Plant structure corrosion prediction device, inference device, machine learning device, information prediction device, plant structure corrosion prediction method, inference method, machine learning method, and information prediction method

Abstract

Provided is a plant structure corrosion prediction device which enables prediction accuracy to be simply improved when predicting internal corrosion of a structure.　The Plant structure corrosion prediction device (7) comprises: an information acquisition unit (700) which acquires operation information in a plant in which a structure to be predicted is installed; a first prediction processing unit (701) which predicts corrosion environment information by inputting the operation information to be predicted, which is acquired by the information acquisition unit (700), to a first training model (14A) that has been trained, by means of machine learning, about the relationship between operation information to be trained and corrosion environment information that is a cause of the internal corrosion of the structure; and a second prediction processing unit (702) which predicts corrosion speed information by inputting the operation information to be predicted that is acquired by the information acquisition unit and the corrosion environment information to be predicted, that is predicted by the first prediction processing unit, to a second training model (14B) which has been trained, by means of machine learning, about the relationship between the operation information and corrosion environment information to be trained, and the corrosion speed information that indicates the speed of the internal corrosion of the structure.

Description

Technical Field

[0001] The present invention relates to a plant structure corrosion prediction device, an inference device, a machine learning device, an information prediction device, a plant structure corrosion prediction method, an inference method, a machine learning method, and an information prediction method.

Background Art

[0002] In a plant, as structures through which a predetermined fluid flows inside, a large number of pipes, reaction towers, distillation towers, tanks, boilers, heating furnaces, heat exchangers, etc. are installed, and a predetermined manufacturing process is performed by the operation of the plant. In an operating plant, deterioration and thinning due to internal surface corrosion of the structure progress, leading to plant shutdowns and accidents. Therefore, it is necessary to manage the status of internal surface corrosion of the structure and appropriately perform maintenance.

[0003] Therefore, as a conventional technique for managing the status of internal surface corrosion of a structure, Patent Document 1 discloses an equipment safety management system that predicts the corrosion rate of a target facility based on operation data of the target facility. Further, Patent Document 2 discloses a maintenance support device including a detection unit that detects the environmental state and corrosion state of a structure to be maintained, and a device body that calculates the probability of reaching the end of the life of the structure based on the detection results of the detection unit.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] Patent Document 1 discloses an equipment safety management system in which operating data such as operating temperature, pipe material, pipe size, pipe shape, years of service, corrosion system, flow conditions, fluid velocity, and operating pressure are recorded. However, when predicting the corrosion rate of the target equipment, data other than the above-mentioned operating data is not taken into consideration, so there were limitations to improving the accuracy of the prediction.

[0006] Furthermore, the maintenance support device disclosed in Patent Document 2 exemplifies the use of, for example, a thermometer / hygrometer, a neutron moisture meter, a thermal camera, an ACM sensor, an acoustic emission measuring instrument, an ultrasonic meter, an X-ray imaging device, and an intelligent pig as detection units. However, installing such detection units in each structure requires not only design and construction for the installation of the detection units, but also inspection and calibration to maintain normal operation, and further increases in costs are expected due to each of these tasks.

[0007] In view of the above problems, the present invention aims to provide a plant structure corrosion prediction device, inference device, machine learning device, information prediction device, plant structure corrosion prediction method, inference method, machine learning method, and information prediction method that can easily improve the prediction accuracy when predicting internal corrosion of structures. [Means for solving the problem]

[0008] To achieve the above objective, a plant structure corrosion prediction device according to one aspect of the present invention is: A plant structure corrosion prediction device installed in a plant that predicts corrosion rate information for internal corrosion of a structure through which a predetermined fluid flows, An information acquisition unit that acquires operating information including one or more operating parameters in the plant where the structure to be predicted is installed, A first prediction processing unit predicts the corrosion environment information for the target operation information by inputting the operation information to be predicted, acquired by the information acquisition unit, into a first learning model that has been trained by machine learning to determine the relationship between the operation information to be learned and corrosion environment information including one or more corrosion environment parameters that are factors for internal corrosion of the structure installed in the plant operated with the operation information to be learned, and The system includes a second prediction processing unit that inputs the operation information of the target to be predicted, acquired by the information acquisition unit, and the corrosion environment information of the target to be predicted, predicted by the first prediction processing unit, into a second learning model that has been trained by machine learning on the relationship between the operation information and corrosion environment information to be learned and corrosion rate information representing the rate of internal corrosion of the structure installed in the plant operated with the operation information to be learned, thereby predicting the corrosion rate information for the operation information and corrosion environment information of the target to be predicted. [Effects of the Invention]

[0009] According to one aspect of the present invention, a plant structure corrosion prediction device predicts (supplements) corrosion environment information that contributes to improving the prediction accuracy of corrosion rate information by inputting operating information into a first learning model, and a second prediction processing unit predicts corrosion rate information by inputting operating information and the predicted (supplemented) corrosion environment information into a second learning model, thereby easily improving the prediction accuracy when predicting internal corrosion of a structure.

[0010] Other issues, configurations, and effects will be clarified in the embodiments for carrying out the invention described later. [Brief explanation of the drawing]

[0011] [Figure 1] This is an overall diagram showing an example of plant management system 1 and plant 10. [Figure 2] This is a data configuration diagram showing an example of the operation database 50, the corrosion environment database 51, and the inspection database 52. [Figure 3] It is a block diagram showing an example of the machine learning device 6. [Figure 4] It is a schematic diagram showing the relationship between operation information, corrosion environment information, and structure inspection information, and the first and second learning data 13A and 13B. [Figure 5] It is a diagram showing an example of the first learning model 14A and the first learning data 13A. [Figure 6] It is a diagram showing an example of the second learning model 14B and the second learning data 13B. [Figure 7] It is a block diagram showing an example of the plant structure corrosion prediction device 7. [Figure 8] It is a functional explanatory diagram showing an example of the plant structure corrosion prediction device 7. [Figure 9] It is a schematic diagram showing the relationship between operation information, corrosion environment information, and structure inspection information, and the first and second learning models 14A and 14B. [Figure 10] It is a hardware configuration diagram showing an example of the computer 900. [Figure 11] It is a flowchart showing an example of the machine learning method by the machine learning device 6. [Figure 12] It is a flowchart showing an example of the plant structure corrosion prediction method by the plant structure corrosion prediction device 7. [Figure 13] It is a diagram showing an example of the display screen for displaying the prediction result by the plant structure corrosion prediction device 7.

Embodiments for Carrying Out the Invention

[0012] Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings. In the following, the scope necessary for the description to achieve the object of the present invention is schematically shown, and mainly the scope necessary for the description of the relevant part of the present invention will be described, and the parts where the description is omitted shall be based on known techniques.

[0013] (Plant Management System 1) FIG. 1 is an overall view showing an example of a plant management system 1 and a plant 10. The plant management system 1 according to the present embodiment functions as a system that predicts corrosion rate information of the inner surface corrosion of a structure 11 installed in the plant 10 and manages the remaining life or wall thickness of the structure 11 based on the prediction result. The plant 10 is, for example, an arbitrary plant such as a natural gas plant, an oil refining plant, a chemical treatment plant, a power generation plant, an iron making plant, etc., and is not limited to these examples.

[0014] In the plant 10, for example, a structure 11 through which an arbitrary fluid such as a gas, a liquid, or a granular material having fluidity flows inside is installed, and a predetermined manufacturing process is performed. The structure 11 is, for example, a pipe, a reaction tower, a distillation tower, a tank, a boiler, a heating furnace, a heat exchanger, etc. The inner surface corrosion of the structure 11 is mainly predicted and managed for the pipe, but any structure 11 other than the pipe may be predicted and managed.

