JP7516286B2 - Fluid supply device, inference device, machine learning device, pressure tank diagnosis method, inference method, machine learning method, pressure tank diagnosis program, inference program, and machine learning program - Google Patents

Description

The present invention relates to a fluid supply device, an inference device, a machine learning device, a pressure tank diagnosis method, an inference method,
The present invention relates to a machine learning method, a pressure tank diagnosis program, an inference program, and a machine learning program.

A water supply device, which is an example of a fluid supply device, is used to supply water (
Water supply devices are widely used as devices for supplying water (fluids). A water supply device generally includes a pump for transporting water, a control unit for controlling the operation of the pump, and a pressure tank provided in the water supply pipe on the discharge side of the pump.

A pressure tank generally has a diaphragm that divides the inside of the container into an air chamber and a water chamber. Air is sealed inside the air chamber beforehand, and water transferred by the operation of the pump flows into the water chamber,
The air is compressed through a diaphragm. A check valve is placed between the pressure tank and the pump.
The check valve and pressure tank ensure that the water pressure in the water supply pipe is maintained at an appropriate level even after the pump has stopped, and water is supplied to the destination. When the water is used at the destination, the water flows out of the water chamber and the diaphragm is pushed back from the air chamber side to the water chamber side.

In the water supply system, the pump is turned on and off depending on the water supply to the destination.
The diaphragm of a pressure tank repeatedly expands and contracts, gradually deteriorating. Similarly, the fixing portion of the diaphragm to the container also gradually deteriorates due to repeated bending and stretching. However, since the deterioration of the diaphragm differs depending on the usage conditions, it has been difficult to replace the diaphragm at the optimal timing. Therefore, an inspection device has been proposed that can diaphragm deterioration state as a measure of the soundness of a pressure tank (see, for example, Patent Documents 1 and 2).

Patent No. 3596909 Japanese Patent Application Publication No. 8-86278

The inspection devices disclosed in Patent Document 1 and Patent Document 2 each include a pressure sensor (the "pressure switch 9" in Patent Document 1 and the "pressure sensor 21" in Patent Document 2) used to diagnose the soundness of the pressure tank. On the other hand, the water supply device also includes a pressure sensor (the "pressure switch 9" in Patent Document 1 and the "pressure sensor 21" in Patent Document 2) other than the above-mentioned diagnostic pressure sensor in order to switch the pump on and off in accordance with the supply of water to the supply destination.
In the inspection devices disclosed in Patent Documents 1 and 2, a pressure switch 6 in the pressure tank and a pressure gauge 6 in the pressure tank 2 are provided on the discharge side of the pump. Therefore, in the inspection devices disclosed in Patent Documents 1 and 2, a pressure sensor for diagnosis must be newly provided just to diagnose the soundness of the pressure tank.

In view of the above-mentioned problems, the present invention aims to provide a fluid supply device, an inference device, a machine learning device, a pressure tank diagnostic method, an inference method, a machine learning method, a pressure tank diagnostic program, an inference program, and a machine learning program that enable the diagnosis of the health of a pressure tank without the need to install a new diagnostic pressure sensor.

In order to achieve the above object, a fluid supply device according to an aspect of the present invention comprises:
A pump that transfers a fluid using an electric motor as a drive source;
a pressure tank provided on a discharge side of the pump and configured to hold a pressure of the fluid;
a power generation amount measuring device for measuring the amount of power generated when the electric motor operates as a generator;
A control unit that controls the supply operation of the fluid,
The control unit is
When the fluid stored in the pressure tank is caused to flow back toward the suction side of the pump, the amount of electricity generated measured by the power generation measuring device is obtained as the amount of electricity generated during backflow, and the integrity of the pressure tank is diagnosed based on the amount of electricity generated during backflow.

According to the fluid supply device of the present invention, the control unit diagnoses the soundness of the pressure tank based on data on the amount of power generated during reverse flow when the electric motor operates as a generator by causing the fluid stored in the pressure tank to flow back to the suction side of the pump. Therefore, it is possible to diagnose the soundness of the pressure tank without providing a new pressure sensor for diagnosis.

Other objects, configurations and effects will become apparent from the detailed description of the invention described below.

1 is an overall configuration diagram showing an example of a water supply system 1 to which a fluid supply device 3 according to a first embodiment is applied. 1 is a block diagram showing an example of a water supply system 1 to which a fluid supply device 3 according to a first embodiment is applied. A hardware configuration diagram showing an example of a fluid supply device 3 (mainly the control panel 36 and the operation display panel 37) and a computer 200 that constitutes the machine learning device 4. FIG. 2 is a block diagram showing an example of a machine learning device 4 according to the first embodiment. FIG. 2 is a data configuration diagram showing an example of data (supervised learning) used by the machine learning device 4 according to the first embodiment. FIG. 2 is a schematic diagram illustrating an example of a neural network model used in the machine learning device 4 according to the first embodiment. 4 is a flowchart showing an example of a machine learning method performed by the machine learning device 4 according to the first embodiment. A block diagram showing an example of a pressure tank diagnosis device 5 (pressure tank diagnosis unit 363) according to the first embodiment. 13 is a flowchart showing an example of a pressure tank diagnosis method by a fluid supply device 3 having a pressure tank diagnosis unit 363 (pressure tank diagnosis device 5) according to the first embodiment. FIG. 11 is a data configuration diagram showing an example of data (unsupervised learning) used by the machine learning device 4 according to the second embodiment. FIG. 13 is a schematic diagram illustrating an example of an autoencoder model used in a machine learning device 4 according to a second embodiment. 13 is a flowchart showing an example of a machine learning method performed by the machine learning device 4 according to the second embodiment. 13 is a flowchart showing an example of a pressure tank diagnosis method by a fluid supply device 3 having a pressure tank diagnosis unit 363 (pressure tank diagnosis device 5) according to a second embodiment.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
The scope necessary for the explanation to achieve the objective of the present invention is shown diagrammatically, and the scope necessary for the explanation of the relevant parts of the present invention will be mainly explained, and the parts where explanation is omitted will be based on publicly known technology.

FIG. 1 is an overall configuration diagram showing an example of a water supply system 1 to which a fluid supply device 3 according to a first embodiment is applied.

The water supply system 1 is a system that supplies water (fluid) to buildings and facilities such as condominiums, office buildings, public facilities, commercial facilities, factories, etc. The water supply system 1 shown in Fig. 1 employs a water tank system and includes a water tank 10 connected to a water main 11 that serves as a water supply source, and a fluid supply device 3 that serves as a water supply device.

In this embodiment, the fluid supply device 3 will be described as supplying water as a fluid, but the fluid supply device 3 may also supply liquids other than water, such as fuel and chemicals, as well as fluids containing gases such as gas, hydrogen, and oxygen.

A water main pipe 11 is connected to the upper part of the water tank 10 via a distribution pipe 12, and water flows in from the water main pipe 11. A fluid supply device 3 is connected to the lower part of the water tank 10 via a connection pipe 13, and water stored in the water tank 10 flows out to the fluid supply device 3.

The water tank 10 is provided therein with a water level gauge 100 and a float switch 101. The water level gauge 100 is a water level measuring device for stepwise or continuous measurement of the water level stored in the water tank 10. The float switch 101 is a water level detector for detecting that the water level stored in the water tank 10 is below a predetermined water level.

An inlet valve 120 is provided in the middle of the water pipe 12.
The valve is opened or closed based on the water level (water level in the receiving tank) detected by the float switch 101 or the water level detected by the float switch 101.
(not shown) may be provided.

The fluid supply device 3 includes a pump 31 for transporting water, a suction side pipe 30a connected to the suction side (upstream side) of the pump 31, and a discharge side pipe 30b connected to the discharge side (downstream side) of the pump 31.
b, a branch pipe 33 connected to the suction side pipe 30a, and a bypass pipe 34 connected to the discharge side pipe 30b.

The inlet of the suction side pipe 30a is connected to the water tank 10 via the connection pipe 13.
The water in the suction pipe 30a flows into the suction pipe 30a.
3 is provided.

The outlet of the discharge side piping 30b is connected via a water supply pipe 14 to multiple fluid supply destinations 15 corresponding to multiple water taps, etc. installed in a building or facility, and the water transported by the pump 31 is supplied to each of the fluid supply destinations 15.

