Method for generating learning model, computer program, learning model, control device, and control method

A learning model using neural networks automates the determination of control parameters for powder processing apparatuses, addressing the variability in powder properties and ensuring consistent particle size production.

JP7584409B2Active Publication Date: 2024-11-15HOSOKAWA MICRON CORP
View PDF 8 Cites 0 Cited by

Patent Information

Application Number
JP2021524544
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2019-06-04
Publication Date
2024-11-15
Estimated Expiration
2039-06-04

AI Technical Summary

Technical Problem

The operating conditions of powder processing apparatuses are often set based on the experience and intuition of field engineers, which can lead to inconsistencies due to the varied properties of powders depending on their type, size, and shape.

Method used

A learning model is generated using measurement data from powder processing apparatuses to calculate control parameters based on desired particle sizes, incorporating a neural network to determine optimal operating conditions without relying on human intuition.

Benefits of technology

This approach allows for the determination of control parameters that consistently produce desired particle sizes by automating the setting of operating conditions, reducing reliance on human expertise.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007584409000001
    Figure 0007584409000001
  • Figure 0007584409000002
    Figure 0007584409000002
  • Figure 0007584409000003
    Figure 0007584409000003
Patent Text Reader

Abstract

Provided is a learning model generation method, a computer program, a learning model, a control device, and a control method. Using a computer, a powder processing apparatus equipped with at least a function of pulverizing powder raw materials acquires measurement data that indicates the operation state of the powder processing apparatus and particle diameter data of powder obtained from the powder processing apparatus, and generates a learning model that outputs a computation result for a control parameter for controlling the operation state of the powder processing apparatus in accordance with input of the particle diameter desired by the user by using the acquired measurement data and particle diameter data as training data.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a method for generating a learning model, a computer program, a learning model, a control device, and a control method. [Background technology]

[0002] The powder processing process is composed of a combination of various processes such as storage, supply, transportation, pulverization, classification, and mixing (see, for example, Patent Document 1). In order to stably obtain products or intermediates with the quality desired by users, it is necessary to appropriately set the operating conditions of the powder processing equipment. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2008-194592 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, the powders handled in the powder processing process exhibit a wide variety of properties depending on the type, size, shape, etc. For this reason, there is a problem in that the operating conditions set in the powder processing device are largely dependent on the experience and intuition of the on-site engineer.

[0005] The present invention aims to provide a method for generating a learning model, a computer program, a learning model, a control device, and a control method that can determine control parameters for a powder processing device without relying on the experience or intuition of a field engineer. [Means for solving the problem]

[0006] A method for generating a learning model according to one embodiment of the present invention uses a computer to acquire measurement data indicating the operating state of a powder processing device having at least a function of grinding powder raw materials, and powder particle size data obtained from the powder processing device, and generates a learning model that uses the acquired measurement data and particle size data as training data to output calculation results for control parameters that control the operating state of the powder processing device in response to input of a particle size desired by a user.

[0007] A computer program according to one embodiment of the present invention is a computer program for causing a computer to execute a process of acquiring measurement data indicating the operating state of a powder processing device having at least a function of grinding powder raw materials, and powder particle diameter data obtained from the powder processing device, and using the acquired measurement data and particle diameter data as training data to generate a learning model that outputs calculation results for control parameters that control the operating state of the powder processing device in response to input of a particle diameter desired by a user.

[0008] A learning model according to one embodiment of the present invention includes an input layer in which a particle diameter desired by a user is input, an output layer in which a calculation result for a control parameter that controls the operating state of a powder processing apparatus is output, and an intermediate layer in which measurement data indicating the operating state of the powder processing apparatus and powder particle diameter data obtained from the powder processing apparatus are used as training data to learn the relationship between the particle diameter and the control parameters. When a particle diameter desired by a user is input to the input layer, the intermediate layer performs calculations and causes a computer to function to output the calculation result for the control parameter that controls the operating state of the powder processing apparatus.

[0009] A control device according to one embodiment of the present invention includes a receiving unit that receives input of a particle diameter desired by a user, a learning model that uses powder particle diameter data obtained from a powder processing device and measurement data indicating the operating state of the powder processing device as training data to learn the relationship between the particle diameter of the powder and control parameters that control the operating state of the powder processing device, a calculation processing unit that inputs the particle diameter data received by the receiving unit to the learning model and performs calculations using the learning model, and a control unit that controls the operation of an apparatus including the powder processing device based on the calculation results using the learning model.

[0010] A control method according to one aspect of the present invention receives input of a particle diameter desired by a user, inputs the received particle diameter data into a learning model that has learned the relationship between the particle diameter of the powder and control parameters that control the operating state of the powder processing device using powder particle diameter data obtained from a powder processing device and measurement data indicating the operating state of the powder processing device as training data, performs calculations using the learning model, and controls the operation of an apparatus including the powder processing device based on the calculation results obtained from the learning model. Effect of the Invention

[0011] According to the present application, control parameters for a powder processing apparatus can be determined without relying on the experience or intuition of a field engineer. [Brief description of the drawings]

[0012] [Figure 1] 1 is a schematic diagram showing an overall configuration of a powder processing system according to a first embodiment. [Diagram 2] FIG. 1 is a schematic cross-sectional view showing a configuration of a powder processing apparatus according to a first embodiment. [Diagram 3] 2 is a block diagram showing an internal configuration of a control device according to the first embodiment. FIG. [Figure 4] FIG. 2 is a block diagram showing the internal configuration of a terminal device. [Diagram 5] FIG. 2 is a schematic diagram showing an example of the configuration of a learning model in the first embodiment. [Figure 6]FIG. 4 is a conceptual diagram showing an example of data collected by a control device. [Figure 7] 11 is a flowchart illustrating a procedure for generating a learning model by a control device. [Figure 8] 4 is a flowchart illustrating a control procedure performed by the control device. [Figure 9] FIG. 11 is a schematic cross-sectional view showing a configuration of a powder processing apparatus according to a second embodiment. [Figure 10] FIG. 11 is a schematic diagram showing an example of the configuration of a learning model in the second embodiment. [Figure 11] FIG. 11 is a schematic cross-sectional view showing a configuration of a powder processing apparatus according to embodiment 3. [Figure 12] FIG. 11 is a schematic cross-sectional view showing a configuration of a powder processing apparatus according to embodiment 4. [Figure 13] FIG. 13 is a schematic diagram showing an example of the configuration of a learning model in embodiment 4. [Figure 14] FIG. 11 is a schematic cross-sectional view showing a configuration of a powder processing apparatus according to embodiment 5. [Figure 15] FIG. 13 is a schematic diagram showing an example of the configuration of a learning model in embodiment 5. [Figure 16] FIG. 23 is a schematic diagram showing an example of the configuration of a learning model in the sixth embodiment. [Figure 17] FIG. 23 is a schematic diagram showing an example of the configuration of a learning model in embodiment 7. [Figure 18] FIG. 23 is a conceptual diagram showing an example of data collection in the eighth embodiment. [Figure 19] 23 is a flowchart illustrating a procedure for generating a learning model in the eighth embodiment. [Figure 20] 4 is a flowchart illustrating a control procedure performed by the control device. [Figure 21] 11 is a flowchart illustrating a procedure for re-learning a learning model. [Figure 22] 11 is a flowchart illustrating a procedure for adjusting a control parameter. [Diagram 23] 4 is a flowchart illustrating a procedure of a process executed by a terminal device and a control device. [Figure 24]FIG. 13 is a schematic diagram showing an example of an input screen. [Diagram 25] FIG. 13 is a schematic diagram showing an example of a monitoring screen. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Hereinafter, the present invention will be described in detail with reference to the drawings showing embodiments thereof. (Embodiment 1) Fig. 1 is a schematic diagram showing the overall configuration of a powder processing system 1 according to embodiment 1. The powder processing system 1 according to embodiment 1 includes, for example, a raw material supplying machine 2, a hot air generator 3, a powder processing device 4A, a cyclone 5, a dust collector 6, a blower 7 (or a pump), and a control device 100 (see Fig. 3).

[0014] The raw material supplying machine 2 is a device for supplying powder raw material to the powder processing device 4A. The powder raw material supplied by the raw material supplying machine 2 to the powder processing device 4A is, for example, a fine powder toner raw material used for coloring paper in a copy machine or a laser printer. The powder raw material is not limited to a toner raw material, and may be a powder raw material for producing powder of inorganic materials, organic materials, or metal materials, such as powder paint, battery materials, magnetic materials, dyes, resins, wax, polymers, pharmaceuticals, catalysts, metal powders, silica, solder, cement, and food.

[0015] The raw material supplying machine 2 is connected to the powder processing device 4A via a raw material supplying path TP1. A screw feeder 21 (see FIG. 2) for conveying the powder raw material is provided in the raw material supplying path TP1. The screw feeder 21 is preferable when the powder raw material is solid and the powder raw material is continuously fed at a constant speed. Instead of the screw feeder 21, a double damper, a rotary valve, or the like may be used. Also, hot air generated by the hot air generator 3 may be introduced into the raw material supplying path TP1, and the powder raw material may be supplied to the powder processing device 4A together with the hot air. Furthermore, the raw material supplying machine 2 may perform weight management using a weight sensor S1 (see FIG. 3) such as a load cell, and adjust the supply amount of the powder raw material so that the retention amount in the device is constant even when continuous processing is performed in the powder processing device 4A. Also, the raw material supplying machine 2 may measure the supply amount of the powder raw material per unit time (i.e., the supply speed) based on the output of a built-in timer (not shown) and the supply amount of the powder raw material measured by the weight sensor S1. The weight sensor S1 may be provided not only in the raw material supplying device 2 but also in the powder processing device 4A, the cyclone 5, and the dust collector 6.

[0016] The hot air generator 3 is a device for generating hot air to be introduced into the powder processing device 4A, and includes a heat source such as a heater, a blower, and a control device for controlling the temperature and volume of the hot air. The hot air generator 3 is not limited to the above configuration, and may have any known configuration. For example, a portion of the hot air introduced into the powder processing device 4A may be recovered and circulated between the hot air generator 3 and the powder processing device 4A.

[0017] The hot air generator 3 is connected to the powder processing device 4A via a gas introduction path TP2. The hot air generated by the hot air generator 3 is introduced into the powder processing device 4A via the gas introduction path TP2. The temperature of the hot air generated by the hot air generator 3 is appropriately set according to the powder to be processed in the powder processing device 4A. For example, hot air of about 200°C to 600°C may be generated to dry the powder processed in the powder processing device 4A.

[0018] The powder processing device 4A is a device for producing powder having a particle size smaller than a predetermined size by pulverizing the powder raw material supplied from the raw material supplying device 2 and classifying the obtained powder. The powder processing device 4A according to the present embodiment will be described as a device (for example, ACM Pulverizer (registered trademark) manufactured by Hosokawa Micron Corporation) that mainly performs a pulverization process for pulverizing the powder raw material and a classification process for classifying the pulverized powder.

[0019] The powder processed by the powder processing device 4A is transported to the cyclone 5 via the powder transport path TP3. In this embodiment, a particle size sensor S2 is installed midway along the powder transport path TP3 from the powder processing device 4A to the cyclone 5, and the particle size sensor S2 measures the particle size of the powder passing through the powder processing device 4A constantly or periodically (for example, every 5 seconds). The powder collected by the cyclone 5 is taken out to a product tank 8 and recovered as a product.

[0020] The particle size sensor S2 is a device that measures particle size distribution using, for example, a laser diffraction / scattering method, and outputs values ​​of D10, D50, and D90. Here, D10, D50, and D90 represent particle sizes corresponding to 10%, 50%, and 90% cumulative from the small diameter side of the cumulative volume distribution in the particle size distribution, respectively. The cumulative volume distribution is a distribution that represents the relationship between the particle size (μm) of the powder and the cumulative frequency (volume %) from the small diameter side. D50 is generally also called the average particle size (median size). Instead of D10, D50, and D90, a mode diameter that represents the particle size with the highest occurrence ratio in the frequency distribution of particle sizes may be used, or various arithmetic average values ​​(number average, length average, area average, volume average, etc.) may be used. In this embodiment, the particle size sensor S2 is installed in the powder transport path TP3, but it may also be installed in the raw material supply machine 2, the path from the cyclone 5 to the product tank 8, the path from the dust collector 6 to the dust collection tank 9, etc.

[0021] The dust collector 6 is connected to the cyclone 5 via a dust collection path TP4. The dust collector 6 is equipped with a bag filter for collecting fine particles and the like that have passed through the cyclone 5. The gas that has passed through the bag filter of the dust collector 6 flows to the blower 7 through the exhaust passage TP5 and is discharged from the exhaust port of the blower 7. Meanwhile, the fine particles and the like that have been collected by the bag filter of the dust collector 6 are taken out to the dust collection tank 9 and collected.

[0022] The blower 7 is connected to the dust collector 6 via an exhaust air passage TP5. By driving the blower 7, a gas flow from the powder processing device 4A to the cyclone 5 (i.e., a gas flow for taking out the powder from the powder processing device 4A) and a gas flow from the cyclone 5 to the dust collector 6 are formed. In this embodiment, a flow sensor S3 (see FIG. 3) is installed in the exhaust air passage TP5 from the dust collector 6 to the blower 7, and the discharge / suction flow rate when taking out the powder from the powder processing device 4A is measured constantly or periodically (for example, every 5 seconds). Instead of the configuration in which the flow sensor S3 is installed in the exhaust air passage TP5, the flow sensor S3 may be installed in the raw material supply passage TP1, the gas introduction passage TP2, the powder transport passage TP3, the dust collection route TP4, the exhaust port of the blower 7, etc.

[0023] The control device 100 is a device that controls the operation of devices including the powder processing device 4A, and is connected to these devices so as to be able to send and receive various data. Here, the device including the powder processing device 4A means that the device includes the powder processing device 4A and may further include at least one of the raw material supplying device 2, the hot air generator 3, the cyclone 5, the dust collector 6, and the blower 7. In the following description, the device including the powder processing device 4A is also referred to as the device to be controlled.

[0024] 2 is a schematic cross-sectional view showing the configuration of a powder processing apparatus 4A according to embodiment 1. The powder processing apparatus 4A includes a cylindrical casing 410 in which powder processing is performed. The casing 410 is provided with a raw material inlet 411, a gas inlet 412, a grinding rotor 413, a guide ring 414, a classifying rotor 415, a powder outlet 416, and the like.

[0025] The material of the casing 410 may be a known material that has been used for the casing of a conventional powder processing device. Specifically, metals such as iron-based steel materials such as SS400, S25C, S45C, and SPHC (Steel Plate Hot Commercial), stainless steel materials such as SUS304 and SUS316, iron casting materials such as FC20 and FC40, and stainless steel casting materials such as SCS13 and 14, or ceramics, glass, etc. may be used. Furthermore, aluminum, wood, and synthetic resins may be used if an abrasion-resistant material is attached to the inner wall surface.

[0026] In order to improve the durability of the device, the inner surface of the casing 410 may be subjected to abrasion-resistant treatments such as plating, e.g., hard chrome plating, abrasion-resistant spraying, e.g., tungsten carbide spraying, metal deposition under vacuum, and diamond-structured carbon deposition.

