Diagnostic device, diagnostic method, and diagnostic program

Abstract

To improve the diagnostic accuracy of flow meters. [Solution] The server device 10 acquires image data from the measuring tube MP of the flow meter 20, calculates an image data evaluation value that shows the difference from the normal state of the flow meter 20 based on the acquired image data, and diagnoses the flow meter 20 using a machine learning model that outputs state information indicating the state of the flow meter 20 in response to the input of the image data evaluation value.

Description

[Technical Field]

[0001] This disclosure relates to a diagnostic device, a diagnostic method, and a diagnostic program. [Background technology]

[0002] Regarding flow meters, which are a type of field instrument (e.g., electromagnetic flow meters, vortex flow meters, Coriolis flow meters), techniques are known to diagnose their health using diagnostic parameters (e.g., zero adjustment value, electrode potential value) and detection signals held by the flow meter. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Patent No. 5574191 [Patent Document 2] Patent No. 6361641 [Overview of the project] [Problems that the invention aims to solve]

[0004] However, the above technology makes it difficult to improve the diagnostic accuracy of flow meters. For example, the above technology has a wide variety of flow meter measuring tube diameters and fluids to be measured, and there is unique wear and deterioration inside the measuring tube, but it is difficult to obtain visual information (as appropriate, "visual information") from parameters. Furthermore, the above technology makes it difficult to quantitatively indicate signs of flow meter failure based on the visual information inside the measuring tube.

[0005] This disclosure is made in view of the above and aims to improve the diagnostic accuracy of flow meters. [Means for solving the problem]

[0006] A diagnostic device according to one embodiment of the present disclosure includes a processor, which performs the following actions: acquiring image data inside the measuring tube of a flow meter; calculating a first evaluation value indicating the difference from the normal state of the flow meter based on the acquired image data; and diagnosing the flow meter using a machine learning model that outputs state information indicating the state of the flow meter in response to the input of the first evaluation value.

[0007] A diagnostic method according to one embodiment of the present disclosure involves a computer performing the following actions: acquiring image data inside the measuring pipe of a flow meter; calculating a first evaluation value indicating the difference from the normal state of the flow meter based on the acquired image data; and diagnosing the flow meter using a machine learning model that outputs state information indicating the state of the flow meter in response to the input of the first evaluation value.

[0008] A diagnostic program according to one embodiment of the present disclosure causes a computer to perform the following actions: acquire image data inside the measuring pipe of a flow meter; calculate a first evaluation value indicating the difference from the normal state of the flow meter based on the acquired image data; and diagnose the flow meter using a machine learning model that outputs state information indicating the state of the flow meter in response to the input of the first evaluation value. [Effects of the Invention]

[0009] According to this disclosure, there is an effect of improving the diagnostic accuracy of flow meters. [Brief explanation of the drawing]

[0010] [Figure 1] This figure shows an example configuration and processing example of a flow meter diagnostic system according to an embodiment. [Figure 2] This block diagram shows an example of the configuration of each device in the flow meter diagnostic system according to the embodiment. [Figure 3] This figure shows an example of an image data storage unit of a server device according to the embodiment. [Figure 4] This figure shows an example of a parameter storage unit of a server device according to an embodiment. [Figure 5] This is a diagram showing an example of the diagnostic result storage unit of the server device according to the embodiment. [Figure 6] This is a diagram showing an example of the machine learning model storage unit of the information processing device according to the embodiment. [Figure 7] This is a diagram showing a specific example of the image data acquisition process of the flow meter diagnostic system according to the embodiment. [Figure 8] This is a diagram showing a specific example of the parameter acquisition process of the flow meter diagnostic system according to the embodiment. [Figure 9] This is a flowchart showing an example of the flow of the entire flow meter diagnostic system according to the embodiment. [Figure 10] This is a flowchart showing an example of the flow of the training control process of the flow meter diagnostic system according to the embodiment. [Figure 11] This is a flowchart showing an example of the flow of the acquisition control process of the flow meter diagnostic system according to the embodiment. [Figure 12] This is a flowchart showing an example of the flow of the calculation control process of the flow meter diagnostic system according to the embodiment. [Figure 13] This is a flowchart showing an example of the flow of the diagnostic control process of the flow meter diagnostic system according to the embodiment. [Figure 14] This is a diagram showing an example of the hardware configuration according to the embodiment.

Embodiments for Carrying Out the Invention

[0011] Hereinafter, a diagnostic device, a diagnostic method, and a diagnostic program according to an embodiment of the present disclosure will be described in detail with reference to the drawings. Note that the present disclosure is not limited to the embodiments described below. [[ID=3​​​​​ [1. Configuration and operation of the flow meter diagnostic system 100] The configuration and processing of the flow meter diagnostic system 100 according to the embodiment will be described using Figure 1. Figure 1 is a diagram showing an example of the configuration and processing of the flow meter diagnostic system 100 according to the embodiment. Below, an example of the overall configuration of the flow meter diagnostic system 100, an example of the processing of the flow meter diagnostic system 100, and the effects of the flow meter diagnostic system 100 will be described.

[0014] (1-1. Example of the overall configuration of the flow meter diagnostic system 100) An example of the overall configuration of the flow meter diagnostic system 100 will be described. The flow meter diagnostic system 100 consists of a server device 10 and a flow meter 20. Here, the server device 10 and the flow meter 20 are connected via a predetermined communication network (not shown) via wired or wireless communication. Various communication networks such as the Internet or dedicated lines can be used as the predetermined communication network.

[0015] (1-1-1. Server device 10) The server device 10 is an information providing device that provides diagnostic results DR to the worker W who performs the pickup calibration of the flow meter 20. For example, the server device 10 can be implemented in a cloud environment, on-premise environment, edge environment, etc. The server device 10 may also be implemented using a desktop PC (Personal Computer), notebook PC, tablet terminal, smartphone, etc. Note that the flow meter diagnostic system 100 shown in Figure 1 may include multiple server devices 10.

[0016] (1-1-2.Flowmeter 20) The flow meter 20 is a field instrument installed in piping in a plant or the like to measure the flow rate of the fluid FL flowing through the measuring pipe MP. For example, the flow meter 20 may be an electromagnetic flow meter 20A, a vortex flow meter 20B, a Coriolis flow meter 20C, etc., but there are no particular limitations on the type or principle of the flow meter 20. Note that the flow meter diagnostic system 100 shown in Figure 1 may include multiple flow meters 20.

[0017] (1-2. Example of the entire process of the flow meter diagnostic system 100) An example of the overall processing of the flow meter diagnostic system 100 will be described below. Note that some of the processes in steps S1 to S4 below may be omitted.

[0018] (1-2-1. Model Training Process) First, the server device 10 performs model training (step S1). For example, the server device 10 uses the image data evaluation value IE calculated from the image data ID in the measuring tube MP of the flow meter 20, and the parameter evaluation value PE calculated from the parameters PM of the flow meter 20, as explanatory variables. At this time, the server device 10 generates training data by adding the state information CI of the flow meter 20 as ground truth data to the data combining the explanatory variables, the image data evaluation value IE and the parameter evaluation value PE, and trains the machine learning model LM using the generated training data. Note that the training data may be pre-generated.

[0019] (1-2-2. Image data acquisition process) Secondly, the server device 10 performs image data acquisition processing (step S2). For example, in the take-up calibration, the server device 10 acquires an image data ID of the inside of the measuring tube MP of the flow meter 20 that is the subject of calibration.

[0020] At this time, the server device 10 obtains an image data ID that includes the diameter of the measuring tube MP of the flow meter 20, characteristic parts of the flow meter 20, etc. For example, if the flow meter 20 is an electromagnetic flow meter 20A, the server device 10 obtains an image data ID that includes the electrode EL as a characteristic part. If the flow meter 20 is a vortex flow meter 20B, the server device 10 obtains an image data ID that includes the vortex generator VG as a characteristic part. If the flow meter 20 is a Coriolis flow meter 20C, the server device 10 obtains an image data ID that includes the measuring tube MP or the vibrating tube VT as a characteristic part.

[0021] (1-2-3. Parameter acquisition process) Thirdly, the server device 10 performs a parameter acquisition process (step S3). For example, the server device 10 acquires the parameter PM held by the flow meter 20, which is the object of calibration during the take-up calibration.

[0022] At this time, the server device 10 acquires all parameters PM, which include the flow rate signal from the flow meter 20, deposit information regarding deposits, alarm history, operating time, fluid information regarding the flowing fluid FL, and self-diagnosis results.

[0023] (1-2-4. Processing the output of diagnostic results) Fourth, the server device 10 performs diagnostic result output processing (step S4). For example, the server device 10 inputs the image data evaluation value IE calculated from the acquired image data ID and the parameter evaluation value PE calculated from the acquired parameter PM into a trained machine learning model LM, and provides the operator W with a diagnostic result DR indicating the health and lifespan of the flow meter 20 based on the output status information CI. The server device 10 can also perform diagnostic result output processing by inputting only the image data evaluation value IE into the trained machine learning model LM. Furthermore, the server device 10 can also perform diagnostic result output processing by inputting only the parameter evaluation value PE into the trained machine learning model LM. In other words, the server device 10 can diagnose the health and lifespan of the flow meter 20 using not only data combining the image data evaluation value IE and the parameter evaluation value PE, but also data using only one of them.

[0024] (1-3. Effects of the Flow Meter Diagnostic System 100) The following section will describe the overview and problems of the flow meter diagnostic system 100-P related to the reference technology, and then explain the effects of the flow meter diagnostic system 100.

[0025] (1-3-1. Overview of the Flow Meter Diagnostic System 100-P) In the flow meter diagnostic system 100-P, operator W calibrates the flow meter 20 in the take-up calibration procedure in order to obtain a calibration value with guaranteed traceability. First, operator W installs the flow meter 20 to be calibrated in the calibration equipment and performs a full-fill operation by filling the measuring tube MP with test fluid. Second, operator W performs a de-aeration operation to remove air bubbles from the flow path. Third, operator W performs a leak check operation to confirm that there are no leaks in the mounting part or other piping of the flow meter 20 to be calibrated. Fourth, operator W performs a real-flow calibration operation, which is measurement and calibration using the fluid FL that actually flows through the measuring tube MP of the flow meter 20. At this time, operator W can diagnose the soundness of the flow meter 20 using the diagnostic parameters (e.g., zero adjustment value, electrode potential value) and detection signals held by the flow meter 20.