The water supply pipe 14 is provided with a usage flow rate sensor 140 that measures the amount of water used at a plurality of fluid supply destinations 15. When a plurality of fluid supply destinations 15 exist, the usage flow rate sensor 14
0 may be obtained by measuring the total usage flow rate obtained by adding up the flow rates of the multiple fluid supply destinations 15. For example,
1, the usage flow rate sensor 140 is provided on the upstream side where the fluid is supplied to a plurality of fluid supply destinations 15. In addition, the usage flow rate sensor 140 may be provided in each room of an apartment building or the like, with each room serving as a fluid supply destination 15, and for example, a water meter may be used.

Instead of directly measuring the usage flow rate, the usage flow rate sensor 140 may estimate the usage flow rate from other measurement data. For example, records of people entering and leaving a plurality of fluid supply destinations 15 may be measured, and the usage flow rate may be estimated from the entry and exit records. Furthermore, a human presence sensor may be installed in the toilet, bathroom, kitchen, etc. of each room, and the usage flow rate may be estimated from the detection results of the human presence sensor. Furthermore, the usage flow rate may be estimated from the amount of electricity usage measured by an electric meter or the amount of gas usage measured by a gas meter installed in each room.

In the discharge side pipe 30b, from the upstream side to the downstream side, a check valve 300, a flow sensor 301, a pump secondary pressure sensor 302, and a pressure tank 32 are provided. In Fig. 1, the fluid supply device 3 includes one set of the pump 31, the check valve 300, and the flow sensor 301, but the suction side pipe 30a and the discharge side pipe 30b may be branched and merged into a plurality of sets, so that a plurality of sets of the pump 31, the check valve 300, and the flow sensor 301 are provided in parallel.

The check valve 300 prevents backflow of water when the pump 31 stops.
Reference numeral 1 denotes a flow rate measuring device for measuring the flow rate of water flowing through the discharge side pipe 30b. The pump secondary pressure sensor 302 is a pressure measuring device for measuring the pressure of the water flowing through the discharge side pipe 30b, i.e., the discharge side pressure of the pump 31.

The pressure tank 32 is a pressure retainer that maintains the pressure of the water to be supplied to the fluid supply destination 15. The pressure tank 32 has a diaphragm 320 that divides the inside of the container into an air chamber and a water chamber. The air chamber is filled with air in advance, and the water chamber is connected to the discharge side pipe 30b.

When the pump 31 is operated and the water discharged to the discharge side of the pump 31 flows into the water chamber of the pressure tank 32 via the discharge side piping 30b, the air in the air chamber is compressed via the diaphragm 320. At this time, since the check valve 300 is disposed between the pressure tank 32 and the pump 31, the water pressure in the discharge side piping 30b and the water supply pipe 14 downstream of the check valve 300 is appropriately maintained even if the pump 31 stops.

The branch pipe 33 branches off from the suction side pipe 30a via a three-way valve 303 and is connected to the upper part of the water tank 10. The three-way valve 303 is a first valve that connects the connection pipe 13 to the suction side of the pump 31.
1 ) in which the branch pipe 33 and the suction side of the pump 31 communicate with each other, and a second communication state (indicated by a dashed arrow in FIG. 1 ) in which the branch pipe 33 and the suction side of the pump 31 communicate with each other.

The bypass pipe 34 is connected to the discharge side pipe 30b so as to bypass the check valve 300, and a bypass valve 340 is provided midway along the pipe.

The fluid supply device 3 also includes an electric motor 310 as a drive source for driving the pump 31, an inverter 311 for supplying drive power to the electric motor 310, a pump operation status sensor 312 for acquiring the operation status of the pump 31, a power generation amount sensor 313 for measuring the amount of power generated when the electric motor 310 operates as a generator, a control panel 36 that functions as a control unit for controlling the water supply (fluid supply operation), and an operation display panel 37 that accepts user operations and displays various information.
Each part of the fluid supply device 3 is supplied with power from a commercial power source (not shown).

The pump operation status sensor 312 is a pump operation status acquisition unit that acquires the operation status of the pump 31. As the operation status of the pump 31, for example, the operation state of the pump 31 (for example, either operating (ON) or stopped (OFF)), the rotation speed of the pump 31, or the rotation speed of the electric motor 31
The pump operation status sensor 312 acquires the motor current value supplied to the pump 31.
, the electric motor 310 , the inverter 311 or the control panel 36 .

The pump operation status sensor 312 is composed of, for example, a rotation angle detector such as a rotary encoder provided on the electric motor 310, a current measuring device that measures the motor current value supplied to the electric motor 310, etc. The pump operation status sensor 312 may acquire the operation status of the pump 31 based on, for example, a control signal from the control panel 36 to the inverter 311 (a control command value for turning the pump 31 ON or OFF, a control command value for the rotation speed of the electric motor 310, etc.), or may acquire the operation status of the pump 31 based on the control signal from the control panel 36 to the inverter 311 based on, for example, a control command value for turning the pump 31 ON or OFF, a control command value for the rotation speed of the electric motor 310, etc.
The operating status of the pump 31 may be obtained from the output frequency of the pump 31 .

The power generation amount sensor 313 is an electric power generation amount measuring device that measures the amount of power generated by the electric motor 310, and is provided in the electric motor 310 or the inverter 311. The power generation amount sensor 313 is composed of, for example, a current measuring device that measures the value of a motor current flowing through the electric motor 310, and a voltage measuring device that measures the value of a motor voltage applied to the electric motor 310. The power generation amount sensor 313 may also serve as the pump operation status sensor 312.

FIG. 2 is a block diagram showing an example of a water supply system 1 to which the fluid supply device 3 according to the first embodiment is applied.

The control panel 36 and the operation display panel 37 included in the fluid supply device 3 are configured, for example, by a general-purpose or dedicated computer (see FIG. 3 described later).

The control panel 36 is electrically connected to the flow sensor 301, the pump secondary pressure sensor 302, the pump operation status sensor 312, the power generation amount sensor 313, the water level gauge 100, the float switch 101, and the usage flow rate sensor 140 as a sensor group 3a for measuring physical quantities and state quantities of each part of the water supply system 1 and the fluid supply device 3. The sensor group 3a includes the water level gauge 100,
, the float switch 101 and the usage flow sensor 140 may not be included.

The control panel 36 includes a control device group 3 for physically or electrically operating each part of the fluid supply device 3.
The inverter 311, the three-way valve 303, the bypass valve 340 and the inlet valve 120 are electrically connected to the control device group 3b. The inlet valve 120 does not necessarily have to be included in the control device group 3b.

The control panel 36 controls the water supply operation by sending a control signal to the control device group 3b in response to the measurement value indicated by the detection signal from the sensor group 3a. The detection signal and the control signal may be either an analog signal or a digital signal.

The control panel 36 specifically includes a mode switching unit 360 and a normal mode execution unit 361.
The pressure tank diagnostic unit 363 includes a backflow mode execution unit 362 and a pressure tank diagnosis unit 363.

The mode switching unit 360 determines whether or not it is necessary to diagnose the soundness of the pressure tank 32 and whether or not the fluid is scheduled to be used at the fluid supply destination 15, and switches the operation mode based on the determination result. The operation modes include, for example, at least a normal operation mode and a backflow operation mode. If the mode switching unit 360 determines that there is no need for diagnosis or that there is a plan for use, it switches to the normal operation mode, and if it determines that there is a need for diagnosis and there is no plan for use, it switches to the backflow operation mode.

Various methods can be used as a method for the mode switching unit 360 to determine whether or not a diagnosis is necessary. For example, the mode switching unit 360 determines that a diagnosis is necessary when a predetermined period of time has passed since the installation of the fluid supply device 3, a predetermined period of time has passed since the previous diagnosis, or a diagnosis reservation operation by an operator is accepted.

The mode switching unit 360 can use various methods to determine whether or not there is a plan to use the fluid. For example, the mode switching unit 360 determines that there is no plan to use the fluid when the current time is late at night or when there is no past use record. The past use record can be obtained, for example, by statistically analyzing past measurements from the pump operation status sensor 312 or the usage flow rate sensor 140.

The normal mode execution unit 361 controls the operation state of the pump 31 based on the discharge side pressure of the pump 31 in the normal operation mode.
When the discharge side pressure of the three-way valve 30 drops to a predetermined starting pressure, the normal mode execution unit 361 sends a control signal to the inverter 311 to start the operation of the pump 31. When the discharge side pressure of the pump 31 reaches a predetermined stop pressure, the normal mode execution unit 361 sends a control signal to the inverter 311 to stop the operation of the pump 31. In the normal operation mode, the bypass valve 340 is in a closed state, and the three-way valve 30 is in a closed state.
3 is controlled to the first communication state, and water flows in the direction of the arrow shown by the solid line in FIG.

