JP7526571B2 - State estimation device, control valve, state estimation program, and state estimation method - Google Patents

Description

The present invention relates to a state estimation device for estimating the state of a control valve, a control valve, a state estimation program, and a state estimation method.

Control valves such as hydraulic servo valves are used to control engines installed on moving objects such as ships. By using electrically controllable control valves to precisely control the supply of fuel to the engine and the exhaust from the engine, the thermal efficiency of the engine can be improved and fuel consumption can be reduced.

Patent No. 5465365

Because ships sail on the ocean, it is not always possible to immediately deal with a malfunction or other problem that occurs in the onboard control valves. Conventionally, when the ship is anchored in port, the control valve with the malfunction or problem would be removed and brought to a factory or other facility for inspection, and any necessary repairs or replacements would be carried out. However, in order to prevent the ship from becoming unable to sail and causing significant damage, it is extremely important to accurately grasp the condition of the control valve and take appropriate measures before it becomes inoperable due to a malfunction or problem.

The present invention was made in consideration of these problems, and its purpose is to provide a technology for more accurately estimating the state of a control valve.

To solve the above problem, a state estimation device according to one embodiment of the present invention includes an acquisition unit that acquires detection information of a data set including a target position and an actual position of a first movable part in a control valve that controls the flow rate of a working fluid depending on the position of the first movable part and a value of a current supplied to a drive unit that drives the first movable part, and an estimation unit that estimates the cleanliness of the working fluid based on the detection information acquired by the acquisition unit.

Yet another aspect of the present invention is a control valve. This control valve includes a first movable part whose position changes in response to a control signal for specifying a position, and whose flow rate of a working fluid is controlled in response to the position, an acquisition unit that acquires detection information of a data set including a target position and an actual position of the first movable part and a value of a current supplied to a drive unit that drives the first movable part, and an estimation unit that estimates the cleanliness of the working fluid based on the detection information acquired by the acquisition unit.

Yet another aspect of the present invention is a state estimation program. This state estimation program causes a computer to function as an acquisition unit that acquires detection information of a data set including a target position and an actual position of a first movable part in a control valve that controls the flow rate of a working fluid depending on the position of the first movable part and a value of a current supplied to a drive unit that drives the first movable part, and an estimation unit that estimates the cleanliness of the working fluid based on the detection information acquired by the acquisition unit.

Yet another aspect of the present invention is a state estimation method. This state estimation method causes a computer to execute a step of acquiring detection information of a data set including a target position and an actual position of a first movable part in a control valve that controls a flow rate of a working fluid depending on the position of the first movable part and a value of a current supplied to a drive part that drives the first movable part, and a step of estimating the cleanliness of the working fluid based on the detection information acquired in the acquiring step.

Yet another aspect of the present invention is a state estimation device. This state estimation device includes an acquisition unit that acquires detection information of a data set including a target position or actual position of a first movable part in a control valve that controls the flow rate of a working fluid according to the position of the first movable part and an actual position of a second movable part that changes its position according to the flow rate of the working fluid, and an estimation unit that estimates an abnormality in a detection unit for detecting the position of the second movable part based on the detection information acquired by the acquisition unit.

Yet another aspect of the present invention is a control valve. This control valve includes a first movable part whose position changes in response to a control signal for specifying a position and whose flow rate of a working fluid is controlled in response to the position, an acquisition unit that acquires detection information of a data set including a target position or an actual position of the first movable part and an actual position of a second movable part whose position changes in response to the flow rate of the working fluid, and an estimation unit that estimates an abnormality in the detection unit for detecting the position of the second movable part based on the detection information acquired by the acquisition unit.

Yet another aspect of the present invention is a state estimation program. This state estimation program causes a computer to function as an acquisition unit that acquires detection information of a data set including a target position or actual position of a first movable part in a control valve that controls the flow rate of a working fluid according to the position of the first movable part and an actual position of a second movable part that changes its position according to the flow rate of the working fluid, and an estimation unit that estimates an abnormality in a detection unit for detecting the position of the second movable part based on the detection information acquired by the acquisition unit.

Yet another aspect of the present invention is a state estimation method. This state estimation method causes a computer to execute a step of acquiring detection information of a data set including a target position or actual position of a first movable part in a control valve that controls the flow rate of a working fluid according to the position of the first movable part and an actual position of a second movable part that changes its position according to the flow rate of the working fluid, and a step of estimating an abnormality of a detection part for detecting the position of the second movable part based on the detection information acquired in the acquiring step.

In addition, any combination of the above components, or mutual substitution of the components or expressions of the present invention among methods, devices, programs, temporary or non-temporary storage media on which programs are recorded, systems, etc. are also valid aspects of the present invention.

The present invention allows the state of the control valve to be estimated more accurately.

1 is a diagram showing a configuration of a management system according to an embodiment of the present invention; 1 is a diagram illustrating a schematic configuration of a hydraulic servo valve and its surroundings mounted on a ship. FIG. 2 is a diagram illustrating a schematic configuration of a hydraulic servo valve. 4 is a schematic diagram showing the position of a valve body of a pilot valve in a first direction and the open/closed state of a port. FIG. FIG. 2 is a diagram showing a configuration of a servo valve control device. FIG. 2 is a diagram illustrating a configuration of a learning device and a state estimation device. 3 is a flowchart showing the procedure of a state estimation method according to the present embodiment. FIG. 2 is a diagram illustrating the configuration of a learning device and a state estimation device according to an embodiment 1-1. FIG. 1 is a diagram illustrating a configuration of a learning device and a state estimation device according to an embodiment 1-2. FIG. 13 is a diagram illustrating the configuration of a learning device and a state estimation device according to Examples 1-3. FIG. 1 is a diagram illustrating the configuration of a learning device and a state estimation device according to Examples 1-4. FIG. 13 is a diagram illustrating the configuration of a learning device and a state estimation device according to Examples 1-5. FIG. 11 is a diagram illustrating a configuration of a learning device and a state estimation device according to a second embodiment. 5A and 5B are diagrams illustrating the operation of a spool of a pilot valve and a spool of a main valve. FIG. 11 is a diagram illustrating a configuration of a learning device and a state estimation device according to a third embodiment. FIG. 13 is a diagram illustrating the configuration of a learning device and a state estimation device according to a fourth embodiment. 5A and 5B are diagrams illustrating the operation of a spool of a pilot valve and a spool of a main valve. FIG. 13 is a diagram illustrating the configuration of a learning device and a state estimation device according to a fifth embodiment. FIG. 11 is a diagram showing the state of a spool position sensor of the main valve. FIG. 13 is a diagram illustrating the configuration of a learning device and a state estimation device according to a sixth embodiment. FIG. 13 is a diagram illustrating the configuration of a learning device and a state estimation device according to a seventh embodiment.

The present invention will be described below based on a preferred embodiment with reference to the drawings. In the embodiments and modified examples, identical or equivalent components and parts are given the same reference numerals, and duplicated explanations will be omitted as appropriate. Furthermore, the dimensions of the parts in each drawing are shown enlarged or reduced as appropriate to facilitate understanding. Furthermore, some of the parts that are not important for explaining the embodiment are omitted in each drawing.

In addition, terms including ordinal numbers such as first and second are used to describe various components, but these terms are used only for the purpose of distinguishing one component from another and do not limit the components.

Figure 1 shows the configuration of a management system according to an embodiment of the present invention. The management system 1 manages a control valve based on information related to the operation of the control valve. Although the management system 1 can be used to manage any control valve, this embodiment will mainly describe an example of managing a hydraulic servo valve for controlling an engine mounted on a ship 2. The management system 1 includes a learning device 300 and a state estimation device 400. The learning device 300 learns an estimation model for estimating the state of the hydraulic servo valve based on information related to the hydraulic servo valve. The state estimation device 400 estimates the state of the hydraulic servo valve using the estimation model learned by the learning device 300.

The learning device 300 learns the estimation model using learning data for learning the estimation model. The learning data may be log data recorded when the hydraulic servo valve or another hydraulic servo valve of the same type as the hydraulic servo valve is actually used in the ship 2, or test data recorded when the hydraulic servo valve or another hydraulic servo valve of the same type as the hydraulic servo valve is used in a test environment other than the ship 2, or simulation data generated by a simulator for simulating the operation of the hydraulic servo valve, or a combination of two or more of them. Hereinafter, unless otherwise specified, the log data or test data may be log data or test data recorded when the hydraulic servo valve itself, the state of which is to be estimated, is used or tested, or log data or test data recorded when another hydraulic servo valve of the same type as the hydraulic servo valve is used or tested, or a combination of them.

When log data is used as learning data, the estimation model can be learned based on log data collected when the hydraulic servo valve is actually used in an environment similar to that of the hydraulic servo valve whose state is to be estimated, thereby improving the accuracy of the estimation model. In this case, the ship 2 may be equipped with a log data storage device for storing the log data, and the log data may be read from the log data storage device when the ship 2 is docked, for example, and supplied to the learning device 300. In addition, the ship 2 may be equipped with a communication device for ship-to-shore communication, and the log data may be transmitted from the ship 2 to the learning device 300 via the communication network 3.

When test data or simulation data is used as learning data, a large amount of data can be generated when the hydraulic servo valve is used in various conditions or environments, thereby improving the accuracy and versatility of the estimation model. For example, even if it is difficult to obtain log data when a fault that occurs very rarely actually occurs, an estimation model that can accurately estimate such faults can be generated by training the estimation model using test data when the fault is experimentally generated or simulation data when the fault is simulated.

The learning device 300 may further learn and update the learned estimation model used in the state estimation device 400.

The state estimation device 400 acquires information related to the hydraulic servo valve mounted on the ship 2 and estimates the state of the hydraulic servo valve using an estimation model. The state estimation device 400 may read log data from a log data storage device mounted on the ship 2 and estimate the state of the hydraulic servo valve in the past navigation of the ship 2 based on the read log data. The state estimation device 400 may also receive information related to the hydraulic servo valve from a communication device for ship-to-shore communication mounted on the ship 2 via the communication network 3 and estimate the state of the hydraulic servo valve in the ship 2 currently sailing based on the received information. The state estimation device 400 may also be mounted on the ship 2 and acquire information related to the hydraulic servo valve in real time to estimate the state of the hydraulic servo valve. In this case, the estimation result by the state estimation device 400 may be transmitted to the owner, management entity, maintenance entity, etc. of the ship 2 via ship-to-shore communication.

According to the technology of this embodiment, the past or present state of the hydraulic servo valve can be accurately estimated, so that even if a failure or malfunction occurs in the hydraulic servo valve, measures can be taken quickly and appropriately. Even if there is no failure or malfunction in the hydraulic servo valve, the possibility of a failure or malfunction occurring in the hydraulic servo valve during future navigation and the lifespan of the hydraulic servo valve can be accurately predicted from the current or past state of the hydraulic servo valve. In this way, the technology of this embodiment is extremely significant for improving the safety and efficiency of moving bodies such as the ship 2.

Figure 2 shows a schematic diagram of the configuration around the hydraulic servo valve 100 mounted on the ship 2. The engine 80 mounted on the ship 2 includes a plurality of cylinders 81 and a sensor 82. The sensor 82 detects the engine 80's speed, load, pressure, exhaust temperature, and the like. The hydraulic servo valve 100 is provided corresponding to each of the plurality of cylinders 81, and controls the fuel injection and exhaust of each cylinder 81. In this embodiment, the hydraulic servo valve 100 includes a pilot valve that controls the flow rate of hydraulic oil supplied to the actuator by electrically controlling the position of a spool, and a main valve, which is an example of an actuator controlled by the pilot valve. The hydraulic servo valve 100 controls the flow rate of hydraulic oil supplied to another actuator provided to drive an injection valve, an exhaust valve, and the like, according to the position of the spool of the main valve. In another example, the hydraulic servo valve 100 may directly drive an injection valve, an exhaust valve, and the like by moving the spool of the main valve.

The engine control device 91 determines the rotation speed of the engine 80 and inputs instructions to the servo valve control device 110 in response to instructions input from a control panel (not shown) for controlling the navigation of the ship 2. The servo valve control device 110 calculates the target positions of the spools of the main valves of the multiple hydraulic servo valves 100 in response to instructions from the engine control device 91, and controls the positions of the spools of the pilot valves so that the positions of the spools of the main valves become the calculated target positions. The servo valve control device 110 obtains information indicating the actual positions of the spools of the main valves from each of the multiple hydraulic servo valves 100 via the junction box 92, and calculates the target position of the spools of the pilot valves based on the target and actual positions of the spools of the main valves and outputs the calculated target positions to the hydraulic servo valves 100, thereby feedback-controlling the position of the spools of the main valves. The servo valve control device 110 may feedback-control the position of the spools of the main valves using any method, such as P control, PI control, or PID control.

The servo valve control device 110 records input data from the engine control device 91, output data to the hydraulic servo valve 100, and various detection data indicating the status of the hydraulic servo valve 100 and the engine 80 obtained via the junction box 92 in the log data storage device 90.

In this figure, an example in which the log data storage device 90 is connected to the servo valve control device 110 is shown, but the log data storage device 90 may be connected to the junction box 92, may be connected between the junction box 92 and the hydraulic servo valve 100, or may be mounted within the hydraulic servo valve 100. When the log data storage device 90 is connected to the servo valve control device 110, the log data storage device 90 can be installed at a position relatively far from the engine 80, so that the effects of vibrations and heat generated in the engine 80 can be suppressed. In addition, the amount of wiring and other changes required when installing the log data storage device 90 in an existing ship 2 can be reduced. When the log data storage device 90 is connected to the junction box 92, even if the data indicating the actual position of the spool of the pilot valve of the hydraulic servo valve 100 is configured not to be transmitted to the servo valve control device 110, the data indicating the actual position of the spool of the pilot valve can be acquired and recorded, so that the state of the hydraulic servo valve 100 can be more accurately estimated. When the log data storage device 90 is connected between the hydraulic servo valve 100 and the junction box 92, in addition to the data indicating the actual position of the spool of the pilot valve of the hydraulic servo valve 100, the value of the voltage or current supplied to the hydraulic servo valve 100 can be obtained, so that the state of the hydraulic servo valve 100 can be estimated more accurately. The same applies when the log data storage device 90 is installed inside the hydraulic servo valve 100.

When the state estimation device 400 is installed on the ship 2, the state estimation device 400 may acquire data from the log data storage device 90, or may be installed in place of the log data storage device 90. In this case, as described above, the state estimation device 400 may be connected to the servo valve control device 110, may be connected to the junction box 92, may be connected between the junction box 92 and the hydraulic servo valve 100, or may be installed within the hydraulic servo valve 100.

Figure 3 shows a schematic configuration of the hydraulic servo valve 100. The hydraulic servo valve 100 includes a pilot valve 10 and a main valve 20. The pilot valve 10 controls the position of the spool 28 of the main valve 20 by changing the state of delivery of hydraulic oil 48 to the main valve 20, which is a controlled device, based on a command from the servo valve control device 110.

The pilot valve 10 is connected to the main valve 20 by a number of bolts B1. The pilot valve 10 is provided with a number of through holes 10h for passing the bolts B1 through. The main valve 20 is provided with a number of female threads 20h for screwing the bolts B1 into. The number of through holes 10h are arranged at positions corresponding to the positions of the number of female threads 20h. The pilot valve 10 is connected to the main valve 20 by screwing the bolts B1 into the female threads 20h through the through holes 10h. The pilot valve 10 is separated from the main valve 20 by removing the bolts B1.

The hydraulic system of the main valve 20 in FIG. 3 includes a drain tank 44 that stores hydraulic oil 48, and a hydraulic pump 42 that pressurizes and discharges the hydraulic oil 48 from the drain tank 44. The hydraulic oil 48 discharged from the hydraulic pump 42 is supplied to the inside of the main valve 20 and the pilot valve 10 through the pump side piping section 22p in the main valve 20. The hydraulic oil 48 discharged from the pilot valve 10 and the inside of the main valve 20 is returned to the drain tank 44 through the tank side piping section 22t in the main valve 20. The pump side piping section 22p and the tank side piping section 22t are collectively referred to as the main valve piping section.

The pilot valve 10 mainly includes a main body 10b, a spool 12, a port 16, and a spool drive unit 18. The spool 12 functions as a first movable unit and has a shaft 12s and a number of valve bodies 14 that move integrally with the shaft 12s. The spool 12 is driven by the spool drive unit 18 to move forward and backward in a first direction. Hereinafter, for convenience, the direction in which the spool 12 extends from the spool drive unit 18 along the first direction (downward in FIG. 1) will be referred to as the "extension direction" and "extension side", and the direction opposite to the extension direction will be referred to as the "counter-extension direction" and "counter-extension side".

A biasing member 12h is provided on the extension side of the spool 12 to bias the spool 12 in the direction opposite to the extension direction. The biasing member 12h may be, for example, a coil spring that expands and contracts in a first direction. The spool drive unit 18 includes an electromagnetic actuator (not shown), such as a coil, that moves the shaft 12s forward and backward in the first direction. The spool drive unit 18 moves the shaft 12s forward and backward based on a command from the servo valve control device 110, and controls the position of the valve body 14 by balancing with the biasing force of the biasing member 12h.

