WO2022018975A1

WO2022018975A1 - Machine-learning device, data processing system, and machine-learning method

Info

Publication number: WO2022018975A1
Application number: PCT/JP2021/020912
Authority: WO
Inventors: 路明杉浦; 徹也荻原
Original assignee: 巴工業株式会社
Priority date: 2020-07-21
Filing date: 2021-06-01
Publication date: 2022-01-27
Also published as: JP2022021244A

Abstract

A machine-learning device according to the present disclosure is for a centrifugal separation system, the machine-learning device comprising: a learning data set storage unit that stores a plurality of packs of learning data sets including input data including image data obtained by capturing, from a prescribed view angle, a liquid-containing solid substance discharged from a bowl and output data which is associated with the input data and includes control parameters including at least one among supply amounts of additives added to a liquid to be treated and a differential speed controlled by a centrifugal force of the bowl and a differential speed generation device; a training unit that trains a learning model that infers a correlation between the input data and the output data by inputting the plurality of packs of the learning data sets; and a trained model storage unit that stores the trained model for which the training has been completed.

Description

Machine learning equipment, data processing systems and machine learning methods

This disclosure relates to machine learning devices, data processing systems and machine learning methods.

Centrifugal separation devices that perform solid-liquid separation using centrifugal force have been conventionally used in water treatment equipment for water and sewage, industrial wastewater, or urine. Various types of centrifuges are known, but a centrifuge called a decanter is widely used, especially in equipment that requires continuous processing for a long time.

This decanter generally has a drive motor, a bowl that is rotated by the drive motor and the liquid to be processed is charged inside, a screw conveyor that is coaxially arranged in the bowl, and the rotation speed of the bowl and the rotation of the screw conveyor. It may include a differential speed generator that generates a differential speed between the speed and the speed. In such a decanter, the solid component in the liquid to be treated charged into the bowl is discharged from one end side of the bowl, and the liquid component in the liquid to be treated is discharged from the other end side of the bowl. As a result, solid-liquid separation is realized. Then, various suitable motion control methods for this decanter have been studied. (For example, refer to Japanese Patent No. 5442099.)

According to the method described in Japanese Patent No. 5442099, the differential speed generated by the differential speed generator and the centrifugal force of the bowl are controlled based on the transport torque of the screw conveyor, so that the liquid to be treated is controlled. It is possible to partially realize the automatic operation of the centrifuge according to the properties of. However, there are various information and parameters related to the control of the centrifuge system including such a centrifuge device other than the transfer torque of the screw conveyor, the differential speed generated by the differential speed generator, and the centrifugal force of the bowl. In particular, there are a wide variety of information (specifically, state variables) that affect the control of the controlled object, and such information may include information that is difficult to quantify. A method for effectively and mechanically considering these state variables has not been established at this time, and as a result, such state variables are centrifuged based on the experience and tacit knowledge of an engineer (operator). The reality is that it is reflected in the operation control of the system.

In view of the above points, it is an object of the present disclosure to provide a machine learning device, a data processing system and a machine learning method for realizing suitable operation control of a centrifugal separation system without depending on the judgment of an operator. ..

In order to achieve the above object, the machine learning device 20 according to the first aspect of the present disclosure has centrifugal force on the liquids to be treated PL1 (not shown), PL2 (see FIG. 15 and the like), for example, as shown in FIG. A bowl 2 for centrifuging the solid matter M and the separation liquid SL (not shown), a screw conveyor 3 for transporting the solid matter M in the bowl 2 toward the discharge port 2a, and the bowl 2 For the centrifugal separation system 1 including a drive motor 4 for rotating the screw conveyor 3 and a differential speed generator 5 for rotating the screw conveyor 3 with a relative speed difference from the bowl 2, the discharge port 2a. A plurality of sets of training data sets including input data including image data obtained by capturing the liquid-containing solid material M discharged from the above from a predetermined angle and output data including control parameters associated with the input data are stored. In the learning data set storage unit 22, the control parameters are the supply amount of the additive added to the liquid to be treated PL1, the centrifugal force of the bowl 2, and the difference controlled by the differential speed generator 5. With the training data set storage unit 22 including at least one of the speeds; a learning model for inferring the correlation between the input data and the output data by inputting a plurality of sets of the training data sets. It includes a learning unit 23 for learning; and a trained model storage unit 24 for storing the learning model learned by the learning unit 23.

With this configuration, it is possible to provide a trained model that can infer the control parameters of the centrifugal separation system from the image data of the separation solution.

The machine learning device 20A according to the second aspect of the present disclosure is, for example, as shown in FIG. 9, in the machine learning device according to the first aspect of the present disclosure, the learning unit is in the learning data set. The first learning unit 231 that learns the first learning model that infers the feature amount of the image data by inputting the image data; and the control parameter is inferred by inputting the feature amount of the image data. Includes a second learning unit 232 and; to learn the second learning model.

With this configuration, when estimating the control parameters of the centrifugation system from the image data of the dehydrated solid, the learning unit is divided into two to obtain the trained model generated in each learning unit. The number of required training data sets can be relatively reduced.

The machine learning device 20B according to the third aspect of the present disclosure is, for example, as shown in FIG. 11, in the machine learning device according to the first aspect of the present disclosure, the input data is the concentration of the separation liquid SL. Further includes at least one of the slurry concentration of the liquid to be treated PL1 and the torque value of the screw conveyor 3.

With this configuration, the number of input data increases, and the number of training data sets required to obtain the trained model generated in the training unit can be relatively suppressed.

The machine learning device 20C according to the fourth aspect of the present disclosure is, for example, as shown in FIG. 13, in the machine learning device according to the first aspect of the present disclosure, the input data is the concentration of the separation liquid SL. The learning unit further includes at least one of the slurry concentration of the liquid to be treated PL1 and the torque value of the screw conveyor 3, and the learning unit inputs the image data in the learning data set. A first learning unit 231 that learns a first learning model for inferring a feature amount of image data; a feature amount of the image data, a concentration of the separation liquid SL, and a slurry concentration of the liquid to be treated PL1. It includes a fourth learning unit 234 that learns a fourth learning model for inferring the control parameters by inputting a torque value of the screw conveyor 3.

With this configuration, when estimating the control parameters of the centrifugation system from the image data of the dehydrated solid, the learning unit is divided into two, and the learning data set to be learned in one of the learning units is input. By increasing the number of data, it is possible to relatively reduce the number of training data sets required to obtain the trained model generated in each training unit.

The machine learning device 120 according to the fifth aspect of the present disclosure is, for example, as shown in FIG. 15, a bowl 2 that applies centrifugal force to the liquids PL1 and PL2 to be treated to centrifuge the solid M and the separation liquid SL. A screw conveyor 3 that conveys the solid material M in the bowl 2 toward the discharge port 2a, a drive motor 4 that rotates the bowl 2, and a differential speed of the screw conveyor 3 relative to the bowl 2. The first image data obtained by capturing the liquid-containing solid material M discharged from the discharge port 2a from a predetermined angle, which is for the centrifugal separation system 1 including the differential speed generator 5 to be rotated by The input data including the second image data obtained by capturing the liquid to be treated PL2 from a predetermined angle of view before being supplied to the bowl 2 and after the predetermined additive is added, and the input data. A training data set storage unit 225 that stores a plurality of sets of training data sets including output data including control parameters associated with the control parameters, wherein the control parameters are the supply amount of the additive and the bowl 2. With the training data set storage unit 225 containing at least one of the centrifugal force and the differential speed controlled by the differential speed generator 5; the input by inputting a plurality of sets of the learning data set. It includes a learning unit 235 that learns a learning model that infers the correlation between the data and the output data; and a trained model storage unit 124 that stores the learning model learned by the learning unit 235.

With this configuration, it is possible to provide a trained model that can infer the control parameters of the centrifugation system from the image data of the dehydrated solid matter and the image data of the liquid to be treated containing flocs.

The machine learning device 120A according to the sixth aspect of the present disclosure is, for example, as shown in FIG. 19, in the machine learning device according to the fifth aspect of the present disclosure, the learning unit is in the learning data set. A first learning unit 231 that learns a first learning model that infers a feature amount of the first image data by inputting the first image data; and a second image in the training data set. A sixth learning unit 236 that learns a sixth learning model that infers the feature amount of the second image data by inputting data; the feature amount of the first image data and the second image data. It includes a seventh learning unit 237 that learns a seventh learning model that infers the control parameter by inputting a feature amount of image data.

With this configuration, when the control parameters of the centrifugation system are estimated from the image data of the dehydrated solid matter and the image data of the liquid to be treated containing flocs, the learning unit is divided into three in each learning unit. The number of training data sets required to obtain the generated trained model can be relatively reduced.

The machine learning device 120B according to the seventh aspect of the present disclosure is, for example, as shown in FIG. 21, in the machine learning device according to the fifth aspect of the present disclosure, the input data is the concentration of the separation liquid SL. Further includes at least one of the slurry concentration of the liquid to be treated PL1 and the torque value of the screw conveyor 3.

The machine learning device 120C according to the eighth aspect of the present disclosure is, for example, as shown in FIG. 23, in the machine learning device according to the fifth aspect of the present disclosure, the input data is the concentration of the separation liquid SL. The learning unit further includes at least one of the slurry concentration of the liquid to be treated PL1 and the torque value of the screw conveyor 3, and the learning unit inputs the first image data in the learning data set. With the first learning unit 231 learning the first learning model for inferring the feature amount of the first image data; by inputting the second image data in the learning data set, the first learning unit A sixth learning unit 236 that learns a sixth learning model that infers the feature amount of the second image data; the feature amount of the first image data, the feature amount of the second image data, and the separation. A ninth learning unit 239 that learns a ninth learning model for inferring the control parameters by inputting the concentration of the liquid SL, the slurry concentration of the liquid to be treated PL1, and the torque value of the screw conveyor 3. And;

With this configuration, the learning unit is divided into three and one of the learning units is used when estimating the control parameters of the centrifugation system from the image data of the dehydrated solid matter and the image data of the liquid to be treated containing flocs. By increasing the number of input data of the training data set trained in, the number of training data sets required to obtain the trained model generated in each training unit can be relatively suppressed.

The machine learning device according to the ninth aspect of the present disclosure is the machine learning device according to the first to eighth aspects of the present disclosure, and the control parameter further includes the supply amount of the liquid to be treated.

With this configuration, finer parameter adjustment can be realized by increasing the number of control parameters output by the trained model.

The machine learning device according to the tenth aspect of the present disclosure is the machine learning device according to the first to ninth aspects of the present disclosure, and the control parameter further includes the dam set diameter of the bowl.

In the data processing systems 80 and 80A according to the eleventh aspect of the present disclosure, for example, as shown in FIGS. 7 and 10, centrifugal force is applied to the liquids PL1 and PL2 to be treated into the solid material M and the separation liquid SL. The bowl 2 for centrifugal separation, the screw conveyor 3 for transporting the solid matter M in the bowl 2 toward the discharge port 2a, the drive motor 4 for rotating the bowl 2, and the screw conveyor 3 for the bowl 2. A first image of a liquid-containing solid material M discharged from a discharge port 2a, which is used in a centrifugal separation system 1 including a differential speed generator 5 that rotates with a relative difference speed, from a predetermined angle. The first image data acquisition unit 81 for acquiring the image data of 1; and the trained model generated by the machine learning devices 20 and 20A according to the first or second aspect, the first image data acquisition unit. It includes

inference units

87, 871, 872 and; which infer the control parameters of the centrifugal separation system by inputting the data acquired by 81.

With this configuration, control parameters that can realize suitable operation control of the centrifugal separation system can be inferred based on the first image data obtained by imaging the dehydrated solid matter, so that the centrifugal separation does not depend on the judgment of the operator. It will be possible to automatically control the operation of the system.

In the data processing systems 80B and 80C according to the twelfth aspect of the present disclosure, for example, as shown in FIGS. 12 and 14, centrifugal force is applied to the liquids PL1 and PL2 to be treated to form a solid M and a separation liquid SL. The bowl 2 for centrifugal separation, the screw conveyor 3 for transporting the solid matter M in the bowl 2 toward the discharge port 2a, the drive motor 4 for rotating the bowl 2, and the screw conveyor 3 for the bowl 2. A first image of a liquid-containing solid material M discharged from a discharge port 2a, which is used in a centrifugal separation system 1 including a differential speed generator 5 that rotates with a relative difference speed, from a predetermined angle. The first image data acquisition unit 81 for acquiring the image data of 1; at least one of the concentration of the separation liquid SL, the slurry concentration of the liquid to be processed PL1, and the torque value of the screw conveyor 3. The first image data acquisition unit 81 and the additional variable acquisition are added to the trained model generated by the machine learning devices 20B and 20C according to the third or fourth aspect. It includes

inference units

873, 871, 874 and; which infer the control parameters of the centrifugal separation system by inputting the data acquired by the unit 89.

With this configuration, centrifugation is performed based on the concentration of the separation liquid, the slurry concentration of the liquid to be treated, at least one of the torque values of the screw conveyor, and the first image data obtained by imaging the dehydrated solid. Since the control parameters that can realize the suitable operation control of the system can be inferred, the operation control of the centrifugal separation system can be automatically performed without depending on the judgment of the operator.

In the data processing systems 180 and 180A according to the thirteenth aspect of the present disclosure, for example, as shown in FIGS. 17 and 20, centrifugal force is applied to the liquids PL1 and PL2 to be treated into the solid material M and the separation liquid SL. The bowl 2 for centrifugation, the screw conveyor 3 for transporting the solid material M in the bowl 2 toward the discharge port 2a, the drive motor 4 for rotating the bowl 2, and the screw conveyor 3 for the bowl 2. A first image of a liquid-containing solid material M discharged from the discharge port 2a, which is used in a centrifugal separation system 1 including a differential speed generator 5 that rotates with a relative differential speed, from a predetermined angle. The first image data acquisition unit 81 for acquiring the image data of 1; and the liquid to be treated PL2 before being supplied to the bowl 2 and after the predetermined additive is added have a predetermined angle of view. The first image data acquisition unit 181 for acquiring the second image data imaged from the above; and the trained model generated by the machine learning devices 120, 120A according to the fifth or sixth aspect. Including

inference units

875, 871, 876, 877 that infer the control parameters of the centrifugal separation system 1 by inputting the data acquired by the image data acquisition unit 81 and the second image data acquisition unit 181; It's a waste.

With this configuration, the operator can infer control parameters that can realize suitable operation control of the centrifugation system based on the first and second image data obtained by imaging the dehydrated solid matter and the floc-containing liquid to be treated. It will be possible to automatically control the operation of the centrifuge system without depending on the judgment of.

In the data processing systems 180B and 180C according to the 14th aspect of the present disclosure, for example, as shown in FIGS. 22 and 24, centrifugal force is applied to the liquids PL1 and PL2 to be treated to form a solid M and a separation liquid SL. The bowl 2 for centrifugation, the screw conveyor 3 for transporting the solid material M in the bowl 2 toward the discharge port 2a, the drive motor 4 for rotating the bowl 2, and the screw conveyor 3 for the bowl 2. A first image of a liquid-containing solid material M discharged from the discharge port 2a, which is used in a centrifugal separation system 1 including a differential speed generator 5 that rotates with a relative differential speed, from a predetermined angle. The first image data acquisition unit 81 for acquiring the image data of 1; and the liquid to be treated PL2 before being supplied to the bowl 2 and after the predetermined additive is added have a predetermined angle of view. With the second image data acquisition unit 181 for acquiring the second image data imaged from the above; the concentration of the separation liquid SL, the slurry concentration of the liquid to be processed PL1, and the torque value of the screw conveyor 3. An additional variable acquisition unit 89 for acquiring at least one of them; a trained model generated by the machine learning devices 120B and 120C according to the seventh or eighth aspect, and the first image data acquisition unit 81. The

inference units

878, 871, 876, 879 that infer the control parameters of the centrifugal separation system 1 by inputting the data acquired by the second image data acquisition unit 181 and the additional variable acquisition unit 89; It includes.