The reverse flow mode execution unit 362 causes the water stored in the pressure tank 32 to flow back to the suction side of the pump 31 in the reverse flow operation mode.
The control signal for opening the bypass valve 340 is sent to the bypass valve 340, and a control signal for switching the three-way valve 303 to the second communication state is sent to the three-way valve 303. As a result, the suction side of the pump 31 is opened, and the water stored in the pressure tank 32 due to the residual pressure in the pressure tank 32 flows back to the suction side of the pump 31, so that the water flows in the direction of the dashed arrow in Fig. 1. At this time, the electric motor 310 operates as a generator in response to the rotation of the impeller of the pump 31, and a generated current flows to the inverter 311 side. The generated power sensor 313 measures the generated power based on the generated current flowing to the inverter 311 side.

The pressure tank diagnosis unit 363 acquires the amount of power generation measured by the power generation sensor 313 as data on the amount of power generation during backflow when the water stored in the pressure tank 32 is caused to flow back to the suction side of the pump 31 in the backflow operation mode. Then, the pressure tank diagnosis unit 363 diagnoses the soundness of the pressure tank 32 based on the data on the amount of power generation during backflow.

The pressure tank diagnostic unit 363 may acquire the discharge side pressure measured by the pump secondary pressure sensor 302 as data on the discharge side pressure during backflow in addition to the data on the power generation amount during backflow, and diagnose the soundness of the pressure tank 32 based on the data on the discharge side pressure during backflow in addition to the data on the power generation amount during backflow. This allows the pressure in the discharge side piping 30b during backflow to be taken into consideration, making it possible to diagnose the soundness of the pressure tank 32 with higher accuracy.

Furthermore, the pressure tank diagnostic unit 363 may acquire the water level of the water tank 10 measured by the water level gauge 100 as data on the water level during backflow in addition to the data on the amount of power generation during backflow, and diagnose the soundness of the pressure tank 32 based on the data on the water level during backflow together with the data on the amount of power generation during backflow. This allows for consideration of the effect that the water stored in the water tank 10 has on the suction side piping 30a during backflow, making it possible to diagnose the soundness of the pressure tank 32 with higher accuracy. This is particularly effective when the fluid supply device 3 is configured without the branch piping 33 and the three-way valve 303.

In this embodiment, a case will be described in which a pressure tank diagnostic device 5 that diagnoses the soundness of a pressure tank 32 using a learning model 2 based on machine learning is incorporated into a control panel 36 as a pressure tank diagnostic unit 363. Therefore, the control panel 36 (pressure tank diagnostic unit 363) operates as a subject of the inference phase of machine learning.

The machine learning device 4 operates as a subject of the learning phase of machine learning, and generates, by machine learning, a learning model 2 used when diagnosing the soundness of the pressure tank 32. The learned learning model 2 is provided to the control panel 36 via any communication network, recording medium, or the like. The machine learning device 4 can employ either "supervised learning" or "unsupervised learning" as a machine learning method. In this embodiment, "supervised learning" is employed, and in the second embodiment described later, a case will be described in which "unsupervised learning" is employed.

During the learning phase of the machine learning, the sensor group 3a is connected to the machine learning device 4 and used to measure data for learning, and during the inference phase of the machine learning, the sensor group 3a is connected to the control panel 36 and used to measure data for inference.

The sensor group 3a is configured to be able to output, for example, each time a predetermined measurement period elapses, a measurement value at the time of measurement as a detection signal to the control panel 36 and the machine learning device 4. Note that the measurement periods of the sensors constituting the sensor group 3a may be the same or different.
Further, instead of outputting discrete measurement values at the end of each measurement period as described above, the sensor group 3a may output continuous measurement values such as analog signals.

FIG. 3 is a hardware configuration diagram showing an example of the fluid supply device 3 (mainly the control panel 36 and the operation display panel 37) and the computer 200 that constitutes the machine learning device 4.

Fluid supply device 3 (mainly the control panel 36 and the operation display panel 37), and machine learning device 4
Each of the above is configured with a general-purpose or dedicated computer 200.
As shown in FIG. 2, the main components of the system include a bus 210, a processor 212, and a
The computer 200 includes a memory 214, an input device 216, a display device 218, a storage device 220, a communication I/F (interface) unit 222, an external device I/F unit 224, an I/O (input/output) device I/F unit 226, and a media input/output unit 228. Note that the above components may be omitted as appropriate depending on the application of the computer 200.

The processor 212 is one or more arithmetic processing devices (CPU, MPU, GPU, DSP, etc.).
The memory 21 is made up of a plurality of memory units, each of which functions as a control unit for controlling the entire computer 200.
The internal memory 4 stores various data and programs 230 and is composed of, for example, a volatile memory (DRAM, SRAM, etc.) that functions as a main memory, and a non-volatile memory (ROM, flash memory, etc.).

The input device 216 is, for example, a keyboard, a mouse, a numeric keypad, an electronic pen, etc. The display device 218 is, for example, a liquid crystal display, an organic EL display, electronic paper, a projector, etc. The input device 216 and the display device 218 may be integrally configured, such as a touch panel display. The storage device 220 is, for example, a HDD, an SSD, etc., and stores various data necessary for executing the operating system and the program 230.

The communication I/F unit 222 is connected to a network 240 such as the Internet or an intranet by wire or wirelessly, and transmits and receives data to and from other computers in accordance with a predetermined communication standard. The external device I/F unit 224 is connected to an external device 250 such as a printer or scanner by wire or wirelessly, and transmits and receives data to and from the external device 250 in accordance with a predetermined communication standard. The wireless communication means typically uses an international standard communication means. International standard communication means include IEEE802.15.4, IEEE802.15.1, IEEE802.15.2, IEEE802.15.3, IEEE802.15.4, IEEE802.15.5, IEEE802.15.6, IEEE802.15.7, IEEE802.15.8, IEEE802.15.9, IEEE802.15.10, IEEE802.15.11, IEEE802.15.12, IEEE802.15.13, IEEE802.15.14, IEEE802.15.15, IEEE802.15.16, IEEE802.15.17, IEEE802.15.18, IEEE802.15.19, IEEE802.15.26, IEEE802.15.27, IEEE802.15.28, IEEE802.15.29, IEEE802.16.29, IEEE802.16.30, IEEE802.16.40, IEEE802.16.51, IEEE802.16.52, IEEE802.16.63, IEEE802.16.74, IEEE802.16.85, IEEE802.16.96
02.15.11a, 11b, 11g, 11n, 11ac, 11ad, ISO/IEC1
There are also other methods such as Bluetooth 4513-3-10 and IEEE802.15.4g.
Bluetooth (registered trademark), Bluetooth Low Energy, Wi-Fi, ZigBe
e (registered trademark), Sub-GHz, EnOcean (registered trademark), etc. can also be used.

The I/O device I/F unit 226 is connected to an I/O device 260 such as various sensors and actuators, and transmits and receives various signals and data, such as detection signals from sensors and control signals to actuators, between the I/O device 260 and the I/O device 260.
28 is composed of a drive device such as a DVD drive, a CD drive, etc.
Data is read from and written to a medium 270 such as a CD.

In the computer 200 having the above configuration, the processor 212 executes the program 23.
2, the processor 212 loads the program 230 into a work memory area of the memory 214, executes the program 230, and controls each unit of the computer 200 via the bus 210. The program 230 may be stored in the storage device 220 instead of the memory 214. The program 230 may be recorded on a non-transitory recording medium such as a CD or DVD in an installable file format or an executable file format, and provided to the computer 200 via the media input/output unit 228. The program 230 may be provided to the computer 200 by being downloaded via the network 240 through the communication I/F unit 222. The computer 200 may also control various functions that are realized by the processor 212 executing the program 230, for example, as an FP.
It may be realized by hardware such as GA or ASIC.

The computer 200 is, for example, a desktop computer or a portable computer, and is an electronic device of any type. The computer 200 may be a client computer, a server computer, or a cloud computer.
0 may be applied to devices other than the fluid supply device 3 and the machine learning device 4.

(Machine learning device 4)
FIG. 4 is a block diagram showing an example of the machine learning device 4 according to the first embodiment.

The machine learning device 4 includes a learning data acquisition unit 40, a learning data storage unit 41, a machine learning unit 42, and a trained model storage unit 43. The machine learning device 4 is configured, for example, with a computer 200 shown in FIG. 3. In this case, the learning data acquisition unit 40 includes a communication I/F
The machine learning unit 42 is composed of the processor 212, and the learning data storage unit 41 and the trained model storage unit 43 are composed of the storage device 220.