[0015] At each position of the structure 11, a plurality of devices 12 for controlling the manufacturing process are installed. The device 12 includes, for example, an instrument 12A such as a sensor that measures operation parameters such as the temperature, pressure, flow velocity, flow rate, liquid level, etc. of the fluid flowing inside the structure 11 and outputs a sensor signal indicating the measurement result of the operation parameters, and a controller 12B such as a valve, a pump, a compressor, etc. that controls the flow rate, pressure, temperature, liquid level, component, etc. of the fluid flowing inside the structure 11 according to the control parameter when a control signal indicating the control parameter is input. Note that the device 12 is not limited to the above examples, and may include a part of the structure 11, or may be an instrument 12A that measures physical quantities such as temperature and humidity in the surrounding environment of the structure 11, or may include a controller 12B that controls those physical quantities.

[0016] The plant management system 1, as shown in Figure 1, comprises a plant operation management device 2, a worker terminal device 3, a fluid analyzer 4A, a structural inspection device 4B, a data management device 5, a machine learning device 6, and a plant structural corrosion prediction device 7. Each of the devices 2 to 7 is, for example, composed of a general-purpose or dedicated computer (see Figure 10 below) and connected to a wired or wireless network 8, enabling the mutual transmission and reception of various types of data. The number of each device 2 to 7 and the connection configuration of the network 8 are not limited to the example in Figure 1 and may be changed as appropriate.

[0017] The plant operation management device 2 is a device that manages the control of the manufacturing process, emergency shutdowns, and safety monitoring of the plant 10 by sending and receiving various device signals (sensor signals and control signals) with a plurality of devices 12 installed at various locations in the plant 10 (structure 11). Specifically, the plant operation management device 2 receives sensor signals (also called input signals) from a device 12 that functions as an instrument 12A among the plurality of devices 12, and based on the sensor information indicated by the sensor signals, transmits control signals (also called output signals) to a device 12 that functions as a controller 12B among the plurality of devices 12.

[0018] The worker terminal device 3 is a terminal device used by workers (operators, maintenance workers, inspection workers, managers, etc.) working at the plant 10, and may be a stationary or portable device. The worker terminal device 3 has programs such as applications and browsers installed on it, accepts various input operations, and outputs various information via a display screen or voice.

[0019] The fluid analyzer 4A is a device that analyzes corrosion environment parameters such as the concentration, pressure (partial pressure), density, and content of predetermined substances and ions contained in a sample (specimen) taken from a portion of the fluid flowing inside the structure 11, and outputs the analysis results of those corrosion environment parameters. The fluid analyzer 4A is a device that consists of, for example, an analysis kit and an analyzer, and is capable of analyzing one or more corrosion environment parameters.

[0020] Corrosion environment parameters are parameters that represent the factors causing internal corrosion of the structure 11. Corrosion environment parameters are specified, for example, according to the type of fluid flowing inside the structure 11 and the type of internal corrosion of the structure 11. For example, it is preferable that corrosion environment parameters for high-temperature sulfide corrosion include sulfur concentration, corrosion environment parameters for naphthene corrosion include sulfur concentration and total acid number, corrosion environment parameters for high-temperature hydrogen / hydrogen sulfide corrosion include hydrogen sulfide concentration, corrosion environment parameters for acid corrosion include acid concentration, corrosion environment parameters for alkaline corrosion include alkali concentration, and corrosion environment parameters for carbonate corrosion include CO2 concentration, CO2 partial pressure, moisture content, and fluid density. Note that the substances causing internal corrosion are not limited to the above examples, as long as they are parameters that represent the factors causing internal corrosion of the structure 11, and the types of internal corrosion are not limited to the above examples.

[0021] If the fluid analyzer 4A is configured to transmit the analysis results of corrosion environment parameters as data, it transmits corrosion environment information, including the analysis results of those corrosion environment parameters, to the data management device 5. If the fluid analyzer 4A does not have the function to transmit the above-mentioned corrosion environment information, the operator performs an input operation to input the analysis results of corrosion environment parameters to the operator terminal device 3, and the operator terminal device 3 transmits the corrosion environment information based on that input operation to the data management device 5. Note that multiple types of fluid analyzers 4A may be used, each with different target corrosion environment parameters and analysis methods.

[0022] The structural inspection device 4B is a device that inspects the thickness of the structure 11 during large-scale inspections of the plant 10 or when replacing the structure 11, and outputs the inspection result of the thickness. If the structural inspection device 4B is configured to transmit the thickness inspection result as data, it transmits structural inspection information, including the thickness inspection result, to the data management device 5. If the structural inspection device 4B does not have the function to transmit such structural inspection information, the worker performs an input operation to input the thickness inspection result to the worker terminal device 3, and the worker terminal device 3 transmits structural inspection information based on that input operation to the data management device 5.

[0023] The data management device 5 is a device that manages various types of data related to the plant 10 as a database, and includes an operation database 50, a corrosion environment database 51, and an inspection database 52.

[0024] The operation database 50 is a database in which measurement results (operation information) of operating parameters measured at predetermined measurement intervals by equipment 12 (mainly instrument 12A) when a manufacturing process is carried out at plant 10 are registered and stored. The corrosion environment database 51 is a database in which analysis results (corrosion environment information) of corrosion environment parameters are registered and stored when a sample is analyzed by the fluid analyzer 4A at predetermined analysis intervals (>measurement intervals). The inspection database 52 is a database in which inspection results (structural inspection information) of the wall thickness of the structure 11 are registered and stored when the wall thickness of the structure 11 is inspected by the structural inspection device 4B at predetermined inspection intervals (>analysis intervals). Note that the operation database 50, corrosion environment database 51, and inspection database 52 may register and store data from multiple plants 10, and the details of the data configuration will be described later.

[0025] The machine learning device 6 is the device that operates as the main component of the machine learning learning phase. For example, the machine learning device 6 acquires first and second training data 13A and 13B (details described later) from the data management device 5, and generates first and second learning models 14A and 14B (details described later) used in the plant structure corrosion prediction device 7 using machine learning based on the first and second training data 13A and 13B. The trained first and second learning models 14A and 14B are provided to the plant structure corrosion prediction device 7 via the network 8, recording media, etc.

[0026] The plant structure corrosion prediction device 7 is a device that operates as the main component of the machine learning inference phase. The plant structure corrosion prediction device 7 uses the first and second learning models 14A and 14B generated by the machine learning device 6 to predict corrosion rate information representing the corrosion rate of internal corrosion in the target structure 11. The corrosion rate information as a result of this prediction is provided, for example, to the worker terminal device 3 and presented to the worker. The corrosion rate information may also be provided to the data management device 5 and registered in the inspection database 52, etc.

[0027] Figure 2 is a data configuration diagram showing an example of the operation database 50, the corrosion environment database 51, and the inspection database 52. Structures 11 installed in the plant 10, one or more inspection points and equipment 12 contained within structures 11 are each assigned structure IDs, inspection point IDs, and equipment IDs, respectively, as information (numbers, letters, or combinations thereof) representing, for example, a unique identification number, identification code, identification name, identification tag, etc., thereby linking the information between each database 50-52. In addition, by referring to the plot plan diagram of the plant 10, the P&ID diagram (Piping & Instrument Diagram), and the I / O (input / output) list (none of which are shown), the flow of the manufacturing process and the arrangement and physical connection relationships of structures 11 and equipment 12 can be identified.

[0028] The operation database 50 registers, for example, the measurement date and time, the type of structure 11, and one or more operation parameters included in the operation information, using the structure ID that identifies each structure 11 as the key. In the example in Figure 2, temperature, pressure, and flow rate (or flow velocity) are registered as multiple operation parameters. For example, for a single structure 11, operation parameters measured upstream of the structure 11 and operation parameters measured downstream of the structure 11 may be registered. In addition, the values ​​of the operation parameters may be registered as normalized values ​​in the range of 0 to 1.