The valve body 14 includes a first valve body 14a, a second valve body 14b, and a third valve body 14c that are spaced apart in the first direction. The second valve body 14b is arranged on the opposite side of the first valve body 14a, and the third valve body 14c is arranged on the extending side of the first valve body 14a. The first valve body 14a changes the communication state of the A port 16a (described later) depending on its position in the first direction. The main body 10b has a cylindrical space 10s that extends in the first direction and houses the spool 12. The cylindrical space 10s functions as a cylinder that surrounds the valve body 14 via a narrow gap.

Ports 16 are provided in the main body 10b. In this embodiment, the ports 16 include a P port 16p, an A port 16a, and a T port 16t. The P port 16p is connected to the pump side piping 22p, and is supplied with pressurized hydraulic oil 48 from the hydraulic pump 42. The A port 16a is connected to the hydraulic oil receiving portion 22a of the main valve 20. The spool 28 of the main valve 20 is moved by the pressure of the hydraulic oil 48 supplied to the hydraulic oil receiving portion 22a. The T port 16t is connected to the tank side piping 22t, and discharges the hydraulic oil 48 that has flowed through the main body 10b to the drain tank 44 through the tank side piping 22t.

The main valve 20 includes a spool 28 that functions as a second movable part. Although details of the main valve 20 are omitted in this figure, the main valve 20 may have a structure similar to that of the pilot valve 10.

A position sensor 19 for detecting the position of the spool 12 is provided at the tip of the spool 12 of the pilot valve 10. A position sensor 29 for detecting the position of the spool 28 is provided at the tip of the spool 28 of the main valve 20. Data indicating the actual position of the spool 12 of the pilot valve 10 detected by the position sensor 19 and data indicating the actual position of the spool 28 of the main valve 20 detected by the position sensor 29 are sent to the junction box 92 via wiring.

Figure 4 is a schematic diagram showing the position of the valve body 14 of the pilot valve 10 in the first direction and the open/closed state of the port. Elements that are not important for the explanation are omitted in this figure. Figure 4 (a) shows the state in which the valve body 14 is located in the first region that connects the A port 16a and the P port 16p. In this state, the A port 16a supplies hydraulic oil 48 from the P port 16p to the hydraulic oil receiving portion 22a (hereinafter referred to as the "supply mode"). In the supply mode, the hydraulic oil 48 from the P port 16p is supplied to the hydraulic oil receiving portion 22a of the main valve 20. This operation causes, for example, the spool 28 of the main valve 20 to move in a direction that increases the amount of fuel supplied to the engine 80.

Figure 4 (b) shows the state where the valve body 14 is located in the neutral region where the A port 16a is blocked and the P port 16p and the T port 16t are not connected (hereinafter, the position in the neutral region is also referred to as the "neutral position"). In this state, the A port 16a is blocked and hydraulic oil is neither supplied to nor collected from the hydraulic oil receiving portion 22a (hereinafter, referred to as the "neutral mode"). In the neutral mode, the hydraulic oil pressure in the hydraulic oil receiving portion 22a of the main valve 20 is maintained in the state just before the valve body 14 was located in the neutral region. This operation, for example, stops the spool 28 of the main valve 20 in the previous position, and the amount of fuel supplied to the engine 80 is maintained in the previous state.

Figure 4 (c) shows the state where the valve body 14 is located in the second region that connects the A port 16a and the T port 16t. In this state, the A port 16a recovers hydraulic oil 48 from the hydraulic oil receiving section 22a and returns it to the pump side piping section 22p (hereinafter referred to as the "recovery mode"). In the recovery mode, the hydraulic oil 48 in the hydraulic oil receiving section 22a of the main valve 20 is recovered to the drain tank 44 through the A port 16a, the T port 16t, and the tank side piping section 22t. This action, for example, causes the spool 28 of the main valve 20 to move in a direction that reduces the amount of fuel supplied to the engine 80.

Figure 5 shows the configuration of the servo valve control device 110. The servo valve control device 110 includes a power supply circuit 120, a servo valve control circuit 130, a servo valve drive circuit 140, a voltage detection unit 150, and a data collection circuit 160.

The power supply circuit 120 supplies power supplied from an external power supply to the servo valve control circuit 130 and the servo valve drive circuit 140. The voltage detection unit 150 detects the voltage input to the power supply circuit 120 or the voltage output from the power supply circuit 120. Instead of or in addition to the voltage detection unit 150, a current detection unit may be provided that detects the current input to the power supply circuit 120 or the current output from the power supply circuit 120.

The servo valve control circuit 130 calculates the target position of the spool 28 of the main valve 20 based on a command from the engine control device 91. The servo valve control circuit 130 calculates the target position of the spool 12 of the pilot valve 10 to move the spool 28 of the main valve 20 to the calculated target position. The servo valve control circuit 130 inputs the calculated target position of the spool 12 of the pilot valve 10 to the servo valve drive circuit 140.

The servo valve drive circuit 140 supplies power to the coil of the spool drive unit 18 of the pilot valve 10 to move the spool 12 in accordance with the target position of the spool 12 of the pilot valve 10 input from the servo valve control circuit 130. The servo valve drive circuit 140 may be provided within the hydraulic servo valve 100.

When the spool 12 of the pilot valve 10 moves to change the open/closed state of the port 16, the spool 28 of the main valve 20 moves toward the target position by supplying or withdrawing hydraulic oil 48. When the actual position of the spool 28 of the main valve 20 matches the target position, the servo valve control circuit 130 returns the spool 12 of the pilot valve 10 to the neutral position. This causes the spool 28 of the main valve 20 to stop at the target position. This series of controls is repeated while the engine 80 is operating.

The data collection circuit 160 records data such as the voltage value detected by the voltage detection unit 150 and the target position of the spool 12 of the pilot valve 10 input from the servo valve control circuit 130 to the servo valve drive circuit 140 in the log data storage device 90. The data collection circuit 160 acquires data indicating the state of the engine 80 detected by the sensor 82, the actual position of the spool 12 of the pilot valve 10 detected by the position sensor 19, the actual position of the spool 28 of the main valve 20 detected by the position sensor 29, etc. via the junction box 92, and records them in the log data storage device 90.

Figure 6 shows the configuration of the learning device 300 and the state estimation device 400. The learning device 300 includes a learning data acquisition unit 301, an estimation model generation unit 302, and an estimation model provision unit 303. The state estimation device 400 includes a detection information acquisition unit 401, a state estimation unit 402, and an estimation result output unit 403.

The learning data acquisition unit 301 acquires learning data used to learn an estimation model for estimating the state of the hydraulic servo valve 100. The learning data includes a set of data that can be acquired in relation to the operation of the hydraulic servo valve 100 and data indicating the state of the hydraulic servo valve 100. The learning data acquisition unit 301 may acquire log data stored in the log data storage device 90, may acquire test data recorded when the hydraulic servo valve 100 is used in a test environment other than the ship 2, or may acquire simulation data generated by a simulator for simulating the operation of the hydraulic servo valve 100.

The learning data acquisition unit 301 may acquire data recorded or generated under a specific situation as the learning data. The learning data acquisition unit 301 may select data recorded or generated under a specific situation from the acquired data as the learning data. For example, the learning data acquisition unit 301 may acquire or select log data when a specific state to be estimated by the estimation model occurs, test data when a specific state occurs in a test environment, or simulation data when a specific state is simulated by a simulator. The learning data acquisition unit 301 may acquire data recorded or generated when the hydraulic servo valve 100 is operated under a specific environment as the learning data. For example, the learning data acquisition unit 301 may classify data according to the type of ship 2, route, sailing time, type of engine 80, number of cylinders, cumulative operating time, etc., and learn a separate estimation model for each of those environments.

The learning data acquisition unit 301 may generate learning data by preprocessing the acquired data. For example, the learning data acquisition unit 301 may calculate, from the acquired data, a feature quantity that is correlated with the state to be estimated by the estimation model, and use the calculated feature quantity as learning data. In addition, in order to associate the target position or actual position of the spool 12 of the pilot valve 10 and the target position or actual position of the spool 28 of the main valve 20 with other data, an offset time from when the target position is input until the actual position of the spool 12 or spool 28 reaches the target position may be adjusted.

The estimation model generating unit 302 uses the learning data acquired by the learning data acquiring unit 301 to generate an estimation criterion used to estimate the state of the hydraulic servo valve 100 in the state estimation device 400. The estimation criterion may be a table or a program that associates data that can be acquired in relation to the operation of the hydraulic servo valve 100 with data indicating the state of the hydraulic servo valve 100. The estimation criterion may be an estimation model that models the correspondence between data that can be acquired in relation to the operation of the hydraulic servo valve 100 and data indicating the state of the hydraulic servo valve 100. The estimation model may be a mathematical formula for calculating data indicating the state of the hydraulic servo valve 100 using data that can be acquired in relation to the operation of the hydraulic servo valve 100 as input variables. In this case, the estimation model generating unit 302 may generate the estimation model by a statistical method such as multivariate analysis, multiple regression analysis, or principal component analysis. The estimation model may be a neural network that inputs data indicating the state of the hydraulic servo valve 100 to an input layer and outputs data indicating the state of the hydraulic servo valve 100 from an output layer. In this case, the estimation model generating unit 302 learns the estimation model by adjusting the intermediate layer of the neural network so that when data acquired in relation to the operation of the hydraulic servo valve 100 included in the learning data is input to the input layer, a value approximating the data indicating the state of the hydraulic servo valve 100 corresponding to the data is output from the output layer. The estimation model may be a rule-based estimation algorithm or any form of artificial intelligence.

The estimation model providing unit 303 provides the estimation model generated by the estimation model generating unit 302 to the state estimation device 400. The estimation model providing unit 303 may be provided to the state estimation device 400 when the state estimation device 400 is manufactured. In addition, if the learning device 300 learns the estimation model and updates the estimation model even after the state estimation device 400 is manufactured, the estimation model providing unit 303 may provide the state estimation device 400 with the updated estimation model at a predetermined timing.

The detection information acquisition unit 401 acquires information detected in relation to the operation of the hydraulic servo valve 100. The detection information acquisition unit 401 may acquire data stored in the log data storage device 90, may acquire data directly from the data collection circuit 160, or may acquire data from the ship 2 via ship-to-shore communication. The detection information acquisition unit 401 may perform pre-processing on the acquired data in the same manner as the learning data used when the estimation model was generated.

The state estimation unit 402 estimates the state of the hydraulic servo valve 100 using the estimation model generated by the learning device 300. The state estimation unit 402 inputs the data acquired by the detection information acquisition unit 401 into the estimation model, and acquires the state of the hydraulic servo valve 100 output from the estimation model as an estimation result.

The estimation result output unit 403 outputs the estimation result by the state estimation unit 402. When the state estimation unit 402 estimates that the hydraulic servo valve 100 is in an abnormal state, the estimation result output unit 403 may notify that fact. Whether or not the hydraulic servo valve 100 is in an abnormal state may be determined based on whether or not the state estimation data indicating the state of the hydraulic servo valve 100 output by the state estimation unit 402 is equal to or greater than a predetermined threshold value or less than a predetermined threshold value, or whether or not it is within a predetermined range, or based on whether or not the amount of change from the initial value of the state estimation data is equal to or greater than a predetermined threshold value. In addition, the estimation result output unit 403 may determine whether or not the hydraulic servo valve 100 is in an abnormal state by comparing the state estimation results of multiple hydraulic servo valves 100 provided corresponding to multiple cylinders 81 of the same engine 80. For example, when the value of the state estimation result of a specific hydraulic servo valve 100 is significantly different from the average value of the state estimation results of the multiple hydraulic servo valves 100, the hydraulic servo valve 100 may be determined to be in an abnormal state. The deviation value of the state estimation results for each of the multiple hydraulic servo valves 100 may be calculated, and the hydraulic servo valves 100 whose deviation value is equal to or greater than a predetermined threshold, for example, 80 or less than 30, may be determined to be in an abnormal state.

Figure 7 is a flowchart showing the steps of the state estimation method of this embodiment. First, the steps of the stage of generating an estimation model will be described. When the ship 2 equipped with the hydraulic servo valve 100 is sailing, data related to the operation of the hydraulic servo valve 100 is recorded in the log data storage device 90 (step S10). The learning data acquisition unit 301 of the learning device 300 acquires the log data recorded in the log data storage device 90 as learning data (step S12). The learning data acquisition unit 301 also acquires test data and simulation data as learning data. The estimation model generation unit 302 generates an estimation model based on the acquired learning data (step S14). The estimation model provision unit 303 provides the generated estimation model to the state estimation device 400 (step S16).

Next, the procedure for estimating the state of the hydraulic servo valve 100 using the estimation model will be described. When the ship 2 equipped with the hydraulic servo valve 100 is sailing, data related to the operation of the hydraulic servo valve 100 is recorded in the log data storage device 90 (step S20). The detection information acquisition unit 401 of the state estimation device 400 acquires the detection information recorded in the log data storage device 90 (step S22). The state estimation unit 402 estimates the state of the hydraulic servo valve 100 using the estimation model based on the acquired detection information (step S24). The estimation result output unit 403 outputs the estimation result (step S26).

A specific example of the management system of this embodiment will be described.

[Example 1: Estimation of Internal Oil Leakage Amount]
When the pilot valve 10 is used for a long time, the valve body 14 wears and the moderation of the valve decreases, and even in the neutral mode, the A port 16a, the P port 16p, and the T port 16t communicate with each other slightly. When the hydraulic oil 48 leaks from the P port 16p to the A port 16a in the neutral mode, the hydraulic pressure of the hydraulic oil receiving section 22a gradually increases, the amount of fuel supplied to the engine 80 increases, and the fuel efficiency of the engine 80 deteriorates. Also, when the hydraulic oil 48 leaks from the A port 16a to the T port 16t in the neutral mode, the hydraulic pressure of the hydraulic oil receiving section 22a gradually decreases, the amount of fuel supplied to the engine 80 decreases, and the output of the engine 80 decreases. When the amount of leakage of the hydraulic oil 48 exceeds the allowable amount, the hydraulic servo valve 100 does not function normally and breaks down. If the amount of leakage can be estimated with high accuracy, the hydraulic servo valve 100 can be replaced or repaired before the amount of leakage exceeds the allowable amount, thereby avoiding unexpected breakdowns.

As described above, when the hydraulic servo valve 100 is operating normally, the spool 28 of the main valve 20 is stationary when the spool 12 of the pilot valve 10 is in the neutral position. However, if the amount of leakage of the hydraulic oil 48 increases, the spool 28 of the main valve 20 moves even when the spool 12 of the pilot valve 10 is in the neutral position. Experiments by the inventors have shown that there is a strong correlation between the moving speed of the spool 28 of the main valve 20 when the spool 12 of the pilot valve 10 is in the neutral position and the amount of leakage of the hydraulic oil 48. Therefore, the amount of internal oil leakage in the pilot valve 10 can be estimated from the moving speed of the spool 28 of the main valve 20 when the spool 12 of the pilot valve 10 is in the neutral position.

[Example 1-1]
8 shows the configuration of a learning device and a state estimation device according to Example 1-1. In a learning device 300 of this example, a learning data acquisition unit 301 acquires an actual pilot valve spool position 310, an actual main valve spool position 311, and an actual measured internal oil leakage amount value 312 as learning data.

The pilot valve actual spool position 310 and the main valve actual spool position 311 are time series data when the hydraulic servo valve 100 is used within a specified period before or after the internal oil leakage amount is measured. This time series data may be time series data for one to several cycles of the rotational operation of the engine 80. Since the position of the spool 12 of the pilot valve 10 is in the neutral position at least several times during one cycle of the rotational operation of the engine 80, the time series data of the pilot valve actual spool position 310 and the main valve actual spool position 311 for one cycle includes information indicating the moving speed of the spool 28 of the main valve 20 when the spool 12 of the pilot valve 10 is in the neutral position.

The measured internal oil leakage amount 312 is the value of the internal oil leakage amount in the hydraulic servo valve 100. When log data or test data is used as the learning data, the measured internal oil leakage amount 312 is the value of the internal oil leakage amount actually measured by a flow meter or the like. Since it is considered that the internal oil leakage amount does not change suddenly during the use of the hydraulic servo valve 100, it is assumed that the hydraulic servo valve 100 was used in a state in which the measured amount of hydraulic oil 48 was leaking inside during a predetermined period before or after the time when the internal oil leakage amount was actually measured, and the actual pilot valve spool position 310 and the actual main valve spool position 311 recorded during that predetermined period are associated with the measured internal oil leakage amount 312 to form the learning data. The predetermined period may be a period during which the amount of internal oil leakage remains the same during the use of the hydraulic servo valve 100, and may be determined by the use time of the hydraulic servo valve 100, the operation time of the engine 80, the rotation speed of the engine 80, etc. When simulation data is used as learning data, the actual internal oil leakage amount 312 is the value of the internal oil leakage amount input to the simulator as a simulation condition.

The internal oil leakage amount estimation model generating unit 313 uses the learning data acquired by the learning data acquiring unit 301 to generate an internal oil leakage amount estimation model used to estimate the internal oil leakage amount of the hydraulic servo valve 100 in the state estimation device 400. The internal oil leakage amount estimation model may be a neural network that inputs time series data of the pilot valve actual spool position 310 and the main valve actual spool position 311 for a predetermined period to the input layer and outputs the internal oil leakage amount of the hydraulic servo valve 100 from the output layer. In this case, the internal oil leakage amount estimation model generating unit 313 learns the internal oil leakage amount estimation model by adjusting the intermediate layer of the neural network so that when the time series data of the pilot valve actual spool position 310 and the main valve actual spool position 311 for a predetermined period is input to the input layer, a value approximating the internal oil leakage amount actual measurement value 312 is output from the output layer.

