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

Abstract

The present invention provides a machine learning device that can accurately detect abnormalities and signs of abnormalities in a fluid pressure-driven valve without relying on the operator's experience. A machine learning device (200) is applied to a fluid pressure driven valve (10) including a main valve (11), a driving device (12) including a cylinder and a piston for driving the main valve (11), and a solenoid valve (1) for controlling the supply and discharge of a driving fluid (A) relative to the driving device (12). The machine learning device (200) comprises: a learning data set storage unit (202) for storing a plurality of learning data sets consisting of input data and output data, wherein the input data includes the valve opening of the main valve (11), the pressure of the output-side driving fluid supplied from the solenoid valve (1) to the driving device (12), and the position of the piston (122) relative to the cylinder, and the output data consists of diagnostic information of the driving device (12) having a corresponding relationship with the input data; a learning unit (203) for learning a learning model for inferring the correlation between the input data and the output data by inputting the plurality of learning data sets; and a learned model storage unit (204) for storing the learned model.

Description

Technical Field

[0001] The present invention relates to a machine learning device, a data processing system, an inference device, and a machine learning method for performing abnormality diagnosis of a valve system. Background Art

[0002] Fluid pressure-driven valves that use solenoid valves to control the driving fluid to open and close a main valve are known. For example, Patent Document 1 discloses an emergency shutoff valve device used in piping systems for plant equipment. In the event of an emergency, such as an equipment failure, the device uses a solenoid valve to control the driving fluid to close a ball valve (main valve), thereby shutting off the fluid flowing through the piping.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Application Laid-Open No. 2009-97539 Summary of the Invention

[0006] Problems to be solved by the invention

[0007] In order to improve the overall operational efficiency and reliability of the plant, fluid pressure-driven valves such as the emergency shutoff valve described in Patent Document 1, it is desirable to prevent unexpected anomalies from occurring. Therefore, it is desirable to implement not only post-announcement maintenance to identify anomalies upon their occurrence, but also predictive maintenance to identify signs of anomalies.

[0008] Fluid pressure-driven valve abnormality symptoms can manifest in a variety of ways, but the causal relationship between these symptoms and the symptoms has not been clearly established. Consequently, predictive maintenance for fluid pressure-driven valves relies on judgment based on the operator's experience (including tacit knowledge), resulting in variations in accuracy depending on the operator.

[0009] The present invention is completed in view of the above-mentioned problems, and its purpose is to provide a machine learning device, data processing system, inference device and machine learning method for grasping anomalies and signs of anomalies in fluid pressure-driven valves with high precision (hereinafter collectively referred to as "anomalies" in the present invention).

[0010] Solutions for solving problems

[0011] In order to achieve the above-mentioned object, a machine learning device according to a first embodiment of the present invention is, for example, Figure 1-5, a fluid pressure driven valve 10 is applied to 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 valve 1 having a solenoid portion 3 for controlling the supply and discharge of a driving fluid A for the driving device 12, the machine learning device comprises: a learning data set storage unit 202, which stores a plurality of sets of learning data sets consisting of input data and output data, the input data including the valve opening of the main valve 11, the pressure of the output side driving fluid supplied from the solenoid valve 11 to the driving device 12, and the position of the piston 122 relative to the cylinder 120, and the output data consisting of diagnostic information of the driving device 12 that has established a corresponding relationship with the input data; a learning unit 203, which learns a learning model for inferring the correlation between the input data and the output data by inputting a plurality of sets of the learning data sets; and a learned model storage unit 204, which stores the learning model learned by the learning unit 203.

[0012] Effects of the Invention

[0013] The machine learning device of the present invention provides a learned model that can infer the presence of a drive system anomaly based on various information available during stable operation of a fluid pressure-driven valve. Therefore, by utilizing this learned model, anomalies occurring in the drive system can be inferred with high accuracy, independent of operator experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This 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.

[0015] Figure 2 This is a schematic diagram showing an example of a driving device such as a machine learning device to which one embodiment of the present invention is applied.

[0016] Figure 3 This 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.

[0017] Figure 4 This 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.

[0018] Figure 5 This is a schematic block diagram of a machine learning device according to one embodiment of the present invention.

[0019] Figure 6 This is a diagram showing a data configuration example (supervised learning) used in a machine learning device or the like according to one embodiment of the present invention.

[0020] Figure 7 This is a diagram showing a data configuration example (unsupervised learning) used in a machine learning device or the like according to one embodiment of the present invention.

[0021] Figure 8 This is a diagram showing an example of a neural network model for supervised learning implemented in a machine learning device according to one embodiment of the present invention.

[0022] Figure 9 is a flowchart illustrating an example of a machine learning method according to one embodiment of the present invention.

[0023] Figure 10 This is a schematic block diagram showing a data processing system according to one embodiment of the present invention.

[0024] Figure 11 1 is a flowchart illustrating an example of a data processing procedure of a data processing system according to one embodiment of the present invention. DETAILED DESCRIPTION

[0025] Hereinafter, various embodiments for implementing the present invention will be described with reference to the accompanying drawings. It should be noted that the following schematically illustrates the scope required for the description to achieve the purpose of the present invention, mainly describing the scope required for the description of the corresponding parts of the present invention, and the omitted parts are based on known technologies.

[0026] Before describing a machine learning device, a data processing system, an inference device, and a machine learning method according to an embodiment of the present invention, a fluid pressure driven valve to which the machine learning device and the like are applied will be described below.

[0027] (Fluid pressure actuated valve)

[0028] Figure 1 This is a schematic diagram illustrating an example of a fluid pressure-driven valve 10 according to one embodiment of the present invention. The fluid pressure-driven valve 10 in this embodiment can be used, for example, as an emergency shutoff valve. This valve is installed in a pipe 100 carrying various gases, oil, or other gases within a plant, and is used to shut off the flow of the pipe 100 in the event of an emergency stop, such as when the plant experiences an abnormality. The installation location and application of the fluid pressure-driven valve 100 are not limited to the examples described above.

[0029] Figure 1 The fluid pressure driven valve 10 shown comprises: a main valve 11, which is arranged in the middle of the piping 100; a fluid pressure driving device 12, which drives the valve stem 13a connected to the main valve 11 according to the fluid pressure of the driving fluid, thereby performing the opening and closing operation of the main valve 11; and an electromagnetic valve 1, which has the function of controlling the supply and discharge of the driving fluid to the driving device 12.

[0030] The driving fluid used in the fluid pressure-driven valve 10 is pneumatically compressed air (hereinafter referred to as "air") A. The air A is supplied from the air supply source 14 to the solenoid valve 1 via a first air pipe 140, and then supplied to the drive device 12 via a second air pipe 141. In addition, the fluid pressure-driven valve 10 is connected to a communication cable 150 for transmitting and receiving various data between an external device 15 and the solenoid valve 1, and a power cable 160 for supplying power to the solenoid valve 1 from an external power supply 16. It should be noted that the driving fluid is not limited to the above-mentioned air A, and can also be other gases or liquids (such as oil).

