JP7357878B2 - Machine learning device, data processing system, inference device and machine learning method - Google Patents

Description

The present invention relates to a machine learning device, a data processing system, an inference device, and a machine learning method that can be applied to a system including a centrifuge.

Centrifugal separators that perform solid-liquid separation using centrifugal force have been conventionally used in water treatment facilities for water and sewage, industrial wastewater, human waste, and the like. Various types of centrifugal separators are known, but a centrifugal separator called a decanter is widely used, especially in equipment that requires continuous processing over a long period of time.

Generally, this decanter consists of a drive motor, a cylindrical bowl with one end shaped like a cone, which is rotated by the drive motor and into which the liquid to be treated is poured, and a cylindrical bowl that is arranged coaxially within the bowl and into which the liquid to be treated is placed. It includes a screw conveyor that conveys the solids in the liquid, and a differential speed generating device that generates a differential speed between the rotational speed of the bowl and the rotational speed of the screw conveyor. Solid-liquid separation is achieved by discharging the solids in the liquid to be processed into the bowl from one end of the bowl and the separated liquid in the liquid to be processed from the other end of the bowl. .

By the way, in order to perform solid-liquid separation, this decanter uses a large-sized motor as a drive motor in order to obtain a relatively large centrifugal force to enable this separation process. Therefore, various methods have been studied to appropriately control the power consumption of this large motor. (For example, see Patent Document 1.)

Patent No. 5442099

As in the method described in Patent Document 1, if a control device is adopted that controls the centrifugal separator based on the torque of the screw conveyor and the differential speed generated by the differential speed generating device, the centrifugal separator can be controlled according to the properties of the liquid to be processed. This allows operations to be carried out without the need for operator judgment. However, the objects to be controlled in a centrifugal separation system including such a centrifugal separator exist in addition to the torque of the screw conveyor and the differential speed generated by the differential speed generator, and there is also information (state) that affects the control of the control object. Since the number of variables (variables) is also wide-ranging, there remains room for further improvement, for example, when attempting to fully automate the control of a centrifugal separation system.

In view of the above points, an object of the present invention is to provide a machine learning device, a data processing system, an inference device, and a machine learning method for realizing suitable operation control based on various information in a centrifugal separation system. do.

In order to achieve the above object, the machine learning device 20 according to the first aspect of the present invention applies centrifugal force to the liquid to be treated to separate the solids and the separated liquid, as shown in FIGS. 1 to 3, for example. A screw conveyor 3 that conveys the solids 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. A centrifugal separation system comprising: a slurry concentration of the liquid to be treated; a water content of the dehydrated solid as a solid separated and discharged in the bowl 2; Learning that stores multiple sets of learning data sets consisting of input data including the concentration of the separated liquid and the torque value of the screw conveyor 3, and output data including the control parameters of the centrifugal separation system associated with the input data. a learning unit 23 that learns a learning model that infers the correlation between input data and output data by inputting a plurality of sets of learning data sets; and a learned model storage unit 24 for storing a model; the control parameters include the supply amount of additives added to the liquid to be treated, the centrifugal force of the bowl 2, and the differential speed controlled by the differential speed generator 5. consisting of at least one of the following.

With this configuration, it is possible to provide a trained model that can infer optimal control parameters for operating the centrifugal separation system from various state variables of the centrifugal separation system.

Further, the data processing system according to the second aspect of the present invention, as shown in FIG. A screw conveyor 3 that conveys the solids inside toward the discharge port 2a, a drive motor 4 that rotates the bowl, and a differential speed generator 5 that rotates the screw conveyor 3 at a differential speed relative to the bowl 2. A data processing system 50 used in a centrifugal separation system including: slurry concentration of the liquid to be treated, water content of the dehydrated solid as a solid separated and discharged in the bowl 2, concentration of the separated liquid, and a screw conveyor. Detection unit 51 for acquiring the torque value of No. 3; By inputting the data acquired by the detection unit 51 into the learned model generated by the machine learning device according to the first aspect, the centrifugal separation system an inference unit 57 for inferring control parameters;

With this configuration, it is possible to provide a data processing system that can infer optimal control parameters for operating the centrifugal separation system from various state variables of the centrifugal separation system.