[0026] (1-3-2. Problems with the Flow Meter Diagnostic System 100-P) The flow meter diagnostic system 100-P has the following problems. Firstly, in the flow meter diagnostic system 100-P, the diameter of the measuring tube MP of the flow meter 20 and the fluid FL being measured vary, and there is unique wear and deterioration inside the measuring tube MP, but it is difficult to obtain visual information from the parameter MP. For example, in the flow meter diagnostic system 100-P, visual information inside the measuring tube MP can be obtained by visually inspecting the flow meter 20 to capture whether the physical quantity is being measured correctly, but this is highly dependent on the know-how of the operator W, and the judgment may differ depending on the operator W. Secondly, in the flow meter diagnostic system 100-P, it is difficult to quantitatively present signs of failure of the flow meter 20 based on the visual information inside the measuring tube MP. For example, in the flow meter diagnostic system 100-P, it is possible to perform visual inspections using a large number of machine learning models, but it is difficult to diagnose the lifespan. Furthermore, with the flow meter diagnostic system 100-P, even if thinning or damage occurs in the measuring tube MP, it is difficult to notice the abnormality if it is within the acceptable range, and in many cases, the visual abnormality is judged only when a malfunction occurs in the flow meter 20.

[0027] (1-3-3. Overview of the Flow Meter Diagnostic System 100) The flow meter diagnostic system 100 performs the following processes. First, the server device 10 trains the machine learning model LM using training data including image data evaluation value IE, parameter evaluation value PE, and state information CI. Second, the server device 10 obtains image data IDs taken from inside the measuring tube MP of the flow meter 20 to be calibrated during the pickup calibration. Third, the server device 10 obtains the parameters PM held by the flow meter 20 to be calibrated during the pickup calibration. Fourth, the server device 10 inputs the image data evaluation value IE calculated from the acquired image data ID and the parameter evaluation value PE calculated from the acquired parameters PM into the trained machine learning model LM, and provides the operator W with a diagnostic result DR indicating the health and lifespan of the flow meter 20 based on the output state information CI of the flow meter 20.

[0028] (1-3-4. Effects of the Flow Meter Diagnostic System 100) The flow meter diagnostic system 100 has the following effects. First, by combining the parameter PM and image data IDs that show the characteristic appearance changes specific to the flow meter 20, the flow meter diagnostic system 100 can diagnose the current health and future lifespan of the flow meter 20. Second, the flow meter diagnostic system 100 can detect abnormalities in the flow meter 20 before they occur, contributing to the maintenance of the plant in which the flow meter 20 is used.

[0029] As described above, the flow meter diagnostic system 100 can improve the diagnostic accuracy of the flow meter 20.

[0030] [2. Configuration and operation of each device in the flow meter diagnostic system 100] Using Figures 2 to 6, the configuration and processing of each device in the flow meter diagnostic system 100 shown in Figure 1 will be explained. Below, an example of the overall configuration of the flow meter diagnostic system 100 according to this embodiment, an example of the configuration and processing of the server device 10, and an example of the configuration and processing of the flow meter 20 will be described.

[0031] (2-1. Example of the overall configuration of the flow meter diagnostic system 100) Using Figure 2, an example of the overall configuration of the flow meter diagnostic system 100 shown in Figure 1 will be explained. Figure 2 is a block diagram showing an example of the configuration of each device in the flow meter diagnostic system 100 according to this embodiment. As shown in Figure 2, the flow meter diagnostic system 100 consists of a server device 10 and a flow meter 20. The server device 10 is also connected to a communication network N, which is implemented via the internet or a dedicated line, etc.

[0032] (2-2. Example configuration and processing of server device 10) Using Figure 2, an example of the configuration and processing of the server device 10 will be explained. The server device 10 includes an input unit 11, an output unit 12, a communication unit 13, a storage unit 14, and a control unit 15.

[0033] (2-2-1. Input section 11) The input unit 11 is responsible for inputting various types of information to the server device 10. For example, the input unit 11 can be implemented using a mouse, keyboard, touch panel, etc., and accepts various types of information input to the server device 10.

[0034] (2-2-2. Output section 12) The output unit 12 is responsible for outputting various types of information from the server device 10. For example, the output unit 12 is implemented using a display, speaker, etc., and displays various types of information stored in the server device 10.

[0035] (2-2-3. Communications Section 13) The communication unit 13 is responsible for data communication with other devices. For example, the communication unit 13 performs data communication with each communication device via a router or the like. The communication unit 13 can also perform data communication with terminals (not shown).

[0036] (2-2-4. Storage section 14) The memory unit 14 stores various information that the control unit 15 references when it operates, and various information acquired when the control unit 15 operates. The memory unit 14 is composed of an image data storage unit 14a, a parameter storage unit 14b, a diagnostic result storage unit 14c, and a machine learning model storage unit 14d. Here, the memory unit 14 can be implemented as, for example, a semiconductor memory element such as RAM (Random Access Memory) or flash memory, or a storage device such as a hard disk or optical disc. In the example in Figure 2, the memory unit 14 is installed inside the server device 10, but it may be installed outside the server device 10, or multiple memory units may be installed.

[0037] (2-2-4-1. Image data storage unit 14a) The image data storage unit 14a stores the image data ID. For example, the image data storage unit 14a stores the image data ID acquired by the acquisition unit 15a of the control unit 15, which will be described later. The image data storage unit 14a also stores the image data evaluation value IE. For example, the image data storage unit 14a stores the image data evaluation value IE calculated by the calculation unit 15b of the control unit 15, which will be described later.

[0038] Here, an example of data stored in the image data storage unit 14a will be explained using Figure 3. Figure 3 is a diagram showing an example of the image data storage unit 14a of the server device 10 according to this embodiment. In the example in Figure 3, the image data storage unit 14a has items such as "flow meter," "image data," and "image data evaluation value."

[0039] "Flowmeter" refers to identification information for identifying the flowmeter 20 to be calibrated, such as the identification number or identification symbol of the flowmeter 20. "Image data" is image data ID of the inside of the measuring tube MP of the flowmeter 20, such as digital data captured by camera CM. Examples of this digital data include digital data showing the diameter of the measuring tube MP and electrode EL of the electromagnetic flowmeter 20A, digital data showing the diameter of the measuring tube MP and vortex generator VG of the vortex flowmeter 20B, and digital data showing the diameter of the measuring tube MP and the measuring tube MP or vibrating tube VT of the Coriolis flowmeter 20C. "Image data evaluation value" is a first evaluation value showing the difference from the normal state of the flowmeter 20 calculated based on the image data ID, such as a numerical value calculated for each type of flowmeter 20. An example of this numerical value is a numerical value between 0 and 100, where the evaluation value of a new flowmeter 20 is 0 and the evaluation value of a faulty flowmeter 20 is 100.

[0040] In other words, Figure 3 shows an example in which the flow meter 20 identified as "FM001" has {image data: "ID101", image data evaluation value: "IE101"}, {image data: "ID102", image data evaluation value: "IE102"}, {image data: "ID103", image data evaluation value: "IE103"}, ... stored in the image data storage unit 14a.

[0041] (2-2-4-2. Parameter storage unit 14b) The parameter storage unit 14b stores the parameter PM. For example, the parameter storage unit 14b stores the parameter PM acquired by the acquisition unit 15a of the control unit 15, which will be described later. The parameter storage unit 14b also stores the parameter evaluation value PE. For example, the parameter storage unit 14b stores the parameter evaluation value PE calculated by the calculation unit 15b of the control unit 15, which will be described later.

[0042] Here, an example of the data stored in the parameter storage unit 14b will be explained using Figure 4. Figure 4 is a diagram showing an example of the parameter storage unit 14b of the server device 10 according to this embodiment. In the example in Figure 4, the parameter storage unit 14b has items such as "flow meter", "parameter", and "parameter evaluation value".

[0043] "Flow meter" refers to identification information used to identify the flow meter 20 to be calibrated, such as the flow meter 20's identification number or identification symbol. "Parameters" are parameters PM held by the flow meter 20, such as information acquired during the calibration of the flow meter 20. Examples of this information include the flow signal of the flow meter 20, deposit information regarding deposits, alarm history, operating time, fluid information regarding the flowing fluid FL, and self-diagnosis results. "Parameter evaluation value" is a second evaluation value that shows the difference from the normal state of the flow meter 20, calculated based on the parameters PM. An example of this second evaluation value is a numerical value between 0 and 100, calculated for each type of parameter PM, with the evaluation value of a new flow meter 20 being 0 and the evaluation value of a faulty flow meter 20 being 100.

[0044] In other words, Figure 4 shows an example in which the parameter storage unit 14b stores {parameter: "PM101", parameter evaluation value: "PE101"}, {parameter: "PM102", parameter evaluation value: "PE102"}, {parameter: "PM103", parameter evaluation value: "PE103"}, ... for the flow meter 20 identified as "FM001".

[0045] (2-2-4-3. Diagnostic result storage unit 14c) The diagnostic result storage unit 14c stores the diagnostic result DR. For example, the diagnostic result storage unit 14c stores the diagnostic result DR output by the diagnostic unit 15c of the control unit 15, which will be described later.

[0046] Here, an example of the data stored in the diagnostic result storage unit 14c will be explained using Figure 5. Figure 5 is a diagram showing an example of the diagnostic result storage unit 14c of the server device 10 according to this embodiment. In the example in Figure 5, the diagnostic result storage unit 14c has items such as "flow meter," "status information," and "diagnosis result."

[0047] "Flow meter" refers to identification information for identifying the flow meter 20 to be calibrated, such as the identification number or identification symbol of the flow meter 20. "Status information" is status information CI indicating the state of the flow meter 20 at the time of calibration, such as the distance from the centroid of a normal cluster or an abnormal cluster, or the distance from the hyperplane that classifies the normal cluster or an abnormal cluster. "Diagnostic result" is a diagnostic result DR output from the status information CI of the flow meter 20, such as a numerical value from 0 to 100 indicating the health of the flow meter 20, or a numerical value indicating the lifespan of the flow meter 20. Alternatively, the "diagnostic result" may be a numerical value output by the diagnostic unit 15c indicating the health, lifespan, etc., of the flow meter 20, based on the distance calculated by the diagnostic unit 15c.

[0048] In other words, Figure 5 shows an example in which, for the flow meter 20 identified as "FM001", {status information: "CI101", diagnostic result: "DR101"}, ... is stored in the diagnostic result storage unit 14c.

