Learning model generation device, learning model generation program, and flow rate estimation system

Abstract

To improve the estimation accuracy of a drain flow rate in a drain trap.SOLUTION: A server device 20 includes: a storage unit 21 that stores site data 211 in which vibration information of a steam trap in the site is associated with a drain flow rate of the steam trap when the vibration information was measured; a data generation unit 222 that extracts the plurality of pieces of site data 211 from the storage unit 21 and generates teacher data with the vibration information in the site data 221 as input and the drain flow rate as output; and a model generation unit 223 that generates an estimation model that has machine-learned the teacher data. The site data 221 is further associated with information being specific site data in which the vibration information falls within a specific range having an upper limit value and a lower limit value or non-specific site data in which the vibration information does not fall within the specific range. The data generation unit 222 extracts the site data 221 such that the ratio of specific site data in the site data 221 extracted from the storage unit 21 becomes a predetermined ratio at which an accuracy rate of the estimation model becomes equal to or greater than a target accuracy rate.SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] The technology of the present disclosure relates to a learning model generation device, a learning model generation program, and a flow rate estimation system.

Background Art

[0002] Conventionally, an apparatus for estimating the flow rate of a fluid passing through a valve such as a drain trap has been known. For example, the estimation apparatus disclosed in Patent Document 1 detects the vibration of a valve and estimates the actual flow rate from the correlation between the flow rate and the vibration stored in advance. Further, the measuring apparatus disclosed in Patent Document 2 determines the operating condition of the valve from the relationship between the vibration of the valve such as a drain trap and the amount of steam leakage.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] By the way, in the estimation apparatus such as Patent Document 1 described above, since the correlation between the flow rate and the vibration is mainly generated based on experimental data, the amount of data is small, and in particular, the data corresponding to various environmental conditions is small. Therefore, the estimation accuracy of the flow rate is not sufficient, and there is a desire to improve the estimation accuracy. Further, in the measuring apparatus such as Patent Document 2, when detecting the vibration generated when the drain flow rate is large, there is a risk of misjudging that the state is a malfunction (that is, steam is leaking) even though the drain is being discharged normally. In such a case, if the flow rate of the drain can be grasped, misjudgment regarding the operating condition can be suppressed. From this also, high estimation accuracy of the flow rate is desired.

[0005] The technology of the present disclosure has been made in view of such circumstances, and its object is to improve the estimation accuracy of the drain flow rate in a drain trap.

Means for Solving the Problems

[0006] The learning model generation device of the present disclosure includes a storage unit, a data generation unit, and a model generation unit. The storage unit accumulates on-site data in which the vibration information of the drain trap measured by a diagnostician at the site of the drain trap is associated with the drain flow rate of the drain trap determined by the diagnostician when measuring the vibration information. The data generation unit extracts a plurality of the on-site data from the storage unit and generates the plurality of on-site data as teacher data having the vibration information in the on-site data as an input and the drain flow rate as an output. The model generation unit generates a learning model obtained by machine learning the teacher data. The on-site data in the storage unit is further associated with information for identifying whether the on-site data is specific on-site data in a specific range where the vibration information has an upper limit value and a lower limit value greater than zero, or non-specific on-site data where the vibration information is not in the specific range. The data generation unit extracts the plurality of on-site data so that the ratio of the specific on-site data in the plurality of on-site data extracted from the storage unit becomes a predetermined ratio at which the correct answer rate of the learning model is equal to or higher than a preset target correct answer rate.

[0007] Further, the flow rate estimation system of the present disclosure estimates the drain flow rate of a drain trap to be estimated. The flow rate estimation system includes the above-described learning model generation device and an estimation device. The estimation device has a detection unit that detects the vibration information of the drain trap to be estimated, and an estimation unit that outputs an estimated drain flow rate by inputting the vibration information detected by the detection unit into the learning model generated by the learning model generation device.

[0008] In addition, the learning model generation program of the present disclosure causes a computer to realize a function of generating a learning model for estimating the drain flow rate of a drain trap. The learning model generation program accumulates on-site data in which vibration information of the drain trap measured by a diagnostician at the site of the drain trap is associated with the drain flow rate of the drain trap determined by the diagnostician when measuring the vibration information, and extracts a plurality of pieces of on-site data from the accumulated on-site data, and generates the plurality of pieces of on-site data as teacher data having the vibration information in the on-site data as an input and the drain flow rate as an output, and generates the learning model obtained by machine-learning the teacher data. The accumulated on-site data is further associated with information for identifying whether the on-site data is specific on-site data in a specific range where the vibration information has an upper limit value and a lower limit value greater than zero, or non-specific on-site data where the vibration information is not in the specific range, and the computer is caused to realize a function of extracting the plurality of pieces of on-site data such that the ratio of the specific on-site data in the plurality of pieces of on-site data extracted from the accumulated on-site data becomes a predetermined ratio at which the correct answer rate of the learning model is equal to or higher than a preset target correct answer rate.

Effect of the Invention

[0009] According to the learning model generation device described above, the estimation accuracy of the drain flow rate in the drain trap can be improved.

[0010] According to the flow rate estimation system described above, the estimation accuracy of the drain flow rate in the drain trap can be improved.

[0011] According to the learning model generation program described above, the estimation accuracy of the drain flow rate in the drain trap can be improved.

