JP2022115319A - Base material processing method - Google Patents

Abstract

To provide technique enabling a base material to be continuously processed even when the characteristic and the ambient temperature of the base material, and the like change.SOLUTION: A base material processing method comprises the steps of: a) ejecting a processing material while conveying a base material 9 and acquiring control values including the number of revolutions of a motor 14 and ejection timing, various information In on the base material 9 and images of the base material 9; b) creating a simulation model; c) controlling the motor 14 and heads 21-24 by a first control unit 81; d) in parallel with the step c), inputting the variable control values to the simulation model from a second control unit 82 while updating the control values, and performing reinforcement learning for bring an estimated value of an error of an ejection position of the output processing material closer to zero; and e) switching the control unit controlling the motor 14 and the heads 21-24, from the first control unit to the second control unit when the estimated value of the error is less than a measured value of the error of the ejection position of the processing material calculated by analyzing the image.SELECTED DRAWING: Figure 1

Description

TECHNICAL FIELD The present invention relates to a method for treating a base material, in which a treatment substance is ejected onto the surface of a base material while conveying a long belt-like base material in the longitudinal direction.

2. Description of the Related Art Conventionally, there has been known an inkjet type image recording apparatus that records an image on a print sheet by ejecting ink from a plurality of recording heads while conveying the print sheet in the longitudinal direction. An image recording apparatus ejects inks of different colors from a plurality of recording heads. Then, a multicolor image is recorded on the surface of the printing paper by superimposing monochromatic images formed by inks of respective colors.

In this type of image recording apparatus, a plurality of rollers are required to transport the printing paper at a constant transport speed without meandering. However, slippage between the surface of the roller and the printing paper, or stretching of the printing paper due to ink ejected onto the surface of the printing paper, can cause the position of the printing paper below the recording head to deviate from the ideal position in the transport direction. It may shift in the width direction. As a result, the ejection positions of the inks of the respective colors on the surface of the printing paper are displaced in the transport direction and width direction, which may lead to deterioration in print quality. Therefore, a method for compensating for such a deviation is described in, for example, Japanese Unexamined Patent Application Publication No. 2002-200011.

JP 2019-166832 A

In Japanese Patent Application Laid-Open No. 2002-200000, ink is applied to a printing substrate by continuously aligning and rotating the printing drums during at least one revolution of one of the printing drums, a jetting drum, while transporting the printing substrate. A method of depositing drops to record an image (9) is disclosed. Then, based on the measurement result of the driving torque of the jetting drum and the grayscale value transition (10) of the image data recorded on the printing base material, the periodic compensation torque is calculated, and the compensation torque is taken into consideration. Disclosed is the step of controlling the jetting drum with a It also discloses performing machine learning when calculating the compensating torque.

However, when using the method disclosed in Patent Document 1, if the characteristics of the printing substrate on which the image is recorded, the characteristics of the ink dropped onto the printing substrate, the temperature and humidity around the printing substrate, etc. changes, machine learning must be performed from scratch. For this reason, it is necessary to take steps to resume printing only after the machine learning is completed, which may reduce workability.

The present invention has been made in view of such circumstances, and even if the characteristics of the printing substrate and ink, or the temperature and humidity around the printing substrate change, it is possible to prevent machine learning from completing each time. An object of the present invention is to provide a technology capable of continuously processing a printing base material with high accuracy without waiting.

In order to solve the above-mentioned problems, the first invention of the present application is a substrate treatment method in which a treatment substance is ejected from a head onto the surface of a long strip-shaped substrate while transporting the substrate in the longitudinal direction along a transport path. A method comprising: a) transporting the base material while discharging the treatment substance onto the surface of the base material while actually transporting the base material; The control value and the control value related to ejection including the ejection timing of the treatment substance, the type, shape, or thickness of the substrate, the ambient temperature or humidity of the substrate, or the type or characteristics of the treatment substance a step of acquiring a plurality of such various information and an image of the surface of the base material after the treatment substance has been discharged; a step of creating, on a simulator, a simulation model showing the relationship between the control value related to the ejection, the various information, and the actually measured value of the error in the ejection position of the treatment substance calculated by analyzing the image; and c) after creating the simulation model, inputting predetermined control values set in advance to the motor and the head from the first control unit as control values related to the transport and control values related to the ejection. d) in parallel with the step c) of controlling the motor and the head, from the second control unit to the simulator as a control value related to the transportation and a control value related to the ejection, which are variable e) inputting while updating the control value and performing reinforcement learning to bring the estimated value of the error in the ejection position of the treatment substance, which is output from the simulation model, closer to zero while using the various information; Compare the measured value of the error in the ejection position of the treatment substance calculated by analyzing the image with the estimated value of the error in the ejection position of the treatment substance, which is the output from the simulation model in step d). When the estimated value is lower than the measured value, the input of the predetermined control value from the first control unit to the motor and the head is stopped, and the motor and the head are controlled from the second control unit. automatically switching to initiate input of said variable control value to the head.

A second invention of the present application is the substrate treatment method of the first invention, wherein in the step a), while actually conveying the substrate, the substance to be treated is discharged onto the surface of the substrate; A plurality of measurement values relating to the tension applied to the base material, the conveying speed of the base material, or the position of the edge of the base material in the width direction are obtained, and in step b), based on the results obtained in step a), The relationship between the control value related to the transportation and the control value related to the ejection, the various information, the measured value, and the actual measurement value of the error in the ejection position of the treatment substance calculated by analyzing the image. The simulation model shown is created on the simulator, and in the step d), in parallel with the step c), the variable control value is input from the second control unit to the simulator while updating, While using the various information and the measured values, reinforcement learning is performed so as to bring the estimated value of the error in the discharge position of the treatment substance output from the simulation model closer to zero.

The third invention of the present application is the substrate treatment method of the first invention or the second invention, wherein f) after the step e), the type, shape, or thickness of the substrate, or the type or When the characteristic is changed, the input of the variable control value from the second control unit to the motor and the head is stopped, and the predetermined control value from the first control unit to the motor and the head is stopped. It further includes the step of automatically switching again to start inputting the

The fourth invention of the present application is the substrate treatment method according to any one of the first invention to the third invention, wherein g) after the step e), the temperature or humidity around the substrate is a predetermined When the variable control value changes beyond the range, the input of the variable control value from the second control unit to the motor and the head is stopped, and the predetermined control from the first control unit to the motor and the head is performed. There is also the step of automatically switching back to start entering values.

A fifth invention of the present application is the substrate processing method according to any one of the first invention to the fourth invention, wherein the substrate is stretched over a plurality of rollers, and the substrate is stretched over at least one of the plurality of rollers. is conveyed by the drive roller being driven and rotated by the motor, and the substrate processing method h) after the step e), when the motor vibrates beyond a predetermined range, the Automatic again to stop inputting the variable control value to the motor and the head from the second control unit and start inputting the predetermined control value to the motor and the head from the first control unit It further has a step of switching with .

The sixth invention of the present application is the method for treating a substrate according to any one of the first invention to the fifth invention, wherein: i) based on the acquisition result newly acquired in the step a), periodically , further comprising updating the simulation model.

The seventh invention of the present application is the substrate processing method of any one of the first invention to the sixth invention, wherein the reinforcement learning is PPO, DQN, R2D2, Q learning method, SARSA, Monte Carlo method, epsilon - greedy method, Boltzmann QPolicy method, or machine learning performed by SAC techniques.

An eighth aspect of the invention of the present application is a substrate processing method for ejecting a treatment substance from a head onto the surface of the substrate while conveying a long belt-shaped substrate in the longitudinal direction along a conveying path, comprising: j) A control value related to transportation including the number of rotations of a motor, which is a drive source for transporting the base material, and a control value related to ejection including the ejection timing of the treatment substance are set in advance from the first control unit. inputting a predetermined control value to the motor and the head to control the motor and the head while obtaining an image of the surface of the substrate after the treatment substance is discharged; ) In parallel with the step j), from the second control unit to the simulator, as the control value related to the transportation and the control value related to the ejection, variable control values are input while being updated, and the type of the base material, The treatment on the surface of the base material, which is the output from the simulation model on the simulator, using various information related to the shape or thickness, the ambient temperature or humidity of the base material, or the type or characteristics of the treatment substance. a step of performing reinforcement learning to bring the estimated value of the error in the ejection position of the substance closer to zero; , and an estimated value of the error of the discharge position of the treatment substance, which is the output from the simulation model in the step k), and when the estimated value is lower than the measured value, the first control unit a step of automatically switching from the second control unit to stop inputting the predetermined control value to the motor and the head and start inputting the variable control value to the motor and the head from the second control unit; and in the step k), the simulation model discharges the treatment substance onto the surface of the base material while actually conveying the base material. Based on the results of acquiring a plurality of control values, the various information, and images of the surface of the base material after the treatment substance has been discharged, control values and Based on the control value related to the ejection and the various information, the estimated value of the error in the ejection position of the treatment substance on the surface of the base material has been created by machine learning so as to output with high accuracy.