The learning data acquisition unit 40 is an interface unit that is connected to various external devices via a communication network and acquires learning data including at least input data. The external devices are a sensor group 3a provided in the fluid supply device 3, a sensor group 3a provided in a test device 7 that simulates the fluid supply device 3, and an operator terminal 8 used by an operator. The learning data acquisition unit 40 may acquire learning data from the sensor group 3a provided in the fluid supply device 3 that is the subject of health diagnosis.

The test device 7 includes a flow sensor 301, a pump secondary pressure sensor 302, a pump operation status sensor 312, a power generation amount sensor 313, a usage flow rate sensor 140,
A water level gauge 100 and a float switch 101 are provided. The test device 7 may include a tank charge pressure sensor 321 that measures the charge pressure of the pressure tank 32.

The learning data storage unit 41 is a database that stores a plurality of sets of learning data acquired by the learning data acquisition unit 40. The specific configuration of the database that constitutes the learning data storage unit 41 may be designed as appropriate.

The machine learning unit 42 performs machine learning using the learning data stored in the learning data storage unit 41. That is, the machine learning unit 42 inputs a plurality of sets of learning data to the learning model 2, and causes the learning model 2 to learn the correlation between the input data included in the learning data and the diagnostic information of the pressure tank 32, thereby generating a trained learning model 2. In this embodiment, a case where a neural network is adopted as a specific method of supervised learning by the machine learning unit 42 will be described.

The trained model storage unit 43 is a database that stores the trained learning model 2 generated by the machine learning unit 42. The trained learning model 2 stored in the trained model storage unit 43 can be transmitted to a real system (e.g., a fluid supply device 3) via an arbitrary communication network, a recording medium, or the like.
4, the learning data storage unit 41 and the trained model storage unit 43 are shown as separate storage units, but they may be configured as a single storage unit.

FIG. 5 is a data configuration diagram showing an example of data (supervised learning) used by the machine learning device 4 according to the first embodiment.

The learning data includes, as input data, at least data on the amount of power generation during backflow. The input data may further include at least one of data on the discharge pressure of the pump 31 during backflow and data on the water level of the water tank 10 during backflow. The input data may further include other data in addition to the above.

When the input data includes, for example, data on the amount of power generation during backflow, data on the discharge side pressure during backflow, and data on the water level during backflow, each of these data is based on measurements taken by the power generation amount sensor 313, the pump secondary pressure sensor 302, and the water level meter 100 during a predetermined period when the water stored in the pressure tank 32 is caused to flow back to the suction side of the pump 31 in the backflow operation mode. The start point of the predetermined period is, for example, a time when the backflow mode execution unit 36
The predetermined period is the time point when the bypass valve 340 is opened and the three-way valve 303 is switched to the second communication state by command 2. The end point of the predetermined period is, for example, a time point set after a sufficient amount of time has elapsed for the remaining pressure in the pressure tank 32 to be released. Note that, when a relationship in which a time difference occurs between each piece of data is recognized, the predetermined period for each piece of data may be set to a period corresponding to the time difference.

The data on the amount of power generated during reverse flow is composed of time-series data consisting of multiple measured values (current values, voltage values) measured by the power generation sensor 313 at multiple measurement points within a specified period, or representative data calculated from the multiple measured values. The time-series data is, for example, data in which each estimated value is arranged in order of measurement time indicating the measurement point, and is composed of data showing changes over time. The representative data is, for example, data calculated based on multiple measured values, such as an integrated value of the amount of power generated during a specified period (integrated power value) or data calculated as the maximum value of the amount of power generated (maximum current value).

The data on the discharge side pressure during backflow is composed of time-series data consisting of multiple measured values (discharge side pressure of the pump 31) measured by the pump secondary pressure sensor 302 at multiple measurement points within a predetermined period, or representative data calculated from the multiple measured values. The time-series data is, for example, data arranged in order of measurement time indicating the measurement point, and is composed of data indicating changes over time. The representative data is, for example, data indicating the discharge side pressure at the start point, or data calculated based on the multiple measured values, such as the amount of change in the discharge side pressure over a predetermined period.

The backflow water level data is composed of time-series data consisting of multiple measured values (water level in the water tank 10) measured by the water level meter 100 at multiple measurement points within a specified period, or representative data calculated from the multiple measured values. The time-series data is, for example, data arranged in order of measurement time indicating the measurement point, and is composed of data showing changes over time. The representative data is, for example, data showing the water level at the start point, or data calculated based on the multiple measured values, such as the amount of change in water level over a specified period.

When "supervised learning" is adopted as the machine learning, the learning data further includes, as output data associated with the input data, diagnostic information indicating that the state of the pressure tank 32 is one of a plurality of states. In supervised learning, the output data is referred to as, for example, teacher data or correct answer label.

When diagnosing the health of the pressure tank 32 as the state of the pressure tank 32, the diagnostic information is configured as information indicating that the state of the pressure tank 32 is either normal or abnormal. In this case, the diagnostic information is classified into two values, for example, a value indicating that the pressure tank 32 is in a normal state is "0" and a value indicating that the pressure tank 32 is in an abnormal state is "1".
In this embodiment, the diagnosis information will be described as indicating either normality or abnormality.

Therefore, the learning data according to this embodiment is configured by associating input data including data on the amount of power generation during reverse flow with output data including diagnostic information indicating either normality or abnormality, as shown in Fig. 5. The abnormality here may include not only a retroactive abnormality in which the occurrence of an abnormality is found at the time of diagnosis, but also a sign of an abnormality that is within an acceptable range and is judged to be normal at the time of diagnosis, but which predicts the occurrence of an abnormality in the future.

In addition, the information indicating an abnormality among the diagnostic information may include, for example, multiple abnormalities according to the specific content and degree of the abnormality, corresponding to abnormality 1/abnormality 2/.../abnormality n shown in Figure 5.

Specific examples of the abnormality include an abnormality related to the pressure of the pressure tank 32 or an abnormality related to the diaphragm 320 of the pressure tank 32. In this case, the diagnostic information is multi-valued (
For example, a value indicating that the pressure tank 32 is in a normal state is "0," a value indicating that there is an abnormality related to the enclosed pressure is "1," a value indicating that there is an abnormality related to the diaphragm 320 is "2," and so on, depending on the nature of each abnormality.

For example, if the abnormality is related to the charged pressure, the specific degree of the abnormality may include a classification of the degree of decrease in the charged pressure into multiple levels. In this case, the diagnostic information may be multi-valued (
For example, a value indicating that the pressure tank 32 is in a normal state is "0," a value indicating that the degree of drop in the enclosed pressure is a low level of abnormality is "1," a value indicating that the degree of drop in the enclosed pressure is a medium level of abnormality is "2," and so on, according to the degree of each abnormality.

Here, the correlation between the input data included in the learning data and the diagnostic information of the pressure tank 32 will be described.

The data on the amount of power generation during backflow included as input data is a measurement of the amount of power generation caused by the residual pressure in the pressure tank 32, and therefore captures the magnitude of the remaining pressure in the pressure tank 32 and its change over time. If the diaphragm 320 is damaged or deteriorated, the degree of the damage or deterioration is considered to be reflected in the magnitude of the remaining pressure and its change over time. Therefore, by including the data on the amount of power generation during backflow as input data in the learning data, it becomes possible to diagnose the soundness of the pressure tank 32.

In addition, if the input data further includes data on the discharge side pressure of the pump 31 during backflow, the pressure in the discharge side piping 30b during backflow is also taken into account, making it possible to diagnose the health of the pressure tank 32 with greater accuracy.

In addition, if the input data further includes data on the water level during backflow of the pump 31, the effect of the water stored in the water tank 10 on the suction side piping 30a during backflow is also taken into consideration, making it possible to more accurately diagnose the soundness of the pressure tank 32. This is particularly effective when the fluid supply device 3 is configured without the branch piping 33 and the three-way valve 303.

When acquiring the above-mentioned learning data, the learning data acquisition unit 40 acquires data on the amount of power generation during backflow measured by the sensor group 3a (power generation amount sensor 313) provided in the fluid supply device 3 or the test device 7 as input data from the sensor group 3a.

In addition, when the worker diagnoses the state of the pressure tank 32 when the input data is measured by the sensor group 3 a and inputs the diagnosis value to the worker terminal 8, the learning data acquisition unit 40 outputs the diagnosis value inputted at the worker terminal 8 as output data (teaching data) to the worker terminal 8.
The learning data acquisition unit 40 then associates these input data and output data to form one learning data set, and stores the learning data set in the learning data storage unit 41.

FIG. 6 is a schematic diagram showing an example of a neural network model used in the machine learning device 4 according to the first embodiment.