Brief Description of the Drawings

[0012]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Mode for Carrying Out the Invention

[0013] Hereinafter, exemplary embodiments will be described in detail with reference to the drawings. FIG. 1 is a diagram showing a schematic configuration of the flow rate estimation system 100.

[0014] The flow rate estimation system 100 estimates, for example, the drain flow rate in the steam trap 2 provided in a steam system or the like. The steam trap 2 is an example of a drain trap and is provided, for example, in the drain pipe 1. The steam trap 2 is a so-called automatic valve that allows the drain to flow out to the downstream side when the drain flows in from the drain pipe 1, while preventing the outflow of the steam when the steam flows in from the drain pipe 1.

[0015] The flow rate estimation system 100 includes an estimation device 10 and a server device 20. The estimation device 10 and the server device 20 can communicate with each other via a network N. The network N is a wide area communication network such as the Internet.

[0016] The estimation device 10 is a portable device that estimates the drain flow rate in the steam trap 2, that is, the drain flow rate passing through the steam trap 2. Specifically, the estimation device 10 has a device main body 11 and a probe 12.

[0017] The probe 12 is an example of a detection unit that detects vibration information (for example, vibration level) of the steam trap 2 to be estimated. The probe 12 detects the vibration information of the steam trap 2 by being pressed against the casing of the steam trap 2, for example. The probe 12 is connected to the device main body 11 via a cable 13. The vibration information detected by the probe 12 is output to the device main body 11 via the cable 13. The steam trap 2 to be estimated is the steam trap 2 for which the drain flow rate is estimated.

[0018] Note that the device main body 11 and the probe 12 may be integrally formed. Also, the device main body 11 and the probe 12 may be wirelessly connected by a wireless communication standard such as Bluetooth (registered trademark).

[0019] FIG. 2 is a block diagram showing a schematic configuration of the estimation device 10. The device main body 11 of the estimation device 10 estimates the drain flow rate in the steam trap 2 based on the vibration information detected by the probe 12. The device main body 11 can communicate with the server device 20 via the network N. Specifically, the device main body 11 has an input unit 111, a storage unit 112, a display unit 113, and an estimation unit 114.

[0020] The input unit 111 receives an input operation from a diagnostician who is a user. The input unit 111 outputs an input signal corresponding to the input operation. The input unit 111 is, for example, an input key or a touch panel superimposed on the display unit 113 described later.

[0021] The storage unit 112 is a storage medium readable by a computer (in this example, the estimation unit 114) that stores various programs and various data. The storage unit 112 is formed by a magnetic disk such as a hard disk, an optical disk such as a CD-ROM and a DVD, or a semiconductor memory. The storage unit 112 stores the estimation model M generated by the server device 20 and the like. Further, the storage unit 112 also stores the drain flow rate and the like output by the estimation unit 114.

[0022] The display unit 113 displays the drain flow rate output by the estimation unit 114 and the confidence level of the drain flow rate. The confidence level of the drain flow rate is an example of information indicating the correct answer rate of the estimation model M stored in the storage unit 112. The display unit 113 is, for example, a liquid crystal display or an organic EL display.

[0023] The estimation unit 114 inputs the vibration information detected by the probe 12 into the estimation model M generated by the server device 20, and outputs the estimated drain flow rate. More specifically, the estimation unit 114 inputs the vibration information detected by the probe 12 and the pressure of the steam trap 2 to be estimated when the probe 12 detects into the estimation model M, and outputs the estimated drain flow rate. The estimation unit 114 has various processors and various semiconductor memories.

[0024] Figure 3 is a block diagram showing a schematic configuration of the server device 20. The server device 20 accumulates the on-site data 211 including the vibration information and the drain flow rate of the steam trap 2 acquired at the site S, and generates and updates the estimation model M using the accumulated on-site data 211. The estimation model M is an example of a learning model using machine learning. The server device 20 is, for example, a cloud server and is an example of a learning model generation device.

[0025] Specifically, the server device 20 has a storage unit 21 and a processing unit 22.

[0026] The storage unit 21 is a computer-readable storage medium (in this example, the processing unit 22) that stores various programs and various data. The storage unit 21 is formed by a magnetic disk such as a hard disk, an optical disk such as a CD-ROM and a DVD, or a semiconductor memory.

[0027] Specifically, the storage unit 21 stores on-site data 211, teacher data sets 212, generation programs 213, and the like. That is, the storage unit 21 is an example of a storage unit in which on-site data 211 accumulates.

[0028] FIG. 4 is a diagram showing an example of on-site data 211 in the storage unit 21 of the server device 20. The on-site data 211 is data in which the vibration information of the steam trap 2 measured by the diagnostician at the site S of the steam trap 2 is associated with the drain flow rate of the steam trap 2 determined by the diagnostician when measuring the vibration information. More specifically, the on-site data 211 is further associated with the pressure of the steam trap 2 when the diagnostician measures the vibration information. This pressure is, for example, the inlet pressure of the steam trap 2 measured by a pressure gauge provided in the drain pipe 1 on the inlet side. The site S of the steam trap 2 is a place where the steam trap 2 is actually installed and operating.

