JP2023176204A - Information processing device - Google Patents

Abstract

To enhance estimation accuracy of information related to a liquid discharge head while shortening a processing time.SOLUTION: An information processing device includes a storage part for storing a learned model that performs machine learning on the basis of a data set obtained by associating first information for learning about a liquid discharge head for discharging liquid with second information for learning about the liquid discharge head, a reception part for receiving an input of the first information about the liquid discharge head used by a user, and an estimation part for estimating the second information about the liquid discharge head to be provided to the user on the basis of the learned model and the first information.SELECTED DRAWING: Figure 1

Description

The present invention relates to an information processing device.

Patent Document 1 discloses a drive waveform determination system that determines the waveform of a drive pulse using the results of measuring ink ejection characteristics by appropriately combining simulation and actual measurement.

JP2022-26028A

The technique described in Patent Document 1 has problems in that when a simulation is executed for each waveform candidate for a plurality of drive pulse waveform candidates, the process takes a long time and it is difficult to accurately estimate the ejection characteristics. be.

In order to solve the above problems, one aspect of the information processing device of the present invention associates first learning information about a liquid ejection head that ejects liquid with second learning information about the liquid ejection head. a storage unit that stores a trained model machine-trained based on a data set; a reception unit that receives input of first information regarding the liquid ejection head used by a user; and an estimation unit that estimates second information regarding the liquid ejection head to be provided to a user.

1 is a schematic diagram showing a configuration example of a system using an information processing device according to a first embodiment. FIG. 3 is a diagram showing a configuration example of a liquid ejection head. FIG. 3 is a diagram showing an example of a waveform of a drive pulse. FIG. 3 is a diagram for explaining actual measurement of ink ejection characteristics. 3 is a flowchart showing the operation of the information processing device according to the first embodiment. FIG. 3 is a diagram for explaining a trained model used in the first embodiment. FIG. 3 is a diagram for explaining machine learning for generating a learned model used in the first embodiment. FIG. 2 is a schematic diagram showing a configuration example of a system using an information processing device according to a second embodiment. FIG. 3 is a diagram showing an example of a waveform of a drive pulse for generating residual vibration. FIG. 3 is a diagram showing an example of a waveform of residual vibration. FIG. 7 is a diagram for explaining a learned model used in the second embodiment. FIG. 7 is a diagram for explaining machine learning that generates a learned model used in the second embodiment. FIG. 7 is a diagram showing the relationship between the number of datasets and the estimation accuracy of a trained model in the second embodiment. FIG. 3 is a schematic diagram showing an example of the configuration of a system using an information processing device according to a third embodiment. FIG. 7 is a diagram for explaining a trained model used in a third embodiment. FIG. 7 is a diagram for explaining machine learning for generating a learned model used in the third embodiment. 7 is a flowchart showing the operation of the information processing device according to Modification 1. FIG.

Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings. In the drawings, the dimensions and scale of each part may differ from the actual size, and some parts are shown schematically to facilitate understanding. Further, the scope of the present invention is not limited to these forms unless there is a statement that specifically limits the present invention in the following description.

1. First embodiment 1-1. Outline of System 100 FIG. 1 is a schematic diagram showing a configuration example of a system using an information processing device 400 according to the first embodiment. The system 100 automatically determines the waveform of a drive pulse PD used when ejecting ink, which is an example of a liquid.

As shown in FIG. 1, the system 100 includes a liquid ejecting device 200, a measuring device 300 that is an example of a “measuring unit”, and an information processing device 400. These will be explained in order below.

1-1a. Liquid discharge device 200
The liquid ejection device 200 is a printer that prints on a print medium using an inkjet method. The print medium is not particularly limited as long as it is a medium that can be printed by the liquid ejection device 200, and may be, for example, various types of paper, various types of cloth, or various types of films. Note that the liquid ejecting device 200 may be a serial type printer or a line type printer.

As shown in FIG. 1, the liquid ejection apparatus 200 includes a liquid ejection head 210, a moving mechanism 220, a power supply circuit 230, a drive signal generation circuit 240, a drive circuit 250, a storage circuit 260, and a "control unit". The processing circuit 270 is an example of the processing circuit 270.

The liquid ejection head 210 ejects ink toward a print medium. In FIG. 1, a plurality of piezoelectric elements 211, which are examples of "driving elements", are illustrated as components of the liquid ejection head 210. A configuration example of the liquid ejection head 210 will be described later based on FIG. 2. Note that instead of the piezoelectric element 211, a heater that heats the ink within the cavity may be used as the driving element.

In the example shown in FIG. 1, the number of liquid ejection heads 210 included in the liquid ejection apparatus 200 is one, but the number may be two or more. In this case, for example, two or more liquid ejection heads 210 are unitized. When the liquid ejection apparatus 200 is of a serial type, the liquid ejection head 210 or a unit including two or more thereof is used so that a plurality of nozzles are distributed over a part of the width direction of the print medium. Further, when the liquid ejection apparatus 200 is a line type, a unit including two or more liquid ejection heads 210 is used so that a plurality of nozzles are distributed over the entire width of the print medium.

The moving mechanism 220 changes the relative position of the liquid ejection head 210 and the print medium. More specifically, when the liquid ejection apparatus 200 is a serial type, the moving mechanism 220 includes a transport mechanism that transports the print medium in a predetermined direction, and a transport mechanism that transports the liquid ejection head 210 on an axis perpendicular to the transport direction of the print medium. and a moving mechanism for repeatedly moving along the same. Further, when the liquid ejection apparatus 200 is a line type, the moving mechanism 220 includes a transport mechanism that transports the print medium in a direction intersecting the longitudinal direction of a unit including two or more liquid ejection heads 210.

The power supply circuit 230 receives power from a commercial power supply (not shown) and generates various predetermined potentials. The generated various potentials are supplied to each part of the liquid ejection device 200 as appropriate. For example, the power supply circuit 230 generates a power supply potential VHV and an offset potential VBS. The offset potential VBS is supplied to the liquid ejection head 210 and the like. Further, the power supply potential VHV is supplied to the drive signal generation circuit 240 and the like.

The drive signal generation circuit 240 is a circuit that generates a drive signal Com for driving each piezoelectric element 211 included in the liquid ejection head 210.

Specifically, the drive signal generation circuit 240 includes, for example, a DA conversion circuit and an amplifier circuit. In the drive signal generation circuit 240, the DA conversion circuit converts the waveform designation signal dCom, which will be described later, from the processing circuit 270 from a digital signal to an analog signal, and the amplifier circuit uses the power supply potential VHV from the power supply circuit 230 to generate the analog signal. A drive signal Com is generated by amplifying the signal. Here, among the waveforms included in the drive signal Com, a waveform signal actually supplied to the piezoelectric element 211 is the drive pulse PD. Note that the drive pulse PD will be detailed later.

The drive circuit 250 is a transmission gate that switches whether or not to supply at least a part of the waveform included in the drive signal Com as a drive pulse PD for each of the plurality of piezoelectric elements 211 based on a control signal SI to be described later. etc. circuit.

The storage circuit 260 stores various programs executed by the processing circuit 270 and various data such as recorded data processed by the processing circuit 270. The storage circuit 260 includes, for example, volatile memory such as RAM (Random Access Memory) and non-volatile memory such as ROM (Read Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), or PROM (Programmable ROM). Contains one or both semiconductor memories. The recording data is supplied from the information processing device 400, for example. Note that the storage circuit 260 may be configured as a part of the processing circuit 270.

The processing circuit 270 has a function of controlling the operation of each part of the liquid ejection device 200 and a function of processing various data. The processing circuit 270 includes, for example, one or more processors such as CPUs (Central Processing Units). Note that the processing circuit 270 may include a programmable logic device such as an FPGA (field-programmable gate array) instead of or in addition to the CPU.

The processing circuit 270 controls the operation of each part of the liquid ejection device 200 by executing a program stored in the storage circuit 260. Here, the processing circuit 270 generates signals such as the control signals Sk and SI and the waveform designation signal dCom as signals for controlling the operation of each part of the liquid ejection apparatus 200. Further, the processing circuit 270 performs control based on recording data in order to apply a drive pulse PD having a waveform determined based on information output from the measuring device 300 to the liquid ejection head 210.

