CN115280055A

CN115280055A - Machine learning device, data processing system, inference device, and machine learning method

Info

Publication number: CN115280055A
Application number: CN202180021504.XA
Authority: CN
Inventors: 青山文明
Original assignee: Kaneko Sangyo Co Ltd
Current assignee: Kaneko Sangyo Co Ltd
Priority date: 2020-04-15
Filing date: 2021-04-09
Publication date: 2022-11-01
Also published as: JP6779457B1; JP2021169838A; WO2021210497A1

Abstract

The invention provides a machine learning device and the like capable of accurately grasping abnormality and abnormality signs in a fluid pressure driven valve without depending on experience of an operator and the like. A machine learning device (200) is applied to a fluid pressure-driven valve (10) that includes a main valve (11), a drive device (12) including a cylinder and a piston that drives the main valve (11), and a solenoid valve (1) that controls supply and discharge of a drive fluid (A) to and from the drive device (12), and the machine learning device (200) is provided with: a learning data set storage means (202) for storing a plurality of sets of learning data sets each composed of input data including a valve opening degree of a main valve (11), a pressure of an output side drive fluid supplied from a solenoid valve (1) to a drive device (12), and a position of a piston (122) relative to a cylinder, and output data composed of diagnostic information of the drive device (12) associated with the input data; a learning unit (203) which learns a learning model for inferring a correlation between input data and output data by inputting a plurality of sets of data for learning; and a learning-completed model storage unit (204) that stores the learning-completed model.

Description

Machine learning device, data processing system, inference device, and machine learning method

Technical Field

The present invention relates to a machine learning device, a data processing system, an inference device, and a machine learning method for diagnosing an abnormality of a valve system.

Background

Conventionally, a fluid pressure-driven valve that controls a driving fluid using a solenoid valve to open and close a main valve is known. For example, patent document 1 discloses an emergency shut-off valve device that, as a fluid pressure-driven valve used in a piping of a plant, shuts off a fluid flowing through the piping by controlling the driven fluid using a solenoid valve and closing a ball valve (main valve) in an emergency when an abnormality occurs in the plant.

Documents of the prior art

Patent document

Patent document 1: japanese laid-open patent publication No. 2009-97539

Disclosure of Invention

Problems to be solved by the invention

In a fluid pressure-driven valve such as an emergency shut-off valve used in a plant as disclosed in patent document 1, it is preferable that unexpected abnormality does not occur in order to improve the operation rate and reliability of the whole plant. Therefore, in such a fluid pressure driven valve, it is desired to implement not only post-maintenance for grasping an abnormality when the abnormality occurs, but also predictive maintenance for grasping a sign of the abnormality.

Here, the sign of abnormality of the fluid pressure-actuated valve may be expressed as various events, but the causal relationship between the event and the sign of abnormality that may occur in the fluid pressure-actuated valve has not been clearly determined. As a result, predictive maintenance of the fluid pressure driven valve is performed based on judgment depending on experience (including implicit knowledge) of an operator, and there is a problem that accuracy thereof varies depending on the operator in charge.

The present invention has been made in view of the above-described problems, and an object thereof is to provide a machine learning device, a data processing system, an inference device, and a machine learning method for accurately grasping an abnormality and a sign of the abnormality (hereinafter, these are collectively referred to as "abnormality" in the present invention) in a fluid pressure-driven valve.

Means for solving the problems

In order to achieve the above object, a machine learning device according to a first aspect of the present invention is applied to a fluid pressure-driven valve 10 of a solenoid valve 1 including at least a main valve 11, a driving device 12 including a cylinder 120 and a piston 122 for driving the main valve 11, and a solenoid portion 3 for controlling supply and discharge of a driving fluid a to and from the driving device 12, as shown in fig. 1 to 5, for example, and includes: a learning data set storage unit 202 that stores a plurality of sets of learning data sets including input data including a valve opening degree of the main valve 11, a pressure of an output side driving fluid supplied from the solenoid valve 11 to the driving device 12, and a position of the piston 122 with respect to the cylinder 120, and output data including diagnostic information of the driving device 12 associated with the input data; a learning unit 203 that learns a learning model that infers a correlation between the input data and the output data by inputting a plurality of sets of the data set for learning; and a learned model storage unit 204 that stores the learning model learned by the learning unit 203.

Effects of the invention

The machine learning device according to the present invention can provide a learned model that can estimate the presence or absence of an abnormality in the driving device based on various information and the like that can be acquired when the fluid pressure-driven valve is operating stably. Therefore, by using the learned model, it is possible to accurately estimate an abnormality occurring in the drive device without depending on the experience of the operator.

Drawings

Fig. 1 is a schematic diagram showing an example of a fluid pressure-driven valve to which a machine learning device or the like according to an embodiment of the present invention is applied.

Fig. 2 is a schematic diagram showing an example of a driving device such as a machine learning device to which an embodiment of the present invention is applied.

Fig. 3 is a schematic diagram showing an example of a solenoid valve to which a machine learning device or the like according to an embodiment of the present invention is applied.

Fig. 4 is a block diagram showing an example of a solenoid valve to which a machine learning device or the like according to an embodiment of the present invention is applied.

Fig. 5 is a schematic block diagram of a machine learning apparatus according to an embodiment of the present invention.

Fig. 6 is a diagram showing an example of a configuration of data (supervised learning) used in a machine learning apparatus and the like according to an embodiment of the present invention.

Fig. 7 is a diagram showing an example of a configuration of data (unsupervised learning) used in a machine learning device or the like according to an embodiment of the present invention.

Fig. 8 is a diagram showing an example of a neural network model for supervised learning implemented in the machine learning apparatus according to the embodiment of the present invention.

Fig. 9 is a flowchart showing an example of a machine learning method of an embodiment of the present invention.

Fig. 10 is a schematic block diagram showing a data processing system according to an embodiment of the present invention.

Fig. 11 is a flowchart showing an example of a data processing procedure of the data processing system according to an embodiment of the present invention.

Detailed Description

Embodiments for carrying out the present invention will be described below with reference to the drawings. The following schematically shows a range necessary for description of the object of the present invention, mainly a range necessary for description of a corresponding part of the present invention, and a part for which description is omitted is based on a known technique.

Before describing the machine learning device, the data processing system, the inference device, and the machine learning method according to the embodiment of the present invention, a fluid pressure driving valve to which the machine learning device or the like is applied will be described below.

(fluid pressure actuated valve)

Fig. 1 is a schematic diagram showing an example of a fluid pressure-driven valve 10 according to an embodiment of the present invention. The fluid pressure driven valve 10 according to the present embodiment can be used, for example, as an emergency shutoff valve which is provided in a plant in a pipe 100 through which various gases, petroleum, and the like flow, and which shuts off the flow of the pipe 100 when the plant is brought into an emergency stop such as an abnormality. In addition, the position and use of the fluid pressure-driven valve 10 are not limited to the above example.

The fluid pressure driven valve 10 shown in fig. 1 includes: a main valve 11 disposed in the middle of the pipe 100; a fluid pressure type driving device 12 that drives a valve stem 13a connected to the main valve 11 in accordance with a fluid pressure of a driving fluid to open and close the main valve 11; and a solenoid valve 1 having a function of controlling supply and discharge of a driving fluid to and from the driving device 12.

