JP2021195209A - Learning method, conveyance device and image formation device - Google Patents

Abstract

To appropriately secure consistency when any sheet is discharged.SOLUTION: A learning method includes: an acquisition step of acquiring a sheet discharge consistency state of a sheet detected in the vicinity of a sheet placing tray; an evaluation step of evaluating the sheet discharge consistency state; a creation step of creating a learning model by machine learning by using a speed value corresponding to sheet discharge speed of the sheet of a sheet discharge part and a position of the sheet as inputs, using a speed value as an output, and using an evaluation result in the sheet discharge consistency state as a reward; and a control step of controlling the sheet discharge speed of the sheet discharge part on the basis of the learning model.SELECTED DRAWING: Figure 5

Description

The present invention relates to a learning method for generating a learning model for a transfer device, a transfer device to which the learning method is applied, and an image forming apparatus provided with the transfer device.

In the field of commercial printing equipment, it is required to provide output products that meet the needs of customers. Therefore, there are various demands for media and printing machines used for printing. In order to meet these demands, it is necessary to control the printing press according to individual circumstances, but at present, because it relies on manual design, it is not possible to meet all demands. Traditional manual design is limited to the worst and typical conditions.

In addition, there are a wide variety of paper types for printing on the printing press, and when a paper type that was not expected in the initial design of development is passed through, the paper cannot be properly output on the paper loading tray. Things will happen. For example, when printing on multiple sheets of paper with a printing machine and storing the printed multiple sheets of paper on a paper mounting tray, if the paper transport speed at the time of paper ejection is not appropriate, one sheet 1 The paper loading position on the paper loading tray of the sheets of paper is displaced, and the consistency at the time of paper ejection cannot be ensured, resulting in an unfavorable state.

Patent Document 1 describes a technique for an image forming apparatus that forms an image on paper by controlling the ejection speed according to the ejection angle of the paper or the type of the paper to ensure consistency at the time of ejecting the paper. There is.

Japanese Unexamined Patent Publication No. 2012-206815

As described in Patent Document 1, it has been conventionally known to ensure consistency at the time of paper ejection by switching the ejection speed at the time of ejecting the paper according to the type of the paper and the like.
When considering the image formation of various types of paper with an image forming device as a configuration for switching according to the type of paper in this way, the appropriate ejection speed for various types of paper is determined in advance. It is necessary to measure, confirm, and reflect in the design data. For this reason, in order to support various types of paper, it is necessary to confirm the paper ejection state of various types of paper in advance before designing, and it takes an enormous amount of time to design the image forming apparatus. It ends up.

In addition, when printing is performed on paper of a paper type that was not expected when developing the printing press, it is not possible to determine what kind of ejection speed should be set, so consistency at the time of printing is improved. I couldn't secure it.

An object of the present invention is to provide a learning method, a transport device, and an image forming device that can appropriately secure the consistency when ejecting paper regardless of the paper.

The learning method of the present invention is a learning method for a transport device having a paper ejection unit for ejecting paper to a paper loading tray, and acquires a paper ejection matching state of the paper detected in the vicinity of the paper loading tray. The acquisition step to be performed, the evaluation step to evaluate the paper ejection alignment state, the speed value corresponding to the paper ejection speed of the paper ejection section and the paper position are input, and the speed value is output as the output of the paper ejection matching state. It has a generation step of generating a learning model by machine learning using the evaluation result as a reward, and a control step of controlling the paper ejection speed of the paper ejection unit based on the learning model.

Further, the transport device of the present invention is a transport device provided with a paper ejection unit for discharging paper to the paper loading tray, and is detected by a sensor that detects the position of the paper in the vicinity of the paper loading tray and a sensor. The paper ejection alignment state is evaluated based on the position of the paper, the speed value corresponding to the paper ejection speed of the paper ejection section and the paper position are input, the speed value is output, and the paper ejection alignment state. It is provided with a learning unit that generates a learning model by machine learning using the evaluation result of the above as a reward, and a control unit that controls the paper ejection speed of the paper ejection unit based on the learning model generated by the learning unit.

Further, in the image forming apparatus of the present invention, a transport section for transporting paper, an image forming section for forming an image on the paper transported by the transport section, and a paper image-formed by the image forming section are placed on a paper mounting tray. An image forming apparatus equipped with a paper ejection section for ejecting paper, which evaluates the paper ejection alignment state based on the detected position of the paper in the vicinity of the paper loading tray, and discharges the paper in the paper ejection section. The learning unit that generates a learning model by machine learning with the speed value and paper position corresponding to the input, the speed value as the output, and the evaluation result of the paper ejection matching state as a reward, and the learning model generated by the learning unit. A control unit for controlling the paper ejection speed of the paper ejection unit is provided.

According to the present invention, the generated learning model can be used to have consistency at the time of paper ejection to the paper loading tray, and appropriate paper ejection corresponding to a wide variety of paper types can be performed. ..

