Machine learning apparatus, machine learning method, and program

Abstract

To provide a machine learning apparatus, a method thereof, and a program therefor which select, based on device configuration information, learning data collected for learning of a learning model for estimating treatment for failure based on a kind of device failure, and use it efficiently.SOLUTION: A machine learning apparatus according to the present invention has storage means for storing a kind of failure for each kind of a device, a component of a device as an origin of failure, and treatment for the failure as failure treatment data, and learning means for carrying out learning of a learning model for estimating a treatment content from the kind of the failure for each kind of the device with using the failure treatment data for each kind of the device as the learning data. The learning means uses, as learning data, the failure treatment data of a first kind device, the failure treatment data relating to the component common to the device of the kind different from that of the first kind device having the common component to the component as the failure origin of the device included in the treatment data, and carries out learning of the learning model for estimating the treatment content from the kind of the failure as to the first kind device.SELECTED DRAWING: Figure 8

Description

[Technical Field]

[0001] The present invention relates to a machine learning device, a machine learning method, and a program. Specifically, the present invention relates to the creation of training data to be input into a learning model using artificial intelligence, and to generating a learning model using data collected from multiple models as training data based on device configuration information. [Background technology]

[0002] Conventionally, when an abnormality such as an error or breakdown occurs in equipment, a serviceman or other maintenance personnel is dispatched upon notification of the abnormality. The maintenance personnel checks the manual and performs maintenance such as replacing parts.

[0003] Cloud computing has also become increasingly popular in recent years. Its main feature is that it uses many computing resources to distribute data conversion and processing, and processes requests from many clients in parallel using distributed parallel processing. By using cloud computing, system developers can easily procure the necessary computing resources, allowing them to focus on developing system functions.

[0004] One of the elements that is highly compatible with cloud computing is AI (Artificial Intelligence). Machine learning is one of the core technologies that realizes AI. Machine learning analyzes large amounts of data (big data) using learning algorithms to create learning models that extract data features (traits, patterns, trends, etc.). Safely storing and analyzing such large amounts of data requires a large amount of computing resources, so it is often implemented in cloud computing environments.

[0005] In general, in machine learning, a learning model is trained using training data, and once the training is complete, the model is put into production as a trained model. The training data used to train the learning model is essential for building the trained model, but in reality, depending on the type of defect, it is expected that collecting defect information will take an enormous amount of time.

[0006] Patent Document 1 proposes a learning model that predicts the occurrence of malfunctions based on the detection results of one or more sensors possessed by a home appliance. In Patent Document 1, a trained model is generated for each major classification of basic types of home appliances and a trained model is generated for each minor classification of each model, and fault diagnosis of the home appliance is performed using the two trained models. As a result, even if sufficient training data cannot be obtained from the minor classification of each model, more training data can be obtained from each major classification of basic types, making it possible to improve prediction accuracy. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2021-177319 Summary of the Invention [Problem to be solved by the invention]

[0008] However, as in Patent Document 1, even if the basic format of a home appliance is the same, if the appliance configuration is different, it may be impossible to apply the generated trained model. Furthermore, the situation where sufficient training data cannot be obtained because it takes an enormous amount of time to collect information about a specific defect remains unresolved.

[0009] The present invention has been made in consideration of the above-mentioned problems, and aims to provide a mechanism for selecting collected learning data based on device configuration information and efficiently using the collected learning data. [Means for solving the problem]

[0010] The machine learning device of the present invention comprises a storage means for storing the type of failure for each type of equipment, the component part of the equipment that caused the failure, and the remedial action content for the failure as failure remedial action data, and a learning means for training a learning model for estimating the remedial action content from the type of failure for each type of equipment using the failure remedial action data for each type of equipment as training data, wherein the learning means trains a learning model for estimating the remedial action content from the type of failure for the first type of equipment using, as training data, failure remedial action data for a first type of equipment and failure remedial action data for a common component part of a different type of equipment from the first type of equipment that has a common component part with the failed part that caused the failure of the equipment included in the failure remedial action data of the first type of equipment. [Effects of the Invention]

[0011] In the present invention, when a component included in faulty component information included in learning data collected from a first device is a common component with a component of a second device, the learning data collected from the first device and the learning data collected from the second device are used as learning data for the first device. This provides a mechanism for selecting collected learning data based on device configuration information and efficiently using the collected learning data. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is a configuration diagram of a machine learning system according to the present invention. [Figure 2] FIG. 1 is a hardware configuration diagram of a machine learning system according to the present invention. [Figure 3] FIG. 4 is a sequence diagram of a fault diagnosis process according to an embodiment. [Figure 4] 1 illustrates an example of a scanned image containing an anomaly, according to one embodiment. [Figure 5] An example of the screen of an information input terminal. [Figure 6] FIG. 2 is a software configuration diagram of a fault diagnosis result notification server according to an embodiment. [Figure 7]10 is a flowchart showing a learning data collection process performed by a fault diagnosis result notification server according to an embodiment. [Figure 8] 10 is a flowchart of a machine learning process according to an embodiment. [Figure 9] 1 is a cross-sectional view of an image forming apparatus according to an embodiment of the present invention; [Figure 10] FIG. 2 is a block diagram showing the configuration of an electrical system in one embodiment. [Figure 11] 10A and 10B show examples of a device component parts list and a diagnosis target model series list according to an embodiment. [Figure 12] 10 is a flowchart of a machine learning process according to an embodiment. [Figure 13] 1 is a list of equipment components according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the scope of the invention claimed. Although multiple features are described in the embodiments, not all of these multiple features are necessarily essential to the invention, and multiple features may be combined arbitrarily. Furthermore, in the accompanying drawings, the same reference numerals are used to designate the same or similar components, and redundant explanations will be omitted. [Example]

[0014] Hereinafter, the data generated in this proposal to generate a machine learning model will be referred to as training data, and the data input to a trained model for estimation will be referred to as input data. Data for retraining an already created machine learning model will also be referred to as training data, and retraining will also be included in the term "learning." When we say that a learning model is trained using training data, this includes both creating a learning model using training data and retraining a learning model using training data.

[0015] The machine learning device of this embodiment collects learning data including faulty parts and repair details corresponding to equipment malfunctions, creates a machine learning model using the learning data, and uses the created learning model to estimate the corrective action to be taken when notified of an equipment malfunction. The user who performs the treatment may be a maintenance person who performs the service, or the user of the equipment. Replacement of consumables may be performed not only by a maintenance person but also by the user of the equipment.