The driving fluid used for the fluid pressure-driven valve 10 is air a compressed by an air gauge (hereinafter, simply referred to as "air"). The air a is supplied from the air supply source 14 to the solenoid valve 1 via the first air pipe 140, and then supplied to the drive device 12 via the second air pipe 141. Further, a communication cable 150 for transmitting and receiving various data between the external device 15 and the solenoid valve 1 and a power cable 160 for supplying power from the external power supply 16 to the solenoid valve 1 are connected to the fluid pressure driven valve 10. The driving fluid is not limited to the air a, and may be other gas or liquid (for example, oil).

The external device 15 is a device for transmitting and receiving various information to and from the fluid pressure driven valve 10, and is configured by an external storage unit such as a plant management computer (including a local server and a cloud server), a diagnostic computer or a USB memory used by an operator (maintenance inspector), or an external HDD. The external device 15 can also be connected to a machine learning device 200 described later, and can transmit various data constituting a data set for learning. The external device 15 includes a reporting means, such as a GUI (Graphical User Interface), for reporting to an operator or the like the occurrence of an abnormality when the fluid pressure driven valve 10 is abnormal, and the content thereof. It should be noted that wireless communication may be used for communication between the external device 15 and the solenoid valve 1.

The fluid pressure driven valve 10 of the present embodiment is driven by an airless closing method. Therefore, during steady operation, the main valve 11 is opened by supplying air a (supply air) from the air supply source 14 to the drive device 12 via the solenoid valve 1, and during emergency stop or test operation, the main valve 11 is closed by discharging air a (exhaust air) from the drive device 12 via the solenoid valve 1. In this case, the fluid pressure-driven valve 10 may be opened without air, and the main valve 11 may be closed by supplying air a to the driving device 12 and discharging the air a from the driving device 12.

The main valve 11 is a ball valve. The main valve 11 specifically includes a valve element 110 disposed in the middle of the pipe 100 and a spherical valve element 111 rotatably provided in the valve element 110. Further, a stem 13a is connected to an upper portion of the valve body 111. The valve body 111 rotates in the valve body 110 in response to the rotation of the valve stem 13a by 0 to 90 degrees, and switches between a fully open state (the state shown in fig. 1) and a fully closed state of the main valve 11. The valve used as the main valve 11 is not limited to a ball valve, and may be a butterfly valve or other two-position valve, for example.

The driving device 12 is a single-acting cylinder mechanism disposed between the main valve 11 and the solenoid valve 1. As a specific configuration of the drive device 12, there is provided: a cylindrical cylinder 120; a pair of

pistons

122A, 122B disposed in the cylinder 120 so as to be linearly movable in a reciprocating manner and connected to each other via a piston rod 121; a coil spring 123 provided on the first piston 122A side; an air supply/discharge port 124 formed on the second piston 122B side; and a transmission mechanism 125 provided at a portion where the main shaft 13b disposed so as to penetrate the cylinder 120 in the radial direction is orthogonal to the piston rod 121. The drive device 12 is not limited to the single-action type, and may be configured by another type such as a multi-action type.

The first piston 122A is biased by a coil spring 123 to operate the main valve 11 in the closing direction. The second piston 122B is pressed by air a (supply air) supplied from the air supply/discharge port 124 to operate the main valve 11 in the opening direction (against the biasing force of the coil spring 123). The transmission mechanism 125 is configured by a rack and pinion mechanism, a scotch yoke mechanism, a link mechanism, a cam mechanism, or the like, and converts the reciprocating linear motion of the piston rod 121 into a rotational motion and transmits the rotational motion to the main shaft 13b of the drive device 12.

Fig. 2 is a schematic diagram showing an example of the driving device 12 to which the machine learning device and the like according to the embodiment of the present invention are applied. Fig. 2 (a) shows an example in which the transmission mechanism 125 is a rack and pinion mechanism, and fig. 2 (b) shows an example in which the transmission mechanism 125 is a turn yoke mechanism.

For example, when the transmission mechanism 125 is a rack and pinion mechanism, the piston 122 and the piston rod 121 perform reciprocating linear motion by supplying air a to the cylinder 120 or discharging air a from the cylinder 120. Then, the rack 125a provided on the piston rod 121 makes a reciprocating linear motion. Then, the pinion 125b, which is in contact with and engaged with the rack 125a, rotates. Then, the main shaft 13b, which rotates similarly to the pinion 125b, performs a rotational motion.

In the case where the transmission mechanism 125 is a yoke mechanism, the piston 122 and the piston rod 121 perform reciprocating linear motion by supplying air a to the cylinder 120 or discharging air a from the cylinder 120. Then, the roller pin 125c moving in the same manner as the piston rod 121 makes a reciprocating linear motion. Subsequently, the yoke 125d assembled by being eccentrically contacted with respect to the roller pin 125c is rotated by 90 °. Then, the main shaft 13b rotated by 90 ° similarly to the yoke 125d performs a rotational motion.

The drive device 12 includes

stoppers

126A and 126B capable of changing positions for regulating the operation of the

pistons

122A and 122B, respectively. In the example shown in fig. 2 (a) and 2 (B),

stoppers

126A and 126B are formed by bolts provided on the axis of piston rod 121 in

housings

120A and 120B of cylinder 120, respectively. Various sensors can be attached to the

stoppers

126A and 126B. In the examples shown in fig. 2 (a) and 2 (B),

position sensors

49A and 49B are attached to the

stoppers

126A and 126B, respectively. The

position sensors

49A, 49B detect the position of the piston 122 or piston rod 121 relative to the

housings

120A, 120B of the cylinder 120, respectively. The

position sensors

49A and 49B are configured by an ultrasonic sensor, an infrared sensor, a hall sensor, a reading sensor, and the like.

As described above, in the drive device 12 according to the embodiment of the present invention, the

position sensors

49A and 49B are attached to the

stoppers

126A and 126B, whereby the position of the piston 122 or the piston rod 121 with respect to the

housings

120A and 120B of the cylinder 120 can be directly detected. Therefore, the position of the piston 122 or the piston rod 121 relative to the

housings

120A, 120B of the cylinder 120 can be accurately detected.

The stem 13a of the main valve 11, the main shaft 13b of the driving device 12, and the shaft 13c of the solenoid valve 1 are formed in a shaft shape so as to be rotatable, respectively. The main shaft 13b of the driving device 12 is disposed so as to penetrate the driving device 12. The valve stem 13a of the main valve 11 and the shaft 13c of the solenoid valve 1 are connected to the main shaft 13b of the driving device 12 via a coupling or the like on a straight line, and perform rotational motion in synchronization with driving.

The solenoid valve 1 has a function of controlling supply and discharge of the air a to and from the drive device 12, and is configured as a normally closed two-position (open when energized and closed when not energized) three-way solenoid valve, for example. The electromagnetic valve 1 includes a spool portion 2 that switches a flow path of air a and a solenoid portion 3 that displaces the spool portion 2 according to an energized state (energized state or non-energized state) in an accommodating portion 6 that functions as a housing of the indoor type or explosion-proof type electromagnetic valve 1. The solenoid valve 1 is not limited to the three-way solenoid valve of the type described above, and may be a three-position solenoid valve, a normally open solenoid valve, a four-way solenoid valve, or the like, and may be configured in various forms based on any combination thereof. In the present embodiment, the solenoid valve 1 is used as a pilot valve in the fluid pressure driven valve 10, but the use of the solenoid valve 1 is not limited to this.