With this configuration, the concentration of the separated liquid, the slurry concentration of the liquid to be treated, at least one of the torque values of the screw conveyor, and the first and second images of the dehydrated solid and the liquid to be treated containing flocs are imaged. Since control parameters that can realize suitable operation control of the centrifugation system can be inferred based on the image data, the operation control of the centrifugation system can be automatically performed without depending on the judgment of the operator. Become.

The machine learning method according to the fifteenth aspect of the present disclosure is carried out by a computer, for example, as shown in FIGS. 3 and 6, and a solid substance is subjected to centrifugal force to the liquids PL1 and PL2 to be treated. A bowl 2 for centrifuging the M and the separation liquid SL, a screw conveyor 3 for transporting the solid material M in the bowl 2 toward the discharge port 2a, a drive motor 4 for rotating the bowl 2, and the screw. The liquid-containing solid material M discharged from the discharge port 2a is for a centrifugal separation system 1 including a differential speed generator 5 for rotating the conveyor 3 with a relative speed difference from the bowl 2. In step S11, a plurality of sets of learning data sets including input data including image data captured from a predetermined angle of view and output data including control parameters associated with the input data are stored, and the control parameters are , The step comprising at least one of the supply of additives added to the liquid PL1 to be treated, the centrifugal force of the bowl 2, and the differential speed controlled by the differential speed generator 5. It includes a step S15 for learning a learning model for inferring the correlation between the input data and the output data by inputting a plurality of sets of data sets for use; and a step S17 for storing the learned learning model. ..

With this configuration, it is possible to provide a trained model that can infer the control parameters of the centrifugation system from the image data of the dehydrated solid matter.

The machine learning method according to the 16th aspect of the present disclosure is carried out by a computer, for example, as shown in FIGS. 6 and 15, and is solidified by applying centrifugal force to the liquids PL1 and PL2 to be treated. A bowl 2 for centrifuging the object M and the separation liquid SL, a screw conveyor 3 for transporting the solid substance M in the bowl 2 toward the discharge port 2a, a drive motor 4 for rotating the bowl 2, and the above. The liquid-containing solid material M discharged from the discharge port 2a is for a centrifugal separation system 1 including a differential speed generator 5 for rotating the screw conveyor 3 with a differential speed relative to the bowl 2. The first image data captured from a predetermined angle of view and the second image of the liquid to be treated PL2 before being supplied to the bowl 2 and after the predetermined additive is added are imaged from a predetermined angle of view. In step S11, a plurality of sets of training data sets including input data including the image data of the above and output data including control parameters associated with the input data are stored, and the control parameters are the additives. By inputting a plurality of sets of the training data set with the step including at least one of the supply amount of the bowl 2, the centrifugal force of the bowl 2, and the differential speed controlled by the differential speed generator 5. It includes a step S15 for learning a learning model for inferring the correlation between the input data and the output data; and a step S17 for storing the learned learning model.

According to the present disclosure, it becomes possible to mechanically acquire control parameters that can realize suitable operation control of the centrifugal separation system without depending on the judgment of the operator.

It is a schematic diagram which shows an example of the schematic structure of the decanter main body which concerns on 1st Embodiment of this disclosure. It is a schematic diagram which shows an example of the centrifugal separation system including the piping structure of the decanter which concerns on 1st Embodiment of this disclosure. It is a schematic block diagram which shows an example of the machine learning apparatus which concerns on 1st Embodiment of this disclosure. An example of image data obtained by the dehydrated solid matter monitoring system according to the first embodiment of the present disclosure is shown. An example of image data obtained by the dehydrated solid matter monitoring system according to the first embodiment of the present disclosure is shown. It is a figure which shows the example of the neural network model for supervised learning carried out in the machine learning apparatus which concerns on 1st Embodiment of this disclosure. It is a flowchart which shows the example of the machine learning method which concerns on 1st Embodiment of this disclosure. It is a schematic block diagram which shows an example of the data processing system which concerns on 1st Embodiment of this disclosure. It is a flowchart which shows the example of the parameter adjustment by the data processing system which concerns on 1st Embodiment of this disclosure. It is a schematic block diagram which shows an example of the machine learning apparatus which concerns on the 2nd Embodiment of this disclosure. It is a schematic block diagram which shows an example of the data processing system which concerns on the 2nd Embodiment of this disclosure. It is a schematic block diagram which shows an example of the machine learning apparatus which concerns on 3rd Embodiment of this disclosure. It is a schematic block diagram which shows an example of the data processing system which concerns on 3rd Embodiment of this disclosure. It is a schematic block diagram which shows an example of the machine learning apparatus which concerns on 4th Embodiment of this disclosure. It is a schematic block diagram which shows an example of the data processing system which concerns on 4th Embodiment of this disclosure. It is a schematic block diagram which shows an example of the machine learning apparatus which concerns on 5th Embodiment of this disclosure. An example of image data obtained by the liquid to be treated monitoring system according to the fifth embodiment of the present disclosure is shown. An example of image data obtained by the liquid to be treated monitoring system according to the fifth embodiment of the present disclosure is shown. It is a schematic block diagram which shows an example of the data processing system which concerns on 5th Embodiment of this disclosure. It is a flowchart which shows the example of the parameter adjustment by the data processing system which concerns on 5th Embodiment of this disclosure. It is a schematic block diagram which shows an example of the machine learning apparatus which concerns on 6th Embodiment of this disclosure. It is a schematic block diagram which shows an example of the data processing system which concerns on 6th Embodiment of this disclosure. It is a schematic block diagram which shows an example of the machine learning apparatus which concerns on 7th Embodiment of this disclosure. It is a schematic block diagram which shows an example of the data processing system which concerns on 7th Embodiment of this disclosure. It is a schematic block diagram which shows an example of the machine learning apparatus which concerns on 8th Embodiment of this disclosure. It is a schematic block diagram which shows an example of the data processing system which concerns on 8th Embodiment of this disclosure.

This application is based on Japanese Patent Application No. 2020-124727 filed on July 21, 2020 in Japan, the contents of which form part of the content of this application.
Also, the present invention will be more fully understood by the following detailed description. Further scope of application of the present application will be clarified by the following detailed description. However, detailed description and specific examples are preferred embodiments of the present invention and are provided only for purposes of explanation. From this detailed description, various changes and modifications will be apparent to those skilled in the art within the spirit and scope of the present invention.
The applicant has no intention of presenting any of the described embodiments to the public, and is equivalent to any disclosed modifications or alternatives that may not be literally included in the claims. It is a part of the invention under the argument.

Hereinafter, each embodiment for carrying out the present disclosure will be described with reference to the drawings. In the following, the scope necessary for the explanation to achieve the object of the present disclosure will be schematically shown, and the scope necessary for the explanation of the relevant part of the present disclosure will be mainly explained. It shall be based on known technology.

<First Embodiment>
Before explaining the machine learning device, the data processing system, and the machine learning method according to the first embodiment of the present disclosure, the centrifuge system to which these machine learning device, the data processing system, and the machine learning method are applied. I will explain briefly. As the centrifuge system according to the present embodiment, a centrifuge device including a horizontal decanter 1 is used. The specific embodiment of the centrifugal separation system according to the present disclosure is not limited to those shown below, and may also be applied to a centrifugal separation system including a centrifugal separation device such as a vertical type or straight body type decanter. Applicable.

FIG. 1 is a schematic diagram showing a schematic structure of a decanter main body according to the first embodiment of the present disclosure. Further, FIG. 2 is a schematic diagram showing a centrifugal separation system including a decanter piping structure according to the first embodiment of the present disclosure. As shown in FIGS. 1 and 2, the horizontal decanter 1 shown in the present embodiment mainly includes a bowl 2, a screw conveyor 3, a drive motor 4, a differential speed generator 5, and a casing 6. be able to.

The bowl 2 can be made of a cylindrical member whose one end is processed into a weight shape, and can be rotatably supported around a horizontal axis. Further, one or more solid matter discharge ports 2a are provided at one end of the bowl 2 processed into a weight shape, and one or more separation liquid discharge ports 2b are formed at the other end of the dam 2c. It may be attached. The solid matter component contained in the liquid to be treated to be charged into the bowl 2 is mainly discharged from the solid matter discharge port 2a, and the liquid to be treated which is also charged into the bowl 2 is discharged from the separation liquid discharge port 2b. The contained liquid component is mainly discharged. Further, a pulley 4a for transmitting power from a drive motor 4, which will be described later, may be attached to one end of the bowl 2 (one end processed into a weight in FIG. 1). In the solid matter component discharged from the solid matter discharge port 2a, most of the liquid component in the liquid to be treated is dehydrated and removed, and the liquid component is discharged in a state of containing a relatively small predetermined amount of the liquid (separation liquid) component. Therefore, in the following, this solid component is referred to as a liquid-containing solid (or dehydrated solid) M.

The screw conveyor 3 can be composed of a member that is coaxially arranged in the bowl 2 and has a spiral screw blade 3a formed around the screw conveyor 3. The screw blade 3a may be a member for transporting and / or squeezing the solid matter in the bowl 2. The body 3b of the screw conveyor 3 is provided with a space 3c for receiving the liquid to be treated and a discharge port 3d for charging the liquid to be treated stored in this space into the bowl 2. You may be. Further, a differential speed generator 5, which will be described later, may be connected to one end of the screw conveyor 3.

The drive motor 4 may be a motor for applying a rotational force to the bowl 2, and may be connected to a pulley 4a attached to one end of the bowl 2 via a belt 4b. For the drive motor 4, it is preferable that a relatively large motor is adopted in order to rotate the bowl 2 in the range of, for example, 2000 to 5000 rpm, and the rotation speed thereof can be appropriately changed by inverter control. Further, the differential speed generator 5 is a device for generating a differential speed between the rotation speed of the bowl 2 and the rotation speed of the screw conveyor 3, and the screw conveyor 3 is slightly (for example, for example) relative to the bowl 2. It may be a device capable of rotating slowly (about 1 to 3 rpm). The differential speed generator 5 can include a gearbox 5a connected to the screw conveyor 3 and a differential motor (also referred to as “back drive motor”) 5b that applies a braking force to the screw conveyor 3. Since the specific operating principle of the differential speed generator 5 has been known conventionally, the description thereof will be omitted.

The casing 6 may be a case provided so as to cover the bowl 2 and the screw conveyor 3. The casing 6 guides the liquid-containing solid (dehydrated solid) M as a solid containing a small amount of liquid discharged from the solid discharge port 2a of the bowl 2 to the solid chute 6a provided below. Similarly, the separation liquid SL (see FIG. 3) discharged from the separation liquid discharge port 2b of the bowl 2 may be guided to the separation liquid chute 6b provided below. Further, the bowl 2 and the screw conveyor 3 covered with the casing 6 can be integrally supported by the frame 6c.

The solid matter chute 6a of the casing 6 may be connected to the solid matter discharge conduit 7, and the solid matter discharged from the solid matter chute 6a may be, for example, a drying device or an incinerator (not shown) via the solid matter discharge conduit 7. Etc. can be transported to the main transport path 7a. Further, the separation liquid chute 6b of the casing 6 may be connected to the separation liquid discharge conduit 8, and the separation liquid discharged from the separation liquid chute 6b may be connected to, for example, a water purification facility (not shown) via the separation liquid discharge conduit 8. Can be carried to.

The decanter 1 can further include a liquid supply pipe 9 for charging the liquid to be treated into the bowl 2. One end of the liquid supply pipe 9 communicates with the space 3c in the body 3b of the screw conveyor 3, and the other end communicates with the liquid supply source 10 to be processed and the additive supply source 11, and flows in from the other end. It is possible to form a conduit for charging the liquid to be treated or the like into the bowl 2.

It is preferable that the liquid supply pipe 9 and the liquid to be processed supply source 10 are fluidly connected by a liquid to be treated pipe 10b including a pump 10a for supplying the liquid to be treated. By controlling the pump 10a, it is possible to control the amount of the liquid to be supplied into the bowl. Here, the liquid to be treated supplied from the liquid to be treated source 10 is often a slurry (suspension) containing sludge as a solid substance.

Further, the liquid supply pipe 9 and the additive supply source 11 include an additive added to the liquid to be treated, for example, a pump 11a for supplying a drug, and a plurality of pumps 11a for controlling the supply position and supply timing of the drug. It may be fluidly connected by an additive pipe 11c including a valve 11b. Here, the agent supplied from the additive supply source 11 may be a flocculant that changes the sludge contained in the slurry into a floc shape by administering it to the slurry, particularly a polymer flocculant or an inorganic flocculant. In the following, the liquid to be treated before the drug is added is referred to as “liquid to be treated PL1”, and the liquid to be treated after the drug is added is referred to as “flock-containing liquid to be treated PL2” to distinguish between the two. ..

As shown in FIG. 2, the additive pipe 11c according to the present embodiment is branched into two pipes at an intermediate position thereof, one at an intermediate position of the liquid pipe 10b to be treated and the other at a liquid supply pipe 9. It may be connected. When the chemical is supplied via the pipe connected to the intermediate position of the liquid pipe 10b to be treated, the liquid PL1 to be treated reacts with the chemical in the liquid pipe 10b to be treated (so-called “line chemical injection”). When the drug is supplied via the pipe connected to the liquid supply pipe 9, the liquid to be treated PL1 reacts with the drug mainly inside the space 3c and inside the bowl 2 (so-called “in-flight drug injection”). It will be. By adopting such a piping structure, the position and timing at which the chemical is added to the liquid to be treated PL1 can be adjusted, and thus sludge flocs (“flock”) due to the chemicals mainly affected by the state of the liquid to be treated PL1. It will be possible to optimally adjust the formation effect (also called "aggregate"). The connection position of the pipe shown in FIG. 2 is only an example, and the pipe structure can be appropriately changed in consideration of the state of the liquid to be treated and the like.

Further, it is preferable that the various pipes of the decanter 1 according to the present embodiment are provided with a monitoring system for monitoring the object passing through the pipes. More specifically, this monitoring system is provided in the separation liquid monitoring system 40 provided in the separation liquid discharge conduit 8, the dehydrated solid matter monitoring system 50 provided in the solid matter discharge conduit 7, and the liquid to be treated pipe 10b. The liquid to be treated monitoring system 60 can be included. Further, in addition to these monitoring systems, a screw conveyor torque monitoring system 70 that monitors the torque value of the screw conveyor 3 may be included. As for the specific configuration of each monitoring system, some examples will be described so as to correspond to each embodiment described later. In addition, since the monitoring system other than the dehydrated solid matter monitoring system 50 is not used in the first embodiment, the monitoring system other than the dehydrated solid matter monitoring system 50 is necessary after the second embodiment. Will be explained accordingly.