[0031] The external device 15 is a device for sending and receiving various types of information to and from the fluid pressure-driven valve 10. It can be composed of, for example, a computer for plant management (including a local server and a cloud server), a diagnostic computer used by an operator (maintenance and repair person), or an external storage unit such as a USB memory or an external HDD. The external device 15 can also be connected to the machine learning device 200 described later to send various data that constitute a learning data set. In addition, the external device 15 has a reporting unit composed of a GUI (Graphical User Interface) and the like. This reporting unit is used to report the occurrence of an abnormality and its content to the operator and the like when an abnormality occurs in the fluid pressure-driven valve 10. It should be noted that wireless communication can also be used for communication between the external device 15 and the solenoid valve 1.

[0032] The fluid pressure-driven valve 10 of this embodiment is driven by an airless closing method. Therefore, during stable operation, the main valve 11 is opened by supplying air A from the air supply source 14 via the solenoid valve 1 to the drive device 12 (air supply). During emergency stops or test runs, the main valve 11 is closed by exhausting air A from the drive device 12 via the solenoid valve 1 (exhaust). It should be noted that the fluid pressure-driven valve 10 can also adopt an airless opening method. In this case, the main valve 11 is closed by supplying air A to the drive device 12 and by exhausting air A from the drive device 12.

[0033] The main valve 11 is a ball valve. The specific structure of the main valve 11 comprises a valve body 110 arranged in the middle of the pipe 100 and a spherical valve core 111 rotatably arranged in the valve body 110. In addition, a valve stem 13a is connected to the upper part of the valve core 111. The valve core 111 rotates in the valve body 110 in response to the rotation of the valve stem 13a from 0 to 90 degrees. In the fully open state ( Figure 1 It should be noted that the valve used as the main valve 11 is not limited to a ball valve, and may be, for example, a butterfly valve or other two-position valve.

[0034] The drive device 12 utilizes a single-acting cylinder mechanism positioned between the main valve 11 and the solenoid valve 1. Specifically, the drive device 12 comprises a cylindrical cylinder 120; a pair of pistons 122A and 122B reciprocatingly arranged within the cylinder 120 and connected via a piston rod 121; a coil spring 123 positioned on the side of the first piston 122A; an air supply and exhaust port 124 formed on the side of the second piston 122B; and a transmission mechanism 125 positioned at a point perpendicular to the main shaft 13b, which radially extends through the cylinder 120, and the piston rod 121. It should be noted that the drive device 12 is not limited to a single-acting mechanism and can also be configured in other configurations, such as a multiple-acting mechanism.

[0035] The first piston 122A is biased by a coil spring 123, closing the main valve 11. Furthermore, the second piston 122B is pushed by air A (air supply) supplied from an air supply and exhaust port 124, opening the main valve 11 (overcoming the force of the coil spring 123). Furthermore, the transmission mechanism 125, comprised of a rack and pinion mechanism, a scotch yoke mechanism, a connecting rod mechanism, and a cam mechanism, converts the reciprocating linear motion of the piston rod 121 into rotational motion and transmits it to the main shaft 13b of the drive unit 12.

[0036] Figure 2 This is a schematic diagram showing an example of a driving device 12 to which a machine learning device or the like according to an embodiment of the present invention is applied. Figure 2 (a) shows an example in which the transmission mechanism 125 is a rack and pinion mechanism. Figure 2 (b) shows an example in which the transmission mechanism 125 is a Scotch yoke mechanism.

[0037] For example, if the transmission mechanism 125 is a rack-and-pinion mechanism, the supply or discharge of air A to or from the cylinder 120 causes the piston 122 and piston rod 121 to reciprocate linearly. This causes the rack 125a attached to the piston rod 121 to reciprocate linearly. Subsequently, the pinion 125b, which contacts and meshes with the rack 125a, rotates. This causes the main shaft 13b, which rotates in the same manner as the pinion 125b, to rotate.

[0038] If the transmission mechanism 125 is a Scotch yoke mechanism, the supply or discharge of air A to or from the cylinder 120 causes the piston 122 and piston rod 121 to reciprocate linearly. This causes the roller pin 125c, which moves in concert with the piston rod 121, to reciprocate linearly. Next, the yoke 125d, which is assembled to eccentrically contact the roller pin 125c, rotates 90°. This causes the main shaft 13b, which rotates 90° in concert with the yoke 125d, to rotate.

[0039] The driving device 12 includes stoppers 126A and 126B that can change the positions of the pistons 122A and 122B. Figure 2 (a) and Figure 2 In the example shown in (b), the stoppers 126A and 126B are formed by bolts provided on the shaft of the piston rod 121 of the housings 120A and 120B of the cylinder 120. Various sensors can be mounted on the stoppers 126A and 126B. Figure 2 (a) and Figure 2 In the example shown in (b), position sensors 49A and 49B are attached to stoppers 126A and 126B, respectively. Position sensors 49A and 49B detect the position of piston 122 or piston rod 121 relative to housings 120A and 120B of cylinder 120, respectively. Position sensors 49A and 49B are composed of ultrasonic sensors, infrared sensors, Hall sensors, readout sensors, and the like.

[0040] As described above, the drive device 12 according to one embodiment of the present invention can directly detect the position of the piston 122 or piston rod 121 relative to the housings 120A and 120B of the cylinder 120 by attaching position sensors 49A and 49B to the limiters 126A and 126B. Therefore, the position of the piston 122 or piston rod 121 relative to the housings 120A and 120B of the cylinder 120 can be accurately detected.

[0041] The valve stem 13a of the main valve 11, the main shaft 13b of the drive device 12, and the shaft 13c of the solenoid valve 1 are each rotatably formed into a shaft. The main shaft 13b of the drive device 12 is disposed so as to penetrate the drive 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 drive device 12 in a straight line via a coupling or the like, and rotate synchronously with the drive.

[0042] The solenoid valve 1 has the function of controlling the supply and exhaust of air A with respect to the drive device 12, and is configured, for example, as a normally closed two-position ("open" when energized, "closed" when not energized) three-way solenoid valve. The solenoid valve 1 is provided with a slide valve portion 2 for switching the flow path of the air A and a solenoid portion 3 for displacing the slide valve portion 2 according to the energized state (energized or not energized) inside a housing portion 6 that functions as a shell of the indoor or explosion-proof solenoid valve 1. It should be noted that the solenoid valve 1 is not limited to the three-way solenoid valve of the above type, but may also be a three-position solenoid valve, a normally open solenoid valve, a four-way solenoid valve, etc., and may be configured in various forms based on any combination thereof. In addition, 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.

[0043] 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 .

[0044] The solenoid portion 3 displaces the spool portion 2 when energized to allow communication between the input port 20 and the output port 21 , and displaces the spool portion 2 when de-energized to allow communication between the output port 21 and the exhaust port 22 .

[0045] Through the above-described series of structures, when the solenoid valve 1 is energized, air A (supply air) from the air supply source 14 flows sequentially through the first air pipe 140, the input port 20, the output port 21, and the second air pipe 141, and is supplied to the air supply and exhaust port 124. This pushes the second piston 122B, compressing the coil spring 123. Furthermore, when the valve stem 13a of the main valve 11, which is connected to the main shaft 13b via a connector, rotates via the piston rod 121 and the transmission mechanism 125 by an amount corresponding to the compression of the coil spring 123, the valve element 111 rotates within the valve body 110, and the main valve 11 is fully opened.