Further, the inference device according to the third aspect of the present invention, as shown in FIG. A screw conveyor 3 that conveys the solids toward the discharge port 2a, a drive motor 4 that rotates the bowl 2, and a differential speed generator 5 that rotates the screw conveyor 3 at a differential speed relative to the bowl 2. a centrifugal separation system comprising a memory and at least one processor, the at least one processor comprising: a slurry concentration of the liquid to be treated, a water content of the dehydrated solids as solids separated and discharged in the bowl 2; processing to obtain the rate, the concentration of the separated liquid, and the torque of the screw conveyor 3; the slurry concentration of the liquid to be treated, the water content of the dehydrated solid as a solid separated and discharged in the bowl 2, and the concentration of the separated liquid. , and when the torque value of the screw conveyor 3 is input, a process of inferring control parameters of the centrifugal separation system is executed; , centrifugal force of the bowl 2, and differential speed controlled by the differential speed generating device 5.

With this configuration, it is possible to provide an inference device that can infer optimal control parameters for operating the centrifugal separation system from various state variables of the centrifugal separation system.

Further, the machine learning method according to the fourth aspect of the present invention, as shown in FIG. A screw conveyor 3 that conveys the solid matter inside toward the discharge port 2a, a drive motor 4 that rotates the bowl 2, and a differential speed generator 5 that rotates the screw conveyor 3 at a differential speed relative to the bowl 2. A computer-based machine learning method for a centrifugal separation system comprising: a slurry concentration of a liquid to be treated, a water content of dehydrated solids as solids separated and discharged in a bowl 2, and a concentration of a separated liquid. , a step S11 of storing a plurality of learning data sets each consisting of input data including the torque value of the screw conveyor 3, and output data including the control parameters of the centrifugal separation system associated with the input data; Steps S13 to S15 of learning a learning model for inferring the correlation between input data and output data by inputting a plurality of sets of data sets; and Step S17 of storing the learned learning model; The parameters include at least one of the supply amount of the additive added to the liquid to be treated, the centrifugal force of the bowl 2, and the differential speed controlled by the differential speed generating device 5.

With this configuration, it is possible to obtain a trained model that can infer optimal control parameters for operating the centrifugal separation system from various state variables of the centrifugal separation system.

According to the present invention, it becomes possible to obtain control parameters that can realize suitable operation control of the centrifugal separation system based on various information in the centrifugal separation system, thereby contributing to complete automation of control of the centrifugal separation system. be able to.

FIG. 1 is a schematic diagram showing the general structure of a decanter main body according to an embodiment of the present invention. FIG. 2 is a schematic diagram showing a centrifugal separation system including a decanter piping structure according to an embodiment of the present invention. FIG. 3 is a schematic block diagram of a machine learning device according to an embodiment of the present invention. FIG. 4 is a diagram illustrating an example of a neural network model for supervised learning implemented in a machine learning device according to an embodiment of the present invention. FIG. 5 is a flowchart illustrating an example of a machine learning method according to an embodiment of the present invention. FIG. 6 is a schematic block diagram showing a data processing system according to an embodiment of the present invention.

EMBODIMENT OF THE INVENTION Hereinafter, one embodiment for carrying out the present invention will be described with reference to the drawings. In addition, below, the scope necessary for explanation to achieve the purpose of the present invention will be schematically shown, and the scope necessary for explanation of the relevant part of the present invention will be mainly explained, and the parts where explanation is omitted will be omitted. It shall be based on known technology.

<Centrifugal separation system>
First, a centrifugal separation system according to an embodiment of the present invention will be briefly described. In this embodiment, a centrifugal separation system including a horizontal decanter 1 as a centrifugal separator is used as an object to which the machine learning device, data processing system, inference device, and machine learning method of the present invention are applied. In other words, the present invention aims to automate operation control in a centrifugal separation system by applying the machine learning device, data processing system, inference device, and machine learning method according to the present invention to the centrifugal separation system. The specific embodiments of the centrifugal separation system according to the present invention are not limited to those shown here, and may be applied to centrifugal separation systems including other centrifugal separators, such as vertical or straight-body decanters. It is also applicable.

FIG. 1 is a schematic diagram showing the general structure of a decanter main body according to an embodiment of the present invention. Moreover, FIG. 2 is a schematic diagram showing a centrifugal separation system including a piping structure of a decanter according to an embodiment of the present invention. The horizontal decanter 1 mainly includes a bowl 2, a screw conveyor 3, a drive motor 4, a differential speed generator 5, and a casing 6, as shown in FIGS. 1 and 2.

The bowl 2 is constituted by a cylindrical member whose one end is processed into a conical shape, and is supported rotatably around a horizontal axis. Further, one or more solid material discharge ports 2a are provided at one end of the bowl 2, which is machined into a weight shape, and a dam 2c in which one or more separated liquid discharge ports 2b are formed is provided at the other end. installed. Furthermore, a pulley 4a for transmitting power from a drive motor 4, which will be described later, is attached to one end of the bowl 2 (in FIG. 1, one end processed into a conical shape).