The internal oil leakage amount estimation model providing unit 314 provides the internal oil leakage amount estimation model generated by the internal oil leakage amount estimation model generating unit 313 to the state estimation device 400.

The detection information acquisition unit 401 acquires the pilot valve actual spool position 410 and the main valve actual spool position 411 as detection information. This detection information is time series data for a predetermined period of time that is the same as the learning data used to generate the internal oil leakage amount estimation model.

The internal oil leakage amount estimation unit 412 estimates the internal oil leakage amount of the hydraulic servo valve 100 using the internal oil leakage amount estimation model generated by the learning device 300. The internal oil leakage amount estimation unit 412 inputs the time series data of the pilot valve actual spool position 410 and the main valve actual spool position 411 acquired by the detection information acquisition unit 401 into the internal oil leakage amount estimation model, and acquires the internal oil leakage amount output from the internal oil leakage amount estimation model as an estimation result.

The internal oil leakage amount estimated value output unit 413 outputs an estimated value of the internal oil leakage amount estimated by the internal oil leakage amount estimation unit 412. When it is estimated that the hydraulic servo valve 100 is in an abnormal state, the internal oil leakage amount estimated value output unit 413 may notify that fact. Whether or not the hydraulic servo valve 100 is in an abnormal state may be determined based on whether or not the value of the internal oil leakage amount output by the internal oil leakage amount estimation unit 412 is equal to or greater than a predetermined threshold value, or whether or not the change amount from the initial value of the internal oil leakage amount is equal to or greater than a predetermined threshold value. In addition, the internal oil leakage amount estimated value output unit 413 may determine whether or not the hydraulic servo valve 100 is in an abnormal state by comparing the values of the internal oil leakage amounts of multiple hydraulic servo valves 100 provided corresponding to multiple cylinders 81 of the same engine 80. For example, when the value of the internal oil leakage amount of a specific hydraulic servo valve 100 is significantly different from the average value of the internal oil leakage amounts of the multiple hydraulic servo valves 100, the hydraulic servo valve 100 may be determined to be in an abnormal state. The deviation value of the internal oil leakage amount of each of the multiple hydraulic servo valves 100 may be calculated, and the hydraulic servo valves 100 whose deviation value is equal to or greater than a predetermined threshold value, for example, 80 or less than 30, may be determined to be in an abnormal state.

[Example 1-2]
Fig. 9 shows the configurations of a learning device and a state estimation device according to Example 1-2. The learning device 300 of Example 1-2 includes a feature calculation unit 315 in addition to the configuration of the learning device 300 of Example 1-1 shown in Fig. 8. Moreover, the state estimation device 400 of Example 1-2 includes a feature calculation unit 414 in addition to the configuration of the state estimation device 400 of Example 1-1 shown in Fig. 8. The following mainly describes the differences from Example 1-1. Other points are the same as those of Example 1-1.

The feature calculation unit 315 of the learning device 300 calculates a feature correlated with the amount of internal oil leakage from the pilot valve actual spool position 310 or the main valve actual spool position 311 acquired by the learning data acquisition unit 301. As described above, the inventors' experiments have revealed that the amount of internal oil leakage is correlated with the moving speed of the spool 28 of the main valve 20 when the spool 12 of the pilot valve 10 is in the neutral position. Therefore, the feature calculation unit 315 calculates the moving speed of the spool 28 of the main valve 20 from the main valve actual spool position 311 during the period when the pilot valve actual spool position 310 is in the neutral position. If there is a time lag between when the spool 12 of the pilot valve 10 moves to the neutral position and when the supply of hydraulic oil to the main valve 20 stops and the position of the spool 28 of the main valve 20 stops, the feature calculation unit 315 may adjust the time lag before calculating the moving speed of the spool 28 of the main valve 20. The amount of time lag adjustment may be determined in advance through experiments or the like depending on the type of hydraulic servo valve 100 and the environment in which the hydraulic servo valve 100 is used, such as the temperature, pressure of the hydraulic oil 48, the engine 80 RPM, load, and exhaust temperature.

The internal oil leakage amount estimation model generation unit 313 generates an internal oil leakage amount estimation model using the learning data acquired by the learning data acquisition unit 301 and the feature amount calculated by the feature amount calculation unit 315. The internal oil leakage amount estimation model generation unit 313 may generate an internal oil leakage amount estimation model using the internal oil leakage amount actual measurement value 312 and the feature amount calculated by the feature amount calculation unit 315. For example, the internal oil leakage amount estimation model may be a formula that calculates the internal oil leakage amount using the feature amount as an input variable. In this case, the internal oil leakage amount estimation model generation unit 313 may generate a formula by a statistical method such as regression analysis. The internal oil leakage amount estimation model generation unit 313 may generate an internal oil leakage amount estimation model using the pilot valve actual spool position 310, the main valve actual spool position 311, and the feature amount. For example, the internal oil leakage amount estimation model may be a neural network that inputs time series data of the pilot valve actual spool position 310 and the main valve actual spool position 311 for a predetermined period and feature values for the predetermined period into an input layer and outputs the internal oil leakage amount from an output layer. In this case, the internal oil leakage amount estimation model generation unit 313 learns the internal oil leakage amount estimation model by adjusting the intermediate layer of the neural network so that when the time series data of the pilot valve actual spool position 310 and the main valve actual spool position 311 for a predetermined period and feature values for the predetermined period are input into the input layer, a value that is close to the actual internal oil leakage amount value 312 is output from the output layer.

The feature calculation unit 414 of the state estimation device 400 calculates the feature from the pilot valve actual spool position 410 or the main valve actual spool position 411 acquired by the detection information acquisition unit 401 in the same manner as the feature calculation unit 315 calculated the feature when the internal oil leakage amount estimation model used by the internal oil leakage amount estimation unit 412 was generated. If the pilot valve actual spool position 410 and the main valve actual spool position 411 of the hydraulic servo valve 100 used in an environment different from the learning data used when the internal oil leakage amount estimation model was generated are acquired by the detection information acquisition unit 401, the feature calculation unit 414 may calculate the feature after adjusting for the difference in the environment. For example, the adjustment amount of the time lag may be changed depending on the environment. This can improve the robustness of the estimation of the internal oil leakage amount.

[Examples 1-3]
FIG. 10 shows the configuration of the learning device and state estimation device according to Example 1-3. In the learning device 300 of Example 1-2 shown in FIG. 9, the learning data acquisition unit 301 acquires the pilot valve actual spool position 310, but in the learning device 300 of Example 1-3, the learning data acquisition unit 301 acquires the pilot valve target spool position 316 instead of the pilot valve actual spool position 310. Also, in the state estimation device 400 of Example 1-2 shown in FIG. 9, the detection information acquisition unit 401 acquires the pilot valve actual spool position 410, but in the state estimation device 400 of Example 1-3, the detection information acquisition unit 401 acquires the pilot valve target spool position 415 instead of the pilot valve actual spool position 410. The following mainly describes the differences from Example 1-2. Other points are the same as those of Example 1-1 or 1-2.

As mentioned above, depending on the connection position of the log data storage device 90, it may not be possible to record the actual pilot valve spool position when the hydraulic servo valve 100 is in use. In this case, the pilot valve target spool position is used instead of the actual pilot valve spool position as the learning data and detection information. This makes it possible to generate an internal oil leakage amount estimation model using data that can be obtained regardless of the connection position of the log data storage device 90, and to estimate the internal oil leakage amount using the internal oil leakage amount estimation model. In the example of FIG. 10, a feature amount calculation unit 315 and a feature amount calculation unit 414 are provided as in Example 1-2, but a feature amount calculation unit may not be provided as in Example 1-1.

[Examples 1 to 4]
Fig. 11 shows the configuration of a learning device and a state estimation device according to Example 1-4. The learning device 300 of Example 1-4 includes an offset time calculation unit 319 in addition to the configuration of the learning device 300 of Example 1-3 shown in Fig. 10. Furthermore, in the learning device 300 of Example 1-4, the learning data acquisition unit 301 further acquires a coil drive voltage 317 and a pilot valve physical parameter 318. Furthermore, in the state estimation device 400 of Example 1-4, the detection information acquisition unit 401 further acquires a coil drive voltage 416. The following mainly describes the differences from Example 1-3. Other points are the same as those of Examples 1-1 to 1-3.

There is a time lag between when a command signal for the target position of the spool 12 is input from the servo valve control device 110 to the pilot valve 10, when a current is supplied to the coil of the spool drive unit 18, and when the spool 12 actually reaches the target position. Therefore, when the pilot valve target spool position 316 is used as learning data instead of the pilot valve actual spool position 310, by adjusting the offset time required for the actual position of the spool 12 of the pilot valve 10 to follow the pilot valve target spool position 316, an internal oil leakage amount estimation model can be generated based on the position of the spool 28 of the main valve 20 when the spool 12 of the pilot valve 10 actually follows the neutral position, thereby improving the accuracy of the internal oil leakage amount estimation model.

The offset time calculation unit 319 calculates the offset time based on the coil drive voltage 317 and the pilot valve physical parameters 318. The pilot valve physical parameters 318 include physical quantities such as the resistance value of the elements constituting the drive circuit of the spool drive unit 18, the inductance of the coil, the mass of the spool 12, and the friction coefficient between the main body 10b and the spool 12. The offset time calculation unit 319 calculates the offset time from an equation of motion including these physical parameters and the coil drive voltage 317. If the offset time may vary depending on the position, movement direction, movement speed, etc. of the spool 12 of the pilot valve 10, these data may also be used to calculate the offset time.

When the learning device 300 generates an internal oil leakage amount estimation model for estimating the internal oil leakage amount of a specific type of hydraulic servo valve 100, the pilot valve physical parameter 318 may be treated as a constant. As a result, log data in which the pilot valve physical parameter 318 is not recorded can also be used as learning data. A constant that is predetermined according to the type of hydraulic servo valve 100 may be used as the pilot valve physical parameter 318. A physical parameter measured at the time of shipment of the hydraulic servo valve 100 may be used as the pilot valve physical parameter 318. As a result, the influence of the variation in the physical parameters of the pilot valve due to the individual difference of the hydraulic servo valve 100 can be suppressed, and the offset time can be calculated more accurately. As the pilot valve physical parameter 318, a physical parameter measured at the time of maintenance such as an overhaul of the hydraulic servo valve 100 may be used. As a result, the influence of the aging of the pilot valve 10 can be suppressed, and the offset time can be calculated more accurately.

The internal oil leakage amount estimation model generation unit 313 may generate an internal oil leakage amount estimation model that inputs the pilot valve actual spool position 410 and the main valve actual spool position 411 and outputs the internal oil leakage amount, as in Example 1-1 or 1-2, or may generate an internal oil leakage amount estimation model that inputs the pilot valve target spool position 415 and the main valve actual spool position 411 and outputs the internal oil leakage amount, as in Example 1-3, or may generate an internal oil leakage amount estimation model that inputs the pilot valve target spool position 415, the main valve actual spool position 411, and the coil drive voltage 416 and outputs the internal oil leakage amount, or may generate an internal oil leakage amount estimation model that further inputs pilot valve physical parameters or offset time in addition to them and outputs the internal oil leakage amount. In the example of this figure, the internal oil leakage amount estimation model inputs the pilot valve target spool position 415, the main valve actual spool position 411, and the coil drive voltage 416 and outputs the internal oil leakage amount. The coil drive voltage 416 supplied to the coil from the power supply circuit 120 of the servo valve control device 110 can vary depending on the engine 80 rotation speed, etc., so by generating an internal oil leakage estimation model that inputs the coil drive voltage 416 in addition to the pilot valve target spool position 415 and the main valve actual spool position 411 and outputs the internal oil leakage amount, the effect of fluctuations in the offset time can be suppressed and the estimation accuracy can be improved.

The detection information acquisition unit 401 of the state estimation device 400 acquires detection information that needs to be input to the internal oil leakage amount estimation model used by the internal oil leakage amount estimation unit 412. In the example of this figure, the detection information acquisition unit 401 acquires the pilot valve target spool position 415, the main valve actual spool position 411, and the coil drive voltage 416. If it is necessary to input pilot valve physical parameters to the internal oil leakage amount estimation model, the detection information acquisition unit 401 may further acquire pilot valve physical parameters. Alternatively, the internal oil leakage amount estimation unit 412 may hold in advance physical parameters according to the type of hydraulic servo valve 100 to be estimated for the internal oil leakage amount. Also, if it is necessary to input an offset time to the internal oil leakage amount estimation model, the detection information acquisition unit 401 may further acquire the offset time. Alternatively, the state estimation device 400 may further include an offset time calculation unit that calculates the offset time based on the coil drive voltage 416, the pilot valve physical parameters, etc.

[Examples 1 to 5]
Fig. 12 shows the configurations of a learning device and a state estimation device according to Example 1-5. The learning device 300 of Example 1-5 includes a data selection unit 320 in addition to the configuration of the learning device 300 of Example 1-1 shown in Fig. 8. Furthermore, the learning device 300 of Example 1-5 includes a data selection unit 417 in addition to the configuration of the state estimation device 400 of Example 1-1 shown in Fig. 8. The following mainly describes the differences from Example 1-1. Other points are the same as those of Examples 1-1 to 1-4.

In this embodiment, in order to improve the estimation accuracy of the internal oil leakage amount estimation model, the state estimation device 400 generates the internal oil leakage amount estimation model using log data obtained when the hydraulic servo valve 100 is used in an environment similar to the environment in which the hydraulic servo valve 100 for which the internal oil leakage amount is to be estimated is used. This makes it possible to reduce the influence of the environment and improve the robustness of the estimation of the internal oil leakage amount.

The data selection unit 320 selects the learning data recorded when the hydraulic servo valve 100 is used in a specific environment from the learning data acquired by the learning data acquisition unit 301. The data selection unit 320 may select the learning data recorded when the hydraulic servo valve 100 is used in a specific environment by referring to data indicating the usage environment of the hydraulic servo valve 100 included in the log data, test data, simulation data, etc., such as the exhaust temperature and pressure of the cylinder 81 of the engine 80. The data selection unit 320 may classify the learning data by environment. In this case, an internal oil leakage amount estimation model may be generated for each environment. The learning data such as the log data, test data, and simulation data may be recorded or generated in a specific environment, or may be classified by usage environment when recorded or generated. In this case, the learning data acquisition unit 301 only needs to acquire the learning data recorded, generated, or classified in a specific environment, so the data selection unit 320 may not be provided.

The data selection unit 417 selects detection information recorded when the hydraulic servo valve 100 was used in a specific environment from the detection information acquired by the detection information acquisition unit 401. The data selection unit 417 may select detection information recorded in an environment similar to the environment in which the learning data used when the internal oil leakage amount estimation model used by the internal oil leakage amount estimation unit 412 was generated was recorded. When multiple internal oil leakage amount estimation models generated for each usage environment are stored in the state estimation device 400, the internal oil leakage amount estimation unit 412 may select the internal oil leakage amount estimation model to be used by referring to data indicating the usage environment of the hydraulic servo valve 100 contained in the detection information. In this case, the data selection unit 417 may not be provided.

The internal oil leakage amount estimation model may be generated using learning data recorded when the hydraulic servo valve 100 is operated in a test mode in which the spool 12 of the pilot valve 10 is operated in a specific pattern. In this case, the state estimation device 400 estimates the internal oil leakage amount using the detection information recorded when the hydraulic servo valve 100 is operated in the test mode. This can improve the robustness of the estimation of the internal oil leakage amount.

In order to further reduce the influence of the operating environment, log data, test data, simulation data, etc. recorded when the engine 80 is stopped may be used as learning data. While the engine 80 is stopped, the engine control device 91 issues a stop command to the servo valve control device 110. However, when the hydraulic pump 42 pressurizes the hydraulic oil 48, the servo valve control device 110 repeatedly moves the spool 12 of the pilot valve 10 in a small amplitude pattern such that fuel is not supplied to the engine 80 in order to prevent the spool 12 of the pilot valve 10 and the spool 28 of the main valve 20 from sticking. The learning data recorded during this sticking prevention operation is used to generate an internal oil leakage amount estimation model, and when the state estimation device 400 estimates the internal oil leakage amount of the hydraulic servo valve 100, the log data recorded during the sticking prevention operation is also used to reduce the influence of disturbances such as vibrations accompanying the operation of the engine 80, thereby improving the estimation accuracy. The sticking prevention operation pattern may be a sine wave pattern, a sawtooth wave pattern, a dither pattern, etc. When anti-sticking operation is performed using multiple different patterns, learning data and detection information recorded when anti-sticking operation is performed using a specific pattern may be used.

[Modification of the first embodiment]
In addition to the data used as the learning data and detection information in Examples 1-1 to 1-5, any of other data such as engine speed, engine load, hydraulic oil pressure, hydraulic oil temperature, or any combination thereof may be used as the learning data and detection information. Of these data, data that may affect data such as the pilot valve actual spool position, the pilot valve target spool position, and the main valve actual spool position as an environmental factor may be selected and used. By generating an internal oil leakage amount estimation model that reflects the influence of these data, the robustness of the estimation of the internal oil leakage amount can be further improved.