[0029] The corrosion environment database 51 registers, for example, the date and time of analysis, the type of fluid, and one or more corrosion environment parameters included in the corrosion environment information, using the structure ID that identifies each structure 11 as the key. In the example in Figure 2, multiple corrosion environment parameters registered include pH, sulfur concentration, hydrogen sulfide concentration, acid concentration, alkali concentration, CO2 concentration, CO2 partial pressure, moisture content, and fluid density. Note that the corrosion environment parameters registered for each structure 11 may be changed according to the type of structure 11, the type of structure 11 (material, dimensions, shape, installation location, etc.), the type of fluid flowing inside the structure 11, the type of internal corrosion of the structure 11, the type of fluid analyzer 4A (inspection target and inspection method), etc. Furthermore, the values ​​of the corrosion environment parameters may be registered as normalized values ​​in the range of 0 to 1, for example.

[0030] The inspection database 52 registers, for example, the inspection date and time, structure type, the wall thickness of the structure 11 included in the structure inspection information, and the corrosion rate (calculated value) of the structure 11 calculated from the structure inspection information, using the structure ID that identifies each structure 11 and the inspection point ID that identifies each inspection point as keys. The corrosion rate (calculated value) is a calculated value obtained by dividing the difference between the previous wall thickness inspection result and the current wall thickness inspection result by the time difference between the previous inspection time and the current inspection time. One or more inspection points are set for each structure 11, and each inspection point is associated with an inspection point table (not shown) using the structure ID and inspection point ID as keys. The inspection point table registers information such as pipe diameter, pipe shape (straight, elbow, etc.), and flow condition (whether it is a stagnant area, etc.) for each inspection point. The flow rate registered in the operation database 50 is converted to flow velocity using the information of the inspection points registered in the inspection point table (pipe diameter, pipe shape, flow condition, etc.), so that the flow velocity can be used as the operation information for each inspection point included in each structure 11.

[0031] (Machine learning device 6) Figure 3 is a block diagram showing an example of a machine learning device 6. The machine learning device 6 comprises a control unit 60, a communication unit 61, a training data storage unit 62, and a trained model storage unit 63.

[0032] The control unit 60 functions as a training data acquisition unit 600 and a machine learning unit 601. The communication unit 61 is connected to an external device via the network 8 and functions as a communication interface for sending and receiving various types of data.

[0033] The learning data acquisition unit 600 is connected to external devices (e.g., operator terminal device 3, data management device 5, etc.) via the communication unit 61 and the network 8, and acquires first learning data 13A, which consists of operation information as input data and corrosion environment information as output data, and second learning data 13B, which consists of operation information and corrosion environment information as input data and corrosion rate information as output data. The first and second learning data 13A and 13B are used as training data, validation data, and test data in supervised learning. In addition, the output data of the first and second learning data 13A and 13B are used as correct data (hereinafter referred to as correct labels) in supervised learning.

[0034] The learning data storage unit 62 is a database that stores multiple sets of the first and second learning data 13A and 13B acquired by the learning data acquisition unit 600. The specific configuration of the database constituting the learning data storage unit 62 can be designed as appropriate.

[0035] The machine learning unit 601 uses multiple sets of first and second training data 13A and 13B stored in the training data storage unit 62 to perform machine learning on the first and second training models 14A and 14B, respectively. When performing machine learning, the machine learning unit 601 can employ any method, such as online learning, batch learning, or mini-batch learning. The machine learning unit 601 may also perform predetermined preprocessing on the input data to be input to the first and second training models 14A and 14B, or perform predetermined postprocessing on the output data output from the first and second training models 14A and 14B.

[0036] The trained model storage unit 63 stores the trained first and second trained models 14A and 14B (specifically, the adjusted weight parameter groups) generated by the machine learning unit 601. The trained trained models 14A and 14B stored in the trained model storage unit 63 are provided to the actual system (for example, the plant structure corrosion prediction device 7) via the network 8 or a recording medium. In Figure 3, the training data storage unit 62 and the trained model storage unit 63 are shown as separate storage units, but they may be composed of a single storage unit. Furthermore, the training data storage unit 62 and the trained model storage unit 63 may be replaced by the storage unit of an external computer (which may be the data management device 5 in this embodiment), in which case the machine learning unit 601 only needs to access the external computer.

[0037] Figure 4 is a schematic diagram showing the relationship between operating information, corrosion environment information, and structural inspection information, and the first and second learning data 13A and 13B. In the example in Figure 4, the operating information includes operating parameters such as temperature, pressure, and flow velocity; the corrosion environment information includes corrosion environment parameters for carbonate corrosion such as CO2 concentration, CO2 partial pressure, moisture content, and fluid density; and the structural inspection information includes wall thickness and corrosion rate (calculated value). In Figure 4, the horizontal axis is time, and the acquisition timing of each piece of information is: measurement cycle for operating information < analysis cycle for corrosion environment information < inspection cycle for structural inspection information. In principle, these acquisition timings should occur regularly, but they may also occur irregularly.

[0038] Operating information (temperature, pressure, flow rate) is the measurement result obtained by instrument 12A at a predetermined measurement cycle (seconds, minutes, hours, etc.). Corrosion environment information (CO2 concentration, CO2 partial pressure, moisture content, fluid density) is the analysis result obtained when a sample taken from the fluid flowing inside the structure 11 is analyzed by fluid analyzer 4A at a predetermined analysis cycle (>measurement cycle). Corrosion rate information (wall thickness, corrosion rate (calculated value)) is the inspection result obtained when the wall thickness of the structure 11 is inspected by fluid analyzer 4A at a predetermined inspection cycle (>analysis cycle).

[0039] The first training data 13A may be a combination of operating information and corrosion environment information at the same point in time (reference numeral 130 in Figure 4), or a combination of operating information and corrosion environment information at the same time period (reference numerals 131 and 132 in Figure 4). The second training data 13B may be a combination of operating information, corrosion environment information, and corrosion rate information at the same point in time (reference numeral 130 in Figure 4), or a combination of operating information, corrosion environment information, and corrosion rate information at the same time period (reference numerals 131 and 132 in Figure 4).

[0040] Figure 5 shows an example of the first learning model 14A and the first training data 13A. The first training data 13A used for machine learning of the first learning model 14A consists of operating information as input data and corrosion environment information as output data. In the example in Figure 5, the operating information includes operating parameters such as temperature, pressure, and flow rate, and the corrosion environment information includes corrosion environment parameters for carbonate corrosion such as CO2 concentration, CO2 partial pressure, moisture content, and fluid density.

[0041] The operating information constituting the first learning data 13A includes one or more operating parameters in the plant 10 where the structure 11 to be learned is installed. The operating parameters are step values ​​or continuous values, and in the case of continuous values, they may be values ​​normalized to a predetermined range (e.g., 0 to 1).

[0042] The operating parameters may be time-series data showing the operating state at a predetermined time, as indicated by reference numeral 130 in Figure 4; time-series data consisting of multiple time-series data included in a predetermined period, as indicated by reference numerals 131 and 132 in Figure 4; or statistical data showing a statistical quantity (e.g., average value) for multiple time-series data included in a predetermined period. The operating information shown in Figure 5 consists of three operating parameters: temperature, pressure, and flow velocity. However, it may also include operating parameters for multiple locations on the structure 11. For example, it may include six operating parameters consisting of temperature, pressure, and flow velocity on the upstream side of the structure 11, and temperature, pressure, and flow velocity on the downstream side of the structure 11. The definition of the operating information may be changed as appropriate in addition to the above example. In that case, the data structure of the input data in the first learning model 14A and the first learning data 13A should be changed as appropriate.

[0043] The corrosion environment information constituting the first training data 13A includes one or more corrosion environment parameters. The corrosion environment parameters are step values ​​or continuous values, and in the case of continuous values, they may be values ​​normalized to a predetermined range (e.g., 0 to 1).