[0049] (2-2-4-4. Machine Learning Model Memory Unit 14d) The machine learning model storage unit 14d stores machine learning models LM. For example, the machine learning model storage unit 14d stores machine learning models LM used by the diagnostic unit 15c of the control unit 15, which will be described later. The machine learning model storage unit 14d also stores machine learning models LM that are trained by the training unit 15d of the control unit 15, which will be described later.

[0050] Here, an example of the information stored in the machine learning model storage unit 14d will be explained using Figure 6. Figure 6 is a diagram showing an example of the machine learning model storage unit 14d of the server device 10 according to the embodiment. In the example in Figure 6, the machine learning model storage unit 14d has an item such as "machine learning model".

[0051] A "machine learning model" is the model data of a trained machine learning model LM, and includes, for example, execution data for running the algorithm of the machine learning model LM, model parameters which are settings, hyperparameters, etc.

[0052] Figure 6 shows an example where trained machine learning models LM, namely "LM001," "LM002," "LM003," etc., are stored in the machine learning model memory unit 14d. Furthermore, "LM001," "LM002," "LM003," etc., can be used as different machine learning models LM depending on the type and amount of input data.

[0053] (2-2-5. Control Unit 15) The control unit 15 is responsible for controlling the entire server device 10. The control unit 15 consists of an acquisition unit 15a, a calculation unit 15b, a diagnostic unit 15c, and a training unit 15d. Here, the control unit 15 can be implemented by electronic circuits such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit), or integrated circuits such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0054] (2-2-5-1. Acquisition part 15a) The acquisition unit 15a acquires various types of information. The acquisition unit 15a then stores the acquired information in the storage unit 14. The image data acquisition control process and the parameter acquisition control process will be described below.

[0055] (Image data acquisition control processing) The acquisition unit 15a executes image data acquisition control processing. For example, the acquisition unit 15a acquires image data IDs of the measuring tube MP of the flow meter 20. The acquisition unit 15a also acquires image data IDs that include characteristic parts of the measuring tube MP according to the type of flow meter 20 during calibration of the flow meter 20. At this time, the acquisition unit 15a acquires image data IDs that include the diameter of the measuring tube MP of the electromagnetic flow meter 20A, the electrode EL inside the measuring tube MP, etc. The acquisition unit 15a also acquires image data IDs that include the diameter of the measuring tube MP of the vortex flow meter 20B, the vortex generator VG inside the measuring tube MP, etc. The acquisition unit 15a also acquires image data IDs that include the diameter of the measuring tube MP of the Coriolis flow meter 20C, the measuring tube MP, or the vibrating tube VT inside the measuring tube MP, etc.

[0056] A specific example of the image data acquisition control process will be described. First, the acquisition unit 15a acquires "ID101", "ID102", "ID103", ... from the camera CM as image data IDs for the measuring tube MP of the flow meter 20 identified as "FM001". Second, the acquisition unit 15a stores the acquired image data IDs in the image data storage unit 14a.

[0057] (Parameter acquisition control process) The acquisition unit 15a performs parameter acquisition control processing. For example, the acquisition unit 15a acquires the parameter PM held by the flow meter 20. The acquisition unit 15a also acquires the parameter PM during calibration of the flow meter 20, which includes at least one of the following: the flow signal of the flow meter 20, deposit information regarding deposits, alarm history, operating time, fluid information regarding the flowing fluid FL, and self-diagnosis results.

[0058] A specific example of the parameter acquisition control process will be described. First, the acquisition unit 15a acquires parameters PM held by the flow meter 20, which is identified as "FM001", such as "PM101", "PM102", "PM103", ... from the flow meter 20. Second, the acquisition unit 15a stores the acquired parameters PM in the parameter storage unit 14b.

[0059] (2-2-5-2. Calculation section 15b) The calculation unit 15b calculates various types of information. The calculation unit 15b may store the calculated information in the storage unit 14. Alternatively, the calculation unit 15b may refer to the information stored in the storage unit 14. The image data evaluation value calculation control process and the parameter evaluation value calculation control process will be described below.

[0060] (Image data evaluation value calculation control processing) The calculation unit 15b executes image data evaluation value calculation control processing. For example, the calculation unit 15b calculates the image data evaluation value IE, which is a first evaluation value indicating the difference from the normal state of the flow meter 20, based on the acquired image data ID. At this time, the calculation unit 15b calculates the image data evaluation value IE using the image data ID corresponding to the type of flow meter 20 at the time of shipment as the normal state. Details of the image data evaluation value calculation control processing will be described later.

[0061] A specific example of the image data evaluation value calculation control process will be described below. First, the calculation unit 15b obtains "ID101", "ID102", "ID103", ... for the flow meter 20 identified as "FM001" as the image data ID stored in the image data storage unit 14a. Second, the calculation unit 15b calculates "IE101", "IE102", "IE103", ... as the image data evaluation values ​​IE for the flow meter 20 identified as "FM001". Third, the calculation unit 15b stores the calculated image data evaluation values ​​IE in the image data storage unit 14a.

[0062] (Parameter evaluation value calculation control process) The calculation unit 15b executes parameter evaluation value calculation control processing. For example, the calculation unit 15b calculates a parameter evaluation value PE, which is a second evaluation value indicating the difference from the normal state of the flow meter 20, based on the acquired parameter PM. At this time, the calculation unit 15b calculates the parameter evaluation value PE using the theoretical value of the flow meter 20 or the parameter PM corresponding to the type of flow meter 20 at the time of the previous calibration as the normal state. Details of the parameter evaluation value calculation control processing will be described later.

[0063] A specific example of the parameter evaluation value calculation control process will be described below. First, the calculation unit 15b obtains "PM101", "PM102", "PM103", ... for the flow meter 20 identified as "FM001" as parameters PM to be stored in the parameter storage unit 14b. Second, the calculation unit 15b calculates "PE101", "PE102", "PE103", ... as parameter evaluation values ​​PE for the flow meter 20 identified as "FM001". Third, the calculation unit 15b stores the calculated parameter evaluation values ​​PE in the parameter storage unit 14b.

[0064] (2-2-5-3. Diagnostic Department 15c) The diagnostic unit 15c outputs a diagnostic result DR. The diagnostic unit 15c may store the outputted diagnostic result DR in the storage unit 14. The diagnostic unit 15c may also refer to various information stored in the storage unit 14. The diagnostic result output control processing (health output control processing, lifespan output control processing) will be described below.

[0065] (Diagnostic result output control processing) The diagnostic unit 15c executes diagnostic result output control processing. For example, the diagnostic unit 15c diagnoses the flowmeter 20 using a machine learning model LM that outputs state information CI indicating the state of the flowmeter 20 in response to the input of an image data evaluation value IE. The diagnostic unit 15c also diagnoses the flowmeter 20 using a machine learning model LM that outputs state information CI indicating the state of the flowmeter 20 in response to at least one input of an image data evaluation value IE and a parameter evaluation value PE. Furthermore, the diagnostic unit 15c diagnoses the flowmeter 20 using a machine learning model LM that outputs the distance from data indicating the normal or abnormal state of the flowmeter 20 in a multidimensional space as state information CI. The diagnostic unit 15c also diagnoses the electrode EL of the electromagnetic flowmeter 20A. Furthermore, the diagnostic unit 15c diagnoses the vortex generator VG of the vortex flowmeter 20B. Furthermore, the diagnostic unit 15c diagnoses the measuring tube MP or vibrating tube VT of the Coriolis flowmeter 20C.

[0066] A specific example of the diagnostic result output control process will be explained. First, the diagnostic unit 15c refers to "IE101" for the flow meter 20 identified as "FM001" as the image data evaluation value IE stored in the image data storage unit 14a. Second, the diagnostic unit 15c inputs the referenced image data evaluation value IE into "LM001", which is the machine learning model LM stored in the machine learning model storage unit 14d. Third, the diagnostic unit 15c refers to "PE101" for the flow meter 20 identified as "FM001" as the parameter evaluation value PE stored in the parameter storage unit 14b. Fourth, the diagnostic unit 15c inputs the referenced parameter evaluation value PE into "LM001", which is the machine learning model LM stored in the machine learning model storage unit 14d. Fifth, the diagnostic unit 15c obtains "CI101", which is the state information CI output from "LM001", which is the machine learning model LM. Sixth, the diagnostic unit 15c outputs a diagnostic result DR, "DR101," from the acquired state information CI, "CI101." Seventh, the diagnostic unit 15c stores the acquired state information CI and the output diagnostic result DR in the diagnostic result storage unit 14c.

[0067] (Health output control processing) The diagnostic unit 15c performs a health output control process as a diagnostic result output control process. For example, the diagnostic unit 15c diagnoses the health of the flow meter 20, indicating the degree of normal or abnormal state of the flow meter 20, using the distance from a predetermined position in a subspace of a normal cluster that indicates the normal state of the flow meter 20 in a multidimensional space. Details of the health output control process will be described later.

[0068] (Lifetime output control processing) The diagnostic unit 15c performs a lifespan output control process as a diagnostic result output control process. For example, the diagnostic unit 15c diagnoses the lifespan of the flow meter 20, which indicates the period until the abnormal state is reached, using the distance from a predetermined position in a subspace of an abnormal cluster that indicates the abnormal state of the flow meter 20 in a multidimensional space. Details of the lifespan output control process will be described later.

[0069] (2-2-5-4. Training Department 15d) The training unit 15d trains the machine learning model LM. The training unit 15d may also refer to various information stored in the memory unit 14. The machine learning model training control process is described below.

[0070] The training unit 15d performs machine learning model training control processing. For example, the training unit 15d trains a machine learning model LM for each type of flow meter 20 by inputting at least one of the image data evaluation value IE and parameter evaluation value PE as explanatory variables.

[0071] A specific example of machine learning model training control processing will be described. First, the diagnostic unit 15c obtains "ID111-N", which is the image data ID of a new flow meter 20, for the flow meter 20 identified as "FM001", and calculates "IE111-N", which is the image data evaluation value IE of the new flow meter 20. Second, the diagnostic unit 15c obtains "PM111-N", which is the parameter PM of the new flow meter 20, for the flow meter 20 identified as "FM001", and calculates "PE111-N", which is the parameter evaluation value PE of the new flow meter 20. Third, for the flow meter 20 identified as "FM001", "CI111-N", which is the state information CI indicating that it is operational, is added as ground truth data, and a dataset of normal state data is generated as {Image data evaluation value IE: "IE111-N", Parameter evaluation value PE: "PE111-N", State information CI: "CI111-N"}.