It is a figure which shows the internal structure of the image forming apparatus by one Embodiment of this invention. It is a block diagram in the case where a plurality of post-processing devices are connected to the image forming apparatus according to the example of one Embodiment of this invention. It is a figure which shows the example of the paper ejection state of the paper to the paper placing tray. It is a block diagram which shows the control structure of the image forming apparatus by one Embodiment of this invention. It is a block diagram which shows the structure of the learning block and the state control block of the image forming apparatus by one Embodiment of this invention. It is a figure which shows the example of the paper transport path of the image forming apparatus by one Embodiment of this invention. It is a figure which shows the example of the relationship between the paper position and the roller drive speed by one Embodiment of this invention. It is a figure is an example of the rigidity and the size of the paper according to the example of one Embodiment of this invention. It is a figure which shows the example of the image printing rate by one Embodiment of this invention. It is a figure which shows the example of the environment by one Embodiment of this invention. It is a figure which shows the relationship between the paper ejection state and the reward by one Embodiment of this invention. It is a flowchart which shows the example of the process flow at the time of learning by the example of one Embodiment of this invention. It is a flowchart which shows the example of the reward calculation process by one Embodiment of this invention. It is a figure which shows the example of transporting a plurality of sheets by one Embodiment of this invention. It is a figure which shows the example of the relationship between the paper transport speed and the paper transport interval according to the example of one Embodiment of this invention. It is a figure which shows the example of the mode setting screen by one Embodiment of this invention. It is a block diagram which shows the control structure (the example which learns with an opportunity learning CPU) of the image forming apparatus by the modification of the embodiment of this invention. It is a block diagram which shows the structure (the example which the server on the cloud learns) by the modification of one Embodiment of this invention. It is a block diagram which shows the structure by the modification of one Embodiment of this invention (the example which the server on the cloud has the observation information generation update part). It is a block diagram which shows the structure by the modification of one Embodiment of this invention (the example which the server on the cloud has an observation information generation part). It is a block diagram which shows the structure by the modification of the example of one Embodiment of this invention (the example which learns with an opportunity learning CPU and the server on a cloud has an observation information generation part). It is a block diagram which shows the structure (the example which the image forming apparatus includes the observation information generation update part) by the modification of one Embodiment of this invention. It is a block diagram which shows the structure by the modification of the example of one Embodiment of this invention (the example which learns by the opportunity learning CPU in an image forming apparatus, and is provided with the observation information generation update part).

Hereinafter, an example of an embodiment of the present invention (hereinafter referred to as “this example”) will be described with reference to the accompanying drawings.

[Internal configuration of image forming apparatus]
FIG. 1 shows the internal configuration of the image forming apparatus 1 of this example.
The image forming apparatus 1 includes an image forming apparatus main body 2, a large-capacity paper feeding device 5, and a post-processing apparatus 6 connected to the subsequent stage of the image forming apparatus main body 2. The image forming apparatus 1 has a copy function of optically reading a document and forming a duplicate image on the paper, and receives print data from an external device such as a personal computer, forms a corresponding image on the paper, and outputs the print data. It is a device called a Multifunction Peripheral (MFP) that has a print function and the like.

The image forming apparatus main body 2 has a scanner unit 11 that reads the paper of the original image, an automatic document feeder 12 that feeds the originals placed on the original tray 12a one by one to the original reading position of the scanner unit 11, and a paper. It includes a printer unit 13 that forms and outputs an image, and an operation display unit 14.

The scanner unit 11 is a reading unit for inputting an image on paper, and has a function as an image input unit. That is, the scanner unit 11 includes a light source that irradiates the document with light, a line image sensor that reads the document for one line in the width direction thereof, and a lens or mirror that guides the reflected light from the document to the line image sensor 11a to form an image. It is composed of an optical path consisting of.

The scanner unit 11 performs a so-called scan-reading type reading process in which a document is conveyed by the automatic document feeder 12 and is scanned while being moved relative to the scanning position. Further, the scanner unit 11 can also read the original while it is placed on the platen glass.

The printer unit 13 is composed of a plurality of paper feed trays 15 for storing image-forming paper, an image forming unit 16 for forming an image on the paper, and a paper transport unit 17. The paper transport unit 17 feeds out paper from the large-capacity paper feed device 5 or the paper feed tray 15, passes it through the image forming unit 16, and discharges it to the post-processing device 6 in the subsequent stage.

The image forming unit 16 includes a photoconductor 21, a charging device 22, a laser unit 23, a developing device 24, a transfer device 25, a separating device 26, a cleaning device 27, a fixing device 28, and the like, and an image is printed on paper by an electrophotographic process. Form to.