The solid-liquid separation by the decanter 1 including the above configuration is generally performed as follows. That is, first, when the floc-containing liquid to be treated PL2 is put into the bowl 2 that rotates at a predetermined rotation speed via the liquid supply pipe 9, the floc-containing liquid to be treated is treated by the action of the centrifugal force generated by the drive motor 4. The solid matter (which is floc-like due to the effect of the chemical) in the liquid PL2 settles on the inner wall surface of the bowl 2. The settled solid matter is formed into a weight shape of the bowl 2 by the screw blade 3a of the screw conveyor 3 which is rotated with respect to the bowl 2 at a rotation speed slightly smaller than the rotation speed of the bowl 2 by the action of the differential speed generator 5. It is conveyed while being squeezed toward the processed one end side, and is discharged from the solid matter discharge port 2a to the solid matter chute 6a together with a predetermined amount of fluid components. Further, the residue in the bowl 2 after the solid matter has been settled and removed as described above stays in the bowl 2 as the separation liquid SL for a certain period of time, and then overflows from the separation liquid discharge port 2b. It is discharged to the chute 6b.

<Machine learning device>
The above-mentioned solid-liquid separation process realizes a suitable process by adjusting various control parameters of the decanter 1. The control parameter referred to here may mainly consist of the centrifugal force of the bowl 2, the supply amount of the drug from the additive supply source 11, and the differential speed between the bowl 2 and the screw conveyor 3. Therefore, the machine learning device 20 according to the first embodiment of the present disclosure for learning an inference model (learned model) capable of estimating the optimum control parameters in the decanter 1 will be described below. ..

FIG. 3 is a schematic block diagram of the machine learning device 20 according to the first embodiment of the present disclosure. As shown in FIG. 3, the machine learning device 20 according to the present embodiment includes a learning data set acquisition unit 21, a learning data set storage unit 22, a learning unit 23, and a trained model storage unit 24. Can include.

The learning data set acquisition unit 21 may be an interface unit that acquires a plurality of data constituting the learning (training) data set via, for example, a wired or wireless communication line. The specific contents of the plurality of data acquired here can be appropriately changed according to the trained model to be generated. In the present embodiment, as a plurality of data, the image data of the dehydrated solid M and the data in which the control parameters associated with the image data are acquired are exemplified. Further, the control parameters can include the supply amount of the drug added to the liquid to be treated PL1, the centrifugal force of the bowl 2, and the differential speed controlled by the differential speed generator 5. In this embodiment, the control parameters for acquiring the above-mentioned three data are exemplified, but the control parameters constituting the learning data set include at least one of the above-mentioned three data. I just need to be there. Further, the control parameters are not limited to the above three, and may include other parameters such as the supply amount of the liquid to be treated PL1 from the liquid to be treated supply source 10 and the dam set diameter of the dam 2c of the bowl 2. May include. Therefore, the number of control parameters constituting the output data may be four or more. Generally, if the number of control parameters increases, fine parameter adjustment can be realized when applied to the operation control of the decanter 1 described later. However, as the number of control parameters increases, the number of training data sets required to obtain a trained model that can be reasoned with sufficient accuracy tends to increase. Therefore, it is preferable to determine the number of control parameters as output data in consideration of the number of training data sets that can be prepared.

An example of an acquisition method of a plurality of data acquired by the learning data set acquisition unit 21 will be described. As shown in FIG. 3, the learning data set acquisition unit 21 is connected to the computer PC1 and acquires desired data from the computer PC1. The computer PC 1 may be, for example, a computer that constitutes at least a part of the control unit 30 that controls various operations of the decanter 1 described above, or is communicably connected to the control unit 30. As a result, various control parameters of the decanter 1 can be acquired in the computer PC1. In addition, as shown in FIG. 2, the computer PC 1 is directly connected to the dehydrated solids monitoring system 50 provided in the solids discharge conduit 7 or indirectly via the control unit 30, and the dehydrated solids. Image data of the dehydrated solid M can be acquired from the monitoring system 50. The control unit 30 of the decanter 1 referred to here is a device for controlling the entire decanter 1, and is well known including a processor, a memory, and the like connected to various sensors and drive means included in the decanter 1. It may be composed of a computer or the like.

As shown in FIG. 3, the dehydrated solid matter monitoring system 50 according to the present embodiment includes a window 52 provided at an arbitrary position of the solid matter discharge conduit 7 and a solid matter discharge system 50 installed in the window 52 at a predetermined angle of view. It is possible to adopt a camera including a dehydrated solid image capturing camera 53 capable of capturing image data of the dehydrated solid M passing through the conduit 7. Of these, a well-known camera capable of capturing a two-dimensional image can be adopted as the dehydrated solid matter imaging camera 53. Further, in the dehydrated solids monitoring system 50, the dehydrated solids M pass through the solids discharge conduit 7 so that the image data captured by the dehydrated solids imaging camera 53 can easily reflect the state of the surface of the dehydrated solids M. A light source (not shown) that illuminates the dehydrated solid matter M and a polarizing filter (not shown) that can be attached to the dehydrated solid matter imaging camera 53 can be appropriately adopted. Further, in FIG. 2, the solid matter discharge conduit 7 extends substantially vertically, and the dehydrated solid matter M exemplifies a solid matter passing through the solid matter discharge conduit 7 so as to fall naturally. The installation angle of the discharge conduit 7 and the like can be changed as appropriate. Further, in the present embodiment, the case where the dehydrated solid matter monitoring system 5 is installed in the solid matter discharge conduit 7 has been described, but the present disclosure is not limited to this. That is, the dehydrated solids monitoring system 5 can be installed at any other position through which the dehydrated solids M passes, such as the main transport path 7a and the solids chute 6a. The configuration of the dehydrated solids monitoring system 50 is not limited to this, and various configurations can be adopted as long as the image data of the dehydrated solids M can be acquired. As a specific example, for example, a configuration such as the dehydrated solid matter monitoring system 50A (see FIG. 9) described later may be adopted.

FIG. 4 shows an example of image data obtained by the dehydrated solids imaging camera 53 of the dehydrated solids monitoring system 50 according to the first embodiment of the present disclosure. By the way, in the technical field of water treatment equipment, the amount of liquid component contained in the dehydrated solid M discharged after the solid-liquid separation treatment by the centrifugal separation system, that is, the water content is small, the solid-liquid separation treatment is performed. It can be a guideline that can be inferred to be well done. Further, in the image data obtained by the dehydrated solids monitoring system 50, if the amount of the liquid component contained in the dehydrated solids M is large, the color tends to be deep and the brightness tends to increase. Therefore, the image data having a relatively high water content of the dehydrated solid M (for example, the one shown in FIG. 4B) has a color tint between pixels as compared with the image data having a relatively low water content (for example, the one shown in FIG. 4A). It can be inferred that the amount of change in is likely to increase.

The training data set storage unit 22 associates a plurality of data constituting the training data set acquired by the training data set acquisition unit 21 with related input data and output data (also referred to as “teacher data”). It may be a database for storing one learning data set. The learning data set stored in the present embodiment uses image data obtained by capturing the dehydrated solid material M discharged from the solid material discharge port 2a from a predetermined angle as input data, and corresponds to the image data as the input data. The attached control parameter can be used as output data. Further, the specific configuration of the database constituting the learning data set storage unit 22 can be appropriately adjusted. For example, in FIG. 3, for convenience of explanation, the training data set storage unit 22 and the trained model storage unit 24 described later are shown as separate storage means, but these are used as a single storage medium (database). ) Can also be configured.

By the way, as described above, the learning data set stored in the learning data set storage unit 22 can be composed of one image data and control parameters corresponding to the one image data. On the other hand, in order to generate one trained model in the learning unit 23, it is usually necessary to perform training using many training data sets (for example, thousands to tens of thousands of sets). Therefore, in order to prepare a large amount of training data set in a relatively short time, it is preferable to perform data augmentation in the learning data set storage unit 22. As a specific method of this data augmentation, for example, a (for example, 100) partial image data of a square smaller than the original image data is randomly extracted from the original image data of one rectangle and extracted. By associating each of the a partial image data with the control parameter associated with the original image data, it is possible to adopt a method of acquiring a learning data set.

The learning unit 23 infers the correlation between the input data and the output data in the learning data set by inputting a plurality of sets of the plurality of learning data sets stored in the learning data set storage unit 22. It may be one that learns a learning model. In this embodiment, as will be explained in detail later, supervised learning using a neural network is adopted as a specific method of machine learning. However, the specific method of machine learning is not limited to this, and other learning methods may be adopted as long as the correlation between input and output can be learned from the training data set. It is possible. For example, ensemble learning (random forest, boosting, stacking, etc.) can also be used.

The trained model storage unit 24 may be a database for storing the trained model generated by the training unit 23. The trained model stored in the trained model storage unit 24 can be applied to a real system via a communication line including the Internet or a storage medium, if requested. The specific application mode of the trained model to the actual system (data processing system 80) will be described in detail later.

By the way, when preparing the learning data set in the decanter 1 according to the present embodiment, it is advisable to specify the optimum control parameters to be the teacher data. Various methods can be adopted as a method for specifying the optimum control parameter. For example, the operator EN manually takes into consideration the image data acquired by the dehydrated solid matter monitoring system 50 and the actual control parameter at that time. The method specified by can be adopted. In this case, it is preferable to assign a skilled engineer or a plurality of engineers to the operator EN that specifies the optimum control parameter to be the teacher data. The optimum control parameters identified in this way are organized and stored in the learning data set storage unit 22 in a state associated with the above-mentioned image data as teacher data of the learning data set in the machine learning device 20. To.

Here, the optimum control parameters specified are, as described above, at least one of the centrifugal force of the bowl 2, the supply amount of the drug from the additive supply source 11, and the difference speed between the bowl 2 and the screw conveyor 3. It may be one. It can be said that these control parameters are considered to be the control parameters having the greatest influence on the solid-liquid separation process by the decanter 1. Therefore, it is preferable that the control parameters specified here are all three control parameters described above as exemplified in the present embodiment. Of course, only one or only two of these three control parameters may be adopted as output data, and parameters other than these three control parameters (for example, drug supply position, etc.) may be used as output data. It is also possible to add more. By the way, the centrifugal force of the bowl 2 among the above three control parameters can be adjusted mainly by controlling the rotation speed of the bowl 2 by the drive motor 4, but the control for changing the rotation speed is other. It is characterized by low responsiveness compared to the two controls (control of drug supply amount and differential speed). Therefore, when only two of the above three control parameters are adopted for the output data, the supply amount of the drug from the additive supply source 11 and the differential speed between the bowl 2 and the screw conveyor 3 are adopted. good.

Next, an example of the learning method in the learning unit 23 using the plurality of learning data sets obtained as described above will be briefly described. FIG. 5 is a diagram showing an example of a neural network model for supervised learning implemented in the machine learning device according to the first embodiment of the present disclosure. The neural network in the neural network model shown in FIG. 5 includes l neurons (x1 to xl) in the input layer, m neurons (y11 to y1m) in the first intermediate layer, and n neurons in the second intermediate layer. It is composed of neurons (y21 to y2n) and o neurons (z1 to zo) in the output layer. The first intermediate layer and the second intermediate layer are also called hidden layers, and the neural network may have a plurality of hidden layers in addition to the first intermediate layer and the second intermediate layer. Or, only the first intermediate layer may be used as a hidden layer.

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

The neural network in the neural network model according to the present embodiment learns the correlation between the input data and the output data of the training data set by using the training data set. Specifically, the image data of the dehydrated solid M as a state variable constituting the input data is associated with the neurons in the input layer, and the values of the neurons in the output layer are calculated as the output values of a general neural network. That is, the value of the neuron on the output side is multiplied by the value of the neuron on the input side connected to the neuron and the weight wi associated with the node connecting the neuron on the output side and the neuron on the input side. It is calculated as the sum of a number of values by using a method performed for all neurons other than the neurons in the input layer. When inputting the above state variables to the neurons of the input layer, the format of the information acquired as the state variables may be appropriately set in consideration of the accuracy of the generated trained model. can.

Then, it is composed of the values of o neurons z1 to zo in the calculated output layer, that is, the values corresponding to the control parameters in the present embodiment, and the control parameters that form a part of the learning data set. The teacher data t1 to to are compared with each other to obtain an error, and the weight wi associated with each node is adjusted (backpropagation) so that the obtained error becomes small.

When a predetermined condition such as repeating the above-mentioned series of steps a predetermined number of times or the error becoming smaller than the allowable value is satisfied, the learning is terminated and the neural network model (node of the node) is satisfied. All the weights wi) associated with each are stored in the trained model storage unit 24 as a trained model.

<Machine learning method>
In connection with the above, the present disclosure provides a machine learning method. FIG. 6 is a flowchart showing an example of the machine learning method according to the first embodiment of the present disclosure. The machine learning method shown below will be described based on the machine learning device 20 described above, but the premise configuration is not limited to the machine learning device 20 described above. Further, although this machine learning method is realized by using a computer, various computers can be applied, for example, a computer constituting the control unit 30 and a server device arranged on a network. , Or the computer PC1 shown in FIG. 3 and the like. Regarding the specific configuration of this computer, for example, an arithmetic unit composed of at least a CPU, a GPU, etc., a storage device composed of a volatile or non-volatile memory represented by RAM or ROM, a network, or other devices. It is possible to adopt a device including a communication device for communicating with the device and a bus connecting each of these devices. Furthermore, this machine learning method is provided in the form of a program containing one or more instructions for performing a predetermined operation on a computer, or in the form of a non-temporary computer-readable medium containing the program. You may.

In supervised learning as a machine learning method according to the present embodiment, as a preliminary preparation for starting machine learning, a desired number of learning data sets are first prepared, and a plurality of prepared learning data sets are prepared. It is stored in the learning data set storage unit 22 (step S11). The number of training data sets prepared here may be set in consideration of the inference accuracy required for the finally obtained trained model. Further, since an example of the method of preparing the training data set has already been illustrated above, the description thereof will be omitted here.

When step S11 is completed, a neural network model before learning is prepared in order to start machine learning in the learning unit 23 (step S12). As the pre-learning neural network model prepared here, for example, a neural network model having the structure shown in FIG. 4 and having the weight of each node set to the initial value can be adopted. Then, for example, one learning data set is randomly selected from the plurality of learning data sets stored in the learning data set storage unit 22 (step S13), and the input data in the one learning data set is selected. Is input to the input layer (see FIG. 4) of the prepared neural network model before learning (step S14). As a method of inputting the input data in the training data set to the input layer of the neural network model before training, various methods can be adopted. As a specific example, a method of inputting a luminance value and / or a color value (for example, an RGB value) for each pixel of image data into a neuron of each input layer can be adopted. Further, before inputting image data as input data to the input layer, predetermined preprocessing such as dimension reduction processing and noise removal for adjusting the number of data may be executed.

Here, since the control parameters of the output layer (see FIG. 4) generated as a result of step S14 are generated by the neural network model before training, they are different from the desired results in most cases. .. Therefore, next, machine learning is performed using the output data as teacher data in one learning data set acquired in step S13, that is, the control parameters and the control parameters of the output layer generated in step S13. It is carried out (step S15). The machine learning performed here is, for example, to compare the control parameters constituting the teacher data and the control parameters constituting the output layer, detect an error between them, and obtain an output layer in which this error becomes small. , It may be a process (backpropagation) for adjusting the weight associated with each node in the neural network model before learning. Further, the number and format of the control parameters output to the output layer of the neural network model before training are the same number and format as the teacher data in the training data set as the training target. Therefore, for example, when the control parameter as the teacher data in the training data set is composed of the data corresponding to the two control parameters, the control parameter output to the output layer by the neural network model is also the two control parameters. It should be the corresponding data.