The learning model 2 is configured as a neural network model shown in Fig. 6. The neural network model includes l neurons (x1 to xl) in the input layer, m neurons (y11 to y1m) in the first hidden layer, and n neurons (y2
The input layer is made up of o neurons (z1 to zo) in the output layer.

Each neuron in the input layer is associated with a respective input data included in the learning data. Each neuron in the output layer is associated with a respective output data included in the learning data. Note that a predetermined pre-processing may be performed on the input data before it is input to the input layer, and a predetermined post-processing may be performed on the output data after it is output from the output layer.

The first and second hidden layers are also called hidden layers. In addition to the first and second hidden layers, the neural network may further include a plurality of hidden layers.
Only the first hidden layer may be a hidden layer. Synapses that connect the neurons of each layer are laid 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, and a weight wi (i is a natural number) is assigned to each synapse.

A neural network model uses training data to learn the correlation between the input data and output data by inputting input data contained in the training data into an input layer, and comparing the output data output from the output layer as the inference result with the output data (teacher data) contained in the training data.

Specifically, each neuron in the input layer is input with input data included in the learning data, and the value of each neuron in the output layer is calculated by performing a process for all neurons other than the input layer, in which the value of the neuron on the input side connected to the neuron in question is calculated as the sum of a sequence of multiplication values of the value of the neuron on the input side connected to the neuron in question and the weight wi associated with the synapse connecting the neuron on the output side and the neuron on the input side.

Then, the values (z1 to zo) output to each neuron in the output layer as the inference results are compared with the values (t1 to to) of the teacher data corresponding to each of the output data included in the learning data to determine the error, and a process (backpropagation) is performed to adjust the weights wi associated with each synapse so that the error is reduced.

When a predetermined learning end condition is met, such as the above series of steps being repeated a predetermined number of times or the above error being smaller than a tolerance, the machine learning is terminated and the trained neural network model (all weights w i associated with each synapse)
is generated as:

(Machine learning methods)
7 is a flowchart showing an example of a machine learning method performed by the machine learning device 4 according to the first embodiment. The machine learning method corresponds to the learning phase in FIG.

First, in step S100, the learning data acquisition unit 40 prepares a desired number of pieces of learning data as advance preparation for starting machine learning, and stores the prepared learning data in the learning data storage unit 41. The number of pieces of learning data to be prepared here may be set in consideration of the inference accuracy required for the learning model 2 to be finally obtained.

There are several methods for preparing training data. For example,
When an abnormality occurs in the pressure tank 32 in a specific fluid supply device 3 or test device 7, or when an operator recognizes a symptom of an abnormality, the water stored in the pressure tank 32 is made to flow backward to the suction side of the pump 31, and various measurement values for a predetermined period at that time are acquired by the sensor group 3a. Then, the operator uses the operator terminal 8 to input the diagnosis result in a manner corresponding to these measurement values, and prepares input data (including at least data on the amount of power generated during backward flow) and output data (for example, the value of the output data in this case is "1") that constitute the learning data. Then, by repeating such operations, it is possible to prepare multiple sets of learning data. As another method, for example, it is also possible to acquire learning data by intentionally causing an abnormal state in the pressure tank 32 of the fluid supply device 3 or test device 7. Furthermore, as the learning data, multiple sets of learning data consisting of input data and output data (for example, the value of the output data in this case is "0") are prepared not only when an abnormality occurs, but also when no abnormality occurs, that is, when the fluid supply device 3 or test device 7 is in a normal state.

Next, in step S110, the machine learning unit 42 prepares a pre-learning learning model 2 to start machine learning. The pre-learning learning model 2 prepared here is configured with the neural network model exemplified in FIG. 6, and the weights of each synapse are set to initial values. Each neuron in the input layer is associated with data on the amount of power generation during reverse flow as input data contained in the learning data. Each neuron in the output layer is associated with diagnostic information as output data contained in the learning data.

Next, in step S<b>120 , the machine learning unit 42 acquires, for example, one piece of learning data randomly from the multiple sets of learning data stored in the learning data storage unit 41 .

Next, in step S130, the machine learning unit 42 inputs the input data included in one learning data to the input layer of the prepared learning model 2 before learning (or during learning). As a result, output data is output as an inference result from the output layer of the learning model 2, but the output data is generated by the learning model 2 before learning (or during learning). Therefore, in the state before learning (or during learning), the output data output as an inference result indicates information different from the output data (teacher data) included in the learning data.

Next, in step S140, the machine learning unit 42 performs machine learning by comparing the output data (teacher data) included in the one learning data acquired in step S120 with the output data output from the output layer as an inference result in step S130 and adjusting the weight of each synapse. In this way, the machine learning unit 42 causes the learning model 2 to learn the correlation between the input data and the output data (diagnosis information of the pressure tank 32).

For example, in a case where the diagnostic information constituting the teacher data is defined as a binary classification in which a normal state is indicated by "0" and an abnormal state is indicated by "1," the value of the output data included in one learning data set selected in step S120 is "1," but the value of the output data output from the output layer is a predetermined value between 0 and 1, specifically, for example, a value such as "0.63." In this case, in step S140, the weights associated with each synapse of the learning model 2 being trained are adjusted so that the value output from the output layer approaches "1" if similar input data is input to the input layer of the learning model 2 being trained.

Next, in step S150, the machine learning unit 42 determines whether or not it is necessary to continue machine learning based on, for example, the error between the output data and the teacher data and the remaining number of unlearned learning data stored in the learning data memory unit 41.

In step S150, when the machine learning unit 42 determines to continue the machine learning (No in step S150), the process returns to step S120, and the process of steps S120 to S140 is performed multiple times on the learning model 2 being trained using untrained learning data.
In step S150, if the machine learning unit 42 determines that the machine learning is to be ended (Yes in step S150), the process proceeds to step S160.

Then, in step S160, the machine learning unit 42 stores the trained learning model 2 generated by adjusting the weights associated with each synapse in the trained model storage unit 43.
7, and ends the series of machine learning method steps shown in Fig. 7. In the machine learning method, step S100 corresponds to a learning data storage step, steps S110 to S150 correspond to a machine learning step, and step S160 corresponds to a trained model storage step.

As described above, the machine learning device 4 and machine learning method of this embodiment can provide a learning model 2 that can infer (estimate) diagnostic information for the pressure tank 32 with high accuracy from data on the amount of power generation during backflow.

(Pressure Tank Diagnostic Device 5)
FIG. 8 is a block diagram showing an example of the pressure tank diagnosis device 5 (pressure tank diagnosis unit 363) according to the first embodiment.

The pressure tank diagnostic device 5 includes an input data acquisition unit 50, an inference unit 51, a trained model storage unit 52, and an output processing unit 53. The pressure tank diagnostic device 5 is configured, for example, by a computer 200 shown in FIG. 3. In this case, the input data acquisition unit 50 includes a communication I/F unit 2
22 or an I/O device I/F unit 226, and the inference unit 51 and the output processing unit 53 are
The pressure tank diagnostic device 5 is configured with a processor 212, and the learned model storage unit 52 is configured with a storage device 220. In this embodiment, the pressure tank diagnostic device 5 is incorporated in the control panel 36 of the fluid supply device 3, but may be configured as a device separate from the fluid supply device 3, for example, a general-purpose or dedicated computer (see FIG. 3).

The input data acquisition unit 50 is an interface unit that is connected to the sensor group 3a provided in the fluid supply device 3 and acquires input data (data on the amount of power generation during backflow) based on measurements taken by the sensor group 3a. The input data includes at least the data on the amount of power generation during backflow.

The inference unit 51 inputs the input data acquired by the input data acquisition unit 50 into the learning model 2, and performs an inference process to infer diagnostic information for the pressure tank 32. For the inference process, the machine learning device 4 and the trained learning model 2 on which supervised learning has been performed using the machine learning method are used.

The inference unit 51 not only has a function of performing inference processing using the learning model 2, but also has a pre-processing function of adjusting the input data acquired by the input data acquisition unit 50 into a desired format, etc., and inputting the adjusted data to the learning model 2 as pre-processing of the inference processing, and a post-processing function of applying a predetermined logical formula or calculation formula to the value of the output data output from the learning model 2 as post-processing of the inference processing, thereby finally determining the state of the pressure tank 32. Note that it is preferable that the inference results of the inference unit 51 are stored in the trained model storage unit 52 or another storage device (not shown), and past inference results can be used as learning data to be used for online learning or re-learning, for example, in order to further improve the inference accuracy of the learning model 2.