The photoconductor 21 has a cylindrical shape and is rotated in a fixed direction (direction of arrow A in the figure) by a driving unit (not shown). During rotation, the photoconductor 21 is uniformly charged by the corona discharge by the charging device 22, and then receives scanning of laser light that is turned on / off according to the image data from the laser unit 23, and an electrostatic latent image is formed on the surface thereof. Is formed. The developing device 24 visualizes an electrostatic latent image formed on the surface of the photoconductor 21 as a toner image.

The paper transport unit 17 feeds the paper from the large-capacity paper feed device 5 or the paper feed tray 15, and feeds the paper between the photoconductor 21 and the transfer device 25 at an appropriate timing (arrow B). The transfer device 25 electrostatically transfers the toner image formed on the surface of the photoconductor 21 onto the paper transported by the paper transport unit 17. The separating device 26 separates the paper from the photoconductor 21. The cleaning device 27 removes the toner remaining on the photoconductor 21 after transfer by rubbing it with a blade or the like. The paper on which the toner image is transferred is further conveyed (arrow C), is pressurized and heated when passing through the fixing device 28, the toner image is fixed on the paper, and then discharged to the post-processing device 6 (arrow). D).

The post-processing device 6 has a function of bundling a plurality of sheets and binding them with staples, a function of punching holes, and the like. The post-processing device 6 includes paper trays 6a and 6b for discharging paper. In the case of this example, when the paper is ejected to the paper trays 6a and 6b, the papers are neatly overlapped with each other so as to have consistency.
It should be noted that the provision of two paper mounting trays 6a and 6b is an example, and only one of the trays may be provided.
Further, the configuration of each part of the image forming apparatus 1 described so far is an example, and an image forming may be performed on paper with other configurations. For example, the image forming unit 16 may form an image on paper by a method other than the electrophotographic method.

[Another configuration example of the image forming apparatus]
The image forming apparatus 1 may be configured by connecting a plurality of post-processing apparatus.
For example, as shown in FIG. 2, the first post-processing device 40, the second post-processing device 50, the third post-processing device 60, and the fourth post-processing device 70 are connected in order to the subsequent stage of the image forming apparatus 1. The image forming apparatus 1 may be configured to sequentially convey the paper on which the image is formed.
In the case of the configuration of FIG. 2, the second post-processing device 50 includes the paper loading tray 51, the third post-processing device 60 includes the paper loading tray 61, and the fourth after-processing device 70 includes the paper loading tray. It is equipped with 71 and 72.
When the paper is ejected from the paper trays 51, 61, 71, 72, the paper may be neatly overlapped to have consistency.
In the following description, the process of ejecting the paper to the paper loading tray 6a shown in FIG. 1 will be described, but the same processing will be performed when the paper is ejected from the paper loading trays 51, 61, 71, 72. May be done.

[Example of the state when paper is ejected to the tray]
FIG. 3 shows an example of a state in which papers P1 and P2 are ejected from the paper loading tray 6a.
The paper loading tray 6a shown in FIG. 3 includes a roller 6c for discharging paper. The papers P1 and P2 whose images have been formed by being driven by the paper ejection roller 6c are ejected from the paper ejection portion 6d to the outside of the device in order, and are on the paper loading tray 6a connected to the paper ejection portion 6d. Can be placed on.

The examples of FIGS. 3A and 3B show a case where the second sheet of paper P2 output from the output point 6d is not in an appropriate output state and the sheets do not overlap neatly.
In the example of FIG. 3A, the first sheet of paper P1 is correctly placed on the paper loading tray 6a. On the other hand, the next conveyed paper P2 is conveyed farther than the proper position due to the air resistance when the paper is ejected from the paper ejection portion 6d to the outside of the apparatus, and is conveyed to the right side of FIG. 3A. As shown, the two sheets P1 and P2 are misaligned, and the consistency cannot be maintained.

In the example of FIG. 3B, the first sheet of paper P1 is correctly placed on the paper loading tray 6a. On the other hand, the next conveyed paper P2 is in a state where it is not conveyed to an appropriate position due to the frictional resistance when the paper is ejected from the paper ejection portion 6d to the outside of the apparatus and comes into contact with the paper P1. Therefore, as shown on the right side of FIG. 3B, the paper P2 is in a bent state in contact with the paper ejection portion 6d.

In the case of this example, by executing the process described below, it is possible to prevent the state shown in FIG. 3A and the state shown in FIG. 3B from being obtained, and the paper loading tray 6a is clean. The consistency of the paper is maintained by overlapping with the paper.

[Control configuration of image forming apparatus]
FIG. 4 is a block diagram showing a control configuration of the image forming apparatus main body 2.
The image forming apparatus main body 2 includes an MFP controller 103 that collectively controls the operation of the image forming apparatus main body 2, and the control by the MFP controller 103 is an engine control SoC (System-on-a-chip) 102 in the engine control unit 101. It is executed based on the command from.