[0027] Furthermore, when powder processing of a low melting point resin component such as toner is performed inside casing 410, contamination of the product with adhering components results in poor quality. Therefore, in order to prevent disturbance of the airflow or blockage inside casing 410 due to adhesion or adhering of the processed powder, the inner surface of casing 410 may be subjected to buff polishing, electrolytic polishing, coating with PTFE (Polytetrafluoroethylene) or plating with nickel or the like.

[0028] The casing 410 is provided with a raw material inlet 411 for introducing the powdered raw material supplied from the raw material supply device 2 into the casing 410. This raw material inlet 411 is preferably provided at a position above the rotating disk 413A of the grinding rotor 413. The powdered raw material supplied from the raw material supply device 2 is transported by the screw feeder 21 in the raw material supply path TP1 and introduced into the casing 410 through the raw material inlet 411.

[0029] The casing 410 is provided with a gas inlet 412 for introducing hot air (gas) from the hot air generator 3 into the casing 410. The gas inlet 412 is connected to the hot air generator 3 via a gas inlet path TP2. The position of this gas inlet 412 is not particularly limited, but it is preferable that the gas inlet 412 is provided at a position below the pulverization rotor 413 so that gas is introduced into the casing 410 via the rotating pulverization rotor 413. In this embodiment, the gas is introduced from a direction intersecting the rotation direction of the pulverization rotor 413, but the gas may be introduced along the rotation direction of the pulverization rotor 413.

[0030] The gas introduced from the gas inlet 412 forms an airflow that circulates while swirling inside the casing 410, and travels from inside the casing 410 through the classifying rotor 415 to the cyclone 5 and the dust collector 6 through the powder outlet 416. The airflow inside the casing 410 may be formed by suction using the blower 7 connected via the cyclone 5 and the dust collector 6, or may be formed by blowing (pressurizing) from the gas inlet 412 side. The type of gas introduced into the casing 410 may be appropriately determined depending on the target product to be treated. For example, air may be used, or an inert gas such as nitrogen or argon may be used to prevent oxidation.

[0031] Depending on the powder to be processed, a temperature rise inside casing 410 may cause the powder to soften, resulting in the powder fusing together and causing variation in particle size or a decrease in yield. Therefore, a temperature sensor S4 (see FIG. 3) may be provided at one or more locations inside casing 410 to control the temperature inside casing 410. For example, when the powder to be processed is toner with a low melting point, the temperature of the gas introduced into casing 410 may be adjusted so that the exhaust temperature at powder outlet 416 is 35 to 55°C. Furthermore, the cyclone 5, the dust collector 6, the product tank 8, the dust collection tank 9, etc. may be provided with a temperature sensor S4.

[0032] In addition, in this embodiment, hot air is introduced into casing 410 from hot air generator 3, but a cold air generator (not shown) may be used to introduce cold air of about -20°C to 5°C into casing 410. In this case, it is preferable that the gas introduced into casing 410 is dehumidified gas to prevent condensation. In addition, when processing food or the like that loses flavor or is easily deteriorated by heat, cold air adjusted to 0 to 15°C may be used.

[0033] Furthermore, in order to adjust the internal temperature of the casing 410, a jacket portion may be provided around the casing 410. The jacket portion adjusts the internal temperature of the casing 410 by circulating and supplying a heating fluid or a cooling fluid from a separately provided tank.

[0034] The pulverizing rotor 413 is a rotor including a rotating disk 413A and a plurality of hammers 413B protruding upward from the upper peripheral edge of the rotating disk 413A, and is configured to rotate at a desired rotation speed by the power of a pulverizing motor 413M (see FIG. 3). Here, the pulverizing motor 413M is one of the driving units included in the powder processing system 1. The rotation speed of the pulverizing motor 413M is measured by a rotation speed sensor S5 (see FIG. 3) all the time or at regular intervals (for example, every 5 seconds). The hammers 413B of the pulverizing rotor 413 are arranged at equal intervals in the circumferential direction on the upper peripheral edge of the rotating disk 413A. The shape, dimensions, number, and material of the hammers 413B are appropriately designed according to the particle diameter, circularity, etc. of the required product powder. For example, although FIG. 2 shows a rod-shaped hammer 413B, it may be a rectangular parallelepiped hammer or a trapezoid hammer in a plan view. Moreover, a blade-like structure may be used in place of the hammer 413B.

[0035] The grinding rotor 413 rotates by the power of the grinding motor 413M, generating a swirling airflow inside the casing 410, and by the action of the hammer 413B, imparts mechanical energy such as impact, compression, grinding, shearing, etc. to the powdered raw material introduced into the casing 410, thereby grinding the powdered raw material.

[0036] The material of the grinding rotor 413 may be a known material that has been used for grinding rotors of powder processing equipment in the past. For example, SS400, S25C, S45C, SUS304, SUS316, SUS630, etc. may be used. For the hammer 413B, a tip of a cemented carbide alloy may be attached so that it can withstand impact force, or a composite of metal and ceramics such as ceramics or cermet, which has wear resistance and toughness, may be used.

[0037] Furthermore, in order to improve the durability of the device, the surface of the pulverization rotor 413 may be subjected to plating treatment such as hard chrome plating, wear-resistant spraying treatment such as tungsten carbide spraying, metal deposition under vacuum, wear-resistant treatment such as diamond-structured carbon deposition, SUS630 quenching hardening treatment, etc. In addition, the surface of the pulverization rotor 413 may be subjected to buff polishing, electrolytic polishing, coating with PTFE or the like, plating with nickel or the like.

[0038] In this embodiment, the configuration of the hammer 413B protruding upward from the peripheral portion of the upper surface of the rotating disk 413A has been described, but a hammer protruding downward from the peripheral portion of the lower surface of the rotating disk 413A may also be used. Although a hammer protruding downward in this manner does not directly pulverize the powder raw material in the casing 410, it can form a strong swirling air current in the casing 410 and indirectly pulverize the powder raw material by causing the powder raw material to collide with each other.

[0039] A grinding liner may be provided on the inner circumferential surface of the casing 410 at a position facing the hammer 413B. The grinding liner is a cylindrical member having a central axis aligned with the rotational axis direction of the grinding rotor 413, and triangular, wavy or wedge-shaped grooves may be provided on the inner circumferential surface of this cylindrical member.

[0040] The guide ring 414 is a cylindrical member for generating a swirling airflow in the casing 410 and guiding the powder to be processed in the casing 410 to the classification rotor 415. The guide ring 414 is arranged above the grinding rotor 413 and coaxially with the grinding rotor 413, and is fixed inside the casing 410. The fixing method of the guide ring 414 is not particularly limited, but it is necessary to fix it without rotating inside the casing 410 during the operation of the powder processing device 4A. This is to control the flow state of the powder to be processed to an appropriate state inside the casing 410. In the example of FIG. 2, the guide ring 414 whose inner diameter is continuously increased from the lower side to the upper side in the casing 410 is shown, but the guide ring may be one whose inner diameter is continuously decreased toward the upper side, or may be one whose inner diameter does not change in the vertical direction.

[0041] The classification rotor 415 is a rotor equipped with a plurality of radially arranged classification blades 415A, and is configured to rotate at a desired rotation speed by the power of the classification motor 415M (see FIG. 3). Here, the classification motor 415M is one of the driving parts equipped in the powder processing system 1. The rotation speed of the classification motor 415M is measured by a rotation speed sensor S6 (see FIG. 3) all the time or at regular intervals (for example, every 5 seconds). The classification rotor 415 is provided above the grinding rotor 413, and by centrifugal force due to high-speed rotation, it passes only powders that are smaller than a predetermined particle size among the powders processed in the casing 410, and leads only the passed powders to the powder outlet 416.

[0042] Here, the particle size of the powder passing through the classifying rotor 415 can be set by controlling the rotation speed, etc., of the classifying rotor 415. That is, by controlling the rotation speed of the classifying rotor 415, powder having a particle size smaller than a predetermined size can be taken out from inside the casing 410. On the other hand, powder that cannot pass through the classifying rotor 415 circulates inside the casing 410 and is repeatedly processed.

[0043] The material of the classification rotor 415 may be any known material that has been used for classification rotors of powder processing devices. For example, SS400, S25C, S45C, SUS304, SUS316, titanium, titanium alloy, aluminum alloy, etc. may be used. In addition, the surface of the classification rotor 415 may be subjected to wear-resistant treatment such as carburizing and quenching, tungsten carbide spraying material treatment, plating treatment such as hard chrome plating, special spraying material treatment for performing heat hardening treatment after spraying, metal deposition under vacuum, and diamond-structured carbon deposition in order to improve the durability of the device.

[0044] In order to prevent adhesion or adhesion of powder to the classifying rotor 415, the surface of the classifying rotor 415 may be subjected to buff polishing, electrolytic polishing, coating with PTFE or the like, or plating with nickel or the like.

[0045] The powder processing device 4A according to the present embodiment includes a pulverizing rotor 413 and a classifying rotor 415, but may be used as a classifier without using the pulverizing rotor 413. In addition, since hot air can be introduced from the hot air generator 3, the powder processing device 4A may be used as a dryer. In addition, the powder processing device 4A may be used as a particle design device by providing a coarse powder recovery port in the casing 410 and recovering the coarse powder. Furthermore, the powder processing device 4A may be used as a continuous mixer that supplies multiple types of raw materials to the casing 410 and mixes the multiple types of raw materials in the casing 410.

[0046] In addition to the various sensors described above, at least one of the devices including the powder processing device 4A, or at least one of the paths between the devices, may be provided with a dust concentration sensor that measures dust concentration, a measurement sensor such as an NIR sensor that measures the composition of the powder, a moisture sensor that measures the moisture in the powder, a pressure sensor that measures pressure, and a sound pressure / frequency sensor that measures sound pressure or frequency.

[0047] The configuration of a control system in the powder processing system 1 will be described below. 3 is a block diagram showing the internal configuration of the control device 100 according to embodiment 1. The control device 100 is configured by a general-purpose or dedicated computer, and includes a control unit 101, a storage unit 102, an input unit 103, an output unit 104, a communication unit 105, an operation unit 106, and a display unit 107.

[0048] The control unit 101 includes, for example, a central processing unit (CPU), a read only memory (ROM), and a random access memory (RAM). The ROM included in the control unit 101 stores a control program that controls the operation of each hardware unit included in the control device 100. The CPU in the control unit 101 executes the control program stored in the ROM and various computer programs stored in a storage unit 102 (described later) and controls the operation of each hardware unit, thereby realizing the function of the control device according to the present invention. The RAM included in the control unit 101 temporarily stores data used during the execution of calculations, etc.

[0049] The control unit 101 is configured to include a CPU, a ROM, and a RAM, but may be one or more arithmetic circuits or control circuits including a GPU (Graphics Processing Unit), an FPGA (Field Programmable Gate Array), a DSP (Digital Signal Processor), a quantum processor, a volatile or non-volatile memory, etc. The control unit 101 may also include functions such as a clock that outputs date and time information, a timer that measures the elapsed time from when a measurement start instruction is given until when a measurement end instruction is given, and a counter that counts numbers.

[0050] The storage unit 102 includes a storage device using a hard disk, a flash memory, etc. The storage unit 102 stores computer programs executed by the control unit 101, various data acquired from the outside, various data generated within the device, and the like.

[0051] The computer programs stored in the memory unit 102 include a control program PG1 for controlling the operation of the device to be controlled, a learning program PG2 for learning the relationship between the particle diameter of the powder processed by the powder processing device 4A and control parameters related to the powder processing device 4A, etc.

[0052] The computer programs including the control program PG1 and the learning program PG2 may be provided by a non-transitory recording medium M1 on which the computer programs are readably recorded. The recording medium M1 is, for example, a portable memory such as a CD-ROM, a USB memory, a Compact Flash (registered trademark), an SD (Secure Digital) card, or a micro SD card. The control unit 101 reads the various programs from the recording medium M1 using a reading device (not shown) and stores the various programs thus read in the storage unit 102.

[0053] In this embodiment, the control unit 101 executes the learning program PG2, whereby the control device 100 transitions to the learning phase. The learning phase is performed, for example, when the control device 100 is introduced. The learning phase may also be performed when a user's instruction is received via the operation unit 106, or at regular intervals.

[0054] In the learning phase, the control unit 101 acquires measurement data including the rotation speed of the pulverizing rotor 413 and the rotation speed of the classifying rotor 415 of the powder processing device 4A, and the discharge and suction flow rate when the powder is taken out from the casing 410 (processing chamber), and the particle size data of the powder obtained from the powder processing device 4A. These measurement data and particle size data may be collected in advance or may be collected after the transition to the learning phase. The control unit 101 uses the collected measurement data and particle size data as teacher data to learn the relationship between the particle size of the powder obtained from the powder processing device 4A and the control parameters related to the powder processing device 4A, thereby generating the learning model 200. Here, the control parameters related to the powder processing device 4A include control parameters for directly controlling the operation of the powder processing device 4A (for example, the rotation speed of the pulverizing rotor 413 and the rotation speed of the classifying rotor 415). Furthermore, the control parameters related to the powder processing device 4A may include control parameters for controlling the operation of the raw material supplying device 2, the hot air generator 3, the dust collector 6, and the blower 7 associated with the powder processing device 4A. In this embodiment, a configuration is described for learning the relationship between the particle diameter obtained from the powder processing device 4A and the control parameters that control the rotation speed of the grinding rotor 413 and the rotation speed of the classifying rotor 415 provided in the powder processing device 4A, and the discharge / suction flow rates when removing powder from the processing chamber of the powder processing device 4A.

[0055] The learning model 200 generated in the learning phase is stored in the storage unit 102. The learning model 200 is defined by its definition information. The definition information of the learning model 200 includes, for example, structural information of the learning model 200, and parameters such as weights and biases between nodes.

[0056] In this embodiment, the control device 100 transitions to an operation phase as a result of the control unit 101 executing the control program PG1. The operation phase is implemented after the learning model 200 is executed.

[0057] In the operation phase, the control device 100 controls the operation of the device including the powder processing device 4A. At this time, the control unit 101 may receive an input of a particle diameter (desired particle diameter) desired by the user. The control unit 101 inputs the received particle diameter data to the learning model 200 and executes a calculation using the learning model 200 to obtain a calculation result related to the control parameters that control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate when extracting the powder. The control unit 101 controls the operation of the device to be controlled based on the calculation result obtained from the learning model 200.

[0058] The input unit 103 includes a connection interface for connecting devices to be controlled. The devices connected to the input unit 103 include the raw material supplying machine 2, the hot air generator 3, the powder processing device 4A, the cyclone 5, the dust collector 6, and the blower 7. The connection interface provided in the input unit 103 may be a wired interface or a wireless interface. The input unit 103 receives data sent from the raw material supplying machine 2, the hot air generator 3, the powder processing device 4A, the cyclone 5, the dust collector 6, and the blower 7. The data input to the input unit 103 includes measurement data measured by the weight sensor S1, the particle size sensor S2, the flow rate sensor S3, the temperature sensor S4, the rotation speed sensor S5 for the grinding rotor 413, and the rotation speed sensor S6 for the classifying rotor 415.

[0059] In this embodiment, the device to be controlled is connected to the input unit 103, but at least some of the sensors S1 to S6 may be directly connected to the input unit 103. In this case, the measurement data output from the sensors S1 to S6 is input directly to the input unit 103 without passing through the device to be controlled. In addition, if the sensors S1 to S6 have a communication interface, the control device 100 may acquire the measurement data through the communication unit 105 described later.