The control signal Sk is a signal for controlling the drive of the moving mechanism 220. Control signal SI is a signal for controlling driving of drive circuit 250. Specifically, the control signal SI determines whether or not the drive circuit 250 supplies the drive signal Com from the drive signal generation circuit 240 as a drive pulse PD to the liquid ejection head 210 based on the stored data. Specify for each unit period. This specification specifies the amount of ink etc. to be ejected from the liquid ejection head 210. The waveform designation signal dCom is a digital signal for defining the waveform of the drive signal Com generated by the drive signal generation circuit 240.

1-1b. Measuring device 300
The measurement device 300 is a device for measuring the ejection characteristics of ink from the liquid ejection head 210 when the drive pulse PD is actually used. Examples of the ejection characteristics include ejection speed, ink amount, number of satellites, and stability. Note that, hereinafter, the ejection characteristics of ink from the liquid ejection head 210 may be simply referred to as "ejection characteristics."

The measuring device 300 of this embodiment is an imaging device that images the state of the ink ejected from the liquid ejecting head 210 while it is in flight. Specifically, the measuring device 300 includes, for example, an imaging optical system and an imaging element. The imaging optical system is an optical system that includes at least one imaging lens, and may include various optical elements such as a prism, or may include a zoom lens, a focus lens, or the like. The image sensor is, for example, a CCD (Charge Coupled Device) image sensor or a CMOS (Complementary MOS) image sensor. The measurement of ejection characteristics using a captured image by the measuring device 300 will be described in detail later based on FIG. 3.

Note that in this embodiment, the measuring device 300 images the ink in flight, but the ejection characteristics such as the amount of ink ejected from the liquid ejection head 210 are measured based on the result of imaging the ink that has landed on a print medium or the like. It is also possible to do so. Furthermore, the measurement device 300 is not limited to an imaging device as long as it can obtain measurement results according to the ejection characteristics of ink from the liquid ejection head 210. An electronic balance or the like for measurement may also be used. Furthermore, as an information source for measuring the ink ejection characteristics from the liquid ejection head 210, in addition to the information from the measuring device 300, the results of detecting the waveform of residual vibration generated in the liquid ejection head 210 may be used. . The residual vibration is a vibration that remains in the ink flow path in the liquid ejection head 210 after the piezoelectric element 211 is driven, and is detected as a voltage signal from the piezoelectric element 211, for example.

1-1c. Information processing device 400
The information processing device 400 is a computer that controls the operations of the liquid ejection device 200 and the measurement device 300. Here, the information processing device 400 is connected to each of the liquid ejecting device 200 and the measuring device 300 wirelessly or by wire so that they can communicate with each other. Note that this connection may involve a communication network including the Internet.

As shown in FIG. 1, the information processing device 400 includes a display device 410, an input device 420, a storage circuit 430, and a processing circuit 440. These are communicably connected to each other.

Display device 410 displays various images under the control of processing circuit 440. Here, the display device 410 includes various display panels such as a liquid crystal display panel or an organic EL (electro-luminescence) display panel. Note that the display device 410 may be provided outside the information processing device 400. Further, the display device 410 may be a component of the liquid ejection device 200.

Input device 420 is a device that accepts operations from a user. For example, the input device 420 includes a pointing device such as a touch pad, a touch panel, or a mouse. Here, when the input device 420 has a touch panel, it may also serve as the display device 410. Note that the input device 420 may be provided outside the information processing device 400. Further, the input device 420 may be a component of the liquid ejection device 200.

The storage circuit 430 is a device that stores various programs executed by the processing circuit 440 and various data processed by the processing circuit 440. Storage circuit 430 includes, for example, a hard disk drive or a semiconductor memory. Note that part or all of the storage circuit 430 may be provided in a storage device, a server, or the like outside the information processing device 400.

The storage circuit 430 of this embodiment stores a program PG1, a learned model PJ1, first information D1, second information D2, measurement information D3, and determined waveform information DP. Note that part or all of the program PG1, the learned model PJ1, the first information D1, the second information D2, the measurement information D3, and the determined waveform information DP are stored in a storage device, a server, etc. external to the information processing device 400. You can.

Each of the first information D1 and the second information D2 is information regarding the liquid ejection head 210. However, the first information D1 is used as an input in estimation using the learned model PJ1, and is obtained by reception by the reception unit 441, which will be described later. In the present embodiment, the first information D1 includes information indicating waveform candidates of the drive pulse PD used to drive the liquid ejection head 210. On the other hand, the second information D2 is information different from the first information D1, and is generated by the estimation unit 442, which will be described later, as an output in estimation using the learned model PJ1. In the present embodiment, the second information D2 includes the amount of droplets ejected from the liquid ejection head 210, the ejection speed of the droplets ejected from the liquid ejection head 210, and the amount of liquid ejected from the liquid ejection head 210. It includes information indicating an estimated value of at least one of the stability of the droplet and the feature quantity of the image representing the state of the droplet ejected from the liquid ejection head 210. Details of the first information D1 and the second information D2 will be explained later based on FIG. 6.

The measurement information D3 is information regarding the results of measuring the ink ejection characteristics from the liquid ejection head 210 by actual measurement using the measurement device 300 described above. The measurement information D3 includes information indicating measurement results, as well as information regarding measurement conditions such as the waveform and temperature used for measurement, as appropriate.

The program PG1 causes the computer to execute a drive waveform determining method for determining the waveform of a drive pulse PD to be applied to a piezoelectric element 211 provided in a liquid ejection head 210 that ejects ink, which is an example of a liquid. The learned model PJ1 is used in this drive waveform determination method.

The learned model PJ1 is an estimation model obtained by machine learning the correspondence between the first information D1 and the second information D2. The device learning is performed based on a data set DST in which first learning information D1 and second learning information D2 are associated with each other, as described later. Details of the trained model PJ1 will be explained later based on FIGS. 6 and 7.

The determined waveform information DP is information regarding parameters such as voltage, potential, and slope of potential change that define the waveform, and indicates the waveform of the drive pulse PD. The determined waveform information DP is generated by the determining unit 454, which will be described later, based on the second information D2.

The processing circuit 440 is a device that has a function of controlling each part of the information processing device 400, the liquid ejecting device 200, and the measuring device 300, and a function of processing various data. The processing circuit 440 includes, for example, a processor such as a CPU (Central Processing Unit). Note that the processing circuit 440 may be composed of a single processor or a plurality of processors. Further, some or all of the functions of the processing circuit 440 may be realized by hardware such as a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array). You can.

The processing circuit 440 functions as a reception section 441, an estimation section 442, a determination section 443, and a determination section 444 by reading the program PG1 from the storage circuit 430 and executing it. Note that in this embodiment, an aspect will be described in which the receiving unit 441, the estimating unit 442, the determining unit 443, and the determining unit 444 are realized as functions of the same processing circuit 440; however, when the number of processing circuits 440 is plural , the reception unit 441, the estimation unit 442, the determination unit 443, and the determination unit 444 may be implemented as functions of different processing circuits 440.

The reception unit 441 receives input of first information D1 regarding the liquid ejection head 210 used by the user. For example, the receiving unit 441 displays an image for a GUI (Graphical User Interface) on the display device 410, and receives input of the first information D1 or information necessary for its generation according to the operation of the input device 420 based on the image. .

The estimation unit 442 estimates second information D2 to be provided to the user based on the learned model PJ1 and the first information D1.

The determination unit 443 determines whether the waveform candidate indicated by the first information D1 should be the waveform of the drive pulse PD applied to the liquid ejection head 210, based on the estimated value indicated by the second information D2.

The determining unit 444 generates determined waveform information DP by determining the waveform of the drive pulse PD based on the first information D1 and the second information D2. Specifically, the determining unit 444 has a function of evaluating the ejection characteristics indicated by the second information D2 and the ejection characteristics measured by the measuring device 300, and based on the result of evaluating the ejection characteristics indicated by the second information D2, It has a function of determining whether or not to perform measurement using the measuring device 300, and adjusting to optimize waveform candidates used for measurement.