The machine learning unit 104, the device state grasping unit 105, the paper ejection control unit 106, and the paper transport control unit 107 are connected to the engine control unit 101.
The machine learning unit 104 generates a learning model for maintaining consistency at the time of discharging the paper which is the object to be conveyed, and performs machine learning with the learning model.
The device state grasping unit 105 grasps the device state such as the paper state, the paper position, and the paper ejection speed control amount. The details of this device state will be described later.
The paper ejection control unit 106 controls the operation of ejecting paper to the paper loading tray 6a or the like.
The paper transport control unit 107 controls the transport of paper in the image forming apparatus 1.

FIG. 5 is a functional block diagram showing a configuration as seen from the function of the machine learning unit 104 to perform machine learning.
The machine learning unit 104 includes a learning block 110, a state control block 120, and a learning result output unit 130.

The learning block 110 includes a state observation unit 111, a reward calculation unit 112, a learning unit 113, and a decision-making unit 114.
The state observation unit 111 acquires observation information about the state of paper transfer in the image forming apparatus 1 from the state control block 120. The observation information acquired by the state observation unit 111 is supplied to the learning unit 113 as a state variable. The state variables here include the paper position information, the paper ejection speed value, the physical characteristics of the paper, the image forming conditions, the usage state of the apparatus, and the like.
Further, the information about the paper discharge matching state in the observation information is supplied from the state observation unit 111 to the reward calculation unit 112 as a state variable.

The reward calculation unit 112 calculates the reward value based on the paper ejection matching state. A specific example of calculating the reward value will be described later.
The learning unit 113 executes the calculation of the prepared learning model based on the state variables supplied from the state observation unit 111, and acquires the state variables and actions indicating the paper ejection speed values.
Further, the learning unit 113 updates the learning model based on the reward value calculated by the reward calculation unit 112.

The decision-making unit 114 determines the paper ejection speed value based on the calculation result in the learning unit 113, and supplies the determined paper ejection speed value to the state control block 120.
Further, the learning model updated (generated) by the learning unit 113 is output from the learning result output unit 130.

The state control block 120 includes a state acquisition unit 121 that changes directly by an action, an observation information generation unit 122, and a state acquisition unit 123 that indirectly changes by an action.
The state acquisition unit 121 instructs the paper ejection control unit 106 (FIG. 4) to use the paper ejection speed value supplied from the decision-making unit 114, and observes that the paper ejection speed value is set. Notify the generation unit 122.
The observation information generation unit 122 observes the state of each unit such as the paper ejection speed value and the paper transport state, and supplies the observation result to the state acquisition unit 123.

The state acquisition unit 123 acquires each observation information such as paper position information, paper ejection speed value, paper physical properties, image formation conditions, device usage status, and paper ejection matching state, and obtains the acquired observation information as a learning block. It is supplied to the state observation unit 111 of 110.

[Example of input during learning]
FIG. 6 shows an outline of a paper transport path from the fixing roller 31 included in the fixing device 28 (FIG. 1) in the image forming device 1 to the paper loading tray 6a.
As shown in FIG. 6, the paper passes through the place PS1 where the fixing roller 31 is installed, the place PS2 immediately before the roller 32 in the middle of the transport path, and the place PS3 immediately before the roller 6c for paper ejection. A sensor is installed to detect.
Here, in the case of this example, the drive speed of the roller 32 and the roller 6c on the downstream side of the transport path from the fixing roller 31 is controlled to set the speed at which the paper is discharged to the paper loading tray 6a. This ensures that the paper is properly ejected.

The table shown in FIG. 7 shows the state (paper position) determined from the paper detection state by the sensors of PS1, PS2, and PS3 at each location, the action showing the relationship between the roller speed and the switching timing, and the roller speed. Shows the action value corresponding to the combination of speed switching.
The table shown in FIG. 7 is a Q-table for performing so-called Q-learning, and is used when the learning unit 113 learns input parameters and output parameters.
The roller 32 and the roller 6c on the downstream side of the fixing roller 31 are set to the same speed so as not to damage the passing paper.

Here, as shown in the lower part of FIG. 7, the roller speed has 11 steps from "-5 (slow side)" to "+5 (fast side)" with the reference speed as "0". "-5" is the slowest speed compared to the reference speed, and "+5" is the fastest speed compared to the reference speed.
Further, the switching timing is set to three stages of immediate switching after the state transition, early switching, and late switching.
Therefore, regarding the behavior, there are 11 stages of roller speed × 3 stages of switching timing = 33 types (actions of a1 to a33 in FIG. 7).

The score shown in the table of FIG. 7 is given according to the detection state of the paper by the sensors of each location PS1, PS2, PS3 and any of the 33 types of actions. For example, in the state S1 and the action a1, a score of 0.38 is given.
Regarding the speed switching timing, the time for switching early and the time for switching late are preset times. Further, although the speed switching timing is set to three stages here, the speed switching timing may be changed more finely.