The trained model storage unit 52 is a database that stores trained learning models 2 used in the inference process of the inference unit 51. The number of learning models 2 stored in the trained model storage unit 52 is not limited to one. For example, multiple learning models 2 with different amounts of input data or different machine learning methods may be stored and selectively available.

The output processing unit 53 performs an output process for outputting the inference result of the inference unit 51, that is, the diagnosis information of the pressure tank 32. As a specific output means, various means can be adopted.
The output processing unit 53 notifies the operator of the diagnostic information by display or sound via the operation display panel 37, stores the diagnostic information in the storage unit of the control panel 36 as a diagnostic history of the pressure tank 32, transmits the diagnostic information to a higher-level management device (not shown) of the fluid supply device 3, and
The diagnostic information may be used for water supply control.

(Pressure tank diagnostic method)
FIG. 9 is a flowchart showing an example of a pressure tank diagnosis method by the fluid supply device 3 including the pressure tank diagnosis unit 363 (pressure tank diagnosis device 5) according to the first embodiment.
Here, the diagnosis information of the pressure tank 32 is assumed to be defined as a binary classification representing either normal "0" or abnormal "1." The pressure tank diagnosis method corresponds to the inference phase of FIG.

When the power of the fluid supply device 3 is turned on and an operator operates the operation display panel 37 to start automatic operation, a series of steps of the pressure tank diagnosis method shown in Figure 9 is executed by the pressure tank diagnosis device 5.

First, in step S200, the mode switching unit 360 of the control panel 36
As a result, if the mode switching unit 360 determines that the diagnosis is necessary (Yes in step S200), the process proceeds to step S210. If the mode switching unit 360 determines that the diagnosis is not necessary (No in step S200), the process proceeds to step S210.
Go to 220.

Next, in step S210, the mode switching unit 360 determines whether or not there is a plan to use the fluid at the fluid supply destination 15. As a result, if the mode switching unit 360 determines that there is no plan to use the fluid (No in step S210), the process proceeds to step S230, and if it determines that there is a plan to use the fluid (Yes in step S210), the process proceeds to step S220.

Then, in step S220, the mode switching unit 360 switches the operation mode to the normal operation mode, whereby the normal mode execution unit 361 controls the operation state of the pump 31 based on the discharge side pressure of the pump 31.

On the other hand, in step S230, the mode switching unit 360 switches the operation mode to the reverse flow operation mode, whereby the reverse flow mode execution unit 362 opens the bypass valve 340 and closes the three-way valve 30.
The pressure tank 32 is switched to the second communication state of 3, and the water stored in the pressure tank 32 is caused to flow back to the suction side of the pump 31.

At this time, in step S240, in parallel with step S230, the power generation amount sensor 313
measures the amount of power generated when the electric motor 310 operates as a generator. Then, the pressure tank diagnosis unit 363 (input data acquisition unit 50) acquires data on the amount of power generated during backflow as input data based on the measured amount of power generated.

Next, in step S250, the pressure tank diagnosis unit 363 (inference unit 51) preprocesses the input data and inputs it to the input layer of the learning model 2, and obtains the output data output from the output layer of the learning model 2.

Next, in step S260, the pressure tank diagnosis unit 363 (inference unit 51), as an example of post-processing of supervised learning, compares the value of the output data (a number between 0 and 1) with a predetermined threshold value, and, for example, if the value of the output data is less than the predetermined threshold value, determines that the diagnostic information is "normal," and if the value is equal to or greater than the predetermined threshold, determines that the diagnostic information is "abnormal," and outputs the determination result as the inference result.

Next, in step S270, the pressure tank diagnosis unit 363 (inference unit 51) determines whether the diagnostic information of the pressure tank 32, which is the inference result of the inference unit 51, indicates “normal” or “abnormal”. If it determines that the diagnostic information is “normal”, the process proceeds to step S280, and if it determines that the diagnostic information is “abnormal”, the process proceeds to step S290.

Then, in step S280, the pressure tank diagnosis unit 363 (output processing unit 53)
The information indicating "normal" is output. Note that step S280 may be omitted.

In step S290, the pressure tank diagnosis unit 363 (output processing unit 53)
Then, information indicating "abnormality" is output, and the series of pressure tank diagnostic methods shown in Fig. 9 is terminated. In the pressure tank diagnostic method, step S230 corresponds to a fluid backflow process, step S240 corresponds to an input data acquisition process, steps S250 to S260 correspond to a diagnosis process, and steps S270 to S290 correspond to an output processing process.

As described above, according to the fluid supply device 3 and pressure tank diagnostic method of this embodiment, it is possible to diagnose the soundness of the pressure tank 32 without providing a new pressure sensor for diagnostic purposes.

Second Embodiment
In the first embodiment, a case where "supervised learning" is adopted as a machine learning technique is described, but in the present embodiment, a case where "unsupervised learning" is adopted is described.
The basic configuration and operation of the pressure tank diagnosis unit 363 are the same as those in the first embodiment.
The following description will focus on the differences from the first embodiment.

(Machine learning device 4)
The machine learning device 4 includes a learning data acquisition unit 40, a learning data storage unit 41, a machine learning unit 42, and a trained model storage unit 43, similar to the first embodiment (see FIG. 4).

FIG. 10 shows data (unsupervised learning) used by the machine learning device 4 according to the second embodiment.
FIG. 13 is a data configuration diagram showing an example of the data structure.

When "unsupervised learning" is adopted as the machine learning, the learning data includes only the input data when the diagnostic information indicates that the state of the pressure tank 32 is a predetermined state. In other words, the learning data may be configured not to include output data, and the presence or absence of output data and the format of the output data can be appropriately selected depending on the machine learning method adopted by the machine learning device 4.

When diagnosing the soundness of the pressure tank 32 as the state of the pressure tank 32, the diagnosis information is configured as information indicating that the state of the pressure tank 32 is either normal or abnormal. In this case, the learning data according to this embodiment is configured only with data on the amount of power generation during backflow when the state of the pressure tank 32 is normal, as shown in Fig. 10 .

When acquiring the above learning data, the learning data acquiring unit 40 acquires data on the amount of power generation during backflow measured by the sensor group 3a (power generation sensor 313) provided in the fluid supply device 3 or the testing device 7 as input data from the sensor group 3a. Then, when an operator diagnoses the soundness of the pressure tank 32 from which the input data was measured by the sensor group 3a and inputs to the operator terminal 8 that the diagnosis result is normal, the learning data acquiring unit 40 constructs one piece of learning data from only that input data and stores it in the learning data storage unit 41.

The machine learning unit 42 inputs one or more sets of learning data into the learning model 2, and generates a trained learning model 2 by having the learning model 2 learn the correlation between the input data contained in the learning data and diagnostic information indicating that the condition of the pressure tank 32 is normal.
In this embodiment, a case will be described in which an autoencoder is used as a specific method for unsupervised learning by the machine learning unit 42.

FIG. 11 is a schematic diagram illustrating an example of an autoencoder model used in the machine learning device 4 according to the second embodiment.

Learning model 2 is configured as an autoencoder model shown in Fig. 10. The autoencoder model is configured with l neurons (x1 to xl) in the input layer, m neurons (y1 to ym) in the intermediate layer, and o neurons (z1 to zo) in the output layer. The number of neurons in the input layer and the output layer is the same, and is greater than the number of neurons in the intermediate layer.

Each neuron in the input layer is associated with a respective input data included in the learning data. The intermediate layer is also called a hidden layer, and may have multiple hidden layers. In addition, between the input layer and the intermediate layer, and between the intermediate layer and the output layer, synapses that connect the neurons of each layer are laid, and a weight wi (i is a natural number) is associated with each synapse.

The autoencoder model uses training data to input input data contained in the training data into an input layer, compares the output data output from the output layer as the inference result with the input data contained in the training data, and performs a process of adjusting the weights wi, thereby learning patterns and trends in the input data.

When a specified learning termination condition is met, such as the above series of steps being repeated a specified number of times or the above error being smaller than an allowable value, the machine learning is terminated and a learned autoencoder model (all weights wi associated with each synapse) is generated.

(Machine learning methods)
FIG. 12 is a flowchart showing an example of a machine learning method performed by the machine learning device 4 according to the second embodiment.

First, in step S300, the learning data acquisition unit 40 prepares a desired number of learning data as a preliminary preparation for starting machine learning, and stores the prepared learning data in the learning data storage unit 41.

There are several methods for preparing training data. For example,
Input data (including at least data on the amount of power generated during backflow) constituting the learning data is prepared by acquiring various measurement values by the sensor group 3a when the pressure tank 32 in a specific fluid supply device 3 or test device 7 is in a normal state. By repeating this process, multiple sets of learning data can be prepared.