[0060] The output unit 104 includes a connection interface for connecting the device to be controlled. The devices connected to the output unit 104 include the raw material supplying machine 2, the hot air generator 3, the powder processing device 4A, the cyclone 5, the dust collector 6, and the blower 7. The connection interface of the output unit 104 may be a wired interface or a wireless interface. The control unit 101 controls the operation of the device to be controlled by outputting a control command through the output unit 104. For example, when controlling the rotation speed of the grinding rotor 413, the control unit 101 generates a control command for the grinding motor 413M, which is the driving unit of the grinding rotor 413, and outputs the control command to the powder processing device 4A through the output unit 104 to control the rotation speed of the grinding rotor 413. The same is true when controlling the rotation speed of the classifying rotor 415, and the control unit 101 generates a control command for the classifying motor 415M, which is the driving unit of the classifying rotor 415, and outputs the control command to the powder processing device 4A through the output unit 104 to control the rotation speed of the classifying rotor 415. In addition, when controlling the discharge / suction flow rate from the casing 410 provided in the powder processing device 4A, the control unit 101 generates a control command for the blower 7 and outputs it to the blower 7 via the output unit 104, thereby controlling the discharge / suction flow rate from the casing 410.

[0061] The communication unit 105 includes a communication interface for transmitting and receiving various communication data. The communication interface of the communication unit 105 is, for example, a communication interface conforming to a communication standard of a LAN (Local Area Network) used in WiFi (registered trademark) or Ethernet (registered trademark). Alternatively, the communication interface may be a communication interface conforming to a communication standard such as Bluetooth (registered trademark), ZigBee (registered trademark), 3G, 4G, 5G, or LTE (Long Term Evolution).

[0062] The communication unit 105 communicates with, for example, a terminal device 500 used by a user of the powder processing system 1. The communication unit 105 may receive operation data or setting data transmitted from the terminal device 500 in order to accept remote operation of the powder processing system 1. When the control unit 101 receives operation data or setting data transmitted from the terminal device 500 through the communication unit 105, the control unit 101 executes processing according to the received operation data or setting data. For example, when data of a particle diameter desired by a user is received, the control unit 101 may execute a calculation using the learning model 200 and execute a process of acquiring a control parameter related to the powder processing device 4A. In addition, the control unit 101 may generate screen data of a user interface screen to be displayed on the terminal device 500 and transmit the generated screen data to the terminal device 500 through the communication unit 105.

[0063] The operation unit 106 includes an input interface such as a keyboard and a mouse, and receives various operations and settings. The control unit 101 performs appropriate processing based on the various operations and settings received through the operation unit 106, and stores setting information in the storage unit 102 as necessary. Note that, although the control device 100 is configured to include the operation unit 106 in this embodiment, the operation unit 106 is not essential, and the operation may be received through an externally connected computer (for example, the terminal device 500).

[0064] The display unit 107 includes a display panel such as a liquid crystal panel or an organic EL (Electro-Luminescence) panel, and displays information to be notified to the user. The display unit 107 may display, for example, measurement data of the various sensors S1 to S6 received through the communication unit 105, or may display information based on various operations and various settings received through the operation unit 106. The display unit 107 may also display the results of calculations performed by the learning model 200. Note that, although the control device 100 is configured to include the display unit 107 in this embodiment, the display unit 107 is not essential, and information to be notified to the user may be output to an external computer (for example, the terminal device 500) and the information may be displayed on the output destination computer.

[0065] In the present embodiment, the control device 100 has been described as a single computer, but it does not have to be a single computer, and may be configured with multiple computers or multiple virtual computers.

[0066] 4 is a block diagram showing the internal configuration of the terminal device 500. The terminal device 500 is a computer such as a personal computer, a smartphone, or a tablet terminal, and includes a control unit 501, a storage unit 502, a communication unit 503, an operation unit 504, and a display unit 505.

[0067] The control unit 501 includes, for example, a CPU, a ROM, and a RAM. The ROM included in the control unit 501 stores a control program for controlling the operation of each hardware unit included in the terminal device 500. The CPU in the control unit 501 executes the control program stored in the ROM and various computer programs stored in a storage unit 502 described later, and controls the operation of each hardware unit. The control unit 501 may also include functions such as a clock that outputs date and time information, a timer that measures the elapsed time from when a measurement start instruction is given until when a measurement end instruction is given, and a counter that counts numbers. The RAM included in the control unit 501 temporarily stores data used during the execution of calculations.

[0068] The storage unit 502 includes a storage device using a hard disk, a flash memory, etc. The storage unit 502 stores computer programs executed by the control unit 501, various data acquired from the outside, various data generated within the device, etc. The computer programs stored in the storage unit 502 may include an application program for accessing the control device 100 from the terminal device 500.

[0069] The communication unit 503 includes a communication interface for transmitting and receiving various data. The communication interface included in the communication unit 503 is, for example, a communication interface conforming to a LAN communication standard used in WiFi (registered trademark) or Ethernet (registered trademark). Alternatively, the communication interface may be a communication interface conforming to a communication standard such as Bluetooth (registered trademark), ZigBee (registered trademark), 3G, 4G, 5G, or LTE.

[0070] The communication unit 503 communicates with, for example, the control device 100 of the powder processing system 1. The data received by the terminal device 500 through the communication unit 503 includes screen data for displaying an interface screen of the control device 100 on the display unit 505, data indicating the setting state and control state of the device including the powder processing device 4A, and the like. Here, the data indicating the setting state includes a set value related to the rotation speed of the grinding rotor 413, a set value related to the rotation speed of the classifying rotor 415, and a set value related to the discharge and suction flow rate by the blower 7, and the like. The data indicating the control state includes a measured value related to the rotation speed of the grinding rotor 413, a measured value related to the rotation speed of the classifying rotor 415, a measured value related to the discharge and suction flow rate by the blower 7, and alarm information output from the device including the powder processing device 4A, and the like. The data received by the communication unit 503 is output to the control unit 501. The data transmitted by the terminal device 500 through the communication unit 503 includes operation data or setting data when remotely operating the powder processing system 1, and the like.

[0071] The operation unit 504 includes an input interface such as a keyboard and a mouse, and receives various operations and settings. The control unit 501 performs appropriate processing based on the various operations and settings received through the operation unit 504, and stores setting information in the storage unit 502 as necessary.

[0072] The display unit 505 includes a display panel such as a liquid crystal panel or an organic electroluminescence panel, and displays information to be notified to the user. For example, the display unit 505 displays an interface screen of the control device 100 based on screen data received by the communication unit 503. The display unit 505 may also display the setting state and control state of the device including the powder processing device 4A based on data indicating the setting state and control state of the device including the powder processing device 4A received by the communication unit 503.

[0073] The learning model 200 used in the control device 100 will be described below. Fig. 5 is a schematic diagram showing a configuration example of the learning model 200 in the first embodiment. The learning model 200 is, for example, a machine learning learning model including deep learning, and is configured by a neural network. The learning model 200 includes an input layer 201, intermediate layers 202A and 202B, and an output layer 203. In the example of Fig. 5, two intermediate layers 202A and 202B are described, but the number of intermediate layers is not limited to two, and may be three or more.

[0074] The input layer 201, the intermediate layers 202A and 202B, and the output layer 203 each have one or more nodes, and the nodes of each layer are connected in one direction with the nodes in the previous and next layers with desired weights and biases. The input layer 201 of the learning model 200 receives data in the same number as the number of nodes in the input layer 201. In this embodiment, the data input to the nodes of the input layer 201 is data on particle diameters (desired particle diameters) desired by the user. Here, an example of the data on particle diameters input to the nodes of the input layer 201 is a median diameter. It is not limited to the median diameter, but may be a mode diameter or various arithmetic mean values. Furthermore, the data on particle diameters given to the nodes of the input layer 201 is not limited to a single value such as a median diameter, a mode diameter, or various arithmetic mean values, but may be the values ​​of D10, D50, and D90, or may be ranges set for each of D10, D50, and D90 (i.e., the upper and lower limits of the range of particle diameters allowed by the user).

[0075] The particle size data input to the learning model 200 is output to the nodes of the first intermediate layer 202A through the nodes constituting the input layer 201. The data input to the first intermediate layer 202A is output to the nodes of the next intermediate layer 202B through the nodes constituting the intermediate layer 202A. At this time, the output is calculated using an activation function including weights and biases set between the nodes. In the same manner, calculations are performed using activation functions including weights and biases set between the nodes, and are transmitted to subsequent layers one after another until a calculation result is obtained by the output layer 203. Parameters such as weights and biases connecting the nodes are learned by a predetermined learning algorithm. For example, a deep learning learning algorithm is used as the learning algorithm for learning various parameters. In this embodiment, particle diameter data regarding particle diameter obtained from the powder processing device 4A (i.e., particle diameter data measured by the particle diameter sensor S2) and measurement data including the rotation speed of the grinding rotor 413 and the rotation speed of the classifying rotor 415 equipped in the powder processing device 4A, and the discharge / suction flow rates when removing powder from the casing 410 are used as teacher data, and various parameters including weights and biases between nodes can be learned by a predetermined learning algorithm.

[0076] The output layer 203 outputs the calculation results related to the control parameters that control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate from the casing 410. As the calculation results, for example, a probability indicating whether the combination of the above-mentioned multiple control parameters is good or bad may be output. Specifically, the output layer 203 may be configured with n nodes from the first node to the n-th node, and the first node may output a probability P1 that the rotation speed of the grinding rotor 413 is G1, the rotation speed of the classifying rotor 415 is C1, and the discharge / suction flow rate is V1, the second node may output a probability P2 that the rotation speed of the grinding rotor 413 is G2, the rotation speed of the classifying rotor 415 is C2, and the discharge / suction flow rate is V2, ..., and the n-th node may output a probability Pn that the rotation speed of the grinding rotor 413 is Gn, the rotation speed of the classifying rotor 415 is Cn, and the discharge / suction flow rate is Vn. The number of nodes constituting the output layer 203 and the calculation results assigned to each node are not limited to the above example, and can be designed as appropriate.

[0077] The control device 100 collects particle diameter data regarding particle diameter obtained from the powder processing device 4A, and measurement data including the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rates from the casing 410 of the powder processing device 4A, and uses these data as training data to generate the learning model 200 as described above.

[0078] FIG. 6 is a conceptual diagram showing an example of data collected by the control device 100. The control unit 101 of the control device 100 acquires measurement data from the particle size sensor S2, the rotation speed sensor S5 of the grinding rotor 413, the rotation speed sensor S6 of the classifying rotor 415, and the flow rate sensor S3 through the input unit 103, stores the acquired data together with a time stamp in the memory unit 102, and collects data to be used as teacher data. FIG. 6 shows an example of data collected by the control device 100. Among the data shown in FIG. 6, the first record indicates that the actual measured value of the rotation speed of the grinding rotor 413 is 4749.6 rpm, the actual measured value of the rotation speed of the classifying rotor 415 is 3195.1 rpm, and the actual measured values ​​of the discharge and suction flow rates are 2281.2 m / s. 3 / h, the actual measured values ​​of D10, D50, and D90 are 0.71 μm, 2.09 μm, and 9.06 μm, respectively. The same is true for the second and subsequent records, and the actual measured values ​​of the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge and suction flow rates from the casing 410 when the powder processing system 1 is operating are associated with the data (D10, D50, and D90) regarding the particle diameter of the powder obtained from the powder processing device 4A and stored in the storage unit 102. Note that, in the example of FIG. 6, an example is shown in which data is collected at 5-second intervals, but the time interval for collecting data is not limited to 5 seconds and may be set arbitrarily.

[0079] The control unit 101 uses the collected data as training data to learn the relationship between the particle size of the powder obtained from the powder processing device 4A and the control parameters related to the powder processing device 4A, and generates the learning model 200 as described above.

[0080] The operation of the control device 100 in the learning phase will now be described. 7 is a flowchart for explaining the procedure for generating the learning model 200 by the control device 100. The control unit 101 of the control device 100 collects measurement data including the rotation speed of the grinding rotor 413 and the rotation speed of the classifying rotor 415 of the powder processing device 4A, and the discharge and suction flow rates when the powder is taken out from the casing 410, and the particle size data of the powder obtained from the powder processing device 4A (step S101). The collected measurement data and particle size data are stored in the memory unit 102 together with a time stamp.

[0081] After collecting the measurement data and particle size data, the control unit 101 selects one set of teacher data from the data stored in the memory unit 102 (step S102). That is, the control unit 101 selects only one set of the rotation speeds of the grinding rotor 413 and the classifying rotor 415, and the actual measured values ​​of the discharge and suction flow rates from the casing 410, and the particle size values ​​(D10, D50, D90) of the powder obtained at that time.

[0082] Next, the control unit 101 inputs the particle diameter value included in the selected teacher data to the learning model 200 (step S103), and executes a calculation by the learning model 200 (step S104). That is, the control unit 101 inputs the particle diameter value to the node constituting the input layer 201 of the learning model 200, performs a calculation using weights and biases between the nodes in the intermediate layers 202A and 202B, and outputs the calculation result from the node in the output layer 203. Note that, in the initial stage before learning is started, initial values ​​are given to the definition information describing the learning model 200.

[0083] Next, the control unit 101 evaluates the calculation result obtained in step S104 (step S105) and determines whether learning is complete or not (step S106). Specifically, the control unit 101 can evaluate the calculation result using an error function (also called an objective function, loss function, or cost function) based on the calculation result obtained in step S104 and the teacher data. The control unit 101 determines that learning is complete when the error function becomes equal to or smaller than a threshold (or equal to or larger than a threshold) in the process of optimizing (minimizing or maximizing) the error function by a gradient descent method such as the steepest descent method. In order to avoid the problem of overlearning, a method such as cross-validation or early termination may be adopted to terminate learning at an appropriate timing.

[0084] When it is determined that the learning is not completed (S106: NO), the control unit 101 updates the weights and biases between the nodes of the learning model 200 (step S107), returns the process to step S102, and continues the learning using other teacher data. The control unit 101 can update the weights and biases between the nodes by using an error backpropagation method that sequentially updates the weights and biases between the nodes from the output layer 203 to the input layer 201 of the learning model 200.

[0085] When it is determined that the learning is completed (S106: YES), the control unit 101 stores the learning model 200 as a trained learning model in the storage unit 102 (step S108), and ends the process according to this flowchart.

[0086] As described above, in the learning phase, the control device 100 according to this embodiment collects particle size data on particle size obtained from the powder processing device 4A, and measurement data including the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge and suction flow rates when taking out powder from the casing 410. By using the collected data as teacher data, the control device 100 can generate a learning model 200 that outputs calculation results on control parameters that control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge and suction flow rates to the casing 410 in response to an input of a particle size desired by the user.

[0087] In this embodiment, the control device 100 is configured to generate the learning model 200, but an external server (not shown) that generates the learning model 200 may be provided, and the external server may generate the learning model 200. In this case, the control device 100 may acquire the learning model 200 from the external server by communication or the like, and store the acquired learning model 200 in the memory unit 102.

[0088] Next, a description will be given of the operation of the control device 100 in the operation phase. Note that in the operation phase, it is assumed that the learning model 200 has already been trained.