The slide valve portion 2 includes an input port 20 connected to the air supply source 14 via a first air pipe 140, an output port 21 connected to the drive device 12 via a second air pipe 141, and an exhaust port 22 for discharging exhaust gas from the drive device 12.

The solenoid portion 3 displaces the spool portion 2 to communicate between the input port 20 and the output port 21 when energized, and displaces the spool portion 2 to communicate between the output port 21 and the exhaust port 22 when de-energized.

With the above-described series of configurations, when the solenoid valve 1 is in the energized state, the air a (supply air) from the air supply source 14 flows through the first air pipe 140, the input port 20, the output port 21, and the second air pipe 141 in this order, and is supplied to the air supply/discharge port 124, whereby the second piston 122B is pressed, and the coil spring 123 is compressed. When the stem 13a of the main valve 11 coupled to the main shaft 13b via the connector is rotated by an amount corresponding to the displacement of the piston rod 121 in response to the compression of the coil spring 123 via the piston rod 121 and the transmission mechanism 125, the valve body 110 rotates the valve body 111, and the main valve 11 is operated to the full open state.

On the other hand, when the solenoid valve 1 is in the non-energized state, the air a (exhaust air) in the cylinder 120 flows through the second air pipe 141, the output port 21, and the exhaust port 22 in this order from the air supply/exhaust port 124, and is exhausted to the outside air, whereby the pressing force of the second piston 122B is reduced, and the coil spring 123 is restored from the compressed state. When the stem 13a of the main valve 11 coupled to the main shaft 13b via the coupling is rotated by the transmission mechanism 125 by an amount corresponding to the movement of the piston rod 121 in response to the return of the coil spring 123, the valve body 110 rotates the valve body 111, and the main valve 11 is operated to the fully closed state.

Fig. 3 is a cross-sectional view showing an example of the solenoid valve 1 according to the embodiment of the present invention. As shown in fig. 3, the solenoid valve 1 of the present embodiment includes, in addition to the spool portion 2 and the solenoid portion 3: a plurality of sensors 4 that acquire the states of the respective portions of the solenoid valve 1; a substrate 5 on which at least one of the plurality of sensors 4 is mounted; and an accommodating portion 6 that accommodates the slide valve portion 2, the solenoid portion 3, the plurality of sensors 4, and the substrate 5.

The storage section 6 includes: a first receiving portion 60 that receives the slide valve portion 2; a second housing portion 61 adjacent to the first housing portion 60 and housing the solenoid portion 3, the plurality of sensors 4, and the substrate 5; and a junction box 62 connecting the communication cable 150 and the power cable 160. The first receiving portion 60 and the second receiving portion 61 are made of a metal material such as aluminum.

The first housing portion 60 has openings (not shown) that function as the input port 20, the output port 21, and the exhaust port 22, respectively.

The second storage portion 61 includes: a cylindrical housing 610 having both ends (a first housing end 610a and a second housing end 610 b) open; a main body 611 disposed inside the housing 610; a solenoid cover 612 that covers the solenoid portion 3 fixed to the first case end 610a to isolate the outside air; and a terminal box cover 613 for covering the terminal box 62 fixed to the second housing end 610b to isolate the external air.

The housing 610 has: a shaft insertion opening 610c formed at a lower portion of the housing 610, into which the shaft 13c is inserted; a body insertion opening 610d formed at an upper portion of the housing 610 for inserting the body 611; and a cable insertion port 610e formed on the second housing end 610b side into which the communication cable 150 and the power cable 160 are inserted.

The first receiving portion 60 and the second receiving portion 61 are formed so as to penetrate the main body 611: a first flow passage 63 that branches from the input-side flow passage 26 and communicates between the input-side flow passage 26 and the first pressure sensor 40; a second flow passage 64 which branches from the output side flow passage 27 and communicates between the output side flow passage 27 and the second pressure sensor 41; and a spool flow path 65 through which air a for interlocking the spool portion 2 and the solenoid portion 3 flows.

The slide valve portion 2 includes: a spool hole 23 formed in the second housing portion 61 functioning as a spool housing; a spool valve 24 movably disposed in the spool hole 23; a spool spring 25 that biases the spool 24; an input-side flow path 26 communicating between the input port 20 and the spool hole 23; an output-side flow path 27 communicating between the output port 21 and the spool hole 23; and an exhaust passage 28 communicating between the exhaust port 22 and the spool hole 23.

The solenoid portion 3 includes: a solenoid housing 30; a solenoid coil 31 housed in the solenoid case 30; a movable core 32 disposed movably in the solenoid coil 31; a fixed core 33 disposed in a fixed state in the solenoid coil 31; and a solenoid spring 34 that biases the movable core 32.

When the solenoid valve 1 is switched from the non-energized state to the energized state, the solenoid coil 31 generates an electromagnetic force by a coil current flowing through the solenoid coil 31 in the solenoid portion 3, and the movable core 32 is attracted to the fixed core 33 against the biasing force of the solenoid spring 34 by the electromagnetic force, thereby switching the flow state of the air a flowing through the spool flow path 65. In the spool valve portion 2, by switching the state of flow of the air a flowing through the spool valve flow path 65, the spool valve 24 moves against the biasing force of the spool valve spring 25, and the state is switched from the state between the communication input port 20 and the exhaust port 22 to the state between the communication input port 20 and the output port 21.

The substrate 5 includes: a first substrate 50 disposed with

substrate surfaces

500A and 500B along a shaft 13c inserted from a shaft insertion port 610 c; a second substrate 51 disposed close to the junction box 62; and a third substrate 52 disposed close to the solenoid portion 3.

The body 611, the solenoid portion 3, and the third substrate 52 are disposed on the first substrate surface 500A side of the substrate surfaces 500A and 500B of the first substrate 50. The second substrate 51 and the junction box 62 are disposed on the second substrate surface 500B opposite to the first substrate surface 500A.

The sensor 4 is disposed at an appropriate position on the first substrate 50, the second substrate 51, and the third substrate 52. The sensor 4 includes, for example: a first pressure sensor 40 that measures a fluid pressure of the air a flowing in the input-side flow path 26 and the first flow path 63; a second pressure sensor 41 that measures the fluid pressure of the air a flowing through the output-side flow path 27 and the second flow path 64; and a main valve opening sensor 42 that measures a rotation angle of the shaft 13c of the solenoid valve 1 when rotated by rotation of the valve stem 13a from the main valve 11 via the main shaft 13b of the drive device 12, and acquires valve opening information of the main valve 11 from the rotation angle.

The main valve opening sensor 42 is, for example, a magnetic sensor, measures the magnetic field intensity generated by the permanent magnet 131 attached to the shaft 13c, and acquires the valve opening information of the main valve 11 based on the magnetic field intensity. The main valve opening sensor 42 is preferably placed on the first substrate surface 500A of the first substrate 5, which is disposed along the shaft 13c inserted from the shaft insertion opening 610c, at a position facing the outer periphery of the shaft 13c around the shaft. Thus, the main valve opening sensor 42 and the shaft 13c mounted on the first substrate 50 can be disposed close to each other in the housing portion 6 without wasting the disposition space, and the valve opening information can be accurately acquired.

