JP2023005298A - Machine learning device, data processing system, machine learning method, and data processing method - Google Patents

Abstract

To make it possible to obtain high-quality printing results with few printing failures.SOLUTION: A machine learning device includes: a state variable acquisition unit for acquiring feature quantity information and medium information, as state variables, among the feature quantity information in an actual printed matter actually printed by an image forming device, first control information that is control information when the actual printed matter is printed, and the medium information of a printing medium used for the actual printed matter; a teacher data acquisition unit for acquiring, as teacher data, difference control information that indicates differences between second control information that sets the feature quantity information within a predetermined threshold and the first control information when the actual printed matter is printed; and a learned model generation unit for generating a learned model by performing machine learning based on the state variables acquired by the state variable acquisition unit and the teacher data acquired by the teacher data acquisition unit.SELECTED DRAWING: Figure 2

Description

The present invention relates to a machine learning device, data processing system, machine learning method, and data processing method.

Conventionally, there is a machine learning device that learns the correlation between control information of an image forming apparatus and printed matter output by the image forming device, and a data processing system that uses a trained model obtained by the machine learning device. (See Patent Document 1, for example).

A machine learning device of this type obtains feature amount information (that is, information about printing defects occurring in the actual printed matter) in an actual printed matter actually printed on a printing medium by an image forming apparatus, and the printing medium used in the actual printed matter. and the first control information, which is a control parameter when printing the actual printed material, as state variables; and the second control information ( That is, a teacher data acquisition unit that acquires control parameters for obtaining high-quality printing results with few printing defects as teacher data, the state variables acquired by the state variable acquisition unit, and the teacher data acquired by the teacher data acquisition unit. and a trained model generation unit that generates a trained model by performing machine learning based on the data.

In addition, this type of data processing system utilizes a learned model generated by a machine learning device to set a control parameter that prevents printing defects from occurring in actual printed matter when printing on an unlearned print medium. 3 to output control information.

By printing with the image forming apparatus using the third control information output from the data processing system, it is possible to obtain high-quality printing results with few printing defects.

JP 2006-254386 A

However, with the conventional technology, due to deterioration over time of the image forming apparatus, even if printing is performed using the third control information output from the data processing system, it may not be possible to obtain a high-quality print result with few print defects. had a problem.

SUMMARY OF THE INVENTION The present invention takes the above into consideration, and proposes a machine learning device, a data processing system, a machine learning method, and a data processing method capable of obtaining high-quality printing results with few printing defects. It is.

A machine learning apparatus according to the present invention includes feature amount information in an actual printed material actually printed by an image forming apparatus, first control information that is control information when the actual printed material is printed, and A state variable acquisition unit that acquires the feature amount information and the medium information among the medium information of the print medium as state variables, and second control information that makes the feature amount information within a predetermined threshold value and the actual printed matter are printed. a teacher data acquisition unit that acquires, as teacher data, differential control information that is a difference from the first control information when the and a trained model generation unit that generates a trained model by performing machine learning based on the data.

A data processing system according to the present invention includes an actual printed matter information acquisition unit that acquires feature amount information of an actual printed matter actually printed by an image forming apparatus and medium information of a printing medium used in the actual printed matter; The feature amount information and the medium information acquired by the printed matter information acquisition unit are input to the learned model generated by the above-described machine learning device, and the actual printed matter is added to the differential control information output by the learned model. A data processing unit that adds the first control information at the time of printing to generate third control information, and a third control information storage unit that stores the third control information.

A machine learning method of the present invention is a machine learning method using a computer, and includes a first a step of acquiring the feature amount information and the medium information out of the control information and the medium information of the print medium used for the actual printed material as state variables; a step of obtaining, as teacher data, difference control information that is a difference between the control information and the first control information when the actual printed material is printed; and performing machine learning based on the obtained state variables and the teacher data. and generating a trained model by doing so.

The data processing method of the present invention includes the step of obtaining, using a computer, feature amount information of an actual printed matter actually printed by an image forming apparatus and medium information of a printing medium used for the actual printed matter; The feature amount information and the medium information acquired by the actual printed matter information acquisition unit are input to the learned model generated by the above-described machine learning method, and the actual printed matter information is added to the difference control information output by the learned model. generating third control information by adding the first control information at the time of printing; and storing the third control information.

In this way, by not including the first control information in the state variables used in the trained model, difference control information indicating how much the control information of the image forming apparatus from the trained model deviates from the optimum control information. is obtained, and the third control information is obtained by adding the difference control information to the first control information when the actual printed matter is printed by the image forming apparatus. By doing so, it is possible to increase the degree of contribution of the feature amount information of the actual printed matter in the trained model, and even if the control information changes to obtain high-quality printing results with few printing defects due to deterioration over time of the image forming apparatus. , the optimum third control information can be obtained.

INDUSTRIAL APPLICABILITY The present invention can realize a machine learning device, a data processing system, a machine learning method, and a data processing method that enable obtaining high-quality printing results with few printing defects.

1 is a diagram showing a mechanical structure of an image forming apparatus according to a first embodiment; FIG. 1 is a block diagram showing the configuration of a machine learning device according to a first embodiment; FIG. 1 is a diagram showing an example of a neural network model for supervised learning according to the first embodiment; FIG. 1 is a block diagram showing the configuration of a data processing system incorporated in an image forming apparatus according to a first embodiment; FIG. 4 is a flow chart showing a machine learning method according to the first embodiment; 4 is a flow chart showing a data processing method according to the first embodiment; 7 is a graph showing results of image quality evaluation according to the first embodiment; FIG. 10 is a block diagram showing the configuration of a data processing system incorporated in a server device according to a second embodiment; FIG.

EMBODIMENT OF THE INVENTION Hereinafter, the form (this is called embodiment hereafter) for implementing invention is demonstrated in detail using drawing. In the following, the parts required to achieve the object of the present invention will be mainly described, and the description of other known techniques will be omitted as appropriate.

First, the basic configuration of an electrophotographic image forming apparatus that performs print output as a learning target of the machine learning apparatus and machine learning method according to the present embodiment will be described.

[1. First Embodiment]
[1-1. Configuration of Image Forming Apparatus]
FIG. 1 is a schematic structural diagram of an image forming apparatus used in the first embodiment of the invention. The image forming apparatus 10 exemplified here is a so-called intermediate transfer type full-color LED printer, and as shown in FIG. , an output unit 60 and a control unit 70 .

The print medium supply unit 20 is for supplying the print medium PM into the image forming apparatus 10 , and has a paper tray 21 , a manual feed tray 22 and a plurality of paper feed rollers 23 . The paper tray 21 accommodates a plurality of stacked print media PM. The print medium PM accommodated in the paper tray 21 is generally general paper (A4 plain paper, B4 plain paper, etc.). The manual feed tray 22 is storably provided on the side surface of the main body of the image forming apparatus 10, and is a tray for feeding the special print medium PM mainly when printing is performed on a special print medium PM different from general paper. is. Therefore, when performing industry printing, this manual feed tray 22 is mainly used. Incidentally, industry printing refers to printing specialized for specific business in a specific industry in an industrial field such as the medical field, manufacturing industry, or distribution industry. The plurality of paper feed rollers 23 are rollers arranged at appropriate positions to transport the print medium PM placed in the paper tray 21 or on the manual feed tray 22 to the transport path PL.