The features described in the above Examples 1-1 to 1-5 may be applied in any combination.

[Example 2: Determining a Power Supply Failure]
In the second embodiment, a technique for determining failure of the power supply circuit 120 and the power supply that supplies power to the power supply circuit 120 will be described. If the power supply or the power supply circuit 120 fails, the hydraulic servo valve 100 will not operate normally, causing abnormalities such as a drop in the output of the engine 80. If a failure of the power supply or the power supply circuit 120 can be predicted with high accuracy, the power supply or the power supply circuit 120 can be replaced or repaired before the power supply or the power supply circuit 120 fails, thereby avoiding unexpected abnormalities.

The hydraulic servo valve 100 for controlling the fuel supply and exhaust to the engine 80 repeats a certain operation while the engine 80 is running, so a certain pattern is observed between the operating waveform of the spool 12 of the pilot valve 10 or the spool 28 of the main valve 20 and the power supply voltage waveform. Therefore, by modeling the difference between the operating waveform and power supply voltage waveform pattern under normal conditions and the operating waveform and power supply voltage waveform pattern when symptoms of a failure are present, a failure in the power supply or power supply circuit 120 can be determined.

Figure 13 shows the configuration of a learning device and a state estimation device according to Example 2. In the learning device 300 of this example, the learning data acquisition unit 301 acquires the power supply voltage 321, the pilot valve actual spool position 310, and the power supply failure symptom index value 322 as learning data.

The pilot valve actual spool position 310 is time series data when the hydraulic servo valve 100 is used within a specified period before or after the power supply failure symptom index value 322 is actually measured. The power supply voltage 321 is time series data of the power supply voltage detected by the voltage detection unit 150 when the pilot valve actual spool position 310 is recorded. The power supply voltage 321 may be the input voltage to the power supply circuit 120, the output voltage from the power supply circuit 120, or both. These time series data may be time series data for one to several cycles of the rotational operation of the engine 80.

The power supply failure symptom index value 322 is calculated based on the cumulative operating time of the power supply circuit 120, the deterioration state of elements such as the capacitors that make up the power supply circuit 120, and the actual measured value of the emission noise emitted from the power supply circuit 120. The method of calculating the power supply failure symptom index value 322 may be determined by experiments, field tests, etc.

When log data or test data is used as the learning data, the power supply failure symptom index value 322 is calculated based on the actual measured values of the above variables. Since the power supply failure symptom index value 322 is not considered to change rapidly during use of the hydraulic servo valve 100, it is assumed that the power supply failure symptom index value 322 does not change during a predetermined period before or after the time when the variables for calculating the power supply failure symptom index value 322 are actually measured, and the power supply failure symptom index value 322 is associated with the power supply voltage 321 and the pilot valve actual spool position 310 recorded during the predetermined period and used as learning data. The predetermined period may be a period during which the power supply failure symptom index value 322 remains roughly the same during use of the hydraulic servo valve 100, and may be determined by the use time of the hydraulic servo valve 100, the operation time of the engine 80, the rotation speed of the engine 80, etc. When simulation data is used as the learning data, the power supply failure symptom index value 322 is calculated based on the values of the variables input to the simulator as simulation conditions.

The power supply failure judgment model generating unit 323 uses the learning data acquired by the learning data acquiring unit 301 to generate a power supply failure judgment model used to judge a failure of the power supply or the power supply circuit 120 in the state estimation device 400. The power supply failure judgment model may be a neural network that inputs time series data of the power supply voltage 321 and the pilot valve actual spool position 310 for a predetermined period to an input layer and outputs a power supply failure symptom index value from an output layer. In this case, the power supply failure judgment model generating unit 323 learns the power supply failure judgment model by adjusting the intermediate layer of the neural network so that when time series data of the power supply voltage 321 and the pilot valve actual spool position 310 for a predetermined period is input to the input layer, a value approximating the power supply failure symptom index value 322 is output from the output layer.

The power supply failure judgment model providing unit 324 provides the power supply failure judgment model generated by the power supply failure judgment model generating unit 323 to the state estimation device 400.

The detection information acquisition unit 401 acquires the power supply voltage 420 and the pilot valve actual spool position 410 as detection information. This detection information is time series data for a predetermined period of time that is the same as the learning data used to generate the power supply failure judgment model.

The power supply failure judgment unit 421 judges the failure of the power supply or the power supply circuit 120 of the hydraulic servo valve 100 using the power supply failure judgment model generated by the learning device 300. The power supply failure judgment unit 421 inputs the time series data of the power supply voltage 420 and the pilot valve actual spool position 410 acquired by the detection information acquisition unit 401 into the power supply failure judgment model, and acquires the power supply failure symptom index value output from the power supply failure judgment model as a judgment result. The power supply failure judgment unit 421 may judge the presence or absence of a failure of the power supply or the power supply circuit 120 based on the output power supply failure symptom index value. For example, the power supply failure judgment unit 421 may judge that the power supply or the power supply circuit 120 is faulty when the output power supply failure symptom index value is equal to or greater than a predetermined threshold value, or may judge that the power supply or the power supply circuit 120 is faulty when the amount of change from the initial value of the power supply failure symptom index value becomes equal to or greater than a predetermined threshold value. The power supply failure judgment unit 421 may estimate the period until the power supply or the power supply circuit 120 fails, i.e., the life, based on the output power supply failure symptom index value. For example, the power supply failure determination unit 421 may estimate the lifespan using a formula or the like that uses the output power supply failure symptom index value as a variable. The power supply failure determination model may be configured to output the presence or absence of a failure in the power supply or power supply circuit 120, or the lifespan of the power supply or power supply circuit 120, instead of or in addition to the power supply failure symptom index value.

When a power supply circuit 120 is provided for each of a plurality of hydraulic servo valves 100 provided corresponding to a plurality of cylinders 81 of the same engine 80, the power supply failure determination unit 421 may determine whether the power supply circuit 120 has failed by comparing the power supply failure symptom index values of each hydraulic servo valve 100. For example, when the power supply failure symptom index value of a specific hydraulic servo valve 100 is significantly different from the average value of the power supply failure symptom index values of the plurality of hydraulic servo valves 100, it may be determined that the power supply circuit 120 of that hydraulic servo valve 100 has failed. The deviation value of the power supply failure symptom index value of each of the plurality of hydraulic servo valves 100 may be calculated, and the power supply circuit 120 of the hydraulic servo valve 100 whose deviation value is a predetermined threshold value, for example, greater than or equal to 80 or less than 30, may be determined to have failed.

The power supply failure judgment result output unit 422 outputs the result judged by the power supply failure judgment unit 421. When the power supply failure judgment unit 421 judges that the power supply or the power supply circuit 120 of the hydraulic servo valve 100 has failed, the power supply failure judgment result output unit 422 may notify this fact. The power supply failure judgment result output unit 422 may output the life of the power supply or the power supply circuit 120 judged by the power supply failure judgment unit 421.

The inventors' experiments have shown that the amount of drop in the power supply voltage relative to the peak value of the load current is correlated with a failure of the power supply or power supply circuit 120. Therefore, if an ammeter is provided to detect the value of the current supplied to the hydraulic servo valve 100, the current value measured by the ammeter may be used as learning data and detection information instead of or in addition to the pilot valve actual spool position 310. This can improve the accuracy of failure determination.

The amount of drop in the power supply voltage relative to the peak value of the load current may be calculated from the power supply voltage and the actual pilot valve spool position. In this case, similar to Example 1-2, the learning device 300 may further include a feature calculation unit 315, and the state estimation device 400 may further include a feature calculation unit 414.

The feature calculation unit 315 of the learning device 300 calculates the amount of drop in the power supply voltage relative to the peak value of the load current as a feature correlated with a failure of the power supply or the power supply circuit 120 from the power supply voltage 321 and the pilot valve actual spool position 310 acquired by the learning data acquisition unit 301. The feature calculation unit 315 may calculate the value of the current supplied to the hydraulic servo valve 100 based on the acceleration or speed of the spool 12 of the pilot valve 10 calculated from the fluctuation of the pilot valve actual spool position 310. The feature calculation unit 315 may calculate the amount of drop in the power supply voltage based on the value of the power supply voltage 321 when the calculated current value peaks. If there is a time lag between when power is supplied from the power supply circuit 120 to the coil of the spool drive unit 18 of the pilot valve 10 and when the spool 12 of the pilot valve 10 moves, the feature calculation unit 315 may adjust the time lag and then calculate the acceleration or speed of the spool 12 of the pilot valve 10. The amount of time lag adjustment may be determined in advance through experiments or the like depending on the type of hydraulic servo valve 100 and the environment in which the hydraulic servo valve 100 is used, such as the temperature, pressure of the hydraulic oil 48, the engine 80 RPM, load, and exhaust temperature.

The power supply failure judgment model generation unit 323 generates a power supply failure judgment estimation model using the learning data acquired by the learning data acquisition unit 301 and the feature amount calculated by the feature amount calculation unit 315. The power supply failure judgment model generation unit 323 may generate a power supply failure judgment model using the power supply failure symptom index value 322 and the feature amount calculated by the feature amount calculation unit 315. For example, the power supply failure judgment model may be a formula that calculates the power supply failure symptom index value using the feature amount as an input variable. In this case, the power supply failure judgment model generation unit 323 may generate the formula by a statistical method such as regression analysis. The power supply failure judgment model generation unit 323 may generate a power supply failure judgment model using the power supply voltage 321, the pilot valve actual spool position 310, and the feature amount. For example, the power supply failure judgment model may be a neural network that inputs time series data of the power supply voltage 321 and the pilot valve actual spool position 310 for a predetermined period and the feature amount for the predetermined period into an input layer and outputs the power supply failure symptom index value from an output layer. In this case, the power supply failure judgment model generation unit 323 learns the power supply failure judgment model by adjusting the intermediate layer of the neural network so that when the time series data for a specified period of time of the power supply voltage 321 and the pilot valve actual spool position 310 and the feature values for that specified period are input to the input layer, a value approximating the power supply failure symptom index value 322 is output from the output layer.

The feature amount calculation unit 414 of the state estimation device 400 calculates the feature amount from the power supply voltage 420 and the pilot valve actual spool position 410 acquired by the detection information acquisition unit 401 in the same manner as the feature amount calculation unit 315 calculated the feature amount when the power supply failure judgment model used by the power supply failure judgment unit 421 was generated. If the power supply voltage 420 and the pilot valve actual spool position 410 of the hydraulic servo valve 100 used in an environment different from the learning data used when the power supply failure judgment model was generated are acquired by the detection information acquisition unit 401, the feature amount calculation unit 414 may calculate the feature amount after adjusting for the difference in the environment. For example, the adjustment amount of the time lag may be changed depending on the environment. This can improve the robustness of the power supply failure judgment. In addition, since the amount of drop in the power supply voltage relative to the peak value of the load current can be calculated as a feature amount without implementing an ammeter, it is possible to improve the efficiency of the circuit and reduce costs.

As the characteristic quantity, a physical quantity that represents the characteristics of the operation of the spool 12 of the pilot valve 10 may be calculated. For example, the acceleration, speed, amount of overshoot, amount of delay, etc. of the spool 12 of the pilot valve 10 may be calculated as the characteristic quantity.

As the learning data and the detection information, the pilot valve target spool position or the main valve actual spool position may be used instead of or in addition to the pilot valve actual spool position. In this case, similar to Example 1-4, the learning device 300 may be provided with an offset time calculation unit 319 to adjust the offset time required for the actual spool position of the pilot valve 10 to follow the target spool position of the pilot valve. The offset time calculation unit 319 may calculate the offset time between the fluctuation of the current supplied to the hydraulic servo valve 100 and the fluctuation of the power supply voltage in order to calculate the amount of drop in the power supply voltage when the value of the current supplied to the hydraulic servo valve 100 peaks. The method of calculating the offset time may be the same as in Example 1-2.

As the learning data and the detection information, data recorded when the hydraulic servo valve 100 is used in a specific environment may be used. In this case, similar to the first to fifth embodiments, the learning device 300 may further include a data selection unit 320, and the state estimation device 400 may further include a data selection unit 417. The data selection unit 320 and the data selection unit 417 may select data recorded when the hydraulic servo valve 100 is operated in a test mode in which the spool 12 of the pilot valve 10 is operated in a specific pattern. This can improve the robustness of the fault determination. The data selection unit 320 and the data selection unit 417 may select data recorded when the engine 80 is stopped and the spool is prevented from sticking. This can reduce the influence of disturbances and improve the accuracy of the fault determination.

[Example 3: Detection of foreign object entrapment]
In the third embodiment, a technique for detecting that a foreign object has become caught between the main body 10b and the spool 12 of the pilot valve 10 will be described. When a foreign object becomes caught between the main body 10b and the spool 12, the spool 12 of the pilot valve 10 will not operate normally, causing the actual spool position of the main valve 20 to deviate from the target spool position and triggering an alarm. Conventionally, it was not possible to identify the cause of this alarm, but if it is possible to detect that a foreign object has become caught with high accuracy, it is possible to identify the cause of the alarm when the alarm is triggered and take appropriate measures promptly.

Figure 14 shows a schematic diagram of the operation of the spool 12 of the pilot valve 10 and the spool 28 of the main valve 20. Figure 14 (a) shows an example of the operation when the hydraulic servo valve 100 is normally controlled. When the target spool position of the main valve 20 is input at time t1, a positive value is input as the target spool position of the pilot valve 10 at time t2 to move the spool 28 of the main valve 20 in the positive direction. When the spool 12 of the pilot valve 10 follows the target spool position, hydraulic oil is supplied to the main valve 20, and the spool 28 of the main valve 20 starts to move in the positive direction at time t3. When the target spool position of the main valve 20 is fixed to a constant value at time t4 and the actual spool position of the main valve 20 follows the target spool position, a neutral position is input as the target spool position of the pilot valve 10, and the spool 28 of the main valve 20 stops at the target spool position at time t5.

Figure 14 (b) shows an example of operation when a foreign object is caught in the pilot valve 10. The operation up to time t4 is the same as in Figure 14 (a), but assume that a foreign object is caught in the pilot valve 10 at time t4. The neutral position is input as the target spool position of the pilot valve 10, but since the foreign object is caught and the spool 12 cannot move, hydraulic oil continues to be supplied to the main valve 20, and the spool 28 of the main valve 20 continues to move in the positive direction. Since the feedback signal of the main valve actual spool position deviates from the target position, a negative value continues to be input as the target spool position of the pilot valve 10 to return the spool 28 of the main valve 20 to the target position. When the spool 28 of the main valve 20 reaches the limit position in the positive direction, it does not move any further and stops at the limit position.

In this way, the operating patterns of the spool 12 of the pilot valve 10 and the spool 28 of the main valve 20 are significantly different between normal and when a foreign object is caught. Therefore, by modeling the difference between the operating pattern in normal and when a foreign object is caught, it is possible to detect the ingestion of a foreign object. Since the operating pattern may differ depending on the type, size, amount, etc. of the foreign object, it is possible to detect not only the presence or absence of the ingestion of a foreign object, but also the type, size, amount, etc. of the ingestion of the foreign object. Even if the foreign object is naturally removed after being caught once, the history of the ingestion of the foreign object can be detected from the log data, so it can be used as evidence of the cause of the malfunction. In a ship 2, etc., the hydraulic oil is circulated not only through the hydraulic servo valve 100 for controlling the engine 80, but also through hydraulic servo valves for controlling other devices, etc., so even if a foreign object once caught in the hydraulic servo valve 100 is naturally removed, the foreign object may be caught in another hydraulic servo valve, etc. The technology of this embodiment can detect the presence of foreign matter in the hydraulic servo valve 100 and perform appropriate maintenance such as changing the hydraulic oil, thereby preventing foreign matter from becoming caught in other devices.

Figure 15 shows the configuration of a learning device and a state estimation device according to Example 3. In the learning device 300 of this example, the learning data acquisition unit 301 acquires the main valve actual spool position 311, the main valve target spool position 330, the pilot valve target spool position 316, and foreign object data 331 as learning data.

The main valve actual spool position 311, the main valve target spool position 330, and the pilot valve target spool position 316 are time series data when the hydraulic servo valve 100 is used within a specified period before or after the time when the foreign object data 331 is actually measured. These time series data may be time series data for one to several cycles of the rotational operation of the engine 80.

Foreign object data 331 is data that indicates the presence, type, size, amount, etc. of a foreign object caught in the pilot valve 10. When log data is used as the learning data, foreign object data 331 is the actual measurement value of a foreign object caught in the pilot valve 10. When test data is used as the learning data, foreign object data 331 is the actual measurement value of a foreign object caught in the pilot valve 10. When simulation data is used as the learning data, foreign object data 331 is data on a foreign object input to the simulator as a simulation condition.