[0044] The corrosion environment parameters may be time-series data showing the corrosion environment parameters at a predetermined point in time, as indicated by reference numeral 130 in Figure 4; time-series data consisting of multiple time-series data included in a predetermined period, as indicated by reference numerals 131 and 132 in Figure 4; or statistical data showing the statistical values ​​(e.g., average values) of multiple time-series data included in a predetermined period. The corrosion environment information shown in Figure 5 consists of four corrosion environment parameters: CO2 concentration, CO2 partial pressure, moisture content, and fluid density. However, it may also include corrosion environment parameters for multiple locations on the structure 11. For example, it may include eight corrosion environment parameters consisting of CO2 concentration, CO2 partial pressure, moisture content, and fluid density upstream of the structure 11, and CO2 concentration, CO2 partial pressure, moisture content, and fluid density downstream of the structure 11. The definition of the corrosion environment information may be changed as appropriate in addition to the above example. In that case, the data structure of the output data in the first learning model 14A and the first learning data 13A should be changed as appropriate.

[0045] The learning data acquisition unit 600 acquires first learning data 13A by referring to various information registered in the operation database 50 and the corrosion environment database 51, and by receiving input operations from the worker terminal device 3. When the learning data acquisition unit 600 refers to the operation database 50 and the corrosion environment database 51, it acquires first learning data 13A by, for example, acquiring operation information and corrosion environment information associated with a specific structure ID.

[0046] The first learning model 14A employs, for example, a neural network structure and comprises an input layer 140, a hidden layer 141, and an output layer 142. Synapses (not shown) connect each neuron between each layer, and each synapse is associated with a weight. The weight parameters, consisting of the weights of each synapse, are adjusted by machine learning.

[0047] The input layer 140 has a number of neurons corresponding to the driving parameters included in the driving information as input data, and each value of the driving parameter is input to each neuron. The output layer 142 has a number of neurons corresponding to the corrosion environment parameters included in the corrosion environment information as output data, and the prediction result (inference result) of the corrosion environment parameters for the driving information is output as output data. If the first learning model 14A is composed of a regression model, the corrosion environment parameters are output as numerical values ​​normalized to a predetermined range (e.g., 0 to 1). If the first learning model 14A is composed of a classification model, the corrosion environment parameters are output as scores (accuracies) for each class, as numerical values ​​normalized to a predetermined range (e.g., 0 to 1).

[0048] The machine learning unit 601 inputs multiple sets of the first training data 13A into the first learning model 14A and generates a trained first learning model 14A by having the first learning model 14A learn the relationship between the operating information and the corrosion environment information contained in the first training data 13A.

[0049] Figure 6 shows an example of the second learning model 14B and the second training data 13B. The second training data 13B used for machine learning of the second learning model 14B consists of operating information and corrosion environment information as input data, and corrosion rate information as output data. In the example in Figure 6, the operating information includes operating parameters such as temperature, pressure, and flow rate; the corrosion environment information includes corrosion environment parameters for carbonate corrosion such as CO2 concentration, CO2 partial pressure, moisture content, and fluid density; and the structural inspection information includes corrosion rate (predicted value).

[0050] The operating information constituting the second training data 13B includes one or more operating parameters of the structure 11 to be trained, similar to the input data of the first training data 13A. Since the operating information constituting the second training data 13B is the same as the operating information constituting the first training data 13A, a detailed explanation is omitted.

[0051] The corrosion environment information constituting the second training data 13B includes one or more corrosion environment parameters for the structure 11 being trained, similar to the output data of the first training data 13A. Since the corrosion environment information constituting the second training data 13B is the same as that constituting the first training data 13A, a detailed explanation is omitted.

[0052] The corrosion rate information constituting the second training data 13B represents the corrosion rate of internal corrosion in the structure 11 being trained. The corrosion rate can be a stepped value or a continuous value, and in the case of a continuous value, it may be a value normalized to a predetermined range (for example, 0 to 1).

[0053] The learning data acquisition unit 600 acquires second learning data 13B by referring to various information registered in the operation database 50, the corrosion environment database 51, and the inspection database 52, and by receiving input operations from the worker terminal device 3. When the learning data acquisition unit 600 refers to the operation database 50, the corrosion environment database 51, and the inspection database 52, it acquires second learning data 13B by, for example, acquiring operation information, corrosion environment information, and corrosion rate (calculated value) associated with a specific structure ID and inspection point ID.

[0054] The machine learning unit 601 inputs multiple sets of the second training data 13B into the second learning model 14B and trains the second learning model 14B on the relationship between the operating information and corrosion environment information contained in the second training data 13B and the corrosion rate information, thereby generating a trained second learning model 14B. The second learning model 14B is configured in the same way as the first learning model 14A, so its explanation is omitted.

[0055] In this embodiment, the data configurations of the first and second learning models 14A and 14B and the first and second learning data 13A and 13B are described as being as shown in Figures 5 and 6. However, multiple data configurations with different conditions may be adopted, for example, such as the machine learning method, the type of structure 11 (material, dimensions, shape, installation location, etc.), the type and flow characteristics of the fluid flowing inside the structure 11, the type of internal corrosion of the structure 11, the type of fluid analyzer 4A (inspection target and inspection method), the type of data included in the operation information, and the type of data included in the corrosion environment information. In such cases, the learning data acquisition unit 600 acquires multiple types of learning data corresponding to the multiple data configurations with different conditions, and the machine learning unit 601 performs machine learning on each learning model using these learning data.

[0056] (Plant structure corrosion prediction device 7) Figure 7 is a block diagram showing an example of a plant structure corrosion prediction device 7. Figure 8 is a functional diagram showing an example of a plant structure corrosion prediction device 7. Figure 9 is a schematic diagram showing the relationship between operating information, corrosion environment information, structural inspection information, and the first and second learning models 14A and 14B.

[0057] The plant structure corrosion prediction device 7 comprises a control unit 70, a communication unit 71, and a storage unit 72. The control unit 70 functions as an information acquisition unit 700, a first prediction processing unit 701, a second prediction processing unit 702, a third prediction processing unit 703, and an output processing unit 704. The communication unit 71 is connected to an external device via a network 8 and functions as a communication interface for sending and receiving various types of data.

[0058] The information acquisition unit 700 is connected to external devices (e.g., plant operation management device 2, worker terminal device 3, data management device 5, etc.) via the communication unit 71 and network 8, and acquires operation information including one or more operation parameters at the plant 10 where the structure 11 to be predicted is installed.

[0059] For example, the information acquisition unit 700 acquires operational information for the predicted structure by receiving sensor signals from each instrument 12A (in the example in Figure 8, a temperature sensor, a pressure sensor, and a flow velocity sensor) installed on the structure 11 to be predicted. Furthermore, if the measurement results of the instruments 12A installed on the structure 11 to be predicted are registered in the operation database 50, the information acquisition unit 700 acquires operational information for the predicted structure by referring to the operation database 50. In addition, the information acquisition unit 700 may acquire operational information for the predicted structure by referring to the operation plan of the plant 10 managed by the plant operation management device 2. The operation plan, for example, specifies time-series data of control parameters for each controller 12B.

[0060] The information acquisition unit 700 acquires operational information for the structure 11 to be predicted, as shown in the first prediction method indicated by reference numeral 140 in Figure 9. However, it may also acquire corrosion environment information for the structure 11 to be predicted along with the operational information for the structure 11 to be predicted. In this case, if the corrosion environment information is time-series data or statistical data that includes multiple corrosion environment parameters, the information acquisition unit 700 may acquire some of the multiple corrosion environment parameters for the structure 11 to be predicted, as shown in the second prediction method indicated by reference numeral 141 in Figure 9. Furthermore, if the corrosion environment information is time-series data that includes corrosion environment parameters at multiple points in time, the information acquisition unit 700 may acquire some of the corrosion environment parameters at multiple points in time for the structure 11 to be predicted, as shown in the third prediction method indicated by reference numeral 142 in Figure 9.

[0061] The first prediction processing unit 701 inputs the operating information of the target structure 11, acquired by the information acquisition unit 700, as input data to the first learning model 14A, thereby predicting corrosion environment information for the target structure. The first prediction processing unit 701 may perform predetermined pre-processing on the input data (operating information) input to the first learning model 14A, or it may perform predetermined post-processing on the output data (corrosion environment information) output from the first learning model 14A.