When machine learning is performed in step S15, whether or not it is necessary to continue machine learning is determined based on, for example, the remaining number of unlearned learning data sets stored in the learning data set storage unit 22. (Step S16). Then, when the machine learning is continued (No in step S16), the process returns to step S13, and when the machine learning is finished (Yes in step S16), the process proceeds to step S17. When the machine learning is continued, the steps S13 to S15 are performed a plurality of times on the neural network model being trained by using the unlearned learning data set. The accuracy of the finally generated trained model generally tends to increase in proportion to this number of times.

When machine learning is terminated (Yes in step S16), the neural network generated by adjusting the weights associated with each node by a series of steps is stored in the trained model storage unit 24 as a trained model. (Step S17), a series of learning processes is completed. The trained model stored here can be applied to various data processing systems and used, and specific examples of the data processing systems will be described later.

In the learning process and machine learning method of the machine learning device described above, in order to generate one trained model, one machine learning process is repeatedly executed for one (pre-learning) neural network model. The present disclosure is not limited to this acquisition method, although the method of improving the inference accuracy and obtaining a trained model sufficient for application to a data processing system is explained in the above. For example, a plurality of trained models that have undergone machine learning a predetermined number of times are stored as one candidate in a plurality of trained model storage units 24, and a common data set for determining validity is stored in the plurality of trained model groups. One of the best trained models to apply to a data processing system by inputting the input data of to generate an output layer (value of neurons) and comparing the accuracy of the control parameters identified in this output layer. You may choose. The validity determination data set may be any data set that is the same as the learning data set used for learning and is not used for learning.

As an option, instead of the pre-learning neural network model in which the node weight prepared in step S12 is set to the initial value, for example, a pre-generated trained model for arbitrary image recognition may be used. can. In this case, a desired trained model is generated by so-called fine tuning or transfer learning. Therefore, a highly accurate trained model can be generated with a smaller number of training data sets as compared with the above-mentioned normal machine learning process.

As described above, by applying the machine learning device 20 and the machine learning method according to the present embodiment, it is possible to derive the optimum control parameters from the image data acquired by the dehydrated solid matter monitoring system 50. You can get a trained model.

<Data processing system>
Next, with reference to FIG. 7, an application example of the trained model generated by the machine learning device and the machine learning method described above will be described. FIG. 7 is a schematic block diagram showing a data processing system 80 according to the first embodiment of the present disclosure. As the data processing system 80 according to the present embodiment, the one applied in the control unit 30 of the horizontal decanter 1 described above is exemplified.

As shown in FIG. 7, the data processing system 80 mainly includes a first image data acquisition unit 81, a parameter adjustment unit 82, an arithmetic unit 83, a database (DB) 84, a user interface 85, and the like. It may include an internal bus 86 for interconnecting. Note that, in FIG. 7, only the components related to the data processing system 80 are shown, and the description of other components not directly related to the data processing system 80 according to the present embodiment of the control unit 30 is omitted. ing.

The first image data acquisition unit 81 may be for acquiring image data, specifically, image data of the dehydrated solid M. Specifically, it can be connected to the above-mentioned dehydrated solids monitoring system 50 and acquire image data captured by the dehydrated solids imaging camera 53 of the dehydrated solids monitoring system 50.

The parameter adjustment unit 82 may be for adjusting various control units adjusted by the control unit 30 to realize optimum operation control based on the inference result of the inference unit 87 described later. The parameter adjusting unit 82 controls a drive motor control unit 31 which is connected to the drive motor 4 and can control the rotation speed thereof, and a pump 11a provided in the additive pipe 11c to change the supply amount of the drug. A pump control unit 32 capable of controlling the pump control unit 32, and a differential motor control unit 33 connected to the differential motor 5b and capable of controlling the rotation speed thereof to control the rotation speed of the screw conveyor 3 (relative to the bowl 2). It should be connected to. The parameter adjustment unit 82 according to the present embodiment exemplifies those connected to the above-mentioned three control units, but the connection destinations are set according to the number and types of control parameters output by the inference unit 87. It can be adjusted as appropriate.

The arithmetic unit 83 may constitute a processor for realizing various processes in the data processing system 80, and may include at least an inference unit 87. Further, the arithmetic unit 83 may be connected to the trained model storage unit 88 that stores the trained model used in the inference unit 87. The inference unit 87 uses the parameter adjustment unit 82 to obtain image data as a state variable acquired by the first image data acquisition unit 81 by referring to one trained model stored in the trained model storage unit 88. It may infer the control parameters to be adjusted. A plurality of trained models stored in the trained model storage unit 88 are stored according to the intended use and various conditions (for example, environmental conditions such as season, weather and temperature / humidity, type of liquid to be treated, etc.). It is preferable to have it. In relation to this, the work of selecting an appropriate trained model from a plurality of trained models may be automatically selected using various sensors or the like, or may be manually selected by an operator EN or the like. You may be able to select it.

The database 84 is made of a well-known recording medium, and temporarily or continuously stores various data handled by the data processing system 80, such as image data acquired by the first image data acquisition unit 81 and calculation results by the calculation unit 83. It may be for the purpose of doing. Further, the user interface 85 is composed of, for example, a GUI (graphical user interface), and may be for receiving a status display of the decanter 1, an input operation from an operator EN, or the like.

FIG. 8 is a flowchart showing an example of parameter adjustment by the data processing system 80 according to the first embodiment of the present disclosure. When the drive of the decanter 1 to which the data processing system 80 is applied is started, as shown in FIG. 8, the data processing system 80 first determines whether or not it is the timing of parameter adjustment (step S21). Here, the timing for adjusting the control parameters can be automatically specified or manually adjusted. Further, the adjustment may be performed periodically at predetermined time intervals while the decanter 1 is operating. For example, when the operation of the decanter 1 is started, the type of the liquid to be treated by the decanter 1 may be changed. When the change is made, when a predetermined time set in advance has elapsed after the operation, or the change in the concentration of the liquid to be treated or the separation liquid can be detected by the liquid to be monitored system 60 or the separation liquid monitoring system 40, the change is predetermined. The adjustment may be made only at a specific timing such as when the threshold value of is exceeded.

When it is detected that it is the timing of parameter adjustment (Yes in step S21), the dehydrated solid imaging camera 53 in the dehydrated solid monitoring system 50 is operated to detect the dehydrated solid M passing through the solid discharge conduit 7. Image is taken (step S22). The image data captured here is acquired by the first image data acquisition unit 81 in the data processing system 80 (step S23).

Next, the image data acquired by the first image data acquisition unit 81 is sent to the inference unit 87 via the internal bus 86, and in the inference unit 87, this image data is stored in the trained model previously specified in the inference unit 87. The control parameters are inferred by being input to the input layer of one trained model in the unit 88 (step S24). Then, the parameter adjustment unit 82 adjusts each control unit using the value of the control parameter inferred here (step S25). After that, the process returns to step S21 to enter the standby state.

As described above, according to the data processing system 80 according to the present embodiment, the preferable control parameters of the decanter 1 can be estimated based on the relatively easy-to-acquire information such as the image data of the dehydrated solid matter. Therefore, the data processing system 80. Can be applied to Decanter 1 relatively easily. Then, the adjustment of the control parameter can be easily performed without depending on the judgment of the operator EN or the like.

As an option, in the above embodiment, as a trained model referred to in the inference unit 87, a model generated through learning by so-called batch learning as shown in FIG. 6 is explained, but the data of the present disclosure is explained. The processing system 80 is not limited to this. Specifically, for example, the accuracy may be further improved by further applying online learning to the trained model generated through the above-mentioned learning by batch learning. As a specific method thereof, first, the image data as a state variable acquired by the first image data acquisition unit 81 and the control parameters inferred by the inference unit 87, particularly the adjustment in the parameter adjustment unit 82. As a result of reflecting the above, the control parameters when the solid-liquid separation processing result is improved are temporarily stored in the database 84 as a set of online learning data sets. Then, machine learning (specifically, fine tuning) similar to that shown in FIG. 6 may be executed using the online learning data set accumulated in the database 84 at an arbitrary timing.

Further, the above-mentioned data processing system 80 is applied in the control unit 30 of the horizontal decanter 1, but instead of this, for example, a computer communicably connected to the control unit 30 of the decanter 1 or a decanter 1 can be used. It can also be applied to a server device or the like connected to the control unit 30 via a network or the like. However, the centrifugal separation system mainly assumed for sludge treatment in this embodiment is based on the treatment content (type of liquid to be treated, treatment amount per unit time, etc.) and surrounding environment (climate, etc.). It is usually the case that it varies greatly from one to another. Therefore, generating a general-purpose trained model that can be applied to all of these centrifuge systems requires a very large number of training datasets to generate, and the content of the data is also various conditions. Since the ones acquired below are required evenly, the generation cost tends to be high. Therefore, it is often preferable that the data processing system 80 and the above-mentioned trained model are applied and operated for each centrifuge system in terms of cost.

<Second embodiment>
In the machine learning device, the machine learning method, and the data processing system according to the first embodiment, the case where the control parameter is inferred from the image data of the dehydrated solid M using one trained model has been described. .. However, although there is a correlation between the image data of the dehydrated solid M and the control parameters, it is difficult to specify the degree of the correlation between the two. Generally, in the machine learning process of a neural network model, until a trained model that can be inferred with sufficient accuracy is obtained in proportion to the degree of correlation between the input data and the output data of the training data set. The number of training datasets required for a neural network tends to be small. Therefore, in the first embodiment described above, it may be necessary to prepare a relatively large number of training data sets in order to learn the correlation between the image data of the dehydrated solid M and the control parameters. possible. Therefore, as one aspect to enable the generation of a trained model capable of inferring with sufficient accuracy with a smaller number of training data sets, there are two ways to infer control parameters from the image data of the dehydrated solid M. The case of using the trained model will be described below as a second embodiment of the present disclosure. Of the components of the machine learning device 20A and the data processing system 80A according to the second embodiment shown below, each component of the machine learning device 20 and the data processing system 80 according to the first embodiment. The same reference numerals are given to those common to the above, and the description thereof will be omitted. Further, the machine learning device, the machine learning method, and the data processing system according to the second embodiment shown below are the same as those according to the first embodiment, and the centrifugal separation system shown in FIGS. 1 and 2 is used. The explanation is given by taking the case of application as an example. Furthermore, all the modifications described in each of the above-mentioned or later embodiments can be applied to the present embodiment as long as there is no contradiction.

FIG. 9 is a schematic block diagram of the machine learning device 20A according to the second embodiment of the present disclosure. As shown in FIG. 9, the machine learning device 20A according to the present embodiment has, as its components, a first learning data set storage unit 221 and a second learning data as a learning data set storage unit. It may be the same as the machine learning device 20 according to the first embodiment except that the set storage unit 222 is included and the learning unit includes the first learning unit 231 and the second learning unit 232. Further, in connection with this, the plurality of data acquired by the learning data set acquisition unit 21 may also be different from those of the first embodiment.

A plurality of data acquired by the learning data set acquisition unit 21 according to the present embodiment can be acquired from a computer PC 1 or the like, and the plurality of data include image data of the dehydrated solid M and control parameters. In addition to this, the feature amount of the image data of the dehydrated solid matter M (hereinafter, also referred to as “characteristic amount of the dehydrated solid matter M”) may be included. As the feature amount of the dehydrated solid M, various information specified based on the imaged dehydrated solid M can be adopted. Specifically, for example, one or a plurality of water content, color value, density and the like of the dehydrated solid M can be included. This feature amount can be specified, for example, by actually sampling the dehydrated solid M and measuring and analyzing it using various measuring instruments or the like, or by visually determining by the operator EN.

In order to measure the above-mentioned feature amount, it is preferable that the dehydrated solid M monitoring system 50A according to the present embodiment adopts a configuration capable of sampling and extracting the dehydrated solid M. Specifically, as shown in FIG. 9, the dehydrated solids monitoring system 50A has a tray 54 including a predetermined depth in which the dehydrated solids M can be filled, and the tray 54 can be placed inside. It can include a photographing box 55 having a light source inside, and a camera 53 for capturing a dehydrated solid that can image the surface of the dehydrated solid M filled in the tray 54 from a predetermined angle of view. The dehydrated solid M filled in the tray 54 is extracted at a predetermined timing using a sampling unit (not shown) provided at an appropriate position in the dehydrated solid M transport path such as the solid discharge conduit 7 or the main transport path 7a. It may be the one that was done. Then, image data is generated by imaging the dehydrated solid M extracted in the tray 54 with the dehydrated solid imaging camera 53, and a desired control parameter corresponding to the image data is specified by the operator EN or the like. .. Further, by measuring and analyzing the dehydrated solid M sampled in the tray 54, the characteristic amount of the dehydrated solid M can be specified. As a matter of course, also in the present embodiment, the dehydrated solid matter monitoring system 50 described in the above-described first embodiment can be adopted as an alternative.

The plurality of data acquired in the learning data set acquisition unit 21 are the first learning data set storage unit 221 and the second learning data set as two learning data sets while considering their respective correspondences. It may be stored separately in the storage unit 222. The first learning data set stored in the first learning data set storage unit 221 is the image data of the dehydrated solid M obtained by imaging the dehydrated solid M discharged from the discharge port 2a from a predetermined angle. It may be included as the input data of No. 1 and may include the feature amount of the dehydrated solid M associated with the first input data as the first output data. Further, the second learning data set stored in the second learning data set storage unit 222 includes the feature amount of the dehydrated solid M as the second input data and is associated with the second input data. The control parameter may be included as the second output data. When two training data sets are separately generated from a plurality of data in this way, for example, the image data of the dehydrated solid M associated with the same dehydrated solid M, the feature amount of the dehydrated solid M, and the feature amount of the dehydrated solid M The control parameters are divided into a set of the image data of the dehydrated solid M and the feature amount of the dehydrated solid M and a set of the feature amount of the dehydrated solid M and the control parameter, and each of them is the first and the first. It may be the training data set of 2. At this time, since the first and second learning data sets after the division are referred to in different learning units described later, they are not stored in a format that maintains the relationship. You can do it.

The first and second learning data sets stored in the first learning data set storage unit 221 and the second learning data set storage unit 222, respectively, are referred only to different learning units. It may be there. The first learning unit 231 learns a learning model that infers the correlation between the first input data and the first output data by inputting a plurality of sets of the first learning data sets. good. In other words, the first learning unit 231 inputs the image data of the dehydrated solid M in the first learning data set, so that the feature amount of the dehydrated solid M in the image data, that is, the image data It may be the one that learns the first learning model that infers the feature quantity of. Then, the second learning unit 232 learns a learning model that infers the correlation between the second input data and the second output data by inputting a plurality of sets of the second learning data sets. It may be there. In other words, the second learning unit 232 may learn the second learning model for inferring the control parameter by inputting the feature amount of the image data in the second learning data set. ..

The specific machine learning methods in the first and

second learning units

231 and 232 are the same as the supervised learning process shown in FIG. 6, although the learning data set used for learning is different. be able to. Then, the first and second trained models obtained through a series of machine learning steps may be stored in the trained model storage unit 24, respectively.