[0029] During the inspection or maintenance of the steam trap 2 at each site S, the diagnostician diagnoses the operating state of the steam trap 2 and reports the diagnosis results to the customer, etc. For example, in order to diagnose the steam trap 2, the diagnostician measures the vibration information (hereinafter also simply referred to as "vibration information") of the steam trap 2 using a vibration measuring instrument. In addition, the diagnostician determines the drain flow rate based on the operating state of the steam trap 2. The operating state of the steam trap 2 is, for example, the operating sound of the steam trap 2 or the flow state of the drain. The operating sound of the steam trap 2 is measured, for example, using a stethoscope. The flow state of the drain is grasped, for example, by visually observing the inside of the drain pipe 1 connected to the downstream of the steam trap 2. The diagnostician determines the drain flow rate from these operating states of the steam trap 2. In this way, various information regarding the steam trap 2 acquired at each site S is sequentially accumulated in the storage unit 21 as on-site data 211.

[0030] The drain flow rate is expressed not by a numerical value but by a degree. In this example, the degree of the drain flow rate is expressed in three levels: "large", "medium", and "small". Note that the degree of the drain flow rate may be expressed in two levels or four levels or more. Also, although it is conceivable to measure the drain flow rate using a flow measuring instrument, since the flow measuring instrument is expensive and the installation work of the flow measuring instrument to the drain pipe 1, etc. is troublesome, it is effective for the diagnostician to determine the drain flow rate.

[0031] FIG. 5 is a graph showing the concept of the specific range RC of the vibration information. The on-site data 211 is further associated with information for identifying whether it is specific on-site data or non-specific on-site data. The specific on-site data is on-site data in which the vibration information is in a specific range RC having an upper limit value f2 and a lower limit value f1 greater than zero. The non-specific on-site data is on-site data in which the vibration information is not in the specific range RC, that is, on-site data in the non-specific range RA or the non-specific range RB. In this example, as shown in FIG. 4, in the case of specific on-site data, it is represented by "1", and in the case of non-specific on-site data, it is represented by "0".

[0032] As shown in FIG. 5, the correlation between the vibration information corresponding to the drain flow rate and the pressure is predetermined by experiments or the like. For example, the correlation between the vibration information corresponding to the drain flow rate "small" and the pressure is shown by graph Gb, and the correlation between the vibration information corresponding to the drain flow rate "large" and the pressure is shown by graph Gc. Note that graph Gd is the correlation between the vibration information corresponding to the drain flow rate "medium". The lower limit value f1 is the vibration information corresponding to the predetermined pressure Pa in graph Gb and is greater than zero. The upper limit value f2 is the vibration information corresponding to the predetermined pressure Pa in graph Gc.

[0033] The vibration information in graphs Gb, Gc, and Gd is a threshold value for determining the operating state of the steam trap 2. For example, in the steam trap 2 with a "small" drain flow rate, when the measured vibration information is less than the lower limit value f1, that is, in the non-specific range RA, the steam trap 2 is determined to be in a normal state. When the measured vibration information is greater than or equal to the lower limit value f1, the steam trap 2 is determined to be in an abnormal state (that is, a state where steam leakage has occurred, the same applies hereinafter). When it is determined to be in an abnormal state, the steam leakage amount is estimated based on graph Ga showing the correlation between the vibration information and the steam leakage amount. Graph Ga is predetermined by experiments or the like. On the other hand, in the steam trap 2 with a "large" drain flow rate, when the measured vibration information is less than or equal to the upper limit value f2, the steam trap 2 is determined to be in a normal state. When the measured vibration information is greater than the upper limit value f2, that is, in the non-specific range RB, the steam trap 2 is determined to be in an abnormal state. When it is determined to be in an abnormal state, the steam leakage amount is estimated based on graph Ga.

[0034] Thus, when the measured vibration information is within the non-specific range RA, the steam trap 2 is presumed to be in a normal state regardless of the drain flow rate. When the measured vibration information is within the non-specific range RB, the steam trap 2 is presumed to be in an abnormal state regardless of the drain flow rate. Also, when the measured vibration information is within the specific range RC, the determination as to whether the steam trap 2 is in a normal state or an abnormal state changes depending on the drain flow rate. That is, the specific range RC is a region where a more stringent estimation of the drain flow rate is required than in the non-specific ranges RA and RB. Put another way, the specific range RC is the intermediate band within the band of vibration information that can occur in the steam trap 2 to be estimated.

[0035] In the storage unit 21, a plurality of on-site data 211 associating such vibration information, pressure, drain flow rate, and information on whether the vibration information is within the specific range RC are classified for each type of the steam trap 2 (see "Type A", "Type B", and "Type C" shown in FIG. 4).

[0036] The teacher data set 212 means a collection of a plurality of teacher data. The teacher data is data for training the estimation model M, and is also referred to as learning data or training data. The teacher data has vibration information as an input and drain flow rate as an output, and more specifically, has vibration information and pressure as inputs and drain flow rate as an output. That is, the teacher data is data in which vibration information and pressure are paired with the drain flow rate.

[0037] The generation program 213 is an example of a learning model generation program. The generation program 213 is a program for causing a computer (in this example, the processing unit 22) to realize various functions for generating the estimation model M. The generation program 213 is read and executed by the processing unit 22.