[0089] FIG. 8 is a flowchart explaining the control procedure by the control device 100. The control unit 101 of the control device 100 receives an input of a particle diameter (desired particle diameter) desired by the user through the operation unit 106 (step S121). Here, the control unit 101 may receive a value of D50 as the particle diameter desired by the user. The control unit 101 may also receive the values ​​of D10, D50, and D90, respectively, instead of the value of D50, and may receive upper and lower limit values ​​for each of D10, D50, and D90. That is, data on the particle diameter may be input according to the configuration of the input layer 201 included in the learning model 200.

[0090] Next, the control unit 101 inputs the received data of the particle diameter to the input layer 201 of the learning model 200, and executes the calculation by the learning model 200 (step S122). At this time, the control unit 101 provides the received data of the particle diameter to the nodes of the input layer 201. The data provided to the nodes of the input layer 201 is output to the nodes of the adjacent intermediate layer 202A. In the intermediate layer 202A, a calculation is performed using an activation function including weights and biases between the nodes, and the calculation result is output to the subsequent intermediate layer 202B. In the intermediate layer 202B, a calculation is further performed using an activation function including weights and biases between the nodes, and the calculation result is output to each node of the output layer 203. Each node of the output layer 203 outputs a calculation result related to the control parameters of the powder processing device 4A. Specifically, each node of the output layer 203 outputs a calculation result related to the control parameters that control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate from the casing 410.

[0091] Next, the control unit 101 obtains the calculation result from the learning model 200 (step S123) and determines the control parameters to be used for control (step S124). The output layer 203 of the learning model 200 outputs the calculation result regarding the control parameters to control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate from the casing 410. More specifically, the output layer 203 outputs the probability Pi (i=1 to n) that the rotation speed of the grinding rotor 413 is Gi, the rotation speed of the classifying rotor 415 is Ci, and the discharge / suction flow rate is Vi from each node. The control unit 101 determines the control parameters to be used for control by identifying the combination of the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate from the casing 410 that has the highest probability.

[0092] Next, the control unit 101 executes control based on the control parameters determined in step S124 (step S125). That is, the control unit 101 generates control commands for the pulverizing motor 413M and the classifying motor 415M so that the rotation speeds of the pulverizing rotor 413 and the classifying rotor 415 become the values ​​determined in step S124, and outputs the control commands to the powder processing apparatus 4A through the output unit 104. The control unit 101 also generates a control command for the blower 7 so that the discharge / suction flow rate from the casing 410 becomes the value determined in step S124, and outputs the control command to the blower 7 through the output unit 104.

[0093] As described above, the control device 100 according to this embodiment can determine the control parameters for controlling the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate from the casing 410 by receiving an input of a particle diameter desired by the user. That is, in this embodiment, the control parameters for the powder processing apparatus 4A can be determined without relying on the experience or intuition of a field engineer. The control device 100 executes control based on the determined control parameters so as to obtain powder having a desired particle diameter.

[0094] In this embodiment, a configuration has been described in which calculation results regarding control parameters are obtained using a machine learning learning model 200 configured by a neural network, but the learning model 200 is not limited to a model obtained using a specific method. For example, instead of a neural network based on deep learning, a learning model based on a perceptron, a convolutional neural network, a recurrent neural network, a residual network, a self-organizing map, or the like may be used. In addition, instead of the learning model using the above neural network, the learning model may be a model trained using other methods including regression analysis methods including linear regression, logistic regression, support vector machines, etc.; methods using search trees such as decision trees, regression trees, random forests, gradient boosting trees, etc.; Bayesian estimation methods including naive Bayes, etc.; time series prediction methods including AR (Auto Regressive), MA (Moving Average), ARIMA (Auto Regressive Integrated Moving Average), state space models, etc.; clustering methods including K-nearest neighbor methods, etc.; methods using ensemble learning including boosting, bagging, etc.; clustering methods including hierarchical clustering, non-hierarchical clustering, topic models, etc.; association analysis, emphasis filtering, etc. Furthermore, a learning model may be constructed using multivariate analysis including PLS (Partial Least Squares) regression, multiple regression analysis, principal component analysis, factor analysis, cluster analysis, and the like.

[0095] (Embodiment 2) In the second embodiment, an application example to a pin mill (for example, a fine impact mill manufactured by Hosokawa Micron Corporation) will be described. Note that the powder processing system 1 in the second embodiment includes a pin mill (powder processing device 4B shown in FIG. 9) instead of the powder processing device 4A, but other configurations are the same as those in the first embodiment, so detailed description thereof will be omitted.

[0096] 9 is a schematic cross-sectional view showing the configuration of a powder processing apparatus 4B according to embodiment 2. The powder processing apparatus 4B includes a cylindrical casing 420 in which powder processing is performed. The casing 420 is provided with a raw material inlet 421, a fixed grinding rotor 422, a rotating grinding rotor 423, a powder outlet 424, and the like.

[0097] The casing 420 may be made of the same material as the casing of the powder processing apparatus 4A described in embodiment 1. That is, metals such as ferrous steels such as SS400, S25C, S45C, SPHC, etc., stainless steels such as SUS304, SUS316, iron castings such as FC20, FC40, stainless castings such as SCS13, 14, etc., ceramics, glass, etc. may be used. In addition, aluminum, wood, synthetic resin, etc. may be used if an abrasion-resistant material is attached to the inner wall surface.

[0098] In addition, in order to improve the durability of the device, the inner surface of casing 420 may be subjected to abrasion-resistant treatments such as plating such as hard chrome plating, abrasion-resistant spraying material treatment such as tungsten carbide spraying, metal deposition under vacuum, and diamond-structured carbon deposition.

[0099] The casing 420 is provided with a raw material inlet 421 for injecting the powder raw material supplied from the raw material supply device 2 into the casing 420. This raw material inlet 421 is provided above the pulverizing rotors 422, 423, for example.

[0100] The grinding rotor 422 is a fixed rotor, and includes a disk 422A fixed in the casing 420 and a plurality of pins 422B protruding toward the grinding rotor 423. On the other hand, the grinding rotor 423 is a rotating rotor, and includes a rotating disk 423A and a plurality of pins 423B protruding toward the grinding rotor 422, and is configured to rotate at a desired rotation speed by the power of the grinding motor 413M (see FIG. 3). The pins 423B of the grinding rotor 423 are positioned so as not to collide with the pins 422B of the fixed grinding rotor 422 when the rotating disk 423A rotates. In this embodiment, the shape, dimensions, arrangement, number, and material of the pins 422B, 423B of the grinding rotors 422, 423 are appropriately designed according to the particle size, circularity, etc. of the required product powder.

[0101] The grinding rotor 423 rotates by the power of the grinding motor 413M, generating a swirling airflow inside the casing 420, and through the action of the pins 422B, 423B, imparts mechanical energy such as impact, compression, grinding, shearing, etc. to the powdered raw material introduced into the casing 420, thereby grinding the powdered raw material.

[0102] The material of the grinding rotors 422, 423 may be a known material that has been used for grinding rotors of powder processing equipment in the past. For example, SS400, S25C, S45C, SUS304, SUS316, SUS630, etc. may be used. The pins 422B, 423B may be provided with a tip of a super hard alloy to withstand impact force, or may be made of a ceramic or a composite of metal and ceramic, such as a cermet, that has wear resistance and toughness.

[0103] Furthermore, in order to improve the durability of the device, the surfaces of the pulverizing rotors 422, 423 may be subjected to plating treatment such as hard chrome plating, wear-resistant spraying treatment such as tungsten carbide spraying, metal deposition under vacuum, wear-resistant treatment such as diamond-structured carbon deposition, SUS630 quenching hardening treatment, etc. In addition, the surfaces of the pulverizing rotors 422, 423 may be subjected to buff polishing, electrolytic polishing, coating with PTFE or the like, plating with nickel or the like.

[0104] The powder outlet 424 is connected to the cyclone 5, the dust collector 6, and the blower 7 via the powder transport path TP3, etc. The powder pulverized by the pulverizing rotors 422, 423 is taken out from the powder outlet 424 together with the airflow.

[0105] In order to generate a learning model that outputs, in response to an input of a desired particle diameter, the calculation results related to the control parameters of the powder processing device 4B, the control device 100 collects the rotation speed of the grinding rotor 423 measured by the rotation speed sensor S5, the discharge / suction flow rates measured by the flow rate sensor S3, and the particle diameter data measured by the particle diameter sensor S2 as teacher data.

[0106] The control unit 101 of the control device 100 uses the collected teacher data to generate a learning model 210 (see FIG. 10) that outputs calculation results relating to control parameters that control the rotation speed of the grinding rotor 423 and the discharge / suction flow rate when extracting powder from the powder outlet 424 in response to input of a particle diameter desired by the user. The procedure for generating the learning model 210 is the same as in embodiment 1, and therefore a description thereof will be omitted. The learning model 210 may be generated by the control device 100 or an external server device (not shown).

[0107] 10 is a schematic diagram showing a configuration example of a learning model 210 in embodiment 2. Similar to the learning model 200 described in embodiment 1, the learning model 210 includes an input layer 211, intermediate layers 212A and 212B, and an output layer 213, each of which includes one or more nodes. The number of intermediate layers is not limited to two, and may be three or more.

[0108] The learning model 210 according to the second embodiment is configured to output calculation results relating to the control parameters that control the rotation speed of the grinding rotor 423 and the discharge / suction flow rate from the casing 420 in response to the input of the particle diameter (desired particle diameter) desired by the user.

[0109] When performing a calculation using the learning model 210, the control device 100 inputs a particle diameter desired by a user (desired particle diameter) to the learning model 210. The data of the particle diameter input to the learning model 210 is output to the nodes of the first intermediate layer 212A through the nodes constituting the input layer 211. The data input to the first intermediate layer 212A is output to the nodes of the next intermediate layer 212B through the nodes constituting the intermediate layer 212A. At this time, an output is calculated using an activation function including weights and biases set between the nodes. In the same manner, calculations are performed using activation functions including weights and biases set between the nodes, and the results are transmitted to the subsequent layers one after another until a calculation result is obtained by the output layer 213.

[0110] The output layer 213 outputs the calculation results related to the control parameters that control the rotation speed of the pulverizing rotor 423 and the discharge / suction flow rate when the powder is taken out of the casing 420. As the calculation results, for example, a probability indicating whether the combination of the above-mentioned multiple control parameters is good or bad may be output. Specifically, the output layer 213 may be configured with n nodes from the first node to the n-th node, and the first node may output a probability P1 that the rotation speed of the pulverizing rotor 423 is G1 and the discharge / suction flow rate is V1, the second node may output a probability P2 that the rotation speed of the pulverizing rotor 423 is G2 and the discharge / suction flow rate is V2, ..., and the n-th node may output a probability Pn that the rotation speed of the pulverizing rotor 423 is Gn and the discharge / suction flow rate is Vn. The number of nodes that configure the output layer 213 and the calculation results assigned to each node are not limited to the above example, and can be designed appropriately.

[0111] The control unit 101 of the control device 100 determines the control parameters to be used for control by identifying the combination with the highest probability of being output from the output layer 213 (in this embodiment, the combination of the rotation speed of the grinding rotor 423 and the discharge / suction flow rate). Based on the determined control parameters, the control unit 101 generates control commands to control the operation of the grinding motor 413M and the blower 7, and outputs them to each device via the output unit 104.

[0112] As described above, in the second embodiment, for the pin mill (powder processing device 4B), a learning model 210 is generated that outputs calculation results relating to control parameters for controlling the rotation speed of the pulverization rotor 423 and the discharge and suction flow rates when extracting powder from the powder outlet 424 in response to an input of a particle diameter desired by a user. In addition, by using the learning model 210, calculation results relating to control parameters for controlling the rotation speed of the pulverization rotor 423 and the discharge and suction flow rates from the casing 420 are obtained. The control device 100 can control the operation of the powder processing system 1 based on the calculation results of the learning model 210 so as to obtain powder having a desired particle diameter.

[0113] (Embodiment 3) In the third embodiment, an application example to an impact crusher (for example, Glacis (registered trademark) manufactured by Hosokawa Micron Corporation) will be described. Note that the powder processing system 1 in the third embodiment includes an impact crusher (powder processing device 4C shown in FIG. 11) instead of the powder processing device 4A, but other configurations are the same as those in the first embodiment, so detailed description thereof will be omitted.

[0114] 11 is a schematic cross-sectional view showing the configuration of a powder processing apparatus 4C according to embodiment 3. The powder processing apparatus 4C includes a cylindrical casing 430 in which powder processing is performed. The casing 430 is provided with a raw material inlet 431, a grinding rotor 432, a liner 433, a powder outlet 434, and the like.

[0115] The casing 430 may be made of the same material as the casing of the powder processing apparatus 4A described in embodiment 1. That is, metals such as ferrous steels such as SS400, S25C, S45C, SPHC, etc., stainless steels such as SUS304, SUS316, iron castings such as FC20, FC40, stainless castings such as SCS13, 14, etc., ceramics, glass, etc. may be used. In addition, aluminum, wood, synthetic resin, etc. may be used if an abrasion-resistant material is attached to the inner wall surface.

[0116] In addition, in order to improve the durability of the device, the inner surface of the casing 430 may be subjected to abrasion-resistant treatments such as plating treatments such as hard chrome plating, abrasion-resistant spraying treatments such as tungsten carbide spraying, metal deposition under vacuum, and diamond-structured carbon deposition.

[0117] The casing 430 is provided with a raw material inlet 431 for feeding the powder raw material supplied from the raw material supply device 2 into the casing 430. The raw material inlet 431 is provided, for example, above the pulverizing rotor 432. The powder raw material supplied from the raw material supply device 2 is transported, for example, by a screw feeder (not shown) in the raw material supply path TP1, and fed into the casing 430 through the raw material inlet 431. A coolant may also be introduced into the casing 430. The coolant may be introduced, for example, through a coolant inlet provided on the circumferential surface of the casing 430 or through the rotating shaft of the pulverizing rotor 432.

[0118] The grinding rotor 432 includes a plurality of rotating disks 432A, 432A, ..., 432A arranged coaxially along the rotation axis, and is configured to rotate at a desired rotation speed by the power of a grinding motor 413M (see FIG. 3). Triangular, wave-shaped and wedge-shaped grooves are provided on the peripheral surface of the rotating disk 432A. A liner 433 is provided on the inner peripheral surface of the casing 430 at a position facing the peripheral surface of the rotating disk 432A. The liner 433 is a cylindrical member having a central axis aligned with the rotation axis direction of the grinding rotor 432, and the inner peripheral surface of this cylindrical member is provided with triangular, wave-shaped and wedge-shaped grooves.

[0119] The material of the grinding rotor 432 and the liner 433 may be a conventionally used known material, such as SS400, S25C, S45C, SUS304, SUS316, SUS630, etc. Also, in order to withstand impact force, a tip of a cemented carbide alloy may be attached, or a composite of metal and ceramics, such as ceramics or cermet, having wear resistance and toughness may be used.

[0120] Furthermore, in order to improve the durability of the device, the surfaces of the pulverization rotor 432 and the liner 433 may be subjected to plating treatment such as hard chrome plating, wear-resistant spraying treatment such as tungsten carbide spraying, wear-resistant treatment such as metal deposition under vacuum and carbon deposition with a diamond structure, SUS630 quenching hardening treatment, etc. In addition, the surfaces of the pulverization rotor 432 and the liner 433 may be subjected to buff polishing, electrolytic polishing, coating with PTFE or the like, plating with nickel or the like.