The determining unit 444 performs the estimation by the estimating unit 442 as much as possible by determining whether or not to perform the measurement using the measuring device 300 based on the result of evaluating the ejection characteristics indicated by the second information D2. Actual measurement is performed by the measuring device 300. Therefore, it is possible to reduce the possibility that the actual measurement is performed more times than necessary or under inappropriate conditions. In addition, the determining unit 444 finally determines the waveform of the drive pulse PD that provides the desired ejection characteristics by optimizing the waveform candidates used for measurement based on the results of the evaluation.

Therefore, when the determining unit 443 determines that the waveform candidate indicated by the first information D1 should be the waveform of the drive pulse PD applied to the liquid ejection head 210, the measuring device 300 sets the drive pulse PD based on the waveform candidate. By applying the voltage to the liquid ejection head 210, information indicating actual measured values regarding ejection characteristics is output. Here, the ejection characteristics include the ejection amount of liquid droplets ejected from the liquid ejection head 210, the ejection speed of the liquid droplets ejected from the liquid ejection head 210, and the ejection rate of the liquid droplets ejected from the liquid ejection head 210. It includes at least one of stability and a feature amount of an image representing the state of a droplet ejected from the liquid ejection head 210.

1-2. Configuration Example of Liquid Discharge Head 210 FIG. 2 is a diagram showing a configuration example of the liquid discharge head 210. The liquid ejection head 210 includes a flow path substrate 212, a pressure chamber substrate 213, a nozzle plate 214, a vibration absorber 215, a vibration plate 216, a plurality of piezoelectric elements 211, a cover 217, a case 218, and a wiring board 219.

Here, a pressure chamber substrate 213, a diaphragm 216, a plurality of piezoelectric elements 211, a case 218, and a cover 217 are installed in a region located in the Z1 direction from the flow path substrate 212. On the other hand, a nozzle plate 214 and a vibration absorber 215 are installed in a region located in the Z2 direction relative to the channel substrate 212. Each element of the liquid ejection head 210 is generally a plate-like member elongated in the direction along the Y-axis, and is bonded to each other using, for example, an adhesive.

As shown in FIG. 2, the nozzle plate 214 is a plate-like member provided with a plurality of nozzles N arranged in the direction along the Y-axis. Each nozzle N is a through hole through which ink passes. The nozzle plate 214 is manufactured, for example, by processing a silicon single crystal substrate using a semiconductor manufacturing technique using a processing technique such as dry etching or wet etching. However, other known methods and materials may be used as appropriate to manufacture the nozzle plate 214.

The channel substrate 212 is a plate-like member for forming an ink channel. As shown in FIG. 2, the channel substrate 212 is provided with an opening R1, a plurality of supply channels Ra, and a plurality of communication channels Na. The opening R1 is an elongated through hole that extends in the direction along the Y axis when viewed from above in the direction along the Z axis so as to be continuous across the plurality of nozzles N. On the other hand, each of the supply channel Ra and the communication channel Na is a through hole provided for each nozzle N. Each of the plurality of supply channels Ra communicates with the opening R1. Similarly to the nozzle plate 214 described above, the channel substrate 212 is manufactured, for example, by processing a silicon single crystal substrate using semiconductor manufacturing technology. However, other known methods and materials may be used as appropriate to manufacture the channel substrate 212. Note that a part of the supply channel Ra may be formed on the pressure chamber substrate 213.

The pressure chamber substrate 213 is a plate-like member in which a plurality of pressure chambers C corresponding to a plurality of nozzles N are formed. The pressure chamber C is a space called a cavity located between the flow path substrate 212 and the diaphragm 216 and for applying pressure to the ink filled in the pressure chamber C. The plurality of pressure chambers C are arranged in the direction along the Y-axis. Each pressure chamber C is composed of holes opening on both sides of the pressure chamber substrate 213, and has a long shape extending in the direction along the X axis. The end of each pressure chamber C in the X2 direction communicates with the corresponding supply channel Ra. Note that the cross-sectional area of the supply flow path Ra is narrower than that of the pressure chamber C, and this portion functions as a flow path resistance, so that backflow when pressure is applied to the ink is suppressed. On the other hand, the end of each pressure chamber C in the X1 direction communicates with the corresponding communication channel Na. Like the nozzle plate 214 described above, the pressure chamber substrate 213 is manufactured, for example, by processing a silicon single crystal substrate using semiconductor manufacturing technology. However, other known methods and materials may be used as appropriate to manufacture each of the pressure chamber substrates 213.

A diaphragm 216 is arranged on the surface of the pressure chamber substrate 213 facing the Z1 direction. The diaphragm 216 is an elastically deformable plate member. In the example shown in FIG. 4, the diaphragm 216 has a first layer 216a that is an elastic film and a second layer 216b that is an insulating film, which are stacked in this order in the Z1 direction. The first layer 216a is, for example, an elastic film made of silicon oxide (SiO ₂ ). The elastic film is formed, for example, by thermally oxidizing one surface of a silicon single crystal substrate. The second layer 216b is, for example, an insulating film made of zirconium oxide (ZrO ₂ ). The insulating film is formed, for example, by forming a zirconium layer by sputtering and thermally oxidizing the layer. Note that the configuration of the diaphragm 216 is not limited to the example shown in FIG. 2.

A plurality of piezoelectric elements 211 corresponding to different nozzles N or pressure chambers C are arranged on the surface of the diaphragm 216 facing the Z1 direction. Each piezoelectric element 211 has an elongated shape extending in the direction along the X-axis. The plurality of piezoelectric elements 211 are arranged in the direction along the Y-axis so as to correspond to the plurality of pressure chambers C. When the diaphragm 216 vibrates in conjunction with the deformation of the piezoelectric element 211, the pressure within the pressure chamber C fluctuates, causing ink to be ejected from the nozzle N.

The case 218 is a case for storing ink to be supplied to the plurality of pressure chambers C, and is bonded to the surface of the channel substrate 212 facing the Z1 direction with an adhesive or the like. Case 218 is made of, for example, a resin material and manufactured by injection molding. The case 218 is provided with a housing portion R2 and a supply port IH. The housing portion R2 is a recessed portion having an outer shape corresponding to the opening portion R1 of the channel substrate 212. The supply port IH is a through hole that communicates with the housing portion R2. The space formed by the opening R1 and the accommodating part R2 functions as a liquid storage chamber R that is a reservoir for storing ink. Ink from the liquid container 10 is supplied to the liquid storage chamber R via the supply port IH.

The vibration absorber 215 is an element for absorbing pressure fluctuations within the liquid storage chamber R. The vibration absorber 215 is, for example, a compliance substrate that is a flexible sheet member that can be elastically deformed. Here, the vibration absorber 215 is attached to a surface of the channel substrate 212 facing in the Z2 direction so as to close the opening R1 of the channel substrate 212 and the plurality of supply channels Ra and constitute the bottom surface of the liquid storage chamber R. will be placed in

The cover 217 is a structure that protects the plurality of piezoelectric elements 211 and reinforces the mechanical strength of the pressure chamber substrate 213 and the diaphragm 216. The cover 217 is bonded to the surface of the diaphragm 216 using, for example, an adhesive. The cover 217 is provided with a recess that accommodates the plurality of piezoelectric elements 211.

A wiring board 219 is bonded to the surface of the pressure chamber board 213 or the diaphragm 216 facing the Z1 direction. The wiring board 219 is a mounting component on which a plurality of wirings for electrically connecting the processing circuit 270 and the like to the liquid ejection head 210 are formed. The wiring board 219 is, for example, a flexible wiring board such as FPC (Flexible Printed Circuit) or FFC (Flexible Flat Cable). A drive circuit 250 is mounted on the wiring board 219.

1-3. Waveform Example of Drive Pulse PD FIG. 3 is a diagram showing an example of the waveform of the drive pulse PD. FIG. 3 shows a temporal change in the potential of the drive pulse PD, that is, a voltage waveform of the drive pulse PD. Note that the waveform of the drive pulse PD is not limited to the example shown in FIG. 3, but is arbitrary.

As shown in FIG. 3, the drive pulse PD is included in the drive signal Com every unit period Tu. In the example shown in FIG. 3, the potential E of the drive pulse PD increases from the reference potential E1 to the potential E2, then decreases to a potential E3 lower than the potential E1, and then returns to the potential E1.