Fig. 4 is a block diagram showing an example of the solenoid valve 1 according to the embodiment of the present invention. As shown in fig. 4, the electrical configuration example of the solenoid valve 1 includes a control unit 7 for controlling the solenoid valve 1, a communication unit 8 for communicating with an external device 15, and a power supply circuit unit 9 connected to an external power supply 16, in addition to the substrate 3 and the sensor 4.

The plurality of sensors 4 are a sensor group for measuring physical quantities of each part, and include, in addition to the first pressure sensor 40, the second pressure sensor 41, and the main valve opening degree sensor 42: a voltage sensor 43 that measures a supply voltage to the solenoid portion 3; a current/resistance sensor 44 that measures a current value at the time of energization and a resistance value at the time of non-energization in the solenoid portion 3; a temperature sensor 45 that measures the internal temperature of the housing portion 6; a magnetic sensor 46 that measures the intensity of the magnetic field generated by the solenoid portion 3; and a position sensor 49 that measures the position of the piston 122 relative to the cylinder 120.

The plurality of sensors 4 are a sensor group for acquiring information on the operation history of each part, and include: an operation timer (timer) 47 that measures at least one of the current energization interlock time and the sum of energization times to the solenoid portions, which is the operation time of the solenoid portion 3; and an operation counter (counter) 48 that counts the number of operations of each of the solenoid valve 1, the drive device 12, and the main valve 11.

These sensors 40 to 49 are not limited to the individual sensors as described above, and may be provided independently, or may be provided as a function of a specific sensor as well as another sensor. For example, the current/resistance sensor 44 may not be separately provided by measuring the magnetic field intensity generated by the solenoid portion 3 by the magnetic sensor 46 and obtaining the current value at the time of energization to the solenoid portion 3 from the magnetic field intensity. The microcontroller 70 may have the sensor function or realize a part of the sensor function, and for example, the microcontroller 70 may have the running timer 47 and the operation counter 48 built therein, so that the running timer 47 and the operation counter 48 are not separately provided.

The control unit 7 includes: a microcontroller 70 that processes information indicating the states of the respective portions of the solenoid valve 1 acquired by the plurality of sensors 4 and controls the respective portions of the solenoid valve 1; and a valve test switch 71 that controls the energization state of the solenoid portion 3 and performs an opening/closing operation of the main valve 11 during a test operation.

The microcontroller 70 includes a processor (not shown) such as a CPU (Central Processing Unit) and a Memory including a ROM (Read Only Memory), a RAM (Random Access Memory) and the like. The microcontroller 70 can include a function to realize the data processing system 300 described later in this embodiment.

The valve test switch 71 is configured to receive a command from the microcontroller 70 when a predetermined test operation condition is satisfied, and to execute a stroke test of the fluid pressure driven valve 10 as a test operation.

The stroke test is performed by, for example, any one of a full stroke test and a partial stroke test. The full stroke test is performed as follows: the main valve 11 is switched from a current-carrying state to a non-current-carrying state in a fully open state to be operated in a fully closed state, and is switched from the non-current-carrying state to a current-carrying state in the fully closed state to be returned to the fully open state. The partial stroke test is performed as follows: the main valve 11 is not operated in the fully closed state (i.e., the plant is not stopped), and the main valve 11 is switched from the energized state to the non-energized state in the fully open state and partially closed to a predetermined opening degree, and is switched from the non-energized state to the energized state in the partially closed state and returned to the fully open state.

As the test operation conditions, for example, when an execution time based on an execution frequency (for example, 1 time per year) designated by a manager as a set value, a specific designated date and time come, an execution command is received from the external device 15, or a test execution button (not shown) provided in the solenoid valve 1 is operated by the manager, the test operation (stroke test) may be executed as if the test operation conditions are satisfied.

(machine learning device)

In the fluid pressure driven valve 10 having the above-described series of configurations, by providing the plurality of sensors 4, it is possible to acquire various information of the fluid pressure driven valve 10, for example, during a steady operation and during an unsteady operation (for example, during a test operation including an opening/closing operation or during an emergency stop). Therefore, the machine learning device 200 that learns an inference model (learned model) that can infer diagnostic information of the fluid pressure driven valve 10 based on information (state variables) that can be acquired from the fluid pressure driven valve 10 will be described below. The machine learning device 200 described here includes not only a machine learning device provided as a device that operates alone, but also a machine learning device provided in the form of a non-transitory computer-readable medium storing a program for causing an arbitrary processor to execute the operation described below or one or more instructions for causing an arbitrary processor to execute the operation.

Fig. 5 is a schematic block diagram of a machine learning apparatus 200 according to an embodiment of the present invention. As shown in fig. 5, the machine learning device 200 of the present embodiment includes a learning data set acquisition unit 201, a learning data set storage unit 202, a learning unit 203, and a learned model storage unit 204.

The learning data set acquisition unit 201 is an interface unit for acquiring a plurality of data constituting a learning (training) data set from various devices connected via, for example, a wired or wireless communication line. Here, examples of the various devices connected to the learning dataset acquisition means 201 include the external device 15 and an operator computer PC1 used by an operator of the fluid pressure driven valve 10. In fig. 5, an example is shown in which the learning dataset acquisition means 201 is connected to the external device 15 and the computer PC1, respectively, but the external device 15 and the operator computer PC1 may be configured by the same computer. In the learning data set acquisition unit 201, it is possible to acquire, as input data, detection data of the plurality of sensors 4 of the fluid pressure driving valve 10 from, for example, the external device 15, and acquire, as output data, diagnostic information of the fluid pressure driving valve 10 associated with the input data from, for example, the operator computer PC 1. These correlated input data and output data constitute a learning data set described later.

Fig. 6 is a diagram showing an example of a configuration of data (supervised learning) used in the machine learning device 200 according to the embodiment of the present invention. Fig. 7 is a diagram showing an example of a configuration of data (unsupervised learning) used in the machine learning device 200 according to the embodiment of the present invention. Fig. 6 and 7 are also referred to as appropriate in the description of the data processing system and the inference device.

The learning data set is a data set used for machine learning described later, and as shown in fig. 6 and 7, input data includes at least the valve opening degree of the main valve 11, the pressure of the air a, and the position of the piston 122 relative to the cylinder 120, and output data includes diagnostic information of the driving device 12. An example will be described below for details of these various data, but the present invention is not limited to this.

The valve opening degree of the main valve 11 is a value indicating an open/close state of the main valve 11, and can be acquired from the main valve opening degree sensor 42.

The pressure of the air a is preferably the pressure of the air a flowing through each part inside the fluid pressure driven valve 10, and more specifically, preferably includes the solenoid valve output-side pressure of the air a supplied from the solenoid valve 1 to the driving device 12. The solenoid valve output-side pressure of the air a is the pressure of the air a supplied from the solenoid valve 1 to the drive device 12, and includes the pressure of the air a (supply air) when the air a is supplied from the solenoid valve 1 to the drive device 12, and the pressure of the air a (discharge air) when the air a is discharged from the drive device 12 to the outside air via the solenoid valve 1. The solenoid valve output side pressure of the air a can be acquired by the second pressure sensor 41.