The foreign object detection model generating unit 333 uses the learning data acquired by the learning data acquiring unit 301 to generate a foreign object detection model used to detect a foreign object caught in the state estimation device 400. The foreign object detection model may be a neural network that inputs time series data of the main valve actual spool position 311, the main valve target spool position 330, and the pilot valve target spool position 316 for a predetermined period to an input layer and outputs foreign object data from an output layer. In this case, the foreign object detection model generating unit 333 learns the foreign object detection model by adjusting the intermediate layer of the neural network so that a value approximating the foreign object data 331 is output from the output layer when the time series data of the main valve actual spool position 311, the main valve target spool position 330, and the pilot valve target spool position 316 for a predetermined period is input to the input layer.

The foreign object detection model providing unit 334 provides the foreign object detection model generated by the foreign object detection model generating unit 333 to the state estimation device 400.

The detection information acquisition unit 401 acquires the main valve actual spool position 411, the main valve target spool position 430, and the pilot valve target spool position 415 as detection information. This detection information is time-series data for the same predetermined period as the learning data used to generate the foreign object detection model.

The foreign object detection unit 431 detects the presence of a foreign object in the pilot valve 10 using the foreign object detection model generated by the learning device 300. The foreign object detection unit 431 inputs the time series data of the main valve actual spool position 411, the main valve target spool position 430, and the pilot valve target spool position 415 acquired by the detection information acquisition unit 401 into the foreign object detection model, and acquires the foreign object data output from the foreign object detection model as a judgment result. The foreign object detection unit 431 may estimate the period until a foreign object is caught in the pilot valve 10 based on the output foreign object data. In this case, the period until a foreign object is caught may be estimated by comparing the foreign object data of multiple hydraulic servo valves 100 provided corresponding to multiple cylinders 81 of the same engine 80.

The foreign object detection result output unit 432 outputs the result determined by the foreign object detection unit 431. When the foreign object detection unit 431 detects that a foreign object is caught in the pilot valve 10, the foreign object detection result output unit 432 may notify that fact. The foreign object detection result output unit 432 may output the period until the foreign object is caught, which is estimated by the foreign object detection unit 431.

A feature value correlated with the presence or absence of a foreign object trapped, or the type, size, or amount of trapped foreign object may be calculated from the actual main valve spool position, the target main valve spool position, or the target pilot valve spool position. In this case, as in the first and second embodiments, the learning device 300 may further include a feature value calculation unit 315, and the state estimation device 400 may further include a feature value calculation unit 414. If there is a time lag between when the power supply circuit 120 supplies power to the coil of the spool drive unit 18 of the pilot valve 10 and when the spool 12 of the pilot valve 10 moves and when the actual main valve spool position follows the target main valve spool position, the feature value calculation unit 315 may adjust the time lag before calculating the feature value. The amount of adjustment of the time lag may be determined in advance by experiments or the like depending on the type of hydraulic servo valve 100, the temperature, the pressure of the hydraulic oil 48, the engine 80 rotation speed, the load, the exhaust temperature, and other environmental factors in which the hydraulic servo valve 100 is used.

As the learning data and the detection information, the pilot valve actual spool position may be used instead of or in addition to the pilot valve target spool position. As the learning data and the detection information, a combination of any two or more of the pilot valve target spool position, the pilot valve actual spool position, the main valve target spool position, and the main valve actual spool position may be used.

As the learning data and the detection information, data recorded when the hydraulic servo valve 100 is used in a specific environment may be used. In this case, similar to the first to fifth embodiments, the learning device 300 may further include a data selection unit 320, and the state estimation device 400 may further include a data selection unit 417. The data selection unit 320 and the data selection unit 417 may select data recorded when the hydraulic servo valve 100 is operated in a test mode in which the spool 12 of the pilot valve 10 is operated in a specific pattern. This can improve the robustness of the detection of foreign object entrapment. The data selection unit 320 and the data selection unit 417 may select data recorded when the engine 80 is stopped and the spool is being prevented from sticking. This can reduce the influence of disturbances and improve the detection accuracy of foreign object entrapment.

Example 4: Estimation of hydraulic oil cleanliness
In the fourth embodiment, a technique for estimating the cleanliness of hydraulic oil will be described. If foreign matter is mixed into the hydraulic oil during operation of the hydraulic servo valve 100 and the cleanliness of the hydraulic oil is reduced, the mixed foreign matter may hinder the operation of the spool 12 of the pilot valve 10 or reduce the pressure of the hydraulic oil supplied to the main valve 20, causing the engine 80 or other controlled objects to malfunction. In addition, the possibility of foreign matter getting caught in the spool 12 of the pilot valve 10 increases. By accurately estimating the cleanliness of the hydraulic oil and quickly taking appropriate measures when the cleanliness of the hydraulic oil is reduced, the hydraulic servo valve 100 and the controlled objects can be maintained in good condition and the ingestion of foreign matter can be prevented.

If the cleanliness of the hydraulic oil is reduced due to contamination by minute foreign matter, the frictional force generated between the spool 12 and the main body 10b of the pilot valve 10 increases when the spool 12 slides, causing the behavior of the spool 12 to change. Therefore, the cleanliness of the hydraulic oil can be estimated by modeling the difference due to the cleanliness of the hydraulic oil in the waveform of the current supplied to the coil of the spool drive unit 18 of the pilot valve 10 and the operating waveform patterns of the target spool position and actual spool position of the pilot valve 10.

Figure 16 shows the configuration of a learning device and a state estimation device according to Example 4. In the learning device 300 of this example, the learning data acquisition unit 301 acquires the coil applied current 341, the pilot valve target spool position 316, the pilot valve actual spool position 310, and the actual measured value of hydraulic oil cleanliness 342 as learning data.

The coil applied current 341, the pilot valve target spool position 316, and the pilot valve actual spool position 310 are time series data when the hydraulic servo valve 100 is used within a specified period before or after the actual measurement of the hydraulic oil cleanliness value 342. These time series data may be time series data for one to several cycles of the rotational operation of the engine 80.

The measured value of hydraulic oil cleanliness 342 is data representing the number, amount, size, etc. of foreign matter mixed into the hydraulic oil. When log data or test data is used as learning data, the measured value of hydraulic oil cleanliness 342 is an actual measurement value obtained by measuring the cleanliness of the hydraulic oil used in the hydraulic servo valve 100 using a general liquid-borne particle measuring device. The cleanliness of the hydraulic oil may be measured, for example, by a gravimetric method, a microscopic method, a light scattering method, a light blocking method, an electrical resistance method, an acoustic method, a dynamic light scattering method, etc. Since the cleanliness of the hydraulic oil is not considered to change rapidly during the use of the hydraulic servo valve 100, it is assumed that the hydraulic oil of the measured cleanliness was used during a predetermined period before or after the time when the cleanliness of the hydraulic oil was actually measured, and the coil applied current 341, the pilot valve target spool position 316, and the pilot valve actual spool position 310 recorded during that predetermined period are associated with the measured value of hydraulic oil cleanliness 342 to form learning data. The specified period may be any period during which the cleanliness of the hydraulic oil remains the same while the hydraulic servo valve 100 is in use, and may be determined by the usage time of the hydraulic servo valve 100, the operating time of the engine 80, the rotation speed of the engine 80, etc. When simulation data is used as the learning data, the measured value of cleanliness of the hydraulic oil 342 is the value of cleanliness of the hydraulic oil input to the simulator as a simulation condition.

The hydraulic oil cleanliness estimation model generating unit 343 uses the learning data acquired by the learning data acquiring unit 301 to generate a hydraulic oil cleanliness estimation model used to estimate the cleanliness of the hydraulic oil used in the state estimation device 400. The hydraulic oil cleanliness estimation model may be a neural network that inputs time series data of the coil applied current 341, the pilot valve target spool position 316, and the pilot valve actual spool position 310 for a predetermined period into the input layer and outputs the hydraulic oil cleanliness from the output layer. In this case, the hydraulic oil cleanliness estimation model generating unit 343 learns the hydraulic oil cleanliness estimation model by adjusting the intermediate layer of the neural network so that a value approximating the hydraulic oil cleanliness actual measurement value 342 is output from the output layer when the time series data of the coil applied current 341, the pilot valve target spool position 316, and the pilot valve actual spool position 310 for a predetermined period are input into the input layer.

The hydraulic oil cleanliness estimation model providing unit 344 provides the hydraulic oil cleanliness estimation model generated by the hydraulic oil cleanliness estimation model generating unit 343 to the state estimation device 400.

The detection information acquisition unit 401 acquires the coil applied current 441, the pilot valve target spool position 415, and the pilot valve actual spool position 410 as detection information. This detection information is time series data for a predetermined period of time that is the same as the learning data used to generate the hydraulic oil cleanliness estimation model.

The hydraulic oil cleanliness estimation unit 442 estimates the cleanliness of the hydraulic oil using the hydraulic oil cleanliness estimation model generated by the learning device 300. The hydraulic oil cleanliness estimation unit 442 inputs the time series data of the coil applied current 441, the pilot valve target spool position 415, and the pilot valve actual spool position 410 acquired by the detection information acquisition unit 401 into the hydraulic oil cleanliness estimation model, and acquires the hydraulic oil cleanliness output from the hydraulic oil cleanliness estimation model as an estimation result.

The hydraulic oil cleanliness estimation result output unit 443 outputs the result estimated by the hydraulic oil cleanliness estimation unit 442. The hydraulic oil cleanliness estimation result output unit 443 may determine whether or not maintenance such as cleaning or replacing the hydraulic oil is necessary based on the output hydraulic oil cleanliness, and may notify the user when it is determined that maintenance of the hydraulic oil is necessary. For example, the hydraulic oil cleanliness estimation result output unit 443 may determine that maintenance of the hydraulic oil is necessary when the outputted hydraulic oil cleanliness is less than a predetermined threshold, or may determine that maintenance of the hydraulic oil is necessary when the amount of change from the initial value of the hydraulic oil cleanliness becomes equal to or greater than a predetermined threshold. The hydraulic oil cleanliness estimation result output unit 443 may estimate and output the period until maintenance of the hydraulic oil is required. For example, the hydraulic oil cleanliness estimation result output unit 443 may estimate the period until maintenance of the hydraulic oil is required using a formula or the like that uses the outputted hydraulic oil cleanliness as a variable. The hydraulic oil cleanliness estimation model may be configured to output, instead of or in addition to the hydraulic oil cleanliness, whether or not hydraulic oil maintenance is required, or the period until hydraulic oil maintenance is required.

When multiple hydraulic servo valves 100 are provided corresponding to multiple cylinders 81 of the same engine 80, the hydraulic oil cleanliness estimation unit 442 may determine whether or not maintenance of the hydraulic oil is required based on the hydraulic oil cleanliness estimated in each hydraulic servo valve 100. When a common hydraulic oil is circulated and used in multiple hydraulic servo valves 100, if the cleanliness of the hydraulic oil estimated in any one of the hydraulic servo valves 100 is less than a predetermined threshold, even if the cleanliness of the hydraulic oil estimated in another hydraulic servo valve 100 is equal to or greater than the threshold, the hydraulic oil with a low cleanliness may be circulated and affect the operation of the other hydraulic servo valve 100. Therefore, the hydraulic oil cleanliness estimation result output unit 443 may determine that maintenance of the hydraulic oil is required when the cleanliness of the hydraulic oil estimated in at least one hydraulic servo valve 100 is less than a predetermined threshold. The hydraulic oil cleanliness estimation result output unit 443 may further determine whether or not maintenance of the hydraulic oil is required based on the order in which the hydraulic oil circulates. For example, the threshold value for determining whether or not hydraulic oil maintenance is required in the upstream hydraulic servo valve 100 may be set higher than the threshold value for determining whether or not hydraulic oil maintenance is required in the downstream hydraulic servo valve 100. The hydraulic oil cleanliness estimation result output unit 443 may calculate a statistical value such as the average value of the hydraulic oil cleanliness estimated in multiple hydraulic servo valves 100, and determine whether or not hydraulic oil maintenance is required based on the calculated statistical value.

Feature quantities correlated with the cleanliness of the hydraulic oil may be calculated from the coil applied current, the pilot valve target spool position, the pilot valve actual spool position, and the like. In this case, similar to Example 1-2, the learning device 300 may further include a feature quantity calculation unit 315, and the state estimation device 400 may further include a feature quantity calculation unit 414. When the cleanliness of the hydraulic oil decreases and the frictional force between the spool 12 of the pilot valve 10 and the main body portion 10b increases, it is considered that the speed or acceleration of the spool 12 decreases. Therefore, the feature quantity calculation unit may calculate the speed or acceleration of the spool 12 of the pilot valve 10 by calculating the rate of change or the rate of change of the rate of change of the pilot valve actual spool position.

The hydraulic oil cleanliness estimation model generation unit 343 generates a hydraulic oil cleanliness estimation model using the learning data acquired by the learning data acquisition unit 301 and the feature amount calculated by the feature amount calculation unit 315. The hydraulic oil cleanliness estimation model generation unit 343 may generate a hydraulic oil cleanliness estimation model using the hydraulic oil cleanliness actual measurement value 342 and the feature amount calculated by the feature amount calculation unit 315. For example, the hydraulic oil cleanliness estimation model may be a formula that calculates the hydraulic oil cleanliness using the feature amount as an input variable. In this case, the hydraulic oil cleanliness estimation model generation unit 343 may generate a formula using a statistical method such as regression analysis. The hydraulic oil cleanliness estimation model may be a formula that calculates the hydraulic oil cleanliness using the speed or acceleration of the spool 12 when the coil applied current or the pilot valve target spool position satisfies a predetermined condition as an input variable. For example, the speed or acceleration of the spool 12 at the peak value of the coil applied current may be used as the input variable, or the speed or acceleration of the spool 12 when the spool 12 is moved from the neutral position may be used as the input variable. The hydraulic oil cleanliness estimation model generating unit 343 may generate the hydraulic oil cleanliness estimation model using the coil applied current 341, the pilot valve target spool position 316, the pilot valve actual spool position 310, and the feature amount. For example, the hydraulic oil cleanliness estimation model may be a neural network that inputs the time series data of the coil applied current 341, the pilot valve target spool position 316, and the pilot valve actual spool position 310 for a predetermined period and the feature amount for the predetermined period into an input layer and outputs the hydraulic oil cleanliness from an output layer. In this case, the hydraulic oil cleanliness estimation model generation unit 343 learns the hydraulic oil cleanliness estimation model by adjusting the intermediate layer of the neural network so that when the time series data for a specified period of time of the coil applied current 341, the pilot valve target spool position 316, and the pilot valve actual spool position 310 and the feature values for that specified period are input to the input layer, a value approximating the actual measured hydraulic oil cleanliness value 342 is output from the output layer.

The feature calculation unit 414 of the state estimation device 400 calculates the feature from the coil applied current 441, the pilot valve target spool position 415, or the pilot valve actual spool position 410 acquired by the detection information acquisition unit 401 in the same manner as the feature calculation unit 315 calculated the feature when the hydraulic oil cleanliness estimation model used by the hydraulic oil cleanliness estimation unit 442 was generated. If the coil applied current 441, the pilot valve target spool position 415, or the pilot valve actual spool position 410 of the hydraulic servo valve 100 used in an environment different from the learning data used when the hydraulic oil cleanliness estimation model was generated is acquired by the detection information acquisition unit 401, the feature calculation unit 414 may calculate the feature after adjusting for the difference in the environment. This can improve the robustness of the hydraulic oil cleanliness estimation.

As the characteristic quantity, another physical quantity that represents the characteristics of the operation of the spool 12 of the pilot valve 10 may be calculated. For example, the amount of overshoot or delay of the spool 12 of the pilot valve 10 may be calculated as the characteristic quantity.

As the learning data and the detection information, data recorded when the hydraulic servo valve 100 is used in a specific environment may be used. In this case, similar to the first to fifth embodiments, the learning device 300 may further include a data selection unit 320, and the state estimation device 400 may further include a data selection unit 417. The data selection unit 320 and the data selection unit 417 may select data recorded when the hydraulic servo valve 100 is operated in a test mode in which the spool 12 of the pilot valve 10 is operated in a specific pattern. This can improve the robustness of the estimation of the hydraulic oil cleanliness. The data selection unit 320 and the data selection unit 417 may select data recorded when the engine 80 is stopped and the spool sticking prevention operation is being performed. This can suppress the influence of disturbances and improve the estimation accuracy of the hydraulic oil cleanliness.

In addition to the data used as learning data and detection information, any of other data such as engine speed, engine load, pressure of hydraulic oil pressurized by the hydraulic pump 42, pressure or flow rate of hydraulic oil before and after a filter for removing foreign matter mixed in the hydraulic oil, temperature of hydraulic oil, power supply voltage, power supply current, etc., or any combination thereof may be used as learning data and detection information. Among these data, data that may affect data such as coil applied current, pilot valve target spool position, and pilot valve actual spool position as environmental factors may be selected and used. Since the frictional force when the spool 12 of the pilot valve 10 slides may change not only due to the cleanliness of the hydraulic oil but also due to wear of the inner wall of the spool 12 of the pilot valve 10 and the main body part 10b, factors that may affect wear of the inner wall of the spool 12 of the pilot valve 10 and the main body part 10b, such as the cumulative use time of the pilot valve 10 and the cumulative movement distance of the spool 12, may be used as learning data and detection information. By generating a hydraulic oil cleanliness estimation model that reflects the influence of these data, the robustness of the hydraulic oil cleanliness estimation can be further improved.