More specifically, the potential E of the drive pulse PD is first maintained at potential E1 over a period from timing t0 to timing t1, and then increases to potential E2 over a period from timing t1 to timing t2. Then, the potential E of the drive pulse PD is maintained at potential E2 over a period from timing t2 to timing t3, and then drops to potential E3 over a period from timing t3 to timing t4. Thereafter, after being maintained at the potential E3 over a period from timing t4 to timing t5, it rises to potential E1 over a period from timing t5 to timing t6.

The drive pulse PD having such a waveform increases the pressure chamber of the liquid ejection head 210 during the period from timing t1 to timing t2, and rapidly decreases the volume of the pressure chamber during the period from timing t3 to timing t4. Due to such a change in the volume of the pressure chamber, a portion of the ink within the pressure chamber is ejected from the nozzle as droplets.

The waveform of the drive pulse PD as described above can be expressed by a function using parameters p1, p2, p3, p4, p5, p6, and p7 corresponding to each of the above-mentioned periods. When the waveform of the drive pulse PD is defined by the function, the waveform of the drive pulse PD can be adjusted by changing each parameter. By adjusting the waveform of the drive pulse PD, the ejection characteristics of ink from the liquid ejection head 210 can be adjusted.

1-4. Actual Measurement of Ink Ejection Characteristics FIG. 4 is a diagram for explaining actual measurement of ink ejection characteristics. As shown in FIG. 4, the measuring device 300 images the flying state of the ink droplet DR ejected from the nozzle N of the liquid ejection head 210 from a direction perpendicular to or intersecting the ejection direction.

In the example shown in FIG. 4, the liquid ejection head 210 is provided with a nozzle surface 214a through which the nozzle N opens. The nozzle surface 214a is usually installed parallel to the printing surface of the recording medium M.

The droplet DR is the main droplet ejected from the nozzle N. In the example shown in FIG. 4, in addition to the droplet DR, a plurality of droplets DRa called satellites are ejected from the nozzle N, which are generated secondary to the droplet DR as the droplet DR is generated. The droplet DRa has a smaller diameter than the droplet DR, and the presence or absence, number, size, etc. of the droplet DRa vary depending on the type of ink, the waveform of the drive pulse PD, etc.

The measuring device 300 images the flying droplet DR continuously or intermittently at minute time intervals. Based on the result of this imaging, the arrival timing of the droplet DR to the recording medium M can be measured. Also, based on the measurement results of the measuring device 300, the position of the droplet DR at each predetermined timing is measured, and the ejection direction, ejection speed, or landing position of the droplet DR is determined based on the position at a plurality of timings. It can also be measured. Furthermore, when a plurality of droplets DR are continuously ejected from the nozzle N, it is possible to measure whether or not the plurality of droplets DR coalesce, and to measure the order in which they coalesce, based on the measurement results of the measuring device 300. You can also

The timing at which the flight distance of the droplet DR from the liquid ejection head 210 reaches a predetermined distance may be calculated based on the time when the flight distance of the droplet DR actually reaches the predetermined distance, or It may be calculated based on the ejection speed and the predetermined distance. Here, if the predetermined distance is the distance PG between the nozzle surface 214a and the recording medium M, the timing at which the droplet DR reaches the recording medium M is measured.

The ejection amount, which is the amount of the droplet DR from the liquid ejection head 210, is calculated as the volume of the droplet DR based on the diameter LB of the droplet DR, for example, using the captured image of the measuring device 300. Further, the ejection speed of the droplet DR from the liquid ejection head 210 is calculated based on, for example, the distance LC between two arbitrary positions of the flying droplet DR and the time. In FIG. 4, the droplet DR after the predetermined time is indicated by a chain double-dashed line. Furthermore, the aspect ratio (LA/LB) of ink from the liquid ejection head 210 can also be calculated as the ink ejection characteristic. The ejection angle of ink from the liquid ejection head 210 can also be determined from the positional relationship of the droplets DR before and after the predetermined time. Note that the amount of the droplet DR from the liquid ejection head 210 may be calculated as the mass of the droplet DR based on the diameter LB of the droplet DR and the density of the droplet DR.

1-5. Flow of Waveform Determination of Drive Pulse PD FIG. 5 is a flowchart showing the operation of the information processing device 400 according to the first embodiment. As shown in FIG. 5, first, in step S110, the determining unit 444 sets a target value for the ejection characteristics. This setting is automatically performed by a user's input using the input device 420 described above or by executing the program PG1.

Then, in step S120, the determining unit 444 sets waveform candidates based on the target value or evaluation value. With this setting, the receiving unit 441 receives input of the first information D1. Note that in step S120, if there is no evaluation value or a waveform candidate based on the evaluation value, the initial waveform based on the target value is set as the waveform candidate; on the other hand, if there is an evaluation value or a waveform candidate based on it, the waveform candidate based on the evaluation value is set as the waveform candidate. Set. Note that the waveform candidates may be set using other methods, and may be generated randomly, for example.

Next, in step S130, the determining unit 444 causes the estimating unit 442 to estimate the ejection characteristics using the learned model PJ1. Then, in step S140, the determining unit 444 stores the estimation result in the storage circuit 430 as second information D2. Thereafter, in step S150, the determining unit 444 calculates the evaluation value of the evaluation function using the ejection characteristics indicated by the second information D2.

To evaluate the ejection characteristics, for example, an evaluation function is used that is minimum or maximum when a predetermined ejection characteristic is within a desired value or range, and the result of the evaluation is used as an evaluation value that is a calculated value of the evaluation function. expressed. The evaluation function uses a linear sum of terms related to the predetermined ejection characteristics. The evaluation function of this embodiment uses a linear sum of a term related to ejection speed and a term related to ink amount. Further, the parameters of the evaluation function are parameters p1, p2, p3, . . . related to the waveform of the drive pulse PD described above.

To explain more specifically, an example of the evaluation function f(x) is:
f(x)=W1×(Vm(x)−Vmtarget) ² +W2×(Iw(x)−Iwtarget) ²
It is expressed as

Here, in the evaluation function f(x), x is the parameter p1, p2, p3, . Vm(x) is a measured value of ejection speed through simulation. Iw(x) is a measured value of ink amount by simulation. Vmtarget is the target value of the ejection speed. Iwtarget is the target value of the ink amount. W1 and W2 are weighting factors, respectively. In this example, the evaluation function f(x) is evaluated based on the ink amount and the ejection speed, but the evaluation may also be performed using ejection stability, inclination of the ejection direction, etc.

Note that, based on the evaluation value of the evaluation function, the waveform candidate is adjusted so that the measurement result approaches the target ejection characteristic. This adjustment is actually reflected on the waveform candidate when it is determined in step S160, which will be described later, to return to step S120.

To adjust the waveform candidates, an optimization algorithm such as Bayesian optimization or the Nelder-Mead method, which minimizes the evaluation value of the evaluation function based on the measured ejection characteristics, is used, for example.

When Bayesian optimization is used to adjust waveform candidates, acquisition functions such as EI (Expected Improvement), PI (Probability of Improvement), UCB (Upper Confidence Bound), and PES (Predictive Entropy Search) are used to determine the parameters p1, p2, By searching p3, . . . , waveform candidates are found.

Here, the characteristics of the obtained waveform candidates differ depending on the type of acquisition function used. As a general tendency, waveform candidates obtained using the acquisition function EI are waveforms with a high expected value of improvement amount. The waveform candidates obtained using the acquisition function PI are waveforms with a high probability of improvement but with a small amount of improvement. The waveform candidates obtained using the acquisition function UCB are waveforms that have a lot of room for improvement, but also have a lot of room for deterioration.

Since the Nelder-Mead method is a local optimization algorithm, it is suitable when using an existing waveform for the drive pulse PD to slightly change the physical properties of the ink or the desired ejection characteristics.

After the above step S150, in step S160, the determining unit 444 determines whether or not the waveform candidate is worth measuring by the measuring device 300 based on the criteria described below. If there is no value, the process returns to step S120 described above. That is, the determining unit 444 repeats steps S120 to S160 described above until the value is found.