[0121] The grinding rotor 432 rotates by the power of the grinding motor 413M, generating a swirling airflow inside the casing 430, and the action of the grinding rotor 432 and the liner 433 imparts mechanical energy such as impact, compression, grinding, shearing, etc. to the powdered raw material introduced into the casing 430, thereby grinding the powdered raw material.

[0122] The powder outlet 434 is connected to the cyclone 5, dust collector 6, and blower 7 via the powder transport path TP3, etc. The powder pulverized by the pulverization rotor 432 and the liner 433 is taken out from the powder outlet 434 together with the airflow.

[0123] In order to generate a learning model that outputs, in response to an input of a desired particle diameter, the calculation results related to the control parameters of the powder processing device 4C, the control device 100 collects the rotation speed of the grinding rotor 432 measured by the rotation speed sensor S5, the discharge / suction flow rates measured by the flow rate sensor S3, and the particle diameter data measured by the particle diameter sensor S2 as teacher data.

[0124] The control unit 101 of the control device 100 uses the collected teacher data to generate a learning model that outputs calculation results relating to control parameters that control the rotation speed of the pulverization rotor 432 and the discharge / suction flow rates when extracting powder from the powder outlet 434, in response to an input of a particle diameter desired by the user. The learning model may be generated by the control device 100 or an external server device (not shown). Note that the configuration and generation procedure of the learning model generated in the third embodiment are the same as those in the second embodiment, and therefore the description thereof will be omitted.

[0125] As described above, in the third embodiment, a learning model is generated for the powder processing device 4C, which outputs calculation results relating to control parameters for controlling the rotation speed of the pulverization rotor 432 and the discharge and suction flow rates when extracting powder from the powder outlet 434 in response to input of a particle diameter desired by the user. Furthermore, by using the learning model, calculation results relating to control parameters for controlling the rotation speed of the pulverization rotor 432 and the discharge and suction flow rates from the casing 430 are obtained. The control device 100 can control the operation of the powder processing system 1 based on the calculation results of the learning model so as to obtain powder having a desired particle diameter.

[0126] (Embodiment 4) In the fourth embodiment, an application example to a medium stirring type powder processing device (for example, Pulvis (registered trademark) manufactured by Hosokawa Micron Corporation) will be described. Note that the powder processing system 1 in the fourth embodiment includes a medium stirring type powder processing device (powder processing device 4D shown in FIG. 12) instead of the powder processing device 4A, but other configurations are the same as those in the first embodiment, so detailed description thereof will be omitted.

[0127] 12 is a schematic cross-sectional view showing the configuration of a powder processing apparatus 4D according to embodiment 4. The powder processing apparatus 4D includes a cylindrical casing 440 in which powder processing is performed. The casing 440 includes a raw material inlet 441, a gas inlet 442, a pulverizing section 443, a classifying rotor 444, a powder outlet 445, an agitator 446, and the like.

[0128] The casing 440 may be made of the same material as the casing of the powder processing apparatus 4D described in embodiment 1. That is, metals such as ferrous steels such as SS400, S25C, S45C, SPHC, etc., stainless steels such as SUS304, SUS316, iron castings such as FC20, FC40, stainless castings such as SCS13, 14, etc., ceramics, glass, etc. may be used. In addition, aluminum, wood, synthetic resin, etc. may be used if an abrasion-resistant material is attached to the inner wall surface.

[0129] In addition, in order to improve the durability of the device, the inner surface of the casing 440 may be subjected to abrasion-resistant treatments such as plating such as hard chrome plating, abrasion-resistant spraying material treatment such as tungsten carbide spraying, metal deposition under vacuum, and diamond-structured carbon deposition.

[0130] The casing 440 is provided with a raw material inlet 441 for injecting the powdered raw material supplied from the raw material supply device 2 into the casing 440. This raw material inlet 441 is provided, for example, above the grinding section 443. The powdered raw material supplied from the raw material supply device 2 is transported, for example, by a screw feeder (not shown) in the raw material supply path TP1, and is introduced into the casing 440 through the raw material inlet 441.

[0131] In the powder processing device 4D according to the fourth embodiment, a gas such as compressed air is introduced into the casing 440 through the gas inlet 442 to generate an airflow inside the casing 440. The powder processing device 4D also includes a pulverizing section 443 having an agitator 446 for stirring balls (or beads) as a medium. The agitator 446 is configured to rotate at a desired rotation speed by the power of the stirring motor 446M. Here, the stirring motor 446M is one of the driving sections included in the powder processing system 1. The powder raw material introduced into the inside of the casing 440 is in a state of covering the surface of the medium. In the pulverizing section 443, the medium is stirred by rotating the agitator 446, and the powder raw material covering the surface of the medium is subjected to impact, compression, shearing, and grinding actions, thereby pulverizing the powder raw material. The pulverized powder is transported together with the airflow and classified by a classifying rotor 444 provided at the top of the device.

[0132] The configuration of the classifying rotor 444 is the same as that of the first embodiment. That is, the classifying rotor 444 is a rotor equipped with a plurality of radially arranged classifying blades 444A, and is configured to rotate at a desired rotation speed by the power of the classifying motor 415M. The rotation speed of the classifying motor is measured by a rotation speed sensor S6 (see FIG. 3) all the time or at regular intervals (for example, every 5 seconds). The classifying rotor 444 allows only powders having a particle size smaller than a predetermined size among the powders processed in the casing 440 to pass by centrifugal force due to high-speed rotation, and leads only the passed powders to the powder outlet 445. On the other hand, the powders that cannot pass through the classifying rotor 444 circulate in the casing 440 and are repeatedly processed.

[0133] Here, the particle size of the powder passing through the powder outlet 445 can be set by controlling the rotation speed of the agitator 446, the rotation speed of the classifying rotor 444, and the discharge and suction flow rates.

[0134] In order to generate a learning model 220 that outputs control parameters for the powder processing device 4D as calculation results in response to an input of a desired particle diameter, the control device 100 collects, as teacher data, the rotation speed of the agitator 446 measured by the rotation speed sensor S5, the rotation speed of the classifying rotor 444 measured by the rotation speed sensor S6, the discharge / suction flow rates measured by the flow rate sensor S3, and the particle diameter data measured by the particle diameter sensor S2.

[0135] The control unit 101 of the control device 100 uses the collected teacher data to generate a learning model 220 (see FIG. 13) that outputs calculation results related to control parameters that control the rotation speed of the agitator 446, the rotation speed of the classifying rotor 444, and the discharge / suction flow rate when extracting powder from the powder outlet 445, in response to an input of a particle diameter desired by the user. The procedure for generating the learning model 220 is the same as in the first embodiment, and therefore a description thereof will be omitted. The learning model 220 may be generated by the control device 100, or may be generated by an external server device (not shown).

[0136] 13 is a schematic diagram showing a configuration example of a learning model 220 in embodiment 4. As with the learning model 200 described in embodiment 1, the learning model 220 includes an input layer 221, intermediate layers 222A and 222B, and an output layer 223, each of which includes one or more nodes. The number of intermediate layers is not limited to two, and may be three or more.

[0137] The learning model 220 according to the fourth embodiment is configured to output calculation results regarding the control parameters that control the rotation speed of the agitator 446, the rotation speed of the classifying rotor 444, and the discharge / suction flow rate from the casing 440 in response to an input of a particle diameter (desired particle diameter) desired by a user.

[0138] When performing a calculation using the learning model 220, the control device 100 inputs a particle diameter (desired particle diameter) desired by a user to the learning model 220. The data of the particle diameter input to the learning model 220 is output to the nodes of the first intermediate layer 222A through the nodes constituting the input layer 221. The data input to the first intermediate layer 222A is output to the nodes of the next intermediate layer 222B through the nodes constituting the intermediate layer 222A. At this time, an output is calculated using an activation function including weights and biases set between the nodes. In the same manner, calculations are performed using activation functions including weights and biases set between the nodes, and the results are transmitted to the subsequent layers one after another until a calculation result is obtained by the output layer 223.

[0139] The output layer 223 outputs the calculation results related to the control parameters that control the rotation speed of the agitator 446, the rotation speed of the classifying rotor 444, and the discharge / suction flow rate when the powder is taken out of the casing 440. As the calculation results, for example, a probability indicating whether the combination of the above-mentioned multiple control parameters is good or bad may be output. Specifically, the output layer 223 may be configured with n nodes from the first node to the n-th node, and the first node may output a probability P1 that the rotation speed of the agitator 446 is A1, the rotation speed of the classifying rotor 444 is C1, and the discharge / suction flow rate is V1, the second node may output a probability P2 that the rotation speed of the agitator 446 is A2, the rotation speed of the classifying rotor 444 is C2, and the discharge / suction flow rate is V2, ..., and the n-th node may output a probability Pn that the rotation speed of the agitator 446 is An, the rotation speed of the classifying rotor 444 is Cn, and the discharge / suction flow rate is Vn. The number of nodes constituting the output layer 223 and the calculation results assigned to each node are not limited to the above example, and can be designed as appropriate.

[0140] The control unit 101 of the control device 100 determines the control parameters to be used for control by identifying the combination with the highest probability of being output from the output layer 223 (in this embodiment, the combination of the rotation speed of the agitator 446, the rotation speed of the classifying rotor 444, and the discharge / suction flow rate). Based on the determined control parameters, the control unit 101 generates control commands to control the operations of the agitating motor 446M, the classifying motor 415M, and the blower 7, and outputs the control commands to each device via the output unit 104.

[0141] As described above, in the fourth embodiment, for the powder processing apparatus 4D, a learning model 220 is generated that outputs a calculation result regarding the control parameters for controlling the rotation speed of the agitator 446, the rotation speed of the classifying rotor 444, and the discharge and suction flow rate when extracting powder from the powder outlet 445, in response to an input of a particle diameter desired by the user. In addition, by using the learning model 220, a calculation result regarding the control parameters for controlling the rotation speed of the agitator 446, the rotation speed of the classifying rotor 444, and the discharge and suction flow rate from the casing 440 is obtained. The control device 100 can control the operation of the powder processing system 1 based on the calculation result of the learning model 220 so as to obtain powder having a desired particle diameter.

[0142] (Embodiment 5) In the first embodiment, the powder processing device 4A is described as applying mechanical energy such as impact, compression, grinding, shearing, etc. to the powder raw material introduced into the casing 410 to pulverize the powder raw material. However, instead of the mechanical powder processing device 4A, the present invention may be applied to an airflow type powder processing device (for example, Micron Jet (registered trademark) T-type manufactured by Hosokawa Micron Corporation) that performs pulverization processing by the action of a jet airflow introduced into the casing. In the fifth embodiment, an application example to an airflow type powder processing device will be described. The powder processing system 1 in the fifth embodiment includes an airflow type powder processing device 4E (see FIG. 14) instead of the mechanical type powder processing device 4A, but other configurations are the same as those in the first embodiment, so detailed description thereof will be omitted.

[0143] 14 is a schematic cross-sectional view showing the configuration of a powder processing apparatus 4E according to embodiment 5. The powder processing apparatus 4E includes a cylindrical casing 450 in which powder processing is performed. The casing 450 includes a raw material inlet 451, an ejector gas inlet 452, a pulverization gas inlet 453, an impact plate 454, a classification rotor 455, a powder outlet 456, and the like.

[0144] The casing 450 may be made of the same material as the casing of the powder processing apparatus 4A described in embodiment 1. That is, metals such as ferrous steels such as SS400, S25C, S45C, SPHC, etc., stainless steels such as SUS304, SUS316, iron castings such as FC20, FC40, stainless castings such as SCS13, 14, etc., ceramics, glass, etc. may be used. Furthermore, aluminum, wood, synthetic resin, etc. may be used if an abrasion-resistant material is attached to the inner wall surface.

[0145] In addition, in order to improve the durability of the device, the inner surface of the casing 450 may be subjected to abrasion-resistant treatments such as plating such as hard chrome plating, abrasion-resistant spraying material treatment such as tungsten carbide spraying, metal deposition under vacuum, and diamond-structured carbon deposition.

[0146] The casing 450 is provided with a raw material inlet 451 for injecting the powder raw material supplied from the raw material supply device 2 into the casing 450. This raw material inlet 451 is provided, for example, above the ejector gas inlet 452. The powder raw material supplied from the raw material supply device 2 is transported, for example, by a screw feeder (not shown) in the raw material supply path TP1, and is introduced into the casing 450 through the raw material inlet 451.

[0147] In the powder processing apparatus 4E according to the fifth embodiment, a gas such as compressed air is introduced into the casing 450 through the ejector gas inlet 452 and the pulverization gas inlet 453, thereby generating a jet stream inside the casing 450. The powder raw material introduced from the raw material inlet 451 is sucked and accelerated by the jet stream inside the casing 450, and is pulverized by colliding with the collision plate 454. Alternatively, the powder particles sucked and accelerated by the jet stream collide with each other, thereby being pulverized. A pressure sensor (not shown) for measuring the pulverization pressure constantly or periodically (for example, every 5 seconds) is provided inside the casing 450.

[0148] The powder pulverized by the action of the jet air stream is transported together with the air stream and classified by the classification rotor 455 provided at the top of the device. A cylindrical guide ring 457 may be provided around the collision plate 454 to guide the pulverized powder to the classification rotor 455.

[0149] The configuration of the classifying rotor 455 is the same as that of the first embodiment. That is, the classifying rotor 455 is a rotor equipped with a plurality of radially arranged classifying blades 455A, and is configured to rotate at a desired rotation speed by the power of the classifying motor 415M. The rotation speed of the classifying motor is measured by a rotation speed sensor S6 (see FIG. 3) all the time or at regular intervals (for example, every 5 seconds). The classifying rotor 455 passes only powder having a particle size smaller than a predetermined size among the powder processed in the casing 450 by centrifugal force due to high-speed rotation, and leads only the passed powder to the powder outlet 456.

[0150] Here, the particle size of the powder passing through the classifying rotor 455 can be set by controlling the rotation speed, etc., of the classifying rotor 455. That is, by controlling the rotation speed of the classifying rotor 455, powder having a particle size smaller than a predetermined size can be taken out from inside the casing 450. On the other hand, powder that cannot pass through the classifying rotor 455 circulates inside the casing 450 and is repeatedly processed.

[0151] In order to generate a learning model 230 that outputs calculation results related to the control parameters of the powder processing device 4E in response to an input of a desired particle diameter, the control device 100 collects, as teacher data, the grinding pressure measured by a pressure sensor (not shown) provided at the grinding gas inlet 453, the rotational speed of the classifying rotor 455 measured by the rotational speed sensor S6, the discharge / suction flow rates measured by the flow rate sensor S3, and the particle diameter data measured by the particle diameter sensor S2.

[0152] The control unit 101 of the control device 100 uses the collected teacher data to generate a learning model 230 (see FIG. 15) that outputs calculation results related to control parameters that control the grinding pressure in the casing 450, the rotation speed of the classifying rotor 455, and the discharge / suction flow rate when extracting powder from the powder outlet 456, in response to an input of a particle diameter desired by the user. The procedure for generating the learning model 230 is the same as in the first embodiment, and therefore a description thereof will be omitted. The learning model 230 may be generated by the control device 100 or an external server device (not shown).