[Example 5: Position sensor disconnection detection]
In the fifth embodiment, a technique for detecting that the position sensor 29 for detecting the actual position of the spool 28 of the main valve 20 has been disconnected will be described. When the position sensor 29 is disconnected, a signal indicating the correct actual position of the spool 28 cannot be received from the position sensor 29, and the actual spool position of the main valve 20 deviates from the target spool position. As described in the third embodiment, in the past, an alarm was issued when the actual spool position of the main valve 20 deviated from the target spool position, but it was not possible to identify the cause of the alarm. If it were possible to detect that the position sensor 29 has been disconnected with high accuracy, it would be possible to identify the cause of the alarm when an alarm is issued and quickly take appropriate measures.

Figure 17 shows a schematic diagram of the operation of the spool 12 of the pilot valve 10 and the spool 28 of the main valve 20. Figure 17(a) shows an example of operation when the hydraulic servo valve 100 is normally controlled. The example of operation shown in Figure 17(a) is the same as the example of operation shown in Figure 14(a).

Figure 17(b) shows an example of operation when the position sensor 29 of the spool 28 of the main valve 20 is disconnected. The operation up to time t3 is the same as in Figure 17(a), but assume that the position sensor 29 of the spool 28 of the main valve 20 is disconnected at time t6. The feedback signal of the actual spool position of the main valve 20 abruptly changes to an abnormal value (e.g., zero) immediately after the disconnection, and thereafter remains fixed at the abnormal value. In the example of this figure, since the actual spool position of the main valve 20 is fixed while deviating from the target spool position in the positive direction, the negative maximum value continues to be input as the target spool position of the pilot valve 10 in order to return the spool 28 of the main valve 20 to the target spool position. Therefore, in reality, as shown by the dashed line, the spool 28 of the main valve 20 continues to move in the negative direction, and when it reaches the limit position in the negative direction, it stops at the limit position without moving any further.

In this way, the operation patterns of the spool 12 of the pilot valve 10 and the spool 28 of the main valve 20 are significantly different between normal operation and when the position sensor 29 is broken. Therefore, by modeling the difference between the operation pattern in normal operation and the operation pattern when the position sensor 29 is broken, the break in the position sensor 29 can be detected. Even if time has passed since an alarm was issued to notify that the actual spool position of the main valve 20 has shifted from the target spool position, the history of the position sensor 29 being broken can be detected from the log data, and can be used as evidence of the cause of the malfunction. According to the technology of this embodiment, when an alarm is issued to notify that the actual spool position of the main valve 20 has shifted from the target spool position, it is possible to accurately determine whether the cause is the intrusion of foreign matter described in embodiment 3 or the break in the position sensor 29 described in this embodiment, and to carry out appropriate maintenance. In addition, chattering caused by poor contact of the position sensor 29 can be accurately detected by modeling the operation pattern of the actual spool position of the main valve 20 and the target spool position of the pilot valve 10. This allows an abnormality to be detected before the position sensor 29 breaks and appropriate maintenance can be performed, preventing malfunctions caused by breaks in the position sensor 29.

Figure 18 shows the configuration of a learning device and a state estimation device according to Example 5. In the learning device 300 of this example, the learning data acquisition unit 301 acquires the main valve actual spool position 311, the pilot valve target spool position 316, and the position sensor disconnection data 351 as learning data.

The main valve actual spool position 311 and the pilot valve target spool position 316 are time series data when the hydraulic servo valve 100 is used within a specified period before or after the point in time when the position sensor disconnection data 351 is actually measured. These time series data may be time series data for one to several cycles of the rotational operation of the engine 80.

The position sensor breakage data 351 is data that indicates the presence or absence, timing, manner, cause, etc. of a breakage in the position sensor 29 of the spool 28 of the main valve 20. When log data is used as the learning data, the position sensor breakage data 351 is the actual measurement value when the position sensor 29 actually breaks. When test data is used as the learning data, the position sensor breakage data 351 is the actual measurement value when the position sensor 29 is broken. When simulation data is used as the learning data, the position sensor breakage data 351 is data input to the simulator as a simulation condition.

The position sensor disconnection detection model generation unit 353 uses the learning data acquired by the learning data acquisition unit 301 to generate a position sensor disconnection detection model used to detect a disconnection of the position sensor 29 in the state estimation device 400. The position sensor disconnection detection model may be a neural network that inputs time series data of the main valve actual spool position 311 and the pilot valve target spool position 316 for a predetermined period to an input layer and outputs position sensor disconnection data from an output layer. In this case, the position sensor disconnection detection model generation unit 353 learns the position sensor disconnection detection model by adjusting the intermediate layer of the neural network so that a value approximating the position sensor disconnection data 351 is output from the output layer when the time series data of the main valve actual spool position 311 and the pilot valve target spool position 316 for a predetermined period is input to the input layer.

The position sensor disconnection detection model providing unit 354 provides the position sensor disconnection detection model generated by the position sensor disconnection detection model generating unit 353 to the state estimation device 400.

The detection information acquisition unit 401 acquires the main valve actual spool position 411 and the pilot valve target spool position 415 as detection information. This detection information is time-series data for a predetermined period of time that is the same as the learning data used to generate the position sensor disconnection detection model.

The position sensor disconnection detection unit 451 detects a disconnection of the position sensor 29 of the spool 28 of the main valve 20 using the position sensor disconnection detection model generated by the learning device 300. The position sensor disconnection detection unit 451 inputs the time series data of the main valve actual spool position 411 and the pilot valve target spool position 415 acquired by the detection information acquisition unit 401 to the position sensor disconnection detection model, and acquires the position sensor disconnection data output from the position sensor disconnection detection model as a judgment result. The position sensor disconnection detection unit 451 may estimate the period until the position sensor 29 disconnects based on the output position sensor disconnection data. In this case, the period until the position sensor 29 disconnects may be estimated by comparing the position sensor disconnection data of multiple hydraulic servo valves 100 provided corresponding to multiple cylinders 81 of the same engine 80.

The position sensor disconnection detection result output unit 452 outputs the result determined by the position sensor disconnection detection unit 451. When the position sensor disconnection detection unit 451 detects that the position sensor 29 is disconnected, the position sensor disconnection detection result output unit 452 may report this. The position sensor disconnection detection result output unit 452 may output the period until the position sensor 29 is disconnected, which is estimated by the position sensor disconnection detection unit 451.

A feature value correlated with the disconnection of the position sensor 29 may be calculated from the main valve actual spool position or the pilot valve target spool position. In this case, similar to Example 1-2, the learning device 300 may further include a feature value calculation unit 315, and the state estimation device 400 may further include a feature value calculation unit 414. When the position sensor 29 of the spool 28 of the main valve 20 disconnects, the main valve actual spool position suddenly changes to an abnormal value, so the feature value calculation unit may calculate the speed or acceleration of the spool 28 of the main valve 20 by calculating the rate of change or the rate of change of the rate of change of the main valve actual spool position.

The position sensor disconnection detection model generation unit 353 generates a position sensor disconnection detection model using the learning data acquired by the learning data acquisition unit 301 and the feature amount calculated by the feature amount calculation unit 315. The position sensor disconnection detection model generation unit 353 may generate a position sensor disconnection detection model using the position sensor disconnection data 351 and the feature amount calculated by the feature amount calculation unit 315. The position sensor disconnection detection model generation unit 353 may generate a position sensor disconnection detection model using the main valve actual spool position 311, the pilot valve target spool position 316, and the feature amount. For example, the position sensor disconnection detection model may be a neural network that inputs time series data of the main valve actual spool position 311 and the pilot valve target spool position 316 for a predetermined period and the feature amount for the predetermined period to an input layer and outputs the position sensor disconnection data from an output layer. In this case, the position sensor disconnection detection model generation unit 353 learns the position sensor disconnection detection model by adjusting the intermediate layer of the neural network so that when the time series data for a specified period of time of the main valve actual spool position 311 and the pilot valve target spool position 316 and the feature values for that specified period are input to the input layer, a value approximating the position sensor disconnection data 351 is output from the output layer.

The feature amount calculation unit 414 of the state estimation device 400 calculates the feature amount from the main valve actual spool position 411 or the pilot valve target spool position 415 acquired by the detection information acquisition unit 401 in the same manner as the feature amount calculation unit 315 calculated the feature amount when the position sensor breakage detection model used by the position sensor breakage detection unit 451 was generated. When the main valve actual spool position 411 and the pilot valve target spool position 415 of the hydraulic servo valve 100 used in an environment different from the learning data used when the position sensor breakage detection model was generated are acquired by the detection information acquisition unit 401, the feature amount calculation unit 414 may calculate the feature amount after adjusting for the difference in the environment. This can improve the robustness of the position sensor breakage detection.

As the characteristic quantity, another physical quantity that represents the characteristics of the operation of the spool 28 of the main valve 20 may be calculated. For example, the amount of overshoot or delay of the spool 28 of the main valve 20 may be calculated as the characteristic quantity.

As the learning data and detection information, the pilot valve actual spool position may be used instead of or in addition to the pilot valve target spool position. In this case, similar to Example 1-4, the learning device 300 may be provided with an offset time calculation unit 319 to adjust the offset time required for the actual spool position of the pilot valve 10 to follow the target spool position of the pilot valve. The method of calculating the offset time may be similar to Example 1-2. As the learning data and detection information, a combination of any two or more of the pilot valve target spool position, the pilot valve actual spool position, the main valve target spool position, and the main valve actual spool position may be used.

As the learning data and the detection information, data recorded when the hydraulic servo valve 100 is used in a specific environment may be used. In this case, similar to the first to fifth embodiments, the learning device 300 may further include a data selection unit 320, and the state estimation device 400 may further include a data selection unit 417. The data selection unit 320 and the data selection unit 417 may select data recorded when the hydraulic servo valve 100 is operated in a test mode in which the spool 12 of the pilot valve 10 is operated in a specific pattern. This can improve the robustness of the disconnection detection of the position sensor 29. The data selection unit 320 and the data selection unit 417 may select data recorded when the engine 80 is stopped and an operation to prevent the spool from sticking is being performed. This can suppress the influence of disturbances and improve the detection accuracy of the disconnection of the position sensor 29.

[Example 6: Detection of loose position sensor]
In the sixth embodiment, a technique for detecting a deviation in the position of the position sensor 29 for detecting the actual position of the spool 28 of the main valve 20 due to, for example, loosening of the screws for fixing the position sensor 29 will be described. If the position of the position sensor 29 is deviated, a deviation occurs between the feedback signal of the main valve actual spool position and the actual position of the spool 28 of the main valve 20, making it impossible to control the spool 28 of the main valve 20 normally, which may cause a malfunction of a controlled object such as the engine 80. If it is possible to detect a deviation in the position of the position sensor 29 with high accuracy, appropriate measures such as tightening the screws or replacing the position sensor 29 can be quickly taken.

Figure 19 shows the state of the position sensor 29 of the spool 28 of the main valve 20. Figure 19(a) shows the state in which the position sensor 29 is normally attached. The position sensor 29 is screwed into the female thread 30 cut at the upper end of the spool 28. When this thread loosens, as shown in Figure 19(b), a gap is generated between the position sensor 29 and the spool 28, and the position of the spool 28 is offset downward by the amount of this gap d. However, since the position of the spool 28 is detected by the position sensor 29, the same position is detected in both the case of Figure 19(a) and the case of Figure 19(b), and the actual position of the spool 28 offset downward cannot be detected. In the state of Figure 19(b), even if the main valve actual spool position appears to correctly follow the main valve target spool position, the actual position of the spool 28 is shifted downward from the main valve target spool position, causing a malfunction in the operation of the controlled object. In this embodiment, the amount of control oil supplied to the fuel injection actuator for supplying fuel to cylinder 81 of engine 80 is reduced, and the amount of fuel supplied to cylinder 81 is reduced, so the pressure inside cylinder 81 and the temperature of the exhaust gas from cylinder 81 drop, causing an alarm to be issued.

In this way, when the position sensor 29 is displaced due to looseness, the actual main valve spool position follows the main valve target spool position, but the pressure and exhaust temperature of the cylinder 81 drop. Therefore, when an alarm is issued due to an abnormality in the pressure or exhaust temperature of the cylinder 81, it is possible to identify whether the abnormality in the position sensor 29 is the cause by checking whether the actual main valve spool position followed the main valve target spool position. Even if time has passed since an alarm was issued to notify the engine 80 of an abnormality, it is possible to detect from the log data that the position sensor 29 has loosened, and this can be used as evidence of the cause of the malfunction. This allows the looseness of the position sensor 29 to be accurately detected and appropriate maintenance to be performed, thereby reducing the maintenance costs and labor required to improve the malfunction of the engine 80.

Figure 20 shows the configuration of a learning device and a state estimation device according to Example 6. In the learning device 300 of this example, the learning data acquisition unit 301 acquires the main valve actual spool position 311, the main valve target spool position 330, the cylinder exhaust temperature 361, and the position sensor looseness data 362 as learning data.

The main valve actual spool position 311, the main valve target spool position 330, and the cylinder exhaust temperature 361 are time series data when the hydraulic servo valve 100 is used within a specified period before or after the time when the position sensor looseness data 362 is actually measured. These time series data may be time series data for one to several cycles of the rotational operation of the engine 80.

The position sensor looseness data 362 is data that indicates the presence or absence, degree, manner, cause, etc. of position misalignment of the position sensor 29 of the spool 28 of the main valve 20. When log data is used as the learning data, the position sensor looseness data 362 is the actual measurement value when the position sensor 29 is actually misaligned. When test data is used as the learning data, the position sensor looseness data 362 is the actual measurement value when the position sensor 29 is misaligned. When simulation data is used as the learning data, the position sensor looseness data 362 is data input to the simulator as a simulation condition.

The position sensor looseness detection model generating unit 363 uses the learning data acquired by the learning data acquiring unit 301 to generate a position sensor looseness detection model used to detect looseness of the position sensor 29 in the state estimation device 400. The position sensor looseness detection model may be a neural network that inputs time series data of the main valve actual spool position 311, the main valve target spool position 330, and the cylinder exhaust temperature 361 for a predetermined period to an input layer and outputs position sensor looseness data from an output layer. In this case, the position sensor looseness detection model generating unit 363 learns the position sensor looseness detection model by adjusting the intermediate layer of the neural network so that a value approximating the position sensor looseness data 362 is output from the output layer when the time series data of the main valve actual spool position 311, the main valve target spool position 330, and the cylinder exhaust temperature 361 for a predetermined period is input to the input layer.

The position sensor loosening detection model providing unit 364 provides the position sensor loosening detection model generated by the position sensor loosening detection model generating unit 363 to the state estimation device 400.

The detection information acquisition unit 401 acquires the main valve actual spool position 411, the main valve target spool position 430, and the cylinder exhaust temperature 461 as detection information. This detection information is time series data for a predetermined period of time that is the same as the learning data used to generate the position sensor looseness detection model.

The position sensor loosening detection unit 462 detects the loosening of the position sensor 29 of the spool 28 of the main valve 20 using the position sensor loosening detection model generated by the learning device 300. The position sensor loosening detection unit 462 inputs the time series data of the main valve actual spool position 411, the main valve target spool position 430, and the cylinder exhaust temperature 461 acquired by the detection information acquisition unit 401 into the position sensor loosening detection model, and acquires the position sensor loosening data output from the position sensor loosening detection model as a judgment result. The position sensor loosening detection unit 462 may estimate the period until the position sensor 29 deviates by a predetermined value or more based on the output position sensor loosening data. In this case, the position sensor loosening data of multiple hydraulic servo valves 100 provided corresponding to multiple cylinders 81 of the same engine 80 may be compared to estimate the period until the position sensor 29 deviates by a predetermined value or more.

When multiple hydraulic servo valves 100 are provided corresponding to multiple cylinders 81 of the same engine 80, the position sensor looseness detection unit 462 may detect looseness of the position sensor 29 based on position sensor looseness data estimated in each hydraulic servo valve 100. For example, the position sensor looseness detection unit 462 may detect looseness of the position sensor 29 by comparing the exhaust temperatures of each cylinder 81. When the exhaust temperature of a certain cylinder 81 is lower than the average exhaust temperature of the other cylinders 81 by a predetermined value or more, the position sensor looseness detection unit 462 may determine that the position sensor 29 of the hydraulic servo valve 100 provided in that cylinder 81 is loose.

The position sensor looseness detection result output unit 463 outputs the result determined by the position sensor looseness detection unit 462. When the position sensor looseness detection unit 462 detects that the position sensor 29 is loose, the position sensor looseness detection result output unit 463 may report this fact. The position sensor looseness detection result output unit 463 may output the period until the position sensor 29 deviates by a predetermined value or more, as estimated by the position sensor looseness detection unit 462.

A feature quantity correlated with the looseness of the position sensor 29 may be calculated from the main valve actual spool position, the main valve target spool position, the cylinder exhaust temperature, or the like. In this case, as in the first and second embodiments, the learning device 300 may further include a feature quantity calculation unit 315, and the state estimation device 400 may further include a feature quantity calculation unit 414. The greater the degree of looseness of the position sensor 29 of the spool 28 of the main valve 20, the greater the deviation between the actual main valve spool position and the actual position of the spool 28, and therefore the greater the amount of decrease in the exhaust temperature or pressure of the cylinder from its normal value. Therefore, the feature quantity calculation unit may calculate the amount of decrease, the rate of change, the rate of change of the rate of change, or the like of the exhaust temperature or pressure of the cylinder from its normal value as a feature quantity.