The screw conveyor 3 is disposed coaxially within the bowl 2, and is composed of a member around which spiral screw blades 3a are formed. The screw blade 3a is a member for conveying and/or squeezing the solid matter in the bowl 2. The body 3b of the screw conveyor 3 is provided with a space 3c therein for receiving the liquid to be treated, and a discharge port 3d for introducing the liquid stored in this space into the bowl 2. ing. Further, a differential speed generating device 5, which will be described later, is connected to one end of the screw conveyor 3.

The drive motor 4 is a motor for applying rotational force to the bowl 2, and is connected to a pulley 4a attached to one end of the bowl 2 via a belt 4b. It is preferable that a relatively large motor is used as the drive motor 4 in order to rotate the bowl 2 in a range of, for example, 2000 to 5000 rpm, and the rotational speed thereof can be changed as appropriate by inverter control. Further, the differential speed generating device 5 is a device for generating a differential speed between the rotational speed of the bowl 2 and the rotational speed of the screw conveyor 3, and is a device for generating a differential speed between the rotational speed of the bowl 2 and the rotational speed of the screw conveyor 3. It is a device that can rotate slowly (about 1 to 3 rpm). This differential speed generator 5 includes a gear box 5a connected to the screw conveyor 3, and a differential motor (also referred to as a "back drive motor") 5b that applies a braking force to the screw conveyor 3. Since the specific operating principle of the differential speed generating device 5 is conventionally known, its explanation will be omitted.

The casing 6 is provided to cover the bowl 2 and the screw conveyor 3, and collects the solids discharged from the solids discharge port 2a of the bowl 2 (hereinafter referred to as dehydrated solids) from the solids provided below. Similarly, the separated liquid discharged from the separated liquid outlet 2b of the bowl 2 is guided to the separated liquid chute 6b provided below. Further, the bowl 2 covered by the casing 6 and the screw conveyor 3 are integrally supported by a frame 6c.

The solid matter chute 6a of the casing 6 is connected to a solid matter discharge conduit 7, and the solid matter discharged from the solid matter chute 6a is transferred to a conveyance path to an incineration facility (not shown) via the solid matter discharge conduit 7, for example. Transported to 7a. Further, the separated liquid chute 6b of the casing 6 is connected to a separated liquid discharge conduit 8, and the separated liquid discharged from the separated liquid chute 6b is transported to a water purification facility (not shown) via the separated liquid discharge conduit 8, for example. It will be done.

The decanter 1 further includes a liquid supply pipe 9 for introducing the liquid to be treated into the bowl 2. One end of this liquid supply pipe 9 communicates with a space 3c in the body 3b of the screw conveyor 3, and the other end communicates with a to-be-treated liquid supply source 10 and an additive supply source 11, and inflow from the other end. This serves as a conduit for introducing the treated liquid and the like into the bowl 2.

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

The liquid supply pipe 9 and the additive supply source 11 also include a pump 11a for supplying additives, such as chemicals, to be added to the liquid to be processed, and a plurality of pumps for controlling the supply position and timing of the medicine. It is fluidly connected by an additive line 11c with a valve 11b. Here, the agent supplied from the additive supply source 11 is a flocculant, particularly a polymer flocculant or an inorganic flocculant, which changes the sludge contained in the slurry into flocs by administering it to the slurry.

As shown in FIG. 2, the additive pipe 11c according to the present embodiment is branched into two pipes at an intermediate position, one of which is connected to the intermediate position of the liquid to be treated pipe 10b, and the other to the liquid supply pipe 9. has been done. When a chemical is supplied via a pipe connected to an intermediate position of the liquid to be treated piping 10b, the liquid to be treated and the chemical react within the liquid to be treated piping 10b (so-called "line chemical injection"). When a chemical is supplied via a pipe connected to the liquid supply pipe 9, the liquid to be treated and the chemical react inside the bowl 2 (so-called "in-machine chemical injection"). By adopting such a piping structure, it is possible to adjust the position and timing of adding chemicals to the liquid to be treated, thereby optimizing the sludge floc generation effect of the chemicals, which is mainly affected by the condition of the liquid to be treated. be able to adjust. Note that the connection position of the piping shown in FIG. 2 is only an example, and the piping structure can be changed as appropriate in consideration of the state of the liquid to be treated and the like.