According to the first to eighth inventions of the present application, the substrate is treated by controlling the motor for conveying the substrate and the head for ejecting the treatment substance onto the surface of the substrate by the first control unit. At the same time, using the second control unit and the simulator, reinforcement learning is performed so that a control value for setting the error in the discharge position of the treatment substance on the surface of the base material to a value close to zero can be output. Then, after the reinforcement learning is completed, the control subject that controls the motor and the head is switched from the first control section to the second control section. As a result, even if the characteristics of the base material or ink, or the temperature or humidity around the base material change, the base material can be processed continuously with high accuracy without waiting for machine learning to complete each time. be able to.

1 is a diagram conceptually showing the configuration of an image recording apparatus; FIG. FIG. 3 is a partial top view of the image recording device in the vicinity of the image recording section; It is the figure which showed the structure of the edge position detection part typically. 4 is a flow chart showing the flow of simulation model creation, reinforcement learning, and print processing. FIG. 4 is a diagram conceptually showing an example of a data aggregate; FIG. 2 is a diagram conceptually showing an example of a decision tree included in a simulation model; FIG. 2 is a block diagram conceptually showing how reinforcement learning is performed; 4 is a flowchart showing a detailed flow of reinforcement learning; FIG. 4 is a diagram conceptually showing the configuration of an image recording apparatus when printing is performed using a second control unit; 5 is a graph showing an example of a first edge signal and an example of a second edge signal;

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described with reference to the drawings. In one embodiment of the present invention, as an example of a substrate processing apparatus, an image recording apparatus that records an image by ejecting ink droplets (hereinafter referred to as "ink droplets") from a head onto the surface of a conveyed printing paper. will be described as an example.

<1. First Embodiment>
<1-1. Configuration of Image Recording Apparatus>
First, the overall configuration of an image recording apparatus 1, which is an example of the substrate processing apparatus of the present invention, will be described with reference to FIG. FIG. 1 is a diagram conceptually showing the configuration of an image recording apparatus 1. As shown in FIG. The image recording apparatus 1 prints an image on the printing paper 9 by ejecting ink from a plurality of recording heads 21 to 24 toward the printing paper 9 while conveying the printing paper 9, which is a long belt-shaped base material. It is an inkjet printing device for recording. As shown in FIG. 1, the image recording apparatus 1 includes a transport mechanism 10, an image recording section 20, various information acquisition section 30, a tension detection section 40, an encoder 50, two edge position detection sections 60, an image acquisition section 70, and A computer 80 is provided.

The conveying mechanism 10 is a mechanism for conveying the printing paper 9 in the conveying direction along the longitudinal direction thereof. The transport mechanism 10 of this embodiment has a plurality of rollers including an unwinding roller 11, a plurality of transport rollers 12, and a take-up roller 13, and one or more (three in this embodiment) motors 14. . The printing paper 9 is stretched over the plurality of rollers. In this embodiment, motors 14 are connected to the unwinding roller 11, one of the plurality of conveying rollers 12 (the conveying roller 121 in FIG. 1), and the winding roller 13, respectively. A motor 14 is a drive source for conveying the printing paper 9 . The unwinding roller 11, the conveying roller 121, and the winding roller 13 are driven by the motor 14 and rotate about their rotation axes. Each conveying roller 12 guides the printing paper 9 to the downstream side of the conveying path by rotating around the rotation axis. As a result, the print paper 9 is fed out from the feed roller 11 and transported along the transport path formed by the plurality of transport rollers 12 . Further, the printing paper 9 after being conveyed is collected by the take-up roller 13 .

That is, in the present embodiment, the unwinding roller 11, the conveying roller 121, and the winding roller 13 among the plurality of rollers are driving rollers connected to the motor 14. FIG. However, the roller to which the motor 14 is connected is not limited to this. The printing paper 9 may be conveyed by a driving roller, which is at least one of a plurality of rollers, being driven by the motor 14 to rotate.

As shown in FIG. 1, the printing paper 9 moves substantially parallel to the arrangement direction of the plurality of recording heads 21 to 24 below the plurality of recording heads 21 to 24, which will be described later. At this time, the surface (recording surface) of the printing paper 9 faces upward (toward the recording heads 21 to 24). Also, the printing paper 9 is stretched over a plurality of transport rollers 12 while being tensioned. This suppresses slack and wrinkles in the printing paper 9 during transportation.

Further, a vibration detection section 141 for detecting vibration of the motor 14 is attached to the motor 14 connected to the conveying roller 121 of this embodiment. For example, a known vibration type contact sensor is used for the vibration detection unit 141 . The vibration detection section 141 detects vibration occurring in the motor 14 and outputs a vibration signal Vi based on the detection result to the computer 80 (integrated operation section 83 to be described later). Note that the vibration detector 141 may be attached to a member other than the motor 14 connected to the conveying roller 121 inside the image recording apparatus 1 .

The image recording unit 20 is a processing unit that ejects ink droplets, which are processing substances, onto the surface of the printing paper 9 conveyed by the conveying mechanism 10 . The image recording section 20 of this embodiment has a first recording head 21 , a second recording head 22 , a third recording head 23 and a fourth recording head 24 . The first recording head 21 , the second recording head 22 , the third recording head 23 and the fourth recording head 24 are arranged along the transport path of the printing paper 9 .

FIG. 2 is a partial top view of the image recording apparatus 1 in the vicinity of the image recording section 20. FIG. Each of the four recording heads 21 to 24 covers the entire width of the printing paper 9 . 2, a plurality of nozzles 250 arranged in parallel with the width direction of the printing paper 9 are provided on the lower surface of each of the recording heads 21-24. Each of the recording heads 21 to 24 emits K (black), C (cyan), M (magenta), and Y (yellow), which are the color components of a multicolor image, from a plurality of nozzles 250 toward the upper surface of the printing paper 9. Ink droplets of each color are ejected.

That is, the first recording head 21 ejects K-color ink droplets onto the upper surface of the printing paper 9 at the first processing position P1 on the transport path. The second recording head 22 ejects C-color ink droplets onto the upper surface of the printing paper 9 at a second processing position P2 downstream of the first processing position P1 in the transport direction. The third recording head 23 ejects ink droplets of M color onto the upper surface of the printing paper 9 at a third processing position P3 downstream in the transport direction from the second processing position P2. The fourth recording head 24 ejects Y-color ink droplets onto the upper surface of the printing paper 9 at a fourth processing position P4 downstream in the transport direction from the third processing position P3. In this embodiment, the first processing position P1, the second processing position P2, the third processing position P3, and the fourth processing position P4 are arranged at equal intervals along the transport direction of the printing paper 9. FIG.

The four recording heads 21 to 24 each record a monochrome image on the upper surface of the printing paper 9 by ejecting ink droplets. A multicolor image is formed on the upper surface of the printing paper 9 by superimposing the four monochromatic images. Therefore, if the positions of the ink droplets ejected from the four recording heads 21 to 24 on the printing paper 9 in the transport direction or the width direction are shifted from each other, the image quality of the printed matter will be degraded. That is, when the ink droplets ejected from the four recording heads 21 to 24 on the printing paper 9 are shifted in the transport direction or in the width direction (inter-color deviation) among the colors, the print quality is degraded. Therefore, it is an important factor for improving the printing quality of the image recording apparatus 1 to bring the errors in the transport direction and the width direction of the ejection positions of the ink droplets on the printing paper 9 close to zero.

The various information acquisition unit 30 is a device for acquiring various values, conditions, and information within the image recording apparatus 1 . The various information acquisition unit 30 includes, for example, an input interface such as a touch panel. Through the input interface, the operator can input, for example, the type, shape, and thickness of the printing paper 9, the temperature and humidity around the printing paper 9, and the type and characteristics of the ink ejected from the recording heads 21 to 24. and various information (hereinafter referred to as "various information In"). The various information acquisition unit 30 may have its own sensor such as a thermometer and a hygrometer for measuring the temperature and humidity around the printing paper 9 . The various information acquisition unit 30 may automatically acquire information such as the temperature and humidity around the printing paper 9 . Further, the various information acquisition unit 30 acquires these various types of information In via the above-described input interfaces and sensors, and outputs signals related to the acquired various information In to the computer 80 (an overall operation unit 83 and a simulator 84, which will be described later). ). However, the various information In acquired by the various information acquisition unit 30 may include at least one of the items listed above, and may include information other than the items listed above.