[0046] On the other hand, when the solenoid valve 1 is de-energized, air A (exhaust) within the cylinder 120 flows from the air supply and exhaust port 124, sequentially through the second air pipe 141, the output port 21, and the exhaust port 22, and is discharged to the outside air. This reduces the pressing force of the second piston 122B, causing the coil spring 123 to return from its compressed state. Furthermore, when the valve stem 13a of the main valve 11, which is connected to the main shaft 13b via a connector via the transmission mechanism 125, rotates by the amount corresponding to the movement of the piston rod 121 in response to the return of the coil spring 123, the valve element 111 rotates within the valve body 110, and the main valve 11 is operated to a fully closed state.

[0047] Figure 3 1 is a cross-sectional view showing an example of a solenoid valve 1 according to an embodiment of the present invention. Figure 3 As shown, the solenoid valve 1 of this embodiment is provided with, in addition to the above-mentioned slide valve portion 2 and solenoid portion 3, a plurality of sensors 4 for acquiring the status of various parts of the solenoid valve 1; a substrate 5 on which at least one of the plurality of sensors 4 is carried; and a housing portion 6 for housing the slide valve portion 2, the solenoid portion 3, the plurality of sensors 4 and the substrate 5.

[0048] The housing portion 6 includes a first housing portion 60 that houses the spool portion 2; a second housing portion 61 that is adjacent to the first housing portion 60 and houses the solenoid portion 3, the plurality of sensors 4, and the substrate 5; and a junction box 62 that connects the communication cable 150 and the power cable 160. The first housing portion 60 and the second housing portion 61 are made of a metal material such as aluminum.

[0049] 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.

[0050] The second housing portion 61 includes: a cylindrical shell 610, both ends of which (a first shell end 610a and a second shell end 610b) are open; a main body 611, which is arranged inside the shell 610; a solenoid cover 612, which covers the solenoid portion 3 fixed to the first shell end 610a to isolate the outside air; and a junction box cover 613, which covers the junction box 62 fixed to the second shell end 610b to isolate the outside air.

[0051] The housing 610 has: an axis insertion port 610c, which is formed at the lower part of the housing 610 for inserting the axis 13c; a main body insertion port 610d, which is formed at the upper part of the housing 610 for inserting the main body 611; and a cable insertion port 610e, which is formed on the side of the second housing end 610b for inserting the communication cable 150 and the power cable 160.

[0052] In the first housing portion 60 and the second housing portion 61, there are formed in a manner that penetrates the main body 611: a first flow path 63, which branches from the input side flow path 26 and connects the input side flow path 26 and the first pressure sensor 40; a second flow path 64, which branches from the output side flow path 27 and connects the output side flow path 27 and the second pressure sensor 41; and a slide valve flow path 65, which is used to flow air A for linking the slide valve portion 2 and the solenoid portion 3.

[0053] The spool valve portion 2 includes a spool valve hole 23 formed in a second housing portion 61 functioning as a spool valve housing; a spool valve 24 movably disposed in the spool valve hole 23; a spool valve spring 25 biasing the spool valve 24; an input-side flow path 26 communicating between the input port 20 and the spool valve hole 23; an output-side flow path 27 communicating between the output port 21 and the spool valve hole 23; and an exhaust flow path 28 communicating between the exhaust port 22 and the spool valve hole 23.

[0054] The solenoid portion 3 includes a solenoid housing 30 , a solenoid coil 31 housed in the solenoid housing 30 , a movable iron core 32 movably disposed in the solenoid coil 31 , a fixed iron core 33 fixedly disposed in the solenoid coil 31 , and a solenoid spring 34 for urging the movable iron core 32 .

[0055] When the solenoid valve 1 switches from a de-energized state to an energized state, a coil current flows through the solenoid coil 31 in the solenoid unit 3, generating an electromagnetic force. This electromagnetic force causes the movable core 32 to overcome the biasing force of the solenoid spring 34 and be attracted to the fixed core 33, thereby switching the flow state of air A flowing through the spool flow path 65. Furthermore, in the spool valve unit 2, the switching of the flow state of air A flowing through the spool flow path 65 causes the spool 24 to move against the biasing force of the spool spring 25, switching the state from communicating between the input port 20 and the exhaust port 22 to communicating between the input port 20 and the output port 21.

[0056] The substrate 5 includes a first substrate 50 disposed with substrate surfaces 500A and 500B along the shaft 13 c inserted from the shaft insertion port 610 c ; a second substrate 51 disposed near the junction box 62 ; and a third substrate 52 disposed near the solenoid portion 3 .

[0057] The main body 611, solenoid portion 3, and third substrate 52 are arranged on the first substrate surface 500A side of the substrate surfaces 500A and 500B of the first substrate 50. The second substrate 51 and junction box 62 are arranged on the second substrate surface 500B side opposite to the first substrate surface 500A side.

[0058] Sensors 4 are arranged at appropriate locations on the first substrate 50, the second substrate 51, and the third substrate 52. These sensors 4 include, for example, a first pressure sensor 40 that measures the fluid pressure of the air A flowing through 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 the rotational angle of the shaft 13c of the solenoid valve 1 as it rotates via the main shaft 13b of the drive device 12 in response to rotation of the valve stem 13a of the main valve 11. This rotational angle provides information on the valve opening of the main valve 11.

[0059] The main valve opening sensor 42, for example, is comprised of a magnetic sensor. It measures the magnetic field strength generated by the permanent magnet 131 mounted on the shaft 13c and obtains valve opening information of the main valve 11 based on this magnetic field strength. The main valve opening sensor 42 is preferably mounted on the first substrate surface 500A of the first substrate 5, which is arranged along the shaft 13c inserted through the shaft insertion port 610c, at a position opposite the outer circumference of the shaft 13c. This allows the main valve opening sensor 42 mounted on the first substrate 50 to be positioned close to the shaft 13c within the housing 6 without wasting space, enabling accurate acquisition of valve opening information.

[0060] Figure 4 FIG. 1 is a block diagram showing an example of a solenoid valve 1 according to an embodiment of the present invention. Figure 4As shown, the electrical configuration example of the solenoid valve 1 includes, in addition to the substrate 3 and the sensor 4 , a control unit 7 for controlling the solenoid valve 1 , a communication unit 8 for communicating with an external device 15 , and a power circuit unit 9 connected to an external power supply 16 .

[0061] Multiple sensors 4 serve as a sensor group for measuring physical quantities of each part. In addition to the above-mentioned first pressure sensor 40, second pressure sensor 41 and main valve opening sensor 42, they also have: a voltage sensor 43, which measures the supply voltage to the solenoid part 3; a current / resistance sensor 44, which measures the current value when power is supplied and the resistance value when power is not supplied in the solenoid part 3; a temperature sensor 45, which measures the internal temperature of the housing part 6; a magnetic sensor 46, which measures the magnetic field strength generated by the solenoid part 3; and a position sensor 49, which measures the position of the piston 122 relative to the cylinder 120.