Further, the learning unit 113 also acquires the rigidity as the physical characteristics of the paper and the size of the paper shown in FIG. 8 as inputs at the time of learning. As shown in FIG. 8, the input value of the paper rigidity here is set in 16 steps. Further, as for the size of the paper, the input value is set in 16 steps as shown in FIG.
Regarding the rigidity of the paper, for example, an image forming apparatus 1 is provided with a sensor for detecting the rigidity, and an input value is set based on the output of the sensor. Alternatively, the rigidity may be set by user operation. The size of the paper is set as an input value by detecting the size of the paper loaded in the image forming apparatus 1 with a sensor.

The physical characteristics of the paper include the weight of the paper per unit area, the smoothness of the surface of the paper, the water content of the paper, the aspect ratio (ratio of the length and width of the paper), and the flow of the paper (so-called T-th and Y-th). One of the eyes) may be obtained as an input.

Further, the learning unit 113 also acquires an image forming condition such as an image printing rate on the paper shown in FIG. 9 as an input at the time of learning. As shown in FIG. 9, the image printing rate on the paper here is set to 9 levels based on the image printing rate of the immediately preceding paper (front paper) and the image printing rate of the current paper (own paper). Will be done. This image print rate is acquired from, for example, the MFP controller 103 that controls the image forming unit 16.
In addition to the image printing rate, at least one of the paper passing mode (single-sided printing / double-sided printing), fixing temperature, paper curl correction amount, and charge control amount is acquired as an image forming condition, and the learning unit. It may be used at the time of learning in 113. In the case of these paper passing modes (single-sided printing / double-sided printing), fixing temperature, paper curl correction amount, and charge control amount, as in the image printing rate shown in FIG. 9, a plurality of steps are taken based on each state. Set the input value of.

Further, the learning unit 113 also acquires the environment as the usage state of the image forming apparatus 1 shown in FIG. 10 as an input at the time of learning. The environment here is an input value of 9 stages of a combination of temperature (3 stages of high, medium, and low) and humidity (3 stages of high, medium, and low). The input value of this environment is set from, for example, the measured value by the temperature sensor and the humidity sensor installed in the image forming apparatus 1.
Further, as the usage state of the image forming apparatus 1, at least one of the amount of wear of the drive roller and the number of sheets loaded on the paper loading tray is acquired in addition to the temperature and humidity, and the learning unit 113 acquires at least one of them at the time of learning. You may use it. Similarly to the temperature and humidity shown in FIG. 10, in the case of the amount of wear of these drive rollers and the number of sheets loaded on the paper tray, input values in a plurality of stages are set based on the respective states.
It should be noted that each of the physical characteristics of the paper, the image formation conditions, and the usage state described so far may be input, but any of them may be selected as inputs based on the device configuration and the like.

[Example of reward setting]
FIG. 11 shows an example of a case where a positive reward is given and a case where a negative reward is given when the reward value is calculated by the reward calculation unit 112.
FIG. 11 shows three examples in which paper is ejected to the paper loading tray 6a. In the case of this example, the image pickup unit 33 is installed close to the paper mounting tray 6a, and the observation information generation unit 122 (FIG. 5) maintains consistency as a paper ejection state from the image captured by the image pickup unit 33. Determine if it is.

The state shown in FIG. 11A is an example in which consistency is maintained. In the case of FIG. 11A, the three sheets of paper P1, P2, and P3 discharged from the paper loading tray 6a via the roller 6c are all overlapped at the same position. When the observation information generation unit 122 detects such a state from the image of the image pickup unit 33, a positive reward is given.

The state shown in FIG. 11B is an example (No. 1) in which consistency is not maintained. In the case of FIG. 11B, the paper P3 among the three sheets P1, P2, P3 discharged to the paper loading tray 6a via the roller 6c is larger than the other papers P1 and P2. The paper has been ejected to a distant position. When the observation information generation unit 122 detects such a state from the image of the image pickup unit 33, a negative reward is given.

The state shown in FIG. 11 (c) is an example (No. 2) in which consistency is not maintained. In the case of FIG. 11C, the paper P3 among the three sheets P1, P2, P3 discharged to the paper loading tray 6a via the roller 6c is larger than the other papers P1 and P2. The paper is ejected in the front position. Even when the observation information generation unit 122 detects such a state from the image of the image pickup unit 33, a negative reward is given.

[Learning process in the machine learning department]
FIG. 12 is a flowchart showing the flow of learning processing in the machine learning unit 104.
First, the decision-making unit 114 gives the paper ejection speed value to the paper ejection control unit 106 (step S11). The paper ejection control unit 106 that has acquired this paper ejection speed value operates the rollers 32, 6c (FIG. 6) at the corresponding speed (step S12).

Next, the state observation unit 111 acquires the observation information generated by the state control block 120 (step S13). The observation information acquired here includes the paper ejection speed value of the roller, the paper position information, the physical characteristics of the paper, the image formation conditions, the usage state of the apparatus, the paper ejection matching conditions, and the like.
Next, the reward calculation unit 112 acquires the observation information and then calculates the reward value (step S14).