The criteria for determination in step S160 are, for example, that the discharge is possible normally without the inclusion of air bubbles, that no discharge failure occurs later, and that it is worth actually measuring. . That is, the determining unit 444 causes the measuring device 300 to perform actual measurement if the ink can be ejected normally, does not cause subsequent ejection failure, and is worth actual measurement, and if not, returns to step S120. Using the waveform candidate adjusted as described above as the next waveform candidate, ejection characteristic estimation using the learned model PJ1 is performed again. Specific examples of whether or not there is value to actually measure are as follows.

For example, if the amount of ink indicated by the estimation result by the learned model PJ1 is less than a predetermined threshold value, it is presumed that normal ejection has not been performed, and therefore it is determined that there is no value in actual measurement yet. On the other hand, if the amount of ink is equal to or greater than a predetermined threshold value, it is determined that it is worth actually measuring.

In addition, for example, the range of waveforms that are likely to cause ejection failures or that are impractical due to constraints based on hardware lifespan, safety, etc. can be defined in advance using the parameter inequalities described above, and the waveform candidates can be set within that range. If it is within the range, it is presumed that there is nonconformity without actually measuring, and it is determined that there is no value in actually measuring yet. On the other hand, if the waveform candidate is outside the range, it is determined that it is worth actually measuring.

For example, if the difference between the ejection characteristics shown by the estimation results using the trained model PJ1 and the target value is greater than a predetermined value, it is assumed that there is sufficient room for improvement of the waveform candidates, so it is still worth actually measuring. It is determined that there is no. On the other hand, if the difference between the ejection characteristic and the target value is less than a predetermined value, it is determined that it is worth actually measuring.

If it is determined in step S160 that there is value, in step S170, the determining unit 444 causes the measuring device 300 to perform measurement. Then, in step S180, the determining unit 444 stores the measurement result in the storage circuit 430 as measurement information D3.

Thereafter, in step S190, the determining unit 444 calculates the evaluation value of the evaluation function using the ejection characteristics indicated by the measurement information D3. The evaluation in step S190 uses the same evaluation function f(x) as that used in the evaluation in step S150. However, in step S190, Vm(x) is the actually measured value of the ejection speed, and Iw(x) is the actually measured value of the ink amount. Further, based on the evaluation value of the evaluation function, the waveform candidate is adjusted so that the measurement result approaches the target ejection characteristic. The adjustment method is the same as step S150. This adjustment is actually reflected on the waveform candidate when it is determined in step S200, which will be described later, to also return to step S120.

Next, in step S200, the determining unit 444 determines whether to end. This determination is made based on whether the measurement result in step S180 is within a predetermined range with respect to the target value. If the measurement result is not within the predetermined range with respect to the target value, the determining unit 444 returns to step S120 described above. On the other hand, if the determined result is within a prescribed range with respect to the target value, the determination unit 444 determines the last set waveform candidate as the waveform of the drive pulse PD, and ends the process.

1-6. Trained Model FIG. 6 is a diagram for explaining the trained model PJ1 used in the first embodiment. The learned model PJ1 is an estimation model that outputs second information D2 in response to input of first information D1. Specifically, the learned model PJ1 is realized by a combination of a program that causes the processing circuit 440 to execute an operation to generate the second information D2 from the first information D1, and a plurality of coefficients applied to the operation. . The program is, for example, a program module that constitutes artificial intelligence software. The plurality of coefficients are set, for example, by deep learning using the data set DST, which will be described later. In FIG. 6, a case where the trained model PJ1 is a mathematical model such as a deep neural network having an input layer, an output layer, and an intermediate layer is shown as a preferred example. Note that the number of intermediate layers is not limited to the example shown in FIG. 6, but is arbitrary.

In the example shown in FIG. 6, the first information D1 includes waveform candidate information D1a, head type information D1b, liquid type information D1c, and environment information D1d. On the other hand, the second information D2 includes ejection amount information D2a, ejection speed information D2b, stability information D2c, and image characteristic information D2d.

Note that the first information D1 is not limited to the example shown in FIG. 6, and may not include at least one of the head type information D1b, the liquid type information D1c, and the environment information D1d. Further, the second information D2 is not limited to the example shown in FIG. 6, and may include at least one of the ejection amount information D2a, the ejection speed information D2b, the stability information D2c, and the image characteristic information D2d. .

The waveform candidate information D1a is information indicating a waveform candidate of the drive pulse PD used to drive the liquid ejection head 210. More specifically, the waveform candidate information D1a is information indicating a plurality of mutually different candidate waveforms, and is information regarding parameters such as voltage, potential, and slope of potential change that define the waveform.

Head type information D1b is information indicating the type of liquid ejection head 210. More specifically, the head type information D1b is identification information such as a serial number unique to the liquid ejection head 210.

The liquid type information D1c is information indicating the type of liquid ejected from the liquid ejection head 210. For example, the liquid type information D1c is information regarding the ink product number, viscosity, residual vibration, ejection speed, etc.

The environment information D1d is information indicating the environment in which the liquid ejection head 210 is used. For example, the environmental information D1d is information about the temperature in the installation environment of the liquid ejection head 210, or information about the temperature detected by a temperature sensor provided at or around the liquid ejection head 210.

The ejection amount information D2a is information indicating an estimated value of the ejection amount of droplets ejected from the liquid ejection head 210. The ejection speed information D2b is information indicating an estimated value of the ejection speed of droplets ejected from the liquid ejection head 210. The stability information D2c is information indicating an estimated value of the stability of droplets ejected from the liquid ejection head 210. The image feature information D2d is information indicating an estimated value of the feature amount of an image representing the state of a droplet ejected from the liquid ejection head 210.

FIG. 7 is a diagram for explaining machine learning for generating the trained model PJ1 used in the first embodiment. The trained model PJ1 is generated by machine learning using a learning device having a learning processing section 500, as shown in FIG. Multiple datasets DST are used for this machine learning. Each data set DST includes first information D1 and second information D2 corresponding thereto. The first information D1 in the data set DST is an example of "first information for learning." Similarly, the second information D2 in the data set DST is an example of "second information for learning." Note that the first information D1 and the second information D2 in the data set DST may be generated using a device separate from the information processing device 400 described above. An example of generating a plurality of data sets DST will be described below.

In order to generate multiple data sets DST, first, a set of parameters such as voltage and time parameters that define the waveform, physical property values of ink viscosity, surface tension and density, and dimensions of the flow path of the liquid ejection head 210 are required. Multiple sets are generated using a sampling method such as random sampling or Latin Hypercube Sampling. The number of such groups is, for example, 1000 or more and 10000 or less. In this sampling method, sampling distributed over the entire parameter space is desirable. Next, the behavior of the ink is simulated by inputting the set of parameters obtained through sampling into software for computational fluid dynamics simulation. As a result of the simulation, data such as pressure distribution, mass distribution, flow velocity distribution, etc. are obtained in time series. A plurality of data sets DST can be obtained by using a database of combinations of parameters used in the simulations and their results.

Here, the aforementioned database is appropriately classified into learning data, test data, verification data, etc., and the learning data is used as the data set DST. Note that data may be added to the data set DST during the learning process described later. Data set DST may also include data such as ink droplet velocity, mass, number of satellites, etc. obtained by post-processing of simulation results. Further, the data set DST may include data showing an image such as a contour diagram that visualizes the simulation result.

Note that although the plurality of data sets DST are generated by simulation in the above, they may be generated by actual measurement. When the data set DST is generated through actual measurement, the cost increases due to the amount of ink consumed, but the accuracy of the data is improved compared to when the data set DST is generated through simulation. Note that the data set DST may be generated using both simulation and actual measurement.

The learning processing unit 500 sets a plurality of coefficients of the learned model PJ1 by supervised machine learning using a plurality of data sets DST. Specifically, the learning processing unit 500 uses the second information D2X output by the provisional learned model PJ1 in response to the input of the first information D1 in the dataset DST, and the second information D2 included in the dataset DST. A plurality of coefficients of the learned model PJ1 are updated so that the difference between the learned model PJ1 and the learned model PJ1 is reduced. For example, the learning processing unit 500 iteratively updates a plurality of coefficients of the learned model PJ1 using the error backpropagation method so that the evaluation function representing the difference between the first information D1 and the second information D2 is minimized. do. The learned model PJ1 after the above machine learning is statistically applied to the unknown first information D1 based on the latent tendency between the first information D1 and the second information D2 in the plurality of data sets DST. The appropriate second information D2 is output.