[0153] 15 is a schematic diagram showing a configuration example of a learning model 230 in embodiment 5. As with the learning model 200 described in embodiment 1, the learning model 230 includes an input layer 231, intermediate layers 232A and 232B, and an output layer 233, each of which includes one or more nodes. The number of intermediate layers is not limited to two, and may be three or more.

[0154] The learning model 230 according to the fifth embodiment is configured to output calculation results relating to the control parameters that control the grinding pressure in the casing 450, the rotation speed of the classifying rotor 455, and the discharge / suction flow rate from the casing 450 in response to an input of a particle diameter (desired particle diameter) desired by a user.

[0155] When performing a calculation using the learning model 230, the control device 100 inputs a particle diameter (desired particle diameter) desired by the user to the learning model 230. The data of the particle diameter input to the learning model 230 is output to the nodes of the first intermediate layer 232A through the nodes constituting the input layer 231. The data input to the first intermediate layer 232A is output to the nodes of the next intermediate layer 232B through the nodes constituting the intermediate layer 232A. At this time, the output is calculated using an activation function including weights and biases set between the nodes. In the same manner, calculations are performed using activation functions including weights and biases set between the nodes, and the results are transmitted to the subsequent layers one after another until the calculation result is obtained by the output layer 233.

[0156] The output layer 233 outputs the calculation results related to the control parameters that control the pulverization pressure of the pulverization gas inlet 453, the rotation speed of the classifying rotor 455, and the discharge / suction flow rate when the powder is taken out of the casing 450. As the calculation results, for example, a probability indicating whether the combination of the above-mentioned multiple control parameters is good or bad may be output. Specifically, the output layer 233 may be configured with n nodes from the first node to the n-th node, and the first node may output a probability P1 that the pulverization pressure of the pulverization gas inlet 453 is B1, the rotation speed of the classifying rotor 455 is C1, and the discharge / suction flow rate is V1, the second node may output a probability P2 that the pulverization pressure of the pulverization gas inlet 453 is B2, the rotation speed of the classifying rotor 455 is C2, and the discharge / suction flow rate is V2, ..., and the n-th node may output a probability Pn that the pulverization pressure of the pulverization gas inlet 453 is Bn, the rotation speed of the classifying rotor 455 is Cn, and the discharge / suction flow rate is Vn. The number of nodes constituting the output layer 233 and the calculation results assigned to each node are not limited to the above example, and can be designed as appropriate.

[0157] The control unit 101 of the control device 100 determines the control parameters to be used for control by identifying the combination with the highest probability of being output from the output layer 233 (in this embodiment, the combination of the grinding pressure of the grinding gas inlet 453, the rotation speed of the classifying rotor 455, and the discharge / suction flow rate). Based on the determined control parameters, the control unit 101 generates control commands to control the grinding pressure of the compressed air or the like that generates a jet stream in the casing 450, the classifying motor 415M, and the operation of the blower 7, and outputs the commands to each device via the output unit 104.

[0158] As described above, in the fifth embodiment, for the airflow type powder processing device 4E, the learning model 230 is generated to output the calculation results regarding the control parameters for controlling the pulverization pressure of the pulverization gas inlet 453, the rotation speed of the classifying rotor 455, and the discharge and suction flow rate when the powder is taken out from the powder outlet 456, in response to the input of the particle diameter desired by the user. In addition, by using the learning model 230, the calculation results regarding the control parameters for controlling the pulverization pressure of the pulverization gas inlet 453, the rotation speed of the classifying rotor 455, and the discharge and suction flow rate from the casing 450 are obtained. The control device 100 can control the operation of the powder processing system 1 based on the calculation results of the learning model 230 so as to obtain powder having a desired particle diameter.

[0159] (Embodiment 6) In embodiment 6, with respect to the powder processing apparatus 4A, the configuration of the learning model 240 is described which outputs calculation results regarding the control parameters that control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, the discharge / suction flow rate from the casing 410, the supply amount of powder raw material by the raw material supply machine 2, and the temperature inside the casing 410 in response to input of the particle diameter desired by the user. The overall configuration of the powder processing system 1 and the configuration of each device in the powder processing system 1 are similar to those in the first embodiment, and therefore the description thereof will be omitted.

[0160] 16 is a schematic diagram showing a configuration example of a learning model 240 in embodiment 6. Similar to the learning model 200 described in embodiment 1, the learning model 240 includes an input layer 241, intermediate layers 242A and 242B, and an output layer 243, each of which includes one or more nodes. The number of intermediate layers is not limited to two, and may be three or more.

[0161] The learning model 240 according to the sixth embodiment is configured to output calculation results relating to the control parameters that control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, the discharge / suction flow rates from the casing 410, the supply amount of powdered raw material by the raw material supply device 2, and the temperature inside the casing 410, in response to an input of a particle diameter (desired particle diameter) desired by a user. The supply amount of powdered raw material may be the supply amount per unit time (i.e., the supply rate).

[0162] Such a learning model 240 is generated by learning the relationship between the particle diameter and the control parameters using the particle diameter data (particle diameter data measured by the particle diameter sensor S2) obtained from the powder processing device 4A and the measurement data including the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, the discharge and suction flow rate when taking out the powder from the casing 410, the supply amount of the powder raw material obtained by the weight sensor S1, and the temperature inside the casing 410 obtained by the temperature sensor S4 as teacher data. The supply amount of the powder raw material may be the supply amount per unit time (i.e., the supply rate). Alternatively, only one of the supply amount of the powder raw material and the temperature inside the casing 410 may be adopted as the control parameter. The learning model 240 may be generated by the control device 100 or an external server device (not shown).

[0163] When performing a calculation using the learning model 240, the control device 100 inputs a particle diameter (desired particle diameter) desired by the user to the learning model 240. The data of the particle diameter input to the learning model 240 is output to the nodes of the first intermediate layer 242A through the nodes constituting the input layer 241. The data input to the first intermediate layer 242A is output to the nodes of the next intermediate layer 242B through the nodes constituting the intermediate layer 242A. At this time, the output is calculated using an activation function including weights and biases set between the nodes. In the same manner, calculations are performed using activation functions including weights and biases set between the nodes, and the results are transmitted to the subsequent layers one after another until the calculation result is obtained by the output layer 243.

[0164] The output layer 243 outputs the calculation results related to the control parameters that control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, the discharge / suction flow rates when taking out the powder from the casing 410, the supply amount of the powder raw material, and the temperature inside the casing 410. As the calculation result, for example, a probability indicating whether the combination of the above-mentioned multiple control parameters is good or bad may be output. Specifically, the output layer 243 may be composed of n nodes from the first node to the nth node, and the first node may output a probability P1 that the rotation speed of the grinding rotor 413 is G1, the rotation speed of the classifying rotor 415 is C1, the discharge / suction flow rate is V1, the supply amount of powdered raw material is F1, and the temperature inside the casing 410 is T1; the second node may output a probability P2 that the rotation speed of the grinding rotor 413 is G2, the rotation speed of the classifying rotor 415 is C2, the discharge / suction flow rate is V2, the supply amount of powdered raw material is F2, and the temperature inside the casing 410 is T2; ..., the nth node may output a probability Pn that the rotation speed of the grinding rotor 413 is Gn, the rotation speed of the classifying rotor 415 is Cn, the discharge / suction flow rate is Vn, the supply amount of powdered raw material is Fn, and the temperature inside the casing 410 is Tn. The number of nodes constituting the output layer 243 and the calculation results assigned to each node are not limited to the above example, and can be designed as appropriate.

[0165] The control unit 101 of the control device 100 determines the control parameters to be used for control by identifying the combination with the highest probability of being output from the output layer 243 (in this embodiment, the combination of the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, the discharge / suction flow rates, the supply amount of powder raw material, and the temperature inside the casing 410). Based on the determined control parameters, the control unit 101 generates control commands to control the operation of each of the grinding motor 413M, the classifying motor 415M, the blower 7, the raw material feeder 2, and the hot air generator 3, and outputs these to each device via the output unit 104.

[0166] As described above, when the learning model 240 according to the sixth embodiment is used, calculation results are obtained for the control parameters that control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, the discharge / suction flow rate from the casing 410, as well as the supply amount of the powder raw material and the temperature inside the casing 410. The control device 100 can control the operation of the powder processing system 1 based on the calculation results of the learning model 240 so as to obtain powder having a desired particle size.

[0167] In the sixth embodiment, an application example to the mechanical powder processing apparatus 4A has been described, but it goes without saying that a similar method can be applied to the powder processing apparatuses 4B to 4E.

[0168] (Embodiment 7) In embodiment 7, a learning model 250 is described which outputs calculation results for the powder processing device 4A in response to input of at least one of the particle diameter desired by the user and the circularity of the powder, the moisture content of the powder raw material, the driving power or driving current of the powder processing device 4A, the heat medium temperature, the medium weight, the environmental temperature, and the environmental humidity. The overall configuration of the powder processing system 1 and the configuration of each device in the powder processing system 1 are similar to those in the first embodiment, and therefore the description thereof will be omitted.

[0169] 17 is a schematic diagram showing a configuration example of a learning model 250 in embodiment 7. Similar to the learning model 200 described in embodiment 1, the learning model 250 includes an input layer 251, intermediate layers 252A and 252B, and an output layer 253, each of which includes one or more nodes. The number of intermediate layers is not limited to two, and may be three or more.

[0170] The learning model 250 according to the seventh embodiment is configured to output calculation results relating to control parameters for controlling the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate from the casing 410 in response to input of a particle diameter (desired particle diameter) desired by the user and at least one of the circularity of the powder, the moisture content of the powder raw material, the driving power or driving current of the powder processing device 4A, the temperature (heat medium temperature) of the fluid introduced into the powder processing device 4A, the weight of the medium (medium weight) in a media stirring type powder processing device, the environmental temperature, and the environmental humidity.

[0171] Such a learning model 250 is generated by learning the relationship between input and output using, as teacher data, particle diameter data on particle diameter obtained from the powder processing device 4A, data on at least one of the circularity of the powder, the moisture content of the powder raw material, the driving power or driving current of the powder processing device 4A, the heat medium temperature, the medium weight, the environmental temperature, and the environmental humidity, and measurement data including the rotation speed of the grinding rotor 413 and the rotation speed of the classifying rotor 415 of the powder processing device 4A, and the discharge and suction flow rate when taking out the powder from the casing 410. Here, the circularity of the powder is a value expressed by, for example, (the perimeter of a circle having a projected area equal to the projected area of ​​the particle) / (the perimeter of the particle), and can be collected using a known measuring device. The moisture content of the powder raw material is a value expressed by, for example, mass%, and can be collected using a known measuring device such as an infrared moisture meter. The driving power and driving current of the powder processing device 4A may be obtained by acquiring values ​​measured inside the powder processing device 4A. The heat medium temperature may be a value measured inside the powder processing device 4A, and the medium weight may be a value measured in advance. The environmental temperature and environmental humidity are the temperature and humidity of the surrounding environment in which the powder processing device 4A is set, and can be collected using a known temperature sensor and humidity sensor. The learning model 250 may be generated by the control device 100 or an external server device (not shown).

[0172] When the control device 100 performs a calculation using the learning model 250, the control device 100 inputs at least one of the particle diameter (desired particle diameter) of the powder desired by the user, the circularity of the powder desired by the user, the moisture content of the powder desired by the user, the driving power or driving current of the powder processing device 4A set by the user, the heat medium temperature, the medium weight, the actual measured value of the environmental temperature, and the actual measured value of the environmental humidity into the learning model 250. The data input to the learning model 250 is output to the nodes of the first intermediate layer 252A through the nodes constituting the input layer 251. The data input to the first intermediate layer 252A is output to the nodes of the next intermediate layer 252B through the nodes constituting the intermediate layer 252A. At this time, the output is calculated using an activation function including weights and biases set between the nodes. In the same manner, calculations are performed using activation functions including weights and biases set between the nodes, and the calculation results are transmitted to the subsequent layers one after another until the calculation results are obtained by the output layer 253.

[0173] The output layer 253 outputs the calculation results related to the control parameters that control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, the discharge / suction flow rate when taking out the powder from the casing 410, the supply amount of the powder raw material, and the temperature inside the casing 410. As the calculation result, for example, a probability indicating whether the combination of the above-mentioned multiple control parameters is good or bad may be output. Specifically, the output layer 253 may be configured with n nodes from the first node to the n-th node, and the first node may output a probability P1 that the rotation speed of the grinding rotor 413 is G1, the rotation speed of the classifying rotor 415 is C1, and the discharge / suction flow rate is V1, the second node may output a probability P2 that the rotation speed of the grinding rotor 413 is G2, the rotation speed of the classifying rotor 415 is C2, and the discharge / suction flow rate is V2, ..., and the n-th node may output a probability Pn that the rotation speed of the grinding rotor 413 is Gn, the rotation speed of the classifying rotor 415 is Cn, and the discharge / suction flow rate is Vn. The number of nodes constituting the output layer 253 and the calculation results assigned to each node are not limited to the above example, and may be designed as appropriate.

[0174] The control unit 101 of the control device 100 determines the control parameters to be used for control by identifying the combination with the highest probability of being output from the output layer 253 (in this embodiment, the combination of the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate). Based on the determined control parameters, the control unit 101 generates control commands to control the operation of each of the grinding motor 413M, the classifying motor 415M, and the blower 7, and outputs them to each device via the output unit 104.

[0175] As described above, when the learning model 250 of embodiment 7 is used, powder can be generated based on conditions such as the particle size of the powder desired by the user, the circularity of the powder desired by the user, the moisture content of the powder desired by the user, and the driving power or driving current of the powder processing device 4A set by the user.

[0176] The learning model 250 according to the seventh embodiment is configured to output a calculation result in response to input of at least one of the particle diameter desired by the user, the circularity of the powder, the moisture content of the powder raw material, the driving power or driving current of the powder processing device 4A, the heat medium temperature, the medium weight, the ambient temperature, and the ambient humidity, but may also be configured to further input at least one of the dust concentration, the powder composition, the powder moisture, the pressure, the sound pressure, or the frequency.

[0177] In addition, the learning model 250 in embodiment 7 is configured to output calculation results regarding the control parameters that control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate from the casing 410, but it may also be configured to output calculation results regarding control parameters that control at least one of the supply amount of powdered raw material and the temperature inside the casing 410 in addition to these control parameters.

[0178] Furthermore, when at least one of the control parameters input to the learning model 250 is changed in a time series manner, the learning model 250 may be a model constructed by a recurrent neural network.

[0179] In the seventh embodiment, an application example to the mechanical powder processing apparatus 4A has been described, but it goes without saying that a similar method can be applied to the powder processing apparatuses 4B to 4E.

[0180] (Embodiment 8) In the eighth embodiment, a configuration for generating a learning model 200 according to the type of powder raw material will be described with respect to the powder processing apparatus 4A. The overall configuration of the powder processing system 1 and the configuration of each device in the powder processing system 1 are similar to those in the first embodiment, and therefore the description thereof will be omitted.