[0072] Fourth, the diagnostic unit 15c obtains "ID121-D", which is the image data ID of the faulty flow meter 20, for the flow meter 20 identified as "FM001", and calculates "IE121-D", which is the image data evaluation value IE of the faulty flow meter 20. Fifth, the diagnostic unit 15c obtains "PM121-D", which is the parameter PM of the faulty flow meter 20, for the flow meter 20 identified as "FM001", and calculates "PE121-D", which is the parameter evaluation value PE of the faulty flow meter 20. Sixth, for the flow meter 20 identified as "FM001", it assigns "CI121-D", which is the state information CI indicating that it is inoperable, as the correct data, and generates an abnormal state dataset {Image data evaluation value IE: "IE121-D", Parameter evaluation value PE: "PE121-D", State information CI: "CI121-D"}. Seventh, the diagnostic unit 15c trains the machine learning model LM, "LM001," stored in the machine learning model storage unit 14d, by inputting the generated training data.

[0073] (2-3. Example configuration and processing of the flow meter 20) Using Figure 2 again, we will explain examples of the configuration and processing of the flow meter 20. For example, the flow meter 20 is an electromagnetic flow meter 20A having a measuring tube MP and an electrode EL, which measures the flow rate by detecting the voltage value induced inside the fluid FL by the law of electromagnetic induction. Alternatively, the flow meter 20 is a vortex flow meter 20B having a measuring tube MP and a vortex generator VG, which measures the flow rate by utilizing the proportional relationship between the frequency generated by the Karman vortex and the fluid velocity FL. Furthermore, the flow meter 20 is a Coriolis flow meter 20C having a measuring tube MP and a vibrating tube VT, which measures the flow rate by utilizing the proportional relationship between the Coriolis force and the mass and velocity of the fluid FL.

[0074] Furthermore, the flow meter 20 stores parameters PM. For example, the flow meter 20 stores the flow signal, adhesion information, alarm history, operating time, fluid information, self-diagnosis results, etc.

[0075] [3. Specific examples of each process of the flow meter diagnostic system 100] Specific examples of each process of the flow meter diagnostic system 100 according to the embodiment will be described below.Specific examples of image data acquisition process, image data evaluation value calculation process, parameter acquisition process, parameter evaluation value calculation process, diagnostic result output process, and model training process will be described below.

[0076] (3-1. Specific Examples of Image Data Acquisition Processes) A specific example of the image data acquisition process will be explained using Figure 7. Figure 7 is a diagram showing a specific example of the image data acquisition process of the flow meter diagnostic system 100 according to the embodiment. Below, an example of the configuration of each device used in the image data acquisition process and an example of the image data acquisition process will be explained.

[0077] (3-1-1. Example configuration of each device used for image data acquisition processing) The flow meter 20 is an electromagnetic flow meter 20A, a vortex flow meter 20B, a Coriolis flow meter 20C, etc. The measuring tube MP is the tube in which the flow meter 20 is installed in the piping. The light LT is used to illuminate the inside of the measuring tube MP and to standardize the quality of the captured images. The camera CM is an imaging device for capturing images inside the measuring tube MP. The support column PT supports the camera CM and the light LT, and its length is adjusted according to the type and diameter of the flow meter 20. The stage ST is on which the flow meter 20 is installed and is sized to stably install the flow meter 20 up to its maximum diameter. If the flow meter 20 is a Coriolis flow meter 20C, a supporter for fixing the Coriolis flow meter 20C may be included as a component due to the length of the measuring tube MP. The server device 10 can control the imaging inside the measuring tube MP, the illumination of the light LT, the adjustment of the length of the support column PT, etc.

[0078] (3-1-2. Example of image data acquisition process) Firstly, the server device 10 instructs the camera CM to photograph the measuring tube MP of the flow meter 20 (see Figure 7(1)). At this time, the server device 10 controls the adjustment of the length of the support column PT, controls the light LT to shine light, and controls the camera CM to photograph the inside of the measuring tube MP.

[0079] Secondly, camera CM photographs the measuring tube MP of the flow meter 20 (see Figure 7(2)). When the flow meter 20 is an electromagnetic flow meter 20A, camera CM photographs the measuring tube MP in such a way that the electrode EL is included as a feature part in the image, along with the diameter of the measuring tube MP. When the flow meter 20 is a vortex flow meter 20B, camera CM photographs the measuring tube MP in such a way that the vortex generator VG is included as a feature part in the image, along with the diameter of the measuring tube MP. When the flow meter 20 is a Coriolis flow meter 20C, camera CM photographs the measuring tube MP or the vibrating tube VT as a feature part in the image, along with the diameter of the measuring tube MP.

[0080] Thirdly, the server device 10 receives image data IDs from the camera CM (see Figure 7(3)). At this time, the server device 10 stores the received image data IDs for each flow meter 20. The server device 10 can also instruct the camera CM to re-image the measuring tube MP of the flow meter 20.

[0081] (3-2. Specific Examples of Image Data Evaluation Value Calculation Process) A specific example of the image data evaluation value calculation process will be explained. For example, the server device 10 calculates an image data evaluation value IE, which quantifies the degree of damage or deterioration, based on the image data ID inside the measuring tube MP of a new flow meter 20. In another example, the server device 10 uses an autoencoder, which is a generative model trained only on the image data ID inside the measuring tube MP of a flow meter 20 in a normal state, to calculate the PSNR (Peak Signal-to-Noise Ratio), SSIM (Structural Similarity Index Measure), etc. of the generated image and the original image as the image data evaluation value IE.

[0082] (3-3. Specific Examples of Parameter Acquisition Processes) A specific example of the parameter acquisition process will be explained using Figure 8. Figure 8 is a diagram showing a specific example of the parameter acquisition process of the flow meter diagnostic system 100 according to the embodiment. Below, the types of parameters PM will be explained, followed by an example of the parameter acquisition process.

[0083] (3-3-1. Types of parameters) The parameters PM include, for example, the flow rate signal from the flow meter 20, adhesion information, alarm history, operating time, fluid information, and self-diagnosis results.

[0084] (3-3-1-1.Flow rate signal) The flow rate signal is raw data detected by the flow meter 20, and is a detection signal such as a voltage value detected by the electromagnetic flow meter 20A, vortex flow meter 20B, Coriolis flow meter 20C, etc., based on their respective measurement principles.

[0085] (3-3-1-2. Adhesion Information) The deposit information refers to information about deposits adhering to the measuring tube MP of the flow meter 20, such as scale, corrosion products, and dust and particles contained in the fluid FL. Here, scale adheres strongly to the measuring tube MP and requires removal with a special cleaning solution or high-pressure water, which may cause deterioration of the measuring tube MP or make complete removal impossible. Furthermore, corrosion products are substances produced by the oxidation of metals, which may change the designed shape and prevent the intended operation. In addition, dust and particles may cause wear and damage to the equipment.

[0086] (3-3-1-3. Alarm History) The alarm history is historical information of alarms from the flow meter 20, and includes information such as the detection date and time of an alarm that is triggered when the flow rate threshold set in the flow meter 20 is exceeded.

[0087] (3-3-1-4. Operating Hours) The operating time is historical information related to the operation of the flow meter 20, and includes information such as the start date and time of use of the flow meter 20, the calibration date and time, continuous operating time, and total operating time.

[0088] (3-3-1-5. Fluid Information) The fluid information is information about the fluid FL that the flow meter 20 measures, and includes, for example, information about the type of fluid FL (e.g., components, properties, presence or absence of conductivity) and its state (e.g., liquid, gas, vapor).

[0089] (3-3-1-6. Self-diagnosis results) The self-diagnosis results are the diagnostic results output by the flow meter 20 through its self-diagnosis function, and consist of the diagnostic parameters output by the electromagnetic flow meter 20A, vortex flow meter 20B, Coriolis flow meter 20C, etc., as their respective diagnostic results.

[0090] The electromagnetic flowmeter 20A stores the results of its self-diagnosis of the electrode EL, including the detection of insulating material adhesion, the diagnosis of insulation degradation, and the diagnosis of the integrity of the magnetic and excitation circuits, as diagnostic parameters.

[0091] The vortex flowmeter 20B stores the results of self-diagnosis of the vortex generator VG, including noise diagnosis, vibration diagnosis, resonance diagnosis, blockage diagnosis, predictive diagnosis, and soundness diagnosis of the detection and calculation circuits, as diagnostic parameters.

[0092] The Coriolis flow meter 20C stores the results of self-diagnosis of the measuring tube MP and vibrating tube VT, such as corrosion detection and vibrating tube self-diagnosis, as diagnostic parameters.

[0093] (3-3-2. Example of parameter acquisition process) Firstly, worker W installs the flow meter 20 to be calibrated in the calibration equipment, fills the measuring tube MP with the test solution, removes air bubbles from the flow path, and performs a leak check to confirm that there are no leaks in the mounting part of the flow meter 20 or other parts of the piping (see Figure 8(1)).

[0094] Secondly, the server device 10 receives the parameter PM from the flow meter 20 (see Figure 8(2)). At this time, the server device 10 saves the received parameter PM for each flow meter 20. In addition, the operator W performs a real-flow calibration operation, which is measurement and calibration using the fluid FL flowing through the measuring tube MP of the flow meter 20.

[0095] (3-4. Specific Examples of Parameter Evaluation Value Calculation Process) A specific example of the parameter evaluation value calculation process will be explained. For example, the server device 10 uses the theoretical values ​​of each parameter PM of the flow meter 20 and the numerical values ​​of each parameter PM from the previous calibration as a reference, and calculates the difference between these values ​​and the acquired numerical values ​​of each parameter PM as the parameter evaluation value PE. The server device 10 can also be adjusted so that the number of types of parameters PM does not become excessively large through dimensionality reduction or selection by the operator W.

[0096] (3-5. Specific Examples of Diagnostic Result Output Processing) This section explains specific examples of diagnostic result output processing. Below, we will first describe an overview of the diagnostic result output processing, and then provide examples of how it is performed.

[0097] (3-5-1. Overview of the diagnostic result output process) The flow meter diagnostic system 100 uses a machine learning model LM, which has learned the relationship between image data ID, parameter PM, and measured values, to perform diagnostic result output processing for each flow meter 20.