FIG. 10 is a schematic block diagram showing a data processing system 80A according to the second embodiment of the present disclosure. As shown in FIG. 10, the data processing system 80A according to the present embodiment has the above-mentioned first inference unit except that the inference unit 83 includes the first inference unit 871 and the second inference unit 872. It may include the same components as the data processing system 80 according to the first embodiment.

The first inference unit 871 may execute inference using the first trained model generated by the machine learning device 20A described above and stored in the trained model storage unit 88. Therefore, when the image data of the dehydrated solid M acquired by the first image data acquisition unit 81 is input, the first inference unit 871 can output the feature amount of the image data to the output layer. can. Further, the second inference unit 872 may execute inference using the second trained model generated by the machine learning device 20A described above and stored in the trained model storage unit 88. Therefore, the second inference unit 872 can output the control parameter to the output layer when the feature amount of the image data inferred by the first inference unit 871 is input.

When adjusting the control parameters in the decanter 1 to which the data processing system 80A including the first and

second inference units

871 and 872 described above is applied, the same processing as that shown in FIG. 8 may be performed. .. However, in the process shown in FIG. 8, when inferring the control parameter in step S24, first, the image data of the dehydrated solid M is input to the first inference unit 871, and the feature amount of the output image data is determined. It will be input to the inference unit 872 of 2.

As described above, according to the machine learning device, the machine learning method, and the data processing system according to the present embodiment, the feature amount of the image data having a large degree of correlation with the image data of the dehydrated solid matter M as compared with the control parameters can be obtained. After inferring, the control parameters are inferred using the feature amount of the image data having a larger degree of correlation with the control parameters than the image data of the dehydrated solid M, and each learning capable of inferring with sufficient accuracy is possible. It can be expected that the number of training data sets required to generate a completed model will be relatively reduced.

<Third embodiment>
In the machine learning device and the machine learning method according to the first and second embodiments, the one in which only the image data of the dehydrated solid M is adopted as the input data of the learning data set has been described. However, if only the image data of the dehydrated solid M is used as the input data, the number of training data sets required to generate a trained model capable of inferring with sufficient accuracy tends to increase. Therefore, as one aspect for generating a trained model capable of inferring with sufficient accuracy with a smaller number of training data sets, the input data of the training data set shown in the first embodiment described above is used. The case where the number is increased will be described below as a third embodiment of the present disclosure. Of the components of the machine learning device 20B and the data processing system 80B according to the third embodiment shown below, the components of the machine learning device 20 and the data processing system 80 according to the first embodiment. The same reference numerals are given to those common to the above, and the description thereof will be omitted. Further, the machine learning device, the machine learning method, and the data processing system according to the third embodiment shown below are the same as those according to the first and second embodiments, and the centrifuge shown in FIGS. 1 and 2. The explanation is given by taking the case of applying to a separation system as an example. Further, all the modifications described in the above-mentioned or later embodiments can be applied to the present embodiment as long as there is no contradiction.

FIG. 11 is a schematic block diagram of the machine learning device 20B according to the third embodiment of the present disclosure. As shown in FIG. 11, the machine learning device 20B according to the present embodiment includes a third learning data set storage unit 223 as a learning data set storage unit for each component, and a third learning unit as a learning unit. It may be the same as the machine learning apparatus 20 according to the first embodiment described above except that the learning unit 233 is included. Further, the plurality of data acquired by the learning data set acquisition unit 21 may also be different from those according to the first embodiment. As shown in FIG. 11, the learning data set acquisition unit 21 according to the present embodiment may be connected to the computer PC1 and can acquire desired data from the computer PC1. The computer PC 1 is connected to the control unit 30 and the dehydrated solid matter monitoring system 50 as in the computer PC 1 according to the first embodiment, but in addition to these, the separation liquid monitoring system 40 and the object to be processed. It may be connected to the liquid monitoring system 60 and the screw conveyor torque monitoring system 70.

As shown in FIGS. 2 and 11, the separation liquid monitoring system 40 is provided in, for example, the separation liquid discharge conduit 8 and monitors the solid content concentration of the separation liquid SL passing through the separation liquid discharge conduit 8. Is preferable. As a specific method for monitoring the concentration of the separated liquid, for example, the concentration value of the separated liquid SL may be directly detected using a well-known laser type, optical type, or ultrasonic type concentration sensor.

As shown in FIGS. 2 and 11, the liquid to be monitored system 60 is arranged, for example, in the liquid pipe 10b to be treated, and monitors the slurry concentration of the liquid to be treated PL1 supplied from the liquid supply source 10 to be treated. It may be there. As a specific method for monitoring the slurry concentration, for example, a well-known laser-type, optical-type, or ultrasonic-type concentration sensor may be used to directly detect the concentration value of the liquid to be treated. In the present embodiment, the liquid to be treated monitoring system 60 is provided at a position where it merges with the liquid to be treated pipe 10b, particularly the additive pipe 11c. Instead of this, for example, the liquid to be treated monitoring system 60 is used. The liquid pipe 10b to be treated may be provided at a position before merging with the additive pipe 11c, in the liquid supply pipe 9, or in the space 3c. Further, it is also possible to measure the slurry concentration not with the liquid to be treated PL1 before the chemical is added, but with the floc-containing liquid to be treated PL2 after the chemical is added to the liquid PL1 to be treated.

As shown in FIGS. 2 and 11, the screw conveyor torque monitoring system 70 can be attached to, for example, the rotating shaft of the screw conveyor 3, and by detecting the reaction force acting on the screw conveyor 3, the screw conveyor 3 can be attached to the screw conveyor 3. The torque value can be monitored.

Various data acquired by these four

monitoring systems

40, 50, 60 and 70, specifically, image data of the dehydrated solid M, the concentration of the separation liquid SL, and the liquid to be treated PL1 (or the liquid to be treated PL2 containing flocs). The slurry concentration and the torque value of the screw conveyor 3 can be sent to the computer PC 1 directly or via the control unit 30. Then, it can be sent from the computer PC 1 to the learning data set acquisition unit 21 together with the corresponding desired control parameters. The plurality of data constituting the learning data set acquired by the learning data set acquisition unit 21 are the image data of the dehydrated solid M, the concentration of the separation liquid SL, the slurry concentration of the liquid to be treated PL1, and the torque value of the screw conveyor 3. Is stored in the third training data set storage unit 223 in the form of a third training data set in which is used as the third input data and the control parameters associated with these data are used as the third output data. It's okay. Then, in the third learning unit 233, machine learning is performed by the same method as the machine learning method of the first embodiment described above using the third learning data set, and the obtained third learning unit is obtained. The trained model may be stored in the trained model storage unit 24. In the present embodiment, of the third input data, the concentration of the separation liquid SL, the slurry concentration of the liquid to be treated PL1, and the torque value of the screw conveyor 3 are data of one value. On the other hand, since the image data of the dehydrated solid M is the brightness and / or the color value of all the pixels constituting the image data, the total number of data differs greatly. Therefore, if these input data are directly associated with the input layer, the obtained inference result will be relatively greatly affected by the image data of the dehydrated solid M. Therefore, in the present embodiment, in order to adjust the number of data before associating the third input data with the input layer so that the degree of influence of the input data other than the image data of the dehydrated solid M is not too small. It is particularly preferable to perform preprocessing of the data in. Further, the preprocessing can be similarly performed at the time of inference by storing it in the trained model storage unit 24 together with the neural network model obtained as a part of the trained model.

FIG. 12 is a schematic block diagram showing a data processing system 80B according to a third embodiment of the present disclosure. As shown in FIG. 12, the data processing system 80B according to the present embodiment has the first embodiment described above except that the inference unit includes the third inference unit 873 and the additional variable acquisition unit 89. It may include the same components as the data processing system 80 according to the form.

As shown in FIG. 12, the additional variable acquisition unit 89 is connected to the separation liquid monitoring system 40, the liquid to be processed monitoring system 60, and the screw conveyor torque monitoring system 70, and the concentration of the separation liquid SL and the subject to be covered from these monitoring systems. The slurry concentration of the processing liquid PL1 and the torque value of the screw conveyor 3 may be acquired. The concentration of the separation liquid SL acquired by the additional variable acquisition unit 89, the slurry concentration of the liquid to be processed PL1 and the torque value of the screw conveyor 3 are acquired by the first image data acquisition unit 81 from the dehydrated solid matter monitoring system 50. It is preferable that the data is acquired at the same time as the image data is captured.

When adjusting the control parameters in the decanter 1 to which the data processing system 80B including the additional variable acquisition unit 89 described above is applied, the same processing as that shown in FIG. 8 described above may be performed. However, among the processes shown in FIG. 8, the additional variable acquisition unit 89 acquires the concentration of the separation liquid SL, the slurry concentration of the liquid to be treated PL1, and the torque value of the screw conveyor 3 at the same time as or at a predetermined timing before and after step S23. do. Then, these three values are associated with the input layer of the trained model in the third inference unit 873 together with the image data of the dehydrated solid M.

In the third embodiment described above, in addition to the control unit 30 and the dehydrated solid matter monitoring system 50, the computer PC 1 includes a separation liquid monitoring system 40, a liquid to be processed monitoring system 60, and a screw conveyor torque. An example is illustrated in which three values are connected to the monitoring system 70 and the three values obtained from these are used as input data of a training data set together with image data, but the present disclosure is limited to this. Not done. Specifically, if at least one of the values acquired by the separation liquid monitoring system 40, the liquid to be processed monitoring system 60, and the screw conveyor torque monitoring system 70 is used as input data of the learning data set. good. The same applies to the number of information acquired by the additional variable acquisition unit 89.

As described above, according to the machine learning device, the machine learning method, and the data processing system according to the present embodiment, in addition to the image data of the dehydrated solid M as input data, the concentration of the separation liquid SL and the liquid to be treated PL1 By adopting the slurry concentration and the torque value of the screw conveyor 3, it becomes easy to identify the correlation between the input data and the output data in the learning stage of the neural network model. This can be expected to reduce the number of training data sets required to generate each trained model that can be reasoned with sufficient accuracy.

<Fourth Embodiment>
As described above, in each of the second and third embodiments, the number of training data sets required to generate a trained model capable of inferring with sufficient accuracy is determined by the first embodiment. It is an example showing an embodiment that can be less than the number in the thing. And both embodiments can be combined. Therefore, the case where the technical ideas shown in the second embodiment and the third embodiment described above are combined will be described below as the fourth embodiment of the present disclosure. Of the components of the machine learning device 20C and the data processing system 80C according to the fourth embodiment shown below, the machine learning devices 20, 20A and 20B according to the first to third embodiments, and , The same reference numerals are given to those common to the respective components of the data processing systems 80, 80A and 80B, and the description thereof will be omitted. Further, the machine learning device, the machine learning method, and the data processing system according to the third embodiment shown below are the same as those according to the first to third embodiments, and the centrifuge shown in FIGS. 1 and 2. The explanation is given by taking the case of applying to a separation system as an example. Further, all the modifications described in the above-mentioned or later embodiments can be applied to the present embodiment as long as there is no contradiction.

FIG. 13 is a schematic block diagram of the machine learning device 20C according to the fourth embodiment of the present disclosure. As shown in FIG. 13, the machine learning device 20C according to the present embodiment has, as its components, a first learning data set storage unit 221 and a fourth learning data as a learning data set storage unit. It may be the same as the machine learning device 20 according to the first embodiment except that the set storage unit 224 is included and the learning unit includes the first learning unit 231 and the fourth learning unit 234. Further, in connection with this, the plurality of data acquired by the learning data set acquisition unit 21 may also be different from the first embodiment.

As shown in FIG. 13, the learning data set acquisition unit 21 according to the present embodiment may be connected to the computer PC1 and can acquire desired data from the computer PC1. The computer PC1 is connected to the control unit 30 and the dehydrated solid matter monitoring system 50A as in the computer PC1 according to the second embodiment, but in addition to these, the separation liquid monitoring system 40 and the liquid to be treated It may also be connected to at least one of the monitoring system 60 and the screw conveyor torque monitoring system 70. The specific configurations of the control unit 30, the dehydrated solid matter monitoring system 50A, the separated liquid monitoring system 40, the processed liquid monitoring system 60, and the screw conveyor torque monitoring system 70 connected to the computer PC 1 are the other implementations already described above. It may be the same as that exemplified in the form of. The plurality of data acquired by the learning data set acquisition unit 21 in the present embodiment include the image data of the dehydrated solid M and the control parameters, the concentration of the separation liquid SL, and the slurry of the liquid to be treated PL1. At least one of the concentration and the torque value of the screw conveyor 3 and the characteristic amount of the dehydrated solid M can be included. In the present embodiment, as in the third embodiment, the computer PC 1 is connected to all of the separation liquid monitoring system 40, the liquid to be processed monitoring system 60, and the screw conveyor torque monitoring system 70 for learning. It is exemplified that the plurality of data acquired by the data set acquisition unit 21 include all of the concentration of the separation liquid SL, the slurry concentration of the liquid to be processed PL1, and the torque value of the screw conveyor 3.

The plurality of data acquired in the learning data set acquisition unit 21 are the first learning data set storage unit 221 and the fourth learning data set as two learning data sets while considering their respective correspondences. It may be stored separately in the storage unit 224. The first learning data set stored in the first learning data set storage unit 221 is the image data of the dehydrated solid M obtained by imaging the dehydrated solid M discharged from the discharge port 2a from a predetermined angle. It may be included as the input data of No. 1 and may include the feature amount of the dehydrated solid M associated with the first input data as the first output data. Further, the fourth learning data set stored in the fourth learning data set storage unit 224 includes the feature amount of the dehydrated solid M, the concentration of the separation liquid SL, and the slurry concentration of the liquid to be treated PL1. The torque value of the screw conveyor 3 may be included as the fourth input data, and the control parameter associated with the fourth input data may be included as the fourth output data. When dividing and generating two learning data sets from a plurality of data in this way, for example, the image data of the dehydrated solid M associated with the same dehydrated solid M, the concentration of the separation liquid SL, and the object to be processed. The slurry concentration of the liquid PL1, the torque value of the screw conveyor 3, the feature amount and the control parameter of the dehydrated solid M, the set of the image data of the dehydrated solid M and the feature amount of the dehydrated solid M, and the concentration of the separated liquid SL. , The slurry concentration of the liquid to be treated PL1, the torque value of the screw conveyor 3, the feature amount of the separation liquid SL, and the control parameters are divided into sets, which are used as one first and fourth learning data sets. Just do it.

The first and fourth learning data sets stored in the first learning data set storage unit 221 and the fourth learning data set storage unit 224, respectively, are referred only to different learning units. It may be there. The first learning unit 231 learns a learning model for inferring the correlation between the first input data and the first output data by inputting a plurality of sets of the first learning data sets. good. In other words, the first learning unit 231 learns the first learning model for inferring the feature amount of the image data by inputting the image data of the dehydrated solid M in the first learning data set. It may be something to do. Then, the fourth learning unit 234 learns a learning model that infers the correlation between the fourth input data and the fourth output data by inputting a plurality of sets of the fourth learning data sets. It may be there. In other words, the fourth learning unit 234 has the feature amount of the image data in the fourth learning data set, the concentration of the separation liquid SL, the slurry concentration of the liquid to be treated PL1, and the torque value of the screw conveyor 3. By inputting and, the fourth learning model for inferring the control parameter may be learned.

The specific machine learning methods in the first and

fourth learning units

231 and 234 are the same as the supervised learning process shown in FIG. 6, although the learning data set used for learning is different. You can do it. Then, the first and fourth trained models obtained through a series of machine learning steps can be stored in the trained model storage unit 24, respectively.