The position of the piston 122 relative to the cylinder 120 is a value at which the piston 122 moves due to the air a being supplied to and discharged from the cylinder 120, and can be acquired by the position sensor 49.

The valve opening degree of the main valve 11, the pressure of the air a, and the position of the piston 122 with respect to the cylinder 120, which constitute input data, may be constituted by one piece of data (time data) at a certain specific time, or may be constituted by a plurality of pieces of data (time series data) acquired at a plurality of different times within a predetermined period, as shown in parentheses in fig. 6 and 7. When each data is constituted by time series data, the time series data of the valve opening degree of the main valve 11, the time series data of the pressure of the air a, and the time series data of the position of the piston 122 with respect to the cylinder 120 may be data acquired at a plurality of times at the same sampling cycle and the same phase (in a state without a phase difference), or may be data in which at least one of the sampling cycle and the phase is different. In order to effectively improve the accuracy of the inference, the latter form of time series data is preferable.

The diagnostic information of the drive device 12 is information showing whether or not some abnormality has occurred in the drive device 12 when the abnormality diagnosis is performed on the drive device 12, and various forms can be adopted as the data form thereof. The abnormality includes not only a post-event abnormality in which the occurrence of an abnormality is determined at the time of abnormality diagnosis, but also an abnormality sign in which the occurrence of an abnormality is expected in the future when the abnormality diagnosis is performed within an allowable range determined to be normal.

For example, as shown in fig. 6, the diagnostic information as one embodiment may be configured by information indicating whether the drive device 12 is normal (no abnormality) or abnormal (abnormal). In this case, the diagnostic information may be classified into 2 values, for example, a value indicating that the drive device 12 is in the normal state may be set to "0", a value indicating that the drive device 12 is in the abnormal state may be set to "1", and the operator may input the corresponding value in a form correlated with the input data using the work computer PC 1. In this case, information on specific abnormal content is not necessarily required.

As indicated by the broken line in fig. 6, the information indicating the abnormality of the drive device 12 in the diagnostic information may include specific contents of the abnormality, and the specific contents of the abnormality may include, for example, an air a circuit failure, an abnormality in the supply pressure of the air a from the air supply source 14, a malfunction of the cylinder 120 and the piston 122, wear of the piston 122, deterioration and breakage of the piston rod 121 or the coil spring 123, clogging of the air supply/discharge port 124, a malfunction of the transmission mechanism 125, and the like. In this case, the diagnostic information may be classified into a plurality of values (3 or more), for example, a value indicating that the driving device 12 is in a normal state is "0", a value indicating that an abnormality of the driving device 12 belongs to an operation failure of the main valve 11 is "1", a value indicating that an abnormality of the driving device 12 belongs to an air a circuit failure is "2", and the values are uniquely specified in advance according to the contents of the respective abnormalities in the following manner, and then the operator may input the corresponding values in a form correlated with the input data using the working computer PC 1. By setting such diagnostic information, it is possible to prepare a learning dataset including not only the presence or absence of the occurrence of an abnormality but also information (corresponding to abnormality 1/abnormality 2/\8230;/abnormality n shown in fig. 7) including the specific contents of the abnormality at the time of the occurrence of the abnormality. The diagnostic information according to the above-described one embodiment is used in the case where supervised learning (see fig. 6) is performed in machine learning described below.

In addition, as the diagnosis information of the driving device 12, information other than the above may be adopted. For example, as shown in fig. 7, the diagnostic information of another mode may be information indicating that only the drive device 12 is normal, not abnormal. In this case, since only the information indicating that the drive device 12 is normal is included in the diagnostic information, the learning data set including the diagnostic information as the output data is only a data set including the input data and the output data when the drive device 12 is normal, as a matter of course. Therefore, since the output data of the data set for learning is always the same in this case, it is obvious to those skilled in the art that the data set for learning does not necessarily have the output data as the data. The diagnostic information of the other mode is used in the case of performing unsupervised learning (see fig. 7) in machine learning described below.

Alternatively, the input data in the learning data set may optionally include the total operation time of the fluid pressure driven valve 10, the operation time after the fluid pressure driven valve 10 is finally powered on, the number of operations of the main valve 11, the number of operations of the driving device 12, the number of operations of the solenoid portion 3, and the opening/closing time of the main valve 11. The total operation time of the fluid pressure driven valve 10 and the operation time after the fluid pressure driven valve 10 is finally powered on can be obtained by the operation timer 47, the number of times the main valve 11 is operated, the number of times the driving device 12 is operated, and the number of times the solenoid portion 3 is operated can be obtained by the operation counter 48, and the opening/closing time of the main valve 11 can be obtained by a timer, not shown, or the like. Increasing the types of input data substantially contributes to improving the estimation accuracy of the learned model obtained after machine learning, but using input data having a low degree of correlation with diagnostic information may adversely hinder improving the estimation accuracy of the learned model. Therefore, the number and type of data used as input data should be appropriately selected in consideration of the state of the fluid pressure driven valve 10 to which the learned model is applied, and the like.

In particular, the drive device 12 is subjected to a large torque at a contact portion between the cylinder 120 and the piston 122 whose relative position changes due to the high-pressure air, a contact portion of the transmission mechanism 125 that converts the reciprocating linear motion into the rotational motion, or the like, and may have an abnormality such as wear, deterioration, or breakage. Such an abnormality of the contact portion of the drive device 12 is assumed to affect a change in the drive characteristic of the fluid pressure driven valve 10. Therefore, in the present embodiment, it is preferable to obtain the pressure of the output side driving fluid supplied from the solenoid valve 1 to the driving device 12, the position of the piston 122 with respect to the cylinder 120, and the valve opening degree of the main shaft 13b, and perform diagnosis. That is, by following a series of flows in which the piston 122 moves relative to the cylinder 120 by the pressure of the output side driving fluid supplied from the solenoid valve 1 to the driving device 12 and the valve opening degree of the main valve 13 changes, the state of the driving device 12 can be grasped, and the sign of an abnormality occurring in the driving device 12 can be detected. For example, in the case where the transmission mechanism 125 is a rack and pinion mechanism, an abnormality in a contact portion between a rack and a pinion can be detected, and in the case of the yoke mechanism, an abnormality in a contact portion between a shaft and a yoke can be detected.

The learning data set storage unit 202 is a database for storing a plurality of data constituting the learning data set acquired by the learning data set acquisition unit 201 as one learning data set in association with input data and output data related thereto. The specific configuration of the database constituting the learning data set storage means can be appropriately adjusted. For example, in fig. 5, the learning data set storage section 202 and the learned model storage section 204 described later are shown as storage sections independent of each other for convenience of explanation, but they may be constituted by a single storage medium (database).

The learning unit 203 executes machine learning by using the plurality of learning data sets stored in the learning data set storage unit 202, thereby generating a learned model in which the correlation between the input data and the output data included in the plurality of learning data sets is learned. In the present embodiment, as will be described in detail later, a specific method of machine learning is supervised learning using a neural network. However, the specific method of machine learning is not limited to this, and other learning methods can be employed as long as the correlation between the input and the output can be learned from the learning data set. For example, ensemble learning (random forest, boosting algorithm, etc.) can also be used.