[0062] In addition, multiple sensors 4 serve as a sensor group for obtaining information related to the operation history of each part, and include: an operation timer (timer) 47, which measures at least one of the total power-on time of the solenoid part 3 as the operation time of the solenoid part 3 and the current power-on linkage time; and an action counter (counter) 48, which counts the number of actions of each of the solenoid valve 1, the drive device 12 and the main valve 11.

[0063] Furthermore, these sensors 40-49 are not limited to being provided as separate sensors as described above; specific sensors may also function as other sensors, eliminating the need for separate sensors. For example, the magnetic sensor 46 may measure the magnetic field strength generated by the solenoid portion 3 and, based on this magnetic field strength, determine the current value when power is flowing through the solenoid portion 3, thereby eliminating the need for a separate current / resistance sensor 44. Furthermore, the microcontroller 70 may have the sensor functions built into it, or may implement a portion of the sensor functions. For example, the operation timer 47 and the motion counter 48 may be built into the microcontroller 70, eliminating the need for separate operation timer 47 and motion counter 48.

[0064] The control unit 7 includes: a microcontroller 70, which processes information representing the status of each part of the solenoid valve 1 obtained by multiple sensors 4 and controls each part of the solenoid valve 1; and a valve test switch 71, which controls the power supply state of the solenoid part 3 and implements the opening and closing operation of the main valve 11 during trial operation.

[0065] The microcontroller 70 includes a processor (not shown) such as a CPU (Central Processing Unit) and memories such as ROM (Read Only Memory) and RAM (Random Access Memory). The microcontroller 70 can include functions for implementing the data processing system 300 described later in this embodiment.

[0066] The valve test switch 71 is used to receive a command from the microcontroller 70 when a predetermined test operation condition is satisfied, and to perform a stroke test of the fluid pressure driven valve 10 as a test operation.

[0067] The stroke test is performed, for example, using either a full-stroke test or a partial-stroke test. The full-stroke test is performed by switching the main valve 11 from the fully-open state to the non-energized state to fully close it, and then switching it from the fully-closed state to the energized state to return it to the fully-open state. The partial-stroke test is performed by switching the main valve 11 from the fully-open state to the non-energized state to partially close it to a predetermined opening without fully closing it (i.e., without stopping the plant). The partial-stroke test is performed by switching the main valve 11 from the fully-open state to the non-energized state to return it to the fully-open state, and then switching it from the non-energized state to the energized state to return it to the fully-open state.

[0068] It should be noted that as a test run condition, for example, when an execution period based on an execution frequency specified by the administrator as a set value (for example, once a year) arrives, a specific designated date and time arrives, or an execution command is received from an external device 15, or the administrator operates a test execution button (not shown) provided on the solenoid valve 1, the test run condition is deemed to be met and a test run (stroke test) can be performed.

[0069] (Machine Learning Device)

[0070] In the fluid pressure driven valve 10 having the above-mentioned series of structures, by having the above-mentioned multiple sensors 4, various information about the fluid pressure driven valve 10 can be obtained, for example, during stable operation and unstable operation (for example, during test operation including opening and closing operations, and during emergency stop). Therefore, the following will describe a 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 obtained from the fluid pressure driven valve 10. It should be noted that the machine learning device 200 mentioned here includes not only a machine learning device provided as a device that acts alone, but also a machine learning device provided in the form of a non-transitory computer-readable medium that stores a program for causing any processor to perform the actions described below or one or more instructions for causing any processor to perform the actions.

[0071] Figure 5 FIG. 2 is a schematic block diagram of a machine learning device 200 according to an embodiment of the present invention. Figure 5 As shown, the machine learning device 200 of this embodiment includes a learning dataset acquisition unit 201 , a learning dataset storage unit 202 , a learning unit 203 , and a learned model storage unit 204 .

[0072] The learning dataset acquisition unit 201 is an interface unit for acquiring a plurality of data constituting a learning (training) dataset from various devices connected via, for example, a wired or wireless communication line. Here, as various devices connected to the learning dataset acquisition unit 201, for example, an external device 15 or an operator's computer PC1 used by an operator of the fluid pressure driven valve 10 can be cited. Figure 5 shows an example where the learning dataset acquisition unit 201 is connected to the external device 15 and the computer PC1 separately. However, the external device 15 and the operator's computer PC1 can also be composed of the same computer. The learning dataset acquisition unit 201 can acquire detection data from multiple sensors 4 of the fluid pressure-driven valve 10, for example, from the external device 15 as input data, and acquire diagnostic information about the fluid pressure-driven valve 10 associated with this input data from the operator's computer PC1 as output data. These associated input and output data constitute a single learning dataset, described below.

[0073] Figure 6 This is a diagram showing a data configuration example (supervised learning) used in the machine learning device 200 according to one embodiment of the present invention. Figure 7 This is a diagram showing an example of the structure of data used in the machine learning device 200 according to one embodiment of the present invention (unsupervised learning). Figure 6 、 Figure 7 Reference is also made, where appropriate, in the description of the data processing system and the inference device.

[0074] The learning dataset refers to the dataset used in the machine learning described below, such as Figure 6 、 Figure 7 As shown, the input data includes at least the valve opening 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 output data includes diagnostic information of the drive device 12. The details of these various data are described below in an example, but the present invention is not limited to this.

[0075] The valve opening of the main valve 11 refers to the value of the open / close state of the main valve 11 , and can be acquired from the main valve opening sensor 42 .

[0076] The pressure of air A is preferably the pressure of air A flowing through various parts of the fluid pressure-driven valve 10. Specifically, it preferably includes the solenoid valve output-side pressure of air A supplied and discharged from the solenoid valve 1 to the drive device 12. Furthermore, the solenoid valve output-side pressure of air A refers to the pressure of air A supplied and discharged from the solenoid valve 1 to the drive device 12. This pressure includes the pressure of air A when air A is supplied from the solenoid valve 1 to the drive device 12 (supply pressure) and the pressure of air A when air A is discharged from the drive device 12 to the outside air through the solenoid valve 1 (exhaust pressure). The solenoid valve output-side pressure of air A can be obtained by the aforementioned second pressure sensor 41.

[0077] The position of the piston 122 relative to the cylinder 120 refers to a value of movement of the piston 122 due to the supply and discharge of the air A into the cylinder 120 , and can be acquired by the position sensor 49 .

[0078] It should be noted that the valve opening of the main valve 11, the pressure of the air A and the position of the piston 122 relative to the cylinder 120 constituting the input data can each be constituted by a data (time data) at a specific time, or can be constituted by a data (time data) at a specific time. Figure 6 、 Figure 7 As shown in the brackets, it consists of multiple data (time series data) acquired at different times within a specified period. It should be noted that when each data consists of time series data, the time series data of the valve opening of the main valve 11, the time series data of the pressure of air A, and the time series data of the position of the piston 122 relative to the cylinder 120 can be data acquired at multiple times with the same sampling period and the same phase (no phase difference), or data with at least one of the sampling period and phase varying. To effectively improve inference accuracy, the latter time series data format is preferred.