FIG. 14 is a schematic block diagram showing a data processing system 80C according to a fourth embodiment of the present disclosure. As shown in FIG. 14, the data processing system 80C according to the present embodiment includes a first inference unit 871 and a fourth inference unit 874 as inference units in the arithmetic unit 83, and is an additional variable acquisition unit. Other than the point including 89, it may include the same components as the data processing system 80 according to the first embodiment described above.

As shown in FIG. 14, the additional variable acquisition unit 89 can be connected to the separation liquid monitoring system 40, the processing liquid monitoring system 60, and the screw conveyor torque monitoring system 70, and from these monitoring systems, the concentration of the separation liquid SL and the subject. The slurry concentration of the processing liquid PL1 and the torque value of the screw conveyor 3 may be acquired. Further, the first inference unit 871 may execute inference using the first trained model generated by the machine learning device 20C described above and stored in the trained model storage unit 88. Therefore, when the image data of the dehydrated solid M acquired by the first image data acquisition unit 81 is input, the first inference unit 871 can output the feature amount of the image data to the output layer. can. Further, the fourth inference unit 874 may execute inference using the fourth trained model generated by the machine learning device 20C described above and stored in the trained model storage unit 88. Therefore, the fourth inference unit 874 has the feature amount of the image data inferred by the first inference unit 871, the concentration of the separation liquid SL, the slurry concentration of the liquid to be treated PL1, and the torque value of the screw conveyor 3. When is input, the control parameters can be output to the output layer.

When the control parameters are adjusted in the decanter 1 to which the data processing system 80C including the additional variable acquisition unit 89 described above and the first and

fourth inference units

871 and 874 are applied, the control parameters are adjusted as shown in FIG. The same processing may be performed. However, among the processes shown in FIG. 8, the additional variable acquisition unit 89 acquires the concentration of the separation liquid SL, the slurry concentration of the liquid to be treated PL1, and the torque value of the screw conveyor 3 at the same time as or at a predetermined timing before and after step S23. Further includes steps to be performed. Further, when inferring the control parameters in step S24, first, the image data of the dehydrated solid M is input to the first inference unit 871, and the feature amount of the output image data is acquired by the additional variable acquisition unit 89. It is input to the fourth inference unit 874 together with the two values.

As described above, according to the machine learning device, the machine learning method, and the data processing system according to the present embodiment, the training data set required to generate each trained model capable of inferring with sufficient accuracy. It can be expected that the number will be further suppressed as compared with the second and third embodiments described above.

<Fifth Embodiment>
The image data included in the input data in the first to fourth embodiments described above is only the image data of the dehydrated solid M. However, the image data that can be adopted as the input data exists in addition to the image data of the dehydrated solid matter M. Therefore, as the fifth embodiment, the image data included in the input data including the image data of the floc-containing liquid to be treated PL2 in addition to the image data of the dehydrated solid M will be described below. Of the components of the machine learning device 120 and the data processing system 180 according to the fifth embodiment shown below, the machine learning devices 20, 20A to 20C, and the machine learning devices 20, 20A to 20C according to the first to fourth embodiments. , The same reference numerals are given to the components common to the components of the data processing systems 80, 80A to 80C, and the description thereof will be omitted. Further, the machine learning device, the machine learning method, and the data processing system according to the fifth embodiment shown below are the same as those according to the first to fourth embodiments, and the centrifuge shown in FIGS. 1 and 2. The explanation is given by taking the case of applying to a separation system as an example. Further, all the modifications described in the above-mentioned or later embodiments can be applied to the present embodiment as long as there is no contradiction.

FIG. 15 is a schematic block diagram of the machine learning device 120 according to the fifth embodiment of the present disclosure. As shown in FIG. 15, the machine learning device 120 according to the present embodiment has learned the learning data set acquisition unit 121, the fifth learning data set storage unit 225, and the fifth learning unit 235. It may include a model storage unit 124.

The learning data set acquisition unit 121 is the same as the learning data set acquisition unit 21 shown in the first embodiment, and a plurality of learning data sets constituting the learning data set via a wired or wireless communication line. It may be an interface unit that acquires the data of. In this learning data set acquisition unit 121, as a plurality of data, the first image data in which the dehydrated solid M is imaged, the second image data in which the floc-containing liquid to be treated PL2 is imaged, and these image data are used. The associated control parameters can be acquired. Further, as the control parameters according to the present embodiment, those including the supply amount of the chemicals added to the liquid to be treated PL1, the centrifugal force of the bowl 2, and the differential speed controlled by the differential speed generator 5 are exemplified.

An example of an acquisition method of a plurality of data acquired by the learning data set acquisition unit 121 will be described. As shown in FIG. 15, the learning data set acquisition unit 121 can acquire desired data from the computer PC1 by being connected to the computer PC1. The computer PC 1 is directly or indirectly connected to the liquid to be monitored system 60 installed in the liquid to be treated pipe 10b in addition to the control unit 30 and the dehydrated solid matter monitoring system 50, either directly or via the control unit 30. It is good to be there. Thereby, each control parameter can be acquired from the control unit 30, the first image data can be acquired from the dehydrated solid matter monitoring system 50, and the second image data can be acquired from the liquid to be processed monitoring system 60.

Here, as shown in FIG. 15, the liquid to be monitored system 60 according to the present embodiment is connected to an arbitrary position downstream of the connection position of the liquid to be treated 10b with at least the additive pipe 11c. A camera 63 for capturing the floc-containing liquid to be processed, which is installed at a predetermined angle on the sight glass 61 and the window 62 provided in the sight glass 61 and can capture the image data of the floc-containing liquid to be treated PL2 flowing in the sight glass 61. Those including and can be adopted. Of these, a well-known camera capable of capturing a two-dimensional image can be adopted as the flock-containing liquid to be image-imaging camera 63. Further, in the liquid to be processed monitoring system 60, the flock-containing subject in the sight glass 61 is easily reflected in the image data captured by the camera 63 for capturing the flock-containing liquid to be treated. A light source (not shown) that illuminates the treatment liquid PL2 and a polarization filter (not shown) that can be attached to the flock-containing liquid to be processed image capturing camera 63 can be appropriately adopted. The configuration of the liquid to be treated monitoring system 60 is not limited to this, and various configurations can be adopted as long as the image data of the liquid to be treated PL2 containing flocs can be acquired. As a specific example, for example, when the drug is supplied via the additive pipe 11c connected to the liquid supply pipe 9, the flock-containing liquid to be imaged camera 63 images the inside of the space 3c inside the screw conveyor 3. A configuration arranged at a possible position may be adopted, or a configuration such as the liquid to be monitored system 60A shown in FIG. 19 described later may be adopted.

FIG. 16 shows an example of image data obtained by the flock-containing liquid to be image imaging camera 63 of the liquid to be treated monitoring system 60 according to the fifth embodiment of the present disclosure. By the way, when the solid-liquid separation treatment is performed by the centrifugal separation system, if the solid component in the liquid to be treated to be charged into the decanter 1 is sufficiently aggregated by the chemical solution, the solid component and the liquid component can be satisfactorily separated. Is generally known. Further, in the second image data obtained by the liquid to be treated monitoring system 60, if the floc aggregate of the floc-containing liquid to be treated PL2 is well performed, one floc aggregate becomes coarse and the floc aggregate becomes coarse. Areas where there is no sol tend to be more transparent. Therefore, in the image data (for example, the one shown in FIG. 16B) in which the floc agglutination reaction in the floc-containing liquid to be treated PL2 is satisfactorily advanced, the reaction is satisfactorily advanced due to, for example, insufficient drug supply. It can be inferred that the amount of change in color between pixels tends to increase as compared with the image data that does not exist (for example, the one shown in FIG. 16B).

The fifth learning data set storage unit 225 associates a plurality of data constituting the learning data set acquired by the learning data set acquisition unit 121 with related input data and output data to form one learning data. It may be a database for storing as a set. As the fifth learning data set stored in the fifth learning data set storage unit 225 in the present embodiment, the first image of the dehydrated solid material M discharged from the discharge port 2a is taken from a predetermined angle. And the second image data obtained by capturing the floc-containing liquid to be treated PL2 from a predetermined angle of view before being supplied to the bowl 2 and after the drug is added, as the fifth input data. The control parameter associated with the image data as the fifth input data can be the fifth output data. The control parameters include at least one of the centrifugal force of the bowl 2, the supply amount of the drug from the additive supply source 11, and the difference speed between the bowl 2 and the screw conveyor 3, as in the first embodiment described above. It may include one.

The fifth learning unit 235 can input the fifth input data and the fifth output data by inputting a plurality of sets of the fifth learning data set stored in the fifth learning data set storage unit 225. It may be one that learns a learning model that infers the correlation between them. Also in this embodiment, as in the first embodiment described above, supervised learning using a neural network is adopted as a specific method of machine learning. The specific machine learning method in the fifth learning unit 235 is the same as the machine learning method according to the first embodiment shown in FIG. 6, although the learning data set used for learning is different. You can do it. Further, the trained model storage unit 124 may be a database for storing the trained model generated by the fifth learning unit 235.

FIG. 17 is a schematic block diagram showing a data processing system 180 according to the fifth embodiment of the present disclosure. As shown in FIG. 17, the data processing system 180 according to the present embodiment mainly includes a first image data acquisition unit 81, a second image data acquisition unit 181, a parameter adjustment unit 82, and an arithmetic unit 183. , A database 84, a user interface 85, and an internal bus 86. Among the above-mentioned components, the first image data acquisition unit 81, the parameter adjustment unit 82, the database 84, the user interface 85, and the internal bus 86 are the data processing system 80 according to the first embodiment described above. The same thing as the one can be adopted. Therefore, these components are designated by the same reference numerals as those of the first embodiment, and the description thereof will be omitted.

The second image data acquisition unit 181 may be for acquiring the second image data. Specifically, it is connected to the liquid to be processed monitoring system 60 and acquires the second image data which is the image data of the liquid to be treated PL2 containing flock imaged by the camera 63 for capturing the liquid to be treated containing flock. It may be there.

The arithmetic unit 183 may constitute a processor for realizing various processes in the data processing system 180, and may include at least a fifth inference unit 875. Further, the arithmetic unit 183 is connected to a trained model storage unit 188 that stores a trained model used in the fifth inference unit 875, that is, a fifth trained model learned in the fifth learning unit 235. ing. The fifth inference unit 875 is the second as a state variable acquired by the first and second image

data acquisition units

81 and 181 by taking into consideration one trained model stored in the trained model storage unit 188. The control parameters to be adjusted by the parameter adjustment unit 82 may be inferred from the first and second image data.

FIG. 18 is a flowchart showing an example of parameter adjustment by the data processing system according to the fifth embodiment of the present disclosure. When the drive of the decanter 1 to which the data processing system 180 is applied is started, as shown in FIG. 18, the data processing system 180 first determines whether or not it is the timing of parameter adjustment (step S21). When it is detected that it is the timing of parameter adjustment (Yes in step S21), the dehydrated solid image imaging camera 53 in the dehydrated solid monitoring system 50 and the floc-containing processed liquid imaging camera in the processed liquid monitoring system 60 are detected. 63 is operated to image the dehydrated solid M passing through the solid discharge conduit 7 and the floc-containing liquid to be treated PL2 flowing in the liquid pipe 10b to be treated (step S221). The first and second image data captured here are acquired in the data processing system 180 by the first and second image data acquisition units 81 and 181 (step S231).

Next, the first and second image data acquired by the first and second image

data acquisition units

81 and 181 are sent to the fifth inference unit 875 via the internal bus 86, and these image data are sent to the fifth inference unit 875. The control parameter is inferred by being input to the input layer of one trained model in the trained model storage unit 188 previously specified in the inference unit 875 of the fifth (step S24). Then, the parameter adjustment unit 82 adjusts each control unit using the value of the control parameter inferred here (step S25). After that, the process returns to step S21 to enter the standby state.

As described above, according to the machine learning device 120 and the machine learning method according to the present embodiment, by adopting two image data as input data, the input data can be compared with the case of only one image data. It is expected that the correlation with the output data will be easy to learn, and it will be possible to generate each trained model that can be inferred with sufficient accuracy with a relatively small number of training data sets. Further, according to the data processing system 180 according to the present embodiment, preferable control parameters of the decanter 1 can be estimated from only two image data, that is, the image data of the dehydrated solid M and the image data of the floc-containing liquid to be treated PL2. Therefore, this data processing system 180 can be applied to the decanter 1 relatively easily. Then, the adjustment of the control parameter can be easily performed without depending on the judgment of the operator EN.

<Sixth Embodiment>
In the machine learning device, the machine learning method, and the data processing system according to the fifth embodiment, the one inferring the control parameters from the first and second image data using one trained model will be described. rice field. In the following, as a modification of the fifth embodiment, the sixth embodiment of the present disclosure is a case where a plurality of trained models are used to infer control parameters from the first and second image data. The form of the above will be described below. Of the components of the machine learning device 120A and the data processing system 180A according to the sixth embodiment shown below, each component of the machine learning device 120 and the data processing system 180 according to the fifth embodiment. The same reference numerals are given to those common to the above, and the description thereof will be omitted. Further, the machine learning device, the machine learning method, and the data processing system according to the sixth embodiment shown below are the same as those according to the first to fifth embodiments, and the centrifuge shown in FIGS. 1 and 2. The explanation is given by taking the case of applying to a separation system as an example. Furthermore, all the modifications described in each of the above-mentioned or later embodiments can be applied to the present embodiment as long as there is no contradiction.

FIG. 19 is a schematic block diagram of the machine learning device 120A according to the sixth embodiment of the present disclosure. As shown in FIG. 19, the machine learning device 120A according to the present embodiment has, as its components, a first learning data set storage unit 221 and a sixth learning data as a learning data set storage unit. A fifth, except that it includes a set storage unit 226 and a seventh learning data set storage unit 227, and includes a first learning unit 231 and a sixth learning unit 236 and a seventh learning unit 237 as learning units. It may be the same as the machine learning device 120 according to the embodiment. Further, in connection with this, the plurality of data acquired by the learning data set acquisition unit 121 may also be different from those of the fifth embodiment. The first learning data set storage unit 221 and the first learning unit 231 can be the same as those described in the machine learning device 20A according to the second embodiment described above. , The same reference numerals as those of the second embodiment are designated, and the description thereof will be omitted.

The plurality of data acquired by the learning data set acquisition unit 121 according to the present embodiment includes the first and second image data, the control parameters, the feature amount of the dehydrated solid M, and the flocs. The feature amount of the second image data obtained by imaging the liquid to be treated PL2 (hereinafter, also referred to as "feature amount of the floc-containing liquid to be treated PL2") may be included. As the feature amount of the flock-containing liquid to be treated PL2, various information specified based on the imaged flock-containing liquid to be treated PL2 can be adopted. Specifically, for example, the size, color value, density, etc. of floc aggregates can be included one or more. This feature amount can be specified, for example, by actually sampling the floc-containing liquid to be treated PL2 and measuring and analyzing it using various measuring instruments or the like, or by visually determining by the operator EN.