[0016] The machine learning device of this embodiment is configured to include a fault diagnosis result notification server 104. The fault diagnosis device may further include an equipment malfunction information collection server 102, a replacement part information collection server 105, and an operation information collection server 106. The equipment malfunction information collection server 102, the fault diagnosis result notification server 104, the replacement part information collection server 105, and the operation information collection server 106 may be configured in any desired manner, and may be configured as a single unit or in a distributed configuration. The fault estimation results by the fault diagnosis result notification server 104 are displayed on a Web UI (Web-based user interface) of the portal site of the information input terminal 103. A user such as a maintenance technician views the estimation results and uses them as a reference for the content of maintenance work.

[0017] <System configuration> The configuration of a fault diagnosis system for providing online fault diagnosis services for implementing the present invention will be described with reference to Figure 1. The fault diagnosis system of the present invention includes an equipment malfunction information collection server 102, an information input terminal 103, a fault division result notification server 104, a replacement parts information collection server 105, an operation information collection server 106, an image forming apparatus 110, and a network 101.

[0018] The multiple image forming devices 110 are, for example, digital multifunction peripherals, facsimile machines, laser beam printers, scanners, etc. In this embodiment, the image forming devices 110 are detection targets that detect a malfunction that occurs in the image forming devices 110 as a predetermined event. Note that the predetermined device applied as the detection target is not limited to the image forming devices 110, but may be, for example, other information processing devices such as personal computers or mobile terminals. Furthermore, the predetermined event is not limited to a malfunction, but may also be, for example, a command to a device.

[0019] The device malfunction information collection server 102 is a server that collects information from the image forming apparatuses 110. It collects and stores malfunction history information such as errors and paper jams from the multiple image forming apparatuses 110 via the network 101, as well as image abnormality information related to image abnormalities contained in images.

[0020] The replacement part information collection server 105 stores information on replacement parts replaced by users during maintenance work. The information on replacement parts may be acquired from the image forming apparatus 110 or may be input by the user.

[0021] The information input terminal 103 is an information input terminal used by a user. The user inputs the details of the measures that were actually taken from the information input terminal 103, and transmits the details to the fault diagnosis result notification server 104 via the network 101.

[0022] The fault diagnosis result notifying server 104 is a server that creates and stores learning data based on the malfunction information and the user's response, and performs diagnosis. The fault diagnosis result notifying server 104 receives various information via the network 101. The various information includes feedback information sent by the information input terminal 103, malfunction history information held by the equipment malfunction information collecting server 102, and replacement part information held by the replacement part information collecting server 105. Learning data is created and stored based on the various received information.

[0023] The replacement part information collection server 105 receives the replacement part information transmitted by the information input terminal 103 via the network 101, and collects and stores the acquired information. The replacement part information may be acquired from the image forming apparatus 110. The operation information collection server 106 receives information on the operation of the image forming apparatuses 110, such as sensor data transmitted from the image forming apparatuses 110, via the network 101, and collects and stores the information as operation information.

[0024] In the following, as an example, the equipment malfunction information collecting server 102, the fault diagnosis result notifying server 104, the replacement part information collecting server 105, and the operation information collecting server 106 will be described as separate servers. The servers 102, 104, 105, and 106 may be configured as a single server, or may be configured as multiple servers with the functions of the servers 102, 104, 105, and 106 distributed. The server configuration is not limited to the form shown in FIG. 1.

[0025] <Hardware Configuration of Image Forming Apparatus 110> 2A, the hardware configuration of an image forming apparatus 110 according to an embodiment of the present invention will be described. The image forming apparatus 110 includes a CPU 201, a ROM 203, a RAM 204, a network interface card 205, an external memory 206, an operation panel 207, a storage device 208, a device interface 209, a printer 210, and a scanner 202. These components are connected via a system bus 200.

[0026] The CPU 201 performs overall control of access to various devices connected to the system bus 200. The CPU 201 performs control by reading out control programs stored in a ROM 203 or control programs and resource data (resource information) stored in an external memory 206 connected via a disk controller or the like into a RAM 204 and executing them.

[0027] The ROM 203 stores various data such as programs such as the basic I / O program, font data used in document processing, template data, etc. The RAM 204 functions as the main memory, work area, etc. of the CPU 201, and is configured so that the memory capacity can be expanded by an optional RAM connected to an expansion port (not shown).

[0028] The network interface card 205 is an interface with external devices, and the image forming apparatus 110 exchanges data with external devices via the network interface card 205. The operation panel 207 displays screens and accepts user operation instructions via the screen. It also has buttons for setting the operation mode of the printing apparatus, displaying the operating status of the printing apparatus, and performing operations such as copy designation, as well as a display unit such as an LCD panel.

[0029] The storage device 208 is an external storage means that functions as a large-capacity memory. The device interface 209 is a connection interface with external devices that can be connected via USB or the like. The printer 210 uses known printing technology, and suitable implementations include electrophotography (laser beam), inkjet, and dye sublimation (thermal transfer) methods. The printer 210 prints image data converted from PDL (Page Description Language), PDF (Portable Document Format), or the like as print data onto paper.

[0030] The scanner 202 uses known image reading technology to optically scan a paper document placed on a transparent top plate and convert it into an image, and also successively reads multiple paper documents placed on an automatic document feeder (ADF) and converts them into images.

[0031] <Hardware configuration of each server> 2(B), the hardware configurations of the equipment malfunction information collecting server 102, the fault diagnosis result notifying server 104, the replacement part information collecting server 105, and the operation information collecting server 106 according to an embodiment of the present invention will be described. The hardware configurations of the servers 102, 104, 105, and 106 are basically the same.

[0032] The servers 102, 104, and 105 each include a CPU 221, a GPU 222, a ROM 223, a RAM 224, a network interface card 225, an external memory 226, an input / output interface 227, a storage device 228, and a device interface 229. The respective components are connected to each other via a system bus 220.

[0033] The CPU 221 controls the entire device and comprehensively controls access to various devices connected to the system bus 220. The CPU 221 performs control by loading control programs stored in a ROM 223 or control programs and resource data (resource information) stored in an external memory 226 connected via a disk controller or the like into a RAM 224 and executing them. The GPU 222 is a computing device specialized for vector operations such as image processing and machine learning.

[0034] The ROM 223 is a storage means and stores various data such as basic I / O programs. The RAM 224 functions as the main memory, work area, etc. of the CPU 221 and GPU 222, and is configured so that the memory capacity can be expanded by an optional RAM connected to an expansion port (not shown). The network interface card 225 is an interface with external devices, and the server exchanges data with external devices via the network interface card 225 .

[0035] The input / output interface 227 can display screens and receive user operation instructions via devices such as a display, keyboard, mouse, smartphone, or tablet. The storage device 228 is an external storage means that functions as a large-capacity memory. The device interface 229 is a connection interface with an external device that can be connected via USB or the like.

[0036] <Cross-sectional view of image forming apparatus 110> Fig. 9(A) is a cross-sectional view of an image forming apparatus 110 according to an embodiment of the present invention. With reference to Fig. 9(A), an operation of forming an image on a recording material P will be described. The image forming apparatus 110 forms images by an electrophotographic method and employs a so-called tandem method. Fig. 10 is a block diagram showing the configuration of the electrical system in this embodiment.