The image forming section 30 is for forming a toner image, and includes a plurality of (five in this embodiment) image forming units 31C, 31M, 31Y, 31K, and 31S (hereinafter referred to as these) arranged in parallel. collectively referred to as an image forming unit 31). The plurality of image forming units 31C, 31M, 31Y, 31K, and 31S are basically the same in construction except that the toner colors are different. , an LED head, and a toner tank. As for the image forming process by each of these configurations, a process similar to a well-known process can be adopted, and detailed description thereof will be omitted here.

The plurality of image forming units 31C, 31M, 31Y, 31K, and 31S are units for forming toner images of cyan, magenta, yellow, black, and special colors, respectively. By adopting a plurality of these image forming units 31, full-color printing is possible. A toner image formed by each image forming unit 31 is transferred onto an intermediate transfer belt 41 of a transfer section 40, which will be described later. As the special color toner, for example, white toner, clear toner, or fluorescent color (neon yellow, etc.) toner can be used.

The transfer unit 40 is for transferring the toner image formed by the image forming unit 30 onto the print medium PM, and includes an intermediate transfer belt 41 and a plurality of primary transfer rollers 42C, 42M, 42Y, 42K, and 42S (hereinafter referred to as , which are also collectively referred to as a primary transfer roller 42 ), a backup roller 43 , and a secondary transfer roller 44 . The intermediate transfer belt 41 is an endless elastic belt supported by a plurality of rollers including drive rollers and made mainly of a resin material such as rubber. A toner image of each color formed by the image forming unit 31 is transferred (primary transfer) onto the surface of the intermediate transfer belt 41, and the formed toner image is later transferred (secondary transfer) onto the print medium PM. ) is done. A plurality of primary transfer rollers 42C, 42M, 42Y, 42K, and 42S are for transferring the toner images of respective colors formed by the image forming units 31 onto the intermediate transfer belt 41, and the plurality of image forming units 31C and 31M are provided. , 31Y, 31K, and 31S, and the intermediate transfer belt 41 is sandwiched between them. A predetermined primary transfer voltage is applied to the primary transfer roller 42 . The primary transfer voltage is controlled by the controller 70, which will be described later.

The backup roller 43 is one of a plurality of rollers that support the intermediate transfer belt 41, and is arranged at a position facing a secondary transfer roller 44, which will be described later, with the intermediate transfer belt 41 interposed therebetween. The secondary transfer roller 44 is arranged in the middle of the conveying path PL at a position facing the backup roller 43 with the intermediate transfer belt 41 interposed therebetween. functions to transfer a pre-formed toner image to the intermediate transfer belt 41 when the print medium PM is passed between them. A predetermined secondary transfer voltage is applied to the secondary transfer roller 44 . The secondary transfer voltage is controlled by the controller 70, which will be described later.

The fixing unit 50 applies heat and pressure to the printing medium PM to which the toner image has been transferred by the transfer unit 40 to fix the toner image onto the printing medium PM. and a roller 52 . The fixing roller 51 has a built-in heater (not shown), and the fixing temperature is controlled by the current value supplied to this heater. The current value supplied to this heater is controlled by a control section 70, which will be described later. The pressure roller 52 is biased against the fixing roller 51 . As a result, a predetermined fixing pressure is applied to the print medium PM passing between the fixing roller 51 and the pressure roller 52 . In this embodiment, the pressure roller 52 is urged against the fixing roller 51 . may be urged against the backup roller.

The output unit 60 outputs the print medium PM on which the toner image is fixed by the fixing unit 50 to the outside of the image forming apparatus 10 as an actual print, and includes a discharge tray 61 and a plurality of transport rollers 62 . The discharge tray 61 is formed in the upper part of the image forming apparatus 10, and the actual printed matter AP outputted through the transport path PL is placed thereon. A plurality of transport rollers 62 are provided at various locations on the transport path PL to transport the print medium PM to the discharge tray 61 . Cooling means for removing heat generated when the toner image is fixed may optionally be provided at any position of the output section 60. Alternatively, a well-known heat pipe, heat sink, fan, or the like may be arranged at a predetermined position of the output section 60 as a cooling means.

The control section 70 is for controlling each section of the image forming apparatus 10, and includes a well-known CPU, storage means, and the like. As described above, the control section 70 has the function of controlling the secondary transfer voltage applied to the secondary transfer roller 44 and the current supplied to the heater inside the fixing roller 51 .

As described above, the intermediate transfer type full-color LED printer has been described as the image forming apparatus 10 used in the present embodiment, but the image forming apparatus to which the present invention is applicable is not limited to this. . Specific examples of this point include adopting a tandem type or rotary type in which the photosensitive drum is directly transferred to the printing medium instead of the intermediate transfer method, and using monochrome or special color toner instead of full color. Adopting a full-color imager that does not emit light, adopting a laser head instead of an LED head as exposure means, and adopting a copier, a facsimile machine, or a digital multi-function machine that combines these functions instead of a printer. etc. is possible.

When printing is performed using the image forming apparatus 10 having the configuration described above, after the control parameters have been adjusted, the optimum control parameters may change due to deterioration over time, resulting in print defects in the actual printed matter. be. When a printing defect occurs in the actual printed material, it is necessary to readjust the control parameters of the image forming apparatus 10 to derive the optimum control parameters to obtain a high-quality print result with few printing defects. comes out.

Here, consider a case where a conventional learned model is used for this series of operations (that is, control parameter readjustment operations). First, a conventional machine learning device (not shown) that generates a conventional trained model and a conventional data processing system (not shown) that uses the conventional trained model generated by the machine learning device will be briefly described. do. As described in Japanese Patent Application Laid-Open No. 2006-254386, a conventional machine learning apparatus includes feature amount information in an actual printed material actually printed on a print medium by the image forming apparatus 10, and a state variable acquisition unit that acquires, as state variables, medium information of a print medium and first control information that is control information (control parameters) when printing the actual printed material; a teacher data acquisition unit that acquires, as teacher data, second control information, which is control information (control parameters) that and a trained model generation unit that generates a trained model by performing machine learning based on the model.

On the other hand, a conventional data processing system uses a conventional trained model generated by a conventional learning device to print on an unlearned print medium, as described in Japanese Patent Application Laid-Open No. 2006-254386. Third control information, which is control information (control parameters) for preventing printing defects from occurring in actual printed matter, is output.

Since the conventional learned model generated by the conventional machine learning apparatus includes the control parameters (first control information) of the reference machine of the image forming apparatus 10 in the state variables as described above, this learning The control parameters (third control information) obtained using the final model tend to be influenced by the control parameters (first control information) of the reference machine. Therefore, the control parameters (third control information) obtained using the conventional learned model may deviate from the optimum control parameters for each image forming apparatus 10, and readjustment may not be successful.