In solid-liquid separation using the decanter 1 having the above configuration, first, when a liquid to be treated to which a chemical has been added is introduced into the bowl 2 which rotates at a predetermined rotational speed via the liquid supply pipe 9, the drive motor 4 Due to the centrifugal force generated by the centrifugal force, solid matter in the liquid to be treated (which has become floc-like due to the effect of the chemical) settles on the inner wall surface of the bowl 2. The settled solids are shaped like a weight in the bowl 2 by the screw blades 3a of the screw conveyor 3, which rotates around the bowl 2 at a rotation speed slightly lower than the rotation speed of the bowl 2 due to the action of the differential speed generator 5. The solid material is conveyed while being squeezed toward the processed one end side, and is discharged from the solid material discharge port 2a to the solid material chute 6a. In addition, the liquid to be treated after the solids have been sedimented and removed as described above stays as a separated liquid in the bowl 2 for a certain period of time, and then overflows from the separated liquid outlet 2b and is discharged to the separated liquid chute 6b. be done.

<Machine learning device>
The solid-liquid separation process described above is achieved by adjusting various control parameters of the decanter 1. The control parameters herein mainly consist of the centrifugal force of the bowl 2, the amount of medicine supplied from the additive supply source 11, and the differential speed between the bowl 2 and the screw conveyor 3. Therefore, a machine learning device of the present invention that learns an inference model (trained model) capable of estimating optimal control parameters according to the type of liquid to be treated, etc. will be described below.

FIG. 3 is a schematic block diagram of a machine learning device according to an embodiment of the present invention. As shown in FIG. 3, the machine learning device 20 according to the present embodiment includes a learning dataset acquisition unit 21, a learning dataset storage unit 22, a learning unit 23, and a learned model storage unit 24. We are prepared.

The learning data set acquisition unit 21 is an interface unit that acquires a plurality of data forming a learning (training) data set from the computer PC1 connected, for example, via a wired or wireless communication line. The computer PC1 may be, for example, a computer that constitutes a control unit of the decanter 1 described above or is communicably connected to this control unit, and is capable of acquiring various control parameters of the decanter 1. Further, this computer PC1 may be inputted with various control information of the decanter 1 by the operator EN, and at least a part of the control information may be output data (also referred to as "teacher data") in a learning data set. The information is transmitted to the learning data set acquisition unit 21 in the same way as various data of the decanter 1, which will be described later. The control unit of the decanter 1 referred to herein is a device for controlling the entire decanter 1, and is composed of a well-known computer or the like including a processor, memory, and the like.

The learning data set acquired by the learning data set acquisition unit 21 includes, as input data , the slurry concentration of the liquid to be treated supplied from the liquid to be treated source 10 and the solid waste separated in the bowl 2. The moisture content of the dehydrated solid that is the solid after being discharged from the outlet 2a, the concentration of the separated liquid separated in the bowl 2 and discharged from the separated liquid discharge port 2b, and the torque value of the screw conveyor 3 are calculated. The output data includes the main control parameters of the decanter 1, specifically the amount of medicine supplied into the bowl 2, the centrifugal force of the bowl 2, and the differential speed controlled by the differential speed generator. Contains at least one. A specific method for acquiring various data constituting these learning data sets will be described later.

The learning data set storage unit 22 associates the plurality of data forming the learning data set acquired by the learning data set acquisition unit 21 with related input data and output data (also referred to as "teacher data"). This is a database for storing one learning data set. The specific configuration of the database that constitutes this learning data set storage unit can be adjusted as appropriate. For example, in FIG. 3, for convenience of explanation, the training data set storage unit 22 and the trained model storage unit 24, which will be described later, are shown as separate storage means, but these are stored in a single storage medium (database ) can also be configured.

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

The trained model storage unit 24 is a database for storing the trained model generated by the learning unit 23. The trained model stored in the trained model storage unit 24 is applied to the actual system via communication lines and storage media, including the Internet, in response to requests. Specific application aspects of the learned model to the actual system (data processing system) will be described in detail later.

Here, an example of a method for collecting a learning data set will be described with reference to FIGS. 2 and 3. The decanter 1 according to the present embodiment is provided with a monitoring system for specifying the operating state of the decanter 1 at appropriate locations of its components. Specifically, as shown in FIG. 2, the monitoring system includes a treated liquid monitoring system 31, a dehydrated solids monitoring system 32, a separated liquid monitoring system 33, and a screw conveyor torque monitoring system 34. The data detected by these monitoring systems 31 to 34 correspond to the input data forming the learning data set.

The liquid to be treated monitoring system 31 is disposed in the liquid to be treated piping 10b, and monitors the liquid to be treated, especially the slurry concentration, supplied from the liquid to be treated source 10. As a specific method for monitoring the slurry concentration, for example, a well-known concentration sensor may be used to directly detect the concentration value of the liquid to be treated.