The position sensor loosening detection model generation unit 363 generates a position sensor loosening detection model using the learning data acquired by the learning data acquisition unit 301 and the feature amount calculated by the feature amount calculation unit 315. For example, the position sensor loosening detection model may be a formula that calculates the position sensor loosening data using the feature amount as an input variable. In this case, the position sensor loosening detection model generation unit 363 may generate the formula by a statistical method such as regression analysis. The position sensor loosening detection model generation unit 363 may generate the position sensor loosening detection model using the position sensor loosening data 362 and the feature amount calculated by the feature amount calculation unit 315. The position sensor loosening detection model generation unit 363 may generate the position sensor loosening detection model using the main valve actual spool position 311, the main valve target spool position 330, and the feature amount. For example, the position sensor looseness detection model may be a neural network that inputs time series data for a predetermined period of time of main valve actual spool position 311 and main valve target spool position 330 and feature values for the predetermined period to an input layer and outputs position sensor looseness data from an output layer. In this case, the position sensor looseness detection model generation unit 363 learns the position sensor looseness detection model by adjusting the intermediate layer of the neural network so that when the time series data for a predetermined period of time of main valve actual spool position 311 and main valve target spool position 330 and feature values for the predetermined period are input to the input layer, a value approximating position sensor looseness data 362 is output from the output layer.

The feature calculation unit 414 of the state estimation device 400 calculates the feature from the main valve actual spool position 411, the main valve target spool position 430, or the cylinder exhaust temperature 461 acquired by the detection information acquisition unit 401 in the same manner as the feature calculation unit 315 calculated the feature when the position sensor looseness detection model used by the position sensor looseness detection unit 462 was generated. If the main valve actual spool position 411, the main valve target spool position 430, or the cylinder exhaust temperature 461 of the hydraulic servo valve 100 used in an environment different from the learning data used when the position sensor looseness detection model was generated is acquired by the detection information acquisition unit 401, the feature calculation unit 414 may calculate the feature after adjusting for the difference in the environment. This can improve the robustness of the position sensor looseness detection.

As the characteristic quantity, another physical quantity that represents the characteristics of the operation of the spool 28 of the main valve 20 may be calculated. For example, the amount of overshoot or delay of the spool 28 of the main valve 20 may be calculated as the characteristic quantity.

As the learning data and detection information, the pressure or maximum pressure of cylinder 81 may be used instead of or in addition to the cylinder exhaust temperature. Also, as the learning data and detection information, various data indicating the state and operation of the control object controlled by hydraulic servo valve 100 may be used.

As the learning data and the detection information, data recorded when the hydraulic servo valve 100 is used in a specific environment may be used. In this case, similar to the first to fifth embodiments, the learning device 300 may further include a data selection unit 320, and the state estimation device 400 may further include a data selection unit 417. The data selection unit 320 and the data selection unit 417 may select data recorded when the hydraulic servo valve 100 is operated in a test mode in which the spool 12 of the pilot valve 10 is operated in a specific pattern. This can improve the robustness of the position sensor loosening detection. The data selection unit 320 and the data selection unit 417 may select data recorded when the engine 80 is stopped and the spool sticking prevention operation is being performed. This can suppress the influence of disturbances and improve the detection accuracy of the position sensor loosening.

[Example 7: Detection of deviation of pilot valve from neutral position]
In the seventh embodiment, a technique for detecting the shift of the neutral position of the spool 12 of the pilot valve 10 will be described. When the neutral position is commanded as the pilot valve target spool position, if the pilot valve 10 is operating normally, the spool 12 is moved to the neutral position, the supply and recovery of hydraulic oil to the main valve 20 is stopped, and the spool 28 of the main valve 20 stops at the main valve target spool position. However, if the neutral position of the spool 12 is shifted due to some cause, the supply and recovery of hydraulic oil to the main valve 20 cannot be controlled normally, and the controlled object such as the engine 80 may malfunction. If the shift of the neutral position of the spool 12 of the pilot valve 10 can be detected with high accuracy, appropriate measures such as maintaining the control system of the pilot valve 10 or replacing the pilot valve 10 can be quickly taken.

There are two main reasons why the neutral position of the spool 12 of the pilot valve 10 is shifted. One is an electrical reason that the spool 12 cannot be moved accurately to the neutral position due to an abnormality in the position sensor 19 for detecting the position of the spool 12 of the pilot valve 10 or an abnormality in the electrical system for driving the spool 12. The other is a mechanical reason that the valve body 14 cannot close the A port 16a and prevent communication with both the P port 16p and the T port 16t even when the spool 12 is in the neutral position due to uneven wear of the spool 12 or the main body 10b. In either case, when the neutral position is specified as the pilot valve target spool position, the spool 28 of the main valve 20 does not stop but moves, so that the feedback signal of the main valve actual spool position shifts from the main valve target spool position, and the pilot valve 10 is controlled to correct the shift. In this way, the operation patterns of the spool 12 of the pilot valve 10 and the spool 28 of the main valve 20 differ greatly between normal operation and when the neutral position of the spool 12 of the pilot valve 10 is shifted. Therefore, by modeling the difference between the operation pattern in normal operation and the operation pattern when the neutral position of the spool 12 is shifted, it is possible to detect the shift of the neutral position of the spool 12. Since the operation pattern can differ depending on the degree and cause of the shift of the spool 12 from the neutral position, it is possible to detect not only whether the spool 12 is shifted from the neutral position, but also the cause of the shift and the magnitude of the shift.

When the spool 12 is unevenly worn, the valve body 14 cannot close the A port 16a even if the spool 12 is in the correct neutral position, and hydraulic oil leaks out from the A port 16a. By actually measuring the amount of this internal oil leakage, it is possible to determine whether the neutral position shift of the spool 12 is due to uneven wear or an electrical cause. The amount of internal oil leakage may be measured by installing a flow meter in the pilot valve 10, may be measured with an external flow meter when maintaining the pilot valve 10, or may be estimated using the technology of Example 1.

When the neutral position of the spool 12 of the pilot valve 10 is shifted, an abnormality occurs in the pressure or exhaust temperature of the cylinder 81 in the engine 80 to be controlled, and an alarm is issued. When an alarm is issued, it is possible to determine whether the cause is an abnormality in the position sensor 29 described in the sixth embodiment or a neutral position shift of the spool 12 of the pilot valve 10 by checking the operation patterns of the spool 12 of the pilot valve 10 and the spool 28 of the main valve 20. In addition, if the cause is a neutral position shift of the spool 12 of the pilot valve 10, it is possible to determine whether the neutral position is shifted due to an electrical cause or a mechanical cause by checking the amount of internal oil leakage. This allows the cause of the malfunction of the engine 80 to be accurately determined and appropriate maintenance to be performed according to the cause, thereby reducing the maintenance costs and labor required to improve the malfunction of the engine 80. When PID control is performed in the feedback control of the hydraulic servo valve 100, the deviation is corrected to a certain extent by the integral component (I component), but the technology of this embodiment can accurately detect the neutral position deviation of the spool 12 of the pilot valve 10 even when the neutral position deviation is corrected by PID control. Also, when P control or PD control is performed, the deviation is not corrected, so the technology of this embodiment is particularly effective.

Figure 21 shows the configuration of a learning device and a state estimation device according to Example 7. In the learning device 300 of this example, the learning data acquisition unit 301 acquires the main valve actual spool position 311, the main valve target spool position 330, the pilot valve actual spool position 310, the cylinder exhaust temperature 361, and the neutral position deviation data 371 as learning data.

The main valve actual spool position 311, the main valve target spool position 330, the pilot valve actual spool position 310, and the cylinder exhaust temperature 361 are time series data when the hydraulic servo valve 100 is used within a specified period before or after the neutral position deviation data 371 is actually measured. These time series data may be time series data for one to several cycles of the rotational operation of the engine 80.

The neutral position deviation data 371 is data that indicates the presence or absence, the degree, and the cause of neutral position deviation of the spool 12 of the pilot valve 10. When log data is used as the learning data, the neutral position deviation data 371 is the actual measurement value when the neutral position deviation of the spool 12 actually occurs. When test data is used as the learning data, the neutral position deviation data 371 is the actual measurement value when the neutral position deviation of the spool 12 occurs. When simulation data is used as the learning data, the neutral position deviation data 371 is data that is input to the simulator as a simulation condition.

The neutral position deviation detection model generation unit 373 uses the learning data acquired by the learning data acquisition unit 301 to generate a neutral position deviation detection model used to detect a neutral position deviation of the spool 12 of the pilot valve 10 in the state estimation device 400. The neutral position deviation detection model may be a neural network that inputs time series data of the main valve actual spool position 311, the main valve target spool position 330, the pilot valve actual spool position 310, and the cylinder exhaust temperature 361 for a predetermined period to an input layer and outputs neutral position deviation data from an output layer. In this case, the neutral position deviation detection model generation unit 373 learns the neutral position deviation detection model by adjusting the intermediate layer of the neural network so that a value approximating the neutral position deviation data 371 is output from the output layer when the time series data of the main valve actual spool position 311, the main valve target spool position 330, the pilot valve actual spool position 310, and the cylinder exhaust temperature 361 for a predetermined period is input to the input layer.

The neutral position deviation detection model providing unit 374 provides the neutral position deviation detection model generated by the neutral position deviation detection model generating unit 373 to the state estimation device 400.

The detection information acquisition unit 401 acquires the main valve actual spool position 411, the main valve target spool position 430, the pilot valve actual spool position 410, and the cylinder exhaust temperature 461 as detection information. This detection information is time series data for a predetermined period of time that is the same as the learning data used to generate the neutral position deviation detection model.

The neutral position deviation detection unit 472 detects the neutral position deviation of the spool 12 of the pilot valve 10 using the neutral position deviation detection model generated by the learning device 300. The neutral position deviation detection unit 472 inputs the time series data of the main valve actual spool position 411, the main valve target spool position 430, the pilot valve actual spool position 410, and the cylinder exhaust temperature 461 acquired by the detection information acquisition unit 401 into the neutral position deviation detection model, and acquires the neutral position deviation data output from the neutral position deviation detection model as a judgment result. The neutral position deviation detection unit 472 may estimate the period until the neutral position of the spool 12 of the pilot valve 10 deviates from the actual neutral position by a predetermined value or more based on the output neutral position deviation data. In this case, the neutral position deviation data of multiple hydraulic servo valves 100 provided for multiple cylinders 81 of the same engine 80 may be compared to estimate the period until the neutral position of the spool 12 of the pilot valve 10 deviates from the actual neutral position by a predetermined value or more.

When multiple hydraulic servo valves 100 are provided corresponding to multiple cylinders 81 of the same engine 80, the neutral position deviation detection unit 472 may detect the neutral position deviation of the spool 12 of the pilot valve 10 based on the neutral position deviation data estimated in each hydraulic servo valve 100. For example, the neutral position deviation detection unit 472 may detect the neutral position deviation of the spool 12 of the pilot valve 10 by comparing the exhaust temperatures of each cylinder 81. When the exhaust temperature of a certain cylinder 81 is lower than the average value of the exhaust temperatures of the other cylinders 81 by a predetermined value or more, the neutral position deviation detection unit 472 may determine that the neutral position of the spool 12 of the pilot valve 10 of the hydraulic servo valve 100 provided in that cylinder 81 is shifted.

The neutral position deviation detection result output unit 473 outputs the result determined by the neutral position deviation detection unit 472. When the neutral position deviation detection unit 472 detects that the neutral position of the spool 12 of the pilot valve 10 is shifted, the neutral position deviation detection result output unit 473 may report this. The neutral position deviation detection result output unit 473 may output the period until the neutral position of the spool 12 of the pilot valve 10 estimated by the neutral position deviation detection unit 472 shifts by a predetermined value or more.

A feature quantity correlated with the neutral position deviation of the spool 12 of the pilot valve 10 may be calculated from the actual main valve spool position, the target main valve spool position, the actual pilot valve spool position, or the cylinder exhaust temperature. In this case, as in the first and second embodiments, the learning device 300 may further include a feature quantity calculation unit 315, and the state estimation device 400 may further include a feature quantity calculation unit 414. The greater the degree of neutral position deviation of the spool 12 of the pilot valve 10, the greater the degree to which the actual main valve spool position deviates from the target main valve spool position, and therefore the greater the deviation of the exhaust temperature or pressure of the cylinder from its normal value. Therefore, the feature quantity calculation unit may calculate the deviation, rate of change, or rate of change of the rate of change of the exhaust temperature or pressure of the cylinder from its normal value as a feature quantity.

The neutral position deviation detection model generation unit 373 generates a neutral position deviation detection model using the learning data acquired by the learning data acquisition unit 301 and the feature amount calculated by the feature amount calculation unit 315. For example, the neutral position deviation detection model may be a formula that calculates neutral position deviation data using the feature amount as an input variable. In this case, the neutral position deviation detection model generation unit 373 may generate the formula by a statistical method such as regression analysis. The neutral position deviation detection model generation unit 373 may generate a neutral position deviation detection model using the neutral position deviation data 371 and the feature amount calculated by the feature amount calculation unit 315. The neutral position deviation detection model generation unit 373 may generate a neutral position deviation detection model using the main valve actual spool position 311, the main valve target spool position 330, the pilot valve actual spool position 310, and the feature amount. For example, the neutral position deviation detection model may be a neural network that inputs time series data of the main valve actual spool position 311, the main valve target spool position 330, and the pilot valve actual spool position 310 for a predetermined period and feature values for the predetermined period into an input layer and outputs neutral position deviation data from an output layer. In this case, the neutral position deviation detection model generation unit 373 learns the neutral position deviation detection model by adjusting the intermediate layer of the neural network so that when the time series data of the main valve actual spool position 311, the main valve target spool position 330, and the pilot valve actual spool position 310 for a predetermined period and feature values for the predetermined period are input into the input layer, a value approximating the neutral position deviation data 371 is output from the output layer.

The feature calculation unit 414 of the state estimation device 400 calculates the feature from the main valve actual spool position 411, the main valve target spool position 430, the pilot valve actual spool position 410, or the cylinder exhaust temperature 461 acquired by the detection information acquisition unit 401 in the same manner as the feature calculation unit 315 calculated the feature when the neutral position deviation detection model used by the neutral position deviation detection unit 472 was generated. If the main valve actual spool position 411, the main valve target spool position 430, the pilot valve actual spool position 410, or the cylinder exhaust temperature 461 of the hydraulic servo valve 100 used in an environment different from the learning data used when the neutral position deviation detection model was generated is acquired by the detection information acquisition unit 401, the feature calculation unit 414 may calculate the feature after adjusting for the difference in the environment. This can improve the robustness of the neutral position deviation detection.

As the characteristic quantity, another physical quantity that represents the characteristics of the operation of the spool 12 of the pilot valve 10 or the spool 28 of the main valve 20 may be calculated. For example, the acceleration, speed, amount of overshoot, amount of delay, etc. of the spool 12 of the pilot valve 10 or the spool 28 of the main valve 20 may be calculated as the characteristic quantity.

As the learning data and detection information, the pressure or maximum pressure of cylinder 81 may be used instead of or in addition to the cylinder exhaust temperature. Also, as the learning data and detection information, various data indicating the state and operation of the control object controlled by hydraulic servo valve 100 may be used.

As the learning data and detection information, the pilot valve target spool position may be used instead of or in addition to the pilot valve actual spool position. In this case, similar to Example 1-4, the learning device 300 may be provided with an offset time calculation unit 319 to adjust the offset time required for the actual spool position of the pilot valve 10 to follow the target spool position of the pilot valve. The method of calculating the offset time may be similar to Example 1-2. As the learning data and detection information, a combination of any two or more of the pilot valve target spool position, the pilot valve actual spool position, the main valve target spool position, and the main valve actual spool position may be used.

As the learning data and the detection information, data recorded when the hydraulic servo valve 100 is used in a specific environment may be used. In this case, similar to the first to fifth embodiments, the learning device 300 may further include a data selection unit 320, and the state estimation device 400 may further include a data selection unit 417. The data selection unit 320 and the data selection unit 417 may select data recorded when the hydraulic servo valve 100 is operated in a test mode in which the spool 12 of the pilot valve 10 is operated in a specific pattern. This can improve the robustness of the detection of the neutral position deviation. The data selection unit 320 and the data selection unit 417 may select data recorded when the engine 80 is stopped and the spool is prevented from sticking. This can suppress the influence of disturbances and improve the detection accuracy of the neutral position deviation.

The features described in the above embodiments 1 to 7 may be applied in any combination.

Above, examples of embodiments of the present invention have been described in detail. All of the above-mentioned embodiments merely show specific examples of how to put the present invention into practice. The contents of the embodiments do not limit the technical scope of the present invention, and many design changes such as changing, adding, or deleting components are possible within the scope of the idea of the invention defined in the claims. In the above-mentioned embodiments, the contents for which such design changes are possible are explained with the notation "in the embodiment" or "in the embodiment", but this does not mean that design changes are not permitted for contents without such notation.