Note that in addition to deep learning, Gaussian process regression, random forest regression, or the like may be used for learning. Furthermore, when outputting an animation using deep learning, learning is performed to generate an image at time t+Δt from an image at time t. Here, the inference program receives as input an image at time t=0 in addition to waveform parameters, etc., and sequentially outputs images after Δt.

With the trained model PJ1 as described above, detailed ejection characteristics of complex waveforms can be predicted in a short time. On the other hand, when using a method such as the equivalent circuit method that can be simulated within a short time of several seconds, it is difficult to estimate the ejection characteristics of a complicated waveform. Furthermore, in this case, the types of information that can be estimated are limited, and it is not possible to obtain the number of satellites, detailed pressure distribution, etc. On the other hand, when using computational fluid dynamics simulation software such as the finite volume method, it is possible to estimate the discharge characteristics of minute waveforms, but it takes a long time, from 10 minutes to several hours, to obtain the simulation results. Therefore, there are practical problems.

As described above, the information processing device 400 includes the storage circuit 430, which is an example of a “storage unit,” the reception unit 441, and the estimation unit 442. The storage circuit 430 stores the learned model PJ1 that has been subjected to machine learning based on the data set DST. The data set DST associates first information D1 for learning about the liquid ejection head 210 that ejects ink, which is an example of "liquid", and second information D2 for learning about the liquid ejection head 210. be. The reception unit 441 receives input of first information D1 regarding the liquid ejection head 210 used by the user. The estimation unit 442 estimates second information D2 regarding the liquid ejection head 210 to be provided to the user, based on the learned model PJ1 and the first information D1.

In the above information processing device 400, the second information D2 is estimated based on the learned model PJ1 and the first information D1, so compared to the case where the second information D2 is estimated based on the first information D1 using simulation. As a result, processing time can be shortened and the accuracy of estimating ejection characteristics can be improved.

In this embodiment, as described above, the first information D1 includes information indicating waveform candidates of the drive pulse PD used to drive the liquid ejection head 210. The second information D2 includes the amount of droplets ejected from the liquid ejection head 210, the ejection speed of the droplets ejected from the liquid ejection head 210, and the stability of the droplets ejected from the liquid ejection head 210. , a feature amount of an image representing the state of a droplet ejected from the liquid ejection head 210, and information indicating an estimated value of at least one of the following. Therefore, the ejection characteristics of the liquid ejection head 210 can be estimated based on the waveform candidates.

Further, as described above, the first information D1 indicates any one of the type of liquid ejection head 210, the type of liquid ejected from the liquid ejection head 210, and the usage environment of the liquid ejection head 210. Contains more information. Therefore, the accuracy of estimating the ejection characteristics based on the waveform candidates can be improved.

Furthermore, as described above, the information processing device 400 further includes the determination unit 443. The determination unit 443 determines whether the waveform candidate indicated by the first information D1 should be the waveform of the drive pulse PD applied to the liquid ejection head 210, based on the estimated value indicated by the second information D2. Therefore, actual measurements of ejection characteristics are less likely to be performed more times than necessary or under inappropriate conditions. As a result, it is possible to prevent ink from being wasted and to shorten the time required to determine the PD waveform of the drive pulse.

Further, as described above, the information processing device 400 is communicably connected to the measuring device 300, which is an example of a “measuring unit”, and the processing circuit 270, which is an example of a “control unit”. When the determining unit 443 determines that the waveform candidate indicated by the first information D1 should be the waveform of the drive pulse PD applied to the liquid ejection head 210, the measuring device 300 uses the drive pulse PD based on the waveform candidate to eject the liquid. By applying a voltage to the head 210, the amount of droplets ejected from the liquid ejection head 210, the ejection speed of the droplets ejected from the liquid ejection head 210, and the stability of the droplets ejected from the liquid ejection head 210 are controlled. The liquid ejection head 210 outputs information indicating an actual measured value of at least one of the characteristics of the droplet and the feature amount of the image representing the state of the droplet ejected from the liquid ejection head 210. The processing circuit 270 performs control based on recording data in order to apply a drive pulse PD having a waveform determined based on information output from the measuring device 300 to the liquid ejection head 210. Therefore, an image can be recorded by applying the drive pulse PD of the determined waveform to the liquid ejection head 210 based on the recording data.

2. Second Embodiment FIG. 8 is a schematic diagram showing a configuration example of a system 100A using an information processing device 400A according to a second embodiment. The system 100A is the same as the system 100 of the first embodiment described above, except that the measuring device 300 is omitted and the information processing device 400A is provided instead of the information processing device 400. Further, the information processing device 400A is the same as the information processing device 400 of the first embodiment described above, except that it uses a program PG2 instead of the program PG1 and uses a learned model PJ2 instead of the learned model PJ1. .

The storage circuit 430 of this embodiment stores a program PG2, a learned model PJ2, first information D1A, and second information D2A.

Each of the first information D1A and the second information D2A is information regarding the liquid ejection head 210. However, the first information D1A is used as an input in estimation using the learned model PJ2, and is obtained by reception by the reception unit 441A, which will be described later. In the present embodiment, the first information D1A includes information indicating residual vibrations generated within the pressure chamber C after the drive pulse PD is applied to the liquid ejection head 210. On the other hand, the second information D2A is different from the first information D1A, and is generated by the estimation unit 442A as an output in estimation using the trained model PJ2. In this embodiment, the second information D2A includes information indicating the physical properties of the liquid ejected from the liquid ejection head 210. Details of the first information D1A and the second information D2A will be explained later based on FIG. 11.

In this embodiment, the processing circuit 440 functions as a receiving section 441A and an estimating section 442A by reading the program PG2 from the storage circuit 430 and executing it.

The reception unit 441A receives input of first information D1A regarding the liquid ejection head 210 used by the user.

The estimation unit 442A estimates second information D2A to be provided to the user based on the learned model PJ2 and the first information D1A. The second information D2A that is the result of this estimation is used, for example, as the first information D1 of the first embodiment described above, as described in Modification 1 below. Furthermore, the processing circuit 440 may perform error notification, flushing, or cleaning based on the second information D2A.

FIG. 9 is a diagram showing an example of the waveform of the drive pulse PDt for generating residual vibration. Note that the waveform of the drive pulse PDt is not limited to the example shown in FIG. 9, but is arbitrary.

As shown in FIG. 9, the potential E of the drive pulse PDt increases from the reference potential E1 to the potential E2, and then returns to the potential E1.

More specifically, the potential E of the drive pulse PDt is first maintained at potential E1 over a period from timing t0 to timing t1, and then rises to potential E2 over a period from timing t1 to timing t2. Then, the potential E of the drive pulse PDt is maintained at the potential E2 over a period from timing t2 to timing t5, and then decreases to the potential E1 over a period from timing t5 to timing t6.

With the drive pulse PDt having such a waveform, it is possible to cause pressure fluctuations in the pressure chamber C without causing ink to be ejected from the nozzle N. Therefore, the waveform of the residual vibration obtained can be simplified and the accuracy of estimating the physical properties described later can be improved. The waveform of the residual vibration is obtained as a signal based on the electromotive force of the piezoelectric element 211 after applying the drive pulse PDt to the liquid ejection head 210. The residual vibration is measured, for example, every 0.1 [μs] over a period of 45 [μs].

FIG. 10 is a diagram showing an example of a waveform of residual vibration. As shown in FIG. 10, the waveform has a shape that periodically changes and attenuates. The residual vibration is a vibration with a natural frequency determined by the flow path resistance of the flow path through which ink flows in the liquid ejection head 210, the inertance of the ink in the flow path, the elastic compliance of the diaphragm 216, and the like. Here, the residual vibration of the diaphragm 216 is equivalent to the residual vibration of the ink (liquid).