[0037] Drum cartridges for forming toner images in four colors, yellow (Y), magenta (M), cyan (C), and black (K), are installed side by side. In the following explanation, parts that are common to the four color drum cartridges in Figure 9(A) are shown with only numbers, omitting the Y, M, C, and K symbols.

[0038] The image forming apparatus 110 has one or more paper feed cassettes 10. A pick roller 11 picks up recording materials P stored in the paper feed cassette 10 and feeds them to a conveyance path. A separation roller 12 is a conveyance roller that separates only the topmost recording material P when multiple sheets of recording materials P are taken out and conveys it further downstream. A pre-registration roller 13, provided downstream of the separation roller 12, is a conveyance roller that conveys the recording material P further downstream. "REGI" is an abbreviation for "registration."

[0039] The registration rollers 14, provided downstream of the pre-registration rollers 13, are conveying rollers that convey the recording material P further downstream. A motor M1 is connected to the pick roller 11, a motor M2 to the separation roller 12, a motor M3 to the pre-registration rollers 13, and a motor M4 to the registration rollers 14. Each roller is driven by its own motor. In this embodiment, the motors M1 to M4 are brushless DC motors. Further, a torque detector 16 is connected to the motor M2, and measures the driving torque of the separation roller 12.

[0040] A registration sensor 15 provided downstream of the registration roller 14 outputs a signal indicating that the recording material P is passing during the period from when the leading edge of the recording material P is detected until when the trailing edge is detected. The transport time from when the drive of the pick roller 11 is instructed to be driven until the registration sensor 15 detects the leading edge of the recording material P is monitored to detect transport delays and jams.

[0041] Image formation in this image forming apparatus 110 is performed as follows. First, the surface of the photosensitive drum 21 is uniformly charged by the charger 22. The charged surface is exposed to a laser 23, thereby forming an electrostatic latent image on the photosensitive drum 21. Toner is applied from a developer 24 to the electrostatic latent image thus obtained, thereby developing it into a toner image. This toner image is transferred onto an intermediate transfer belt 26 by a primary transfer roller 25.

[0042] In parallel with the toner image formation operation, recording material P is conveyed one sheet at a time from paper feed cassette 10 through a conveying path to registration rollers 14. Registration rollers 14 correct any skew of the recording material P. After the skew has been corrected, the recording material P is conveyed by registration rollers 14 to a secondary transfer section. The multiple color toner images transferred onto intermediate transfer belt 26 in a superimposed state are transferred onto the conveyed recording material P at the secondary transfer section where inner secondary transfer roller 31 and outer secondary transfer roller 32 abut against each other. The toner image on this recording material P is heated and pressed by fixing device 4 to be fixed, and then discharged outside image forming apparatus 110.

[0043] <Secondary transfer unit configuration> The secondary transfer unit includes an inner secondary transfer roller 31 and an outer secondary transfer roller 32. The inner secondary transfer roller 31 is disposed opposite the outer secondary transfer roller 32 across the intermediate transfer belt 26. A secondary transfer power supply 33 is connected to the outer secondary transfer roller 32. A voltage detection sensor 33a that detects the output voltage and a current detection sensor 33b that detects the output current are connected to the secondary transfer power supply 33.

[0044] The secondary transfer power supply 33 applies a DC voltage as a secondary transfer voltage to the outer secondary transfer roller 32. The outer secondary transfer roller 32 contacts the intermediate transfer belt 26 to form a secondary transfer portion between the outer secondary transfer roller 32 and the intermediate transfer belt 26. When a secondary transfer voltage of a polarity opposite to that of the toner is applied to the secondary transfer portion, the outer secondary transfer roller 32 performs a second transfer of the toner image on the intermediate transfer belt 26 onto the recording material P supplied to the secondary transfer portion. The core metal of the inner secondary transfer roller 31 is connected to a ground potential.

[0045] <Configuration of Fixing Device 4> 9(B) shows a cross-sectional configuration of the fixing device 4. The fixing device 4 includes a fixing film 42 with a heater 41 disposed therein, and a pressure roller 43 that forms a fixing nip with the heater 41 and transmits driving force to the fixing film 42. The heater 41 heats the fixing film 42 in the fixing nip portion. The fixing device 4 also includes a first thermistor 44a and a second thermistor 44b as temperature detection means for detecting the temperature of the heater 41.

[0046] <Learning data collection> In this embodiment, when a malfunction occurs in the image forming device 110, feedback information that associates information about the faulty part that caused the malfunction with information indicating the details of the treatment for the faulty part is collected as learning data based on the treatment performed by the user. Hereinafter, image abnormalities contained in image data will be taken as examples of defects, and learning data collection related to the image abnormalities will be described.

[0047] In this embodiment, the CPU 201 of the image forming apparatus 110 transmits a scanned image read by the scanner 202 to the device malfunction information collection server 102 via the network 101. The CPU 221 of the fault diagnosis result notification server 104 is capable of executing image diagnosis processing to identify the type and position information of an image abnormality contained in the scanned image, a candidate faulty part indicating the faulty part that caused the image abnormality, and candidate treatment contents indicating treatment contents for the faulty part. Treatment contents for the faulty part include replacement, cleaning, adjustment, repair, etc. of the faulty part. An overview of the image diagnosis processing executed in this embodiment will be described below. Fig. 3 is a sequence diagram relating to the image diagnosis processing shown in this embodiment. In the following, the step numbers of each process included in the sequence diagram are indicated by numbers beginning with "S."

[0048] First, in S301, the CPU 201 reads a paper document using the scanner 202. The scanner 202 outputs an actual image on a paper document printed by the image forming apparatus 110, or a printed test chart, as a scanned image 401. For example, the operation panel 207 performs an abnormal image reception function to receive a scanned image containing an image abnormality. The operation panel 207 displays a message prompting the user to set the paper document on the scanner 202 and then press the start button. This causes the user to set the paper document on the scanner 202 and press the start button on the operation panel 207. Note that, normally, when there is an image abnormality in the paper document, the abnormal image reception function is used. It is also assumed that the user will scan a paper document containing an image abnormality from the image forming apparatus 110 using the abnormal image reception function.

[0049] In S302, the CPU 201 acquires a scanned image 401 obtained by reading a paper document. In S303, the CPU 201 transmits the scanned image 401 and a timestamp (scan date and time information) to the device malfunction information collection server 102. Fig. 4A shows an example of the scanned image 401 for explaining an abnormal image according to this embodiment. Image abnormalities 402 and 403 are image abnormalities contained in the scanned image 401. In S304, the CPU 221 of the failure diagnosis result notifying server 104 acquires the scanned image 401 of the diagnosis target from the device malfunction information collecting server 102.