The tension detector 40 is attached to one of the plurality of transport rollers 12 (in this embodiment, the transport roller 122 in FIG. 1). The tension detector 40 intermittently measures the force that the conveying roller 122 receives from the printing paper 9 . However, the tension detector 40 may continuously measure the force that the conveying roller 122 receives from the printing paper 9 . Thereby, the tension detector 40 detects the tension applied to the printing paper 9 and outputs a tension signal Te related to the detection result to the computer 80 (a simulator 84 described later). The tension signal Te is data that reflects changes over time in the tension applied to the printing paper 9 conveyed by a plurality of conveying rollers 12 including the conveying roller 122 while being in contact with the conveying roller 122 . However, the tension detector 40 may not necessarily be provided.

The encoder 50 is attached to the shaft core of one of the plurality of transport rollers 12 (the transport roller 123 in FIG. 1). In this embodiment, the encoder 50 detects the rotational drive amount of the transport roller 123 and outputs a continuous pulse signal En synchronized with the rotation of the transport roller 123 to the computer 80 (simulator 84 described later). The continuous pulse signal En is data that reflects changes over time in the conveying speed of the printing paper 9 conveyed by the plurality of conveying rollers 12 including the conveying roller 123 . However, the encoder 50 may not necessarily be provided.

The two edge position detection units 60 are detection units that detect the width direction positions of the edges (width direction end portions) 91 of the printing paper 9 . In this embodiment, a first detection position Pa on the transport direction upstream side of the first processing position P1 on the transport path, and a fourth processing position P4 on the transport path away from the first detection position Pa on the transport direction downstream side. An edge position detector 60 is arranged at a second detection position Pb on the downstream side in the transport direction. However, the edge position detector 60 may not necessarily be provided.

FIG. 3 is a diagram schematically showing the structure of the edge position detection section 60. As shown in FIG. As shown in FIG. 3 , the edge position detection section 60 has a light projector 601 positioned above the edge 91 of the printing paper 9 and a line sensor 602 positioned below the edge 91 . The light projector 601 emits parallel light downward. The line sensor 602 has a plurality of light receiving elements 621 arranged in the width direction. As shown in FIG. 3, outside the edge 91 of the printing paper 9, the light emitted from the light projector 601 is incident on the light receiving element 621, and the light receiving element 621 detects the light. On the other hand, inside the edge 91 of the printing paper 9, the light emitted from the light projector 601 is blocked by the printing paper 9, so the light receiving element 621 does not detect the light. The edge position detection section 60 detects the position of the edge 91 of the printing paper 9 in the width direction based on the presence or absence of light detection by the plurality of light receiving elements 621 .

As shown in FIGS. 1 and 2 , the edge position detection section 60 arranged at the first detection position Pa is hereinafter referred to as a first edge position detection section 61 . Also, the edge position detection section 60 arranged at the second detection position Pb is referred to as a second edge position detection section 62 . The first edge position detector 61 intermittently detects the position of the edge 91 of the printing paper 9 in the width direction at the first detection position Pa. As a result, the detection result indicating the temporal change in the position of the edge 91 in the width direction at the first detection position Pa is obtained. Then, a detection signal indicating the obtained detection result (hereinafter referred to as "first edge signal Ed1") is output to the computer 80 (simulator 84 to be described later). The second edge position detector 62 intermittently detects the position of the edge 91 of the printing paper 9 in the width direction at the second detection position Pb. As a result, the detection result indicating the temporal change in the position of the edge 91 in the width direction at the second detection position Pb is obtained. Then, a detection signal indicating the obtained detection result (hereinafter referred to as "second edge signal Ed2") is output to computer 80 (simulator 84 described later). However, the first edge position detection section 61 and the second edge position detection section 62 may each continuously detect the position of the edge 91 of the printing paper 9 in the width direction. Further details such as the waveforms of the first edge signal Ed1 and the second edge signal Ed2 will be described later in <2-1. First Modification> and FIG.

The image acquisition unit 70 is located downstream of the recording heads 21 to 24 in the transport direction. The image acquiring unit 70 acquires an image of the surface of the printing paper 9 by imaging the surface of the printing paper 9 after ink droplets have been ejected from the plurality of recording heads 21 to 24 of the image recording unit 20 . The image acquisition unit 70 also outputs the acquired data related to the image of the printing paper 9 (hereinafter referred to as “image data Di”) to the computer 80 (integrated operation unit 83 to be described later). Note that the image acquisition unit 70 is equipment that has already been introduced in many general image recording apparatuses 1, so that it can be used without requiring additional introduction costs.

A drying processing section for drying the ink ejected onto the surface of the printing paper 9 may be further provided on the downstream side of the recording heads 21 to 24 in the transport direction. The drying unit dries the ink by, for example, blowing heated gas toward the printing paper 9 to evaporate the solvent in the ink adhering to the printing paper 9 . However, the drying processing section may dry the ink by other methods such as heating with a heat roller or light irradiation.

Computer 80 includes a first control section 81 , a second control section 82 , a general operation section 83 and a simulator 84 . The first control section 81 , the second control section 82 , and the overall operation section 83 are means for controlling the operation of each section in the image recording apparatus 1 .

As conceptually shown in FIG. 1, the first control unit 81 has a processor 811 such as a CPU, a memory 812 such as a RAM, and a storage unit 813 such as a hard disk drive. Further, in the storage unit 813, a program 81P and data 81D for controlling the operation of the image recording apparatus 1 are stored. 1, the first control unit 81 is electrically and selectively connected to the three motors 14 and the four recording heads 21 to 24, respectively. Also, the first control unit 81 is electrically connected to the overall operation unit 83 and the simulator 84, respectively. The first control unit 81 reads the program 81P and data 81D stored in the storage unit 813 into the memory 812 while being connected to the three motors 14 and the four recording heads 21 to 24, respectively, and stores the program 81P and the data 81D. , the processor 811 performs arithmetic processing to control the three motors 14 to convey the printing paper 9, and to control the recording heads 21 to 24 to eject ink onto the printing paper 9. Let As a result, the printing process in the image recording apparatus 1 proceeds.

The second control unit 82 has a processor 821 such as a CPU, a memory 822 such as a RAM, and a storage unit 823 such as a hard disk drive. Further, in the storage unit 823, a program 82P and data 82D for controlling the operation of the image recording apparatus 1 are stored. Also, the second control unit 82 is electrically and selectively connected to the three motors 14 and the four recording heads 21 to 24 (see FIG. 9, which will be described later). Also, the second control unit 82 is electrically connected to the overall operation unit 83 and the simulator 84, respectively. Further, the storage unit 823 stores a trained model M2 acquired by reinforcement learning, which will be described later. While being connected to the three motors 14 and the four recording heads 21 to 24, the second control unit 82 conveys the printing paper 9 by controlling the three motors 14 using the learned model M2. The recording heads 21 to 24 are respectively controlled to eject ink toward the printing paper 9 . As a result, the printing process of the printing paper 9 can be performed with high precision.

The integrated operation section 83 has a processor 831 such as a CPU, a memory 832 such as a RAM, and a storage section 833 such as a hard disk drive. A program 83P and data 83D are stored in the storage unit 833 . Also, the general operation unit 83 is electrically connected to the various information acquisition unit 30, the image acquisition unit 70, and the vibration detection unit 141, respectively. Also, the integrated operation unit 83 is electrically connected to the first control unit 81, the second control unit 82, and the simulator 84, respectively. Further, the overall operation unit 83 reads the program 83P and the data 83D stored in the storage unit 833 into the memory 832, and the processor 831 performs arithmetic processing based on the program 83P and the data 83D to obtain the first data as will be described later. The operations of the control unit 81 and the second control unit 82 are switched.

The simulator 84 has a processor 841 such as a CPU, a memory 842 such as a RAM, and a storage unit 843 such as a hard disk drive. A program 84P and data 84D are stored in the storage unit 843 . Further, as described above, the simulator 84 is electrically connected to the first control section 81, the second control section 82, and the overall operation section 83, respectively. Further, the storage unit 843 stores a simulation model SiMo created by machine learning, which will be described later. Then, the program 84P and data 84D stored in the storage unit 843 are read into the memory 842, and the processor 841 performs arithmetic processing based on the program 84P and data 84D, thereby executing a simulation to be described later.