[0079] Diagnostic information about the drive unit 12 indicates whether any abnormality has occurred in the drive unit 12 during abnormality diagnosis. This data can take various formats. Abnormalities include not only post-error abnormalities, where the occurrence of an abnormality is identified at the time of abnormality diagnosis, but also abnormality indicators, where the abnormality is within the normal allowable range at the time of abnormality diagnosis but a future abnormality is foreseeable.

[0080] For example, Figure 6 As shown, one form of diagnostic information can consist of information indicating whether the drive unit 12 is normal (no abnormality) or abnormal (abnormality is present). In this case, the diagnostic information is categorized into two values: for example, "0" indicates that the drive unit 12 is normal, and "1" indicates that the drive unit 12 is abnormal. The operator simply enters the corresponding value using the work computer PC1 in a format associated with the input data. It should be noted that in this case, information related to the specific abnormality is not necessarily required.

[0081] In addition, if Figure 6 As shown by the dotted line, the information indicating the abnormality of the drive device 12 in the above diagnostic information may also include the specific content of the abnormality. The specific content of the abnormality may include, for example, a poor air A circuit, an abnormal supply pressure of air A from the air supply source 14, poor operation 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, blockage of the air supply and exhaust port 124, poor operation of the transmission mechanism 125, etc. In this case, the diagnostic information is classified into multiple values ​​(3 or more), for example, the value indicating that the drive device 12 is in a normal state is set to "0", the value indicating that the abnormality of the drive device 12 is due to poor operation of the main valve 11 is set to "1", and the value indicating that the abnormality of the drive device 12 is due to a poor air A circuit is set to "2". The following values ​​are uniquely predetermined in accordance with the content of each abnormality, and then the operator can use the work computer PC1 to input the corresponding value in a form associated with the input data. By implementing such a setting of diagnostic information, it is possible to prepare information that includes not only whether an abnormality has occurred but also the specific content of the abnormality when the abnormality has occurred (corresponding to Figure 7 As the diagnostic information of one of the above methods, it is used to implement supervised learning in the machine learning described below (see Figure 6 ) situation.

[0082] In addition, information other than the above may be used as the diagnostic information of the drive device 12. For example, Figure 7As shown, other forms of diagnostic information can use information that only indicates that the drive device 12 is normal rather than abnormal. In this case, since the diagnostic information only includes information indicating that the drive device 12 is normal, it is inevitable that the learning data set that includes this diagnostic information as output data is only a data set consisting of input data and output data when the drive device 12 is normal. Therefore, the output data of the learning data set in this case is always the same, so those skilled in the art can certainly understand that the learning data set does not necessarily have output data as data. As this other form of diagnostic information, in the machine learning described below, it is used to implement unsupervised learning (see Figure 7 ) situation.

[0083] Alternatively, the input data within the learning dataset can also selectively include the total operating time of the fluid pressure-driven valve 10, the operating time since the last power-on of the fluid pressure-driven valve 10, the number of actuations of the main valve 11, the number of actuations of the driver 12, the number of actuations of the solenoid 3, and the opening and closing time of the main valve 11. The total operating time of the fluid pressure-driven valve 10 and the operating time since the last power-on of the fluid pressure-driven valve 10 can be obtained using the aforementioned operating timer 47, the number of actuations of the main valve 11, the number of actuations of the driver 12, and the number of actuations of the solenoid 3 can be obtained using the aforementioned actuation counter 48, and the opening and closing time of the main valve 11 can be obtained using a timer (not shown). Increasing the types of input data generally helps improve the inference accuracy of the learned model obtained after machine learning. However, using input data with a low degree of relevance to diagnostic information may hinder the improvement of the learned model's inference accuracy. Therefore, the amount and type of input data should be appropriately selected, taking into account, for example, the state of the fluid pressure-driven valve 10 to which the learned model is applied.

[0084] In particular, the contact areas between the cylinder 120 and the piston 122 of the drive device 12, where the relative positions change due to high-pressure air, and the contact areas of the transmission mechanism 125 that converts reciprocating linear motion into rotational motion, are subject to significant torque and may experience abnormalities such as wear, degradation, or breakage. Such abnormalities in the contact areas of the drive device 12 are expected to affect changes in the drive characteristics of the fluid pressure-driven valve 10. Therefore, in this embodiment, it is preferred to obtain the pressure of the output-side drive fluid supplied from the solenoid valve 1 to the drive device 12, the position of the piston 122 relative to the cylinder 120, and the valve opening of the main shaft 13b for diagnosis. That is, by using the pressure of the output-side drive fluid supplied from the solenoid valve 1 to the drive device 12, the piston 122 moves relative to the cylinder 120, and the valve opening of the main valve 13 changes. By following this series of processes, the state of the drive device 12 can be understood, and signs of abnormalities in the drive device 12 can be detected. For example, when the transmission mechanism 125 is a rack and pinion mechanism, an abnormality in the contact portion between the rack and the pinion can be detected. When it is a Scotch yoke mechanism, an abnormality in the contact portion between the shaft and the yoke can be detected.

[0085] The learning dataset storage unit 202 is a database for storing a plurality of data constituting the learning dataset acquired by the learning dataset acquisition unit 201 in association with the relevant input data and output data as one learning dataset. The specific structure of the database constituting the learning dataset storage unit can be adjusted as appropriate. For example, Figure 5 In the figure, for convenience of explanation, the learning dataset storage unit 202 and the learned model storage unit 204 described later are shown as independent storage units, but they can also be composed of a single storage medium (database).

[0086] The learning unit 203 performs machine learning using the multiple learning datasets stored in the learning dataset storage unit 202, thereby generating a learned model that has learned the correlation between the input data and the output data included in the multiple learning datasets. In this embodiment, as described in detail later, the specific method of machine learning adopts supervised learning using a neural network. However, the specific method of machine learning is not limited to this. As long as the correlation between input and output can be learned from the learning dataset, other learning methods can also be used. For example, ensemble learning (random forest, boosting algorithm, etc.) can also be used.

[0087] The learned model storage unit 204 is a database for storing the learned models generated by the learning unit 203. The learned models stored in this learned model storage unit 204 are applied to the actual system upon request via communication lines, including the Internet, and storage media. The specific application of the learned models to the actual system (data processing system 300) will be described in detail later.

[0088] Next, a learning method in the learning unit 203 using the plurality of learning data sets obtained as described above will be described, focusing on supervised learning. Figure 8 This is a diagram showing an example of a neural network model for supervised learning implemented in a machine learning device according to one embodiment of the present invention. Figure 8 The neural network in the neural network model shown is composed of l neurons (x1~x1) in the input layer, m neurons (y11~y1m) in the first intermediate layer, n neurons (y21~y2n) in the second intermediate layer, and o neurons (z1~zo) in the output layer. The first and second intermediate layers are also called hidden layers. As a neural network, in addition to the first and second intermediate layers, it can also have multiple hidden layers, or only the first intermediate layer can be used as a hidden layer. It should be noted that in Figure 8 In the figure, a neural network model is illustrated in which multiple (o) output layers are set. However, for example, when the above-mentioned diagnostic information is determined by one value, that is, when the number of training data described later is 1 (only t1), the number of neurons in the output layer can also be 1 (only z1).