[0062] Furthermore, if the information acquisition unit 700 acquires some of the multiple corrosion environment parameters of the structure 11 to be predicted (second prediction method 141 shown in Figure 9), the first prediction processing unit 701 may input the operating information of the structure to be predicted acquired by the information acquisition unit 700 as input data to the first learning model 14A, thereby predicting the corrosion environment parameters other than some of the multiple corrosion environment parameters.

[0063] Furthermore, if the information acquisition unit 700 acquires corrosion environment parameters for some of the multiple time points of the structure 11 to be predicted (third prediction method 142 shown in Figure 9), the first prediction processing unit 701 may input the operation information of the structure to be predicted, including the operation parameters for the remaining time points acquired by the information acquisition unit 700, as input data to the first learning model 14A, thereby predicting the corrosion environment parameters for the remaining time points among the multiple time points of corrosion environment parameters.

[0064] The second prediction processing unit 702 inputs the operating information of the target structure 11 acquired by the information acquisition unit 700 and the corrosion environment information of the target structure 11 predicted by the first prediction processing unit 701 as input data to the second learning model 14B, thereby predicting corrosion rate information for the operating information and corrosion environment information of the target structure. The second prediction processing unit 702 may perform predetermined pre-processing on the input data (operating information and corrosion environment information) input to the second learning model 14B, or may perform predetermined post-processing on the output data (corrosion rate information) output from the first learning model 14A.

[0065] Furthermore, if the information acquisition unit 700 acquires some of the multiple corrosion environment parameters in the structure 11 to be predicted (second prediction method 141 shown in Figure 9), the second prediction processing unit 702 may input the operating information of the structure to be predicted acquired by the information acquisition unit 700, and the corrosion environment information of the structure to be predicted, which includes some of the corrosion environment parameters acquired by the information acquisition unit 700 and the remaining corrosion environment parameters predicted by the first prediction processing unit 701, as input data to the second learning model 14B to predict the corrosion rate information.

[0066] Furthermore, if the information acquisition unit 700 acquires corrosion environment parameters for some of the multiple time points of corrosion environment parameters at the structure 11 to be predicted (third prediction method 142 shown in Figure 9), the second prediction processing unit 702 may input the operating information of the target to be predicted acquired by the information acquisition unit 700, the corrosion environment information of the target to be predicted which includes the corrosion environment parameters at some time points acquired by the information acquisition unit 700 and the corrosion environment parameters at other time points predicted by the first prediction processing unit 701, as input data to the second learning model 14B to predict corrosion rate information.

[0067] The third prediction processing unit 703 predicts structural prediction information representing the remaining life or wall thickness of the target structure 11 based on the corrosion rate information (predicted corrosion rate value) of the target structure predicted by the second prediction processing unit 702. The structural prediction information is used, for example, in formulating a maintenance plan. The data necessary for predicting the structural prediction information (for example, the initial value and usage limit value of the wall thickness of the structure 11) is stored in advance in, for example, the storage unit 72 or the data management device 5.

[0068] The output processing unit 704 performs output processing to output at least one of the corrosion environment information predicted by the first prediction processing unit 701, the corrosion rate information predicted by the second prediction processing unit 702, and the structural lifespan information predicted by the third prediction processing unit 703. For example, the output processing unit 704 may transmit screen information for displaying the corrosion environment information, corrosion rate information, and structural lifespan information to be output to the worker terminal device 3, thereby displaying the information on the display screen of the worker terminal device 3, or it may transmit the corrosion rate information, corrosion rate information, and structural lifespan information to the data management device 5, thereby registering the information in the inspection database 52.

[0069] The output processing unit 704 may also output alarm information as part of its output processing, based on the corrosion rate information, corrosion rate information, and structural lifespan information to be output. For example, the storage unit 72 or data management device 5 may store alarm judgment values ​​(data that can be displayed and edited on the worker terminal device 3) that represent upper and lower limits for the corrosion environment information, corrosion rate information, and structural lifespan information, and the output processing unit 704 may output alarm information when it determines that the corrosion environment information, corrosion rate information, and structural lifespan information exceed or fall below those alarm judgment values.

[0070] The memory unit 72 stores the first and second trained models 14A and 14B that are used by the first and second prediction processing units 701 and 702. The number of trained models stored in the memory unit 72 is not limited to the above example, and multiple trained models with different conditions may be stored and selectively used, such as machine learning methods, types of structures 11 (material, dimensions, shape, installation location, etc.), types and flow characteristics of fluids flowing inside the structure 11, types of internal corrosion of the structure 11, types of fluid analyzers 4A (inspection targets and inspection methods), types of data included in operating information, and types of data included in corrosion environment information. The memory unit 72 may also be replaced by the memory unit of an external computer (which may be the data management device 5 in this embodiment), in which case the first and second prediction processing units 701 and 702 only need to access the external computer.

[0071] (Hardware configuration of each device) Figure 10 is a hardware configuration diagram showing an example of computer 900. Each of the devices 2 to 7 of the plant management system 1 is configured by a general-purpose or dedicated computer 900.

[0072] As shown in Figure 10, the computer 900 comprises, as its main components, a bus 910, a processor 912, memory 914, an input device 916, an output device 917, a display device 918, a storage device 920, a communication interface unit 922, an external device interface unit 924, an I / O device interface unit 926, and a media input / output unit 928. Note that the above components may be omitted as appropriate depending on the intended use of the computer 900.

[0073] The processor 912 consists of one or more arithmetic processing units (CPU (Central Processing Unit), MPU (Micro-processing unit), DSP (digital signal processor), GPU (Graphics Processing Unit), GPGPU (General-purpose computing on graphics processing unit), etc.) and operates as a control unit that oversees the entire computer 900. The memory 914 stores various data and programs 930 and consists of volatile memory (DRAM, SRAM, etc.) that functions as main memory, and non-volatile memory (ROM), flash memory, etc.

[0074] The input device 916 consists of, for example, a keyboard, mouse, numeric keypad, or electronic pen, and functions as an input unit. The output device 917 consists of, for example, a sound (voice) output device or a vibration device, and functions as an output unit. The display device 918 consists of, for example, a liquid crystal display, an organic EL display, electronic paper, or a projector, and functions as an output unit. The input device 916 and the display device 918 may be configured as an integrated unit, such as a touch panel display. The storage device 920 consists of, for example, an HDD or SSD (Solid State Drive), and functions as a storage unit. The storage device 920 stores various data necessary for the execution of the operating system and program 930.

[0075] The communication I / F unit 922 is connected by wire or wireless to a network 940 such as the Internet or an intranet (which may be the same as network 8 in Figure 1) and functions as a communication unit that sends and receives data with other computers according to a predetermined communication standard. The external device I / F unit 924 is connected by wire or wireless to external devices 950 such as cameras, printers, scanners, and reader / writers and functions as a communication unit that sends and receives data with external devices 950 according to a predetermined communication standard. The I / O device I / F unit 926 is connected to I / O devices 960 such as various sensors and actuators and functions as a communication unit that sends and receives various signals and data with the I / O devices 960, for example, detection signals from sensors and control signals to actuators. The media input / output unit 928 consists of, for example, a drive device such as a DVD (Digital Versatile Disc) drive or a CD (Compact Disc) drive, a memory card slot, and a USB connector, and reads and writes data to media (non-temporary storage media) 970 such as DVDs, CDs, memory cards, and USB memory.

[0076] In the computer 900 having the above configuration, the processor 912 calls and executes the program 930 stored in the storage device 920 in the memory 914, and controls various parts of the computer 900 via the bus 910. The program 930 may also be stored in the memory 914 instead of the storage device 920. The program 930 may be recorded on the media 970 in an installable or executable file format and provided to the computer 900 via the media input / output unit 928. The program 930 may also be provided to the computer 900 by downloading it via the network 940 through the communication interface unit 922. Furthermore, the computer 900 may implement the various functions realized by the processor 912 executing the program 930 using hardware such as an FPGA or ASIC.