<1-2. Flow of Simulation Model Creation, Reinforcement Learning, and Printing Process>
Next, the creation of the simulation model SiMo and the flow of reinforcement learning and print processing performed after the creation of the simulation model SiMo will be described.

FIG. 4 is a flow chart showing the flow of creation of the simulation model SiMo, and reinforcement learning and print processing performed after the creation of the simulation model SiMo. As shown in FIG. 4, first, in order to create the simulation model SiMo, the printing paper 9 is conveyed experimentally in the image recording apparatus 1, and ink is ejected onto the surface of the printing paper 9 from the plurality of recording heads 21 to 24. do. Then, a plurality of various data in each section in the image recording apparatus 1 at this time are acquired and stored (step S1). Specifically, first, the first control unit 81, the three motors 14, and the four recording heads 21 to 24 are electrically connected to each other.

Then, the first control unit 81 controls each motor 14 to actually convey the printing paper 9 by inputting to each motor 14 while changing a plurality of patterns of values as the control value CVd1 related to conveyance. Here, the control value CVd1 relating to transport is, for example, the number of revolutions of the motor 14 connected to the unwinding roller 11, the transport roller 121, and the winding roller 13, respectively. In addition, the first control unit 81 controls the recording heads 21 to 24 by inputting values to the recording heads 21 to 24 while changing a plurality of patterns as the control value CVe1 related to ejection, thereby controlling the recording heads 21 to 24 to apply ink to the surface of the printing paper 9. is ejected to print a test pattern (for example, a plurality of lines or marks printed while being separated from each other in the transport direction) on the surface of the printing paper 9 . Here, the control value CVe1 relating to ejection is, for example, the ejection timing of ink from the recording heads 21 to 24 (the ejection timing of ink droplets from each nozzle 250).

The first control unit 81 also outputs a control value CVd1 related to transportation and a control value CVe1 related to ejection to the simulator 84 . These pieces of information are accumulated in the storage unit 843 of the simulator 84 as a data aggregate DM. However, in step S1, the main control body for ejecting ink from the plurality of recording heads 21 to 24 onto the surface of the printing paper 9 while conveying the printing paper 9 on a trial basis is other than the first control unit 81 (for example, It may be the second control unit 82). Also, an operator may input control values to the three motors 14 and the four recording heads 21 to 24 by himself, and perform tests and data collection.

In step S1, the number of rotations of the motor 14 may be constant regardless of time, or may change with the passage of time. Further, the ink ejection timing may be regular or irregular. Further, as the control value CVe1 related to ejection, instead of or in addition to the ink ejection timing, the amount of ink ejected from the recording heads 21 to 24 (the amount of ink droplets ejected from each nozzle 250) Ink may be ejected onto the surface of the printing paper 9 while changing a plurality of patterns of . Alternatively, the number of rotations of the motor 14 may be excessively increased to cause slippage between the surfaces of the plurality of rollers and the printing paper 9, or the amount of ink ejected may be excessively increased to expand the printing paper 9 greatly. . Furthermore, a wide variety of patterns may be performed, including cases where the power of the image recording apparatus 1 is turned off, the motor 14 is not driven, and ink is not ejected. That is, the control value CVd1 related to transportation can be freely set as long as it includes at least information related to the number of revolutions of the motor 14 . Further, the control value CVe related to ejection can be freely set as long as it includes at least information related to the timing of ink ejection from the recording heads 21-24. Further, one of the control value CVd1 related to transportation and the control value CVe1 related to ejection may not be treated as a control value but may be a predetermined value.

Further, in step S1, while the printing paper 9 is transported on a trial basis, ink is ejected from the plurality of recording heads 21 to 24 onto the surface of the printing paper 9. , shape and thickness, temperature and humidity around the printing paper 9, and various types of information In relating to the type and characteristics of the ink ejected from the recording heads 21 to 24 are obtained. Then, the various information acquisition unit 30 outputs the acquired various information In to the simulator 84 . Various information In input from the various information acquisition unit 30 is stored as a data aggregate DM in the storage unit 843 of the simulator 84 .

Further, in step S1, while the printing paper 9 is transported on a trial basis, ink is ejected from the plurality of recording heads 21 to 24 onto the surface of the printing paper 9, and tension applied to the printing paper 9 is detected by the tension detector 40. is detected, and a tension signal Te related to the detection result is output to the simulator 84 . The simulator 84 has a tension calculator 844 (see FIG. 7, which will be described later). The functions of the tension calculation unit 844 and the speed calculation unit 845 and the edge position calculation unit 846, which will be described later, temporarily read the program 84P and the data 84D stored in the storage unit 843 of the simulator 84 into the memory 842, It is realized by the processor 841 performing arithmetic processing based on 84P and data 84D. In the simulator 84, the tension calculator 844 calculates the tension measurement value Mt of the printing paper 9 based on the input tension signal Te, and the memory 843 accumulates the measured value Mt as a data aggregate DM.

In step S1, while conveying the printing paper 9 on a trial basis, ink is ejected from the plurality of recording heads 21 to 24 onto the surface of the printing paper 9. A continuous pulse signal En related to the detection result is output to the simulator 84 . The simulator 84 has a velocity calculator 845 (see FIG. 7, which will be described later). Based on the input continuous pulse signal En, the simulator 84 calculates a measurement value Ms relating to the transport speed of the printing paper 9 in the speed calculation unit 845 and stores it in the storage unit 843 as a data aggregate DM. However, in step S1, a plurality of measured values such as the acceleration and vibration frequency of the printing paper 9 being conveyed may be acquired.

Further, in step S1, while the printing paper 9 is transported on a trial basis, ink is ejected from the plurality of recording heads 21 to 24 onto the surface of the printing paper 9, and the first edge position detection unit 61 detects the position of the printing paper 9. The position of the edge 91 in the width direction is detected, and a first edge signal Ed1 related to the detection result is output to the simulator 84 . Further, the second edge position detector 62 detects the position of the edge 91 of the printing paper 9 in the width direction, and outputs a second edge signal Ed<b>2 related to the detection result to the simulator 84 . The simulator 84 has an edge position calculator 846 (see FIG. 7 described later). In the simulator 84, the edge position calculator 846 calculates and stores a measured value Me related to the position of the edge 91 of the printing paper 9 in the width direction based on the input first edge signal Ed1 and second edge signal Ed2. In section 843, it is stored as a data aggregate DM.

In step S1, the test pattern is printed by the image acquiring unit 70 while ejecting ink from the plurality of recording heads 21 to 24 onto the surface of the printing paper 9 while the printing paper 9 is conveyed on a trial basis. A plurality of images of the surface of the printing paper 9 are acquired by capturing images of the surface of the printing paper 9 that has been pressed. Then, the acquired image data Di of the printing paper 9 is output to the integrated operation section 83 . The overall operation unit 83 has an image analysis unit 834 (see FIG. 7, which will be described later). The functions of the image analysis unit 834 and the determination unit 835 described later are to temporarily read the program 83P and the data 83D stored in the storage unit 833 of the general operation unit 83 into the memory 832, and store the program 83P and the data 83D. Based on this, the processor 831 performs arithmetic processing. The image analysis unit 834 calculates the actual measurement value Da of the ink droplet ejection position error on the surface of the printing paper 9 for each image data Di by analyzing the image associated with each input image data Di. Here, the actual measurement value Da of the ejection position error on the surface of the printing paper 9 is the difference between the position on the surface of the printing paper 9 where the ink droplets should be ejected and the position where the ink droplets are actually ejected. , indicate values related to errors in the transport direction and the width direction. Further, the overall operation unit 83 outputs the actual measurement value Da of the calculated ink droplet ejection position error to the simulator 84 . Then, the measured value Da of the ink droplet ejection position error is accumulated in the storage unit 843 of the simulator 84 as a data aggregate DM. However, the actual measurement value Da of the error in the ejection position of the ink droplet may be calculated visually by an operator or the like.

FIG. 5 is a diagram conceptually showing an example of the data aggregate DM. As shown in FIG. 5, a control value CVd1 related to transportation, a control value CVe1 related to ejection, various information In, a measured value Mt related to the tension of the printing paper 9, and transportation of the printing paper 9 are stored as the data aggregate DM. The measured value Ms related to the speed, the measured value Me related to the position in the width direction of the edge 91 of the printing paper 9, and the measured value Da of the error in the ejection position of the ink droplet are associated with each other. However, the data collection DM may also be manually recorded by an operator via an input interface or the like.