[0089] In addition, 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, there are nodes connecting the neurons between layers, and each node establishes a corresponding relationship with the weight wi (i is a natural number).

[0090] The neural network in the neural network model of this embodiment uses a learning dataset to learn the correlation between the valve opening of the main valve 11, the pressure of air A, and the position of the piston 122 relative to the cylinder 120, and the diagnostic information of the drive unit 12. Specifically, the valve opening of the main valve 11, the pressure of air A, and the position of the piston 122 relative to the cylinder 120, as state variables, are associated with neurons in the input layer. The output value of the neurons in the output layer is calculated using a common neural network output value calculation method. Specifically, for all neurons other than the input layer neurons, the value of the output-side neuron connected to the output-side neuron is calculated as the sum of the multiplication of the value of the input-side neuron connected to the output-side neuron and the weight wi, which is associated with the node connecting the output-side neuron to the input-side neuron. It should be noted that the format in which the information obtained as the state variables is input to the neurons in the input layer can be appropriately set, taking into account the accuracy of the generated learned model, etc. Specifically, preprocessing can be performed on specific input data to adjust the number of neurons corresponding to each input data or to adjust the values ​​to match the corresponding neurons.

[0091] Then, the calculated values ​​of the o neurons z1~zo located in the output layer, that is, in this embodiment, one or more diagnostic information are compared with the training data t1~to which are also composed of one or more diagnostic information and constitute part of the learning data set, and the error is calculated. The weight wi that establishes a corresponding relationship with each node is repeatedly adjusted (back propagation) to reduce the calculated error.

[0092] Then, when the above series of processes are repeated for a specified number of times or the specified conditions such as the error is less than the allowable value are met, the learning is terminated and the neural network model (all weights wi corresponding to each node) is stored in the learned model storage unit 204 as a learned model.

[0093] (Machine Learning Methods)

[0094] In connection with the above, the present invention provides a machine learning method. Figure 6 (Learning phase), Figure 7 (Learning phase), Figure 8 、 Figure 9 The machine learning method of the present invention is described. Figure 9: is a flowchart showing an example of a machine learning method according to an embodiment of the present invention. In the machine learning method shown below, the description is based on the above-mentioned machine learning device 200, but the structure as a premise is not limited to the above-mentioned machine learning device 200. In addition, the machine learning method is implemented by using a computer, but various computers can be applied as the computer, for example, a computer device constituting the external device 15, the work computer PC1 or the microcontroller 70, a server device configured on the network, etc. In addition, regarding the specific structure of the computer, for example, a structure including at least a computing device composed of a CPU, a GPU, etc., a storage device composed of a volatile or non-volatile memory, etc., a communication device for communicating with a network or other devices, and a bus connecting these devices can be adopted.

[0095] As a supervised learning method of the machine learning method of this embodiment, as a preliminary preparation for starting machine learning, first, a desired number of learning data sets (see Figure 6 ), the prepared plurality of learning data sets are stored in the learning data set storage unit 202 (step S11). The number of learning data sets prepared here can be set in consideration of the inference accuracy required by the final learning model.

[0096] Several methods can be used to prepare the learning dataset used in this supervised learning. For example, when an abnormality occurs in a specific drive device 12 or when an operator recognizes signs of an abnormality, various information about the stable operation of the fluid pressure-driven valve 10 at that time is obtained using multiple sensors 4, etc. The operator then uses a work computer PC1, etc. to determine and input diagnostic information in a form associated with this information, thereby preparing the input data and output data that constitute the learning dataset (for example, the output data value in this case is "1"). Furthermore, a method can be used to prepare a desired number of learning datasets by repeating this operation. It should be noted that, in addition to this method, various methods can be used to prepare the learning dataset, such as actively creating an abnormal state in the drive device 12 to obtain the learning dataset. However, the various information about the fluid pressure-driven valve 10 often has unique tendencies for each fluid pressure-driven valve 10. Therefore, as the object of obtaining the data constituting the learning dataset, it is preferable to collect the data constituting the learning dataset from only one fluid pressure-driven valve 10 to which the learned model obtained through machine learning, described below, is to be applied. In addition, the learning data set includes not only a data set consisting of input and output data when an abnormality occurs, but also a specified amount of learning data set consisting of input data and output data when no abnormality occurs, that is, in the normal state of the drive device 12 (for example, the value of the output data at this time is "0").

[0097] When step S11 is completed, a neural network model before learning is prepared in order to start learning (S12) of the learning unit 203. The neural network model before learning prepared here has, for example, Figure 8 The structure shown in FIG. 1 is used as its structure, and the weight of each node is set to an initial value. Then, a learning dataset is selected, for example, randomly, from the plurality of learning datasets stored in the learning dataset storage unit 202 (step S13), and the input data in the one learning dataset is input to the input layer of the neural network model prepared before learning (see FIG. Figure 8 )(Step S14).

[0098] Here, the output layer (see Figure 8 ) is a value generated by the neural network model before learning, and therefore in most cases is a value different from the expected result, that is, a value showing information different from the correct diagnostic information. Therefore, next, machine learning is implemented (step S15) using the diagnostic information as training data in a learning data set obtained in step S13 and the value of the output layer generated in step S13. The machine learning implemented here can be, for example, a process (back propagation) of comparing the diagnostic information constituting the training data with the value of the output layer and adjusting the weights that have established a corresponding relationship with each node in the neural network model before learning to obtain a preferred output layer. It should be noted that the number and form of the values ​​output to the output layer of the neural network model before learning are the same number and form as the training data in the learning data set that is the learning object.

[0099] To specifically illustrate the machine learning described here, assume that the diagnostic information constituting the training data consists of a binary classification system where normal conditions are set to "0" and abnormal conditions are set to "1." Furthermore, if the value of the output data in the learning dataset selected in step S13 is "1," the output layer outputs a predetermined value between 0 and 1, specifically, a value such as "0.63." Therefore, in step S15, if the same input data is input to the input layer, the weights associated with the nodes of the neural network model being learned are adjusted so that the value obtained by the learning neural network model approaches "1."

[0100] When machine learning is performed in step S15, it is determined whether machine learning needs to be continued further (step S16), for example, based on the remaining number of unlearned learning data sets stored in the learning data set storage unit 202. Then, in the case of continuing machine learning (step S16 is "No"), the process returns to step S13, and in the case of ending machine learning (step S16 is "Yes"), the process proceeds to step S17. In the case of continuing the above-mentioned machine learning, the process of steps S13 to S15 is performed multiple times on the neural network model being learned using the unlearned learning data sets. The accuracy of the finally generated learned model generally increases in proportion to the number of times.

[0101] When machine learning is complete ("Yes" in step S16), the neural network generated by adjusting the weights corresponding to each node through a series of steps is stored as a learned model in learned model storage unit 204 (step S17), thus completing the learning process. The stored learned model can be used in data processing system 300, which will be described later.