In order to measure the feature amount described above, the liquid to be treated monitoring system 60A according to the present embodiment can acquire the second image data after sampling and extracting the floc-containing liquid to be treated PL2. It is preferable to adopt. Specifically, as the liquid to be treated monitoring system 60A, the branch pipe 64 provided in the liquid to be treated pipe 10b, the valve 65 provided in the branch pipe 64, and the end of the branch pipe 64 are arranged. A transparent container 66 and a flock-containing liquid to be image-imaging camera 63 installed on the side surface of the transparent container 66 can be adopted. The arrangement of the flock-containing liquid to be processed image imaging camera 63 is not limited to this, and for example, the container 66 may be installed at a position where imaging can be performed from above. In the liquid to be treated monitoring system 60A having such a configuration, the valve 65 is opened at an arbitrary timing to sample and extract the floc-containing liquid to be treated PL2 flowing in the liquid to be treated pipe 10b into a transparent container 66. be able to. Then, the image data of the flock-containing liquid to be treated PL2 can be generated by imaging the flock-containing liquid to be treated PL2 in the container 66 with the flock-containing liquid to be image-imaging camera 63, and the corresponding desired control parameter is the operator EN. Can be identified by. Further, by measuring and analyzing the floc-containing liquid to be treated PL2 sampled in the container 66, the characteristic amount of the floc-containing liquid to be treated PL2 can be specified.

The plurality of data acquired in the learning data set acquisition unit 121 are the first learning data set storage unit 221 and the sixth learning data set as three learning data sets while considering their respective correspondences. It may be stored separately in the storage unit 226 and the seventh learning data set storage unit 227. The first learning data set stored in the first learning data set storage unit 221 captures the first image data obtained by capturing the dehydrated solid M discharged from the discharge port 2a from a predetermined angle. It may be included as input data and may include the feature amount of the dehydrated solid M associated with the first input data as the first output data. Further, the sixth learning data set stored in the sixth learning data set storage unit 226 contains the floc-containing liquid to be treated PL2 before being supplied to the bowl 2 and after the drug is added. The second image data captured from a predetermined angle is included as the sixth input data, and the feature amount of the floc-containing liquid to be treated PL2 associated with the sixth input data is included as the sixth output data. It's okay. Further, in the seventh learning data set stored in the seventh learning data set storage unit 227, the feature amount of the dehydrated solid M and the feature amount of the floc-containing liquid to be treated PL2 are used as the seventh input data. It may include and include the control parameter associated with the seventh input data as the seventh output data.

The first, sixth and seventh learning data sets stored in the first learning data set storage unit 221 and the sixth learning data set storage unit 226 and the seventh learning data set storage unit 227, respectively. May only be referenced by different learning units. The first learning unit 231 learns a learning model that infers the correlation between the first input data and the first output data by inputting a plurality of sets of the first learning data sets. good. In other words, the first learning unit 231 learns the first learning model that infers the feature amount of the first image data by inputting the first image data in the first learning data set. It may be something to do. Further, the sixth learning unit 236 learns a learning model for inferring the correlation between the sixth input data and the sixth output data by inputting a plurality of sets of the sixth learning data sets. It may be there. In other words, the sixth learning unit 236 learns a sixth learning model that infers the feature amount of the second image data by inputting the second image data in the sixth learning data set. It may be something to do. Further, the seventh learning unit 237 learns a learning model for inferring the correlation between the seventh input data and the seventh output data by inputting a plurality of sets of the seventh learning data sets. It may be there. In other words, the seventh learning unit 237 learns the seventh learning model that infers the control parameters by inputting the features of the first and second image data in the seventh learning data set. It may be something to do.

Although the specific machine learning methods in the first, sixth and

seventh learning units

231, 236 and 237 differ from the learning data set used for learning, the process is different from the supervised learning process shown in FIG. Both may be the same. Then, the first, sixth, and seventh trained models obtained through a series of machine learning steps can be stored in the trained model storage unit 124, respectively.

FIG. 20 is a schematic block diagram showing a data processing system 180A according to the sixth embodiment of the present disclosure. As shown in FIG. 20, the data processing system 180A according to the present embodiment includes a first inference unit 871, a sixth inference unit 876, and a seventh inference unit 877 as inference units in the arithmetic unit 183. Other than the points, it may include the same components as the data processing system 180 according to the fifth embodiment described above.

The first inference unit 871 may execute inference using the first trained model generated by the machine learning device 120A described above and stored in the trained model storage unit 188. Therefore, when the first image data acquired by the first image data acquisition unit 81 is input, the first inference unit 871 can output the feature amount of the first image data to the output layer. can. Further, the sixth inference unit 876 may execute inference using the sixth trained model generated by the machine learning device 120A described above and stored in the trained model storage unit 188. Therefore, when the second image data acquired by the second image data acquisition unit 181 is input, the sixth inference unit 876 can output the feature amount of the second image data to the output layer. can. Further, the seventh inference unit 877 may execute inference using the seventh inference model generated by the machine learning device 120A described above and stored in the trained model storage unit 188. Therefore, the seventh inference unit 877 outputs a control parameter when the feature amounts of the first and second image data inferred by the first inference unit 871 and the sixth inference unit 876 are input. Can be output to.

When adjusting the control parameters in the decanter 1 to which the data processing system 180A including the first, sixth and

seventh inference units

871, 876 and 877 described above is applied, the same as that shown in FIG. 18 described above is used. All you have to do is process it. However, in the process shown in FIG. 18, when the control parameter is inferred in step S24, the first and second image data are first input to the first inference unit 871 and the sixth inference unit 876 and output. The feature quantities of the first and second image data may be input to the seventh inference unit 877.

As described above, according to the machine learning device, the machine learning method, and the data processing system according to the present embodiment, by obtaining the inference results of the control parameters using a plurality of trained models, it is possible to infer with sufficient accuracy. It can be expected that the number of training data sets required to generate each possible trained model will be relatively small.

<7th embodiment>
In the machine learning device, the machine learning method, and the data processing system according to the fifth embodiment, the case where the control parameter is inferred only from the first and second image data has been described. In the following, as a modification of the fifth embodiment, a case where other variables are adopted in addition to the first and second image data as input data is described in the seventh embodiment of the present disclosure. Will be described below. Of the components of the machine learning device 120B and the data processing system 180B according to the seventh embodiment shown below, the machine learning devices 120, 120A and the data processing system according to the fifth and sixth embodiments. Those common to the components of 180 and 180A are designated by the same reference numerals and the description thereof will be omitted. Further, the machine learning device, the machine learning method, and the data processing system according to the seventh embodiment shown below are the same as those according to the first to sixth embodiments, and the centrifuge shown in FIGS. 1 and 2. The explanation is given by taking the case of applying to a separation system as an example. Furthermore, all the modifications described in each of the above-mentioned or later embodiments can be applied to the present embodiment as long as there is no contradiction.

FIG. 21 is a schematic block diagram of the machine learning device 120B according to the seventh embodiment of the present disclosure. As shown in FIG. 21, the machine learning device 120B according to the present embodiment includes an eighth learning data set storage unit 228 as a learning data set storage unit for each component, and an eighth learning data set storage unit 228 as a learning unit. It may be the same as the machine learning device 120 according to the fifth embodiment described above except that the learning unit 238 is included. Further, the plurality of data acquired by the learning data set acquisition unit 121 may also be different from those of the fifth embodiment. As shown in FIG. 21, the learning data set acquisition unit 121 according to the present embodiment can acquire desired data from the computer PC1 by being connected to the computer PC1. The computer PC 1 is connected to the control unit 30, the dehydrated solid matter monitoring system 50, and the liquid to be treated monitoring system 60, similarly to the computer PC 1 according to the fifth embodiment. In addition to these, the separation liquid is connected. It may also be connected to at least one of the monitoring system 40, the liquid concentration sensor 67 to be processed and the screw conveyor torque monitoring system 70. Of these, the liquid to be treated concentration sensor 67 is arranged at a predetermined position, for example, before merging with the additive pipe 11c of the liquid to be treated 10b, and is a slurry of the liquid to be treated PL1 supplied from the liquid supply source 10 to be treated. The concentration may be measured directly. As the liquid to be treated concentration sensor 67, for example, a well-known laser type, optical type, or ultrasonic type concentration sensor can be used. In the present embodiment, the computer PC 1 is connected to all of the separation liquid monitoring system 40, the liquid to be processed concentration sensor 67, and the screw conveyor torque monitoring system 70, and a plurality of data set acquisition units 121 for learning acquire. The data includes the slurry concentration of the liquid to be treated PL1, the concentration of the separation liquid SL, and the torque value of the screw conveyor 3 to be exemplified. Further, the separation liquid monitoring system 40 and the screw conveyor torque monitoring system 70 may be the same as those exemplified in the above-mentioned third embodiment, and detailed description thereof will be omitted here. Further, the liquid to be treated concentration sensor 67 is not shown in FIG. 2.

Various data acquired by these monitoring systems and the like, specifically, the first and second image data, the concentration of the separation liquid SL, the slurry concentration of the liquid to be treated PL1, and the torque value of the screw conveyor 3 are directly or controlled. It is sent to the computer PC1 via the unit 30. Then, it is sent from the computer PC 1 to the learning data set acquisition unit 121 together with the corresponding desired control parameters. The plurality of data constituting the learning data set acquired by the learning data set acquisition unit 121 include the first and second image data, the concentration of the separation liquid SL, the slurry concentration of the liquid to be treated PL1, and the screw conveyor. Eighth training data set storage in the form of an eighth training data set in which the torque value of 3 is used as the eighth input data and the control parameters associated with these data are used as the eighth output data. It may be stored in unit 228. Then, in the eighth learning unit 238, machine learning is performed by the same method as the machine learning method of the fifth embodiment described above using the eighth learning data set, and the obtained eighth learning unit is obtained. The trained model is stored in the trained model storage unit 124.

FIG. 22 is a schematic block diagram showing a data processing system 180B according to the seventh embodiment of the present disclosure. As shown in FIG. 22, the data processing system 180B according to the present embodiment has the fifth embodiment described above except that the inference unit includes the eighth inference unit 878 and the additional variable acquisition unit 89. It may include the same components as the data processing system 180 according to the form. Of these, the additional variable acquisition unit 89 is connected to the separation liquid monitoring system 40, the liquid to be processed concentration sensor 67, and the screw conveyor torque monitoring system 70, as shown in FIG. 22, from which the concentration of the separation liquid SL and the subject are to be covered. The slurry concentration of the treatment liquid PL1 and the torque value of the screw conveyor 3 may be acquired.

When adjusting the control parameters in the decanter 1 to which the data processing system 180B is applied, the same processing as that shown in FIG. 18 may be performed. However, among the processes shown in FIG. 18, the additional variable acquisition unit 89 acquires the concentration of the separation liquid SL, the slurry concentration of the liquid to be processed PL1, and the torque value of the screw conveyor 3 at the same time as or at a predetermined timing before and after step S231. do. Then, these values are associated with the input layer of the trained model in the eighth inference unit 878 together with the first and second image data.

As described above, according to the machine learning device, the machine learning method, and the data processing system according to the present embodiment, as input data, in addition to the first and second image data, the concentration of the separation liquid SL and the data to be processed. By adopting the slurry concentration of the liquid PL1 and the torque value of the screw conveyor 3, it becomes easy to identify the correlation between the input data and the output data in the learning stage of the neural network model. This can be expected to reduce the number of training data sets required to generate each trained model that can be reasoned with sufficient accuracy.

<8th embodiment>
Next, the case where the technical ideas described in the sixth and seventh embodiments are combined will be described below as the eighth embodiment of the present disclosure. Of the components of the machine learning device 120C and the data processing system 180C according to the eighth embodiment shown below, the machine learning devices 120, 120A, 120B and data according to the fifth to seventh embodiments. Those common to the components of the processing systems 180, 180A, and 180B are designated by the same reference numerals and the description thereof will be omitted. Further, the machine learning device, the machine learning method, and the data processing system according to the eighth embodiment shown below are the same as those according to the first to seventh embodiments, and the centrifuge shown in FIGS. 1 and 2. The explanation is given by taking the case of applying to a separation system as an example. Furthermore, all the modifications described in each of the above-mentioned or later embodiments can be applied to the present embodiment as long as there is no contradiction.

FIG. 23 is a schematic block diagram of the machine learning device 120C according to the eighth embodiment of the present disclosure. As shown in FIG. 23, the machine learning device 120C according to the present embodiment has, as its components, a first learning data set storage unit 221 and a sixth learning data as a learning data set storage unit. No. 1 except that the set storage unit 226 and the ninth learning data set storage unit 229 are included, and the learning units include the first learning unit 231 and the sixth learning unit 236 and the ninth learning unit 239. It may be the same as the machine learning device 120 according to the embodiment of 5. Further, in connection with this, the plurality of data acquired by the learning data set acquisition unit 121 may also be different from the fifth embodiment.

As shown in FIG. 23, the learning data set acquisition unit 121 according to the present embodiment can acquire desired data from the computer PC1 by being connected to the computer PC1. The computer PC1 is connected to the control unit 30, the dehydrated solid matter monitoring system 50A, and the liquid to be treated monitoring system 60A, similarly to the computer PC1 according to the fifth embodiment. In addition to these, the separation liquid is connected. It may also be connected to at least one of the monitoring system 40, the liquid concentration sensor 67 to be processed and the screw conveyor torque monitoring system 70. Specific configurations of the control unit 30, the dehydrated solid matter monitoring system 50A, the processed liquid monitoring system 60A, the separated liquid monitoring system 40, the processed liquid concentration sensor 67, and the screw conveyor torque monitoring system 70 connected to the computer PC1 are. , Which may be the same as those already exemplified in the other embodiments described above. The plurality of data acquired by the learning data set acquisition unit 121 in the present embodiment include the first and second image data, the control parameters, the concentration of the separation liquid SL, and the liquid to be processed PL1. It may contain at least one of the slurry concentration and the torque value of the screw conveyor 3, the characteristic amount of the dehydrated solid M, and the characteristic amount of the floc-containing liquid to be treated PL2. In the present embodiment, as in the seventh embodiment, the computer PC 1 is connected to all of the separation liquid monitoring system 40, the liquid to be processed concentration sensor 67, and the screw conveyor torque monitoring system 70 for learning. Examples of the plurality of data acquired by the data set acquisition unit 121 include the concentration of the separation liquid SL, the slurry concentration of the liquid to be treated PL1, and the torque value of the screw conveyor 3.

The plurality of data acquired in the learning data set acquisition unit 121 are the first learning data set storage unit 221 and the sixth learning data set as three learning data sets while considering their respective correspondences. It may be stored separately in the storage unit 226 and the ninth learning data set storage unit 229. The first learning data set stored in the first learning data set storage unit 221 captures the first image data obtained by capturing the dehydrated solid M discharged from the discharge port 2a from a predetermined angle. The feature amount of the first image data associated with the first input data may be included as the first output data. Further, the sixth learning data set stored in the sixth learning data set storage unit 226 contains a floc-containing liquid to be treated PL2 before being supplied to the bowl 2 and to which a chemical is added. The second image data captured from the corner may be included as the sixth input data, and the feature amount of the second image data associated with the sixth input data may be included as the sixth output data. Further, the ninth learning data set stored in the ninth learning data set storage unit 229 includes the feature amount of the first image data, the feature amount of the second image data, and the concentration of the separation liquid SL. The slurry concentration of the liquid to be treated PL1 and the torque value of the screw conveyor 3 are included as the ninth input data, and the control parameter associated with the ninth input data is included as the ninth output data. It's okay.