[0098] For example, in the electromagnetic flowmeter 20A, wear and adhesion can occur on the electrode EL used to pick up the electromotive force. In this case, since the electromagnetic flowmeter 20A indirectly measures the flow velocity from the electromotive force according to Faraday's law of electromagnetic induction, wear and adhesion on the electrode EL will cause a deviation from the factory-adjusted value, resulting in errors in the measured value. In particular, diagnostic parameters related to the electrode EL have a very strong correlation with the visual information of the electrode EL.

[0099] Furthermore, in the vortex flowmeter 20B, corrosion or rust may occur on the vortex rod, which is the vortex generator VG. In this case, the vortex frequency generation process in the vortex flowmeter 20B will differ from the design, resulting in errors in the measured values. For example, if there is an alarm history for vibration diagnosis or resonance diagnosis, it can be inferred that the vortex generator VG was subjected to a heavy load, and there is a possibility that the vortex generator VG may have been damaged.

[0100] Furthermore, in the Coriolis flowmeter 20C, deposits may occur on the measuring tube MP and the vibrating tube VT. In this case, the Coriolis flowmeter 20C measures mass flow rate by utilizing the Coriolis force generated when the fluid FL passes through the vibrating tube VT. Therefore, wear or deposits change the rigidity of the vibrating tube VT, causing errors in the measurement.

[0101] As described above, the flow meter diagnostic system 100 has a correlation between the measurement principle of the flow meter 20, the acquired parameter PM, and the visual information. By simultaneously analyzing the parameter PM and the image data ID, the reliability of the measured values ​​can be evaluated with higher accuracy. Furthermore, the flow meter diagnostic system 100 defines each parameter PM and image data ID in a multidimensional spatial model, enabling the diagnosis of the health and lifespan of the flow meter 20.

[0102] (3-5-2. Example of processing for outputting diagnostic results) The server device 10 diagnoses the health and lifespan of a flow meter 20 by mapping the features of a specific flow meter 20 into a multidimensional space and calculating the distance to normal or abnormal data in the multidimensional space using a machine learning model LM. Here, the server device 10 causes the machine learning model LM to construct the above-mentioned multidimensional space through the model training process described in (3-6. Specific Examples of Model Training Process) below. Here, the above-mentioned multidimensional space has visual information based on image data IDs and parameter information based on parameters PM. Below, specific examples 1 to 4 of the machine learning model LM for mapping the feature data of the flow meter 20 into a multidimensional space, specific examples 1 and 2 of the distance calculation process, specific examples of the health diagnosis process, and specific examples of the lifespan diagnosis process will be described.

[0103] (3-5-2-1. Specific Example of a Machine Learning Model (LM) 1) Specific example 1 of the machine learning model LM is a model that maps flow meter data X, used for model construction (model training), into a K-dimensional space. Here, the flow meter data X is a matrix of size N × feature K. Note that feature K is a feature that contains at least one of the image data ID and parameter PM. Also, the N data points are either normal flow data or abnormal flow meter data. Specific example 1 of the machine learning model LM similarly maps unknown flow meter data and evaluates the distance between the unknown flow meter data and the normal or abnormal space.

[0104] (3-5-2-2. Specific Example of a Machine Learning Model (LM) 2) Specific example 2 of the machine learning model LM is a model that reduces the dimensionality of flow meter data X used for model construction (model training) using principal component analysis. In specific example 2 of the machine learning model LM, the flow meter data X is mapped to an arbitrary L-dimensional space while reducing dimensionality using equation (1) below. The flow meter data X is a matrix of size N × feature K. Note that feature K is at least one feature from image data ID and parameter PM. Also, the N data points are either normal flow data or abnormal flow meter data. Also, W L This is a K×L weight matrix trained to maximize the variance of the data. Also, Y L This is a compressed N×L size matrix.

[0105] Y L =XW L ...(1)

[0106] In the second specific example of the machine learning model LM, according to the above equation (1), a dimensional space of L dimensions is constructed by combining the flowmeter data X with visual information and parameter information. In the second specific example of the machine learning model LM, since K < L, it becomes possible to explain the flowmeter data X with fewer explanatory variables. In the second specific example of the machine learning model LM, unknown flowmeter data is similarly mapped, and the distance between the unknown flowmeter data and the normal space or the abnormal space is evaluated.

[0107] (3-5-2-3. Third Specific Example of Machine Learning Model LM) The third specific example of the machine learning model LM is a model that compresses dimensions using an autoencoder with a bottleneck. The autoencoder has an encoder that compresses the input data in dimensions and a decoder that restores the data compressed in dimensions. At this time, in the autoencoder, by machine learning the identity mapping of the flowmeter data X for model construction (model training), the encoder of the autoencoder machine-learns a mapping function that compresses the flowmeter data X in dimensions. In the third specific example of the machine learning model LM, the identity mapping can be machine-learned by minimization as in the following equation (2). In the following equation (2), the encoder of the autoencoder is F encoder , and the decoder of the autoencoder is F decoder .

[0108] min|F decoder (F encoder (X)) - X| ···(2)

[0109] In the third specific example of the machine learning model LM, in the process of the encoder and decoder machine-learning the identity mapping of normal or abnormal flowmeter data according to the above equation (2), a method of explaining the characteristics of each flowmeter data with a small number of dimensions can be machine-learned. In the third specific example of the machine learning model LM, the unknown flowmeter data is input to the encoder F encoder of the trained autoencoder to compress the dimensions of the unknown flowmeter data, and then the distance between the unknown flowmeter data and the normal space or the abnormal space is evaluated.

[0110] (3-5-2-4. Specific Example of a Machine Learning Model (LM) 4) Specific example 4 of the machine learning model LM is a model that estimates the probability distribution p(X) of flow meter data X used for model construction (model training) using a variational autoencoder. In specific example 4 of the machine learning model LM, machine learning is performed to find the parameter z that maximizes the likelihood of p(X) using the maximum likelihood method. In this case, specific example 4 of the machine learning model LM maximizes the variational lower bound L(X,z) of the flow meter data X and parameter z using equation (3) below. Note that in equation (3) below, D KL [q(z|X)||p(z)] is a normalization term for the parameter z, and is a term used for machine learning to bring the distributions of q(z|X) and p(z) closer together. Also, in equation (3) below, E q(z|X) [log p(X|z)] is the reconstruction error term, which maximizes the expected value of the log-likelihood of the decoder output p(X|z) with respect to the encoder output q(z|X).

[0111] max(L(X,z))=max(-D KL [q(z|X)||p(z)]+E q(z|X) [log p(X|z)]) ···(3)

[0112] In specific example 4 of the machine learning model LM, the probability distribution of flow meter data X can be estimated by maximizing the value using equation (3) above. In this case, specific example 4 of the machine learning model LM calculates the difference between the encoder output z when unknown flow meter data is input and the flow meter data used for model training.

[0113] (3-5-2-5. Specific Example of Distance Calculation Process 1) Specific examples 1-4 of the above machine learning model LM demonstrate the distance calculation process. Specific example 1 calculates the centroids of normal and abnormal clusters, and then calculates the distance from the calculated centroids to the unknown flow meter data. Note that distances include Manhattan distance, Euclidean distance, Mahalanobis distance, etc.

[0114] (3-5-2-6. Specific Example of Distance Calculation Process 2) Specific examples 1-4 of the above machine learning model LM demonstrate, as specific example 2 of the distance calculation process, defining a hyperplane that classifies normal and abnormal clusters using a support vector machine, and then calculating the distance from the defined hyperplane to the unknown flow meter data. In specific example 2 of the distance calculation process, it is also possible to define the hyperplane using only data from either normal or abnormal clusters. Note that distances include Manhattan distance, Euclidean distance, Mahalanobis distance, etc.

[0115] (3-5-2-7. Specific Examples of Health Assessment Processes) As a concrete example of health diagnosis processing, the server device 10 diagnoses the health of the flow meter 20, indicating the degree of normality or abnormality, by inputting the image data evaluation value IE for the image data ID obtained from the image data acquisition process in (3-1. Concrete Example of Image Data Acquisition Processing) and the image data evaluation value calculation process in (3-2. Concrete Example of Image Data Evaluation Value Calculation Processing), as well as the parameter evaluation value PE for the parameter PM obtained from the parameter acquisition process in (3-3. Concrete Example of Parameter Acquisition Processing) and the parameter evaluation value calculation process in (3-4. Concrete Example of Parameter Evaluation Value Calculation Processing), as input data into one of the concrete examples 1 to 4 of the machine learning model LM described above.

[0116] The server device 10 calculates a health level on a scale of 0 to 100 based on the distance calculated by, for example, Specific Example 1 of the distance calculation process, which is the distance from the centroid of a normal cluster to the unknown flow meter data. In this case, the server device 10 calculates a health level closer to 100 the smaller the distance from the centroid of a normal cluster based on a new flow meter 20. On the other hand, the server device 10 calculates a health level closer to 0 the larger the distance from the centroid of a normal cluster based on a new flow meter 20. The server device 10 can also output the distance calculated by Specific Example 1 of the distance calculation process as a numerical value indicating the health level of the flow meter 20.

[0117] The server device 10 calculates a health rating on a scale of 0 to 100 based on the distance calculated by, for example, the specific example 2 of the distance calculation process, from the unknown flow meter data to the hyperplane. In this case, the server device 10 calculates a health rating closer to 100 as the distance from the unknown flow meter data to the hyperplane increases. On the other hand, the server device 10 calculates a health rating closer to 0 as the distance from the unknown flow meter data to the hyperplane decreases. The server device 10 can also output the distance calculated by the specific example 2 of the distance calculation process as a numerical value indicating the health of the flow meter 20.

[0118] Furthermore, the server device 10 can also calculate the health status based on the distance from the centroid of the abnormal cluster, which is determined by the faulty flow meter 20, to the unknown flow meter data. In addition, the server device 10 can output the distance from the centroid of the abnormal cluster as a numerical value indicating the health status of the flow meter 20.

[0119] (3-5-2-8. Specific Examples of Life Diagnosis Processing) As a concrete example of life diagnosis processing, the server device 10 diagnoses the lifespan of the flow meter 20, which indicates the period until an abnormal state is reached, by inputting the image data evaluation value IE for the image data ID obtained from the image data acquisition process in (3-1. Concrete Example of Image Data Acquisition Processing) and the image data evaluation value calculation process in (3-2. Concrete Example of Image Data Evaluation Value Calculation Processing), as well as the parameter evaluation value PE for the parameter PM obtained from the parameter acquisition process in (3-3. Concrete Example of Parameter Acquisition Processing) and the parameter evaluation value calculation process in (3-4. Concrete Example of Parameter Evaluation Value Calculation Processing), as input data into one of the concrete examples 1 to 4 of the machine learning model LM described above.