A dewatered solids monitoring system 32 is provided in the solids discharge conduit 7 and is intended to monitor, in particular, the moisture content of the dehydrated solids (also referred to as "dehydrated cake") passing through the solids discharge conduit 7. The dewatered solids monitoring system 32 measures or estimates the moisture content of the dehydrated solids passing through the solids discharge conduit 7. Note that various methods can be adopted to measure or estimate the moisture content of the dehydrated solid, but in this embodiment, the moisture content of the dehydrated solid is directly measured using a well-known moisture content meter. The method is adopted.

The separated liquid monitoring system 33 is provided in the separated liquid discharge conduit 8 and monitors the concentration of the separated liquid passing through the separated liquid discharge conduit 8. As a specific method for monitoring the concentration of the separated liquid, for example, a well-known concentration sensor may be used to directly detect the concentration value of the separated liquid.

The screw conveyor torque monitoring system 34 is attached to the rotating shaft of the screw conveyor 3, for example, and monitors the torque value of the screw conveyor 3 by detecting the reaction force acting on the screw conveyor 3.

Various data acquired by these multiple monitoring systems, specifically the concentration of the liquid to be treated, the water content of the sampled dehydrated solids, the concentration of the separated liquid, and the torque value of the screw conveyor 3 are transmitted to the control unit of the decanter 1 and the The data is then sent to the computer PC1, and then organized and stored in the machine learning device 20 as input data for a learning data set.

Further, when preparing a learning data set in the decanter 1 according to the present embodiment, it is necessary to specify optimal control parameters to serve as training data based on information etc. acquired by the above-mentioned monitoring system. The optimal control parameters may be manually specified by the operator EN, for example, based on information obtained by the monitoring system and various other information. The optimal control parameters identified in this way are organized and stored in the machine learning device 20 as training data of a learning data set together with various data acquired by the above-mentioned monitoring system.

Here, as described above, the optimal control parameters identified include at least one of the centrifugal force of the bowl 2, the amount of medicine supplied from the additive supply source 11, and the differential speed between the bowl 2 and the screw conveyor 3. It can be done as one. These control parameters are considered to be the control parameters that have the most influence on the solid-liquid separation process by the decanter 1. Therefore, it is preferable that all the three control parameters mentioned above be specified here. Note that only one or two of these three control parameters may be input, and parameters other than these three control parameters (for example, drug supply position, etc.) may be additionally input. is also possible. In particular, among the three control parameters mentioned above, the centrifugal force of the bowl 2 can be adjusted mainly by controlling the rotation speed of the bowl 2 by the drive motor 4, but other controls can change this rotation speed. The responsiveness is lower than that of the two controls (control of drug supply amount and differential speed). Therefore, when specifying only two of the three control parameters described above, it is preferable to input two of the amount of medicine supplied from the additive supply source 11 and the differential speed between the bowl 2 and the screw conveyor 3.

Next, a learning method in the learning unit 23 using the plurality of learning data sets obtained as described above will be briefly described. FIG. 4 is a diagram illustrating an example of a neural network model for supervised learning implemented in a machine learning device according to an embodiment of the present invention. The neural network in the neural network model shown in Figure 4 consists of l neurons (x1 to xl) in the input layer, m neurons (y11 to y1m) in the first hidden layer, and n neurons in the second hidden layer. 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. Alternatively, only the first intermediate layer may be a hidden layer.

Furthermore, nodes connecting neurons between the layers are connected between the input layer and the first hidden layer, between the first hidden layer and the second hidden layer, and between the second hidden 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 uses the learning data set to determine the concentration of the liquid to be treated, the water content of the sampled dehydrated solid, the concentration of the separated liquid, and the torque value of the screw conveyor 3. , learn the correlation with the control parameters. Specifically, the concentration of the liquid to be treated as state variables, the water content of the sampled dehydrated solid, the concentration of the separated liquid, and the torque value of the screw conveyor 3 are associated with neurons in the input layer, and The value of a certain neuron is calculated using the general neural network output value calculation method, that is, the value of the output neuron is calculated by the value of the input neuron connected to the neuron, the output neuron, and the input neuron. It is calculated as the sum of the sequence of multiplication values of the weight wi associated with the node that connects the neuron of . Note that when inputting the above state variables to neurons in the input layer, the format in which the information acquired as state variables is input can be set as appropriate, taking into consideration the accuracy of the trained model to be generated. can.

Then, the calculated values of the o neurons z1 to zo in the output layer, that is, in this embodiment, one or more control parameters and one or more control parameters that also constitute a part of the learning data set. The following steps are repeated: comparing the teacher data t1 to to to determine the error, and adjusting the weight wi associated with each node (backpropriation) so that the determined error becomes smaller.