When a sufficient number of data is accumulated as the data aggregate DM, next, on the simulator 84, the accumulated control value CVd1 related to transportation, the accumulated control value CVe1 related to ejection, various information In, and the measured value A simulation model SiMo representing the relationship between Mt, Ms, Me and the actually measured value Da of the ink droplet ejection position error is created (step S2). Specifically, while using the measured value Da of the ink droplet ejection position error as teacher data (correct data), the ink droplet A simulation model SiMo is machine-learned for calculating the ejection position error with high accuracy.

The simulation model SiMo includes a decision tree X(a, b, c, f(CVd1, CVe1, In, Mt, Ms, Me)...). FIG. 6 is a diagram conceptually showing an example of the decision tree X (a, b, c, f (CVd1, CVe1, In, Mt, Ms, Me) . . . ) of this embodiment. In step S2, in machine learning, an estimated value De (calculated value) and the actual measurement value Da of the error in the ejection position of the ink droplet, which is the corresponding correct data, the decision tree X (a, b, c, f (CVd1, Mt, Ms, Me) . Here, the estimated value De of the ejection position error on the surface of the printing paper 9 is the difference between the position on the surface of the printing paper 9 where the ink droplets should be ejected and the position where the ink droplets are supposed to be ejected. It shows values related to the error in the transport direction and the width direction between the two.

When the difference between the estimated ink droplet ejection position error value De output from the simulation model SiMo and the corresponding measured ink droplet ejection position error value Da becomes equal to or less than a predetermined value, the machine learning is completed. As a result, using the simulation model SiMo, the estimated value De of the ink droplet ejection position error can be calculated with high accuracy based on the control values CVd1 and CVe1, various information In, and the measurement values Mt, Ms, and Me. becomes possible. However, some of the control values CVd1, CVe1, various information In, and the measured values Mt, Ms, Me may be omitted as the values input to the simulation model SiMo.

Also, the simulation model SiMo may include a model created by machine learning other than the decision tree (for example, GBM (Gradient Boosting Machine)). The simulation model SiMo includes control values CVd1 and CVe1, various types of information In, measured values Mt, Ms, and Me, and estimated ink droplet ejection position error values De, instead of models created by machine learning. may include a calculation formula (function) indicating the relationship of Further, the simulation model SiMo may include a conditional expression such as "whether the power of the image recording apparatus 1 is on or off" (for example, an expression that outputs 0 (zero) when the power is off). good.

Furthermore, in the present embodiment, even after the simulation model SiMo is created, the control value CVd1 related to transport (or the control value CVd2 related to transport described later) and the control value CVe1 related to ejection (or the control value CVe1 related to ejection described later) The value CVe2), various information In, the measured values Mt, Ms, and Me, and the measured value Da of the ink droplet ejection position error are subsequently acquired and accumulated as a data set DM. The simulation model SiMo is then updated periodically (for example, every 30 minutes) based on these newly acquired acquisition results, also by machine learning. As a result, even when the characteristics of the printing paper 9 and the ink droplets, or the ambient temperature and humidity of the printing paper 9 change, the estimated value De of the ink droplet ejection position error can be continuously increased from the simulation model SiMo. It can be calculated with precision.

After creating the simulation model SiMo, as shown in FIG. 4, the process proceeds to the next step S3. In step S3, first, print processing for printing a desired image to be actually recorded on the surface of the printing paper 9 is started (step S31). Specifically, while the first control unit 81, the three motors 14, and the four recording heads 21 to 24 are electrically connected to each other, the first control unit 81 controls the transportation. By inputting a predetermined control value set in advance to each motor 14 as the value CVd1, each motor 14 is controlled to convey the printing paper 9 . Further, by inputting a predetermined control value to the recording heads 21 to 24 as the control value CVe1 related to ejection, the recording heads 21 to 24 are controlled to eject ink onto the surface of the printing paper 9. FIG. Here, the predetermined control value set in advance is a value that should be input based on customs in a normal printing process. In this embodiment, even after starting step S3, the operation of step S1 continues as described above, and the simulation model SiMo created in step S2 is updated.

In parallel with step S31, the estimated value De (calculated value) of the error in the ejection position of the ink droplets on the surface of the printing paper 9 is calculated by using the second control unit 82 and the created simulation model SiMo. A control rule for maintaining the range is autonomously learned (machine learning) (step S32). In step S32, a trained model M2, which will be described later, is created by performing reinforcement learning. FIG. 7 is a block diagram conceptually showing how reinforcement learning is performed in step S32. Also, FIG. 8 is a flow chart showing the detailed flow of step S32.

When performing reinforcement learning, a pre-learning model M0 that serves as a prototype of a trained model M2, which will be described later, is prepared based on a reinforcement learning program. The pre-learning model M0 is installed in the storage unit 823 of the second control unit 82 . Then, the prepared pre-learning model M0 learns the relationship between the control value CVd2 related to transportation and the control value CVe2 related to ejection, and the reward reflecting the output from the simulation model SiMo. Here, the control value CVd2 related to transportation is similarly a variable control value assumed as the number of revolutions of the motor 14 connected to the unwinding roller 11, the transportation roller 121, and the winding roller 13, respectively. Further, the control value CVe2 related to ejection is a variable control value that is assumed as the ejection timing of ink from the recording heads 21 to 24 (the ejection timing of ink droplets from each nozzle 250).

Specifically, as shown in FIGS. 7 and 8, first, the pre-learning model M0 inputs certain values to the simulation model SiMo as the control value CVd2 related to transportation and the control value CVe2 related to ejection. (Step S321). However, either one of the control value CVd2 related to transportation and the control value CVe2 related to ejection may be a predetermined constant value.

At the same time, the various information acquisition unit 30 acquires various information In and inputs the acquired various information In to the simulation model SiMo. The tension detector 40 also detects tension applied to the printing paper 9 and outputs a tension signal Te according to the detection result to the tension calculator 844 of the simulator 84 . The tension calculator 844 calculates the tension measurement value Mt of the printing paper 9 based on the input tension signal Te, and inputs it to the simulation model SiMo. The encoder 50 also detects the amount of rotation of the conveying roller 123 and outputs a continuous pulse signal En related to the detection result to the speed calculator 845 of the simulator 84 . The speed calculator 845 calculates a measured value Ms relating to the transport speed of the printing paper 9 based on the inputted continuous pulse signal En, and inputs it to the simulation model SiMo. Further, the first edge position detection section 61 and the second edge position detection section 62 detect the position of the edge 91 of the printing paper 9 in the width direction, and generate a first edge signal Ed1 and a second edge signal Ed2 related to the detection results. , to the edge position calculator 846 of the simulator 84 . The edge position calculator 846 calculates a measured value Me related to the position of the edge 91 of the printing paper 9 in the width direction based on the input first edge signal Ed1 and second edge signal Ed2, and inputs it to the simulation model SiMo. do. Furthermore, the image acquisition section 70 acquires an image by imaging the surface of the printing paper 9 and outputs the image data Di to the image analysis section 834 of the overall operation section 83 . Based on the input image data Di, the image analysis unit 834 calculates an actual measurement value Da of the error in the ejection position of the ink droplets on the surface of the printing paper 9, and inputs it to the determination unit 835, which will be described later.

At the initial stage of step S321, the value output from the pre-learning model M0 is a random value. The simulation model SiMo estimates the error of the ink droplet ejection position on the surface of the printing paper 9 based on the values input from the pre-learning model M0, the various information In and the measured values Mt, Ms, and Me. De is calculated (step S322). Hereinafter, the pre-learning model M0 after learning has been started in this manner is referred to as a learning model M1.

Then, the simulator 84 inputs the estimated value De of the ink droplet ejection position error output from the simulation model SiMo to the overall operation section 83 . The integrated operation section 83 further has a function as a determination section 835 . Based on the reinforcement learning program, the determination unit 835 gives the learning model M1 a reward RE corresponding to the estimated value De of the ink droplet ejection position error, which is the output value from the simulation model SiMo (step S323). . At this time, the determination unit 835 gives a higher reward RE as the estimated value De approaches zero. On the other hand, the determination unit 835 gives a lower reward RE as the estimated value De is farther from zero.

When the learning model M1 receives the reward RE from the determination unit 835, it outputs the following values as the control value CVd2 related to transportation and the control value CVe2 related to ejection based on the reward RE. For example, when the learning model M1 is given a high reward RE from the determination unit 835, it outputs values close to the control values CVd2 and CVe2 at which the high reward RE is given as the next values. Further, when the learning model M1 is given a low reward RE from the determination unit 835, it outputs a value away from the control values CVd2 and CVe2 at which the low reward RE is given as the next value.