[0050] In S305, the CPU 221 of the fault diagnosis result notification server 104 uses the trained learning model to estimate position information of the image abnormality contained in the scanned image 401, information about the faulty part that caused the image abnormality, and the details of the remedial action to be taken for the faulty part. That is, the image forming apparatus 110 can estimate the position information of the image abnormality, the faulty part, and the details of the remedial action simply by reading a paper document. Here, the machine learning may be performed using deep learning or other known object detection algorithms (object detection models).

[0051] Machine learning uses images containing image anomalies as training data for learning. A maintenance technician visually inspects an image containing an image anomaly, such as the scanned image 401 sent as an image anomaly, and identifies the area containing the anomaly by surrounding the anomaly with a pointing device or the like. Then, the type of image anomaly is indicated for each area containing an anomaly. Types of image anomalies include, for example, circular anomalies (stains (dots)) and streak-like anomalies (stains (streaks)). In this way, a learning model that can identify anomalies from an image can be created.

[0052] Identifying abnormalities can also be done through image recognition processing. For example, abnormalities can be identified by comparing an abnormal image with a properly printed image or RIP image. If there is only one abnormal area in an image, the user simply needs to specify the type of image abnormality. If there are multiple abnormal areas, the user selects the areas and identifies the type of abnormality for each area.

[0053] FIG. 4(B) shows an example of a diagnostic image 411 for explaining the results of executing machine learning according to this embodiment. The CPU 221 of the image diagnosis result notification server 104 outputs a bounding box 412 indicating an image abnormality 402, an image abnormality type "image abnormality," and a certainty factor. Furthermore, because the scanned image 401 contains multiple image abnormalities, a bounding box 413 indicating an image abnormality 403, an image abnormality type "image abnormality," and a certainty factor are output. The bounding boxes 412 and 413 contain area information that identifies an area indicating an image abnormality to be detected. For example, they include position information of the image abnormality and area information indicating the type of the image abnormality. Furthermore, the certainty factor is the likelihood of the detection result and is expressed as a numerical value between 0 and 100, for example.

[0054] If the bounding box or image anomaly type included in the diagnostic image is incorrect, the user can correct it and have the learning model learn using feedback.In addition, by providing feedback that the answer is correct, the confidence level of the image diagnosis can be increased.

[0055] The training data for the treatment details corresponding to the abnormal image is data that associates the type of image abnormality with the treatment details taken by a user such as a maintenance technician. The user selects a bounding box included in the abnormal image and inputs the treatment details. If the abnormal image contains only one abnormal location, there is no need to select a bounding box. By associating the abnormal image with the treatment taken to resolve the abnormal image, the type of abnormality can be identified, and the treatment details corresponding to the type of abnormality can be identified.

[0056] The display of the information input terminal 103 will be described with reference to FIG. The display screen of the information input terminal 103 displays a screen of a portal site 501. The portal site 501 is an example of a portal screen for inputting information to be transmitted to the fault diagnosis result notifying server 104. The portal site 501 includes a target image abnormality detail information display section 510, recommended action content display sections 511 and 513, feedback input sections 512 and 514, a scanned image display section 515, and image abnormality position information display sections 516 and 517.

[0057] The information input terminal 103 is a terminal used by a user. It may be a tablet terminal or a smartphone carried by a maintenance technician when performing maintenance on the image forming apparatus 106. The user can scan a paper document having an image abnormality by receiving an abnormal image report on the image forming apparatus 110 to be maintained, and then check the image abnormality and the recommended action to be taken on the information input terminal 103, which is a tablet terminal.

[0058] The image abnormality detailed information display unit 510 displays the product name, machine number, and date and time when the image abnormality occurred. The product name is the product type of the image forming device 110. The machine number is a unique ID assigned to each image forming device 110. The image abnormality occurrence date and time is the date and time when the image abnormality occurred.

[0059] The recommended action content display unit 511 or 513 displays, for each image anomaly, candidate faulty parts and candidate action content that are estimated by the failure diagnosis result notification server 104 to have a high probability of resolving the image anomaly. If one scanned image 401 contains multiple image anomalies, candidate faulty parts and candidate action content are displayed for all image anomalies. The recommended action content may include replacement, cleaning, repair, etc. of a specific part. Furthermore, the display of candidate faulty parts and candidate action content may indicate the likelihood that the processing will resolve the event based on the estimation results. In the example of the recommended action content display section 511, the image abnormality position information display section 516 of the scanned image display section 515 indicates that the probability that the abnormality will be resolved by replacing part A is 80%, by replacing part B it is 15%, and by cleaning part C it is 5%.

[0060] The feedback input section 512 is a screen for inputting feedback information. The details of the measures taken to resolve the actual abnormal event are input as feedback information. The input method may be a selection format such as a check box, or a free input format using text. If multiple image abnormalities exist in one scanned image 401, feedback information is input for all of the image abnormalities. In the example of the feedback input section 512, the part A displayed in the recommended treatment content display section 511 is replaced with the abnormal part 516 in the image displayed in the scan image display section 515.

[0061] The scanned image display section 515 is a display section that displays the scanned image 401 sent by the user, and if an image abnormality is present, the abnormal area is displayed as a rectangle. If multiple image abnormalities are present in one scanned image 401, rectangles are displayed in the areas of all image abnormalities. The example of the scanned image display section 515 in Figure 5 shows the presence of bounding boxes 516 and 517, which display the abnormal areas included in the scanned image display section 515 as rectangles.

[0062] The software configuration of the fault diagnosis result notifying server 104 of the present invention will be described with reference to Fig. 6. The programs of the fault diagnosis result notifying server 104 are read from the RAM 224, the storage device 228, a secondary storage device connected via the device interface 229, etc., and are executed by the CPU 221 and GPU 222 of the fault diagnosis result notifying server 104. Access to the external devices such as the device malfunction information collecting server 102 and the information input terminal 103 is performed via the network interface card 225.

[0063] The fault diagnosis result notifying server 104 has, as data storage units, an equipment malfunction information storage unit 601, a part replacement information storage unit 602, and a feedback storage unit 603. The fault diagnosis result notifying server 104 also has, as software functional units, a learning / input data management unit 604, a learning execution unit 605, a machine learning model management unit 606, an estimation execution unit 607, and an estimation result storage unit 608.

[0064] The device malfunction information storage unit 601 stores image abnormality information of the image forming device 110 and the like, which the CPU 221 of the failure diagnosis result notification server 104 receives from the device malfunction information collection server 102 via the network 101 . The part replacement information storage unit 602 stores the replacement part information that the CPU 221 of the fault diagnosis result notification server 104 receives from the replacement part information collection server 105 via the network 101 . The feedback storage unit 603 receives and stores the feedback information input by the CPU 221 of the fault diagnosis result notifying server 104 via the feedback input units 512 and 514 of the information input terminal 103 via the network 101 .