Then, if a predetermined condition is met, such as repeating the above-mentioned series of steps a predetermined number of times or the error becomes smaller than a tolerance value, learning is completed and the neural network model ( All weights wi) associated with each of the nodes are stored in the trained model storage unit 24 as a trained model.

<Machine learning method>
In connection with the above, the present invention provides a machine learning method. FIG. 5 is a flowchart illustrating an example of a machine learning method according to an embodiment of the present invention. The machine learning method shown below will be explained based on the machine learning device described above, but the prerequisite configuration is not limited to the machine learning device described above. In addition, although this machine learning method is realized by using a computer, various types of computers can be applied. For example, a computer forming the control unit of the decanter 1, a server arranged on a network, etc. A device or a computer PC1 shown in FIG. 3 can be mentioned. The specific configuration of this computer includes, for example, an arithmetic unit consisting of at least a CPU, a GPU, etc., a storage device consisting of volatile or nonvolatile memory, etc., and a communication device for communicating with a network and other devices. It is possible to employ a system that includes devices and a bus that connects these devices.

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

When step S11 is completed, a pre-learning neural network model is then prepared in order to start learning in the learning unit 23 (S12). The pre-learning neural network model prepared here has the structure shown in FIG. 4 above, and the weights of each node are set to initial values. Then, for example, one learning dataset is randomly selected from the plurality of learning datasets stored in the learning dataset storage unit 22 (step S13), and the input data in the one learning dataset is is input into the input layer (see FIG. 4) of the prepared neural network model before learning (step S14). Note that various methods can be employed for inputting the input data in the learning data set to the input layer of the neural network model before learning. In addition, before inputting the concentration of the liquid to be treated as input data, the water content of the sampled dehydrated solid, the concentration of the separated liquid, and the torque value of the screw conveyor 3 to the input layer, a predetermined value for adjusting the data is provided. Preprocessing may be performed.

Here, the control parameters of the output layer (see FIG. 4) generated as a result of step S14 above are generated by the neural network model before learning, so in most cases, the control parameters are different from the desired result. be. Therefore, next, machine learning is performed using the control parameters as teacher data in the one learning data set acquired in step S13 and the control parameters of the output layer generated in step S13 (step S15). ). The machine learning performed here, for example, compares the control parameters that make up the teaching data and the control parameters that make up the output layer, detects the error between the two, and then tries to obtain an output layer that reduces this error. , may be a process (backpropagation) of adjusting the weights associated with each node in the neural network model before learning. Further, the number and format of control parameters output to the output layer of the neural network model before learning are the same number and format as the teacher data in the learning data set as the learning target. Therefore, for example, if the control parameters as teacher data in the learning dataset consist of data corresponding to two control parameters, the control parameters output by the neural network model to the output layer will also consist of two control parameters. This is 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 datasets stored in the learning dataset storage unit 22. (Step S16). Then, if the machine learning is to be continued (No in step S16), the process returns to step S13, and if the machine learning is to be ended (Yes in step S16), the process is to proceed to step S17. When continuing the machine learning, steps S13 to S15 are performed multiple times on the neural network model that is being trained using an unlearned learning data set. Generally, the accuracy of the trained model finally generated increases in proportion to this number of times.

When ending machine learning (Yes in step S16), the neural network generated by adjusting the weights associated with each node through a series of steps is stored as a trained model in the trained model storage unit 24. (Step S17), the series of learning processes ends. The learned model stored here can be applied and used in various data processing systems, and details of the data processing system 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, machine learning processing is repeatedly performed multiple times on one neural network model (before learning). Although a method for improving the accuracy and obtaining a trained model sufficient for application to a data processing system is described, the present invention is not limited to this acquisition method. For example, a plurality of trained models that have been subjected to machine learning a predetermined number of times are stored as candidates in the trained model storage unit 24, and a data set for validity judgment is input to the group of trained models. The output layer (the values of the neurons) may be generated by comparing the accuracy of the control data specified in the output layer, and the best trained model to be applied to the data processing system may be selected. . Note that the validity determination data set may be any data set that is the same as the learning data set used for learning and that is not used for learning.

As explained above, by applying the machine learning device and machine learning method according to the present embodiment, optimal control parameters can be derived from various data acquired by a plurality of monitoring systems installed at appropriate locations in the decanter 1. You can obtain a trained model that can

<Data processing system>
Next, with reference to FIG. 6, an application example of the learned model generated by the machine learning device and machine learning method described above will be described. FIG. 6 is a schematic block diagram showing a data processing system according to an embodiment of the present invention. As an example of the data processing system 50 according to the present embodiment, a mode is illustrated in which the data processing system 50 is installed in the control unit of the horizontal decanter 1 described above.

