 33... second label sensor, 34... cutter, 35... conveying roller, 40... fixing unit, 53... image forming control unit, 60, 60A... processing unit, 61... sensor value acquisition unit, 62... label pitch detection unit, 63... label pitch displacement detection unit, 64... label pitch estimation unit, 65, 67... label pitch correction unit, 66... ​​write timing calculation unit, 70, 70A... memory unit, 71... label pitch history data, 72... label pitch estimation data, 73... displacement history data, 74... displacement average data, 75... displacement fluctuation data, 76... pre-processing history data, 200, 200A... machine learning device, 210, 210A...processing unit, 211...dataset acquisition unit, 212...teaching data generation unit, 213, 213A...learning model generation unit, 220, 220A...memory unit, 221...label pitch history data, 222...label pitch estimation data, 223...estimated LP difference value data, 224...LP displacement average difference value data, 225...displacement fluctuation data, 226...label pitch data, DS...dataset, DT...teaching data, F1, F2...conveying direction, L...label, LA...reference label, LB...label to be printed, LP...label pitch, LP0, LP1...estimated label pitch, LPD...label pitch displacement, LPDA...label pitch displacement average value, LS...label spacing, LX...input layer, LY1...first hidden layer, LY2...second hidden layer, LZ...output layer, M, M0...correction value generation model, MA...label pitch generation model, NN...neural network, ΔLP...correction value.

Claims

1. a storage unit that stores first data including pitch-related information related to the pitch of a plurality of printing ranges provided along a predetermined transport direction on a medium transported in the image forming apparatus; a learning model generation unit that performs machine learning processing using the first data and second data for correcting the pitch of the printing range to generate a learning model that receives the first data and outputs the second data; Equipped with a part of the pitch-related information included in the first data is a difference value between information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction; The second print range is the print range immediately before the first print range. A machine learning device characterized by:

2. A storage unit that stores first data including pitch-related information related to the pitch of a plurality of printing ranges provided along a predetermined transport direction on a medium transported in the image forming device; a learning model generation unit that performs machine learning processing using the first data and second data for correcting the pitch of the printing range to generate a learning model that receives the first data and outputs the second data; Equipped with a part of the pitch-related information included in the first data is a difference value between information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction; The first data includes, as the pitch-related information, an estimated pitch difference value that is the difference between the estimated value of the pitch in the first print range and the estimated value of the pitch in the second print range. A machine learning device characterized by:

3. A storage unit that stores first data including pitch-related information related to the pitch of a plurality of printing ranges provided along a predetermined transport direction on a medium transported in the image forming device; a learning model generation unit that performs machine learning processing using the first data and second data for correcting the pitch of the printing range to generate a learning model that receives the first data and outputs the second data; Equipped with a part of the pitch-related information included in the first data is a difference value between information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction; The first data includes, as the pitch-related information, a pitch displacement average difference value which is the difference between an average value of pitch displacements of a plurality of printing ranges including a third printing range that is a predetermined number of printing ranges before the first printing range and an average value of pitch displacements of a plurality of printing ranges including a fourth printing range that is the predetermined number of printing ranges before the second printing range, The pitch displacement is This is the difference between the detected value of the pitch of the print range detected using the sensor and the preset value of the pitch of the print range. A machine learning device characterized by:

4. A storage unit that stores first data including pitch-related information related to the pitch of a plurality of printing ranges provided along a predetermined transport direction on a medium transported in the image forming device; a learning model generation unit that performs machine learning processing using the first data and second data for correcting the pitch of the printing range to generate a learning model that receives the first data and outputs the second data; Equipped with a part of the pitch-related information included in the first data is a difference value between information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction; The second data is a difference between the pitch of the first print range and an estimated value of the pitch of the first print range. A machine learning device characterized by:

5. a conveying unit that conveys the medium in a predetermined conveying direction; an image forming unit that forms an image in a plurality of printing ranges provided on the medium along the transport direction; a storage unit that stores a learning model to which first data including pitch-related information related to the pitch of the printing range is input and to which second data for correcting the pitch of the printing range is output; a calculation unit that generates the second data based on the first data using the learning model; an adjustment unit that adjusts the image formation position by the image forming unit in the printing range based on the calculation result of the calculation unit; Equipped with a part of the pitch-related information included in the first data is a difference value between information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction; The first data includes, as the pitch-related information, an estimated pitch difference value that is the difference between the estimated value of the pitch of the first print range and the estimated value of the pitch of the second print range. An image forming apparatus characterized by:

6. A conveying unit that conveys the medium in a predetermined conveying direction; an image forming unit that forms an image in a plurality of printing ranges provided on the medium along the transport direction; a storage unit that stores a learning model to which first data including pitch-related information related to the pitch of the printing range is input and to which second data for correcting the pitch of the printing range is output; a calculation unit that generates the second data based on the first data using the learning model; an adjustment unit that adjusts the image formation position by the image forming unit in the printing range based on the calculation result of the calculation unit; Equipped with a part of the pitch-related information included in the first data is a difference value between information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction; a sensor for detecting the pitch of the printing range; The first data includes, as the pitch-related information, a pitch displacement average difference value which is the difference between an average value of pitch displacements of a plurality of printing ranges including a third printing range that is a predetermined number of printing ranges before the first printing range and an average value of pitch displacements of a plurality of printing ranges including a fourth printing range that is the predetermined number of printing ranges before the second printing range, The pitch displacement is is a difference between the detected value of the pitch of the printing range detected by the sensor and the preset value of the pitch of the printing range, The first data includes, as the pitch-related information, a detected pitch difference value that is a difference between a detected pitch value of each of a plurality of print ranges including the third print range and a detected pitch value of each of a plurality of print ranges including the fourth print range. An image forming apparatus characterized by:

7. Storing first data in a storage unit, the first data including pitch-related information relating to the pitch of a plurality of printing ranges provided along a predetermined transport direction on a medium transported in the image forming apparatus; performing machine learning processing using the first data and second data for correcting the pitch of the printing range, thereby generating a learning model in which the first data is input and the second data is output; Including, Among the pitch-related information included in the first data, a portion of the pitch-related information is a difference value between information related to the pitch of a first printing range and information related to the pitch of a second printing range provided downstream of the first printing range in the transport direction. A machine learning method characterized by:

Images