In other words, in the image forming apparatus 10 in which the optimal control parameters have changed due to deterioration over time, readjustment of the control parameters using the conventional learned model will not work (that is, high-speed printing with few printing defects). quality print results).

Therefore, in the present embodiment, the control parameters of the image forming apparatus 10 are readjusted using a machine learning apparatus 100 and a learned model generated by a machine learning method described below.

By the way, when executing printing using the image forming apparatus 10, various control parameters are adjusted in order to obtain an optimum printing result. As a result of verification of these various control parameters by the present inventors, among the various control parameters, the secondary transfer voltage applied to the secondary transfer roller 44 and the fixing temperature of the fixing roller 51 (that is, the temperature inside the fixing roller 51) The current value supplied to the heater) is a control parameter that directly affects the quality of printing, in other words, it is a control parameter that has a high correlation with printing defects that occur on the printing surface. was done. After conducting a more detailed examination of the relationship between these two control parameters and printing defects, we found that high secondary transfer voltage mainly causes dust (printing defects in which white spots occur). A low transfer voltage is the main cause of blurring (printing defects in which colors become light). Also, if the fixing temperature is high, it mainly causes spots (printing defects in which a speckled pattern occurs). It was found that it causes printing defects in which a thin portion occurs.

Therefore, in the machine learning device 100 according to the first embodiment of the present invention described below, the control parameters to be adjusted are the secondary transfer voltage and the fixing temperature, and these two control parameters are readjusted. A specific configuration and a series of machine learning methods for generating a trained model to be used will be explained.

[1-2. Configuration of machine learning device]
FIG. 2 is a schematic block diagram of the machine learning device 100 according to the first embodiment of the invention. The machine learning device 100 includes a state variable acquisition unit 110 , a teacher data acquisition unit 120 , a trained model generation unit 130 and a storage unit 140 . This machine learning device 100 is a device that generates a trained model by so-called supervised learning. In FIG. 2, for the purpose of facilitating understanding, the machine learning device 100 is illustrated as built in a computer (server device, PC, etc.) separate from the image forming device 10. The device 100 may be built in the image forming device 10 .

The state variable acquisition unit 110 acquires, as state variables, information necessary for generating a learned model capable of readjusting the control parameters of the image forming apparatus 10 . In machine learning, state variables are the most important factor that determines the accuracy of the generated trained model. If the combination of information acquired as state variables is different, naturally the generated trained model is also different. Therefore, it should be especially noted in the present invention that the combination is also an extremely important factor.

In the present embodiment, the state variable acquisition unit 110 acquires, as state variables, two types of information: feature amount information in the actual printed material AP and medium information of the print medium PM (see FIG. 1) used in the actual printed material AP. is. Furthermore, the state variable acquisition unit 110 also acquires first control information, which is a control parameter when outputting the actual printed material AP. As described above, in the machine learning apparatus 100 according to the present embodiment, the feature amount information in the actual printed material AP, the medium information of the printing medium PM used in the actual printed material AP, and the actual Instead of acquiring the three pieces of information of the first control information, which are the control parameters when the printed material AP is output, as state variables, among these three pieces of information, the feature amount information in the actual printed material AP and the information used in the actual printed material AP. Two pieces of medium information of the print medium PM obtained are acquired as state variables, and the first control information, which is a control parameter when outputting the actual printed material AP, is acquired as information different from the state variables. It's becoming

Note that the method of acquiring the state variables and the first control information can be arbitrarily set according to the connection state between the machine learning device 100 and the image forming device 10. be used or obtained through any storage medium. These three acquired pieces of information are stored as one data set in the storage unit 140, which will be described later.

The feature amount information in the actual printed matter AP is information about printing defects occurring in the actual printed matter AP, and more specifically includes information about printing failure information. The information about printing failure is information about the degree of printing failure. Here, the feature amount information actually input to the machine learning apparatus 100 may be image data information obtained by reading the actual printed matter AP using a known scanner. There is no need to specify the type or degree in advance. This is because the learning method of the machine learning device learns the control information as a suitable output result for the state variables regardless of the type and degree of the printing defect, and it is necessary to determine the type and degree of the printing defect. This is because there is no However, it is of course possible to employ a pre-processing process, for example, to pre-condition the information of the image data into information suitable for input to the input layer of the machine learning apparatus. For a specific preprocessing process, a person skilled in the art can easily understand that a method commonly used in the technical field of image recognition can be adopted, so the explanation thereof is omitted here.

The medium information of the print medium PM is various kinds of information about the print medium PM, and is preferably information about the presence or absence of coating, material, thickness, weight and density of the print medium PM. The above five types of media information, that is, presence or absence of coating, material, thickness, weight, and density, were identified by the present inventor as media information that has a particular effect on print quality. By performing machine learning based on it, it is possible to efficiently generate a highly accurate trained model. It should be noted that the material as the medium information is sufficient to specify the main material used in the print medium PM, and does not necessarily require information about the additionally included material. Various types of weight can be used as medium information as long as it indicates characteristics related to the weight of the print medium. Those commonly used in the technical field can be used.

The first control information is a control parameter actually set when the actual printed material AP is output. Preferably, the toner fixing temperature of the fixing roller 51 and the secondary transfer roller 44 is the applied secondary transfer voltage. In the image forming apparatus 10, since the toner image is transferred from the intermediate transfer belt 41 onto the print medium PM, the secondary transfer voltage is used as the first control information. In a tandem-type image forming apparatus that directly transfers a toner image from a photosensitive drum to a printing medium without a The current transfer voltage may be used as the first control information.

The teacher data acquisition unit 120 acquires the second control information (that is, the print result that does not include the printing failure) that is the control parameter that the feature amount information that is the information of the printing failure is within a predetermined threshold value in the actual printed material AP that includes the printing failure. Difference control information (hereinafter referred to as fourth control information) between the first control information and the second control information, which is an improved control parameter such that is obtained, is acquired as teaching data. This second control information is, for example, control information derived by an engineer based on the output result of the actual printed matter AP and control information (control parameters) at that time. Therefore, the above-mentioned "predetermined threshold value" does not necessarily indicate a specific value. It can be said that the feature amount information is within a predetermined threshold. Also, the second control information is acquired by the engineer EN by directly inputting it to the teacher data acquiring unit or by transmitting it via various communication means or an arbitrary storage medium. The second control information is stored in the storage unit 140 in association with the corresponding data set acquired by the state variable acquisition unit 110 and stored in the storage unit 140 as a data set.

The trained model generation unit 130 generates a data set consisting of two pieces of information (feature information and medium information) as state variables acquired by the state variable acquisition unit 110 and a corresponding data set acquired by the teacher data acquisition unit 120. Based on the associated teacher data, machine learning is executed to generate a trained model. A specific method of machine learning performed here will be described in detail below.

Machine learning apparatus 100 according to the present embodiment employs supervised learning using a neural network model as its learning method.