As shown in FIG. 6, the data processing system 50 mainly includes a detection unit 51, an adjustment unit 52, a calculation unit 53, a database (DB) 54, a user interface 55, and a system for interconnecting these. An internal bus 56 is provided. In addition, in FIG. 6, only the components constituting the data processing system 50 are shown, and descriptions of other components that are not directly related to the data processing system 50 according to the present embodiment provided in the control unit are omitted. There is.

The detection unit 51 is for detecting state variables of the decanter 1, and specifically, the above-mentioned liquid to be treated monitoring system 31, dehydrated solids monitoring system 32, separated liquid monitoring system 33, and screw conveyor torque monitoring. system 34. The detection unit 51 uses these monitoring systems to detect the slurry concentration of the liquid to be treated, the water content of the dehydrated solids separated and discharged in the bowl 2, the concentration of the separated liquid, and the screw conveyor. 3 torque values can be obtained.

The adjustment unit 52 is for adjusting various control units that are adjusted in order to realize optimal operation control in the decanter 1. Specifically, a drive motor control unit 41 that is connected to the drive motor 4 and controls its rotation speed; a pump control unit 42 that controls the pump 11b that is installed in the additive pipe 11c and that varies the amount of medicine supplied; The differential motor control unit 43 is connected to the differential motor 5b and controls its rotation speed, thereby controlling the rotation speed of the screw conveyor 3 (with respect to the bowl 2).

The arithmetic unit 53 constitutes a processor for realizing various processes in the data processing system 50, and includes an inference unit 57. The inference unit 57 is connected to a trained model storage unit 58 that stores a trained model prepared in advance, and by taking into account the trained model stored in this trained model storage unit 58, the detection unit 51 The control parameters to be adjusted by the adjustment unit 52 are inferred from the state variables detected by the control unit 52. A plurality of trained models are stored in the trained model storage unit 58, depending on the application and various conditions (for example, season, weather, environmental conditions such as temperature and humidity, or type of liquid to be treated, etc.). It is preferable to have one. Furthermore, the task of selecting an appropriate trained model from a plurality of trained models may be performed automatically using various sensors, or manually selected by the operator EN. It's okay.

The database 54 is made of a well-known recording medium and is used to temporarily or continuously store various data in the data processing system 50, such as detection results by the detection unit 51 and calculation results by the calculation unit 53. Further, the user interface 55 is composed of, for example, a GUI (graphical user interface), and is used to display the status of the decanter 1 and receive input operations from the operator EN and the like.

When adjusting control parameters using the data processing system 50 having the above-described configuration, the following process is executed. That is, first, four state variables of the decanter 1 are acquired in the detection unit 51. Next, by inputting these four state variables to the inference unit 57 of the calculation unit 53, three control parameters are inferred. Then, the three control parameters as the inference results are sent to the adjustment unit 52, and each control unit is operated to adjust the centrifugal force of the bowl 2, the supply amount of medicine, and the adjustment between the bowl 2 and the screw conveyor 3 as control parameters. Adjust differential speed.

The timing for adjusting the control parameters described above can be automatically specified or manually adjusted. Further, the adjustment may be performed whenever the decanter 1 is in operation; for example, when the decanter 1 starts operating, or when the type of liquid to be treated with the decanter 1 is changed, Only when a preset predetermined time has elapsed, or when a change in the concentration of the liquid to be processed or the separated liquid can be detected in the liquid to be processed monitoring system 31 or the separated liquid monitoring system 33, and the change exceeds a predetermined threshold value. The adjustment may be made.

As an option, although the above embodiment describes the case where the trained model is trained by batch learning, in the present invention, the trained model trained by batch learning may be trained online. Further accuracy may be improved by In that case, the state variable detected in the detection unit 51 and the control parameter inferred by the inference unit 57, in particular, the control when the solid-liquid separation processing result is improved as a result of the adjustment in the adjustment unit 52. The data set shown in FIG. A similar machine learning process can be executed.

Further, the data processing system 50 described above is applied within the control unit of the horizontal decanter 1, but instead of this, for example, a computer communicatively connected to the control unit of the decanter 1, or a control unit of the decanter 1 may be used. It is also possible to apply the present invention to a server device or the like connected to the server via a network. However, the centrifugal separation system according to the present embodiment usually differs greatly in processing contents (type of liquid to be processed, processing amount per unit time, etc.) and surrounding environment (climate, etc.) depending on the system. It is not efficient to generate a general-purpose trained model that can be applied to all of these centrifugal separation systems. Therefore, it is practical and preferable to apply and operate the data processing system 50 and the machine learning process described above for each centrifugal separation system.