However, the learning model M1 may output the following values as the control value CVd2 related to transportation and the control value CVe2 related to ejection in consideration of values other than the reward RE. For example, the various information acquisition unit 30 may input various information In to the learning model M1, and the learning model M1 may output the following values in consideration of the input various information In. . More specifically, the learning model M1 outputs the following values depending on whether the ambient temperature of the printing paper 9, which is the various information In, is higher or lower than a predetermined temperature. The control value CVd2 and the control value CVe related to ejection may be distinguished.

Furthermore, in addition to the reward RE, the learning model M1 includes various information In, a measured value Mt of the tension of the printing paper 9, a measured value Ms of the transport speed of the printing paper 9, and a position of the edge 91 of the printing paper 9 in the width direction. , the following values may be output as the control value CVd2 related to transportation and the control value CVe2 related to ejection. In this case, the various information acquisition unit 30 described above also inputs the various information In to the learning model M1. The tension calculator 844 also inputs the calculated measured value Mt of the tension of the printing paper 9 to the learning model M1. The speed calculator 845 also inputs the calculated measured value Ms of the conveying speed of the printing paper 9 to the learning model M1. The edge position calculator 846 also inputs the calculated measured value Me of the position of the edge 91 of the printing paper 9 in the width direction to the learning model M1.

In this way, in step S4, the control value CVd2 related to transport and the control value CVe2 related to ejection, which are variable control values, are updated from the learning model M1 of the second control unit 82 to the simulation model SiMo on the simulator 84. while referring to the various information In and the measured values Mt, Ms, and Me described above, the estimated value De of the error in the ejection position of the ink droplet on the surface of the printing paper 9, which is the output from the simulation model SiMo, is set to zero. Reinforcement learning is performed to bring it closer. Further, in the present embodiment, as reinforcement learning (deep reinforcement learning), for example, machine learning performed by a technique of R2D2 (Recurrent Replay Distributed DQN) is performed. However, as reinforcement learning, we perform machine learning performed by techniques such as PPO (Proximal Policy Optimization), DQN (Deep Q Network), Q-learning, SARSA, Monte Carlo, epsilon-greedy, Boltzmann QPolicy, or SAC. may Further, in step S4, reinforcement learning may be performed without referring to the various information In or the measured values Mt, Ms, and Me.

As described above, in the initial stage of reinforcement learning, the learning model M1 outputs random values as the control values CVd2 and CVe2. However, in the process of repeating the processing of steps S321 to S323, the control values CVd2 and CVe2 are tried to increase, maintain, decrease, etc. according to the reward RE given by the determination unit 835. FIG. This automatically learns the relationship between the control values CVd2 and CVe2 and the reward RE reflecting the output from the simulation model SiMo. Then, the learning model M1 gradually becomes able to obtain a high reward RE. That is, the learning model M1 gradually increases the control values CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVd2, CVe2 can now be output.

After that, the determination unit 835 determines whether or not to once complete the reinforcement learning based on the reinforcement learning program (step S324). As described above, the actual measurement value Da of the ink droplet ejection position error on the surface of the printing paper 9 calculated by the image analysis unit 834 of the overall operation unit 83 is input to the determination unit 835 . The determination unit 835 determines the input actual measurement value Da of the error in the ejection position of the ink droplets on the surface of the printing paper 9 and the error in the ejection position of the ink droplets on the surface of the printing paper 9, which is the calculated value from the simulation model SiMo. Compare with the estimated value De. Then, if the estimated value De is greater than or equal to the measured value Da, the reinforcement learning is continued (step S324: NO). In that case, the above steps S321 to S324 are executed again.

On the other hand, if it is determined in step S324 that the estimated value De is lower than the measured value Da, reinforcement learning is once completed based on the reinforcement learning program (step S324: YES). Then, the learning model M1 for which the above reinforcement learning has been completed becomes the learned model M2. In addition, after this paragraph, for convenience of explanation, the subsequent steps will be described assuming that the reinforcement learning is once completed (stopped) in step S324, but step S324 is to the end of the reinforcement learning at a certain stage. It is a thing, and in fact, reinforcement learning is continued even after that.

When the determination unit 835 determines that the reinforcement learning is completed, at the same time, as shown in FIG. The input of the predetermined control values CVd1 and CVe1 from the control unit 81 to the motors 14 and the recording heads 21 to 24 is stopped, and the second control unit 82, the motors 14 and the recording heads 21 to 24 are electrically connected. By connecting, the input of the variable control values CVd2 and CVe2 from the second control unit 82 to the motors 14 and the recording heads 21 to 24 is automatically switched (step S4). As described above, through reinforcement learning, the second control unit 82 increases the control values CVd2 and CVe2 for making the error (estimated value De) of the ejection position of the ink droplets on the surface of the printing paper 9 close to zero. It has been trained so that it can be output with accuracy. As a result, for example, the characteristics of each actual device (for example, characteristics that appear with aging of the device, differences in thickness due to minute unevenness on the printing paper 9, slight differences in viscosity for each ink, etc.) can achieve good performance.

As described above, in this embodiment, first, the first control unit 81 inputs the control values CVd1 and CVe1 based on customs in the normal printing process to the motors 14 and the recording heads 21 to 24. While carrying out the printing process by ejecting ink droplets while conveying the printing paper 9, in parallel with this, the error in the ejection position of the ink droplets on the surface of the printing paper 9 is detected using the second control unit 82 and the simulator 84. Reinforcement learning is performed so that control values CVd2 and CVe2 for making (estimated value De) close to zero can be output. Then, after the reinforcement learning is completed, the subject controlling each motor 14 and the recording heads 21 to 24 is switched from the first control section 81 to the second control section 82 . As a result, the printing process for printing a desired image on the surface of the printing paper 9 can be started without waiting for the reinforcement learning of the second control unit 82 to be completed, thereby improving work efficiency. .

Further, even after switching the main control unit for the motors 14 and the recording heads 21 to 24 from the first control unit 81 to the second control unit 82, the determination unit 835 can detect various situations in the image recording apparatus 1. Stay on top of changes. First, as described above, the various information In acquired by the various information acquisition section 30 is input to the overall operation section 83 . Therefore, based on the various information In input from the various information acquisition unit 30, the determination unit 835 determines whether the type, shape, or thickness of the printing paper 9 to be used, or the type or characteristics of the ink droplets ejected onto the printing paper 9 is determined. When it is determined that the change has been made, the connection state between the second control unit 82 and the motors 14 and the recording heads 21 to 24 is released, so that the connections from the second control unit 82 to the motors 14 and the recording heads 21 to 24 are canceled. By stopping the input of the predetermined control values CVd2 and CVe2 and electrically connecting the first control unit 81 to the motors 14 and the recording heads 21 to 24, the first control unit 81 controls the motors 14 and The recording heads 21 to 24 are automatically switched again so as to start inputting the variable control values CVd1 and CVe1 (step S5). Then, the above steps S3 and S4 are executed again. As a result, better operation suitable for new types of printing paper 9 or the like can be achieved.

Similarly, when the determination unit 835 recognizes that the temperature or humidity around the printing paper 9 has changed beyond a predetermined range based on the various information In input from the various information acquisition unit 30, the second control unit 82, the motor 14 and the recording heads 21 to 24 are disconnected from each other, the input of the predetermined control values CVd2 and CVe2 from the second control unit 82 to the motor 14 and the recording heads 21 to 24 is stopped. Also, by electrically connecting the first control unit 81 to the motors 14 and the recording heads 21 to 24, the variable control value CVd1 from the first control unit 81 to the motors 14 and the recording heads 21 to 24 can be set. , CVe1 are automatically switched again. Then, the above steps S3 and S4 are executed again. As a result, it is possible to realize better operation in accordance with the new environment after the temperature or the like has changed.

Further, as described above, the vibration signal Vi associated with the vibration generated in the motor 14 detected by the vibration detection section 141 is input to the integrated operation section 83 . Therefore, similarly, based on the vibration signal Vi input from the vibration detection unit 141, the determination unit 835 grasps that one of the motors 14 has vibrated beyond a predetermined range. By disconnecting the motor 14 and the recording heads 21 to 24, input of the predetermined control values CVd2 and CVe2 from the second control unit 82 to the motors 14 and the recording heads 21 to 24 is stopped, and By electrically connecting the first control unit 81 to the motors 14 and the recording heads 21 to 24, variable control values CVd1 and CVe1 from the first control unit 81 to the motors 14 and the recording heads 21 to 24 are controlled. Automatically switch again to start typing. Then, the above steps S3 and S4 are executed again. This makes it possible to achieve better operation in line with the new environment after the vibration has occurred.