The learned model storage unit 204 is a database for storing the learned model generated by the learning unit 203. The learned model stored in the learned model storage unit 24 is applied to an actual system via a communication line including the internet and a storage medium according to a request. The specific application of the learned model to the actual system (data processing system 300) will be described in detail later.

Next, a learning method in the learning unit 203 using the plurality of data sets for learning obtained as described above will be described centering on supervised learning. Fig. 8 is a diagram showing an example of a neural network model for supervised learning implemented in the machine learning apparatus according to the embodiment of the present invention. The neural network in the neural network model shown in fig. 8 is composed of one neuron (x 1 to x 1) located in the input layer, m neurons (y 11 to y1 m) located in the first intermediate layer, n neurons (y 21 to y2 n) located in the second intermediate layer, and o neurons (z 1 to zo) located in the output layer. The first intermediate layer and the second intermediate layer are also referred to as hidden layers, and as the neural network, a plurality of hidden layers may be provided in addition to the first intermediate layer and the second intermediate layer, or only the first intermediate layer may be used as a hidden layer. In fig. 8, a neural network model in which a plurality of (o) output layers are set is illustrated, but for example, when the diagnostic information is determined by one value, that is, when the number of training data to be described later is 1 (only t 1), the number of neurons in the output layers may be 1 (only z 1).

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

The neural network in the neural network model of the present embodiment learns the correlation between the valve opening degree of the main valve 11, the pressure of the air a, and the position of the piston 122 relative to the cylinder 120 and the diagnostic information of the drive device 12 using the learning data set. Specifically, the value of the neuron element located in the output layer is calculated by associating the valve opening degree of the main valve 11, the pressure of the air a, and the position of the piston 122 with respect to the cylinder 120, which are state variables, with the neuron element in the input layer, and by using a method of calculating the output value of a general neural network, that is, a method of calculating the value of the neuron element located in the output layer as the sum of the number sequence of multiplication values of the value of the neuron element located in the input side connected to the neuron element located in the output side and a weight wi associated with a node connecting the neuron element located in the output side and the neuron element located in the input side with respect to all the neuron elements located in the input layer. The form of information to be input as the state variable when the state variable is input to the neuron of the input layer may be set as appropriate in consideration of the accuracy of the generated learned model, and the like. Specifically, in order to adjust the number of neurons corresponding to each input data or to adjust to a value that can correspond to a neuron, preprocessing can be performed on specific input data.

Then, the calculated values of o neurons z1 to zo located in the output layer, that is, in the present embodiment, 1 or more pieces of diagnostic information are compared with training data t1 to which 1 or more pieces of diagnostic information are also included, which form a part of the data set for learning, respectively, to obtain errors, and the weights wi (back propagation) associated with the respective nodes are repeatedly adjusted so that the obtained errors are reduced.

Then, when the series of steps repeated a predetermined number of times or a predetermined condition such as that the error is smaller than the allowable value is satisfied, the learning is terminated, and (all the weights wi corresponding to the respective nodes of) the neural network model is stored as a learned model in the learned model storage unit 204.

(machine learning method)

In connection with the above, the present invention provides a machine learning method. The machine learning method according to the present invention will be described below with reference to fig. 6 (learning phase), fig. 7 (learning phase), fig. 8, and fig. 9. Fig. 9 is a flowchart showing an example of a machine learning method of an embodiment of the present invention. The machine learning method described below is explained based on the machine learning device 200, but the configuration as a premise is not limited to the machine learning device 200. The machine learning method is realized by using a computer, but various computers can be applied to the computer, and examples thereof include a computer device constituting the external device 15, the work computer PC1, or the microcontroller 70, and a server device disposed on a network. The specific configuration of the computer can be configured to include at least an arithmetic device including a CPU, a GPU, and the like, a storage device including a volatile or nonvolatile memory and the like, a communication device for communicating with a network and other devices, and a bus connecting these devices.

Supervised learning, which is a machine learning method according to the present embodiment, is a preliminary preparation for starting machine learning, in which a desired number of learning datasets (see fig. 6) are first stored in the learning dataset storage unit 202, and a plurality of prepared learning datasets are stored (step S11). The number of learning data sets to be prepared here may be set in consideration of the inference accuracy required for the finally obtained learned model.

The method of preparing the learning data set used in supervised learning can employ several methods. For example, when an abnormality occurs in a specific drive device 12 or when an operator recognizes a sign of the abnormality, various information during the steady operation of the fluid pressure driven valve 10 at that time is acquired using the plurality of sensors 4 and the like, and the operator specifies and inputs diagnostic information in a form associated with the information using the work computer PC1 and the like, thereby preparing input data and output data (for example, the value of the output data in this case is "1") constituting a learning data set. Further, a method of preparing a desired number of data sets for learning by repeating such a job can be employed. As a method of preparing the data set for learning, various methods such as acquiring the data set for learning by actively creating an abnormal state in the drive device 12 can be employed in addition to such a method. However, since various information of the fluid pressure driven valve 10 tends to be peculiar to each fluid pressure driven valve 10, it is preferable to collect data constituting a learning data set from only one fluid pressure driven valve 10 as a target of acquiring data constituting the learning data set, and the fluid pressure driven valve 10 is intended to apply a learned model obtained through machine learning described later. The learning data set includes not only a data set including input/output data at the time of occurrence of an abnormality but also a learning data set including a predetermined amount of input data and output data (for example, the value of the output data at this time is "0") in a normal state of the drive device 12 at the time of non-occurrence of an abnormality.

When step S11 is completed, a neural network model before learning is next prepared in order to start learning by the learning unit 203 (S12). The neural network model before learning prepared here has, for example, the structure shown in fig. 8 as its structure, and the weight of each node is set to an initial value. Then, one learning data set is selected, for example, randomly from the plurality of learning data sets stored in the learning data set storage unit 202 (step S13), and input data in the one learning data set is input to the prepared input layer of the neural network model before learning (see fig. 8) (step S14).

Here, the value of the output layer (see fig. 8) generated as a result of the above step S14 is a value generated by the neural network model before learning, and therefore is a value different from the desired result, that is, a value showing information different from the correct diagnostic information in most cases. Therefore, machine learning is then performed using the diagnostic information as training data in one learning data set acquired in step S13 and the value of the output layer generated in step S13 (step S15). The machine learning to be performed here may be, for example, a process (back propagation) of comparing diagnostic information constituting the training data with the value of the output layer and adjusting the weight having a correspondence relationship with each node in the neural network model before learning to obtain a preferred output layer. The number and form of values output to the output layer of the neural network model before learning are the same as those of training data in the learning data set to be learned.

Specifically, the machine learning described here is exemplified, and it is assumed that the diagnostic information constituting the training data is composed of any value (binary classification) in which the normal state is "0" and the abnormal state is "1", and when the value of the output data in one of the learning data sets selected in step S13 is "1", the value of the output layer is a predetermined value of 0 to 1, specifically, a value of "0.63", for example, is output. Therefore, if the same input data is input to the input layer in step S15, the weights associated with the respective nodes of the neural network model under learning are adjusted so that the value obtained by the neural network model under learning approaches "1".