FIG. 3 is a diagram showing an example of a neural network model for supervised learning performed in machine learning apparatus 100 according to the present embodiment. The neural network in the neural network model shown in FIG. 3 has l neurons (x1 to xl) in the input layer, m neurons (y11 to y1m) in the first hidden layer, and n neurons in the second hidden layer. , and two neurons (z1, z2) in the output layer. The first intermediate layer and the second intermediate layer are also called hidden layers, and the neural network may have a plurality of hidden layers in addition to the first intermediate layer and the second intermediate layer. Alternatively, only the first intermediate layer may be a hidden layer.

Between the input layer and the first intermediate layer, between the first intermediate layer and the second intermediate layer, and between the second intermediate layer and the output layer, nodes connecting neurons between the layers are provided. , and each node is associated with a weight wi (i is a natural number).

The neural network in the neural network model according to the present embodiment learns the correlation between the control information of the image forming apparatus and the printed matter output by the image forming apparatus using the learning data set. Specifically, each of the two state variables (feature information and medium information) is associated with a neuron in the input layer, and the value of the neuron in the output layer is calculated by a general neural network output value calculation method, that is, , the value of the output-side neuron is the value of the input-side neuron connected to that neuron, and the weight wi associated with the node connecting the output-side neuron and the input-side neuron. is calculated by using a method of calculating the sum of all neurons other than the neurons in the input layer. When the state variables are associated with the neurons of the input layer, the form in which the information obtained as the state variables is associated can be appropriately set in consideration of the accuracy of the generated trained model. For example, when associating the image data of the actual printed material AP as the feature amount information in the actual printed material AP with the input layer, the image data is divided into predetermined regions, and then the color values (for example, RGB values) of each divided region are calculated. Information can be associated with each input layer.

Then, the values of the two neurons z1 and z2 in the calculated output layer, that is, the difference value of the toner fixing temperature of the fixing roller 51 and the secondary transfer voltage applied to the secondary transfer roller 44 in the present embodiment, are calculated. The difference value and the teacher data t1 and t2, which are also associated with the data set and consist of the difference value of the toner fixing temperature of the fixing roller 51 and the difference value of the secondary transfer voltage applied to the secondary transfer roller 44, respectively. An error is obtained by comparison, and the weight wi associated with each node is adjusted (back promotion) so that the obtained error becomes small.

Then, when a predetermined condition such as repeating the series of steps described above is performed a predetermined number of times or the error is smaller than the allowable value, the learning is terminated and the neural network model (of the All weights wi) associated with each node are stored in the storage unit 140 as learned models.

The learned model stored in the storage unit 140 is applied to a real system upon request via communication means such as the Internet or a storage medium. A specific application of the trained model to the actual system (data processing system) will be detailed later.

[1-3. Configuration of data processing system]
Next, an application example of the learned model generated by the machine learning device 100 described above will be described. FIG. 4 is a schematic block diagram of the data processing system according to this embodiment. In the present embodiment, as an example of a data processing system, a case where a learned model generated through the machine learning method described in detail with reference to FIG. 3 is applied in the image forming apparatus 200 will be described below. .

Image forming apparatus 200 according to the present embodiment is a full-color LED printer and has the same mechanical structure as image forming apparatus 10 (see FIG. 1). Therefore, description of the mechanical structure of the image forming apparatus 200 is omitted. The image forming apparatus 200 has a system configuration shown in FIG. 4 in addition to the mechanical structure similar to that of the image forming apparatus 10 . Specifically, the image forming apparatus 200 has a system configuration including a display/operation unit 210, an output control unit 220, an actual print information acquisition unit 230, a storage unit 250, a data processing unit 260, and a learned and a model selection unit 270 .

The display/operation unit 210 includes display means such as a liquid crystal panel provided at a predetermined position of the image forming apparatus 200 and operation means such as operation buttons or a touch panel. It is a configuration for notifying of and enabling input operation. Further, the output control unit 220 is a configuration for controlling various configurations for realizing print output by the image forming apparatus 200, and is exemplified by a configuration including at least a CPU and memories such as RAM and ROM. . Control by the output control unit 220 includes controlling the secondary transfer voltage applied to the secondary transfer roller 44 and the fixing temperature of the fixing roller 51 (specifically, the current supplied to the heater inside the fixing roller 51). I'm in.

The actual printed matter information acquisition unit 230 is configured to acquire information related to the actual printed matter AP including printing defects, and includes a feature amount information acquisition unit 231 , a medium information acquisition unit 232 , and a first control information acquisition unit 233 . and Various types of information acquired by the actual printed material information acquisition unit 230 are associated with each other as one data set and stored in the data set storage unit 252, which will be described later.

The feature amount information acquisition unit 231 is configured to acquire image data of the actual printed material AP as feature amount information. As shown in FIG. 4, the feature amount information acquisition unit 231 is locally connected to the scanner SC outside the image forming apparatus 200 via an input/output interface (not shown). The image data of the printed matter AP is acquired as feature amount information. The medium information acquisition unit 232 is configured to acquire information on the print medium PM, more specifically information on the presence or absence of coating, material, thickness, weight and density of the print medium PM. Such medium information can be acquired by the user inputting a product code assigned in advance to the print medium PM via the display/operation unit 210, for example. Note that the density as medium information is information that can be specified by calculation if the values of thickness and weight (basis weight) can be specified, so such an input operation is not necessarily required.

The first control information acquisition unit 233 also obtains first control information, which is control parameters when the actual printed material AP is output, specifically, the secondary transfer voltage applied to the secondary transfer roller 44 and the fixing temperature of the fixing roller 51 . It is a configuration for obtaining and. The first control information is information stored in the memory of the output control unit 220 or in the storage unit 250 that stores the control parameters of the output control unit 220, so it can be obtained via the internal bus.

The storage unit 250 constitutes a storage area for storing various information in the image forming apparatus 200, and includes a learned model storage unit 251, a data set storage unit 252, and a third control information storage unit 253. contains. The learned model storage unit 251 performs machine learning using a learned model generated by the machine learning method described in detail with reference to FIG. The learned model is the one that is stored. Further, in this learned model storage unit 251, a conventional learned model, that is, a learned model generated by performing machine learning using three pieces of information as state variables, namely, feature amount information, medium information, and first control information, is stored. is also remembered. A learned model generated by the machine learning method described in detail with reference to FIG. 3 will be referred to as a new learned model hereinafter.

The data set storage unit 252 stores feature amount information, medium information, and first control information related to the common actual printed material AP acquired by the actual printed material information acquisition unit 230 as one data set. The third control information storage unit 253 stores the third control information output from the data processing unit 260, which will be described later.

The data processing unit 260 selects the data set related to the specific actual printed material AP stored in the data set storage unit 252 and the two learned models stored in the learned model storage unit 251 by the trained model selection unit 270 . Various information constituting a data set is input to (each neuron of) the input layer of the selected trained model.

Specifically, when a conventional learned model is selected by the learned model selection unit 270, the data processing unit 260 stores feature information, medium information, and first control information in the input layer of this learned model. Enter 3 pieces of information. Then, the data processing unit 260 outputs the secondary transfer voltage applied to the secondary transfer roller 44 and the fixing temperature of the fixing roller 51 to the output layer of this learned model, thereby generating the third control information. .