[0065] The learning / input data management unit 604 creates and stores learning data and input data based on each element information stored in the fault diagnosis result notifying server 104. Each element information includes the following information. Malfunction information stored in the device malfunction information storage unit 601 Replacement part information stored in the part replacement information storage unit 602 Feedback information stored in the feedback storage unit 603 The learning and input data management unit 604 creates and stores learning data when the CPU 221 of the fault diagnosis result notification server 104 learns a machine learning model, and creates and stores input data when making an estimation using the machine learning model.

[0066] The learning execution unit 605 creates a machine learning model by having the CPU 221 of the fault diagnosis result notification server 104 acquire learning data from the learning and input data management unit 604 and execute learning based on a pre-specified machine learning algorithm. The created machine learning model is stored in the machine learning model management unit 606 by the CPU 221 of the fault diagnosis result notification server 104. Note that as the learning data stored in the learning and input data management unit 604 changes, learning may be executed repeatedly to recreate (re-learn) the machine learning model.

[0067] The machine learning model management unit 606 stores the machine learning model created by the learning execution unit 605 by the CPU 221 of the fault diagnosis result notification server 104. Note that the machine learning model used for estimation may be replaced when triggered by reception of a machine learning model from the learning execution unit 605 or a condition determination within the machine learning model management unit 606. For example, the current machine learning model may be replaced with a new machine learning model when the accuracy rate of the new machine learning model exceeds a certain level. The estimation execution unit 607 executes estimation by having the CPU 221 of the fault diagnosis result notification server 104 acquire input data from the learning / input data management unit 604 and input it into the machine learning model stored in the machine learning model management unit 606.

[0068] The estimation result storage unit 608 stores the results of estimation performed by the estimation execution unit 607, which is performed by the CPU 221 of the fault diagnosis result notifying server 104. The estimation result storage unit 608 also transmits the estimation results to the information input terminal 103 via the network 101. Alternatively, the estimation result storage unit 608 may receive a request from the information input terminal 103 via the network 101 and return the estimation results. The information input terminal 103 displays the estimation results on the portal screen, as shown in recommended action contents 511 and 513 and in the scanned image display unit 515. Taking the bounding box 516 and recommended action contents 511 in the scanned image display unit 515 as an example, the recommended action contents of the bounding box 516 indicate that the likelihood of part A is 80%, the likelihood of part B is 15%, and the likelihood of part C is 5%.

[0069] The diagnosis target model information management unit 609 stores a diagnosis target model series list 1110 of the failure diagnosis target for generating a machine learning model. The device configuration information management unit 610 stores a device configuration parts list 1101 relating to each model of device to be diagnosed for which a machine learning model is to be generated. The diagnosis target model series list 1110 and the device component parts list 1101 will be described later with reference to FIG.

[0070] The proposed method shown in this embodiment will be described with reference to FIGS. Fig. 7(A) is an overall flowchart of the fault diagnosis result notifying server 104 in this embodiment. The processes of Fig. 7(A), Fig. 7(B), and Fig. 8 are realized, for example, by the CPU 221 or GPU 22 of the fault diagnosis result notifying server 104 reading out a program stored in the ROM 223 or the external memory 226 into the RAM 224 and executing it. In the following, the step numbers of each process included in the flowcharts are indicated by numbers beginning with "S". The same applies to the subsequent flowcharts.

[0071] First, in S701, the CPU 221 of the fault diagnosis result notifying server 104 collects learning data using the learning / input data management unit 604. Next, in S702, the CPU 221 of the fault diagnosis result notifying server 104 uses the collected learning data to create a machine learning model using the learning execution unit 605. Finally, in S703, the CPU 221 of the fault diagnosis result notifying server 104 stores the created machine learning model in the machine learning model management unit 606.

[0072] When an image abnormality occurs in the image forming apparatus 110, the fault diagnosis result notifying server 104 receives a notification from the image forming apparatus 110. The CPU 221 of the fault diagnosis result notifying server 104 uses the machine learning model created in the flow of FIG. 7A to cause the estimation execution unit 607 to estimate the content of the corrective action. The CPU 221 of the fault diagnosis result notifying server 104 transmits the estimation result to the information input terminal 103. The information input terminal 103 displays a portal site 501, allowing the user to check the portal site 501. The user can input feedback information via the portal site 501 displayed on the information input terminal 103. For example, the portal site 501 accepts input of feedback information 512 for a bounding box 516 and feedback information 514 for a bounding box 517. The CPU 221 of the fault diagnosis result notifying server 104 stores the accepted feedback information 512 and 514 in the feedback storage unit 603.

[0073] FIG. 7B is a flowchart showing details of the learning data collection by the fault diagnosis result notifying server 104 in this embodiment in S701 of FIG. 7A. In S751, the CPU 221 of the fault diagnosis result notification server 104 acquires image anomaly information from the image anomaly information collection server 102 and stores it in the image anomaly information storage unit 701. An example of image anomaly information collected by the image anomaly information collection server 102 is shown in Table 1. The product name is the product type of the image forming device 110. The machine number is a unique ID that identifies each image forming device 110. The scanned image transmission date and time is the date and time when the user transmitted the scanned image. The scanned image ID is a unique ID assigned to each transmitted scanned image. The image anomaly ID is a unique character string code for recognizing image anomalies contained in the transmitted scanned image.

[0074] For example, the first line of Table 1 means that "The user of the device with product name PRO1001 and serial number DEV0001 sent the scanned image ID SCA0001 at 10:00 on 2022 / 02 / 01. The image abnormality ID is IMA1001." [Table 1]

[0075] Table 2 also shows an example of bounding box information for image anomalies. The scan image ID is a unique ID assigned to each scanned image sent. The image anomaly ID is a unique string code used to identify image anomalies contained in the scanned image sent. The X and Y coordinates are the coordinates (pixel values) of the top left vertex of the bounding box, and the height and width are the height and width (pixel values) of the bounding box. In the example in Table 2, the image data with scan image ID "SCA0001" contains two image anomalies and has two pieces of bounding box information corresponding to the two image anomalies. [Table 2]

[0076] Next, in S752, the CPU 221 of the fault diagnosis result notifying server 104 acquires feedback information of the machine number from the feedback storage unit 603. Table 3 shows an example of the feedback information. The product name is the product type of the image forming device 110. The machine number is a unique ID assigned to each image forming device 110. The treatment date and time is the date and time when the user performed the treatment. The scanned image ID is a unique ID assigned to each scanned image sent. The image abnormality ID is a unique string code for recognizing image abnormalities contained in the scanned image sent. The treatment content is the name of the treatment content actually performed by the user. The faulty part is the name of the part that the user actually repaired or replaced.