When the machine learning is carried out in step S15, it is determined whether or not further continuation of the machine learning is necessary, for example, from the remaining number of the unlearned learning data sets stored in the learning data set storage unit 202 (step S16). When the machine learning is continued (no in step S16), the process returns to step S13, and when the machine learning is ended (yes in step S16), the process proceeds to step S17. When the machine learning is continued, the steps S13 to S15 are performed on the neural network model under learning using the learning data set that is not learned. The accuracy of the finally generated learned model is generally improved in proportion to the number of times.

When machine learning is finished (yes in step S16), a neural network generated by adjusting the weights associated with the respective nodes through a series of steps is stored as a learned model in learned model storage section 204 (step S17), and a series of learning processes is finished. The learned model stored here can be applied to and used by the data processing system 300 described later.

In the learning process and the machine learning method of the machine learning apparatus described above, in order to generate one learned model, the accuracy is improved by repeatedly performing the machine learning process a plurality of times on one (pre-learning) neural network model, and a learned model sufficient for application to the data processing system 300 is obtained. However, the present invention is not limited to this acquisition method. For example, a plurality of learned models subjected to machine learning a predetermined number of times may be stored in advance as one candidate in the learned model storage unit 204, a data set for validity determination may be input to the plurality of learned model groups, an output layer (neuron value) may be generated, and one optimal learned model suitable for the data processing system 300 may be selected by comparing and studying the accuracy of the values specified in the output layer. The validity determination data set may be the same data set as the learning data set used for learning, and may not be used for learning.

As described above, by applying the machine learning apparatus and the machine learning method according to the present embodiment, the following learned model can be obtained: from various data acquired by the plurality of sensors 4 provided at an appropriate portion of the fluid pressure driven valve 10, diagnostic information indicating whether or not an abnormality (including a posterior abnormality and a sign of the abnormality) has occurred can be accurately derived.

In the learning method and the machine learning method of the machine learning device 200 described above, "supervised learning" is explained. However, as a method of generating the learned model, other known "supervised learning" methods such as a Convolutional Neural Network (CNN) may be used, and "unsupervised learning" using a data set for learning including the other forms of diagnostic information described above, that is, information indicating only that the fluid pressure driven valve 10 is not abnormal but normal as diagnostic information constituting output data (see fig. 7) may also be used. By using the "unsupervised learning", even when only information on the normal state of the drive device 12 can be obtained with respect to the diagnostic information in the output data associated with the input data, the learned model can be obtained by learning the correlation indicating the characteristics of the normal state of the input data and the output data as shown in the "learning stage" of fig. 7. In this case, when inference is performed in the data processing system 300 described later, it is possible to infer the diagnostic information by regarding input data that is determined to be not in conformity with a predetermined amount of the characteristic of the normal state as not being in the normal state, that is, as being in the abnormal state. As a specific method of the "unsupervised learning", for example, a known method using an automatic encoder or the like, which is schematically shown in fig. 7, can be used, and a detailed description thereof will be omitted.

(data processing System)

Next, an application example of the learned model generated by the machine learning device 200 and the machine learning method will be described with reference to fig. 10. FIG. 10 is a schematic block diagram illustrating a data processing system according to an embodiment of the present invention.

The data processing system 300 according to the present embodiment is mounted in the microcontroller 70 of the fluid pressure driven valve 10. It should be noted that, with regard to the data processing system 300, at least a part thereof may be applied to other apparatuses, for example, other apparatuses connected to the external device 15 and the fluid pressure-driven valve 10.

The data processing system 300 includes at least an input data acquisition unit 301, an inference unit 302, a learning model storage unit 303, and a reporting unit 304.

The input data acquisition means 301 is an interface means connected to the plurality of sensors 4 included in the fluid pressure driving valve 10 and configured to acquire data output from each sensor 4. The input data acquisition unit 301 acquires at least the valve opening degree of the main valve 11, the pressure of the air a, and the position of the piston 122 with respect to the cylinder 120. In the example shown in fig. 10, all the sensors 4 included in the fluid pressure driven valve 10 are connected so as to be able to acquire all the input data that can be used in the estimation described later, but which sensor 4 is connected to the input data acquisition means 301 can be appropriately selected in accordance with a learned model or the like used in the estimation means 302 described later. It is preferable that the inference result of the inference unit 302 is stored in a storage unit, not shown, and the stored past inference result can be used as a learning data set for on-line learning for further improving the inference accuracy of the learned model in the learned model storage unit 303, for example.

The inference unit 302 is configured to infer whether an abnormality occurs in the drive device 12 based on various data of the fluid pressure driven valve 10 acquired by the input data acquisition unit 301. For this inference, for example, a learned model learned by the machine learning device 200 and the machine learning method described above is used, and the learned model is stored in the learned model storage unit 303 made of an arbitrary storage medium. The inference unit 302 has not only a function of performing an inference process using a learned model, but also a preprocessing function of adjusting input data acquired by the input data acquisition unit 301 to a desired format or the like as a preprocessing of the inference process and inputting the input data to the learned model, and a post-processing function of finally determining the presence or absence of an abnormality (including a post-abnormality and a sign of an abnormality) (absence of an abnormality (normal) or presence of an abnormality (abnormality)) by applying, for example, a predetermined threshold value to an output value output from the learned model as a post-processing of the inference process.

As described above, the learned model storage unit 303 is a storage medium for storing the learned model used in the inference unit 302. The number of learned models stored in the learned model storage unit 303 is not limited to one. For example, a plurality of learning-done models having different amounts of input data or different learning methods (for example, supervised learning and unsupervised learning performed by the machine learning device 200 described above) can be stored, and these learning-done models can be selectively used.

The reporting unit 304 is used to report the inference result of the inference unit 302 to an operator or the like. A specific reporting method can be various methods, and for example, the presence or absence of an abnormality can be reported to an operator or the like by transmitting the inference result to the external device 15 via the communication unit 8, displaying it on a GUI of the external device 15, or by providing a light-emitting member, a speaker, or the like in advance in the fluid pressure driving valve 10, and operating them.

Next, a data processing procedure of the data processing system having the above configuration will be described with reference to fig. 6 (inference phase), fig. 7 (inference phase), and fig. 11. Fig. 11 is a flowchart showing an example of a data processing procedure of the data processing system 300 according to an embodiment of the present invention.

When power supply from the external power supply 16 to the solenoid valve 1 of the fluid pressure driven valve 10 is started and the abnormality diagnosis of the drive device 12 is started accordingly, the input data acquisition unit 301 acquires various data indicating the state of each part of the fluid pressure driven valve 10 acquired by the plurality of sensors 4 (step S21). When the input data acquisition means 301 obtains desired input data (the valve opening degree of the main valve 11, the pressure of the air a, and the position of the piston 122 with respect to the cylinder 120 (see fig. 6 and 7)), the inference means 302 performs inference based on the input data (step S22). In this case, it is preferable to determine a learned model for inference in advance. In addition, when the learned model determined in advance requires predetermined time series data as input data thereof, for example, the necessary amount of data is acquired by the input data acquisition unit 301, and thereafter the inference in step S22 is performed.