As described above, in the present embodiment, when the characteristics of the printing paper 9 and ink, or the temperature and humidity around the printing paper 9 change, the subject that controls the motors 14 and the recording heads 21 to 24 is temporarily changed. , from the second controller 82 back to the first controller 81 . Then, while the first control unit 81 controls each motor 14 and the recording heads 21 to 24, in parallel with this, the second control unit 82 and the simulator 84 are used to print in accordance with the changed conditions. Reinforcement learning is performed so that the control values CVd2 and CVe2 for making the error (estimated value De) of the ink droplet ejection position on the surface of the paper 9 close to zero can be output. After the reinforcement learning is completed, the subject controlling the motors 14 and the recording heads 21 to 24 is switched from the first control section 81 to the second control section 82 again. With such a configuration, even if the characteristics of the printing paper 9 or ink, or the temperature, humidity, etc. around the printing paper 9 change, it is possible to wait for the second control unit 82 to complete the reinforcement learning. Since print processing can be continued without interruption, work efficiency can be improved.

<2. Variation>
Although one embodiment of the present invention has been described above, the present invention is not limited to the above embodiment. In the following, differences from the above-described embodiment will be described for various modifications.

<2-1. First modification>
In the above-described embodiment, in the speed calculation unit 845, based on the continuous pulse signal En output from the encoder 50 connected to the conveying roller 123, the speed of the printing paper 9 conveyed by the plurality of conveying rollers 12 including the conveying roller 123 is calculated. A measured value Ms of the conveying speed was calculated. However, in the speed calculator 845, based on the first edge signal Ed1 output from the first edge position detector 61 and the second edge signal Ed2 output from the second edge position detector 62, the transport speed of the printing paper 9 is calculated. may be calculated. Details are described below.

FIG. 10 is a graph showing an example of the first edge signal Ed1 and an example of the second edge signal Ed2. In FIG. 10, the horizontal axis indicates time. The vertical axis in FIG. 10 indicates the position of the edge 91 in the width direction. Note that the left end of the horizontal axis of the graph in FIG. 10 is the current time, and the time becomes older toward the right side. Therefore, the data line in FIG. 10 moves to the right as time elapses, as indicated by the white arrow. An edge 91 of the printing paper 9 has minute unevenness. The first edge position detection section 61 and the second edge position detection section 62 detect the position of the edge 91 of the printing paper 9 in the width direction at preset minute intervals (for example, at intervals of 50 microseconds). As a result, as shown in FIG. 10, data indicating the change in the position of the edge 91 of the printing paper 9 in the width direction with time is obtained. The first edge signal Ed1 is data reflecting the shape of the edge 91 of the printing paper 9 passing through the first detection position Pa. The second edge signal Ed2 is data reflecting the shape of the edge 91 of the printing paper 9 passing through the second detection position Pb.

The velocity calculator 845 compares the first edge signal Ed1 and the second edge signal Ed2. Then, the position where the same edge 91 of the printing paper 9 is detected is specified by the first edge signal Ed1 and the second edge signal Ed2. Specifically, for each data interval (fixed time range) included in the first edge signal Ed1, data with high matching among a plurality of data intervals (fixed time range) included in the second edge signal Ed2 Identify the interval. A data section included in the first edge signal Ed1 is hereinafter referred to as a first data section DS1. A data section included in the second edge signal Ed2 is referred to as a second data section DS2.

A matching method such as cross-correlation or residual sum of squares is used to identify data segments with high matching. The velocity calculator 845 selects a plurality of second data sections DS2 as corresponding data section candidates for each first data section DS1. Also, for each of the plurality of second data sections DS2 that are selected candidates, an evaluation value indicating the degree of matching with the first data section DS1 is calculated. Then, the second data section DS2 with the highest evaluation value is identified as the second data section DS2 corresponding to the first data section DS1 (the highest match with the first data section DS1).

After that, the speed calculation unit 845 calculates the first detection position based on the time difference between the detection time TM1 of the first data section DS1 and the detection time TM2 of the second data section DS2 that most closely matches the first data section DS1. An actual transport time ΔT (time difference between time TM2 and time TM1) required for transporting the printing paper 9 from Pa to the second detection position Pb is calculated. By dividing the distance from the first detection position Pa to the second detection position Pb by the transport time ΔT, the transport speed of the printing paper 9 below the image recording unit 20 can be calculated.

<2-2. Second modification>
In addition to the methods described in the above embodiments or modifications, a laser speedometer (not shown) for detecting the transport speed of the printing paper 9 may be used. Then, from the laser speedometer, two laser beams are irradiated toward the printing paper 9, and based on the results of analysis of the wavelengths of the reflected light reflected by the printing paper 9, the conveyance of the printing paper 9 is determined. Velocity may be calculated.

<2-3. Third modification>
In the above embodiment, in step S1, while conveying the printing paper 9 on a trial basis, ink is ejected from the plurality of recording heads 21 to 24 onto the surface of the printing paper 9, and each part in the image recording apparatus 1 After acquiring a plurality of various data and creating a simulation model SiMo in step S2, steps S3 to S5 are executed. However, if the simulation model SiMo has already been created, steps S1 and S2 may be omitted.

<2-4. Other Modifications>
The image recording apparatus 1 records an image on the surface of the printing paper 9 by ejecting ink as a treatment substance while conveying the printing paper 9 . However, the image recording apparatus 1 may record an image by ejecting a treatment substance other than ink onto the surface of the printing paper 9 . Further, the image recording apparatus 1 may be an apparatus that records an image on the printing paper 9 by a method other than inkjet (for example, electrophotography, exposure, etc.). Further, the image recording apparatus 1 described above performs print processing on the printing paper 9 as a base material. However, the substrate processing apparatus of the present invention may perform predetermined processing on a long belt-shaped substrate other than general paper (for example, resin film, metal foil, etc.).

That is, the present invention provides a substrate processing method for discharging a treatment substance from a head onto the surface of a substrate while transporting a long belt-shaped substrate in the longitudinal direction along a transport path, comprising: a) actually a control value related to conveyance including the number of rotations of a motor that is a drive source for conveying the substrate while discharging the treatment substance onto the surface of the substrate while conveying the substrate, and the treatment substance A control value related to ejection including the ejection timing of the substrate, the type, shape, or thickness of the substrate, the temperature or humidity around the substrate, or various information related to the type or characteristics of the processing substance, and the processing substance a step of acquiring a plurality of images of the surface of the base material after the is ejected; b) a control value related to the transportation and a control value related to the ejection based on the result obtained in the step a); a step of creating, on a simulator, a simulation model showing the relationship between the various information and the actually measured value of the error in the discharge position of the treatment substance calculated by analyzing the image; and c) creating the simulation model. After that, the motor and the head are operated by inputting predetermined control values, which are set in advance, from the first control unit to the motor and the head as the control value for the transportation and the control value for the ejection. and d) in parallel with the step c), inputting variable control values from the second control unit to the simulator as the control values relating to the transportation and the control values relating to the ejection while updating them. , using the various information, performing reinforcement learning to bring the estimated value of the error in the ejection position of the treatment substance, which is the output from the simulation model, closer to zero; and the estimated value of the error in the ejection position of the treatment substance, which is the output from the simulation model in the step d), and the estimated value is the actual measurement input of the predetermined control value from the first control unit to the motor and the head, and the variable control value from the second control unit to the motor and the head and a step of automatically switching to start inputting.

Further, in the above step a), while actually conveying the substrate and discharging the treatment substance onto the surface of the substrate, the tension applied to the substrate, the transportation speed of the substrate, or the A plurality of measured values relating to the position of the edge of the substrate in the width direction are obtained, and in the step b) above, based on the results obtained in the step a), the control value relating to the transportation and the control value relating to the ejection are obtained. and creating, on the simulator, the simulation model showing the relationship between the various information, the measured values, and the actually measured values of the error in the ejection position of the treatment substance calculated by analyzing the image, In the above-described step d), in parallel with the step c), the variable control value is updated and input from the second control unit to the simulator, and using the various information and the measured value, Reinforcement learning may be performed to bring an estimated error in the discharge position of the treatment substance output from the simulation model closer to zero.