On the other hand, when a new learned model is selected by the learned model selection unit 270, the data processing unit 260 inputs two pieces of information, feature amount information and medium information, to the input layer of this learned model. Then, the data processing unit 260 stores the differential value of the secondary transfer voltage applied to the secondary transfer roller 44 and the differential value of the fixing temperature of the fixing roller 51 as the fourth control information in the output layer of this learned model. to output Furthermore, the data processing unit 260 generates third control information (that is, secondary transfer voltage and fixing temperature) by adding the first control information in the data set regarding the actual printed material AP to the fourth control information. do.

It should be noted that regardless of whether the conventional trained model or the new trained model is used, the third control information generated by the data processing unit 260 is composed of the secondary transfer voltage and the fixing temperature, and in the actual printed material AP This is control information that can improve printing defects that have occurred. That is, data processing is performed so that a desired printed matter can be obtained by printing out with the image forming apparatus 200 using the printing medium PM used in the actual printed matter AP and using the third control information as a control parameter. It is the control information adjusted by the unit 260 .

The learned model selection unit 270 stores and manages the number of times the control parameters have been adjusted. Then, the learned model selection unit 270 selects the two learned models stored in the learned model storage unit 251 when the current adjustment of the control parameters is the first time (that is, when it is the first adjustment). Among them, the conventional trained model is selected. On the other hand, the learned model selection unit 270 selects the two learned models stored in the learned model storage unit 251 when the current adjustment of the control parameter is the second time or later (that is, when it is readjusted). Select a new trained model from among

Here, the reason why the conventional trained model is used when the control parameters are adjusted for the first time and the new trained model is used when the control parameters are readjusted will be briefly described.

As described above, the conventional trained model uses, as state variables, three types of information: feature amount information related to the actual printed material AP, medium information, and control parameters (first control information) of the reference machine of the image forming apparatus 200. and outputs the absolute value of the control parameter (third control information). For this reason, when adjusting the control parameters for the first time in the image forming apparatus 200 that has not deteriorated much over time, it is better to adjust the control parameters using the conventional learned model. There is a good chance that you will be able to adjust it properly without going too far off.

For this reason, the conventional learned model is used when adjusting the control parameters for the first time.

On the other hand, in the conventional trained model, since the control parameters (first control information) of the reference machine are included in the state variables, the control parameters (third control information) to be output are the control parameters of the reference machine. It tends to drag. Therefore, in the image forming apparatus 200 whose optimal control parameters have deviated from those of the reference device due to deterioration over time, if the control parameters are readjusted using the conventional trained model, the control parameters cannot be adjusted appropriately. , it may not be possible to obtain high-quality print results with few print defects.

On the other hand, the new learned model does not include the control parameters (first control information) of the reference machine in the state variables, and indicates how much the control parameters of the image forming apparatus 200 deviate from the optimum control parameters. The difference control information (fourth control information) shown is output. By doing this, the new trained model increases the degree of contribution of the feature amount information of the actual printed material AP, and adds the difference control information (fourth control information) to the control parameter (first control information) of the image forming apparatus 200. It is possible to avoid the control parameter (third control information) obtained by addition being influenced by the control parameter (first control information) of the reference machine. Therefore, when readjusting the control parameters of the image forming apparatus 200 whose optimal control parameters have deviated from those of the reference device due to deterioration over time, it is preferable to adjust the control parameters using a new learned model. However, it is likely that the control parameters can be adjusted appropriately.

For this reason, a new learned model is used for readjustment of the control parameters.

As described above, in the present embodiment, after the control parameters of image forming apparatus 200 are appropriately adjusted using the conventional trained model, the control parameters of image forming apparatus 200 are adjusted using the new trained model. By appropriately readjusting, even if the image forming apparatus 200 deteriorates over time, it is possible to obtain a high-quality print result with few print defects.

[1-4. Machine learning method]
Next, the procedure of the machine learning method according to this embodiment (that is, the procedure of generating a new trained model) will be described using the flowchart shown in FIG. This machine learning method is realized by using a computer, and various computers can be applied. A PC locally connected to the server or a server device arranged on a network can be used.

When executing supervised learning as the machine learning method according to the present embodiment, first, a pre-learning model having initial weights is prepared (step SP11). Next, the feature amount information of the actual printed material AP actually printed by the image forming apparatus 10 and the medium information of the printing medium PM used for the actual printed material AP are acquired as state variables, and the first control information is also obtained. Acquire (step SP12). Then, along with acquiring teacher data corresponding to the state variables acquired in step SP12 (step SP13), or before or after this, the state variables acquired in step SP12 are applied to the input layer of the pre-learning model (see FIG. 3). By associating, an output layer (see FIG. 3) is generated (step SP14).

Since the control information forming the output layer generated in step SP14 is generated by the pre-learning model, it is usually not the control information that produces a print result that satisfies the user's request. Therefore, next, machine learning is performed using the fourth control information forming the teacher data acquired in step SP13 and the control information forming the output layer generated in step SP14 (step SP15). The machine learning performed here is to compare the fourth control information that constitutes the teacher data and the control information that constitutes the output layer, detect the error between the two, and obtain an output layer that reduces this error. , performs machine learning by adjusting the weights associated with each node in the pre-learning model.

When the machine learning is performed in step SP15, it is determined whether or not it is necessary to perform further machine learning (step SP16), and if the machine learning is to be continued (No in step SP16), the process returns to step SP12, If the machine learning is finished (Yes in step SP16), the process proceeds to step SP17. When continuing the above machine learning, the steps SP12 to SP15 are performed a plurality of times, and the accuracy of the finally generated learned model usually increases in proportion to the number of times.

When machine learning ends, the neural network generated by adjusting the weight associated with each node through a series of steps is stored as a trained model in the storage unit 140 (step SP17), and a series of learning processes is performed. finish. A procedure of a data processing method for adjusting control information, which is a control parameter, using the stored learned model (that is, a new learned model) will be described later. Note that the procedure for generating a conventional trained model is the same as the procedure described in Japanese Patent Application Laid-Open No. 2006-254386, so description thereof will be omitted.

[1-5. Data processing method]
Next, using the flowchart shown in FIG. 6, the procedure of the data processing method executed in the data processing unit 260 and the trained model selection unit 270 of the data processing system according to the present embodiment (that is, using the trained model procedure for adjusting the control parameters using the

As shown in FIG. 6, the data processing unit 260 maintains the standby state when no printing error has occurred (No in step SP31), and when printing error has occurred (Yes in step SP31), the following operations are performed. It performs data processing. Whether or not a printing failure has occurred is determined based on a report by the user. The decision is made by a customer service or the like who manages the forming apparatus 200 . In the present embodiment, since various configurations are built in so that the image forming apparatus 200 alone can eliminate printing defects, if the image forming apparatus 200 is equipped with a specific mode such as a printing defect elimination mode, It is preferable in terms of convenience. In this case, when the user selects the mode via the display/operation unit 210, the following data processing is executed to eliminate printing defects.