[Modification]
The following describes the modified examples. In the drawings and description of the modified examples, the same or equivalent components and members as those in the embodiment are denoted by the same reference numerals. Descriptions that overlap with the embodiment will be omitted as appropriate, and the description will focus on configurations that differ from the embodiment.

In the description of the embodiment, an example was shown in which the pilot valve 10 of the hydraulic servo valve 100 is a three-port valve, but the present invention is not limited to this. The hydraulic servo valve 100 may be a valve of another type.

In the description of the embodiment, an example has been shown in which the pilot valve 10 of the hydraulic servo valve 100 controls the main valve 20, but the present invention is not limited to this. The controlled object of the pilot valve 10 may be any other actuator.

In the description of the embodiment, an example has been shown in which the hydraulic servo valve 100 is used to control the engine 80 of the ship 2, but the present invention is not limited to this. The hydraulic servo valve 100 may be used to control any control target.

The above-mentioned modified example provides the same effects and advantages as the embodiment.

Any combination of the above-described embodiments and modifications is also useful as an embodiment of the present invention. The new embodiment resulting from the combination has the combined effects of each of the combined embodiments and modifications.

[Aspects of the present invention]
One aspect of the present invention is a state estimation device. The state estimation device includes an acquisition unit that acquires detection information of a data set including a target position and an actual position of a first movable part in a control valve that controls a flow rate of a working fluid depending on a position of the first movable part and a value of a current supplied to a drive unit that drives the first movable part, and an estimation unit that estimates a cleanliness of the working fluid based on the detection information acquired by the acquisition unit. According to this aspect, it is possible to estimate the cleanliness of the working fluid in the control valve, and therefore it is possible to take an appropriate measure depending on the state of the control valve.

The estimation unit may estimate the cleanliness of the working fluid using an estimation criterion for estimating the cleanliness of the working fluid based on the detection information of the data set. According to this aspect, it is possible to improve the accuracy of estimating the cleanliness of the working fluid.

The estimation criteria may be generated based on the relationship between the values of a data set when the control valve or a control valve of the same type as the control valve is used within a specified period of time and the cleanliness of the working fluid actually measured within the specified period of time. According to this aspect, the accuracy of estimating the cleanliness of the working fluid can be improved.

The estimation criterion is generated based on the relationship between the feature amount calculated from the values of the data set and the actual measured value of the cleanliness of the working fluid, and the estimation unit may calculate the feature amount from the detection information acquired by the acquisition unit and estimate the cleanliness of the working fluid based on the calculated feature amount. According to this aspect, it is possible to improve the estimation accuracy of the cleanliness of the working fluid.

The feature value may be the speed or acceleration of the first movable part. According to this aspect, the cleanliness of the working fluid can be estimated using a feature value that has a high correlation with the cleanliness of the working fluid, thereby improving the accuracy of estimating the cleanliness of the working fluid.

The acquisition unit may acquire detection information of the dataset in a usage environment similar to the usage environment of the control valve or a control valve of the same type as the control valve when the value of the dataset used to generate the estimation criterion was recorded. According to this aspect, it is possible to improve the robustness of the estimation of the cleanliness of the working fluid.

The usage environment may be a situation in which the control valve or a control object of the same type as the control valve is stopped and an anti-sticking operation of the first movable part is being performed. According to this aspect, the robustness of the estimation of the cleanliness of the working fluid can be improved.

The state estimation device may further include a notification unit that notifies the user when the cleanliness of the working fluid estimated by the estimation unit is below a certain level. According to this aspect, an abnormality in the working fluid of the control valve can be accurately detected, and appropriate measures can be taken depending on the state of the control valve.

The notification unit may estimate the time until the cleanliness of the working fluid falls below a threshold and notify the user. According to this embodiment, an abnormality in the working fluid of the control valve can be accurately predicted, and appropriate measures can be taken according to the state of the control valve.

The control valve may be a control valve for adjusting the amount of fuel supplied to the engine. According to this embodiment, an abnormality in the control valve for controlling the engine can be accurately detected, and appropriate measures can be taken according to the state of the control valve.

The control valve may be provided for each cylinder of an engine having multiple cylinders to adjust the amount of fuel supplied to each cylinder, and the estimation unit may determine an abnormality in the working fluid by comparing information about each of the multiple control valves provided for the multiple cylinders. According to this aspect, an abnormality in the control valve that controls the engine can be accurately identified, and appropriate measures can be taken depending on the state of the control valve.

Another aspect of the present invention is a control valve. This control valve includes a first movable part whose position changes in response to a control signal for specifying a position and whose flow rate of a working fluid is controlled in response to the position, an acquisition unit that acquires detection information of a data set including a target position and an actual position of the first movable part and a value of a current supplied to a drive unit that drives the first movable part, and an estimation unit that estimates the cleanliness of the working fluid based on the detection information acquired by the acquisition unit. According to this aspect, since the cleanliness of the working fluid in the control valve can be estimated, appropriate measures can be taken in response to the state of the control valve.

Yet another aspect of the present invention is a state estimation program. This state estimation program causes a computer to function as an acquisition unit that acquires detection information of a data set including a target position and an actual position of a first movable part in a control valve that controls the flow rate of the working fluid depending on the position of the first movable part and a value of a current supplied to a drive part that drives the first movable part, and an estimation unit that estimates the cleanliness of the working fluid based on the detection information acquired by the acquisition unit. According to this aspect, since it is possible to estimate the cleanliness of the working fluid in the control valve, it is possible to take appropriate measures depending on the state of the control valve.

Yet another aspect of the present invention is a state estimation method. This state estimation method causes a computer to execute a step of acquiring detection information of a data set including a target position and an actual position of a first movable part in a control valve that controls the flow rate of the working fluid depending on the position of the first movable part and a value of a current supplied to a drive part that drives the first movable part, and a step of estimating the cleanliness of the working fluid based on the detection information acquired in the acquiring step. According to this aspect, since it is possible to estimate the cleanliness of the working fluid in the control valve, it is possible to take appropriate measures depending on the state of the control valve.

One aspect of the present invention is a state estimation device. This state estimation device includes an acquisition unit that acquires detection information of a data set including a target position or actual position of a first movable part in a control valve that controls the flow rate of a working fluid according to the position of the first movable part and an actual position of a second movable part that changes its position according to the flow rate of the working fluid, and an estimation unit that estimates an abnormality in a detection unit for detecting the position of the second movable part based on the detection information acquired by the acquisition unit. According to this aspect, since it is possible to estimate an abnormality in the detection unit for detecting the position of the second movable part in the control valve, it is possible to take appropriate measures according to the state of the control valve.

The estimation unit may estimate the abnormality of the detection unit using an estimation criterion for estimating the abnormality of the detection unit based on the detection information of the dataset. According to this aspect, it is possible to improve the accuracy of estimating the abnormality of the detection unit.

The estimation criteria may be generated based on the relationship between the values of the data set and information about the state of the detection unit when the control valve or a control valve of the same type as the control valve is used within a specified period of time. According to this aspect, it is possible to improve the accuracy of estimating an abnormality in the detection unit.

The information regarding the state of the detection unit may be the presence or absence, timing, manner, or cause of a disconnection in the detection unit. According to this embodiment, the accuracy of estimating an abnormality in the detection unit can be improved.

The estimation criterion is generated based on the relationship between the feature amount calculated from the values of the dataset and the information on the state of the detection unit, and the estimation unit may calculate the feature amount from the detection information of the dataset acquired by the acquisition unit, and estimate an abnormality in the detection unit based on the calculated feature amount. According to this aspect, it is possible to improve the accuracy of estimating the abnormality in the detection unit.

The feature value may be the speed or acceleration of the second movable part. According to this aspect, the feature value that has a high correlation with the abnormality of the detection part can be used to estimate the abnormality of the detection part, thereby improving the accuracy of estimating the abnormality of the detection part.

The acquisition unit may acquire detection information of a dataset in a usage environment similar to the usage environment of the control valve or a control valve of the same type as the control valve when the value of the dataset used to generate the estimation criterion was recorded. According to this aspect, it is possible to improve the robustness of the estimation of anomalies by the detection unit.

The usage environment may be a situation in which the control valve or a control object of the same type as the control valve is stopped, and an anti-sticking operation of the first movable part and the second movable part is being performed. According to this aspect, the robustness of the estimation of an abnormality of the detection unit can be improved.

The state estimation device may further include a notification unit that notifies the user when the estimation unit estimates that there is an abnormality in the detection unit. According to this aspect, it is possible to accurately grasp the abnormality in the detection unit of the control valve, and to take appropriate measures according to the state of the control valve.

The notification unit may estimate the time until an abnormality occurs in the detection unit and notify the user. According to this embodiment, an abnormality in the detection unit of the control valve can be accurately predicted, and appropriate measures can be taken according to the state of the control valve.

The control valve may be a control valve for adjusting the amount of fuel supplied to the engine. According to this embodiment, an abnormality in the control valve for controlling the engine can be accurately detected, and appropriate measures can be taken according to the state of the control valve.

Another aspect of the present invention is a control valve. This control valve includes a first movable part whose position changes in response to a control signal for specifying a position and whose flow rate of a working fluid is controlled in response to the position, an acquisition unit that acquires detection information of a data set including a target position or actual position of the first movable part and an actual position of a second movable part whose position changes in response to the flow rate of the working fluid, and an estimation unit that estimates an abnormality in the detection unit for detecting the position of the second movable part based on the detection information acquired by the acquisition unit. According to this aspect, since it is possible to estimate an abnormality in the detection unit for detecting the position of the second movable part in the control valve, it is possible to take appropriate measures in response to the state of the control valve.

Yet another aspect of the present invention is a state estimation program. This state estimation program causes a computer to function as an acquisition unit that acquires detection information of a data set including a target position or actual position of a first movable part in a control valve that controls the flow rate of a working fluid according to the position of the first movable part and an actual position of a second movable part that changes its position according to the flow rate of the working fluid, and an estimation unit that estimates an abnormality in a detection unit for detecting the position of the second movable part based on the detection information acquired by the acquisition unit. According to this aspect, since it is possible to estimate an abnormality in the detection unit for detecting the position of the second movable part in the control valve, it is possible to take appropriate measures according to the state of the control valve.

Yet another aspect of the present invention is a state estimation method. This state estimation method causes a computer to execute a step of acquiring detection information of a data set including a target position or actual position of a first movable part in a control valve that controls the flow rate of a working fluid according to the position of the first movable part and an actual position of a second movable part that changes its position according to the flow rate of the working fluid, and a step of estimating an abnormality in a detection part for detecting the position of the second movable part based on the detection information acquired in the acquiring step. According to this aspect, since it is possible to estimate an abnormality in the detection part for detecting the position of the second movable part in the control valve, it is possible to take an appropriate response according to the state of the control valve.

1. Management system, 2. Ship, 10. Pilot valve, 12. Spool, 16. Port, 18. Spool drive unit, 19. Position sensor, 20. Main valve, 28. Spool, 29. Position sensor, 48. Hydraulic oil, 80. Engine, 81. Cylinder, 82. Sensor, 90. Log data storage device, 91. Engine control device, 92. Junction box, 100. Hydraulic servo valve, 1 10. Servo valve control device, 120. Power supply circuit, 130. Servo valve control circuit, 140. Servo valve drive circuit, 150. Voltage detection unit, 160. Data collection circuit, 300. Learning device, 301. Learning data acquisition unit, 302. Estimation model generation unit, 303. Estimation model provision unit, 310. Pilot valve actual spool position, 311. Main valve actual spool position, 315. Feature amount calculation unit, 316 Pilot valve target spool position, 320: data selection unit, 341: coil applied current, 342: actual measured value of hydraulic oil cleanliness, 343: hydraulic oil cleanliness estimation model generation unit, 344: hydraulic oil cleanliness estimation model provision unit, 351: position sensor disconnection data, 353: position sensor disconnection detection model generation unit, 354: position sensor disconnection detection model provision unit, 400: state estimation device, 401: detection information acquisition unit, 402: state State estimation unit, 403... Estimation result output unit, 410... Actual pilot valve spool position, 411... Actual main valve spool position, 414... Feature amount calculation unit, 415... Pilot valve target spool position, 417... Data selection unit, 441... Coil applied current, 442... Hydraulic oil cleanliness estimation unit, 443... Hydraulic oil cleanliness estimation result output unit, 451... Position sensor disconnection detection unit, 452... Position sensor disconnection detection result output unit.

Claims (11)

an acquisition unit that acquires detection information of a data set including a target position and an actual position of a first movable part in a control valve that controls a flow rate of a working fluid in accordance with a position of the first movable part and a value of a current supplied to a drive unit that drives the first movable part;
an estimation unit that estimates a cleanliness level of the working fluid based on the detection information acquired by the acquisition unit ,
The degree of cleanliness is an index that increases as the amount of foreign matter contained in the working fluid decreases,
The estimation unit estimates the cleanliness of the working fluid using an estimation criterion for estimating the cleanliness of the working fluid based on the detection information of the data set;
the acquisition unit acquires detection information of the data set in a usage environment similar to a usage environment of the control valve or a control valve of the same type as the control valve when the value of the data set used to generate the estimation criterion was recorded;
The operating environment is a situation in which the control valve or a control target of the same type as the control valve is stopped and an operation to prevent the first movable part from sticking is being performed.
State estimator. 2. The state estimation device according to claim 1, wherein the estimation standard is generated based on a relationship between a value of the data set when the control valve or a control valve of the same type as the control valve is used within a specified period of time and a cleanliness of the working fluid actually measured within the specified period of time. the estimation criterion is generated based on a relationship between a feature amount calculated from values of the data set and an actual measurement value of the cleanliness of the working fluid,
The state estimating device according to claim 2 , wherein the estimating unit calculates the feature amount from the detection information acquired by the acquiring unit, and estimates the cleanliness of the working fluid based on the calculated feature amount. The state estimating device according to claim 3 , wherein the characteristic amount is a speed or an acceleration of the first movable part. The state estimating device according to claim 1 , further comprising a notification unit that notifies a user when the cleanliness of the working fluid estimated by the estimation unit is equal to or lower than a certain level. The state estimating device according to claim 5 , wherein the notification unit estimates a time until the cleanliness of the working fluid falls below a threshold value and notifies the user. 7. The state estimating device according to claim 1, wherein the control valve is a control valve for adjusting the amount of fuel supplied to the engine. the control valve is provided for each cylinder of a multi-cylinder engine to adjust the amount of fuel supplied to each cylinder,
The state estimating device according to claim 7 , wherein the estimating unit determines whether or not there is an abnormality in the working fluid by comparing information relating to each of a plurality of control valves provided in the plurality of cylinders. a first movable portion whose position changes in response to a control signal for designating a position, and whose flow rate of the working fluid is controlled in response to the position;
an acquisition unit that acquires detection information of a data set including a target position and an actual position of the first movable part and a value of a current supplied to a drive unit that drives the first movable part;
an estimation unit that estimates a cleanliness level of the working fluid based on the detection information acquired by the acquisition unit ,
The degree of cleanliness is an index that increases as the amount of foreign matter contained in the working fluid decreases,
The estimation unit estimates the cleanliness of the working fluid using an estimation criterion for estimating the cleanliness of the working fluid based on the detection information of the data set;
the acquisition unit acquires detection information of the data set in a usage environment similar to a usage environment of the control valve or a control valve of the same type as the control valve when the value of the data set used to generate the estimation criterion was recorded;
The usage environment is a situation in which the control valve or a controlled object of the same type of control valve as the control valve is stopped and an operation to prevent sticking of the first movable part is performed . Computer,
an acquisition unit that acquires detection information of a data set including a target position and an actual position of a first movable part in a control valve that controls a flow rate of a working fluid in accordance with a position of the first movable part and a value of a current supplied to a drive unit that drives the first movable part;
an estimation unit that estimates the cleanliness of the working fluid based on the detection information acquired by the acquisition unit ;
The degree of cleanliness is an index that increases as the amount of foreign matter contained in the working fluid decreases,
The estimation unit estimates the cleanliness of the working fluid using an estimation criterion for estimating the cleanliness of the working fluid based on the detection information of the data set;
the acquisition unit acquires detection information of the data set in a usage environment similar to a usage environment of the control valve or a control valve of the same type as the control valve when the value of the data set used to generate the estimation criterion was recorded;
A state estimating program in which the usage environment is a situation in which the control valve or a control target of the same type as the control valve is stopped and an operation to prevent sticking of the first movable part is being performed . On the computer,
A step of acquiring detection information of a data set including a target position and an actual position of a first movable part in a control valve that controls a flow rate of a working fluid in accordance with a position of the first movable part and a value of a current supplied to a drive part that drives the first movable part;
and a step of estimating the cleanliness of the working fluid based on the detection information acquired in the acquiring step .
The degree of cleanliness is an index that increases as the amount of foreign matter contained in the working fluid decreases,
the estimating step includes estimating a cleanliness level of the working fluid using an estimation criterion for estimating a cleanliness level of the working fluid based on sensor information of the data set;
The step of acquiring includes acquiring detection information of the data set in a usage environment similar to the usage environment of the control valve or a control valve of the same type as the control valve when the values of the data set used to generate the estimation criterion were recorded;
The state estimation method according to the present invention, wherein the usage environment is a situation in which the control valve or a control target of the same type as the control valve is stopped and an operation to prevent sticking of the first movable part is being performed .

Images