[0102] In the learning process and machine learning method of the above-mentioned machine learning device, it is described that in order to generate a learned model, a machine learning process is repeatedly performed multiple times on a neural network model (before learning) to improve its accuracy, thereby obtaining a learned model that is sufficient for application to the data processing system 300. However, the present invention is not limited to this acquisition method. For example, a plurality of learned models that have undergone a predetermined number of machine learning processes may be stored in advance in the learned model storage unit 204 as a candidate, a data set for validity judgment may be input to the plurality of learned model groups, an output layer (the value of the neurons) is generated, the accuracy of the values ​​determined in the output layer is compared and studied, and an optimal learned model suitable for the data processing system 300 is selected. It should be noted that the data set for validity judgment only needs to be composed of the same data set as the learning data set used for learning and has not been used for learning.

[0103] As described above, by applying the machine learning device and machine learning method of this embodiment, the following learned model can be obtained: diagnostic information indicating whether an abnormality (including subsequent abnormalities and signs of abnormalities) has occurred can be correctly derived from various data obtained by multiple sensors 4 arranged at appropriate positions of the fluid pressure driven valve 10.

[0104] In the learning method and machine learning method of the machine learning device 200 described above, "supervised learning" has been described. However, as a method for generating a learned model, other known "supervised learning" methods such as convolutional neural networks (CNNs) can be used. Alternatively, "unsupervised learning" can be used, which uses a learning dataset that includes diagnostic information of the other methods described above, that is, information indicating that the fluid pressure driven valve 10 is normal and not abnormal, as diagnostic information constituting output data (see Figure 7 By using “unsupervised learning”, even when only information on the normal state of the drive device 12 can be obtained from the diagnostic information in the output data corresponding to the input data, Figure 7 As shown in the "learning phase" of the present invention, a learning model is obtained by learning the correlation between the features representing the normal state of the input data and the output data. In this case, during the inference in the data processing system 300 described later, the input data that is judged to be inconsistent with the features of the normal state by a predetermined amount is regarded as not being in the normal state, that is, as being in the abnormal state, thereby enabling the inference of diagnostic information. As a specific method of this "unsupervised learning", for example, Figure 7 The known method using an autoencoder etc. is briefly shown in FIG, and detailed description thereof is omitted here.

[0105] (Data Processing System)

[0106] Next, refer to Figure 10 An application example of the learned model generated by the above-mentioned machine learning device 200 and machine learning method will be described. Figure 10 This is a schematic block diagram showing a data processing system according to one embodiment of the present invention.

[0107] As an example of the data processing system 300 of this embodiment, an embodiment is shown in which the data processing system 300 is installed in the microcontroller 70 of the fluid pressure-driven valve 10. It should be noted that at least a portion of the data processing system 300 may also be applied to other devices, such as other devices connected to the external device 15 or the fluid pressure-driven valve 10.

[0108] 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 .

[0109] The input data acquisition unit 301 is an interface unit connected to the multiple sensors 4 of the fluid pressure driven valve 10 and used to acquire the data output by each sensor 4. The input data acquisition unit 301 acquires at least the valve opening of the main valve 11, the pressure of the air A, and the position of the piston 122 relative to the cylinder 120. Figure 10In the illustrated example, all sensors 4 included in the fluid pressure-driven valve 10 are connected so that all input data usable in the inference described later can be acquired. However, which sensor 4 to connect to the input data acquisition unit 301 can be appropriately selected based on, for example, the learned model used in the inference unit 302 described later. Furthermore, the inference results of the inference unit 302 are preferably stored in a storage unit (not shown). The stored past inference results can be utilized, for example, as a learning dataset for online learning to further improve the inference accuracy of the learned model in the learned model storage unit 303.

[0110] The inference unit 302 is used to infer whether an abnormality has occurred in the drive device 12 based on the various data regarding the fluid pressure-driven valve 10 acquired by the input data acquisition unit 301. This inference utilizes, for example, a learned model learned using the machine learning device 200 and the machine learning method described above. This learned model is stored in a learned model storage unit 303, which comprises an arbitrary storage medium. It should be noted that the inference unit 302 not only performs inference processing using the learned model but also includes a pre-processing function, such as adjusting the input data acquired by the input data acquisition unit 301 into a desired format before inputting it into the learned model. Furthermore, as a post-processing function, the inference unit 302 applies, for example, a predetermined threshold to the output value of the learned model to ultimately determine whether an abnormality (including subsequent abnormalities and signs of abnormalities) has occurred (no abnormality (normal) or an abnormality (abnormal)).

[0111] As described above, the learned model storage unit 303 is a storage medium for storing learned models used by the inference unit 302. The number of learned models stored in the learned model storage unit 303 is not limited to one. For example, multiple learned models with different amounts of input data or different learning methods (e.g., supervised learning and unsupervised learning implemented by the machine learning device 200, etc.) can be stored, and these learned models can be selectively used.

[0112] The reporting unit 304 is used to report the inference results of the inference unit 302 to an operator or the like. Various methods can be used for specific reporting methods, such as transmitting the inference results to the external device 15 via the communication unit 8 and displaying them on the GUI of the external device 15, or providing a light-emitting component, a speaker, etc. in the fluid pressure-driven valve 10 and activating them to report the occurrence of an abnormality to the operator or the like.

[0113] Below, refer to Figure 6 (Inference stage), Figure 7 (Inference stage), Figure 11, the data processing process of the data processing system with the above structure is explained. Figure 11 1 is a flowchart illustrating an example of a data processing procedure of the data processing system 300 according to one embodiment of the present invention.

[0114] When power is supplied from the external power supply 16 to the solenoid valve 1 of the fluid pressure driven valve 10 and the abnormality diagnosis of the driving device 12 is started, the input data acquisition unit 301 acquires various data indicating the status of each part of the fluid pressure driven valve 10 acquired by the multiple sensors 4 (step S21). After the input data acquisition unit 301 has acquired the desired input data (valve opening of the main valve 11, pressure of the air A, and position of the piston 122 relative to the cylinder 120 (see Figure 6 、 Figure 7 )), inference is performed by the inference unit 302 based on the input data (step S22). At this time, it is preferable to predetermine a learned model for inference. Furthermore, if the predetermined learned model requires, for example, predetermined time series data as its input data, inference in step S22 is performed after the input data acquisition unit 301 acquires the necessary amount of data.

[0115] Specifically, the inference unit 302 pre-processes the input data and inputs it into the learning model completion model, and post-processes the output value from the learning model completion model to determine whether an abnormality (including post-abnormality and abnormality signs) occurs as the inference result. Figure 6 In the post-processing of the "inference stage" of the learning model, the inference unit 302 compares the output value of the learning model completion model (if it is a binary classification, it is a value between 0 and 1) with the specified threshold value. For example, if the output value of the learning model completion model is above the specified threshold value, it is judged as "abnormal (abnormal)", and if it is less than the specified threshold value, it is judged as "no abnormality (normal)", and the judgment result is output as the inference result. In addition, in unsupervised learning (see Figure 7 In the post-processing of the "inference stage"), the inference unit 302 calculates the difference (distance) between the output value (feature value) of the completed model of the learning model and the feature value based on the input data. If the difference (distance) is above the specified threshold, it is judged as "abnormal (abnormal)". If the difference (distance) is less than the specified threshold, it is judged as "no abnormality (normal)", and the judgment result is output as the inference result.