[0120] The server device 10 calculates the lifespan in years and months based on the distance calculated by, for example, the specific example 1 of the distance calculation process, which is the distance from the centroid of the abnormal cluster to the unknown flow meter data. In this case, the server device 10 calculates the lifespan in a shorter period of years and months the smaller the distance from the centroid of the abnormal cluster based on the faulty flow meter 20. On the other hand, the server device 10 calculates the lifespan in a longer period of years and months the larger the distance from the centroid of the abnormal cluster based on the faulty flow meter 20. The server device 10 can also output the distance calculated by the specific example 1 of the distance calculation process as a numerical value indicating the lifespan of the flow meter 20.

[0121] The server device 10 calculates the lifespan in years and months based on the distance calculated by, for example, the specific example 2 of the distance calculation process, from the unknown flow meter data to the hyperplane. In this case, the server device 10 calculates the lifespan in a shorter period of years and months as the distance from the unknown flow meter data to the hyperplane increases. On the other hand, the server device 10 calculates the lifespan in a longer period of years and months as the distance from the unknown flow meter data to the hyperplane increases. The server device 10 can also output the distance calculated by the specific example 2 of the distance calculation process as a numerical value indicating the lifespan of the flow meter 20.

[0122] Furthermore, the server device 10 can also calculate the lifespan based on the distance from the centroid of a normal cluster, based on a new flow meter 20, to the unknown flow meter data. In addition, the server device 10 can output the distance from the centroid of a normal cluster as a numerical value indicating the lifespan of the flow meter 20.

[0123] (3-6. Specific Examples of Model Training Processes) A concrete example of the model training process will be explained. For example, the server device 10 uses the image data evaluation value IE calculated from the image data ID in the measuring tube MP of the flow meter 20, and the parameter evaluation value PE calculated from the parameters PM of the flow meter 20, as explanatory variables, and generates training data to which the state information CI of the flow meter 20 is attached as ground truth data, and trains the machine learning model LM. At this time, the server device 10 can construct a model that has a sub-region meaning that the explanatory variables are operational and a sub-region meaning that they are not operational by training concrete examples 1 to 4 of the machine learning model LM in (3-5. Concrete Examples of Diagnostic Result Output Processing) above with operational status (e.g., whether or not it is a faulty product) as ground truth data as state information CI.

[0124] The server device 10, for example, acquires a dataset of image data IDs from the measuring tube MP of a new flow meter 20 and parameters PM of the new flow meter 20, calculates the image data evaluation value IE and the parameter evaluation value PE for a normal state, and generates training data with status information CI indicating that it is operational as ground truth data, and trains a machine learning model LM using supervised machine learning.

[0125] The server device 10, for example, acquires a dataset of image data IDs from the measuring tube MP of the faulty flow meter 20 and parameters PM of the faulty flow meter 20, calculates an image data evaluation value IE and an abnormal parameter evaluation value PE, and generates training data with status information CI indicating that the device is inoperable as ground truth data, and trains a machine learning model LM using supervised machine learning.

[0126] Furthermore, the server device 10 can also generate training data without ground truth data and train the machine learning model LM using unsupervised machine learning.

[0127] [4. Flowchart of each process in the flow meter diagnostic system 100] The processing flow of the flow meter diagnostic system 100 according to the embodiment will be explained using Figures 9 to 13. Below, the overall processing flow of the flow meter diagnostic system 100 will be explained, followed by a description of each process: training control processing, acquisition control processing, calculation control processing, and diagnostic control processing.

[0128] (4-1. Overall processing of the flow meter diagnostic system 100) The overall processing flow of the flow meter diagnostic system 100 according to the embodiment will be explained using Figure 9. Figure 9 is a flowchart showing an example of the overall processing flow of the flow meter diagnostic system 100 according to the embodiment. Note that the processes in steps S101 to S104 below can be executed in a different order. Also, some of the processes in steps S101 to S104 below may be omitted.

[0129] (4-1-1. Training and Control Processing) Firstly, the flow meter diagnostic system 100 performs training control processing (step S101). For example, the flow meter diagnostic system 100 trains a machine learning model LM that outputs state information CI indicating the state of the flow meter 20 by performing the processes described in steps S201 to S204.

[0130] (4-1-2. Acquisition control processing) Secondly, the flow meter diagnostic system 100 performs acquisition control processing (step S102). For example, by performing the processes described in steps S301 to S304, the flow meter diagnostic system 100 obtains the image data ID in the measuring tube MP of the flow meter 20 and obtains the parameter PM held by the flow meter 20.

[0131] (4-1-3. Calculation and control processing) Thirdly, the flow meter diagnostic system 100 performs calculation control processing (step S103). For example, by performing the processing described in steps S401 to S406, the flow meter diagnostic system 100 calculates an image data evaluation value IE, which shows the difference from the normal state of the flow meter 20, based on the image data IM, and calculates a parameter evaluation value PE, which shows the difference from the normal state of the flow meter 20, based on the parameter PM.

[0132] (4-1-4. Diagnostic and Control Processing) Fourth, the flow meter diagnostic system 100 executes a diagnostic control process (step S104) and then terminates the process. For example, the flow meter diagnostic system 100 diagnoses the health and lifespan of the flow meter 20 by executing the processes described in steps S501 to S507.

[0133] (4-2. Training and Control Processing) The flow of the training control process of the flow meter diagnostic system 100 according to the embodiment will be explained using Figure 10. Figure 10 is a flowchart showing an example of the flow of the training control process of the flow meter diagnostic system 100 according to the embodiment. Note that the processes in steps S201 to S204 below can be executed in a different order. Also, some of the processes in steps S201 to S204 below may be omitted.

[0134] (4-2-1. Dataset Acquisition Process) Firstly, the server device 10 performs a dataset acquisition process (step S201). For example, the server device 10 acquires a dataset containing the image data ID and parameter PM of a new flow meter 20.

[0135] (4-2-2. Image Data Evaluation Value Calculation Process) Secondly, the server device 10 performs image data evaluation value calculation processing (step S202). For example, the server device 10 calculates the image data evaluation value IE based on the image data IDs included in the acquired dataset.

[0136] (4-2-3. Parameter evaluation value calculation process) Thirdly, the server device 10 performs a parameter evaluation value calculation process (step S203). For example, the server device 10 calculates a parameter evaluation value PE based on the parameter PM included in the acquired dataset.

[0137] (4-2-4. Model Training Process) Fourth, the server device 10 executes the model training process (step S204) and terminates the training control process. For example, the server device 10 performs supervised machine learning by inputting training data into the machine learning model ML, which includes image data evaluation value IE and parameter evaluation value PE as explanatory variables, and state information CI indicating whether the flow meter 20 is operational as ground truth data.

[0138] (4-3. Acquisition control processing) The flow of the acquisition control process of the flow meter diagnostic system 100 according to the embodiment will be explained using Figure 11. Figure 11 is a flowchart of an example of the flow of the acquisition control process of the flow meter diagnostic system 100 according to the embodiment. Note that the processes in steps S301 to S304 below can be executed in a different order. Also, some of the processes in steps S301 to S304 below may be omitted.

[0139] (4-3-1. Image data acquisition process) Firstly, the server device 10 performs image data acquisition processing (step S301). For example, when the flow meter 20 is taken into calibration, the server device 10 acquires an image data ID that includes the diameter of the measuring tube MP of the flow meter 20, characteristic parts inside the measuring tube MP according to the type of flow meter 20, etc.

[0140] (4-3-2. Image Data Storage Processing) Secondly, the server device 10 performs image data storage processing (step S302). For example, the server device 10 stores the acquired image data ID in the image data storage unit 14a for each flow meter 20.

[0141] (4-3-3. Parameter acquisition process) Thirdly, the server device 10 performs parameter acquisition processing (step S303). For example, when the flow meter 20 is taken into calibration, the server device 10 acquires parameters PM including the flow signal of the flow meter 20, adhesion information, alarm history, operating time, fluid information, self-diagnosis results, etc.

[0142] (4-3-4. Parameter storage process) Fourth, the server device 10 executes parameter storage processing (step S304) and terminates the acquisition control processing. For example, the server device 10 stores the acquired parameter PM in the parameter storage unit 14b for each flow meter 20.

[0143] (4-4. Calculation and control processing) The flow of the calculation control process of the flow meter diagnostic system 100 according to the embodiment will be explained using Figure 12. Figure 12 is a flowchart showing an example of the flow of the calculation control process of the flow meter diagnostic system 100 according to the embodiment. Note that the processes in steps S401 to S406 below can be executed in a different order. Also, some of the processes in steps S401 to S406 below may be omitted.

[0144] (4-4-1. Image data reference processing) Firstly, the server device 10 performs image data reference processing (step S401). For example, the server device 10 references the image data ID stored in the image data storage unit 14a.

[0145] (4-4-2. Image Data Evaluation Value Calculation Process) Secondly, the server device 10 performs image data evaluation value calculation processing (step S402). For example, the server device 10 calculates an image data evaluation value IE, which indicates the difference from the normal state of the flow meter 20, based on the referenced image data ID.

[0146] (4-4-3. Image data evaluation value storage process) Thirdly, the server device 10 performs image data evaluation value storage processing (step S403). For example, the server device 10 stores the calculated image data evaluation value IE in the image data storage unit 14a for each flow meter 20.

[0147] (4-4-4. Parameter Reference Processing) Fourth, the server device 10 performs parameter reference processing (step S404). For example, the server device 10 references the parameter PM stored in the parameter storage unit 14b.

[0148] (4-4-5. Parameter evaluation value calculation process) Fifth, the server device 10 performs a parameter evaluation value calculation process (step S405). For example, the server device 10 calculates a parameter evaluation value IE, which represents the difference from the normal state of the flow meter 20, based on the referenced parameter PM.

[0149] (4-4-6. Parameter evaluation value storage process) Sixth, the server device 10 executes the parameter evaluation value storage process (step S406) and terminates the calculation control process. For example, the server device 10 stores the calculated parameter evaluation value IE in the parameter storage unit 14b for each flow meter 20.