FIG. 11 is a diagram for explaining the learned model PJ2 used in the second embodiment. The learned model PJ2 is an estimation model that outputs second information D2A in response to input of first information D1A. Specifically, the learned model PJ2 is realized by a combination of a program that causes the processing circuit 440 to execute an operation to generate the second information D2A from the first information D1A, and a plurality of coefficients applied to the operation. . The program is, for example, a program module that constitutes artificial intelligence software. The plurality of coefficients are set, for example, by deep learning using the data set DSTA, which will be described later. In FIG. 11, a case where the learned model PJ2 is a mathematical model such as a deep neural network having an input layer, an output layer, and an intermediate layer is shown as a preferred example. Note that the number of intermediate layers is not limited to the example shown in FIG. 11, but is arbitrary.

In the example shown in FIG. 11, the first information D1A includes residual vibration information D1e. On the other hand, the second information D2A includes liquid physical property information D2e.

The residual vibration information D1e is information indicating residual vibrations generated within the pressure chamber C after the drive pulse PDt is applied to the liquid ejection head 210. Specifically, the residual vibration information D1e includes information such as the waveform, amplitude, or period of the residual vibration. The liquid physical property information D2e is information indicating the physical properties of the liquid ejected from the liquid ejecting head 210. Specifically, the liquid physical property information D2e includes information indicating the viscosity of the liquid, information indicating the density of the liquid, and information indicating surface tension.

FIG. 12 is a diagram for explaining machine learning for generating the trained model PJ2 used in the second embodiment. The trained model PJ2 is generated by machine learning using a learning device having a learning processing section 500A, as shown in FIG. 12. Multiple datasets DSTA are used for this machine learning. Each data set DSTA includes first information D1A and second information D2A corresponding thereto. The first information D1A in the data set DSTA is an example of "first information for learning." Similarly, the second information D2A in the data set DSTA is an example of "second information for learning." Note that the first information D1A and the second information D2A in the data set DSTA may be generated using a device separate from the above-mentioned information processing device 400A.

The learning processing unit 500A sets a plurality of coefficients of the learned model PJ2 by supervised machine learning using a plurality of data sets DSTA. Specifically, the learning processing unit 500A uses the second information D2AX outputted by the provisional trained model PJ2 in response to the input of the first information D1A in the dataset DSTA, and the second information D2A included in the dataset DSTA. A plurality of coefficients of the trained model PJ2 are updated so that the difference between the learned model PJ2 and the learned model PJ2 is reduced. For example, the learning processing unit 500A repeatedly updates a plurality of coefficients of the learned model PJ2 using the error backpropagation method so that the evaluation function representing the difference between the first information D1A and the second information D2A is minimized. do. The trained model PJ2 after the above machine learning is statistically based on the unknown first information D1A based on the latent tendency between the first information D1A and the second information D2A in the plurality of data sets DSTA. The appropriate second information D2A is output.

FIG. 13 is a diagram showing the relationship between the number of data sets and the estimation accuracy of the learned model PJ2 in the second embodiment. As shown in FIG. 13, from the viewpoint of increasing the estimation accuracy of the trained model PJ2, the number of data sets in the data set DSTA is preferably 100 or more, and more preferably 200 or more and 500 or less.

When creating the data set DSTA, it is preferable to use sampling in which ink physical property values are distributed over the entire parameter space. Further, although actual measurement results are used to create the data set DSTA, it is preferable to use simulation results in combination. This makes it possible to reduce the consumption of ink required for creating the data set DSTA. Here, it is preferable to increase the database by collecting simulation data for most of the database used to create the data set DSTA, and to use the actual measurement data only for simulation verification. Thereby, it is possible to improve the prediction accuracy of the learned model PJ2 while increasing the number of data in the database.

Also according to the second embodiment described above, the accuracy of estimating information regarding the liquid ejection head 210 can be improved while reducing the processing time. In this embodiment, as described above, the first information D1A includes information indicating the residual vibration that occurs within the pressure chamber C after the drive pulse PD is applied to the liquid ejection head 210. The second information D2A includes information indicating the physical properties of the liquid ejected from the liquid ejection head 210. Therefore, the physical properties can be estimated based on the residual vibration.

Further, as described above, the information indicating the physical properties of the liquid discharged from the liquid discharge head 210 includes information indicating the viscosity of the liquid, information indicating the density of the liquid, and information indicating surface tension.
Therefore, information indicating the viscosity of the liquid, information indicating the density of the liquid, and information indicating the surface tension can be estimated based on the residual vibration.

3. Third Embodiment FIG. 14 is a schematic diagram showing a configuration example of a system 100B using an information processing device 400B according to a third embodiment. The system 100B is the same as the system 100 of the first embodiment described above, except that the measuring device 300 is omitted and the information processing device 400B is replaced with the information processing device 400. Further, the information processing device 400B is the same as the information processing device 400 of the first embodiment described above, except that it uses a program PG3 instead of the program PG1 and uses a learned model PJ3 instead of the learned model PJ1. .

The storage circuit 430 of this embodiment stores a program PG3, a learned model PJ3, first information D1A, and second information D2B.

The second information D2B is information regarding the liquid ejection head 210. More specifically, the second information D2B is information different from the first information D1A, and is generated by the estimation unit 442B as an output in estimation using the learned model PJ3. In this embodiment, the second information D2B includes information indicating the dimensions of the liquid ejection head 210. Details of the second information D2B will be explained later based on FIG. 15.

In this embodiment, the processing circuit 440 functions as a reception unit 441A and an estimation unit 442B by reading the program PG3 from the storage circuit 430 and executing it.

The estimation unit 442B estimates the second information D2B to be provided to the user based on the learned model PJ3 and the first information D1A. The second information D2B that is the result of this estimation is used, for example, as the first information D1 of the first embodiment described above, as described in Modification 1 below.

FIG. 15 is a diagram for explaining the trained model PJ3 used in the third embodiment. The learned model PJ3 is an estimation model that outputs the second information D2B in response to the input of the first information D1A. Specifically, the learned model PJ3 is realized by a combination of a program that causes the processing circuit 440 to execute an operation to generate the second information D2B from the first information D1A, and a plurality of coefficients applied to the operation. . The program is, for example, a program module that constitutes artificial intelligence software. The plurality of coefficients are set, for example, by deep learning using the data set DSTB, which will be described later. In FIG. 15, a case where the trained model PJ3 is a mathematical model such as a deep neural network having an input layer, an output layer, and an intermediate layer is shown as a preferred example. Note that the number of intermediate layers is not limited to the example shown in FIG. 15, but is arbitrary.

In the example shown in FIG. 15, the second information D2B includes head dimension information D2f.

The head dimension information D2f includes information indicating the dimensions of the liquid ejection head 210. Specifically, the head dimension information D2f includes information indicating the diameter of the nozzle N provided in the liquid ejection head 210, and information indicating the diameter of the supply port IH that supplies liquid to the liquid ejection head 210. .

FIG. 16 is a diagram for explaining machine learning for generating the learned model PJ3 used in the third embodiment. The trained model PJ3 is generated by machine learning using a learning device having a learning processing section 500B, as shown in FIG. 16. Multiple datasets DSTB are used for this machine learning. Each data set DSTB includes first information D1A and second information D2B corresponding thereto. The first information D1A in the data set DSTB is an example of "first information for learning." Similarly, the second information D2B in the data set DSTB is an example of "second information for learning." Note that the first information D1A and the second information D2B in the data set DSTB may be generated using a device separate from the above-mentioned information processing device 400B.

The learning processing unit 500B sets a plurality of coefficients of the learned model PJ3 by supervised machine learning using a plurality of data sets DSTB. Specifically, the learning processing unit 500B uses the second information D2BX output by the provisional learned model PJ3 in response to the input of the first information D1A in the data set DSTB, and the second information D2B included in the data set DSTB. A plurality of coefficients of the trained model PJ3 are updated so that the difference between the learned model PJ3 and the learned model PJ3 is reduced. For example, the learning processing unit 500B iteratively updates a plurality of coefficients of the learned model PJ3 using the error backpropagation method so that the evaluation function representing the difference between the first information D1A and the second information D2B is minimized. do. The learned model PJ3 after the above machine learning is statistically applied to the unknown first information D1A based on the latent tendency between the first information D1A and the second information D2B in the plurality of data sets DSTB. The appropriate second information D2B is output.