[0116] Then, in step S22, the reasoning of the reasoning unit 302 is implemented, and when the reasoning result indicates "no abnormality (normal)" (step S23 is "No"), the process returns to step S21 and continues a series of reasoning. Figure 6 、 Figure 7As shown, if the inference result indicates "abnormality (abnormality)" ("Yes" in step S23), the reporting unit 304 reports the inference result as "abnormality (abnormality)" to the operator, etc., indicating that an abnormality (including subsequent abnormalities and abnormality precursors) has occurred in the drive device 12 (step S24). After the abnormality is reported in step S24, the process returns to step S21 and the inference process continues. It should be noted that, depending on the intended use of the drive device 12 and the nature of the detected abnormality, the fluid pressure-driven valve 10 may be stopped at the stage of abnormality detection.

[0117] (Inference Device)

[0118] The present invention can be provided not only in the form of the above-mentioned data processing system 300, but also in the form of an inference device for implementing inference. In this case, the inference device includes a memory and at least one processor, wherein the processor is capable of performing a series of processes. The series of processes includes: the process of obtaining input data including the valve opening of the main valve, the pressure of the driving fluid, and the position of the piston 122 relative to the cylinder 120; and the process of inferring diagnostic information in the fluid pressure driven valve 10 when the input data is input. By providing the present invention in the form of the above-mentioned inference device, it can be simply applied to various fluid pressure driven valves 10 compared to the case where the data processing system 300 is installed. Those skilled in the art will of course understand that: at this time, when the inference device implements the processing of inferring diagnostic information, the inference method previously described in this specification, which is implemented by the inference unit 302 of the data processing system using the learned model obtained by learning through the machine learning device and the machine learning method in the present invention, can also be used.

[0119] The present invention is not limited to the above-described embodiment, and can be implemented with various modifications without departing from the spirit of the present invention, and all such modifications are included in the technical concept of the present invention.

[0120] Description of Reference Numerals

[0121] 1: Solenoid valve; 3: Solenoid unit; 4: Sensor; 10: Fluid pressure driven valve; 11: Main valve; 12: (Fluid pressure type) drive device; 13a: Valve stem; 13b: Main shaft; 13c: Shaft; 14: Air supply source; 15: External device; 26: Input side flow path; 27: Output side flow path; 28: Exhaust flow path; 30: Solenoid housing; 31: Solenoid coil; 32: Movable iron core; 40: First pressure sensor; 41: Second pressure sensor; 42: Main valve opening sensor; 43: Voltage sensor; 44: Current / resistance sensor; 45: Temperature sensor Sensor; 46: Magnetic sensor; 47: Operation timer (timer); 48: Action counter (counter); 70: Microcontroller; 100: Piping; 200: Machine learning device; 201: Learning data set acquisition unit; 202: Learning data set storage unit; 203: Learning unit; 204: Learning completed model storage unit; 300: Data processing system; 301: Input data acquisition unit; 302: Inference unit; 303: Learning completed model storage unit; 304: Reporting unit; A: Air (driving fluid); PC1: Operation computer.

Claims

1. A machine learning device, applied to a fluid pressure driven valve comprising at least a main valve, a drive device including a cylinder and a piston for driving the main valve, and a solenoid valve for controlling the supply and discharge of a driving fluid relative to the drive device, wherein: have: a learning data set storage unit storing a plurality of learning data sets consisting of input data including the valve opening of the main valve, the pressure of the output-side driving fluid supplied from the solenoid valve to the driving device, and the position of the piston relative to the cylinder, and output data consisting of diagnostic information of the driving device corresponding to the input data; a learning unit that learns a learning model for inferring the correlation between the input data and the output data by inputting a plurality of sets of the learning data sets; as well as The learned model storage unit stores the learned model learned by the learning unit.

2. The machine learning device according to claim 1, wherein: The diagnostic information is information indicating whether the drive device is normal or abnormal.

3. The machine learning device according to claim 1, wherein: The diagnostic information is information indicating only that the drive device is normal and not abnormal.

4. The machine learning apparatus according to any one of claims 1 to 3, wherein: The driving device has a transmission mechanism for converting the linear motion of the piston into the rotational motion of the main valve. The diagnostic information is information specific to the transmission mechanism.

5. The machine learning device according to claim 4, wherein: The transmission mechanism is a rack-and-pinion mechanism including a rack that linearly moves together with the piston and a pinion connected to the rack and rotationally moves together with the main valve, and the diagnostic information is information on the rack-and-pinion mechanism.

6. The machine learning device according to claim 4, wherein: The transmission mechanism is a scotch-yoke mechanism including a shaft that linearly moves together with the piston and a yoke that is connected to the shaft and rotationally moves together with the main valve, and the diagnostic information is information about the scotch-yoke mechanism.

7. The machine learning device according to claim 4, wherein: The diagnostic information is information related to wear of a connecting portion that converts linear motion into rotational motion in the transmission mechanism.

8. The machine learning device according to claim 5 or 6, wherein: The diagnostic information is information related to wear of a connecting portion that converts linear motion into rotational motion in the transmission mechanism.

9. A data processing system for a fluid pressure driven valve, the fluid pressure driven valve comprising at least a main valve, a drive device including a cylinder and a piston for driving the main valve, and an electromagnetic valve including a solenoid portion for controlling the supply and discharge of a driving fluid to the drive device, wherein: The data processing system comprises: an input data acquisition unit that acquires input data including a valve opening 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 An inference unit that inputs the input data acquired by the input data acquisition unit into a learned model generated by the machine learning device according to any one of claims 1 to 8, and infers diagnostic information of the drive device.

10. An inference device for a fluid pressure driven valve, the fluid pressure driven valve comprising at least a main valve, a drive device including a cylinder and a piston for driving the main valve, and an electromagnetic valve including a solenoid portion for controlling the supply and discharge of a driving fluid to the drive device, wherein: The inference device comprises a memory and at least one processor, The at least one processor is configured to perform the following processing: acquiring input data including a valve opening 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 When the input data is input, diagnostic information of the drive device is inferred.

11. A machine learning method using a computer applied to a fluid pressure driven valve, the fluid pressure driven valve comprising at least a main valve, a drive device including a cylinder and a piston for driving the main valve, and an electromagnetic valve including a solenoid portion for controlling the supply and discharge of a driving fluid to the drive device, wherein: The machine learning method has the following steps: storing a plurality of learning data sets consisting of input data including the valve opening of the main valve, the pressure of the output-side driving fluid supplied from the solenoid valve to the driving device, and the position of the piston relative to the cylinder, and output data consisting of diagnostic information of the driving device corresponding to the input data; By inputting a plurality of sets of the learning data sets, a learning model for inferring the correlation between the input data and the output data is learned; The learned learning model is stored.

Images