[0150] (4-5. Diagnostic and Control Processing) The flow of the diagnostic control process of the flow meter diagnostic system 100 according to the embodiment will be explained using Figure 13. Figure 13 is a flowchart showing an example of the flow of the diagnostic control process of the flow meter diagnostic system 100 according to the embodiment. Note that the processes in steps S501 to S507 below can be executed in a different order. Also, some of the processes in steps S501 to S507 below may be omitted.

[0151] (4-5-1. Image data evaluation value reference processing) Firstly, the server device 10 performs image data evaluation value reference processing (step S501). For example, the server device 10 references the image data evaluation value IE stored in the image data storage unit 14a.

[0152] (4-5-2. Image data evaluation value input processing) Secondly, the server device 10 performs image data evaluation value input processing (step S502). For example, the server device 10 inputs the referenced image data evaluation value IE into the trained machine learning model LM.

[0153] (4-5-3. Parameter evaluation value reference processing) Thirdly, the server device 10 performs parameter evaluation value reference processing (step S503). For example, the server device 10 references the parameter evaluation value PE stored in the parameter storage unit 14b.

[0154] (4-5-4. Parameter evaluation value input processing) Fourth, the server device 10 performs parameter evaluation value input processing (step S504). For example, the server device 10 inputs the referenced parameter evaluation value PE into the trained machine learning model LM.

[0155] (4-5-5. Processing the output of diagnostic results) Fifth, the server device 10 performs diagnostic result output processing (step S505). For example, the server device 10 outputs a diagnostic result DR indicating the health and lifespan of the flow meter 20 based on the state information CI obtained from the trained machine learning model LM.

[0156] (4-5-6. Diagnosis Result Notification Processing) Sixth, the server device 10 performs diagnostic result notification processing (step S506). For example, the server device 10 notifies the worker W of the diagnostic result DR indicating the health and lifespan of the flow meter 20 by displaying the outputted diagnostic result DR on the worker terminal used by the worker W.

[0157] (4-5-7. Processing to store diagnostic results) Seventh, the server device 10 executes the diagnostic result storage process (step S507) and terminates the diagnostic control process. For example, the server device 10 stores the output diagnostic result DR of the flow meter 20 in the diagnostic result storage unit 14c for each flow meter 20.

[0158] [5. Effects of the Embodiment] The effects of the embodiment will be described below. Effects 1 to 13 corresponding to the processing according to the embodiment will be described below.

[0159] (5-1. Effect 1) Firstly, in this embodiment, the server device 10 acquires image data IDs from the measuring tube MP of the flow meter 20, calculates an image data evaluation value IE that indicates the difference from the normal state of the flow meter 20 based on the acquired image data IDs, and diagnoses the flow meter 20 using a machine learning model LM that outputs state information CI indicating the state of the flow meter 20 in response to the input of the image data evaluation value IE. Therefore, in this embodiment, diagnosis can be performed using visual information of the flow meter 20, thereby improving the diagnostic accuracy of the flow meter 20.

[0160] (5-2. Effect 2) Secondly, in this embodiment, the server device 10 acquires the parameter PM held by the flow meter 20, calculates a parameter evaluation value PE that indicates the difference from the normal state of the flow meter 20 based on the acquired parameter PM, and diagnoses the flow meter 20 using a machine learning model LM that outputs state information CI indicating the state of the flow meter 20 in response to at least one input of the image data evaluation value IE and the parameter evaluation value PE. Therefore, in this embodiment, diagnosis can be performed using the visual information and parameter information of the flow meter 20, thereby improving the diagnostic accuracy of the flow meter 20.

[0161] (5-3. Effect 3) Thirdly, in the embodiment, the server device 10 trains a machine learning model LM for each type of flow meter 20 by inputting at least one of the image data evaluation value IE and the parameter evaluation value PE as explanatory variables. Therefore, in the embodiment, the machine learning model LM, which has been trained on the visual information and parameter information of the flow meter 20, can be used, thereby improving the diagnostic accuracy of the flow meter 20.

[0162] (5-4. Effect 4) Fourth, in this embodiment, the server device 10 acquires an image data ID that includes a characteristic portion within the measuring tube MP, corresponding to the type of flow meter 20, during the calibration of the flow meter 20. Therefore, in this embodiment, diagnosis can be performed using visual information corresponding to the type of flow meter 20, thereby improving the diagnostic accuracy of the flow meter 20.

[0163] (5-5. Effect 5) Fifth, in the embodiment, the server device 10 acquires a parameter PM during calibration of the flow meter 20, which includes at least one of the following: the flow signal of the flow meter 20, deposit information regarding deposits, alarm history, operating time, fluid information regarding the flowing fluid FL, and self-diagnosis results. Therefore, in the embodiment, diagnosis can be performed using parameter information according to the type of flow meter 20, thereby improving the diagnostic accuracy of the flow meter 20.

[0164] (5-6. Effect 6) Sixth, in this embodiment, the server device 10 calculates an image data evaluation value IE by considering the image data ID corresponding to the type of flow meter 20 at the time of shipment as the normal state. Therefore, in this embodiment, diagnosis can be performed using visual information based on a new flow meter 20, thereby improving the diagnostic accuracy of the flow meter 20.

[0165] (5-7. Effect 7) Seventh, in this embodiment, the server device 10 calculates the parameter evaluation value PE by considering the theoretical value of the flow meter 20 or the parameter PM corresponding to the type of flow meter 20 at the time of the previous calibration as the normal state. Therefore, in this embodiment, diagnosis can be performed using parameter information based on a new or calibrated flow meter 20, thereby improving the diagnostic accuracy of the flow meter 20.

[0166] (5-8. Effect 8) Eighth, in this embodiment, the server device 10 diagnoses the flow meter 20 using a machine learning model LM that outputs the distance from data indicating the normal or abnormal state of the flow meter 20 in a multidimensional space as state information CI. Therefore, in this embodiment, diagnosis can be performed using the distance from normal or abnormal data in a multidimensional space, thereby improving the diagnostic accuracy of the flow meter 20.

[0167] (5-9. Effect 9) Ninth, in the embodiment, the server device 10 diagnoses the health of the flow meter 20, indicating the degree of normal or abnormal state, by using the distance from a predetermined position in a subspace of a normal cluster that represents the normal state of the flow meter 20 in a multidimensional space. Therefore, in the embodiment, since the health can be diagnosed using the distance from normal data in a multidimensional space, the diagnostic accuracy of the flow meter 20 can be improved.

[0168] (5-10. Effect 10) Tenth, in the embodiment, the server device 10 diagnoses the lifespan of the flow meter 20, which indicates the period until the abnormal state is reached, using the distance from a predetermined position in a subspace of an abnormal cluster that indicates the abnormal state of the flow meter 20 in multidimensional space. Therefore, in the embodiment, since the lifespan can be diagnosed using the distance from abnormal data in multidimensional space, the diagnostic accuracy of the flow meter 20 can be improved.

[0169] (5-11. Effect 11) Eleventh, in this embodiment, the flow meter 20 is an electromagnetic flow meter 20A, and the server device 10 diagnoses the electrode EL of the electromagnetic flow meter 20A. Therefore, in this embodiment, diagnosis can be performed using visual information including characteristic parts of the electromagnetic flow meter 20A, thereby improving the diagnostic accuracy of the flow meter 20.

[0170] (5-12. Effect 12) Twelfth, in this embodiment, the flow meter 20 is a vortex flow meter 20B, and the server device 10 diagnoses the vortex generator VG of the vortex flow meter 20B. Therefore, in this embodiment, diagnosis can be performed using visual information including characteristic parts of the vortex flow meter 20B, thereby improving the diagnostic accuracy of the flow meter 20.

[0171] (5-13. Effect 13) Thirteenth, in this embodiment, the flow meter 20 is a Coriolis flow meter 20C, and the server device 10 diagnoses the measuring tube MP or vibrating tube VT of the Coriolis flow meter 20C. Therefore, in this embodiment, diagnosis can be performed using visual information including characteristic parts of the Coriolis flow meter 20C, thereby improving the diagnostic accuracy of the flow meter 20.

[0172] [6. Examples of applications of the embodiment] Examples of applications of the embodiment will be described below. Examples of applications 1 to 5 of the embodiment will be described below.

[0173] (6-1. Application Example 1) As an application example 1 of the embodiment, the server device 10 can use the raw data from the sensor section of the flow meter 20 as input data to the machine learning model LM. For example, the server device 10 can further use the excitation waveform if it is an electromagnetic flow meter 20A, the frequency if it is a vortex flow meter 20B, the phase difference if it is a Coriolis flow meter 20C, and the frequency difference if it is an ultrasonic flow meter 20D as input data. In application example 1 of the embodiment, this can be further used as useful parameters for diagnosing the lifespan of the sensor section of the flow meter 20.

[0174] (6-2. Application Example 2) As an application example 2 of the embodiment, the server device 10 can further use the 4-20mA current output signal from the flow meter 20 as input data to the machine learning model LM. Therefore, in application example 2 of the embodiment, the degree of deterioration of the flow meter 20 circuit can be determined.

[0175] (6-3. Application Example 3) As an application example 3 of the embodiment, the server device 10 can further use the operating sound of the flow meter 20 as input data to the machine learning model LM. For example, the server device 10 can further use abnormal sounds from the flow meter 20 as input data. Therefore, in application example 3 of the embodiment, the know-how of worker W, who listens to sounds during inspections to determine whether or not there is an abnormality, can be utilized.

[0176] (6-4. Application Example 4) As application example 4 of the embodiment, the server device 10 can use a machine learning model LM that has learned the probability distribution of normal or abnormal, and a classification model that estimates the probability of failure. Therefore, in application example 4 of the embodiment, the health of the flow meter 20 can be easily diagnosed by two-class classification.

[0177] (6-5. Application Example 5) As application example 5 of the embodiment, the server device 10 can utilize a machine learning model LM that has been trained only on data of faulty products. For example, the server device 10 uses an autoencoder to train the machine learning model LM to learn the identity mapping of data of faulty products. When data of faulty products that have not been trained is input to the machine learning model LM trained as described above, it is possible to restore them correctly, so the difference between the input data and the restored data is small. On the other hand, when data of new products that have not been trained is input, it is not possible to restore them correctly, so the difference between the input data and the restored data is large. For this reason, in application example 5 of the embodiment, by defining the above difference as a distance, it is possible to diagnose the distance to the faulty product, i.e., the lifespan.