[0038] The processing unit 22 generates teacher data and generates the estimation model M using the teacher data. The processing unit 22 includes various processors and various semiconductor memories. Specifically, the processing unit 22 has a setting unit 221, a data generation unit 222, and a model generation unit 223 as functional blocks.

[0039] The setting unit 221 performs various settings according to the user's input operation. Specifically, the setting unit 221 sets the number of pieces of on-site data 211 to be extracted, the ratio of specific on-site data (specifically, the initial ratio), and the target correct rate of the estimation model M. The number of pieces of on-site data 211 to be extracted is the number of pieces of on-site data 211 to be extracted from the storage unit 21. The ratio of specific on-site data is the ratio of specific on-site data among the plurality of pieces of on-site data 211 extracted from the storage unit 21.

[0040] The data generation unit 222 extracts a plurality of pieces of on-site data 211 from the storage unit 21 and generates the plurality of pieces of on-site data 211 as teacher data. The data generation unit 222 extracts the plurality of pieces of on-site data 211 such that the ratio of specific on-site data among the extracted plurality of pieces of on-site data 211 becomes a predetermined ratio at which the correct rate of the estimation model M is equal to or higher than the target correct rate set in advance.

[0041] Specifically, the data generation unit 222 extracts on-site data 211 from the storage unit 21 based on the number of extractions and the initial ratio set in the setting unit 221 and generates it as teacher data. Finally, the ratio of specific on-site data becomes a predetermined ratio at which the correct rate of the estimation model M is equal to or higher than the target correct rate set in the setting unit 221. That is, the data generation unit 222 generates teacher data while changing the ratio of specific on-site data so that the correct rate of the estimation model M is equal to or higher than the target correct rate. The teacher data generated in this way is stored in the storage unit 21 as a teacher data set 212.

[0042] The model generation unit 223 generates an estimated model M obtained by performing machine learning on the teacher data. FIG. 6 is a conceptual diagram showing the input and output of the estimated model M. Specifically, the model generation unit 223 reads out the teacher data set 212 from the storage unit 21, and generates a trained estimated model M obtained by performing machine learning on each piece of teacher data included in the teacher data set 212. The estimated model M takes vibration information and pressure as inputs, and outputs the estimated drain flow rate. Various known methods can be used for machine learning, for example, methods such as neural networks, reinforcement learning, or deep learning can be used. For machine learning, supervised learning or semi-supervised learning is used.

[0043] Next, the generation process of the estimated model M in the server device 20 will be described in detail with reference to FIG. 7. FIG. 7 is a flowchart showing the generation process of the estimated model M by the server device 20.

[0044] First, in step Sa1, the storage unit 21 sequentially accumulates various information of the steam trap 2 acquired at the site S as on-site data 211. In the subsequent step Sa2, the setting unit 221 sets the number of extractions of the on-site data 211, the ratio of specific on-site data (specifically, the initial ratio), and the target accuracy rate of the estimated model M according to the input operation of the user.

[0045] In the subsequent step Sa3, the data generation unit 222 generates teacher data. Specifically, the data generation unit 222 extracts the on-site data 211 from the storage unit 21 based on the number of extractions and the initial ratio set in the setting unit 221, and generates it as teacher data. In the subsequent step Sa4, the teacher data generated by the data generation unit 222 is stored in the storage unit 21. That is, the storage unit 21 stores the teacher data as the teacher data set 212.

[0046] In the subsequent step Sa5, the model generation unit 223 generates the estimation model M. Specifically, the model generation unit 223 reads out the teacher data set 212 from the storage unit 21. Then, the model generation unit 223 generates the estimation model M obtained by performing machine learning using each piece of teacher data in the teacher data set 212.

[0047] In the subsequent step Sa6, the model generation unit 223 determines whether the accuracy rate of the estimation model M is equal to or higher than the target accuracy rate. If the accuracy rate of the estimation model M is equal to or higher than the target accuracy rate, it is considered that the estimation model M has been appropriately generated, and the process proceeds to step Sa7. The estimation model M is stored in the storage unit 21, and the generation process of the estimation model M ends.

[0048] In step Sa6, if it is determined that the accuracy rate of the estimation model M is less than the target accuracy rate, the process proceeds to step Sa8. In step Sa8, the setting unit 221 changes the ratio (initial ratio) of the specific site data in response to the user's input operation. Specifically, the setting unit 221 sets the ratio of the specific site data to a value larger than the initial ratio.

[0049] In the subsequent step Sa9, the data generation unit 222 regenerates the teacher data. Specifically, the data generation unit 222 extracts the site data 211 from the storage unit 21 again so that the ratio of the specific site data becomes the ratio newly set by the setting unit 221. That is, the data generation unit 222 newly extracts the site data 211 by the number of extraction counts set in the setting unit 221. The data generation unit 222 regenerates new teacher data using the newly extracted plurality of site data 211. As a result, teacher data in which the site data in the region where estimation of severe drain flow rate is required is more taken into account is generated.

[0050] In the subsequent step Sa10, the teacher data regenerated by the data generation unit 222 is stored in the storage unit 21. That is, the storage unit 21 stores (i.e., overwrites) the new teacher data as the teacher data set 212. Thus, when the new teacher data set 212 is stored, the process returns to step Sa5 again, and a new estimation model M trained with the new teacher data set 212 is generated. In the subsequent step Sa6, if it is determined that the accuracy rate of the new estimation model M is equal to or higher than the target accuracy rate, the process proceeds to step Sa7; if it is determined that the accuracy rate of the new estimation model M is less than the target accuracy rate, the process proceeds to step Sa8 again.