The present invention also provides a substrate treatment method for ejecting a treatment substance from a head onto the surface of the substrate while conveying a long belt-shaped substrate in the longitudinal direction along a conveying path, comprising: j) 1 from the control unit, as a control value related to transport including the number of rotations of a motor that is a driving source for transporting the substrate, and a control value related to ejection including the ejection timing of the treatment substance, a predetermined value set in advance; to the motor and the head to acquire an image of the surface of the substrate after the treatment substance has been discharged, while controlling the motor and the head; and k) In parallel with the step j), from the second control unit to the simulator, as the control value related to the transportation and the control value related to the ejection, variable control values are input while updating, and the type and shape of the base material are input. Alternatively, the treatment substance on the surface of the base material, which is output from a simulation model on the simulator, using various information related to the thickness, the ambient temperature or humidity of the base material, or the type or characteristics of the treatment substance l) a step of performing reinforcement learning to bring the estimated value of the ejection position error closer to zero; is compared with the estimated value of the error in the discharge position of the treatment substance, which is the output from the simulation model in the step k), and when the estimated value is lower than the measured value, the first control unit a step of automatically switching to stop inputting the predetermined control value to the motor and the head and start inputting the variable control value to the motor and the head from the second control unit; and in the step k), the simulation model discharges the treatment substance onto the surface of the base material while actually conveying the base material, the control value related to the transport and the control related to the discharge values, the various information, and an image of the surface of the base material after the treatment substance has been discharged, the control values relating to the transport and the It is sufficient that the estimated value of the error in the ejection position of the treatment substance on the surface of the base material can be output with high accuracy from the control value related to ejection and the various information.

Also, the elements appearing in the above embodiments and modifications may be appropriately combined as long as there is no contradiction.

1 image recording device 9 printing paper 10 conveying mechanism 11 unwinding roller 12 conveying roller 13 take-up roller 14 motor 20 image recording unit 21 first recording head 22 second recording head 23 third recording head 24 fourth recording head 30 various information Acquisition unit 40 Tension detection unit 50 Encoder 60 Edge position detection unit 70 Image acquisition unit 81 First control unit 82 Second control unit 83 Integrated operation unit 84 Simulator 91 Edge 121 Conveyor roller 122 Conveyor roller 123 Conveyor roller 141 Vibration detector 250 Nozzle 834 Image analysis unit 835 Judgment unit 844 Tension calculation unit 845 Velocity calculation unit 846 Edge position calculation unit CVd1, CVd2 Control values related to transportation CVe1, CVe2 Control values related to ejection Da (Error in ejection position of ink droplet on the surface of printing paper Measured value De Estimated value (error of ink droplet ejection position on the surface of printing paper) Di Image data In Various information M0 Model before learning M1 Model during learning M2 Model after learning Me (Width direction of edge of printing paper Measured value (related to position) Ms Measured value (of printing paper transport speed) Mt Measured value (of tension of printing paper 9) SiMo Simulation model Te Tension signal Vi Vibration signal

Claims (8)

A base material treatment method for ejecting a treatment substance from a head onto the surface of the base material while conveying a long belt-like base material in the longitudinal direction along a conveying path,
a) a control value related to transportation, including the number of revolutions of a motor, which is a drive source for transporting the substrate while actually transporting the substrate and ejecting the treatment substance onto the surface of the substrate; Control values related to ejection including the timing of ejection of the treatment substance; various types of information relating to the type, shape, or thickness of the base material; temperature or humidity around the base material; or the type or characteristics of the treatment substance; a step of acquiring a plurality of images of the surface of the substrate after the treatment substance has been discharged;
b) an error in the ejection position of the treatment substance calculated by analyzing the control value relating to the transportation and the control value relating to the ejection, the various information, and the image, based on the results obtained in the step a); A step of creating a simulation model showing the relationship between the measured values of and on the simulator;
c) after creating the simulation model, by inputting preset control values to the motor and the head from the first control unit as control values relating to the transport and control values relating to the ejection; , controlling the motor and the head;
d) In parallel with the step c), from the second control unit to the simulator, as the control value related to the transportation and the control value related to the ejection, variable control values are input while being updated, and the various information is input. a step of performing reinforcement learning for bringing the estimated value of the error in the discharge position of the treatment substance, which is the output from the simulation model, closer to zero while using the simulation model;
e) a measured value of the error in the ejection position of the treatment substance calculated by analyzing the image, and an estimated value of the error in the ejection position of the treatment substance, which is the output from the simulation model in step d); are compared, and if the estimated value is lower than the measured value, the input of the predetermined control value from the first control unit to the motor and the head is stopped, and the second control unit outputs the motor and automatically switching to start inputting the variable control value to the head;
A substrate treatment method. The substrate treatment method according to claim 1,
In the step a), while actually conveying the substrate and discharging the treatment substance onto the surface of the substrate, the tension applied to the substrate, the transportation speed of the substrate, or the edge of the substrate Acquire multiple measurement values related to the position in the width direction of
In the step b), based on the results obtained in the step a), the control value related to the transportation and the control value related to the ejection, the various information, the measured value, and the image are analyzed and calculated. creating the simulation model showing the relationship between the measured value of the error in the ejection position of the treatment substance and the relationship on the simulator,
In the step d), in parallel with the step c), the variable control value is updated and input from the second control unit to the simulator, and the simulation is performed using the various information and the measured value. A base material processing method, wherein reinforcement learning is performed to bring an estimated value of the error in the ejection position of the processing substance, which is output from the model, closer to zero. The substrate treatment method according to claim 1 or claim 2,
f) after step e), when the type, shape or thickness of the substrate or the type or property of the treatment substance is changed, the variable and automatically switching again to start inputting the predetermined control values from the first control unit to the motor and the head. The substrate treatment method according to any one of claims 1 to 3,
g) inputting the variable control value from the second control unit to the motor and the head when the temperature or humidity around the substrate changes beyond a predetermined range after step e); and automatically switching again to start inputting the predetermined control values from the first controller to the motor and the head. The substrate treatment method according to any one of claims 1 to 4,
The base material is stretched over a plurality of rollers, and conveyed by rotating at least one of the plurality of rollers, a driving roller, driven by the motor,
The base material treatment method includes
h) after step e), if the motor vibrates beyond a predetermined range, stop inputting the variable control value from the second control unit to the motor and the head; The substrate processing method further comprising the step of automatically switching again to start inputting the predetermined control values from a control unit to the motor and the head. The substrate treatment method according to any one of claims 1 to 5,
i) The substrate processing method, further comprising the step of periodically updating the simulation model based on newly obtained results obtained in step a). The substrate treatment method according to any one of claims 1 to 6,
A substrate processing method, wherein said reinforcement learning is machine learning performed by PPO, DQN, R2D2, Q-learning method, SARSA, Monte Carlo method, epsilon-greedy method, Boltzmann QPolicy method, or SAC technique. A base material treatment method for ejecting a treatment substance from a head onto the surface of the base material while conveying a long belt-like base material in the longitudinal direction along a conveying path,
j) preset from the first control unit as a control value related to transportation including the number of revolutions of a motor that is a drive source for transporting the substrate and a control value related to ejection including the ejection timing of the treatment substance; obtaining an image of the surface of the base material after the treatment substance has been discharged while controlling the motor and the head by inputting the determined control value to the motor and the head; ,
k) In parallel with the step j), from the second control unit to the simulator, as the control value related to the transportation and the control value related to the ejection, variable control values are input while being updated, and the type of the base material is input. , shape or thickness, temperature or humidity around the substrate, or the type or characteristics of the substance to be treated, the above on the surface of the substrate that is output from the simulation model on the simulator a step of performing reinforcement learning to bring the estimated value of the error in the discharge position of the treatment substance closer to zero;
l) the measured value of the error in the ejection position of the treatment substance calculated by analyzing the image in step j) and the error in the ejection position of the treatment substance output from the simulation model in step k); When the estimated value is lower than the measured value, input of the predetermined control value from the first control unit to the motor and the head is stopped, and the second automatically switching to initiate input of the variable control value from a controller to the motor and the head;
has
In the step k), the simulation model, while actually conveying the substrate and discharging the treatment substance onto the surface of the substrate, sets a control value related to the transportation and a control value related to the ejection, Based on the results of acquiring a plurality of the various information and the image of the surface of the base material after the treatment substance has been discharged, a control value related to the transport and a control value related to the discharge are set in advance on the simulator. A method of treating a base material, which is prepared by machine learning so as to output an estimated value of the error of the discharge position of the treatment substance on the surface of the base material from the control value and the various information with high accuracy.

Images