[0077] For example, the first line of Table 3 means that "Part A was replaced for image abnormality ID IMA1001 on scan image ID SCA0001 on machine number DEV0001 with product name PRO1001 at 14:02 on 2022 / 02 / 03." [Table 3]

[0078] In S753, the CPU 221 of the fault diagnosis result notification server 104 links the scanned image ID and image abnormality ID with the faulty part and the remedial action content from the feedback information in the feedback storage unit 404. Table 4 shows the fault remedial action data resulting from linking the scanned image ID and image abnormality ID with the faulty part and the remedial action content from the feedback information.

[0079] The product name in the fault correction data in Table 4 is the product type of the image forming device 110. The scanned image ID is a unique ID assigned to each scanned image sent. The image abnormality ID is a unique string code for recognizing image abnormalities in the scanned image sent. The X and Y coordinates are the coordinates (pixel values) of the top left vertex of the bounding box, and the height and width are the height and width (pixel values) of the bounding box. The correction content is the name of the correction content actually performed by the user. The faulty part is the name of the part actually repaired or replaced by the user. [Table 4]

[0080] Table 4 makes it possible to link image abnormalities that occurred in the image forming apparatus 110, the details of the measures taken to resolve the image abnormalities, and the faulty parts. This provides a mechanism for using fault correction data, including information on the measures taken by the user, as learning data for the type of equipment failure (type of image abnormality in the image forming apparatus). FIG. 8 is a flowchart showing an example of a step of creating a machine learning model in the fault diagnosis result notifying server 104 in this embodiment, which corresponds to S702 in FIG. 7(A).

[0081] In S801, the CPU 221 of the fault diagnosis result notification server 104 acquires a list of model series for which a learning model is to be created, based on the diagnosis target model series list 1110 stored in the diagnosis target model information management unit 609. FIG. 11(C) is an example of the diagnosis target model series list 1110. The model series is the model number of the image forming apparatus 110. The diagnosis target FLG indicates whether or not the model series is to be diagnosed. If the diagnosis target FLG is "1", it is a model to be diagnosed, and if it is "0", it is not a model to be diagnosed. In S802, the CPU 221 of the fault diagnosis result notification server 104 acquires the device configuration parts list 1101 for the model for which a learning model is to be created, which is stored in the device configuration information management unit 610.

[0082] FIG. 11A is an example of a device component parts list 1101. The component names are the device component parts that make up the image forming device 110. The model series is the model number of the image forming device 110. For example, parts_A001 listed for component A of model series A indicates the type of component A. In other words, component A is made up of components of the types parts_A001, component F is made up of parts_F001, and component Q is made up of parts_Q002. Here, components A, F, Q, and Y correspond to components that make up the image forming device 110, such as the photosensitive drum 14, intermediate transfer belt 26, and fixing device 4 described in FIG. 9A.

[0083] 11(B) shows the common parts determination results for model series A in device configuration parts list 1101 with other model series' components as common parts 1112. For example, part A in device configuration parts list 1101 means that "model series A is composed of parts of the type parts_A001, model series B is composed of parts_A002, and model series C is composed of parts_A003." In other words, part A is not a common part among model series A, B, and C. Part F means that "model series A is composed of parts of the type parts_F001, model series B is composed of parts_F001, model series C is composed of parts_F002, model series D is composed of parts_F002, and model series E is composed of parts_F002." In other words, part F is a common part, parts_F001, among model series A, B, and E, but is not a common part among model series C and D. On the other hand, in model series C and D, part F is a common part, parts_F002.

[0084] In steps S803 to S805, the CPU 221 of the failure diagnosis result notifying server 104 performs processing for each model and each device component based on the information acquired in steps S801 and S802. In S803, the CPU 221 of the failure diagnosis result notifying server 104 determines, based on the device component list 1101, whether or not each device component is common to each model series.

[0085] For example, if it is determined in S803 that the device components of the first model and the second model are common parts (Yes), the process proceeds to S804. Then, in S804, the CPU 221 of the fault diagnosis result notifying server 104 acquires failure information caused by the model component parts of the first model and failure information caused by the common parts of the second model, which are stored in the device malfunction information storage unit 601. The acquired failure information caused by the model component parts of the first model and failure information caused by the common parts of the second model are then used as learning data for the first model. In this way, for all other models with common parts, failure information caused by the common parts is acquired as learning data for the learning model of the device component parts of the first model. As the learning data, the failure information stored in the device malfunction information storage unit 601, replacement part information stored in the part replacement information storage unit 602, and fault correction data derived from Table 4 are acquired. Either the replacement part information or the fault correction data derived from Table 4 may be used.

[0086] Also, for example, if it is determined in S803 that the device components of the first model and the second model are not common parts (No), the process proceeds to S805. Then, in S805, the CPU 221 of the fault diagnosis result notifying server 104 acquires only the failure information caused by the device components collected from the first model stored in the device malfunction information storage unit 601 as learning data for training the learning model of the first model. The acquired learning data includes the failure information stored in the device malfunction information storage unit 601, the replacement part information stored in the part replacement information storage unit 602, and the failure action data derived from Table 4. Either the replacement part information or the failure action data derived from Table 4 may be used. Failure information caused by the device components that are not common to the second model is not used as learning data for the learning model of the first model. Regarding other models, failure information of device components of models that do not share common parts with the first model is not used as learning data for the learning model of the first model.

[0087] In S806 to S808, the CPU 221 of the fault diagnosis result notification server 104 uses the learning data generated in S803 to S805 to train a learning model for each fault type for each model series. In S806, it is determined whether learning of the learning models for all model series and all fault types has been completed. If learning of the learning models for all model series and all fault types has not been completed (No), proceed to S807. If learning of the learning models for all model series and all fault types has been completed (Yes), end the flow. Learning of the learning models in S807 and S808 is repeated until learning of the learning models for all model series and all fault types has been completed.

[0088] In this embodiment, a learning model is trained for each model series and each fault type. However, multiple fault types may be trained together for each model series. A learning model may also be created for each model series and each faulty component. In this embodiment, learning was performed by acquiring fault information caused by common components collected from a first model and fault information caused by common components collected from another (second) model as learning data for the learning model of the first model. The flow of FIG. 8 can also be modified so that, if a component is determined to be a common component between the first model and the second model, the learning data collected from the first model is used as learning data for the learning models of the first and second models. In this case, if a component of the first model and a component of the second model are determined to be non-common components, the failure information of the first model is used only as learning data for the learning model of the first model, and not as learning data for the learning model of the second model.