Also according to the third embodiment described above, the accuracy of estimating information regarding the liquid ejection head 210 can be improved while reducing the processing time. In the present embodiment, the first information D1A includes information indicating residual vibrations generated within the pressure chamber C after the drive pulse PD is applied to the liquid ejection head 210. The second information D2B includes information indicating the dimensions of the liquid ejection head 210. Therefore, the dimensions can be estimated based on the residual vibration.

Further, as described above, the information indicating the dimensions of the liquid ejection head 210 includes information indicating the diameter of the nozzle N provided in the liquid ejection head 210 and the diameter of the supply port IH that supplies liquid to the liquid ejection head 210. and information indicating. Therefore, the diameter of the nozzle N and the diameter of the supply port IH can be estimated based on the residual vibration.

4. Modifications Although the information processing apparatus of the present invention has been described above based on the illustrated embodiments, the present invention is not limited thereto. Further, the configuration of each part of the present invention can be replaced with any configuration that exhibits the same function as in the embodiment described above, or any configuration can be added.

3-1. Modification example 1
One or both of the second and third embodiments described above may be combined with the first embodiment described above. An example of this will be explained below.

FIG. 17 is a flowchart showing the operation of the information processing device according to the first modification. The information processing device has a configuration that is a combination of the first embodiment, the second embodiment, and the third embodiment. As shown in FIG. 17, the operation of the information processing apparatus is the same as that of the information processing apparatus 400 of the first embodiment, except that step S210 is added.

Step S210 is executed between step S110 and step S120. In step S210, after the reception unit 441A of the second embodiment or the third embodiment receives the first information D1A, the estimation unit 442A of the second embodiment generates the second information D2A, and the estimation unit 442A of the third embodiment generates the second information D2A. The estimation unit 442B generates second information D2B.

Here, the second information D2A is used as the liquid type information D1c of the first information D1 of the first embodiment. The second information D2B is used as the head type information D1b of the first information D1 of the first embodiment. According to the first modification described above, the accuracy of estimating information regarding the liquid ejection head 210 can be improved while reducing the processing time.

3-2. Modification example 2
In the above embodiment, the programs PG1, PG2, and PG3 are executed by a processing circuit provided in the same device as the installed storage circuit, but the configuration is not limited to this configuration, and the installed storage circuit The processing may be executed by a processing circuit provided in a different device. For example, the programs PG1, PG2, and PG3 may be executed by a computer that is communicably connected to the same device as the installed storage circuit via the Internet.

DESCRIPTION OF SYMBOLS 10...Liquid container, 100...System, 100A...System, 100B...System, 200...Liquid discharge device, 210...Liquid discharge head, 211...Piezoelectric element, 212...Flow path board, 213...Pressure chamber board, 214...Nozzle plate , 214a... Nozzle surface, 215... Vibration absorber, 216... Vibration plate, 216a... First layer, 216b... Second layer, 217... Cover, 218... Case, 219... Wiring board, 220... Movement mechanism, 230... Power supply circuit , 240... Drive signal generation circuit, 250... Drive circuit, 260... Memory circuit, 270... Processing circuit (control section), 300... Measuring device (measuring section), 400... Information processing device, 400... Information processing device, 400B... Information processing device, 410...Display device, 420...Input device, 430...Storage circuit (storage unit), 440...Processing circuit, 441...Reception unit, 441A...Reception unit, 442...Estimation unit, 442A...Estimation unit, 442B... Estimating section, 443... Judgment section, 444... Determining section, 454... Determining section, 500... Learning processing section, 500A... Learning processing section, 500B... Learning processing section, C... Pressure chamber, Com... Drive signal, D1... First Information, D1A...first information, D1a...waveform candidate information, D1b...head type information, D1c...liquid type information, D1d...environmental information, D1e...residual vibration information, D2...second information, D2A...second information, D2AX ...Second information, D2B...Second information, D2BX...Second information, D2X...Second information, D2a...Discharge amount information, D2b...Discharge speed information, D2c...Stability information, D2d...Image characteristic information, D2e...Liquid Physical property information, D2f...head dimension information, D3...measurement information, DP...determined waveform information, DR...droplet, DRa...droplet, DST...data set, DSTA...data set, DSTB...data set, E...potential, E1 ... Potential, E2... Potential, E3... Potential, EI... Acquisition function, IH... Supply port, LB... Diameter, LC... Distance, M... Recording medium, N... Nozzle, Na... Communication channel, PD... Driving pulse, PDt ...Drive pulse, PG...Distance, PG1...Program, PG2...Program, PG3...Program, PI...Acquisition function, PJ1...Learned model, PJ2...Learned model, PJ3...Learned model, R...Liquid storage chamber, R1 ...Opening part, R2...Accommodation part, Ra...Supply channel, S110...Step, S120...Step, S130...Step, S140...Step, S150...Step, S160...Step, S170...Step, S180...Step, S190...Step , S200...step, S210...step, SI...control signal, Sk...control signal, Tu...unit period, UCB...acquisition function, VBS...offset potential, VHV...power supply potential, dCom...waveform designation signal, p1...parameter, p2 ...parameter, p3...parameter, p4...parameter, p5...parameter, p6...parameter, t0...timing, t1...timing, t2...timing, t3...timing, t4...timing, t5...timing, t6...timing.

Claims (9)

a storage unit that stores a learned model machine-learned based on a data set in which first learning information regarding a liquid ejection head that ejects liquid and second learning information regarding the liquid ejection head are associated;
a reception unit that receives input of first information regarding the liquid ejection head used by a user;
an estimation unit that estimates second information regarding the liquid ejection head to be provided to a user based on the learned model and the first information;
An information processing device characterized by: The first information includes information indicating waveform candidates of drive pulses used to drive the liquid ejection head,
The second information includes the amount of droplets ejected from the liquid ejection head, the ejection speed of the droplets ejected from the liquid ejection head, and the stability of the droplets ejected from the liquid ejection head. , a feature amount of an image representing a state of a droplet ejected from the liquid ejection head, and information indicating an estimated value of at least one of the following:
The information processing device according to claim 1, characterized in that: The first information further includes information indicating any one of the type of the liquid ejection head, the type of liquid ejected from the liquid ejection head, and the usage environment of the liquid ejection head.
The information processing device according to claim 2, characterized in that: further comprising a determination unit that determines whether or not the waveform candidate should be a waveform of a drive pulse applied to the liquid ejection head, based on the estimated value indicated by the second information;
The information processing device according to claim 3, characterized in that: If the determining unit determines that the waveform candidate should be the waveform of a drive pulse applied to the liquid ejection head, the liquid ejection head is applied to the liquid ejection head by applying a drive pulse based on the waveform candidate to the liquid ejection head. the ejection amount of droplets ejected from the liquid ejection head, the ejection speed of the liquid droplets ejected from the liquid ejection head, the stability of the liquid droplets ejected from the liquid ejection head, and the liquid ejected from the liquid ejection head. a measurement unit that outputs information indicating an actual measurement value of at least one of the feature quantity of the image representing the state of the drop;
communicably connected to a control unit that performs control based on recording data to apply a drive pulse with a waveform determined based on information output from the measurement unit to the liquid ejection head;
The information processing device according to claim 4, characterized in that: The first information includes information indicating residual vibration that occurs within the pressure chamber after applying a drive pulse to the liquid ejection head,
The second information includes information indicating physical properties of the liquid ejected from the liquid ejection head.
The information processing device according to claim 1, characterized in that: The information indicating the physical properties of the liquid discharged from the liquid discharge head includes information indicating the viscosity of the liquid, information indicating the density of the liquid, and information indicating surface tension.
7. The information processing apparatus according to claim 6. The first information includes information indicating residual vibration that occurs within the pressure chamber after applying a drive pulse to the liquid ejection head,
The second information includes information indicating dimensions of the liquid ejection head.
The information processing device according to claim 1, characterized in that: The information indicating the dimensions of the liquid ejection head includes information indicating the diameter of a nozzle provided in the liquid ejection head, and information indicating the diameter of a supply port that supplies liquid to the liquid ejection head.
9. The information processing device according to claim 8.

Images