<Inference device>
The present invention can be provided not only in the form of the data processing system 50 described above, but also in the form of an inference device for performing inference. In that case, the inference device may include a memory and at least one processor, of which the processor may execute a series of processes. The series of processes includes the process of acquiring the state variables of the decanter, that is, the slurry concentration of the liquid to be treated, the water content of the dehydrated solids, the concentration of the separated liquid, and the torque of the screw conveyor 3, and the process of acquiring the state variables of the decanter, and the process of acquiring the state variables of the decanter, that is, the slurry concentration of the liquid to be treated, the water content of the dehydrated solid, the concentration of the separated liquid, and the torque of the screw conveyor 3. and inferring control parameters of the centrifugal separation system when the centrifugal separation system is used. The control parameters herein include at least one of the supply amount of the additive, the centrifugal force of the bowl 2, and the differential speed controlled by the differential speed generating device 5. By providing the present invention in the form of the inference device described above, it can be applied to various centrifugal separation systems more easily than when implementing a data processing system.

The present invention is not limited to the embodiments described above, and can be implemented with various modifications without departing from the spirit of the present invention. All of these are included in the technical idea of the present invention.

1 (horizontal) decanter 2 Bowl 3 Screw conveyor 4 Drive motor 5 Differential speed generator 6 Casing 9 Liquid supply pipe 10 Processed liquid supply source 10a, 11a Pump 11 Additive supply source 20 Machine learning device 21 Learning data set acquisition Unit 22 Learning data set storage unit 23 Learning unit 24 Learned model storage unit 31 Processed liquid monitoring system 32 Dehydrated solids monitoring system 33 Separated liquid monitoring system 34 Screw conveyor torque monitoring system 41 Drive motor control unit 42 Pump control unit 43 Differential motor control unit 50 Data processing system 51 Detection unit 52 Adjustment unit 53 Arithmetic unit 57 Inference unit

Claims (3)

A bowl that applies centrifugal force to a liquid to be treated to centrifugally separate solids and a separated liquid, a screw conveyor that conveys the solids in the bowl toward a discharge port, and a drive motor that rotates the bowl. , a differential speed generating device for rotating the screw conveyor at a differential speed relative to the bowl, the machine learning device for a centrifugal separation system comprising:
Consisting of four data: slurry concentration of the liquid to be treated, water content of the dehydrated solid as the solid separated and discharged in the bowl, concentration of the separated liquid, and torque value of the screw conveyor. a learning data set storage unit that stores a plurality of learning data sets including input data and output data including control parameters of the centrifugal separation system associated with the input data;
a learning unit that learns a learning model that infers a correlation between the input data and the output data by inputting the plurality of training data sets;
a learned model storage unit that stores the learning model learned by the learning unit;
The control parameter includes at least one of a supply amount of an additive added to the liquid to be treated, a centrifugal force of the bowl, and a differential speed controlled by the differential speed generating device.
Machine learning device. A bowl that applies centrifugal force to a liquid to be treated to centrifugally separate solids and a separated liquid, a screw conveyor that conveys the solids in the bowl toward a discharge port, and a drive motor that rotates the bowl. , a differential speed generating device for rotating the screw conveyor at a differential speed relative to the bowl, the data processing system used in a centrifugal separation system comprising:
a detection unit for acquiring a slurry concentration of the liquid to be treated, a water content of the dehydrated solid as the solid separated and discharged in the bowl, a concentration of the separated liquid, and a torque value of the screw conveyor; ;
an inference unit that infers control parameters of the centrifugal separation system by inputting the data acquired by the detection unit into a learned model generated by the machine learning device according to claim 1;
Data processing system. A bowl that applies centrifugal force to a liquid to be treated to centrifugally separate solids and a separated liquid, a screw conveyor that conveys the solids in the bowl toward a discharge port, and a drive motor that rotates the bowl. , a differential speed generating device that rotates the screw conveyor at a differential speed relative to the bowl, the computer-based machine learning method for a centrifugal separation system comprising:
Consisting of four data: slurry concentration of the liquid to be treated, water content of the dehydrated solid as the solid separated and discharged in the bowl, concentration of the separated liquid, and torque value of the screw conveyor. storing a plurality of learning data sets each including input data and output data including control parameters of the centrifugal separation system associated with the input data;
learning a learning model that infers a correlation between the input data and the output data by inputting a plurality of the training data sets;
storing the learned learning model;
The control parameter includes at least one of a supply amount of an additive added to the liquid to be treated, a centrifugal force of the bowl, and a differential speed controlled by the differential speed generating device.
Machine learning methods.

Images