Specifically, the inference unit 302 applies preprocessing to input data and inputs the data to a learning model completion model, and applies post-processing to an output value from the learning model completion model, thereby determining the presence or absence of an abnormality (including a posterior abnormality and an abnormality sign) as an inference result. In the post-processing of supervised learning (see "inference phase" in fig. 6), the inference unit 302 compares the output value of the learning model completion model (a value between 0 and 1 if binary classification) with a predetermined threshold value, for example, determines that there is "abnormality (anomaly)" if the output value of the learning model completion model is equal to or greater than the predetermined threshold value, and determines that there is "no abnormality (normal)" if it is smaller than the predetermined threshold value, and outputs the determination result as an inference result. In the post-processing of unsupervised learning (see "inference stage" in fig. 7), the inference unit 302 calculates a difference (distance) between an output value (feature amount) of the learning model completion model and a feature amount based on the input data, determines "abnormal (abnormal)" if the difference (distance) is equal to or greater than a predetermined threshold, and determines "no abnormal (normal)" if the difference (distance) is smaller than the predetermined threshold, and outputs the determination result as an inference result.

Then, in step S22, the inference unit 302 performs inference, and when the inference result indicates "no abnormality (normal)" (no in step S23), the flow returns to step S21, and a series of inference is continued. On the other hand, as shown in fig. 6 and 7, when the inference result indicates "abnormality (anomaly)" is present (yes in step S23), the reporting unit 304 reports "abnormality (anomaly)" as the inference result, that is, an anomaly (including a post-event anomaly and an anomaly precursor) occurs in the drive device 12 to the operator or the like (step S24). After the occurrence of an abnormality is reported in step S24, the process returns to step S21 to continue a series of inferences. Note that the handling of stopping the fluid pressure driven valve 10 at the stage of abnormality detection may be performed depending on the usage of the driving device 12 and the content of the detected abnormality.

(inference device)

The present invention can be provided not only in the form of the above-described data processing system 300 but also in the form of an inference device for implementing inference. In this case, the inference means comprises a memory and at least one processor, wherein the processor is capable of executing a series of processes. The series of processes includes: a process of acquiring input data including a valve opening degree of a main valve, a pressure of a driving fluid, and a position of the piston 122 with respect to the cylinder 120; and a process of inferring diagnostic information in the fluid pressure driven valve 10 when the input data is input. By providing the present invention as the above-described inference device, it is possible to easily apply the present invention to various fluid pressure-driven valves 10, as compared with the case where the data processing system 300 is installed. The person skilled in the art will certainly understand that: in this case, when the inference device performs the process of inferring the diagnostic information, the inference method described above in this specification, which is performed by the inference unit 302 of the data processing system using the learned model learned by the machine learning device and the machine learning method of the present invention, may be used.

The present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the scope of the present invention. And these are included in the technical idea of the present invention.

Description of the reference numerals

1: an electromagnetic valve; 3: a solenoid portion; 4: a sensor; 10: a fluid pressure actuated valve; 11: a main valve; 12: a (fluid pressure type) driving device; 13a: a valve stem; 13b: a main shaft; 13c: a shaft; 14: an air supply source; 15: an external device; 26: an input-side flow path; 27: an output-side flow path; 28: an exhaust gas flow path; 30: a solenoid housing; 31: a solenoid coil; 32: a movable iron core; 40: a first pressure sensor; 41: a second pressure sensor; 42: a main valve opening sensor; 43: a voltage sensor; 44: a current/resistance sensor; 45: a temperature sensor; 46: a magnetic sensor; 47: running timers (timers); 48: an action counter (counter); 70: a microcontroller; 100: piping; 200: a machine learning device; 201: a learning data set acquisition unit; 202: a learning data set storage unit; 203: a learning unit; 204: a learning completion model storage unit; 300: a data processing system; 301: an input data acquisition unit; 302: an inference unit; 303: a learning completion model storage unit; 304: a reporting unit; a: air (driving fluid); PC1: a work computer.

Claims

1. A machine learning device applied to a fluid pressure-driven valve including at least a main valve, a driving device including a cylinder and a piston for driving the main valve, and a solenoid valve for controlling supply and discharge of a driving fluid to and from the driving device, the machine learning device comprising:

a learning data set storage unit that stores a plurality of sets of learning data sets including input data and output data, the input data including a valve opening degree of the main valve, a pressure of an output-side drive fluid supplied from the solenoid valve to the drive device, and a position of the piston relative to the cylinder, the output data including diagnostic information of the drive device in correspondence 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 a plurality of sets of the data set for learning; and

a learned model storage unit that stores the learning model learned by the learning unit.

2. The machine learning apparatus of claim 1,

the diagnostic information is information indicating any one of normality and abnormality of the drive device.

3. The machine learning apparatus of claim 1,

the diagnostic information is information indicating that only the drive device is not abnormal but normal.

4. The machine learning device of any one of claims 1-3,

the driving device is provided with a transmission mechanism for converting the linear motion of the piston into the rotary motion of the main valve,

the diagnostic information is information for the delivery mechanism.

5. The machine learning apparatus of claim 4,

the transmission mechanism is a rack and pinion mechanism having a rack that moves linearly together with the piston and a pinion that is connected to the rack and rotates together with the main valve, and the diagnostic information is information for the rack and pinion mechanism.

6. The machine learning apparatus of claim 4,

the transmission mechanism is a rotation stop rod yoke mechanism having a shaft that moves linearly together with the piston and a yoke that is connected to the shaft and rotates together with the main valve, and the diagnostic information is information for the rotation stop rod yoke mechanism.

7. The machine learning apparatus of any one of claims 4 to 6,

the diagnostic information is information related to wear of a connecting portion that converts linear motion into rotational motion in the transmission mechanism.

8. A data processing system for a fluid pressure driven valve including at least a main valve, a driving device including a cylinder and a piston that drives the main valve, and an electromagnetic valve including a solenoid portion that controls supply and discharge of a driving fluid to and from the driving device, the data processing system comprising:

an input data acquisition unit that acquires input data including a valve opening degree of the main valve, a pressure of an output side driving fluid supplied from the solenoid valve to the driving device, and a position of the piston with respect to the cylinder; and

an inference unit that inputs the input data acquired by the input data acquisition unit to a learning-completed model generated by the machine learning device according to any one of claims 1 to 8, and infers diagnostic information of the drive device.

9. An inference device for a fluid pressure-driven valve including at least a main valve, a drive device including a cylinder and a piston that drives the main valve, and an electromagnetic valve including a solenoid portion that controls supply and discharge of a drive fluid to and from the drive device,

the inference apparatus is provided with a memory and at least one processor,

the at least one processor is configured to perform the following:

acquiring input data including a valve opening degree of the main valve, a pressure of an output side driving fluid supplied from the solenoid valve to the driving device, and a position of the piston with respect to the cylinder; and

when the input data is input, the diagnostic information of the drive device is inferred.

10. A machine learning method using a computer applied to a fluid pressure driven valve including at least a main valve, a driving device including a cylinder and a piston that drives the main valve, and a solenoid valve including a solenoid portion that controls supply and discharge of a driving fluid to and from the driving device, the machine learning method comprising:

storing a plurality of sets of learning data including input data including a valve opening degree of the main valve, a pressure of an output side driving fluid supplied from the solenoid valve to the driving device, and a position of the piston relative to the cylinder, and output data including diagnostic information of the driving device in correspondence with the input data;

learning a learning model for inferring a correlation between the input data and the output data by inputting a plurality of sets of the data set for learning;

storing the learned learning model.