[0077] Computer 900 is an electronic device of any form, consisting of, for example, a stationary computer or a portable computer. Computer 900 may be a client computer, a server computer, a cloud computer, or a virtual machine. Computer 900 may also be applied to devices other than those specified in devices 2 to 7.

[0078] (Machine learning methods) Figure 11 is a flowchart illustrating an example of a machine learning method using the machine learning device 6. The following description assumes the generation of a learning model 14 using multiple sets of training data 13, but this method is applicable when creating the first and second learning models 14A and 14B, respectively, using the first and second sets of training data 13A and 13B.

[0079] First, in step S100, the training data acquisition unit 600 acquires a desired number of training data 13 as preparation for starting machine learning, and stores the acquired training data 13 in the training data storage unit 62. The number of training data 13 prepared here should be set considering the inference accuracy required for the final learning model 14.

[0080] Next, in step S110, the machine learning unit 601 prepares a pre-training model 14 in order to start machine learning. The pre-training model 14 prepared here is, for example, a neural network model, in which the weights of each synapse are set to initial values.

[0081] Next, in step S120, the machine learning unit 601 randomly selects, for example, one set of training data 13 from the multiple sets of training data 13 stored in the training data storage unit 62.

[0082] Next, in step S130, the machine learning unit 601 inputs the driving information (input data) contained in a set of training data 13 to the input layer 140 of the prepared pre-training (or training) learning model 14. As a result, output data is output as an inference result from the output layer 142 of the learning model 14, but this output data is generated by the pre-training (or training) learning model 14. Therefore, in the pre-training (or training) state, the output data output as an inference result shows information different from the correct labels contained in the training data 13.

[0083] Next, in step S140, the machine learning unit 601 compares the correct labels included in the set of training data 13 acquired in step S120 with the output data output from the output layer as an inference result in step S130, and performs machine learning by adjusting the weight of each synapse (backpropagation). In this way, the machine learning unit 601 trains the learning model 14 on the relationship between the input data and the output data.

[0084] Next, in step S150, the machine learning unit 601 determines whether predetermined learning termination conditions have been met, for example, based on the evaluation value of the error function, which is based on the correct labels included in the training data 13 and the output data output as an inference result, or on the remaining number of untrained training data 13 stored in the training data storage unit 62.

[0085] In step S150, if the machine learning unit 601 determines that the learning termination condition has not been met and that machine learning should continue (No in step S150), it returns to step S120 and repeats steps S120 to S140 multiple times using the untrained training data 13 on the learning model 14 that is currently being trained. On the other hand, in step S150, if the machine learning unit 601 determines that the learning termination condition has been met and that machine learning should be terminated (Yes in step S150), it proceeds to step S160.

[0086] Then, in step S160, the machine learning unit 601 stores the trained model 14 (set weight parameter group) generated by adjusting the weights associated with each synapse in the trained model storage unit 63, and the series of machine learning methods shown in Figure 11 is completed. In the machine learning method, step S100 corresponds to the training data storage step, steps S110 to S150 are the machine learning steps, and step S160 is the trained model storage step.

[0087] As described above, the machine learning apparatus 6 and machine learning method according to this embodiment provide a first learning model 14A capable of predicting (inferring) corrosion environment information of a structure 11 from the operating information of a plant 10 in which the structure 11 is installed, and a second learning model 14B capable of predicting (inferring) corrosion rate information for the structure 11 from the operating information of the plant 10 in which the structure 11 is installed and the corrosion environment information of the structure 11.

[0088] (Method for predicting corrosion of plant structures) Figure 12 is a flowchart illustrating an example of a plant structure corrosion prediction method using the plant structure corrosion prediction device 7. Below, we will describe an example of the operation of the plant structure corrosion prediction device 7 when an operator uses the operator terminal device 3 to predict corrosion rate information and structural life information for the target structure 11.

[0089] First, in step S200, when the operator terminal device 3 receives input from the operator on the display screen, for example, specifying the structure 11 to be predicted (for example, by structure ID and inspection point ID) and specifying the prediction time (for example, by current time, past time, and future time), these specified prediction conditions are transmitted to the plant structure corrosion prediction device 7. Note that multiple prediction times may be specified as prediction times; for example, multiple future time points such as 1 year, 2 years, and 3 years from now may be specified. Furthermore, if future time points are specified, multiple operation plans with different operating parameters may be specified, for example, an operation plan for a severe case and an operation plan for a mild case.

[0090] Next, in step S210, when the information acquisition unit 700 of the plant structure corrosion prediction device 7 receives the prediction conditions, it acquires the information necessary for predicting corrosion rate information and structure life information according to those prediction conditions. For example, if a prediction for the present time is specified, the information acquisition unit 700 receives sensor signals from the instrument 12A installed on the structure 11 to be predicted; if a prediction for a past time is specified, it refers to the operation database 50; and if a prediction for a future time is specified, it refers to the operation plan of the plant 10 to acquire the operation information for the prediction.

[0091] Next, in step S220, the first prediction processing unit 701 inputs the operational information of the target structure 11 acquired by the information acquisition unit 700 as input data to the first learning model 14A, thereby predicting the corrosion environment information for the operational information of the target structure. As a result, even if the corrosion environment information is missing at the time of prediction, since the prediction timing is not the timing of acquisition of the corrosion environment information, it will be supplemented from the operational information of the target structure.

[0092] Next, in step S230, the second prediction processing unit 702 inputs the operating information of the target structure 11 acquired by the information acquisition unit 700 and the corrosion environment information of the target structure 11 predicted by the first prediction processing unit 701 into the second learning model 14B, thereby predicting corrosion rate information for the operating information and corrosion environment information of the target structure. As a result, even if corrosion environment information is missing at the time of prediction because the prediction timing is not the timing of acquisition of corrosion environment information, corrosion rate information can be predicted from the operating information of the target structure and corrosion environment information supplemented from the operating information of the target structure.

[0093] Next, in step S240, the third prediction processing unit 703 predicts structural prediction information representing the remaining life or wall thickness of the structure 11 to be predicted, based on the corrosion rate information of the predicted target predicted by the second prediction processing unit 702.

[0094] Next, in step S250, the output processing unit 704 transmits screen information to the operator terminal device 3 as output processing, which includes corrosion environment information predicted by the first prediction processing unit 701, corrosion rate information predicted by the second prediction processing unit 702, and structural lifespan information predicted by the third prediction processing unit 703.

[0095] Then, in step S260, the worker terminal device 3 displays a prediction result screen for the prediction conditions based on the screen information, presenting the worker with information on the corrosion environment, corrosion rate, and structural lifespan of the target structure 11.

[0096] Figure 13 shows an example of a prediction result display screen 73 that displays the prediction results from the plant structure corrosion prediction device 7. The prediction result display screen 73 includes a prediction condition display area 730 that displays prediction conditions, an operation information display area 731 that displays operation information, a corrosion environment information display area 732 that displays corrosion environment information, a corrosion rate information display area 733 that displays corrosion rate information, a structure lifespan information display area 734 that displays structure lifespan information, and a prediction execution button 735 that instructs the plant structure corrosion prediction device 7 to execute the prediction process.

[0097] In the example shown in Figure 13, the predicted results for the corrosion environment information, corrosion rate information, and structural lifespan information (wall thickness) of the target structure 10 (specified as structure ID=P1001, inspection point ID=T001) at future points in time (1 year, 2 years, 3 years, and 4 years) are displayed in the corrosion environment information display area 732, the corrosion rate information display area 733, and the structural lifespan information display area 734, respectively, when the plant 10 is operated according to the operating information specified in mild case operation plan 1 and the operating information specified in severe case operation plan 2.

[0098] The corrosion environment information display area 732, the corrosion rate information display area 733, and the structural lifespan information display area 734 each display alarm judgment values, and if any of these values ​​exceed or fall below the alarm judgment value, alarm information indicating that an alarm has occurred is displayed. In the structural lifespan information display area 734 shown in Figure 13, alarm information 736 is displayed indicating that the wall thickness will fall below the alarm judgment value three years from the present (11th year from the start of plant operation).