[0181] The control device 100 according to the eighth embodiment collects, for each type of powder raw material, powder particle size data obtained from the powder processing device 4A, and measurement data including the rotation speed of the pulverizing rotor 413, the rotation speed of the classifying rotor 415, and the discharge and suction flow rate when the powder is taken out from the casing 410, which are provided in the powder processing device 4A. FIG. 18 is a conceptual diagram showing an example of data collection in the eighth embodiment. The control unit 101 of the control device 100 acquires measurement data from the particle size sensor S2, the rotation speed sensor S5 of the pulverizing rotor 413, the rotation speed sensor S6 of the classifying rotor 415, and the flow rate sensor S3 through the input unit 103, and stores the acquired data together with a time stamp and information on the type of powder raw material in the memory unit 102. Here, the information on the type of powder raw material may be text information indicating the type name of the powder raw material, or may be any identifier that can identify the type of powder raw material. The information on the type of powder raw material may be received through the operation unit 106 before or after data collection. The example of FIG. 18 shows a state in which data is collected for each type of powder raw material, such as toner raw material, graphite, and manganese dioxide (MnO 2 ), and stored in the storage unit 102.

[0182] The control unit 101 generates a learning model 200 using the data collected for each type of powder raw material as training data.

[0183] 19 is a flowchart illustrating the procedure for generating the learning model 200 according to the eighth embodiment. Prior to generating the learning model 200, the control unit 101 of the control device 100 receives information regarding the type of powdered raw material (step S401). The control unit 101 can, for example, display a screen on the display unit 107 that inquires of the user about the type of powdered raw material, and receive the information regarding the type of powdered raw material through the displayed screen. Alternatively, the control unit 101 may receive the information regarding the type of powdered raw material by sending an inquiry regarding the type of powdered raw material to the terminal device 500 via the communication unit 105, and receiving a reply from the terminal device 500.

[0184] Next, the control unit 101 reads data stored in association with the type received in step S401 from the storage unit 102 (step S402). The control unit 101 selects data to be used as teacher data from the read data (step S403). That is, the control unit 101 selects only one set from the read data: the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the actual measured values ​​of the discharge and suction flow rates from the casing 410, and the particle diameter values ​​(D10, D50, D90) of the powder obtained at that time.

[0185] Next, the control unit 101 inputs the particle diameter value included in the selected teacher data to the learning model 200 (step S404), and executes a calculation by the learning model 200 (step S405). That is, the control unit 101 inputs the particle diameter value to the node constituting the input layer 201 of the learning model 200, performs a calculation using weights and biases between the nodes in the intermediate layers 202A and 202B, and outputs the calculation result from the node in the output layer 203. Note that, in the initial stage before learning is started, initial values ​​are given to the definition information describing the learning model 200.

[0186] Next, the control unit 101 evaluates the calculation result obtained in step S405 (step S406) and determines whether learning is complete or not (step S407). Specifically, the control unit 101 can evaluate the calculation result using an error function (also called an objective function, loss function, or cost function) based on the calculation result obtained in step S405 and the teacher data. The control unit 101 determines that learning is complete when the error function becomes equal to or smaller than a threshold (or equal to or larger than a threshold) in the process of optimizing (minimizing or maximizing) the error function by a gradient descent method such as the steepest descent method. In order to avoid the problem of overlearning, a method such as cross-validation or early termination may be adopted to terminate learning at an appropriate timing.

[0187] When it is determined that the learning is not completed (S407: NO), the control unit 101 updates the weights and biases between the nodes of the learning model 200 (step S408), returns the process to step S403, and continues the learning using other teacher data. The control unit 101 can update the weights and biases between the nodes by using an error backpropagation method that sequentially updates the weights and biases between the nodes from the output layer 203 to the input layer 201 of the learning model 200.

[0188] If it is determined that the learning is complete (S407: YES), the control unit 101 stores the learned learning model 200 in the memory unit 102 in association with information regarding the type of powder raw material (step S409), and ends the processing according to this flowchart.

[0189] As described above, the control device 100 according to the eighth embodiment can generate a learning model 200 according to the type of powder raw material. The learning model 200 in the eighth embodiment is similar to that in the first embodiment, but instead of the learning model 200, the learning models 210 to 250 described in the second to seventh embodiments may be generated for each type of powder raw material.

[0190] In this embodiment, the control device 100 is configured to generate the learning model 200, but an external server (not shown) that generates the learning model 200 may be provided, and the external server may generate the learning model 200. In this case, the control device 100 may acquire the learning model 200 from the external server by communication or the like, and store the acquired learning model 200 in the memory unit 102.

[0191] 20 is a flowchart illustrating the control procedure by the control device 100. The control unit 101 of the control device 100 receives information on the type of powder raw material and an input of a particle diameter desired by the user via the operation unit 106 (step S421).

[0192] Next, the control unit 101 reads out the learning model 200 corresponding to the type of the received powder raw material from the storage unit 102 (step S422). The control unit 101 inputs the received particle size data to the input layer 201 of the learning model 200 read out in step S422, and executes a calculation by the learning model 200 (step S423). At this time, the control unit 101 provides the received particle size data to the nodes of the input layer 201, and executes a calculation by the intermediate layers 202A and 202B. The output layer 203 of the learning model 200 outputs the calculation results related to the control parameters that control the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate from the casing 410.

[0193] Next, the control unit 101 obtains the calculation result from the learning model 200 (step S424), and determines the control parameters to be used for control (step S425). The control unit 101 may determine the combination of the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge and suction flow rates based on the probability of being output as the calculation result.

[0194] Next, the control unit 101 executes control based on the control parameters determined in step S425 (step S426). That is, the control unit 101 generates control commands for the pulverizing motor 413M and the classifying motor 415M so that the rotation speeds of the pulverizing rotor 413 and the classifying rotor 415 become the values ​​determined in step S425, and outputs the control commands to the powder processing apparatus 4A through the output unit 104. The control unit 101 also generates a control command for the blower 7 so that the discharge / suction flow rate from the casing 410 becomes the value determined in step S425, and outputs the control command to the blower 7 through the output unit 104.

[0195] As described above, the control device 100 according to the present embodiment can determine the control parameters using the learning model 200 according to the type of powder raw material. The control device 100 can execute control based on the determined control parameters so as to obtain powder having a desired particle size.

[0196] In this embodiment, the control parameters are determined using a learning model 200 generated for each type of powder raw material. However, the learning models 210-250 described in embodiments 2-7 may be generated for each type of powder raw material, and the control parameters may be determined using the generated learning models 210-250.

[0197] In the eighth embodiment, an example of application to the mechanical powder processing apparatus 4A has been described, but it goes without saying that a similar technique can be applied to the powder processing apparatuses 4B to 4E.

[0198] (Embodiment 9) In the ninth embodiment, a re-learning procedure for the learning model 200 will be described. The overall configuration of the powder processing system 1 and the configuration of each device in the powder processing system 1 are similar to those in the first embodiment, and therefore the description thereof will be omitted.

[0199] FIG. 21 is a flow chart for explaining the re-learning procedure of the learning model 200. As described in the first embodiment, in the operation phase, the control unit 101 of the control device 100 receives an input of a particle diameter (desired particle diameter) desired by the user, and inputs the received particle diameter data to the learning model 200 to obtain a calculation result related to the control parameter. The control unit 101 controls the operation of each device constituting the powder processing system 1 based on the calculation result obtained from the learning model 200. In the operation phase, the control unit 101 may collect measurement data including the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate from the casing 410, and the particle diameter data of the powder obtained from the powder processing device 4A. The collected measurement data and particle diameter data are stored in the memory unit 102 together with a time stamp.

[0200] At an appropriate timing after the start of the operation phase, the control unit 101 compares the particle diameter indicated by the particle diameter data (i.e., the actual measured value of the particle diameter) with the user's desired particle diameter (step S501). When the input to the learning model 200 is only the value of D50, the control unit 101 may compare the value of D50 obtained as the actual measured value with the value of D50 input as the desired particle diameter. When the input to the learning model 200 is the values ​​of D10, D50, and D90 (or their upper and lower limits), the control unit 101 may compare the values ​​of D10, D50, and D90 obtained as the actual measured value with the values ​​of D10, D50, and D90 input as the desired particle diameter (or their upper and lower limits). The control unit 101 determines whether or not to perform re-learning based on the comparison result (step S502).

[0201] If the actual measured particle diameter is close to the desired particle diameter (for example, if the difference between the two is less than 10%), the control unit 101 determines not to execute re-learning (S502: NO) and ends the process according to this flowchart.

[0202] On the other hand, if the actual measured particle size is not close to the desired particle size (for example, if the difference between the two is 10% or more), the control unit 101 determines to execute re-learning (S502: YES).

[0203] When it is determined that re-learning is to be performed, the control unit 101 reads out data collected after the start of operation from the storage unit 102 (step S503), and selects teacher data (step S504). Note that the re-learning procedure from step S503 onward may be performed when the powder processing system 1 is not in operation.

[0204] The control unit 101 inputs the particle diameter values ​​included in the selected teacher data to the learning model 200 (step S505), and executes a calculation by the learning model 200 (step S506). That is, the control unit 101 inputs the particle diameter values ​​to the nodes constituting the input layer 201 of the learning model 200, performs a calculation using weights and biases between the nodes in the intermediate layers 202A and 202B, and outputs the calculation results from the nodes in the output layer 203.

[0205] Next, the control unit 101 evaluates the calculation result obtained by the calculation in step S506 (step S507), and determines whether learning is completed (step S508). Specifically, the control unit 101 can evaluate the calculation result using an error function (also called an objective function, loss function, or cost function) based on the calculation result obtained in step S506 and the teacher data.

[0206] When it is determined that the learning is not completed (S508: NO), the control unit 101 updates the weights and biases between the nodes of the learning model 200 (step S509), returns the process to step S504, and continues the learning using other teacher data. The control unit 101 can update the weights and biases between the nodes by using an error backpropagation method that sequentially updates the weights and biases between the nodes from the output layer 203 to the input layer 201 of the learning model 200.

[0207] When it is determined that the learning is completed (S508: YES), the control unit 101 stores the learning model 200 as a trained learning model in the storage unit 102 (step S510), and ends the process according to this flowchart.

[0208] As described above, the control device 100 according to this embodiment performs re-learning of the learning model 200 as necessary, thereby enabling the accuracy of powder processing by the powder processing system 1 to be improved even after the operation phase has started.

[0209] In addition, in embodiment 9, the re-learning procedure for the learning model 200 has been described, but it is also possible to re-learn the learning models 210 to 250 described in embodiments 2 to 7 and the learning model 200 for each type of powder raw material described in embodiment 8 using a similar procedure.

[0210] In the ninth embodiment, an application example to the mechanical powder processing apparatus 4A has been described, but it goes without saying that a similar method can be applied to the powder processing apparatuses 4B to 4E.

[0211] (Embodiment 10) In the tenth embodiment, a procedure for adjusting control parameters in the powder processing apparatus 4A will be described. The overall configuration of the powder processing system 1 and the configuration of each device in the powder processing system 1 are similar to those in the first embodiment, and therefore the description thereof will be omitted.

[0212] FIG. 22 is a flow chart for explaining the adjustment procedure of the control parameters. As described in the first embodiment, in the operation phase, the control unit 101 of the control device 100 receives the input of the particle diameter (desired particle diameter) desired by the user, and inputs the data of the received particle diameter to the learning model 200 to obtain the calculation result related to the control parameter. The control unit 101 controls the operation of each device constituting the powder processing system 1 based on the calculation result obtained from the learning model 200. In the operation phase, the control unit 101 may collect measurement data including the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate from the casing 410, and the particle diameter data of the powder obtained from the powder processing device 4A. The collected measurement data and particle diameter data are stored in the memory unit 102 together with a time stamp.

[0213] At an appropriate timing after the start of the operation phase, the control unit 101 compares the particle diameter indicated by the particle diameter data (i.e., the actual measured value of the particle diameter) with the user's desired particle diameter (step S601). When the input to the learning model 200 is only the value of D50, the control unit 101 may compare the value of D50 obtained as the actual measured value with the value of D50 input as the desired particle diameter. When the input to the learning model 200 is the values ​​of D10, D50, and D90 (or their upper and lower limits), the control unit 101 may compare the values ​​of D10, D50, and D90 obtained as the actual measured value with the values ​​of D10, D50, and D90 input as the desired particle diameter (or their upper and lower limits). The control unit 101 determines whether to adjust the control parameters based on the comparison result (step S602).

[0214] If the actual particle diameter is close to the desired particle diameter (for example, if the difference between the two is less than 10%), the control unit 101 determines not to adjust the control parameters (S602: NO) and ends the process according to this flowchart.

[0215] On the other hand, if the actual measured particle size is not close to the desired particle size (for example, if the difference between the two is 10% or more), the control unit 101 determines to adjust the control parameters (S602: YES).

[0216] When it is determined that the control parameters are to be adjusted, the control unit 101 adjusts the control parameters according to the comparison result of step S602 (step S603), and controls the operation of the apparatus including the powder processing device 4A based on the adjusted control parameters (step S604). For example, when the actual measured value of the particle diameter is smaller than the desired particle diameter, the control unit 101 adjusts the control parameters so that the rotation speed of the pulverizing rotor 413 and the classifying rotor 415 is lower than the current value and the discharge / suction flow rate is higher than the current value, and controls the operation of the apparatus including the powder processing device 4A based on the adjusted control parameters. Also, when the actual measured value of the particle diameter is larger than the desired particle diameter, the control unit 101 adjusts the control parameters so that the rotation speed of the pulverizing rotor 413 and the classifying rotor 415 is higher than the current value and the discharge / suction flow rate is lower than the current value, and controls the operation of the apparatus including the powder processing device 4A based on the adjusted control parameters.

[0217] As described above, in this embodiment, if a deviation of a certain amount or more occurs between the actual measured particle diameter and the user's desired particle diameter during operation of the powder processing system 1, the control parameters can be adjusted so that the deviation becomes smaller.

[0218] In the tenth embodiment, an example of application to the mechanical powder processing apparatus 4A has been described, but it goes without saying that a similar technique can be applied to the powder processing apparatuses 4B to 4E.

[0219] (Embodiment 11) In the eleventh embodiment, a configuration in which a terminal device 500 controls the operation of devices including the powder processing device 4A will be described. The overall configuration of the powder processing system 1 and the configuration of each device in the powder processing system 1 are similar to those in the first embodiment, and therefore the description thereof will be omitted.

[0220] 23 is a flowchart illustrating the procedure of the process executed by the terminal device 500 and the control device 100. The control unit 501 of the terminal device 500 accesses the control device 100 through the communication unit 503 (step S701).

[0221] When the control unit 101 of the control device 100 receives an access from the terminal device 500 while the powder processing device 4A is not operating, the control unit 101 generates screen data relating to an input screen for receiving an input of a particle diameter desired by the user (step S702), and transmits the generated screen data to the terminal device 500 via the communication unit 105 (step S703). Note that when the control unit 101 receives an access from the terminal device 500 while the powder processing device 4A is operating, the control unit 101 may execute the processes from step S703 onward, which will be described below.

[0222] The control unit 501 of the terminal device 500 receives screen data of the input screen transmitted from the control device 100 through the communication unit 503 (step S704). The control unit 501 displays the input screen on the display unit 505 based on the received screen data (step S705), and accepts input of a particle diameter desired by the user (step S706).

[0223] Fig. 24 is a schematic diagram showing an example of an input screen. This input screen shows a screen for accepting lower and upper limits for D10, D50, and D90. Instead of accepting lower and upper limits for D10, D50, and D90, the values ​​of D10, D50, and D90 desired by the user may be accepted, or only the value of D50 (median diameter) may be accepted. The control unit 501 transmits the particle diameter data accepted in step S706 to the control device 100 (step S707).