[0178] [7. System] Unless otherwise specified, the processing procedures, control procedures, specific names, and various data and parameters shown in the above documents and drawings may be changed at will.

[0179] Furthermore, the components of each illustrated device are functionally conceptual and do not necessarily need to be physically configured as shown. In other words, the specific forms of distribution and integration of each device are not limited to those shown. That is, all or part of them can be functionally or physically distributed and integrated in any unit according to various loads and usage conditions.

[0180] Furthermore, each processing function performed by each device may be implemented, in whole or in part, by a CPU and a program executed for analysis by that CPU, or by wired logic hardware.

[0181] [8. Hardware] An example of the hardware configuration of the server device 10 will be explained using Figure 14. Note that other devices can have a similar hardware configuration. Figure 14 is a diagram showing an example of the hardware configuration according to this embodiment. As shown in Figure 14, the server device 10 includes a communication device 10a, an HDD (Hard Disk Drive) 10b, memory 10c, and a processor 10d. Furthermore, the components shown in Figure 14 are interconnected by a bus or the like.

[0182] The communication device 10a is implemented using a network interface card or the like, and communicates with other servers. The HDD 10b stores programs and databases that operate the functions shown in Figure 2.

[0183] The processor 10d operates a process that performs the functions described in Figure 2 by reading a program that performs the same processing as each processing unit shown in Figure 2 from the HDD 10b or the like and loading it into memory 10c. For example, this process performs the same functions as each processing unit of the server device 10. Specifically, the processor 10d reads a program that has the same functions as the acquisition unit 15a, calculation unit 15b, diagnostic unit 15c, training unit 15d, etc. from the HDD 10b or the like. Then, the processor 10d executes a process that performs the same processing as the acquisition unit 15a, calculation unit 15b, diagnostic unit 15c, training unit 15d, etc.

[0184] As described above, the server device 10 operates as a device that executes various processing methods by reading and executing the program according to the embodiment. Furthermore, the server device 10 can also achieve the same functionality as the embodiment by reading the program from the recording medium using a media reader and executing the read program. Note that the program according to the embodiment is not limited to being executed by the server device 10. For example, this disclosure can be similarly applied when another computer or server executes the program, or when they cooperate to execute the program.

[0185] The program according to this embodiment can be distributed via a network such as the Internet. Furthermore, this program can be recorded on a computer-readable recording medium such as a hard disk, flexible disk (FD), CD-ROM, MO (Magneto-Optical disk), or DVD (Digital Versatile Disc), and executed by reading it from the recording medium by a computer.

[0186] [9. Other] Some examples of the combinations of technical features that will be disclosed are listed below.

[0187] (1) A diagnostic device comprising a processor, the processor performing the following actions: acquiring image data inside the measuring pipe of a flow meter; calculating a first evaluation value indicating the difference from the normal state of the flow meter based on the acquired image data; and diagnosing the flow meter using a machine learning model that outputs state information indicating the state of the flow meter in response to the input of the first evaluation value.

[0188] (2) The diagnostic device according to (1), wherein the processor acquires parameters held by the flow meter, calculates a second evaluation value indicating the difference from the normal state of the flow meter based on the acquired parameters, and diagnoses the flow meter using the machine learning model that outputs state information indicating the state of the flow meter in response to at least one input of the first evaluation value and the second evaluation value.

[0189] (3) The diagnostic apparatus according to (2), wherein the processor performs the task of training the machine learning model for each type of flow meter by inputting at least one of the first evaluation value and the second evaluation value as explanatory variables.

[0190] (4) The diagnostic device according to any one of (1) to (3), wherein the processor acquires the image data including characteristic portions inside the measuring tube according to the type of flow meter during calibration of the flow meter.

[0191] (5) The diagnostic device according to (2), wherein the processor acquires the parameters during calibration of the flow meter, including at least one of the flow rate signal of the flow meter, deposit information regarding deposits, alarm history, operating time, fluid information regarding the flowing fluid, and self-diagnosis results.

[0192] (6) The diagnostic device according to any one of (1) to (5), wherein the processor calculates the first evaluation value with respect to the image data corresponding to the type of flow meter at the time of shipment of the flow meter as the normal state.

[0193] (7) The diagnostic device according to (2), wherein the processor calculates the second evaluation value by taking the theoretical value of the flow meter or the parameter corresponding to the type of flow meter at the time of the previous calibration as the normal state.

[0194] (8) The diagnostic device according to any one of (1) to (7), wherein the processor diagnoses the flow meter using the machine learning model which outputs the distance from data indicating a normal or abnormal state of the flow meter in a multidimensional space as state information.

[0195] (9) The diagnostic device according to (8), wherein the processor diagnoses the health of the flow meter, indicating the degree of normal or abnormal state of the flow meter, using the distance from a predetermined position in a subspace of a normal cluster indicating the normal state of the flow meter in a multidimensional space.

[0196] (10) The diagnostic device according to (8), wherein the processor diagnoses the lifespan of the flow meter, which indicates the period until the abnormal state is reached, using the distance from a predetermined position in a subspace of an abnormal cluster indicating an abnormal state of the flow meter in a multidimensional space.

[0197] (11) The diagnostic device according to any one of (1) to (10), wherein the flow meter is an electromagnetic flow meter, and the processor diagnoses the electrodes of the electromagnetic flow meter.

[0198] (12) The diagnostic device according to any one of (1) to (11), wherein the flow meter is a vortex flow meter, and the processor diagnoses the vortex generator of the vortex flow meter.

[0199] (13) The diagnostic device according to any one of (1) to (12), wherein the flow meter is a Coriolis flow meter, and the processor diagnoses the measuring tube or vibrating tube of the Coriolis flow meter.

[0200] (14) A diagnostic method comprising: a computer acquiring image data from inside the measuring pipe of a flow meter; calculating a first evaluation value indicating the difference from the normal state of the flow meter based on the acquired image data; and diagnosing the flow meter using a machine learning model that outputs state information indicating the state of the flow meter in response to the input of the first evaluation value.

[0201] (15) A diagnostic program that causes a computer to perform the following actions: acquire image data of the inside of the measuring pipe of the flow meter; calculate a first evaluation value indicating the difference from the normal state of the flow meter based on the acquired image data; and diagnose the flow meter using a machine learning model that outputs state information indicating the state of the flow meter in response to the input of the first evaluation value. [Explanation of Symbols]

[0202] 10 Server devices 10a Communication device 10b HDD 10c memory 10d processor 11 Input section 12 Output section 13 Communications Department 14 Storage section 14a Image data storage unit 14b Parameter storage unit 14c Diagnostic Result Storage Unit 14d Machine Learning Model Memory Unit 15 Control Unit 15a Acquisition part 15b Calculation part 15c Diagnostic Department 15d Training Department 20 Flow meter 100 Flow meter diagnostic system W Worker

Claims

1. Processor, Equipped with, The aforementioned processor, To acquire image data from inside the measuring pipe of the flow meter, Based on the acquired image data, a first evaluation value is calculated that shows the difference from the normal state of the flow meter. The flow meter is diagnosed using a machine learning model that outputs state information indicating the state of the flow meter in response to the input of the first evaluation value, A diagnostic device that performs this task.

2. The aforementioned processor, The parameters held by the flow meter are acquired, Based on the acquired parameters, a second evaluation value is calculated that shows the difference from the normal state of the flow meter. The flow meter is diagnosed using the machine learning model that outputs state information indicating the state of the flow meter in response to at least one of the first evaluation value and the second evaluation value. The diagnostic device according to claim 1.

3. The aforementioned processor, The machine learning model is trained for each type of flow meter by inputting at least one of the first evaluation value and the second evaluation value as explanatory variables. A diagnostic device according to claim 2, which performs the following:

4. The aforementioned processor, During the calibration of the flow meter, image data including characteristic portions within the measuring tube according to the type of flow meter is acquired. The diagnostic device according to claim 1.

5. The aforementioned processor, During calibration of the flow meter, the parameters are acquired, including at least one of the flow rate signal of the flow meter, deposit information regarding deposits, alarm history, operating time, fluid information regarding the flowing fluid, and self-diagnosis results. The diagnostic device according to claim 2.

6. The aforementioned processor, The first evaluation value is calculated by using the image data corresponding to the type of flow meter at the time of shipment as the normal state. The diagnostic device according to claim 1.

7. The aforementioned processor, The second evaluation value is calculated by taking the theoretical value of the flow meter or the parameter corresponding to the type of flow meter at the time of the previous calibration as the normal state. The diagnostic device according to claim 2.

8. The aforementioned processor, The flow meter is diagnosed using the machine learning model that outputs the distance from data indicating the normal or abnormal state of the flow meter in a multidimensional space, as state information. The diagnostic device according to claim 1 or 2.

9. The aforementioned processor, The soundness of the flow meter, indicating the degree of normal or abnormal state, is diagnosed using the distance from a predetermined position in a subspace of a normal cluster that represents the normal state of the flow meter in a multidimensional space. The diagnostic device according to claim 8.

10. The aforementioned processor, The lifespan of the flow meter, which indicates the period until the abnormal state is reached, is diagnosed using the distance from a predetermined position in a subspace of an abnormal cluster that indicates the abnormal state of the flow meter in a multidimensional space. The diagnostic device according to claim 8.

11. The flow meter is an electromagnetic flow meter, The aforementioned processor, To diagnose the electrodes of the aforementioned electromagnetic flowmeter, The diagnostic device according to claim 1 or 2.

12. The flow meter is a vortex flow meter, The aforementioned processor, To diagnose the vortex generator of the aforementioned vortex flow meter, The diagnostic device according to claim 1 or 2.

13. The flow meter is a Coriolis flow meter, The aforementioned processor, To diagnose the measuring tube or vibrating tube of the Coriolis flow meter, The diagnostic device according to claim 1 or 2.

14. Computers To acquire image data from inside the measuring pipe of the flow meter, Based on the acquired image data, a first evaluation value is calculated that shows the difference from the normal state of the flow meter. The flow meter is diagnosed using a machine learning model that outputs state information indicating the state of the flow meter in response to the input of the first evaluation value, A diagnostic method to perform.

15. On the computer, To acquire image data from inside the measuring pipe of the flow meter, Based on the acquired image data, a first evaluation value is calculated that shows the difference from the normal state of the flow meter. The flow meter is diagnosed using a machine learning model that outputs state information indicating the state of the flow meter in response to the input of the first evaluation value, A diagnostic program that executes the following:

Images