[0051] In this way, steps Sa8 to 10 are repeated until the accuracy rate of the estimation model M becomes equal to or higher than the target accuracy rate. That is, the data generation unit 222 extracts the on-site data 211 from the storage unit 21 while changing (specifically, increasing) the ratio of the specific on-site data until the accuracy rate of the estimation model M becomes equal to or higher than the target accuracy rate, and generates teacher data.

[0052] Next, the drain flow rate estimation process in the estimation device 10 will be described in detail with reference to FIG. 8. FIG. 8 is a flowchart showing the drain flow rate estimation process by the estimation device 10. FIG. 9 is a diagram showing an example of the display mode on the display unit 113 of the estimation device 10.

[0053] In the flow rate estimation system 100, when a diagnostician operates the estimation device 10, the estimation device 10 downloads the estimation model M from the storage unit 21 of the server device 20 via the network N and stores it in the storage unit 112. More specifically, the storage unit 112 stores a plurality of types of estimation models M corresponding to each type of the steam trap 2.

[0054] First, in step Sb1, the diagnostician inputs the pressure and type into the estimation device 10. Specifically, the diagnostician reads the pressure (i.e., the inlet pressure) of the steam trap 2 to be estimated from, for example, a pressure gauge provided in the drain pipe 1, and inputs the read pressure into the input unit 111. Further, the diagnostician inputs the type of the steam trap 2 to be estimated into the input unit 111.

[0055] In the subsequent step Sb2, the diagnostician detects the vibration information of the steam trap 2 to be estimated by the estimation device 10. Specifically, when the diagnostician presses the probe 12 against the casing of the steam trap 2, the probe 12 detects the vibration information of the steam trap 2. The vibration information detected by the probe 12 is output to the device main body 11.

[0056] In the subsequent step Sb3, the estimation unit 114 estimates the drain flow rate of the steam trap 2 to be estimated. Specifically, the estimation unit 114 reads out the estimation model M corresponding to the type of the steam trap 2 input by the input unit 111 from the storage unit 112. Then, the estimation unit 114 inputs the vibration information output from the probe 12 and the pressure input by the input unit 111 into the estimation model M read out from the storage unit 112, and outputs the estimated drain flow rate (specifically, the degree of the drain flow rate).

[0057] In the subsequent step Sb4, the degree of the drain flow rate output by the estimation unit 114 is displayed on the display unit 113 as shown in FIG. 9. In this example, it is displayed that the degree of the drain flow rate is "large". Also, in this step Sb4, the confidence level of the drain flow rate (the degree of the drain flow rate) output by the estimation unit 114 is also displayed on the display unit 113. That is, the correct answer rate of the estimation model M read out by the estimation unit 114 from the storage unit 112 is displayed on the display unit 113. In this example, it is displayed that the confidence level is "80%".

[0058] In the subsequent step Sb5, the storage unit 112 stores the drain flow rate and the like estimated by the estimation unit 114. Specifically, the drain flow rate output by the estimation unit 114 is associated with the vibration information and the pressure input to the estimation model M when the estimation unit 114 outputs the drain flow rate, and is stored in the storage unit 112. At this time, information on the type of the steam trap 2 and whether the vibration information is within the specific range RC is also associated with the drain flow rate and the like and stored in the storage unit 112.

[0059] In the subsequent step Sb6, the on-site data 211 in the storage unit 21 of the server device 20 is updated. The drain flow rate, vibration information, etc. stored in the storage unit 112 in step Sb5 are accumulated as new on-site data in the storage unit 21 of the server device 20 via, for example, the network N. Therefore, the on-site data 211 in the storage unit 21 is updated with highly accurate data. When step Sb6 is thus completed, the estimation process of the drain flow rate ends.

[0060] As described above, the server device 20 (learning model generation device) includes a storage unit 21 (accumulation unit), a data generation unit 222, and a model generation unit 223. The storage unit 21 accumulates on-site data 211 in which the vibration information of the steam trap 2 (drain trap) measured by the diagnostician at the on-site S of the steam trap 2 is associated with the drain flow rate of the steam trap 2 determined when the diagnostician measures the vibration information. The data generation unit 222 extracts a plurality of on-site data 211 from the storage unit 21 and generates the plurality of on-site data 211 as teacher data having the vibration information in the on-site data 211 as an input and the drain flow rate as an output. The model generation unit 223 generates an estimation model M (learning model) obtained by machine-learning the teacher data. The on-site data 211 in the storage unit 21 is further associated with information for identifying whether it is specific on-site data within a specific range RC in which the vibration information has an upper limit value f2 and a lower limit value f1 greater than zero, or non-specific on-site data in which the vibration information is not within the specific range RC. The data generation unit 222 extracts a plurality of on-site data 211 such that the ratio of the specific on-site data in the plurality of on-site data 211 extracted from the storage unit 21 becomes a predetermined ratio at which the correct answer rate of the estimation model M is equal to or higher than a preset target correct answer rate.