[0089] In S807, the CPU 221 of the fault diagnosis result notification server 104 uses an object detection algorithm to train a learning model for each fault type (image anomaly type) for each model series generated in S804 and S805. The object detection algorithm can use a well-known algorithm called DETEction TRansformer (DETR). For example, a convolutional neural network is used to estimate a bounding box, which is a frame that contains an object. Then, the confidence that the bounding box contains the object and the probability of each object type being included in the bounding box are predicted. During the learning process, the learning results are evaluated using cross-validation, which randomly divides the learning data into analysis data and validation data. This allows the creation of a learning model that estimates the diagnostic image 411 shown in Figure 4(B).

[0090] Here, there are various types of machine learning algorithms. Various object detection models can be used to detect rectangular areas that indicate an image of an object of interest (e.g., an image abnormality). Examples include YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector), and R-CNN (Region Based Convolutional Neural Networks). Hyperparameters of machine learning algorithms also differ depending on the machine learning algorithm. In this embodiment, the machine learning algorithm, the method for evaluating the learning results, and the method for optimizing the hyperparameters of the machine learning algorithm can be changed as appropriate.

[0091] Taking an image forming device as an example, machine learning involves learning to associate the type of abnormality in the diagnostic image (type of equipment failure), the part that caused the abnormality, and the action taken to resolve the abnormality. In the above description, the failure action data from Table 4 is used as learning data. However, in the initial state where the feedback information of Table 3 has not yet been accumulated, the user must create failure action data that associates the type of abnormality in the diagnostic image (type of equipment failure), the part that caused the abnormality, and the action taken to resolve the abnormality. The created failure action data is then used as learning data.

[0092] The creation of troubleshooting data by the user will now be described. In a scanned image containing an anomaly, the user identifies the anomaly type (type of equipment failure) and the anomaly location, and associates the faulty part and the remedial action for that anomaly type. If there is only one anomaly location in the scanned image, the faulty part and the remedial action for that anomaly type, or replacement part information, can be automatically associated.

[0093] When a scanned image contains multiple abnormalities, the types of abnormalities may differ. Different types of abnormalities may result in different components causing the abnormalities, and multiple faulty or replacement components may correspond to a single scanned image. In such cases, the user creates fault correction data by associating the faulty components and their corrective actions, or replacement component information, with the abnormal image areas of the scanned image while checking the scanned image. The created fault correction data is then used as learning data for creating a learning model.

[0094] In S808, the CPU 221 of the fault diagnosis result notification server 104 saves the trained model trained in S807 in a file and registers it in the machine learning model management unit 606. The file in which the trained model is saved includes the type of learning algorithm and the values ​​of the hyperparameters of the learning algorithm.

[0095] As described above, when it is determined that a component matching the faulty component information included in the learning data collected from the first model is a common component with a component of the second model, the learning data collected from the first model and the second model are used as the learning data of the first model. This makes it possible to provide a mechanism for selecting the learning data collected based on the device configuration information and efficiently using the collected learning data. [Example]

[0096] In Example 1, a trained model was generated using failure information of other models that share components common to the first model as training data for the first model. The trained model generated in Example 1 is referred to as trained model LM1.

[0097] However, when comparing the diagnostic accuracy of the trained model LM1 with that of the trained model LM2, which was generated using training data from only the first image forming apparatus, the accuracy of the trained model LM1 is not necessarily better. In Example 1, the trained model LM1 is a trained model generated with an emphasis on collecting training data from as many models as possible. Therefore, even if common parts are used, failure information from models with different structures is also used as training data, so the diagnostic accuracy of the trained model LM1 may be lower than that of the trained model LM2, which was trained using failure data from only one model.

[0098] In this embodiment, attention is focused on the diagnostic accuracy of the learned model LM1 and the learned model LM2. In this embodiment, the trained model LM1 and the trained model LM2 in the first embodiment are compared, and the trained model with the higher diagnostic accuracy is selected and adopted.

[0099] In this embodiment, the machine learning device configuration diagram (FIG. 1), hardware configuration diagram (FIG. 2), input terminal screen example diagram (FIG. 5), software configuration diagram (FIG. 6), and configuration of the image forming device 110 (FIGS. 9A and 9B) are the same as those in the first embodiment, and therefore descriptions thereof will be omitted. In addition, the flowchart for identifying correspondence candidates (FIG. 3), learning data collection processing flow (FIG. 7), and equipment component list (FIG. 11) are also the same as those in the first embodiment, and therefore descriptions thereof will be omitted.

[0100] FIG. 12 is a flowchart showing details of the learning data collection process S702 of the fault diagnosis result notifying server 104 in the second embodiment. Since S801 to S807 overlap with the contents shown in FIG. 7(B), the explanation will be omitted.

[0101] In S1201, the CPU 221 of the fault diagnosis result notification server 104 generates a trained model LM2 for the first model using failure information for only the first model as training data, without taking common parts into consideration. In the trained model LM2, failure information for other models that have common parts with the first model is not used as training data. Figure 13 shows the acquisition target 1302 of training data acquired to train the trained model LM2 for model series A in the equipment configuration parts list 1101. The trained model LM1 is a trained model generated for model series A by learning using training data acquired for target 1112 in Figure 11(B).

[0102] In S1202, the CPU 221 of the fault diagnosis result notification server 104 compares the diagnostic accuracy of the trained model LM1 generated in S807 with that of the trained model LM2 generated in S1201. In the flow of Fig. 12, the learning results are evaluated by cross-validation, which randomly divides the training data into analysis data and validation data. Trained models LM1 and LM2 are generated using the analysis data, and in S1202, the trained models LM1 and LM2 are verified using the validation data to determine whether a correct answer is obtained.

[0103] In S1203, the CPU 221 of the fault diagnosis result notification server 104 saves the trained model with high diagnostic accuracy verified in S1202 in a file and registers it in the machine learning model management unit 606. In S1203, both trained models LM1 and LM2 may be registered, allowing the user to select which model to use, LM1 or LM2.

[0104] As explained above, by comparing the diagnostic accuracy of the trained model LM1 and the trained model through verification and adopting the one with the better verification results, it is possible to perform fault diagnosis with higher accuracy. In addition, both trained models LM1 and LM2 may be used to infer the content of the measures to be taken, and the confidence level may be determined according to the accuracy of the verification results.

[0105] Alternatively, a threshold may be set for the number of training data to be collected, and if the number of training data is less than the threshold, the trained model LM1 may be used to estimate the treatment, and if the number of training data is equal to or greater than the threshold, the trained model LM2 may be used to estimate the treatment. By doing this, while there is little failure information, a lot of training data is collected from models that use common parts, and as failure information accumulates, training data is obtained only from the failure information of the relevant model, allowing for the generation of a highly accurate trained model.

[0106] <Other embodiments> The present invention can also be realized by supplying a program that realizes one or more functions of the above-described embodiments to a system or device via a network or a storage medium, and having one or more processors in the computer of the system or device read and execute the program. It can also be realized by a circuit (e.g., ASIC) that realizes one or more functions.