[0224] The control unit 101 of the control device 100 receives the particle diameter data transmitted from the terminal device 500 through the communication unit 105 (step S708). The control unit 101 inputs the received particle diameter data, for example, to the learning model 200, and executes a calculation by the learning model 200 (step S709). Next, the control unit 101 acquires the calculation result from the learning model 200 (step S710), and determines the control parameters based on the calculation result (step S711). The control unit 101 controls the operation of the device including the powder processing device 4A based on the determined control parameters (step S712).

[0225] During operation of the powder processing system 1, the control unit 101 constantly acquires measurement data on the particle size of the powder obtained from the powder processing device 4A, the rotation speeds of the grinding rotor 413 and the classifying rotor 415, and the discharge and suction flow rates from the casing 410 from the input unit 103. The control unit 101 generates screen data for a monitoring screen including at least one of the above measurement data and an overall view of the powder processing system 1 at an appropriate timing (step S713). The control unit 101 transmits the generated screen data to the terminal device 500 via the communication unit 105 (step S714).

[0226] The control unit 501 of the terminal device 500 receives the screen data transmitted from the control device 100 through the communication unit 503 (step S715). The control unit 501 displays a monitoring screen on the display unit 505 based on the received screen data (step S716). FIG. 25 is a schematic diagram showing an example of the monitoring screen. The example shown in FIG. 25 shows a state in which a monitoring screen including measurement data of particle diameter obtained from the powder processing device 4A (i.e., the actual measurement value of the particle diameter measured by the particle diameter sensor S2) and an overall view of the powder processing system 1 is displayed on the display unit 505. Instead of or together with the measurement data of the particle diameter, measurement data of the rotation speed of the grinding rotor 413, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate from the casing 410 may be displayed on the monitoring screen. Furthermore, the desired particle size input by the user, the rotation speed of the grinding rotor 413 set based on the calculation results of the learning model 200, the rotation speed of the classifying rotor 415, and the discharge / suction flow rate values ​​from the casing 410 may be displayed on the monitoring screen.

[0227] As described above, in this embodiment, the powder processing system 1 can be remotely operated using the terminal device 500, and the operating status of the powder processing system 1 can be monitored by the terminal device 500.

[0228] In the eleventh embodiment, an application example to the mechanical powder processing apparatus 4A has been described, but it goes without saying that a similar technique can be applied to the powder processing apparatuses 4B to 4E.

[0229] The embodiments disclosed herein are illustrative in all respects and should not be considered as limiting. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the scope and meaning equivalent to the claims.

[0230] For example, the present invention is not limited to the powder processing devices 4A to 4E described in embodiments 1 to 11, but can be applied to powder processing devices classified into impact mills, shear type mills, air flow mills, friction type mills, media agitation type mills, screen mills, bead mills, ball mills, pin mills, and jet mills (i.e., powder processing devices having at least a grinding function).

[0231] In addition, in the first to eleventh embodiments, the powder processing apparatus 4A to 4E is configured to introduce gas, but the apparatus may be configured to introduce liquid instead of gas. In this case, a sensor for measuring the flow rate of liquid may be used as the flow rate sensor S3. Also, a pump may be used instead of the blower 7. [Explanation of symbols]

[0232] 1...powder processing system, 2...raw material supply machine, 3...hot air generator, 4A to 4E...powder processing device, 5...cyclone, 6...dust collector, 7...blower, 410...casing, 411...raw material inlet, 412...gas inlet, 413...grinding rotor, 414...guide ring, 415...classifying rotor, 416...powder outlet, S1...weight sensor, S2...particle size sensor, S3...flow rate sensor, S4...temperature sensor, S5...rotational speed sensor, S6...rotational speed sensor, 100...control device, 101...control unit, 102...storage unit, 103...input unit, 104...output unit, 105...communication unit, 106...operation unit, 107...display unit, 500...terminal device, 501...control unit, 502...storage unit, 503...communication unit, 504...operation unit, 505...display unit

Claims

1. The computer With respect to a powder processing apparatus that is provided with a grinding rotor for grinding powder raw material for toner or battery material, and a classifying rotor for classifying the powder, and that produces powder for toner or battery material from the powder raw material, acquiring a plurality of sets of measurement data including a rotation speed of the grinding rotor, a rotation speed of the classifying rotor, and a discharge / suction flow rate when taking out powder from a treatment chamber of the powder processing apparatus as operating states of the powder processing apparatus, as well as particle size data of the powder and circularity data of the powder obtained from the powder processing apparatus; A learning model is generated that uses the acquired multiple sets of measurement data, particle size data, and circularity data as training data, and outputs calculation results for control parameters related to the rotation speed of the grinding rotor, the rotation speed of the classifying rotor, and the discharge / suction flow rate from the processing chamber in response to input of the particle size and circularity of the powder desired by the user. How to generate a learning model.

2. A learning model is generated by using, as training data, measurement data further including at least one of the supply amount of the powder raw material supplied to the treatment chamber of the powder treatment device and the temperature inside the treatment chamber, and outputting a calculation result related to a control parameter further including at least one of the supply amount and the temperature. A method for generating a learning model according to claim 1 .

3. the training data further includes at least one of data on the moisture content of the powder raw material, the driving power of the powder processing apparatus, the driving current of the powder processing apparatus, the heat medium temperature, a medium weight indicating a weight of a medium that is stirred in the processing chamber to pulverize the powder raw material, an environmental temperature, and an environmental humidity; A learning model is generated that outputs a calculation result regarding the control parameters in response to an input of a particle size and powder circularity desired by a user and at least one of the moisture content, the driving power or the driving current, the heat medium temperature, the medium weight, the environmental temperature, and the environmental humidity. A method for generating a learning model according to claim 1 or 2.

4. Reacquire the measurement data, particle size data, and circularity data used as training data; The learning model is retrained using the reacquired measurement data, the particle size data, and the circularity data as training data. A method for generating a learning model according to any one of claims 1 to 3.

5. A learning model that outputs a calculation result related to the control parameters in response to the input of the particle size and circularity desired by the user is generated for each type of powder raw material. A method for generating a learning model according to any one of claims 1 to 4.

6. The fluid flowing in the system including the powder processing device is a gas or a liquid. A method for generating a learning model according to any one of claims 1 to 5.

7. On the computer, With respect to a powder processing apparatus that is provided with a grinding rotor for grinding powder raw material for toner or battery material, and a classifying rotor for classifying the powder, and that produces powder for toner or battery material from the powder raw material, acquiring a plurality of sets of measurement data including a rotation speed of the grinding rotor, a rotation speed of the classifying rotor, and a discharge / suction flow rate when taking out powder from a treatment chamber of the powder processing apparatus as operating states of the powder processing apparatus, as well as particle size data of the powder and circularity data of the powder obtained from the powder processing apparatus; A learning model is generated that uses the acquired multiple sets of measurement data, particle size data, and circularity data as training data, and outputs calculation results for control parameters related to the rotation speed of the grinding rotor, the rotation speed of the classifying rotor, and the discharge / suction flow rate from the processing chamber in response to input of the particle size and circularity of the powder desired by the user. A computer program for executing a process.

8. A computer comprising: an input layer in which the particle size and powder circularity desired by the user are input; an output layer that outputs a calculation result regarding a control parameter that controls an operating state of a powder processing apparatus that is provided with a grinding rotor that grinds powder raw material for toner or battery material and a classifying rotor that classifies the powder, and that produces powder for toner or battery material from the powder raw material; an intermediate layer that learns the relationship between the particle size and the circularity and the control parameters by using training data including a plurality of sets of measurement data including the rotation speed of the grinding rotor, the rotation speed of the classifying rotor, and the discharge and suction flow rate when the powder is taken out from the treatment chamber of the powder treatment device as the operating state of the powder treatment device, and the particle size data and the circularity data of the powder obtained from the powder treatment device; A learning model for functioning as When a particle size and a circularity of a powder desired by a user are input to the input layer, the intermediate layer performs a calculation and causes the computer to function to output a calculation result regarding a control parameter for controlling an operating state of the powder processing device. Learning model.

9. A receiving unit that receives an input of a particle size and a circularity of a powder desired by a user; a learning model that learns the relationship between the particle size and circularity of the powder and control parameters that control the operating state of the powder processing device, using teacher data including a plurality of sets of measurement data including the rotation speed of the grinding rotor, the rotation speed of the classifying rotor, and the discharge and suction flow rates when the powder is taken out from a processing chamber of the powder processing device, the teacher data including a plurality of sets of measurement data including the rotation speed of the grinding rotor, the rotation speed of the classifying rotor, and the discharge and suction flow rates when the powder is taken out from a processing chamber of the powder processing device; a calculation processing unit that inputs the particle size and circularity received by the reception unit into the learning model and executes a calculation using the learning model; A control unit that controls the operation of an apparatus including the powder processing apparatus based on a calculation result by the learning model; A control device comprising:

10. the calculation processing unit inputs the particle size and the circularity of the powder received by the receiving unit into a learning model that has learned the relationship using measurement data that further includes at least one of an amount of the powder raw material supplied to a treatment chamber of the powder processing device and a temperature inside the treatment chamber, thereby executing a calculation using the learning model; The control unit controls the operation of the apparatus including the powder processing apparatus based on the calculation result by the learning model. The control device according to claim 9.

11. the receiving unit receives at least one of a particle size and powder circularity desired by a user, a moisture content of the powder raw material, a driving power of the powder processing device, a driving current of the powder processing device, a heat medium temperature, a medium weight indicating a weight of a medium that is stirred in the processing chamber to pulverize the powder raw material, an environmental temperature, and an environmental humidity; the calculation processing unit executes a calculation using the learning model by inputting the value received by the reception unit into a learning model that has learned the relationship using teacher data that further includes at least one of data on the moisture content, the driving power or the driving current, the heat medium temperature, the medium weight, the environmental temperature, and the environmental humidity; The control unit controls the operation of the apparatus including the powder processing apparatus based on the calculation result by the learning model. The control device according to claim 9 or 10.

12. The receiving unit receives a type of powder raw material, the calculation processing unit selects a learning model corresponding to the type accepted by the accepting unit from a plurality of learning models in which the relationship is learned for each type of the powder raw material, and executes a calculation using the learning model by inputting values ​​including a particle diameter and a circularity of the powder desired by a user into the selected learning model; The control unit controls the operation of the apparatus including the powder processing apparatus based on the calculation result by the learning model. A control device according to any one of claims 9 to 11.

13. An adjustment unit that compares the particle diameter of the powder desired by the user with the particle diameter data of the powder obtained from the powder processing device, and adjusts the control parameters according to the comparison result so that the particle diameter of the powder obtained from the powder processing device approaches the particle diameter of the powder desired by the user. Equipped with The control unit controls the operation of the apparatus including the powder processing apparatus based on the adjusted control parameters. A control device according to any one of claims 9 to 12.

14. a screen generating unit that generates screen data including at least one of the measurement data, the particle diameter data of the powder obtained from the powder processing apparatus, the particle diameter desired by the user received by the receiving unit, and an overall view of the powder processing apparatus; A transmission unit that transmits the generated screen data to an external terminal; The control device according to any one of claims 9 to 13, comprising:

15. The computer The user inputs the particle size and powder circularity desired by the user; the received particle size and circularity are input into a learning model that has learned the relationship between the particle size and circularity of the powder and control parameters for controlling the operating state of the powder processing apparatus, using teacher data including a plurality of sets of powder particle size data and powder circularity data obtained from a powder processing apparatus that is provided with a grinding rotor for grinding powder raw material for toner or battery material and a classifying rotor for classifying the powder, and measurement data including the rotation speed of the grinding rotor, the rotation speed of the classifying rotor, and the discharge / suction flow rates when removing the powder from a processing chamber of the powder processing apparatus as operating states of the powder processing apparatus; Executing a calculation based on the learning model; Based on the calculation results obtained from the learning model, the operation of the device including the powder processing device is controlled. Control methods.

16. inputting the received particle size and circularity into a learning model that has learned the relationship using measurement data that further includes at least one of an amount of powder raw material supplied to a treatment chamber of the powder processing device and a temperature inside the treatment chamber; Based on the calculation results obtained from the learning model, the operation of the device including the powder processing device is controlled. The control method according to claim 15.

17. A particle size and powder circularity desired by a user are received, along with at least one of the moisture content of the powder raw material, the driving power of the powder processing device, the driving current of the powder processing device, the heat medium temperature, a medium weight indicating the weight of a medium that is stirred in the processing chamber to pulverize the powder raw material, an environmental temperature, and an environmental humidity; inputting the received data into a learning model that has learned the relationship using teacher data that further includes at least one of the moisture content, the driving power or the driving current, the heat medium temperature, the medium weight, the environmental temperature, and the environmental humidity; Based on the calculation results obtained from the learning model, the operation of the device including the powder processing device is controlled. A control method according to claim 15 or 16.

18. Accept the type of powdered raw material, Selecting a learning model corresponding to the received type from a plurality of learning models that have learned the relationship for each type of powder raw material; By inputting the received particle size and circularity into a selected learning model, a calculation is performed using the learning model; Based on the calculation results obtained from the learning model, the operation of the device including the powder processing device is controlled. A control method according to any one of claims 15 to 17.

19. Comparing the particle size of the powder desired by the user with the particle size data of the powder obtained from the powder processing device; According to the comparison result, the control parameters are adjusted so that the particle diameter of the powder obtained from the powder processing device approaches the particle diameter of the powder desired by the user; Based on the adjusted control parameters, the operation of the apparatus including the powder processing apparatus is controlled. A control method according to any one of claims 15 to 18.

20. generating screen data including at least one of the measurement data in the powder processing apparatus, the particle size data of the powder obtained from the powder processing apparatus, and an overall view of the powder processing apparatus; Send the generated screen data to an external device A control method according to any one of claims 15 to 19.

21. The apparatus including the powder processing device includes the powder processing device and at least one of a raw material supplying device, a hot air generator, a cyclone, a dust collector, and a blower or a pump connected to the powder processing device. A control method according to any one of claims 15 to 20.

22. On the computer, The user inputs the particle size and powder circularity desired by the user; the received particle size and circularity are input to a learning model that has learned the relationship between the particle size and circularity of the powder and control parameters for controlling the operating state of the powder processing device, using teacher data including a plurality of sets of powder particle size data and powder circularity data obtained from a powder processing device that is provided with a grinding rotor for grinding powder raw material for toner or battery material and a classifying rotor for classifying the powder, and measurement data including the rotation speed of the grinding rotor, the rotation speed of the classifying rotor, and the discharge / suction flow rates when the powder is taken out from a processing chamber of the powder processing device, as operating states of the powder processing device, thereby executing a calculation using the learning model; Based on the calculation result by the learning model, the operation of the device including the powder processing device is controlled. A computer program for executing a process.

Citation Information

Patent Citations

  • Control method of roller mill

    JP1994327990A

  • Vibration estimation device of crusher and control device based on vibration estimation of crusher

    JP1997141116A

  • Crushing and pulverizing treatment system

    JP1999104511A

  • Adaptation control method and apparatus in powdered coal mill

    JP2002361116A

  • Method and apparatus for operating stone crushing plant

    JP2003200079A