[0061] Further, the flow rate estimation system 100 estimates the drain flow rate of the steam trap 2 to be estimated, and includes an estimation device 10 and the aforementioned server device 20. The estimation device 10 has a probe 12 (detection unit) that detects vibration information of the steam trap 2 to be estimated, and an estimation unit 114 that outputs an estimated drain flow rate by inputting the vibration information detected by the probe 12 into an estimation model M generated by the server device 20.

[0062] Also, the generation program 213 (learning model generation program) causes a computer to realize a function of generating an estimation model M (learning model) for estimating the drain flow rate of the steam trap 2. The generation program 213 has a function of accumulating on-site data 211 in which the vibration information of the steam trap 2 measured by a diagnostician at the site S of the steam trap 2 is associated with the drain flow rate of the steam trap 2 determined by the diagnostician when measuring the vibration information, a function of extracting a plurality of on-site data 211 from the accumulated on-site data 211 and generating the plurality of on-site data 211 as teacher data with the vibration information in the on-site data 211 as the input and the drain flow rate as the output, a function of generating an estimation model M obtained by machine learning the teacher data, and the accumulated on-site data 211 is further associated with information for identifying whether it is specific on-site data in a specific range where the vibration information has an upper limit value f2 and a lower limit value f1 greater than zero, or non-specific on-site data where the vibration information is not in the specific range. A function of extracting a plurality of on-site data 211 so that the ratio of the specific on-site data in the plurality of on-site data 211 extracted from the accumulated on-site data 211 becomes a predetermined ratio such that the correct answer rate of the estimation model M is equal to or higher than a preset target correct answer rate is realized by the computer.

[0063] According to these configurations, teacher data is generated based on the on-site data 211 obtained at each site of the steam trap 2, and the estimation model M is generated using the teacher data. A large amount of on-site data 211 taking various environmental conditions into account is accumulated in the storage unit 21. To generate teacher data based on such on-site data 211, the estimation accuracy of the estimation model M is improved.

[0064] Furthermore, when extracting a plurality of on-site data 211 from the storage unit 21 to generate teacher data, the plurality of on-site data 211 are extracted such that the ratio of the specific on-site data becomes a predetermined ratio at which the correct answer rate of the estimation model M is equal to or higher than a preset target correct answer rate. Therefore, teacher data in which the on-site data 211 in the region where strict drain flow rate estimation is required is more taken into account is generated. That is, teacher data that emphasizes the on-site data 211 of the vibration information in a specific range RC where it is difficult to determine the drain flow rate is generated. Therefore, the estimation accuracy of the drain flow rate is improved substantially evenly in the entire band of the vibration information that can occur in a certain steam trap 2. Therefore, combined with the above-described effects, the estimation accuracy of the drain flow rate can be further improved.

[0065] Also, since the estimation accuracy of the drain flow rate can be improved, with a measuring device such as the above-described Patent Document 2, the drain flow rate can be accurately grasped. Therefore, for example, it is possible to suppress misjudging that the state is a malfunction (that is, steam is leaking) even though the drain is being discharged normally. Therefore, the determination accuracy of the operation quality can be improved.

[0066] In the server device 20, the on-site data 211 in the storage unit 21 is further associated with the pressure of the steam trap 2 when the diagnostician measures the vibration information. The teacher data takes the vibration information and pressure in the on-site data 211 as inputs and outputs the drain flow rate. In the flow rate estimation system 100, the estimation unit 114 inputs the vibration information detected by the probe 12 and the pressure of the steam trap 2 to be estimated when the probe 12 detects it into the estimation model M, and outputs the estimated drain flow rate.

[0067] According to this configuration, the pressure of the steam trap 2 when the vibration information is measured is also added as the on-site data 211. Therefore, the on-site data 211 that more conforms to the environmental conditions of the on-site S can be accumulated in the storage unit 21. Therefore, teacher data and thus the estimation model M that more conforms to the environmental conditions of the on-site can be generated, so that the estimation accuracy of the drain flow rate can be further improved.

[0068] <<Modification Example>> In this modification example, in the above-described embodiment, instead of one type of estimation model M, two types of estimation models (that is, the first estimation model M1 and the second estimation model M2) are generated. Here, the points different from the above-described embodiment will be described in detail.

[0069] FIG. 10 is a block diagram showing a schematic configuration of the server device 20 according to the modification example. FIG. 11 is a block diagram showing a schematic configuration of the estimation device 10 according to the modification example.

[0070] In this modification example, the teacher data includes first teacher data in which the ratio of specific on-site data is a first predetermined ratio, and second teacher data in which the ratio of specific on-site data is a second predetermined ratio larger than the first predetermined ratio. The estimation model M includes a first estimation model M1 obtained by machine learning of the first teacher data and a second estimation model M2 obtained by machine learning of the second teacher data.