Then, the learning unit 113 inputs the observation information into the learning model and performs the learning process based on the observation information (behavior) (step S15). At this time, the learning unit 113 updates the learning model based on the reward value obtained by the reward calculation unit 112.
Further, the learning unit 113 determines whether or not the predetermined number of learnings has been completed (step S16). When it is determined in step S16 that the learning has not been performed a predetermined number of times (NO in step S16), the process returns to the paper ejection speed value supply process in step S11.

Further, when it is determined in step S16 that the learning has been performed a predetermined number of times and the positive reward has been continuously obtained a predetermined number of times (YES in step S16), the learning result output unit 130 is updated in step S15. The generated learning model is output as a newly generated learning model (step S17). The predetermined number of times to be determined in step S16 is, for example, 100 times.

[Reward calculation processing in the reward calculation department]
FIG. 13 is a flowchart showing the flow of the calculation process of the reward value in the reward calculation unit 112.
First, the reward calculation unit 112 acquires the observed value of the matching state (any state shown in FIG. 11) of the current paper from the state observation unit 111 (step S21).
Then, the reward calculation unit 112 determines whether or not the acquired observed value of the matching state is within a predetermined range (step S22). Here, for example, when the edge of the discharged paper is within 2 mm from the edge of the paper loading tray 6a, it is defined as within a predetermined range.

When it is determined in step S22 that the range is not within the predetermined range (NO in step S22: in the state shown in FIGS. 11B or 11C), the reward calculation unit 112 sets the reward value of the output this time to "-". 1 ”(step S23).
Further, when it is determined in step S22 that the range is within the predetermined range (YES in step S22: in the state shown in FIG. 11A), the reward calculation unit 112 sets the reward value of the output this time to "+1". (Step S24).
When the reward value is determined in steps S23 or S24, the reward calculation unit 112 ends the calculation process of the reward value by the current paper ejection.

As described above, by controlling the paper ejection speed based on the output value of the learning model obtained by the learning process using the learning model in the machine learning unit 104, what kind of paper is used? However, the paper mounting tray 6a is neatly stacked in the same state, and the consistency is maintained.

[Example of controlling the paper transfer interval]
In the description so far, the paper ejection speed to the paper loading tray 6a is adjusted to adjust the paper ejection state. Here, when the paper transfer interval when continuously conveyed is short, it is necessary to adjust the paper transfer interval in addition to adjusting the paper ejection speed.
That is, for example, as shown in FIG. 14, one sheet of paper P11 passes through the roller 32 and the roller 6c with the fixing roller 31, the roller 32, and the roller 6c arranged in the paper transport path, and the next paper. It is assumed that P12 is close to the fixing roller 31.
In such a state, if the paper ejection speed of the paper P11 is slowed down by the paper ejection roller 6c, the transfer of the next paper P12 may be hindered. In such a case, the paper transport control unit 107 (FIG. 4) temporarily widens the transport interval of each paper in the transport path.

For example, as shown in FIG. 15, when the two sheets of paper P11 and P12 are continuously conveyed, the paper ejection control unit 106 performs matching at the time of paper ejection in a state where the basic paper spacing is set to 100 ms. It is assumed that the speed of the paper ejection roller 6c is reduced in order to ensure the performance. Here, the basic paper spacing of 100 ms is the time difference between the two sheets P11 and P12 passing through the same place.
At this time, the paper transport control unit 107 temporarily stops the transport of the paper P12, and as shown in FIG. 15, performs a process of widening the paper spacing to 120 ms or more.

As described above, according to the image forming apparatus 1 of this example, when the process of ensuring consistency at the time of paper ejection is executed, there is a possibility that the paper transport interval is widened.
However, if the paper transport interval is widened, the number of sheets per unit time in which the image forming apparatus 1 forms an image and is discharged to the paper placing tray 6a or the like decreases, and the productivity of the image forming apparatus decreases.

Therefore, it is preferable that the image forming apparatus 1 allows the user to select either a mode that prioritizes ensuring consistency at the time of paper ejection or a mode that prioritizes productivity. When a mode for giving priority to ensuring consistency at the time of paper ejection is set, the image forming apparatus 1 temporarily widens the paper spacing as necessary, and then secures the consistency at the time of paper ejection. On the other hand, when the mode that prioritizes productivity is set, the image forming apparatus 1 sets the paper spacing even when it is necessary to widen the paper spacing in order to ensure consistency at the time of paper ejection. Limit the process of ensuring consistency when ejecting paper without spreading it.

FIG. 16 shows an example of a mode selection screen displayed by the operation display unit 14.
On the selection screen shown in FIG. 16, a button for selecting on / off of the “paper ejection consistency priority mode” is displayed. On this selection screen, a warning message "In the case of the output consistency priority mode, productivity may decrease" is displayed.
When the on of the "paper ejection consistency priority mode" is selected by the user operation on this selection screen, the paper transport control unit 107 executes the process of widening the paper spacing described with reference to FIG. 15 to 120 ms or more.