When a printing defect occurs, the image forming apparatus 200 obtains actual printed matter information (that is, feature amount information of the actual printed matter AP, medium information and first control information) from the user. The actual printed matter information input by the user based on this request is stored as one data set in the data set storage unit 252, and the data processing unit 260 refers to the data set storage unit 252 to obtain the actual printed matter information. (step SP32). As a request for the actual printed matter information to the user, various information can be input to the user by displaying a method of obtaining various information on the display/operation unit 210, or by outputting a navigation voice through voice output means (not shown). It encourages work.

When the user inputs the necessary actual printed material information, the data processing section 260 causes the learned model selection section 270 to select a learned model to be used for adjusting the current control information, and acquires the selected learned model. (step SP33). Specifically, when the current adjustment of the control parameter (control information) is the first adjustment, the learned model selection unit 270 selects, among the learned models stored in the learned model storage unit 251, A conventional trained model is selected, and if the current control parameter adjustment is not the first adjustment (that is, readjustment), a new trained model is selected.

Then, when the acquired learned model is a conventional learned model, the data processing unit 260 stores the feature amount information of the actual printed matter AP, which is the actual printed matter information acquired in step SP32, in the input layer of this learned model, The medium information and the first control information are input, and the output of this trained model is used as the third control information. On the other hand, if the acquired trained model is a new learned model, the data processing unit 260 stores the feature amount of the actual printed material AP among the actual printed material information acquired in step SP32 in the input layer of this trained model. Information and medium information are input, the output of this learned model is used as difference control information (fourth control information), and further this difference control information (fourth control information) is added to obtain the third control information (step SP34).

The data processing unit 260 temporarily stores this third control information in the third control information storage unit 253 (step SP35), and waits for a reprint instruction from the user (step SP36). Then, for example, when the user issues a reprint instruction by operating the display/operation unit 210, the output control unit 220 replaces the first control information with the third control information in the third control information storage unit 253, After adjusting the secondary transfer voltage applied to the secondary transfer roller 44 and the fixing temperature of the fixing roller 51 (that is, after adjusting the control parameters), the printing process is executed again (step SP37).

[1-6. Summary and Effects]
As described above, in the first embodiment, the machine learning apparatus 100 is provided with the feature amount information in the actual printed material AP actually printed by the image forming apparatus 10 and the control when the actual printed material AP is printed. a state variable acquisition unit 110 for acquiring, as state variables, the feature amount information and the medium information out of the first control information, which is information, and the medium information of the print medium used in the actual printed matter AP; Acquisition of teacher data for acquiring, as teacher data, difference control information (fourth control information) that is a difference between second control information that makes information within a predetermined threshold and the first control information when the actual printed material AP is printed and a trained model generation unit that generates a trained model by performing machine learning based on the state variables acquired by the state variable acquisition unit 110 and the teacher data acquired by the teacher data acquisition unit 120. A section 130 is provided.

Further, in the present embodiment, the image forming apparatus 200, which is an application example of the machine learning apparatus 100, is provided with the feature amount information of the actual printed material AP actually printed by the image forming apparatus 200 and the actual printed material AP as a function of the data processing system. A machine learning device 100 generates an actual printed matter information acquisition unit 230 for acquiring medium information of a printing medium used in a printed matter AP, and the feature amount information and the medium information acquired by the actual printed matter information acquiring unit 230. The third control information is added to the difference control information (fourth control information) output by the learned model, and the first control information when the actual printed matter AP is printed is added to the new learned model. A data processing unit 260 for generating information and a third control information storage unit 253 for storing the third control information are provided.

As described above, in the present embodiment, the first control information is not included in the state variables used in the trained model, and the degree of deviation of the control information of the image forming apparatus 200 from the learned model from the optimum control information is calculated. By obtaining difference control information (fourth control information) indicating whether the actual printed material AP is printed by the image forming apparatus 200 and adding the difference control information (fourth control information) to the first control information when the actual printed material AP is printed, the third control information is obtained. I got the control information. By doing so, in the present embodiment, the degree of contribution of the feature amount information of the actual printed material AP to the learned model can be increased, and even if the optimum control information changes due to deterioration over time of the image forming apparatus 200, the optimum control information can be obtained. third control information can be obtained. Thus, according to the present embodiment, it is possible to obtain high-quality printing results with few printing defects.

Further, in the present embodiment, image forming apparatus 200 has two trained models, a new trained model as the first trained model and a conventional trained model as the second trained model, as functions of the data processing system. A trained model storage unit 251 for storing a model and a trained model selection unit 270 for selecting one of the two trained models stored in the trained model storage unit 251 are provided. The selection unit 270 selects the conventional learned model when adjusting the control parameters in the image forming apparatus 200 for the first time, and selects the new learned model when performing subsequent readjustments. Data processing unit 260 outputs third control information using the learned model selected by learned model selection unit 270, and output control unit 220 of image forming apparatus 200 outputs third control information based on the third control information. Control parameters (secondary transfer voltage and fixing temperature) are adjusted.

As described above, in the present embodiment, when the image forming apparatus 200 adjusts the control parameters for the first time, that is, when the image forming apparatus 200 is assumed to have not degraded so much over time, the conventional learned model is used for the first adjustment of the control parameters. The control parameters can be adjusted more appropriately by using the new learned model. can be used to adjust the control parameters more appropriately.

FIG. 7 is a graph showing the result of predetermined image quality evaluation for evaluating the quality of the actual printed material AP printed by the image forming apparatus 200 . This graph shows the PQ (deterioration) level of the image of the actual printed material AP before readjustment of the control parameters and the PQ level after readjustment of the control parameters for each of a plurality of types (A to M) of print media. there is A new learned model is used for readjustment of the control parameters. The higher the PQ level, the greater the degree of deterioration of the image, and a value of 0.6 or less indicates a practical level of quality (that is, high quality).

In this image quality evaluation, the PQ level was determined using the MT (Mahalanobis Taguchi) system. Specifically, a unit space is created from the feature amount information of a plurality of actual printed materials AP that can be visually judged to have no printing defects (that is, the image quality is good), and the MT system incorporating this unit space, The feature amount information of the actual printed material AP to be evaluated was input, and the result (that is, the output of the MT system) was used as the PQ level of the image of the actual printed material AP. In this way, in an MT system that incorporates a unit space created in advance from the feature amount information of a plurality of actual printed matter APs with good image quality, when the input feature amount information of the actual printed matter AP is close to normal (that is, the input actual If the image quality of the printed matter AP is good), a value of about 0 to 1 is output. ) outputs a value greater than one. In the present embodiment, in order to evaluate the image quality more strictly, if the output value of the MT system is 0.6 or less, the quality is evaluated as practical level.