[0071] The data generation unit 222 generates each of the first teacher data and the second teacher data based on the flowchart of FIG. 7. That is, the data generation unit 222 extracts a plurality of on-site data 211 from the storage unit 21 so that the ratio of the specific on-site data in the plurality of on-site data 211 becomes a first predetermined ratio at which the accuracy rate of the first estimation model M1 is equal to or higher than a preset target accuracy rate, and generates the first teacher data. Further, the data generation unit 222 extracts a plurality of on-site data 211 from the storage unit 21 so that the ratio of the specific on-site data in the plurality of on-site data 211 becomes a second predetermined ratio at which the accuracy rate of the second estimation model M2 is equal to or higher than a preset target accuracy rate, and generates the second teacher data. As shown in FIG. 10, the generated first teacher data and second teacher data are stored in the storage unit 21 as a first teacher data set 214 and a second teacher data set 215, respectively.

[0072] The model generation unit 223 reads the first teacher data set 214 from the storage unit 21, and generates a first estimation model M1 obtained by performing machine learning on each piece of first teacher data included in the first teacher data set 214. Further, the model generation unit 223 reads the second teacher data set 215 from the storage unit 21, and generates a second estimation model M2 obtained by performing machine learning on each piece of second teacher data included in the second teacher data set 215. The generated first estimation model M1 and second estimation model M2 are stored in the storage unit 21.

[0073] As shown in FIG. 11, the estimation device 10 downloads the first estimation model M1 and the second estimation model M2 from the server device 20 via the network N and stores them in the storage unit 112. More specifically, the storage unit 112 stores a plurality of types of first estimation models M1 and second estimation models M2 corresponding to each type of the steam trap 2. When the vibration information detected by the probe 12 is within the specific range RC, the estimation unit 114 of the estimation device 10 inputs the vibration information and the pressure into the second estimation model M2, outputs the estimated drain flow rate, and when the vibration information detected by the probe 12 is not within the specific range RC, inputs the vibration information and the pressure into the first estimation model M1, and outputs the estimated drain flow rate.

[0074] Specifically, the estimation device 10 estimates the drain flow rate according to the flowchart of FIG. 12. FIG. 12 is a flowchart showing the estimation process of the drain flow rate by the estimation device 10 according to the modified example. Here, the differences from the flowchart of FIG. 8, that is, the points of including steps Sb11, 12, and 13 instead of step Sb3 will be described in detail.

[0075] In step Sb2, when the probe 12 detects the vibration information of the steam trap 2, the process proceeds to step Sb11. In step Sb11, the estimation unit 114 determines whether the vibration information is within a specific range RC. If the vibration information is within the specific range RC, the process proceeds to step Sb12, and the estimation unit 114 reads out the second estimation model M2 corresponding to the type of the steam trap 2 from the storage unit 112. Then, the estimation unit 114 inputs the vibration information and the pressure into the second estimation model M2 to output the estimated drain flow rate. Thus, when the drain flow rate is output, the process proceeds to step Sb4, and the drain flow rate and the like are displayed on the display unit 113. Steps Sb5 and subsequent steps are the same as those in the above embodiment.

[0076] In step Sb11, when it is determined that the vibration information is not within the specific range RC, the process proceeds to step Sb13, and the estimation unit 114 reads out the first estimation model M1 corresponding to the type of the steam trap 2. Then, the estimation unit 114 inputs the vibration information and the pressure into the first estimation model M1 to output the estimated drain flow rate. Thus, when the drain flow rate is output, the process proceeds to step Sb4, and the drain flow rate and the like are displayed on the display unit 113. Steps Sb5 and subsequent steps are the same as those in the above embodiment.

[0077] As described above, in the server device 20 of this modification example, first teacher data and second teacher data with different ratios of specific site data are generated, and a first estimation model M1 and a second estimation model M2 obtained by performing machine learning on each of them are generated. Then, in the estimation device 10, the first estimation model M1 and the second estimation model M2 are selectively used depending on whether the vibration information is within a specific range RC. That is, when the vibration information is within the specific range RC, estimation is performed using the second estimation model M2 based on the second teacher data with a higher ratio of specific site data. When the vibration information is not within the specific range RC, that is, when the vibration information is within the non-specific ranges RA and RB, estimation is performed using the first estimation model M1 based on the first teacher data with a lower ratio of specific site data.

[0078] Therefore, it is possible to estimate the drain flow rate using an appropriate estimation model according to the band of the generated vibration information. Accordingly, it is possible to further improve the estimation accuracy of the drain flow rate in the entire band of the vibration information that can occur in a certain steam trap 2.

[0079] 《Other Embodiments》 As described above, the above embodiments have been described as examples of the technology disclosed in the present application. However, the technology in the present disclosure is not limited to this, and is also applicable to embodiments in which changes, replacements, additions, omissions, etc. are appropriately made. Further, it is also possible to form a new embodiment by combining the respective components described in the above embodiments. In addition, among the components described in the accompanying drawings and the detailed description, not only the components essential for solving the problem but also the components not essential for solving the problem for exemplifying the technology may be included. Therefore, it should not be immediately determined that those non-essential components are essential just because they are described in the accompanying drawings and the detailed description.

[0080] For example, as teacher data, only vibration information may be input omitting pressure, and the drain flow rate (degree of drain flow rate) may be output.

[0081] In addition to the on-site data 211, the storage unit 21 may also store various types of information obtained in a laboratory, for example.

[0082] Also, as the on-site data 211, for example, the temperature of the steam trap 2 may be added, and the above-mentioned temperature may be added as the input of the teacher data.