Next, in step S310, the machine learning unit 42 prepares a pre-learning learning model 2 to start machine learning. The pre-learning learning model 2 prepared here is configured with the autoencoder model exemplified in Fig. 11, and the weights of each synapse are set to initial values. Each neuron in the input layer is associated with data on the amount of power generation during reverse flow as input data included in the learning data.

Next, in step S320, the machine learning unit 42 acquires, for example, one piece of learning data randomly from the multiple sets of learning data stored in the learning data storage unit 41.

Next, in step S330, the machine learning unit 42 inputs the input data included in one learning data to the input layer of the prepared learning model 2 before learning (or during learning). As a result, output data is output as an inference result from the output layer of the learning model 2, but the output data is generated by the learning model 2 before learning (or during learning). Therefore, in the pre-learning (or during learning) state, the output data output as an inference result indicates information different from the input data included in the learning data.

Next, in step S340, the machine learning unit 42 compares the input data included in the one learning data acquired in step S320 with the output data output from the output layer as an inference result in step S330, and adjusts the weights of each synapse, thereby
Implement machine learning.

Next, in step S350, the machine learning unit 42 determines whether or not it is necessary to continue the machine learning. As a result, if it is determined that the machine learning should be continued (No in step S350), the process returns to step S320 and executes steps S320 to S340 on the learning model 2 being trained. If it is determined that the machine learning should be terminated (Yes in step S350), the process proceeds to step S360.

Then, in step S360, the machine learning unit 42 stores the trained learning model 2 generated by adjusting the weights associated with each synapse in the trained model storage unit 43.
12, and ends the series of machine learning method steps shown in Fig. 12. In the machine learning method, step S300 corresponds to a learning data storage step, steps S310 to S350 correspond to a machine learning step, and step S360 corresponds to a trained model storage step.

As described above, the machine learning device 4 and machine learning method of this embodiment can provide a learning model 2 that can infer (estimate) diagnostic information for the pressure tank 32 with high accuracy from data on the amount of power generation during backflow.

(Pressure Tank Diagnostic Device 5)
The pressure tank diagnostic device 5 includes an input data acquisition unit 5, similar to the first embodiment (see FIG. 8).
0, an inference unit 51, a learned model storage unit 52, and an output processing unit 53.

The inference unit 51 inputs the input data acquired by the input data acquisition unit 50 to the learning model 2, and performs an inference process to infer diagnostic information for the pressure tank 32. For the inference process, the machine learning device 4 and the trained learning model 2 that has undergone unsupervised learning using the machine learning method are used.

(Pressure tank diagnostic method)
13 is a flowchart showing an example of a pressure tank diagnostic method by the fluid supply device 3 equipped with the pressure tank diagnostic unit 363 (pressure tank diagnostic device 5) according to the second embodiment. In FIG. 13, the diagnostic information of the pressure tank 32 is defined as a binary classification representing either normal or abnormal. In addition, steps S400 to S440 in FIG. 13
Since steps S200 to S240 are similar to steps S200 to S240 in FIG. 9, the following description will focus on steps 450 and after.

In step S450, the pressure tank diagnosis unit 363 (inference unit 51) preprocesses the input data and inputs it to the input layer of the learning model 2, and obtains the output data output from the output layer of the learning model 2.

Next, in step S460, the pressure tank diagnosis unit 363 (inference unit 51) determines the difference between the feature amount based on the input data and the feature amount based on the output data as an example of post-processing of unsupervised learning. The feature amount is, for example, a parameter expressed as a feature vector in a multidimensional space, and the difference is expressed as the distance between the feature vectors. Then, the inference unit 51 determines the difference between the feature amount based on the input data and the feature amount based on the output data as an example of post-processing of unsupervised learning. The feature amount is, for example, a parameter expressed as a feature vector in a multidimensional space, and the difference is expressed as the distance between the feature vectors.
If the difference is less than a predetermined threshold, the diagnostic information is judged to be "normal," and if the difference is greater than or equal to the predetermined threshold, the diagnostic information is judged to be "abnormal," and the judgment result is output as the inference result.

Next, in step S470, the pressure tank diagnosis unit 363 (output processing unit 53) determines whether the diagnosis information of the pressure tank 32, which is the inference result of the inference unit 51, indicates "normal" or "abnormal". If it is determined to be "normal", the process proceeds to step S480.
If it is determined to be "abnormal", the process proceeds to step S490 (optional) and the series of pressure tank diagnostic methods shown in Fig. 13 is terminated. In the pressure tank diagnostic method, step S430 corresponds to a fluid backflow process, step S440 corresponds to an input data acquisition process, steps S450 to S460 correspond to a diagnosis process, and steps S470 to S490 correspond to an output processing process.

As described above, according to the fluid supply device 3 and pressure tank diagnostic method of this embodiment, it is possible to diagnose the soundness of the pressure tank 32 without providing a new pressure sensor for diagnostic purposes.

Other Embodiments
The present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit and scope of the present invention. All such modifications are included in the technical concept of the present invention.

In the above embodiment, the water supply system 1 has been described as adopting a water tank type having a water tank 10, but a direct water supply type not having a water tank 10 may also be adopted. In the direct water supply type, a backflow prevention device is generally provided in the suction side piping 30a. Therefore, when the water stored in the pressure tank 32 is caused to flow back to the suction side of the pump 31, the fluid supply device 3
It is preferable to provide a piping system that opens the suction side of pump 31 without allowing water to flow back into main water pipe 11 while bypassing the backflow prevention device.

In the above embodiment, the fluid supply device 3 diagnoses the soundness of the pressure tank 32 when the mode switching unit 360 switches to the reverse flow operation mode. However, the fluid supply device 3 may diagnose the soundness of the pressure tank 32 when an instruction to execute the diagnosis is received from an operator. In this case, steps S200 and S210 in FIG. 9, step S400 in FIG. 13,
The determination in S410 may be omitted.

In the above embodiment, the pressure tank diagnosis unit 363 uses the learning model 2 to
In the above description, the case where the soundness of the pressure tank 32 is diagnosed has been described, but the soundness of the pressure tank 32 may be diagnosed by other methods without using the learning model 2. The pressure tank diagnosis unit 363 may diagnose the soundness of the pressure tank 32 by, for example, calculating an evaluation value such as a maximum value, a minimum value, an average value, a differential value, an integral value, etc. from the data on the amount of power generation during backflow, and comparing the evaluation value with a reference value such as an upper limit value, a lower limit value, etc.

In the above embodiment, the pressure tank diagnostic unit 363 diagnoses the soundness of the pressure tank 32. However, the soundness of more detailed phenomena of the pressure tank 32 may be diagnosed. For example, the pressure tank diagnostic unit 363 may estimate the charged pressure of the pressure tank 32 based on the data of the power generation amount during backflow, and diagnose the soundness of the charged pressure. In this case, the pressure tank diagnostic unit 363 may estimate the charged pressure using a learning model 2 that has been machine-learned to learn the correlation between the input data and the charged pressure of the pressure tank 32, or may estimate the charged pressure using a predetermined conversion formula for converting the charged pressure of the pressure tank 32 based on an evaluation value calculated from the data of the power generation amount during backflow.

In the above embodiment, the case where a neural network (first embodiment) and an autoencoder (second embodiment) are adopted as specific methods of machine learning by the machine learning unit 42 has been described, but the machine learning unit 42 may adopt any other machine learning method. Examples of other machine learning methods include tree types such as decision trees and regression trees, ensemble learning such as bagging and boosting, neural network types (including deep learning) such as recurrent neural networks and convolutional neural networks, clustering types such as hierarchical clustering, non-hierarchical clustering, k-nearest neighbors, and k-means, multivariate analysis such as principal component analysis, factor analysis, and logistic regression, and support vector machines.

(Machine learning program)
The present invention can also be provided in the form of a program (machine learning program) 230 for causing a computer 200 shown in FIG. 3 to execute each step of the machine learning method according to the above embodiment.

(Pressure tank diagnostic program)
The present invention can also be provided in the form of a program (pressure tank diagnosis program) 230 for causing a computer 200 shown in FIG. 3 to execute each step of the pressure tank diagnosis method according to the above embodiment.

(Inference device, inference method, and inference program)
The present invention can be provided not only in the form of the pressure tank diagnostic device 5 (pressure tank diagnostic method or pressure tank diagnostic program) according to the above embodiment, but also in the form of an inference device (inference method or inference program) used to diagnose the soundness of the pressure tank 32. In that case, the inference device (inference method or inference program) can include a memory and a processor, and the processor can execute a series of processes.
This series of processes includes an input data acquisition process (input data acquisition step) that acquires data on the amount of power generated during backflow (which may further include at least one of data on the discharge side pressure during backflow and data on the water level during backflow) as input data, and an inference process (inference step) that infers diagnostic information for the pressure tank 32 once the input data has been acquired in the input data acquisition process.