From this graph, it can be seen that the PQ level of the actual printed matter AP is lower after readjustment than before the control parameter readjustment (that is, the quality of the actual printed matter AP is lower than that before readjustment of the control parameters) for all of the multiple types (A to M) of print media. is rising). Furthermore, for print media (A to H) where the PQ level before the control parameter adjustment is greater than 0.6 and the quality of the actual printed matter AP has not reached the practical level, for A and C to H excluding B, after readjustment of the control parameter , the PQ level is 0.6 or less, and the quality of the actual printed material AP has reached a practical level.

Although not shown in this graph, in the image quality evaluation when the control parameters are readjusted using the conventional trained model, the PQ level of the actual printed material AP before and after readjustment is was almost unchanged and no improvement in quality was observed.

As is clear from the image quality evaluation shown in FIG. 7, by readjusting the control parameters of the image forming apparatus 200 using the new learned model of the present embodiment, high-quality printing with few printing defects can be achieved. You can get results.

[2. Second Embodiment]
Next, a second embodiment of the invention will be described. FIG. 8 is a schematic block diagram of a data processing system according to the second embodiment. As shown in FIG. 8, the data processing system according to this embodiment comprises a server device 300 connected to the Internet. In addition, a plurality of terminal devices TD (for example, smart phones, tablet terminals, PCs, etc.) and image forming apparatuses 10 are connected to the server device 300 via the Internet.

The server device 300 includes a transmission/reception section 310 , a storage section 320 , a data processing section 330 , and a learned model correction section 340 . The transmitting/receiving unit 310 acquires actual printed material information and feedback information transmitted from the terminal device TD and/or the image forming apparatus 10, and transmits third control information (to be described later) to the terminal device TD and/or the image forming apparatus 10. It is for Various methods can be adopted for transmitting the actual printed material information to the transmitting/receiving unit 310. For example, the feature amount information and the medium information are transmitted via an application pre-installed in the terminal device TD, and the first control information is It may be transmitted by the image forming apparatus 10 operating with the operation of the application in the terminal device TD as a trigger. Further, the third control information transmitted by the transmission/reception unit 310 may be transmitted to the image forming apparatus 10 so that the image forming apparatus 10 reflects the third control information, or the third control information may be transmitted to the terminal device TD, The third control information may be reflected in the image forming apparatus 10 by the user inputting the information received by the terminal device TD into the image forming apparatus 10 .

The storage unit 320 constitutes a storage area for storing various information in the server device 300, and includes a trained model storage unit 321, a data set storage unit 322, and a third control information storage unit 323. contains. The learned model storage unit 321 stores a plurality of learned models so as to be compatible with various types of actual printed matter information and various image forming apparatuses. Note that the plurality of learned models stored in the learned model storage unit 321 are input with the feature amount information and the medium information of the actual printed matter AP, similarly to the learned models in the first embodiment. It outputs differential control information (fourth control information). Here, it is preferable to prepare a plurality of trained models corresponding to the types of image forming apparatuses so as to correspond to a plurality of types of image processing apparatuses having different printing methods, functions, and the like. The data set storage unit 322 stores the feature amount information, the medium information, and the first control information related to the common actual printed matter AP received by the transmission/reception unit 310 as one data set. The third control information storage unit 323 stores the third control information output from the data processing unit 330, which will be described later.

The data processing unit 330 selects a specific data set transmitted via the transmission/reception unit 310 and stored in the data set storage unit 322 and a plurality of trained models stored in the trained model storage unit 321. By using one trained model specified based on the content and the type of image forming apparatus, etc., and inputting various kinds of information constituting a data set into the input layer of this trained model, two data sets are obtained as differential control information. The differential value of the secondary transfer voltage and the differential value of the fixing temperature are output, and the secondary transfer voltage and the fixing temperature are output as third control information obtained by adding the first control information to the difference control information. It is something to do.

The learned model correction unit 340 is for correcting the corresponding learned model in the learned model storage unit 321 based on the feedback information received by the transmission/reception unit 310 so as to have higher accuracy. The feedback information refers to a case where, as a result of reprinting based on the third control information output by the data processing unit 330, the printing failure is not improved, or another printing failure occurs, the terminal device TD and/or information transmitted from the image forming apparatus 10 . The trained model correction unit 340 uses this feedback information as a learning data set for the corresponding trained model, so that the trained model once stored in the trained model storage unit 321 can be corrected as needed. It is possible.

The data processing method by server device 300 according to the present embodiment is generally the same as the method shown in FIG. Therefore, the description is omitted here.

As described above, in the second embodiment, by providing the server device 300 with the functions of the data processing system, the present system can be easily applied to existing image forming apparatuses. . In addition, by providing the learned model correction unit 340 in the server device 300, the learned model can be updated at any time, and the accuracy of data processing is always improved, so that the optimum data processing result can always be provided. .

[3. Other embodiments]
[3-1. Other embodiment 1]
In the first embodiment described above, the machine learning device 100 may be incorporated in the image forming apparatus 10 or a normal PC as an application form. Not limited to this, in the case of machine learning, especially when there are a large number of input parameters, the amount of calculation becomes extremely large. It can take a long time. Therefore, when the machine learning device 100 is incorporated in the image forming device 10 or a normal PC, a high-performance computing device such as a GPU is added, or another PC connected via a network to perform computation during machine learning. It is preferable to take measures to shorten the processing time, such as utilizing the computing power of the server device.

[3-2. Other Embodiment 2]
The information acquisition method in each configuration of the actual printed matter information acquisition unit 230 in the above-described first embodiment is not limited to the above method. As a specific example, the feature amount information acquisition unit 231 uses imaging means other than a scanner (for example, a camera function built into a smartphone, an image reading sensor installed at an outlet of an image forming apparatus, etc.). The feature amount information may be obtained by receiving the read image data of the actual printed matter AP via the Internet or the like. In place of the product code of the print medium PM, the medium information acquisition unit 232 inputs information obtained by actual measurement by the user through an interactive process via the display/operation unit 210 or a predetermined application, or By adopting various sensors for acquiring the medium information of the print medium PM on the image forming apparatus 200 side, the medium information may be acquired by a method such as automatic acquisition inside the image forming apparatus 200. .

[3-3. Other embodiment 3]
In the above-described first embodiment, the image forming apparatus 200 constituting the data processing system is illustrated as an intermediate transfer type full-color LED printer. It may be a digital multi-function device further equipped with a function and a facsimile function. In this case, since the digital MFP itself has a scanner function, there is no need to use the above-described external scanner SC when acquiring the feature amount information. Therefore, in this case, the data processing system according to the first embodiment can be configured even in a completely offline digital multifunction peripheral.

[3-4. Other embodiment 4]
Although the data processing system according to the second embodiment described above is realized by a single server device 300 for convenience of explanation, the number of server devices is not limited. It is also possible to provide this data processing system in the form of a cloud service.

[3-5. Other embodiment 5]
In the first embodiment described above, the conventional learned model is used for the initial adjustment of the control parameters (control information), and the new learned model is used for subsequent readjustments. Alternatively, for example, the initial adjustment of the control parameters may be performed by a method that does not use a conventional learned model. Specifically, the initial adjustment of the control parameters may be performed by the user. Also, an existing trained model different from the conventional trained model or a new trained model may be used when adjusting the control parameters for the first time.