[0083] In addition, in a modification of the above embodiment, the first teacher data with a lower ratio of specific on-site data may randomly extract a predetermined number of on-site data 211 from the storage unit 21. That is, basically, since the ratio of specific on-site data is small in the on-site data 211 stored in the storage unit 21, when randomly extracted, the ratio of specific on-site data in the extracted on-site data 211 will probabilistically be low. Therefore, even if the on-site data 211 is randomly extracted from the storage unit 21 to generate the first teacher data, the ratio of specific on-site data in the first teacher data will be lower than that of the second teacher data.

Industrial Applicability

[0084] As described above, the technology of the present disclosure is useful for a learning model generation device, a learning model generation program, and a flow rate estimation system.

Explanation of Reference Numerals

[0085] 100 Flow rate estimation system 2 Steam trap (drain trap) 10 Estimation device 12 Probe (detection unit) 114 Estimation unit 20 Server device (learning model generation device) 21 Storage unit (accumulation unit) 211 On-site data 213 Generation program (learning model generation program) 222 Data generation unit 223 Model generation unit S Site f1 Lower limit value f2 Upper limit value RC specific range

Claims

1. An accumulation unit that accumulates on-site data associating the vibration information of the drain trap measured by a diagnostician at the site of the drain trap with the drain flow rate of the drain trap determined by the diagnostician when measuring the vibration information; A data generation unit that extracts a plurality of the on-site data from the accumulation unit and generates the plurality of on-site data as teacher data that takes the vibration information in the on-site data as an input and outputs the drain flow rate; A model generation unit that generates a learning model obtained by machine learning the teacher data; The on-site data in the accumulation unit is further associated with information for identifying whether the on-site data is specific on-site data in a specific range where the vibration information has an upper limit value and a lower limit value greater than zero, or non-specific on-site data where the vibration information is not in the specific range; The data generation unit extracts the plurality of on-site data such that the ratio of the specific on-site data in the plurality of on-site data extracted from the accumulation unit becomes a predetermined ratio at which the correct answer rate of the learning model is equal to or higher than a preset target correct answer rate; A learning model generation device characterized by the above.

2. In the learning model generation device according to Claim 1, The teacher data includes first teacher data in which the ratio of the specific on-site data is a first predetermined ratio and second teacher data in which the ratio of the specific on-site data is a second predetermined ratio greater than the first predetermined ratio; The learning model includes a first learning model obtained by machine learning the first teacher data and a second learning model obtained by machine learning the second teacher data; A learning model generation device characterized by the above.

3. In the learning model generation device according to Claim 1 or 2, The on-site data in the accumulation unit is further associated with the pressure of the drain trap when the diagnostician measures the vibration information; The teacher data takes the vibration information and pressure in the on-site data as inputs and outputs the drain flow rate; A learning model generation device characterized by the above.

4. A flow rate estimation system for estimating the drain flow rate of a drain trap to be estimated, The learning model generation device according to Claim 1; A detection unit that detects the vibration information of the drain trap to be estimated, and an estimation device that has an estimation unit that inputs the vibration information detected by the detection unit into the learning model generated by the learning model generation device and outputs the estimated drain flow rate; A flow rate estimation system characterized by the above.

5. In the flow rate estimation system according to claim 4, the teacher data includes first teacher data in which the ratio of the specific site data is a first predetermined ratio, and second teacher data in which the ratio of the specific site data is a second predetermined ratio greater than the first predetermined ratio, the learning model includes a first learning model obtained by performing machine learning on the first teacher data and a second learning model obtained by performing machine learning on the second teacher data, when the vibration information detected by the detection unit is within the specific range, the estimation unit inputs the vibration information into the second learning model to output an estimated drain flow rate, and when the vibration information detected by the detection unit is not within the specific range, the estimation unit inputs the vibration information into the first learning model to output an estimated drain flow rate A flow rate estimation system characterized by the above.

6. In the flow rate estimation system according to claim 4 or 5, the site data in the storage unit is further associated with the pressure of the drain trap when the diagnostician measures the vibration information, the teacher data takes the vibration information and pressure in the site data as inputs and outputs the drain flow rate, the estimation unit inputs the vibration information detected by the detection unit and the pressure of the drain trap to be estimated when the detection unit detects it into the learning model to output an estimated drain flow rate A flow rate estimation system characterized by the above.

7. A learning model generation program for causing a computer to realize a function of generating a learning model for estimating the drain flow rate of a drain trap, a function of accumulating site data in which the vibration information of the drain trap measured by a diagnostician at the site of the drain trap is associated with the drain flow rate of the drain trap determined when the diagnostician measures the vibration information, a function of extracting a plurality of site data from the accumulated site data and generating the plurality of site data as teacher data that takes the vibration information in the site data as an input and outputs the drain flow rate, a function of generating the learning model obtained by performing machine learning on the teacher data, the stored site data is further associated with information for identifying whether the site data is specific site data in which the vibration information is within a specific range having an upper limit value and a lower limit value greater than zero, or non-specific site data in which the vibration information is not within the specific range The computer is caused to implement a function of extracting the plurality of on-site data such that the ratio of the specific on-site data in the plurality of on-site data extracted from the stored on-site data becomes a predetermined ratio at which the correct answer rate of the learning model is equal to or higher than a preset target correct answer rate. A learning model generation program characterized by the above.

Images