[0099] The prediction condition display area 730 displays the prediction conditions and is configured to allow the operator to set the prediction conditions. When the prediction execution button 735 is pressed, step S200 is executed by the operator terminal device 3 according to the prediction conditions set in the prediction condition display area 730 at that time. Then, steps S220 to S250 are executed by the plant structure corrosion prediction device 7, and step S260 is executed by the operator terminal device 3, updating the prediction result display screen 73.

[0100] In the above-described method for predicting corrosion of plant structures, step S210 corresponds to the information acquisition step, step S220 to the first prediction processing step, step S230 to the second prediction processing step, step S240 to the third prediction processing step, and step S250 to the output processing step.

[0101] As described above, according to the plant structure corrosion prediction device 7 and plant structure corrosion prediction method of this embodiment, the first prediction processing unit 701 inputs operating information into the first learning model 14A to predict (supplement) corrosion environment information that contributes to improving the prediction accuracy of corrosion rate information, and the second prediction processing unit 702 inputs the operating information and the corrosion environment information predicted (supplemented) from the operating information into the second learning model 14B to predict corrosion rate information. As a result, by using the corrosion environment information predicted (supplemented) from the operating information, it is possible to prepare input data (operating information and corrosion environment information) for the second learning model 14B even when the analysis of corrosion environment parameters by the fluid analyzer 4A is not being performed, and corrosion rate information can be predicted with higher accuracy compared to when corrosion rate information is predicted from operating information alone. Therefore, the prediction accuracy when predicting the rate of internal corrosion of the structure 11 can be easily improved. Furthermore, since corrosion rate information is predicted by considering operating information and corrosion environment information predicted (supplemented) from that operating information, it is possible to predict internal corrosion in response to changes in operating information, and it can also be used to formulate maintenance plans according to the progress of internal corrosion.

[0102] (Other embodiments) The present invention is not limited to the embodiments described above, and can be implemented with various modifications without departing from the spirit of the invention. All such modifications are included in the technical concept of the present invention.

[0103] In the above embodiment, the plant operation management device 2, data management device 5, machine learning device 6, and plant structure corrosion prediction device 7 were described as being composed of separate devices. However, these four devices may be composed of a single device, or any two or three of these four devices may be composed of a single device. Furthermore, at least one of the machine learning device 6 and the plant structure corrosion prediction device 7 may be incorporated into the plant operation management device 2, the worker terminal device 3, or the data management device 5.

[0104] In the above embodiment, a case in which a neural network is used as the learning model for realizing machine learning by the machine learning unit 601 was described, but other machine learning models may also be used. Other machine learning models include, for example, tree-type models such as decision trees and regression trees, ensemble learning such as bagging, boosting, and XGBoost, recurrent neural networks, convolutional neural networks, and neural network-type models such as LSTM. (including planning), hierarchical clustering, non-hierarchical clustering, k-nearest neighbors, k Examples include clustering methods such as the mean method, multivariate analyses such as principal component analysis, factor analysis, and logistic regression, and support vector machines.

[0105] In the above embodiment, we have described a case in which the plant structure corrosion prediction device 7 performs a prediction process to predict corrosion environment information, corrosion rate information, and structure life information when it receives prediction conditions from the administrator terminal device 4. However, even if the plant structure corrosion prediction device 7 does not receive prediction conditions from the administrator terminal device 4, it may still perform the above prediction process and present the prediction results to the operator, for example, when predetermined prediction execution conditions are met. The prediction execution conditions may be specified, for example, as a specific day of the week (every Monday) or a specific date (the last day of each month).

[0106] (Machine learning program and plant structure corrosion prediction program) The present invention can also be provided in the form of a program (machine learning program) that causes the computer 900 to function as each part of the machine learning device 6, or a program (machine learning program) that causes the computer 900 to execute each process of the machine learning method. Furthermore, the present invention can also be provided in the form of a program (plant structure corrosion prediction program) that causes the computer 900 to function as each part of the plant structure corrosion prediction device 7, or a program (plant structure corrosion prediction program) that causes the computer 900 to execute each process of the plant structure corrosion prediction method according to the above embodiment.

[0107] (Information prediction device, information prediction method, and information prediction program) In the above embodiment, the first prediction processing unit 701 predicts (supplements) corrosion environment information that contributes to improving the accuracy of corrosion rate information prediction by inputting operating information into the first learning model 14A, and the second prediction processing unit 702 predicts corrosion rate information by inputting the operating information and the predicted (supplemented) corrosion environment information into the second learning model 14B. As described above, the method in which the first and second prediction processing units 701 and 702 perform a two-stage prediction process using the first and second learning models 14A and 14B (see Figures 8 and 9) is not limited to the form of the plant structure corrosion prediction device 7 (plant structure corrosion prediction method or information processing program) according to the above embodiment, but can also be provided in the form of an information prediction device (information prediction method and information prediction program) used to predict arbitrary information.

[0108] In this case, the information prediction device (information prediction inference method or information prediction program) comprises: an information acquisition unit (information acquisition step) that acquires first information including one or more first parameters; a first prediction processing unit (first prediction processing step) that inputs the first information of the prediction target acquired by the information acquisition unit into a first learning model to predict second information relating to the first information of the prediction target; and a second prediction processing unit (second prediction processing step) that inputs the first information of the prediction target acquired by the information acquisition unit and the second information of the prediction target predicted by the first prediction processing unit into a second learning model to predict third information relating to the first and second information of the prediction target. Here, the first learning model is trained by machine learning to understand the relationship between the first information of the learning target and the second information of the learning target, which is different from the first information of the learning target and includes one or more second parameters. The second learning model is trained by machine learning to understand the relationship between the first and second information of the learning target and the third information of the learning target, which is different from the first and second information of the learning target and includes one or more third parameters.

[0109] Furthermore, as an information prediction device (information prediction inference method or information prediction program) corresponding to the second prediction method 141 shown in Figure 9, if the second information includes multiple second parameters, the information acquisition unit acquires some of the multiple second parameters included in the second information of the prediction target, along with the first information of the prediction target. The first prediction processing unit inputs the first information of the prediction target acquired by the information acquisition unit into the first learning model, thereby predicting the second parameters other than some of the multiple second parameters as second information for the first information of the prediction target. The second prediction processing unit inputs the first information of the prediction target acquired by the information acquisition unit, the second information of the prediction target including some of the second parameters acquired by the information acquisition unit and the other second parameters predicted by the first prediction processing unit into the second learning model, thereby predicting third information for the first and second information of the prediction target.

[0110] Furthermore, as an information prediction device (information prediction inference method or information prediction program) corresponding to the third prediction method 142 shown in Figure 9, if the first information includes first parameters at multiple time points and the second information includes second parameters at multiple time points, the information acquisition unit acquires the second parameters at some of the multiple time points included in the second information of the prediction target, along with the first information of the prediction target. The first prediction processing unit inputs the first information of the prediction target, which includes the first parameters at the other time points acquired by the information acquisition unit, into the first learning model, thereby predicting the second parameters at the other time points as second information for the first information of the prediction target. The second prediction processing unit inputs the first information of the prediction target acquired by the information acquisition unit, the second information of the prediction target, which includes the second parameters at some time points acquired by the information acquisition unit and the second parameters at the other time points predicted by the first prediction processing unit, into the second learning model, thereby predicting third information for the first and second information of the prediction target.

[0111] As described above, with the information prediction device (information prediction inference method or information prediction program), the first prediction processing unit inputs the first information into the first learning model to predict (complete) the second information which contributes to improving the prediction accuracy of the third information, and the second prediction processing unit inputs the first information and the predicted (completed) second information into the second learning model to predict the third information. Therefore, the prediction accuracy when predicting the third information from the first information can be easily improved.