In the above-described first embodiment, the learned model selection unit 270 stores and manages the number of times the control parameters have been adjusted, and selects the conventional learned model when adjusting the control parameters for the first time. , a new learned model is selected when adjusting the control parameters for the second and subsequent times. Here, for example, the learned model selection unit 270 determines the timing at which a printing medium to be printed from now on is set in the image forming apparatus 200 (for example, the timing at which a different type of printing medium from the printing medium that has been set so far is set). ), the number of times of adjustment of the control parameter may be reset. In this case, after the print medium is set in the image forming apparatus 200, the learned model selection unit 270 selects the conventional learned model during the first control parameter adjustment, and selects the conventional learned model during the second and subsequent control parameter adjustments. A new trained model will be selected.

Further, not limited to this, when the user instructs to use the conventional trained model via the display/operation unit 210, the trained model selection unit 270 selects the conventional trained model and creates a new model. A new trained model may be selected when the user instructs to use a trained model.

[3-6. Other embodiment 6]
In the above-described first embodiment, the medium information included in the state variables includes five pieces of information: presence/absence of coating of the print medium, material, thickness, weight, and density. The medium information is not limited to this, and may contain at least one of these five pieces of information, and may contain information other than these five pieces of information. The same applies to the second embodiment.

Further, in the above-described first embodiment, the control parameters adjusted by learning (that is, the control information used in the learned model) are the secondary transfer voltage applied to the secondary transfer roller 44 and the fixing temperature of the fixing roller 51. was assumed to be Alternatively, either the secondary transfer voltage or the fixing temperature may be adjusted, or control parameters other than the secondary transfer voltage and the fixing temperature may be adjusted together. In other words, the control parameters to be adjusted by learning should include at least one of the secondary transfer voltage and the fixing temperature. The same applies to the second embodiment. When the transfer method of the image forming apparatuses 10 and 200 is a method of transferring directly from the photoreceptor drum to the print medium, the control parameter adjusted by learning is not the secondary transfer voltage, but the transfer voltage from the photoreceptor drum to the print medium. This is the transfer voltage for direct transfer.

[3-7. Other embodiment 7]
Furthermore, the present invention is not limited to the embodiments described above. That is, the scope of the present invention also extends to embodiments obtained by arbitrarily combining a part or all of the above-described embodiments, and embodiments obtained by extracting a part thereof. It is possible to implement various modifications within a range that does not deviate. All of them are included in the technical idea of the present invention.

INDUSTRIAL APPLICABILITY The present invention can be widely used, for example, in an electrophotographic image forming apparatus.

10 image forming apparatus 44 secondary transfer roller 51 fixing roller 70 control unit 100 machine learning device 110 state variable acquisition unit 120 teacher data acquisition unit 130... Learned model generation unit 140... Storage unit 200... Image forming apparatus 220... Output control unit 230... Actual printed matter information acquisition unit 231... Feature amount information acquisition unit 232... Medium information acquisition unit 233 First control information acquisition unit 250 Storage unit 251 Learned model storage unit 252 Data set storage unit 253 Third control information storage unit 260 Data processing unit 270 Trained model selection unit 300 Server device 320 Storage unit 330 Data processing unit 340 Trained model correction unit AP Actual printed matter PM … printed media.

Claims (9)

Of the feature amount information in the actual printed material actually printed by the image forming apparatus, the first control information that is the control information when the actual printed material is printed, and the medium information of the printing medium used in the actual printed material , a state variable acquisition unit that acquires the feature amount information and the medium information as state variables;
a teacher data acquisition unit that acquires, as teacher data, differential control information that is a difference between second control information that sets the feature amount information within a predetermined threshold value and the first control information when the actual printed matter is printed;
a trained model generation unit that generates a trained model by performing machine learning based on the state variables acquired by the state variable acquisition unit and the teacher data acquired by the teacher data acquisition unit. Device. The machine learning device according to claim 1, wherein the feature amount information includes information about printing defects in the actual printed material. 3. The machine learning device according to claim 1, wherein the medium information includes at least one of coating presence/absence, material, thickness, weight, and density of the print medium. The image forming apparatus is an electrophotographic system,
4. The machine learning device according to claim 1, wherein said first control information and said second control information include at least one of toner fixing temperature and transfer voltage in said image forming apparatus. an actual printed material information acquiring unit for acquiring feature amount information of an actual printed material actually printed by an image forming apparatus and medium information of a printing medium used for the actual printed material;
The feature amount information and the medium information acquired by the actual printed matter information acquisition unit are input to a trained model generated by the machine learning device according to any one of claims 1 to 4, and the trained model is a data processing unit that generates third control information by adding first control information obtained when the actual printed matter is printed to the output difference control information;
A data processing system comprising: a third control information storage unit that stores the third control information. a trained model storage unit that stores the first trained model as a first trained model and a second trained model that is different from the first trained model;
a trained model selection unit that selects one of the first trained model and the second trained model stored in the trained model storage unit;
The second trained model is
The feature amount information in the actual printed material actually printed by the image forming apparatus, the first control information when the actual printed material was printed, and the medium information of the printing medium used in the actual printed material are used as state variables, and A learned model generated by performing machine learning based on the state variables and the teacher data, using second control information having feature amount information within a predetermined threshold value as teacher data,
The data processing unit
When the first learned model is selected by the learned model selection unit, the feature amount information of the actual printed material actually printed by the image forming apparatus and the medium information of the printing medium used for the actual printed material are selected. Inputting to the first learned model and adding the first control information when the actual printed material is printed to the difference control information output by the first learned model to generate the third control information;
When the second learned model is selected by the learned model selection unit, feature amount information of an actual printed material actually printed by an image forming apparatus, first control information when the actual printed material is printed, 6. The data processing system according to claim 5, wherein the medium information of the print medium used for the actual printed matter is input to the second trained model, and the output of the second trained model is used as the third control information. The learned model selection unit
7. The data processing according to claim 6, wherein the second learned model is selected when the control information is adjusted for the first time in the image forming apparatus, and the first learned model is selected when the control information is readjusted in the image forming apparatus. system. A machine learning method using a computer,
Of the feature amount information in the actual printed material actually printed by the image forming apparatus, the first control information that is the control information when the actual printed material is printed, and the medium information of the printing medium used in the actual printed material , a step of acquiring the feature amount information and the medium information as state variables;
a step of obtaining, as teacher data, difference control information, which is a difference between second control information in which the feature amount information is within a predetermined threshold and the first control information when the actual printed matter is printed;
A machine learning method comprising: generating a trained model by performing machine learning based on the acquired state variables and the teacher data. using a computer to acquire feature amount information of an actual printed material actually printed by an image forming apparatus and medium information of a printing medium used for the actual printed material;
The obtained feature amount information and the medium information are input to a trained model generated by the machine learning method according to claim 8, and the actual printed material is printed on the differential control information output by the trained model. adding the current first control information to generate third control information;
and storing the third control information.

Images