By providing it in the form of an inference device (inference method or inference program), it can be easily applied to various devices compared to the case of implementing the pressure tank diagnosis device 5. It can be naturally understood by those skilled in the art that when the inference device (inference method or inference program) infers the soundness of the pressure tank 32, an inference method implemented by the inference unit 51 of the pressure tank diagnosis device 5 may be applied using the machine learning device 4 and the trained learning model 2 generated by the machine learning method according to the above embodiment.

1...Water supply system, 2...Learning model,
3...Fluid supply device, 3a...Sensor group, 3b...Control equipment group,
4...machine learning device, 5...pressure tank diagnostic device, 7...test device, 8...operator terminal,
10...water tank, 11...water main pipe, 12...distribution pipe, 13...connection pipe,
14... Water supply pipe, 15... Fluid supply destination,
30a...suction side piping, 30b...discharge side piping, 31...pump, 32...pressure tank,
33: Branch pipe, 34: Bypass pipe, 36: Control panel, 37: Operation display panel, 40: Learning data acquisition unit, 41: Learning data storage unit,
42...machine learning unit, 43...model storage unit,
100...water level gauge, 101...float switch,
120... inlet valve, 140... usage flow rate sensor,
200... computer, 210... bus, 212... processor, 214... memory,
216: input device, 218: display device, 220: storage device,
222: communication I/F unit, 224: external device I/F unit, 226: I/O device I/F unit,
228...media input/output unit, 230...program, 240...network,
250: external device, 260: I/O device, 270: media,
300: check valve; 301: flow sensor; 302: pump secondary pressure sensor;
303: three-way valve; 310: motor; 311: inverter;
312: Pump operation status sensor, 313: Power generation amount sensor, 320: Diaphragm,
321: Tank pressure sensor; 340: Bypass valve;
360: mode switching unit; 361: normal mode execution unit;
362: Backflow operation mode execution unit, 363: Pressure tank diagnosis unit

Claims (15)

A pump that transfers a fluid using an electric motor as a drive source;
a pressure tank provided on a discharge side of the pump and configured to hold a pressure of the fluid;
a power generation amount measuring device for measuring the amount of power generated when the electric motor operates as a generator;
A control unit that controls the supply operation of the fluid,
The control unit is
When the fluid stored in the pressure tank is caused to flow back to the suction side of the pump, the amount of power generated measured by the power generation measuring device is acquired as data on the amount of power generated during backflow, and the soundness of the pressure tank is diagnosed based on the data on the amount of power generated during backflow.
Fluid supply device. The control unit is
a charged pressure of the pressure tank is estimated based on the amount of power generated during backflow, and the soundness of the charged pressure is diagnosed;
The fluid supply device of claim 1 . The control unit is
determining whether or not a fluid supply destination to which the fluid is supplied from the pressure tank has a schedule for using the fluid;
When it is determined that the pump is scheduled to be used, the operation of the pump is controlled based on the discharge side pressure of the pump.
When it is determined that there is no plan to use the pressure tank, the fluid stored in the pressure tank is caused to flow back to the suction side of the pump, and the soundness of the pressure tank is diagnosed.
The fluid supply device according to claim 1 or 2. a discharge side pipe connecting the pump and the pressure tank and having a check valve provided thereon;
a bypass pipe connected to the discharge side pipe so as to bypass the check valve and having a bypass valve provided thereon;
The control unit is
by opening the bypass valve, the fluid stored in the pressure tank is caused to flow back to the suction side of the pump.
The fluid supply device according to any one of claims 1 to 3. The control unit is
The data on the amount of power generation during backflow, obtained from the amount of power generation measured by the power generation measuring device, is input as input data into a learning model that has been machine-learned to learn the correlation between input data including at least the data on the amount of power generation during backflow and diagnostic information of the pressure tank, thereby diagnosing the soundness of the pressure tank when the amount of power generation during backflow is obtained.
A fluid supply device according to any one of claims 1 to 4. a pressure measuring device for measuring a discharge side pressure of the pump;
The control unit is
When the fluid stored in the pressure tank is caused to flow back to the suction side of the pump, the discharge side pressure measured by the pressure measuring instrument is acquired as data of the discharge side pressure during backflow;
diagnosing the soundness of the pressure tank based on the data of the power generation amount during backflow as well as the data of the discharge side pressure during backflow;
A fluid supply device according to any one of claims 1 to 4. The pump comprises:
The suction side is connected to a water tank having a water level gauge for measuring a water level,
The control unit is
When the fluid stored in the pressure tank is caused to flow backward to the suction side of the pump, the water level measured by the water level gauge is acquired as data on the water level during backward flow;
diagnosing the soundness of the pressure tank based on the data of the power generation amount during backflow and the data of the water level during backflow;
A fluid supply device according to any one of claims 1 to 4. 1. An inference device that is provided on a discharge side of a pump that uses an electric motor as a drive source to transfer a fluid, and is used to diagnose the soundness of a pressure tank that holds a pressure of the fluid, comprising:
The inference device comprises a memory and a processor;
The processor,
an input data acquisition process for acquiring data on the amount of power generated during reverse flow as input data based on the amount of power generated when the electric motor operates as a generator by causing the fluid stored in the pressure tank to flow reversely to the suction side of the pump;
When the input data is acquired by the input data acquisition process, an inference process is executed to infer diagnostic information of the pressure tank.
Inference device. 1. A machine learning device that generates a learning model for diagnosing the health of a pressure tank that is provided on a discharge side of a pump that uses an electric motor as a drive source to transport a fluid and that holds a pressure of the fluid,
a learning data storage unit that stores a plurality of sets of learning data including at least data on the amount of power generation during reverse flow as input data, based on the amount of power generation when the electric motor operates as a generator by causing the fluid stored in the pressure tank to flow reversely to the suction side of the pump;
a machine learning unit that inputs a plurality of sets of the learning data to the learning model, thereby causing the learning model to learn a correlation between the input data and the diagnostic information of the pressure tank;
a trained model storage unit that stores the learning model trained by the machine learning unit;
Equipped with
Machine learning device. A pressure tank diagnostic method for diagnosing the soundness of a pressure tank that is provided on a discharge side of a pump that uses an electric motor as a drive source to transport a fluid and that retains pressure of the fluid, comprising:
a fluid backflow process for backflowing the fluid stored in the pressure tank to a suction side of the pump;
an input data acquisition step of measuring an amount of power generated when the electric motor operates as a generator in parallel with the fluid reverse flow step, and acquiring data on the amount of power generated during reverse flow as input data based on the measured amount of power generated;
and a diagnosis step of diagnosing the soundness of the pressure tank based on the amount of power generation during backflow acquired by the input data acquisition step.
Pressure tank diagnostic methods. 1. An inference method used to diagnose the soundness of a pressure tank that is provided on the discharge side of a pump that transfers a fluid using an electric motor as a drive source and that retains pressure of the fluid, comprising:
an input data acquisition process for acquiring data on the amount of power generated during reverse flow as input data based on the amount of power generated when the electric motor operates as a generator by causing the fluid stored in the pressure tank to flow reversely to the suction side of the pump;
When the input data is acquired in the input data acquisition step, an inference step of inferring diagnostic information of the pressure tank is executed.
Inference methods. 1. A machine learning method for generating a learning model for diagnosing the health of a pressure tank that is provided on a discharge side of a pump that uses an electric motor as a drive source to transport a fluid and that holds a pressure of the fluid, comprising:
a learning data storage step of storing, in a learning data storage unit, a plurality of sets of learning data including at least data on the amount of power generated during reverse flow as input data, based on the amount of power generated when the electric motor operates as a generator by causing the fluid stored in the pressure tank to flow reversely to the suction side of the pump;
a machine learning process for inputting a plurality of sets of the learning data into the learning model, thereby causing the learning model to learn a correlation between the input data and the diagnostic information of the pressure tank;
A trained model storage step of storing the trained model trained by the machine learning step in a trained model storage unit.
Machine learning methods. A method for causing a computer to execute each step of the pressure tank diagnosis method according to claim 10,
Pressure tank diagnostic program. A method for causing a computer to execute each step of the inference method according to claim 11,
Inference program. A method for causing a computer to execute each step of the machine learning method according to claim 12,
Machine learning programs.

Images