[Example of another configuration of the image forming apparatus]
The present invention is not limited to the above-described embodiment, and can be modified or applied without departing from the gist of the present invention.

For example, in the configuration of the image forming apparatus 1 shown in FIG. 4, the engine control SoC102 is arranged in the engine control unit 101 so that the machine learning unit 104 or the like operates under the control of the engine control SoC102. .. On the other hand, as shown in FIG. 17, the engine control CPU 102'and the machine learning CPU 108 communicate with each other so that the engine control unit 101 includes the engine control CPU 102'and the machine learning unit 104' includes the machine learning CPU 108. It may be done to acquire the paper alignment state by executing the learning model.

Further, in the configuration of the image forming apparatus 1 shown in FIG. 4, all the steps up to the generation and updating of the learning model are executed in the image forming apparatus 1.
On the other hand, some or all the processing related to the generation and update of the learning model is performed on the server side prepared on the cloud, and the image forming apparatus 1 performs the learning model generated (updated) on the server side. The image matching state may be set as the output of the image formation.

That is, as shown in FIG. 18, the server 200 on the cloud capable of communicating with the image forming apparatus 1 via the network NW includes a learning block 110, a state control block 120, and a learning result output unit 130. , The same processing as the machine learning unit 104 of the example of FIG. 4 may be performed.

Further, as shown in FIG. 19, when the server 200 is provided, the state control block 120 of the server 200 may be configured to include the observation information generation update unit 124. In this case, the observation information generation / updating unit 124 in the server 200 acquires the observation information from the image forming apparatus 1, and sends the acquired observation information to the observation information generating unit 122 of the image forming apparatus 1.

Alternatively, as shown in FIG. 20, the image forming apparatus 1 may include an observation information generation / updating unit 124. Then, the observation information generation / updating unit 124 generates and updates the observation information from the engine control unit 101, and sends the obtained observation information to the observation information generation unit 122 in the server 200 via the network NW. May be good.

Furthermore, as shown in FIG. 21, the machine learning unit 104'provides an observation information generation update unit 124 as a configuration in which the engine control unit 101 includes an engine control CPU 102'and the machine learning unit 104' includes a machine learning CPU 108. Observation information may be generated and updated. Also in the case of the example of FIG. 2, the observation information generation update unit 124 sends the obtained observation information to the observation information generation unit 122 in the server 200.

Further, when the learning model is generated or updated in the image forming apparatus 1, the image forming apparatus 1 may include the observation information generation and updating unit 124.
That is, as shown in FIG. 22, the configuration includes the observation information generation update unit 124 connected to the engine control unit 101 in the image forming apparatus 1, and the observation information generation update unit 124 uses the information from the engine control unit 101 as information. The observation information may be generated and updated based on the above.
Further, as shown in FIG. 23, when the engine control unit 101 includes the engine control CPU 102'and the machine learning unit 104' includes the machine learning CPU 108, the observation information connected to the machine learning unit 104' It may be configured to include the generation update unit 124. In the case of the configuration of FIG. 23, the observation information generation / updating unit 124 generates and updates the observation information via the machine learning unit 104'.

[Other variants]
In the above-described embodiment, the process of maintaining the consistency of the paper when the image is formed by the image forming apparatus 1 is performed by learning by applying the learning model. On the other hand, the state of the object (paper) is observed in a device for transporting an object such as paper separate from the image forming apparatus 1, and the result of learning based on the observed value is used for transport. Control may be performed to maintain the consistency of the object inside when the paper is ejected. For example, in the post-processing device 6 shown in FIG. 1, paper ejection control may be performed by a learning model.

Further, in the above-described embodiment, the generated (updated) learning model is used in the image forming apparatus 1 that generated the learning model, but the generated learning model is used in another image forming apparatus. It may be set and used for paper ejection control in another image forming apparatus.

Further, the physical characteristics of the paper described in the above-described embodiment of the embodiment show an example of a suitable combination, and at least one or a plurality of the physical characteristics of the above-mentioned paper may be combined.