The first, sixth and ninth learning data sets stored in the first learning data set storage unit 221 and the sixth learning data set storage unit 226 and the ninth learning data set storage unit 229, respectively. May only be referenced by different learning units. The first learning unit 231 learns a learning model that infers the correlation between the first input data and the first output data by inputting a plurality of sets of the first learning data sets. good. In other words, the first learning unit 231 learns the first learning model that infers the feature amount of the first image data by inputting the first image data in the first learning data set. It may be something to do. Further, the sixth learning unit 236 learns a learning model for inferring the correlation between the sixth input data and the sixth output data by inputting a plurality of sets of the sixth learning data sets. It may be there. In other words, the sixth learning unit 236 learns a sixth learning model that infers the feature amount of the second image data by inputting the second image data in the sixth learning data set. It may be something to do. Then, the ninth learning unit 239 learns a learning model that infers the correlation between the ninth input data and the ninth output data by inputting a plurality of sets of the ninth learning data sets. It may be there. In other words, the ninth learning unit 239 has the feature amounts of the first and second image data in the ninth learning data set, the concentration of the separation liquid SL, the slurry concentration of the liquid to be treated PL1, and the screw. By inputting the torque value of the conveyor 3, the ninth learning model for inferring the control parameter may be learned.

The specific machine learning methods in the first, sixth and

ninth learning units

231 and 236 and 239 differ from the supervised learning process shown in FIG. 6, although the learning data set used for learning is different. Both may be the same. Then, the first, sixth, and ninth trained models obtained through a series of machine learning steps can be stored in the trained model storage unit 124, respectively.

FIG. 24 is a schematic block diagram showing a data processing system 180C according to the eighth embodiment of the present disclosure. As shown in FIG. 24, the data processing system 180C according to the present embodiment includes a first inference unit 871, a sixth inference unit 876, and a ninth inference unit 879 as inference units in the arithmetic unit 183. It may include the same components as the data processing system 180 according to the fifth embodiment described above, except that it includes the additional variable acquisition unit 89. Of these, as the additional variable acquisition unit 89, the same one as described in the seventh embodiment can be adopted.

The first inference unit 871 may execute inference using the first trained model generated by the machine learning device 120C described above and stored in the trained model storage unit 188. Therefore, when the first image data acquired by the first image data acquisition unit 81 is input, the first inference unit 871 can output the feature amount of the first image data to the output layer. can. Further, the sixth inference unit 876 may execute inference using the sixth trained model generated by the machine learning device 120C described above and stored in the trained model storage unit 188. Therefore, when the second image data acquired by the second image data acquisition unit 181 is input, the sixth inference unit 876 can output the feature amount of the second image data to the output layer. can. Further, the ninth inference unit 879 may execute inference using the ninth trained model generated by the machine learning device 120C described above and stored in the trained model storage unit 188. Therefore, the ninth inference unit 879 includes the feature amount of the first image data inferred by the first inference unit 871 and the feature amount of the second image data inferred by the sixth inference unit 876. When the concentration of the separation liquid SL, the slurry concentration of the liquid to be treated PL1 and the torque value of the screw conveyor 3 are input, the control parameters can be output to the output layer.

When the control parameters are adjusted in the decanter 1 to which the data processing system 180C including the additional variable acquisition unit 89 and the first, sixth and

ninth inference units

871, 876 and 879 is applied, the above figure is shown. The same processing as that shown in 18 may be performed. However, among the processes shown in FIG. 18, the additional variable acquisition unit 89 acquires the concentration of the separation liquid SL, the slurry concentration of the liquid to be processed PL1, and the torque value of the screw conveyor 3 at the same time as or at a predetermined timing before and after step S231. The steps to be performed can be further included. Further, when inferring the control parameters in step S24, first, the first and second image data are input to the first inference unit 871 and the sixth inference unit 876, respectively, and the first and second image data are output. The feature amount of the image data is input to the ninth inference unit 879 together with the three values acquired in the additional variable acquisition unit 89.

As described above, according to the machine learning device, the machine learning method, and the data processing system according to the present embodiment, the training data set required to generate each trained model capable of inferring with sufficient accuracy. It can be expected that the number will be further suppressed as compared with the sixth and seventh embodiments described above.

The above-described embodiment is merely an example, and therefore, the present disclosure is not limited to the above-mentioned embodiment, and various modifications are made without departing from the gist of the present disclosure. It is possible. And all of them are included in the technical idea of the present disclosure. Further, in the present disclosure, each component may be present alone or in combination of two or more as long as there is no contradiction.

All documents, including publications, patent applications and patents cited herein, are individually specifically shown, referenced and incorporated, and all of their content is described herein. To the same extent, refer to and incorporate here.

The use of nouns and similar demonstratives used in connection with the description of the invention (particularly in connection with the following claims) is not particularly pointed out herein or clearly inconsistent with the context. , Singular and plural. The phrases "prepare," "have," "include," and "include" are to be interpreted as open-ended terms (ie, meaning "including, but not limited to," unless otherwise noted). The description of the numerical range herein is intended solely to serve as an abbreviation for individually referring to each value within that range, unless otherwise noted herein. Each value is incorporated herein as if it were individually listed herein. All methods described herein can be performed in any suitable order, as long as they are not specifically pointed out herein or are clearly inconsistent with the context. All examples or exemplary phrases used herein (eg, "etc.") are intended solely to better illustrate the invention and set limits to the scope of the invention, unless otherwise stated. is not it. Nothing in the specification shall be construed as indicating an element not described in the claims as essential to the practice of the present invention.

In the present specification, preferred embodiments of the present invention are described, including the best embodiments known to the inventor for carrying out the present invention. For those skilled in the art, reading the above description will reveal variations of these preferred embodiments. The inventor expects an expert to apply such modifications as appropriate, and intends to implement the invention by methods other than those specifically described herein. .. Accordingly, the invention includes all amendments and equivalents of the content of the claims attached herein, as permitted by applicable law. Moreover, any combination of the above elements in all modifications is included in the invention unless specifically pointed out herein or is clearly inconsistent with the context.

Claims

A bowl that applies centrifugal force to the liquid to be treated to centrifuge into the solid and the separated liquid, a screw conveyor that conveys the solid in the bowl toward the discharge port, and a drive motor that rotates the bowl. A machine learning device for a centrifugal separation system, comprising a differential speed generator that rotates the screw conveyor with a differential speed relative to the bowl.
A plurality of sets of learning data sets including input data including image data obtained by capturing a liquid-containing solid matter discharged from the discharge port from a predetermined angle and output data including control parameters associated with the input data. A data set storage unit for learning to be stored, wherein the control parameters are the supply amount of the additive added to the liquid to be treated, the centrifugal force of the bowl, and the differential speed controlled by the differential speed generator. With the training dataset storage unit comprising at least one of them;
With a learning unit that learns a learning model that infers the correlation between the input data and the output data by inputting a plurality of sets of the training data sets;
A trained model storage unit that stores the learning model learned by the learning unit;
Machine learning device.
The learning unit is
With the first learning unit that learns the first learning model that infers the feature amount of the image data by inputting the image data in the training data set;
A second learning unit for learning a second learning model for inferring the control parameter by inputting a feature amount of the image data;
The machine learning device according to claim 1.
The input data further includes at least one of the concentration of the separation liquid, the slurry concentration of the liquid to be treated, and the torque value of the screw conveyor.
The machine learning device according to claim 1.
The input data further includes at least one of the concentration of the separation liquid, the slurry concentration of the liquid to be treated, and the torque value of the screw conveyor.
The learning unit is
With the first learning unit that learns the first learning model that infers the feature amount of the image data by inputting the image data in the training data set;
By inputting the feature amount of the image data, the concentration of the separation liquid, the slurry concentration of the liquid to be processed, and the torque value of the screw conveyor, a fourth learning model for inferring the control parameter is learned. With a fourth learning unit;
The machine learning device according to claim 1.
A bowl that applies centrifugal force to the liquid to be treated to centrifuge into the solid and the separated liquid, a screw conveyor that conveys the solid in the bowl toward the discharge port, and a drive motor that rotates the bowl. A machine learning device for a centrifugal separation system, comprising a differential speed generator that rotates the screw conveyor with a differential speed relative to the bowl.
The first image data obtained by imaging the liquid-containing solid matter discharged from the discharge port from a predetermined angle, and the liquid to be treated before being supplied to the bowl and after the predetermined additive is added. Data set storage for learning that stores a plurality of sets of training data sets including input data including a second image data captured from a predetermined angle of view and output data including control parameters associated with the input data. The unit, said control parameter, said learning data set storage comprising at least one of the supply of the additive, the centrifugal force of the bowl, and the differential speed controlled by the differential speed generator. With the unit;
With a learning unit that learns a learning model that infers the correlation between the input data and the output data by inputting a plurality of sets of the training data sets;
A trained model storage unit that stores the learning model learned by the learning unit;
Machine learning device.
The learning unit is
With the first learning unit that learns the first learning model that infers the feature amount of the first image data by inputting the first image data in the training data set;
With a sixth learning unit that learns a sixth learning model that infers the features of the second image data by inputting the second image data in the training data set;
A seventh learning unit that learns a seventh learning model for inferring the control parameter by inputting the feature amount of the first image data and the feature amount of the second image data; ,
The machine learning device according to claim 5.
The input data further includes at least one of the concentration of the separation liquid, the slurry concentration of the liquid to be treated, and the torque value of the screw conveyor.
The machine learning device according to claim 5.
The input data further includes at least one of the concentration of the separation liquid, the slurry concentration of the liquid to be treated, and the torque value of the screw conveyor.
The learning unit is
With the first learning unit that learns the first learning model that infers the feature amount of the first image data by inputting the first image data in the training data set;
With a sixth learning unit that learns a sixth learning model that infers the features of the second image data by inputting the second image data in the training data set;
By inputting the feature amount of the first image data, the feature amount of the second image data, the concentration of the separation liquid, the slurry concentration of the liquid to be processed, and the torque value of the screw conveyor. , A ninth learning unit that learns a ninth learning model that infers the control parameters;
The machine learning device according to claim 5.
The control parameter further comprises a supply of the liquid to be treated.
The machine learning device according to any one of claims 1 to 8.
The control parameter further comprises a dam set diameter of the bowl.
The machine learning device according to any one of claims 1 to 9.
A bowl that applies centrifugal force to the liquid to be treated to centrifuge into the solid and the separated liquid, a screw conveyor that conveys the solid in the bowl toward the discharge port, and a drive motor that rotates the bowl. A data processing system used in a centrifugal separation system, comprising a differential speed generator that rotates the screw conveyor with a differential speed relative to the bowl.
With the first image data acquisition unit for acquiring the first image data obtained by capturing the liquid-containing solid matter discharged from the discharge port from a predetermined angle of view;
By inputting the data acquired by the first image data acquisition unit into the trained model generated by the machine learning device according to claim 1 or 2, the control parameters of the centrifugal separation system are inferred. With an inference unit;
Data processing system.
A bowl that applies centrifugal force to the liquid to be treated to centrifuge into the solid and the separated liquid, a screw conveyor that conveys the solid in the bowl toward the discharge port, and a drive motor that rotates the bowl. A data processing system used in a centrifugal separation system, comprising a differential speed generator that rotates the screw conveyor with a differential speed relative to the bowl.
With the first image data acquisition unit for acquiring the first image data obtained by capturing the liquid-containing solid matter discharged from the discharge port from a predetermined angle of view;
An additional variable acquisition unit for acquiring at least one of the concentration of the separation liquid, the slurry concentration of the liquid to be treated, and the torque value of the screw conveyor;
By inputting the data acquired by the first image data acquisition unit and the additional variable acquisition unit into the trained model generated by the machine learning device according to claim 3 or 4, the centrifugal separation is performed. With an inference unit that infers system control parameters;
Data processing system.
A bowl that applies centrifugal force to the liquid to be treated to centrifuge into the solid and the separated liquid, a screw conveyor that conveys the solid in the bowl toward the discharge port, and a drive motor that rotates the bowl. A data processing system used in a centrifugal separation system, comprising a differential speed generator that rotates the screw conveyor with a differential speed relative to the bowl.
With the first image data acquisition unit for acquiring the first image data obtained by capturing the liquid-containing solid matter discharged from the discharge port from a predetermined angle of view;
With a second image data acquisition unit for acquiring a second image data obtained by imaging the liquid to be treated from a predetermined angle of view before being supplied to the bowl and after the predetermined additive is added. ;
By inputting the data acquired by the first image data acquisition unit and the second image data acquisition unit into the trained model generated by the machine learning device according to claim 5 or 6. With an inference unit that infers the control parameters of the centrifuge system;
Data processing system.
A bowl that applies centrifugal force to the liquid to be treated to centrifuge into the solid and the separated liquid, a screw conveyor that conveys the solid in the bowl toward the discharge port, and a drive motor that rotates the bowl. A data processing system used in a centrifugal separation system, comprising a differential speed generator that rotates the screw conveyor with a differential speed relative to the bowl.
With the first image data acquisition unit for acquiring the first image data obtained by capturing the liquid-containing solid matter discharged from the discharge port from a predetermined angle of view;
With a second image data acquisition unit for acquiring a second image data obtained by imaging the liquid to be treated from a predetermined angle of view before being supplied to the bowl and after the predetermined additive is added. ;
An additional variable acquisition unit for acquiring at least one of the concentration of the separation liquid, the slurry concentration of the liquid to be treated, and the torque value of the screw conveyor;
Data acquired by the first image data acquisition unit, the second image data acquisition unit, and the additional variable acquisition unit in the trained model generated by the machine learning device according to claim 7 or 8. With an inference unit that infers the control parameters of the centrifuge system by inputting;
Data processing system.
A bowl that applies centrifugal force to the liquid to be treated to centrifuge into the solid and the separated liquid, a screw conveyor that conveys the solid in the bowl toward the discharge port, and a drive motor that rotates the bowl. , A computer-based machine learning method for a centrifuge system comprising a differential speed generator that rotates the screw conveyor with a differential speed relative to the bowl.
A plurality of sets of learning data sets including input data including image data obtained by capturing a liquid-containing solid matter discharged from the discharge port from a predetermined angle and output data including control parameters associated with the input data. The control parameter is at least one of the supply amount of the additive added to the liquid to be treated, the centrifugal force of the bowl, and the differential speed controlled by the differential speed generator. With steps and;
A step of learning a learning model for inferring a correlation between the input data and the output data by inputting a plurality of sets of the training data sets;
A step of storing the learned learning model;
Machine learning method.
A bowl that applies centrifugal force to the liquid to be treated to centrifuge into the solid and the separated liquid, a screw conveyor that conveys the solid in the bowl toward the discharge port, and a drive motor that rotates the bowl. , A computer-based machine learning method for a centrifuge system comprising a differential speed generator that rotates the screw conveyor with a differential speed relative to the bowl.
The first image data obtained by imaging the liquid-containing solid matter discharged from the discharge port from a predetermined angle of view, and the liquid to be treated before being supplied to the bowl and after the predetermined additive is added. This is a step of storing a plurality of sets of learning data sets including input data including a second image data captured from a predetermined angle of view and output data including control parameters associated with the input data. The control parameter comprises at least one of the supply of the additive, the centrifugal force of the bowl, and the differential speed controlled by the differential speed generator;
A step of learning a learning model for inferring a correlation between the input data and the output data by inputting a plurality of sets of the training data sets;
A step of storing the learned learning model;
Machine learning method.