[0107] The invention is not limited to the above-described embodiments, and various changes and modifications can be made without departing from the spirit and scope of the invention. Accordingly, the following claims are appended to apprise the public of the scope of the invention.

[0108] The disclosure of this specification includes the following machine learning device, machine learning method, and program. (Item 1) a storage means for storing, as failure treatment data, the type of failure for each type of device, the component part of the device that caused the failure, and the treatment content for the failure; a learning means for learning a learning model for estimating a content of a remedy from a type of failure for each type of device, using the failure remedy data for each type of device as learning data; Equipped with the learning means uses, as learning data, failure treatment data of a first type of equipment and failure treatment data of a different type of equipment from the first type of equipment, which has a common component with a faulty part that caused the failure of the equipment included in the failure treatment data of the first type of equipment, and the learning means learns a learning model for estimating treatment content from the type of failure for the first type of equipment. Machine learning device. (Item 2) an estimation means for estimating a countermeasure for a device failure using the learning model; a receiving means for receiving a content of a remedy performed by a user from the estimated candidates of the content of remedy, and storing the content of the remedy performed by the user in a storage means as failure remedy data; Furthermore, Item 1. The machine learning device according to item 1. (Item 3) Further, a means for transmitting the estimated treatment content candidates to an information input terminal is provided, the receiving means receives the content of the action taken by the user from the information input terminal; Item 2. The machine learning device according to item 2. (Item 4) the device is an image forming device, The failure is an image abnormality included in the image data that has been formed. Item 3. The machine learning device according to any one of items 1 to 3. (Item 5) the learning means uses, as learning data, failure action data for a first type of equipment and failure action data for a common component of a type of equipment different from the first type of equipment that has a common component with a failed part that caused the equipment failure included in the failure action data for the first type of equipment, to learn a first learning model for estimating a content of action from the type of failure for the first type of equipment; and uses, as learning data, the failure action data for the first type of equipment and failure action data for the common component of a type of equipment different from the first type of equipment, to learn a second learning model for estimating a content of action from the type of failure for the first type of equipment, without using the failure action data for the common component of a type of equipment different from the first type of equipment; comparing the first learning model and the second learning model and selecting a model with high diagnostic accuracy; 5. The machine learning device according to any one of items 1 to 4. (Item 6) A machine learning method for a machine learning device including a storage means for storing, as failure treatment data, a type of failure for each type of device, a component part of the device that caused the failure, and a treatment content for the failure, the method comprising: a learning means for learning a learning model for estimating a content of a remedy from a type of failure for each type of device using the failure remedy data for each type of device as learning data; the learning means uses, as learning data, failure treatment data of a first type of equipment and failure treatment data of a different type of equipment from the first type of equipment, which has a common component with a faulty part that caused the failure of the equipment included in the failure treatment data of the first type of equipment, and the learning means learns a learning model for estimating treatment content from the type of failure for the first type of equipment. Machine learning methods. (Item 7) a computer of a machine learning device including a storage means for storing the type of failure for each type of equipment, the component part of the equipment that caused the failure, and the content of the treatment for the failure as failure treatment data; the fault correction data for each type of device is used as learning data, and the learning means functions as a learning model for estimating a corrective action from the type of fault for each type of device; the learning means uses, as learning data, failure treatment data of a first type of equipment and failure treatment data of a different type of equipment from the first type of equipment, which has a common component with a faulty part that caused the failure of the equipment included in the failure treatment data of the first type of equipment, and the learning means learns a learning model for estimating treatment content from the type of failure for the first type of equipment. program. [Explanation of symbols]

[0109] 102: Image abnormality information collection server, 103: Information input terminal, 104: Image diagnosis result notification server, 105: Replacement part information collection server, 106: Operation information collection server, 110: Image forming apparatus

Claims

1. a storage means for storing, as failure treatment data, the type of failure for each type of device, the component part of the device that caused the failure, and the treatment content for the failure; a learning means for learning a learning model for estimating a content of a remedy from a type of failure for each type of device, using the failure remedy data for each type of device as learning data; Equipped with the learning means uses, as learning data, failure treatment data for a first type of equipment and failure treatment data for a common component of a different type of equipment from the first type of equipment, which has a common component with a faulty part that caused the failure of the equipment included in the failure treatment data for the first type of equipment, and learns a learning model for estimating treatment content from the type of failure for the first type of equipment; Machine learning device.

2. an estimation means for estimating a countermeasure for a device failure using the learning model; a receiving means for receiving a content of a remedy performed by a user from the estimated candidates of the content of remedy, and storing the content of the remedy performed by the user in a storage means as failure remedy data; Furthermore, The machine learning device according to claim 1 .

3. Further, a means for transmitting the estimated treatment content candidates to an information input terminal is provided, the receiving means receives the content of the action taken by the user from the information input terminal; The machine learning device according to claim 2 .

4. the device is an image forming device, The failure is an image abnormality included in the image data that has been formed. The machine learning device according to claim 1 .

5. The learning means uses, as learning data, failure action data for a first type of equipment and failure action data for a common component of a type of equipment different from the first type of equipment that has a common component with a failed part that caused the equipment failure included in the failure action data for the first type of equipment, and learns a first learning model for estimating a content of action from the type of failure for the first type of equipment, and uses, as learning data, the failure action data for the first type of equipment and learns, without using the failure action data for the common component of a type of equipment different from the first type of equipment, comparing the first learning model and the second learning model and selecting a model with high diagnostic accuracy; The machine learning device according to claim 1 .

6. A machine learning method for a machine learning device including a storage means for storing, as failure treatment data, a type of failure for each type of device, a component part of the device that caused the failure, and a treatment content for the failure, the method comprising: a learning means for learning a learning model for estimating a content of a remedy from a type of failure for each type of device using the failure remedy data for each type of device as learning data; the learning means uses, as learning data, failure treatment data for a first type of equipment and failure treatment data for a common component of a different type of equipment from the first type of equipment, which has a common component with a faulty part that caused the failure of the equipment included in the failure treatment data for the first type of equipment, and learns a learning model for estimating treatment content from the type of failure for the first type of equipment; Machine learning methods.

7. a computer of a machine learning device including a storage means for storing the type of failure for each type of equipment, the component part of the equipment that caused the failure, and the content of the treatment for the failure as failure treatment data; the fault correction data for each type of device is used as learning data, and the learning means functions as a learning model for estimating a corrective action from the type of fault for each type of device; the learning means uses, as learning data, failure treatment data for a first type of equipment and failure treatment data for a common component of a different type of equipment from the first type of equipment, which has a common component with a faulty part that caused the failure of the equipment included in the failure treatment data for the first type of equipment, and learns a learning model for estimating treatment content from the type of failure for the first type of equipment; program.

Images