1 ... Image forming apparatus, 2 ... Image forming apparatus main body, 3 ... External terminal, 4 ... PCI bus, 5 ... Large capacity paper feed device, 6 ... Post-processing device, 6a, 6b ... Paper loading tray, 6c ... Roller, 6d ... Paper ejection location, 11 ... Scanner unit, 11a ... Line image sensor, 12 ... Automatic document feeder, 13 ... Printer unit, 14 ... Operation display unit, 15 ... Paper feed tray, 16 ... Image forming unit, 17 ... Paper Transport unit, 21 ... Photoreceptor, 22 ... Charging device, 23 ... Laser unit, 23a ... Laser diode, 24 ... Developing device, 25 ... Transfer device, 26 ... Separation device, 27 ... Cleaning device, 28 ... Fixing device, 33 ... Imaging unit, 40 ... 1st post-processing device, 50 ... 2nd post-processing device, 60 ... 3rd post-processing device, 70 ... 4th post-processing device, 101 ... engine control unit, 102 ... engine control SoC, 102'... Engine control CPU, 103 ... MFP controller, 104, 104'... Machine learning unit, 105 ... Equipment status grasping unit, 106 ... Paper ejection control unit, 107 ... Paper transfer control unit, 108 ... Machine learning CPU, 110 ... Learning block, 111 ... state observation unit, 112 ... reward calculation unit, 113 ... learning unit, 114 ... decision unit, 120 ... state control block, 121 ... state acquisition unit that changes directly by action, 122 ... observation information generation unit, 123 … State acquisition unit that changes indirectly by action, 130… Learning result output unit

Claims (18)

This is a learning method for a transport device having a paper ejection unit that ejects paper in a paper loading tray.
An acquisition step for acquiring the paper ejection alignment state of the paper detected in the vicinity of the paper loading tray, and
An evaluation step for evaluating the paper ejection consistency state and
A learning model is generated by machine learning by inputting a speed value corresponding to the paper ejection speed of the paper in the paper ejection unit and the position of the paper, outputting the speed value, and using the evaluation result of the paper ejection matching state as a reward. And the generation steps to do
A learning method including a control step for controlling the paper ejection speed of the paper ejection unit based on the learning model. The learning method according to claim 1, wherein the acquisition step acquires a paper ejection alignment state with respect to the paper transport direction on the paper loading tray. The learning method according to claim 1, wherein the input includes the physical properties of the paper. The learning method according to claim 3, wherein the physical characteristics of the paper include at least one of the rigidity, the size, and the weight per unit area of the paper. The learning method according to claim 4, further comprising any one of the smoothness, water content, aspect ratio, and flow of the paper as the physical characteristics of the paper. The learning method according to claim 1, wherein the paper ejection speed is the rotation speed of a drive source that drives the paper ejection unit. The learning method according to claim 1, wherein learning is performed on a plurality of sheets to be ejected in the same job to generate the learning model. When the paper ejection alignment state at the time of the current paper ejection is within the predetermined range, a positive reward value is given to the current output, and a positive reward value is given.
The learning method according to claim 1, wherein a negative reward value is given to the output this time when the paper ejection matching state at the time of this paper ejection is out of a predetermined range. The learning method according to claim 8, wherein learning is terminated when a positive reward value is continuously obtained a predetermined number of times as the reward value. It is a transport device equipped with a paper ejection unit that ejects paper on the paper loading tray.
A sensor that detects the position of the paper in the vicinity of the paper tray, and
The paper ejection alignment state of the paper is evaluated based on the position of the paper detected by the sensor, and the speed value corresponding to the paper ejection speed of the paper ejection portion and the position of the paper are input as described above. A learning unit that generates a learning model by machine learning using the speed value as an output and the evaluation result of the paper ejection matching state as a reward.
A transport device including a control unit that controls the paper ejection speed of the paper ejection unit based on a learning model generated by the learning unit. A transport unit that transports paper and
An image forming unit that forms an image on the paper conveyed by the conveying unit, and an image forming unit.
An image forming apparatus provided with a paper ejection portion for ejecting the paper image-formed by the image forming portion to a paper loading tray.
The paper ejection alignment state of the paper is evaluated based on the position of the paper detected in the vicinity of the paper loading tray, and the speed value corresponding to the paper ejection speed of the paper ejection portion and the position of the paper are determined. A learning unit that generates a learning model by machine learning with the input, the speed value as the output, and the evaluation result of the paper ejection matching state as a reward.
An image forming apparatus including a control unit that controls the paper ejection speed of the paper ejection unit based on a learning model generated by the learning unit. The image forming apparatus according to claim 11, wherein the control unit controls a drive source for driving a roller downstream of the fixing roller included in the image forming unit to control the paper ejection speed. The image forming apparatus according to claim 11, wherein the learning unit performs learning by changing the paper ejection speed of the paper ejection unit while keeping the paper transport speed by the fixing roller provided in the image forming unit constant. The image forming apparatus according to claim 11, wherein the input of the learning unit includes an image forming condition on the paper in the image forming unit. The image forming apparatus according to claim 11, wherein the input of the learning unit includes a usage state of the image forming apparatus. The image forming apparatus according to claim 11, further comprising an imaging unit that detects the position of the paper in the vicinity of the paper loading tray. The image forming apparatus according to claim 12, wherein when the execution of controlling the paper ejection speed by controlling the downstream roller interferes with the next paper transport, the paper transport interval in the transport section is widened. The user can select and set a first mode that prioritizes paper ejection consistency in the paper loading tray and a second mode that prioritizes the paper transport interval in the transport unit.
The image forming apparatus according to claim 11, wherein the control unit performs control based on a set mode.

Images