[0112] (Inference device, inference method, and inference program) The present invention can be provided not only in the form of the plant structure corrosion prediction device 7 (plant structure corrosion prediction method or information processing program) according to the above embodiment, but also in the form of an inference device (inference method or inference program) used to infer corrosion environment information. In that case, the inference device (inference method or inference program) may include a memory and a processor, the processor of which may execute a series of processes. The series of processes includes an information acquisition process (information acquisition step) for acquiring operating information including one or more operating parameters in a plant 10 in which a structure 11 through which a fluid flows is installed, and an inference process (inference step) for inferring corrosion environment information including one or more corrosion environment parameters that are factors in internal corrosion of the structure 11 installed in the plant 10 operated by the operating information, once the operating information has been acquired in the information acquisition process.

[0113] By providing the inference device (inference method or inference program) in the form of an inference device, it becomes easier to apply to various devices compared to implementing the plant structure corrosion prediction device 7. It will be obvious to those skilled in the art that when the inference device (inference method or inference program) infers corrosion environment information, it may apply the inference method performed by the generation processing unit using the first learned model that has been trained by the machine learning device and machine learning method according to the above embodiment. [Explanation of symbols]

[0114] 1...Plant management system, 2...Plant operation management system, 3…Worker terminal device, 4A…Fluid analyzer, 4B…Structural inspection device, 5...Data management device, 6...Machine learning device, 7... Plant structure corrosion prediction device, 8... Network, 10...plant, 11...structure 12...equipment, 12A...instrument, 12B...controller 13A...First training data, 13B...Second training data, 14A...First learning model, 14B...Second learning model, 50...Operational database, 51...Corrosion environment database, 52... Inspection database, 60...Control unit, 61...Communication unit, 62...Learning data storage unit, 63...Trained model memory unit, 70...Control unit, 71...Communication unit, 72...Storage unit, 73...Prediction result display screen, 600...Training data acquisition unit, 601...Machine learning unit, 700... Information acquisition unit, 701... First prediction processing unit, 702...Second prediction processing unit, 703...Third prediction processing unit, 704... Output processing unit, 900... Computer

Claims

1. A plant structure corrosion prediction device installed in a plant that predicts corrosion rate information for internal corrosion of a structure through which a predetermined fluid flows, An information acquisition unit that acquires operating information including one or more operating parameters in the plant where the structure to be predicted is installed, A first prediction processing unit predicts the corrosion environment information for the target operation information by inputting the operation information to be predicted, acquired by the information acquisition unit, into a first learning model that has been trained by machine learning to determine the relationship between the operation information to be learned and corrosion environment information including one or more corrosion environment parameters that are factors for internal corrosion of the structure installed in the plant operated with the operation information to be learned, and The system includes a second prediction processing unit that inputs the operation information to be predicted, acquired by the information acquisition unit, and the corrosion environment information to be predicted, predicted by the first prediction processing unit, into a second learning model that has been trained by machine learning on the relationship between the operation information and corrosion environment information to be learned and corrosion rate information representing the rate of internal corrosion of the structure installed in the plant operated with the operation information to be learned, thereby predicting the corrosion rate information for the operation information and corrosion environment information to be predicted. Corrosion prediction device for plant structures.

2. The system further comprises a third prediction processing unit that predicts structural prediction information representing the remaining lifespan or wall thickness of the structure to be predicted, based on the corrosion rate information of the predicted target predicted by the second prediction processing unit. The plant structure corrosion prediction device according to claim 1.

3. The aforementioned operating parameters are: This data is acquired by instruments installed in the aforementioned plant. The aforementioned corrosion environment parameters are: This data is obtained by a fluid analyzer using the fluid extracted from inside the aforementioned structure as a sample. A plant structure corrosion prediction device according to claim 1 or claim 2.

4. The aforementioned driving information is, This includes the aforementioned operating parameters at multiple points in time, The aforementioned corrosion environment information is, This includes multiple corrosion environment parameters at the aforementioned time points, The aforementioned information acquisition unit, Along with the operational information of the subject to prediction, the corrosion environment parameters for some of the multiple corrosion environment parameters for the subject structure at the aforementioned time points are acquired. The first prediction processing unit, By inputting the operation information of the target of prediction, including the operation parameters at the time point excluding the portion acquired by the information acquisition unit, into the first learning model, the corrosion environment parameters at the time point excluding the portion of the multiple corrosion environment parameters at the time point are predicted as corrosion environment information for the operation information of the target of prediction. The second prediction processing unit, The operation information of the target of prediction acquired by the information acquisition unit, and the corrosion environment information of the target of prediction, which includes some of the corrosion environment parameters at the time acquired by the information acquisition unit and the corrosion environment parameters at the time other than those predicted by the first prediction processing unit, are input to the second learning model to predict the corrosion rate information for the operation information and corrosion environment information of the target of prediction. A plant structure corrosion prediction device according to any one of claims 1 to 3.

5. The aforementioned corrosion environment information is, This includes multiple aforementioned corrosion environment parameters, The aforementioned information acquisition unit, Along with the operational information of the subject to prediction, some of the corrosion environment parameters among the multiple corrosion environment parameters of the structure of the subject to prediction are acquired. The first prediction processing unit, The operating information of the target of prediction acquired by the information acquisition unit is input to the first learning model, thereby predicting the corrosion environment parameters other than the partial set of corrosion environment parameters as the corrosion environment information for the operating information of the target of prediction. The second prediction processing unit, The operation information of the target to be predicted, acquired by the information acquisition unit, and the corrosion environment information of the target to be predicted, including some of the corrosion environment parameters acquired by the information acquisition unit and the other corrosion environment parameters predicted by the first prediction processing unit, are input to the second learning model to predict the corrosion rate information for the operation information and corrosion environment information of the target to be predicted. A plant structure corrosion prediction device according to any one of claims 1 to 3.

6. An information acquisition unit that acquires first information including one or more first parameters, A first prediction processing unit inputs the first information to be predicted, acquired by the information acquisition unit, into a first learning model that has learned the relationship between the first information to be learned and the second information to be learned, which is different from the first information to be learned and includes one or more second parameters, by machine learning, thereby predicting the second information for the first information to be predicted. The system comprises a second prediction processing unit that inputs the first information of the prediction target acquired by the information acquisition unit and the second information of the prediction target predicted by the first prediction processing unit into a second learning model that has been trained by machine learning to understand the relationship between the first and second information of the learning target and a third information of the learning target that is different from the first and second information of the learning target and includes one or more third parameters, thereby predicting the third information of the prediction target for the first and second information of the prediction target. The first piece of information mentioned above is, This includes the first parameter at multiple points in time, The second information mentioned above is, The second parameter includes multiple of the aforementioned time points, The aforementioned information acquisition unit, Along with the first information of the subject to prediction, the second parameters of some of the multiple second parameters of the time points included in the second information of the subject to prediction are obtained. The first prediction processing unit, By inputting the first information of the target to be predicted, including the first parameters of the time period other than the partial obtained by the information acquisition unit, into the first learning model, the second parameters of the time period other than the partial among a plurality of second parameters of the time period are predicted as second information for the first information of the target to be predicted. The second prediction processing unit, The first information of the target to be predicted, acquired by the information acquisition unit, and the second information of the target to be predicted, which includes the second parameters of a portion of the time period acquired by the information acquisition unit and the second parameters of a portion of the time period predicted by the first prediction processing unit, are input into the second learning model to predict the third information for the first and second information of the target to be predicted. Information prediction device.

7. An information acquisition unit that acquires first information including one or more first parameters, A first prediction processing unit inputs the first information to be predicted, acquired by the information acquisition unit, into a first learning model that has learned the relationship between the first information to be learned and the second information to be learned, which is different from the first information to be learned and includes one or more second parameters, by machine learning, thereby predicting the second information for the first information to be predicted. The system comprises a second prediction processing unit that inputs the first information of the prediction target acquired by the information acquisition unit and the second information of the prediction target predicted by the first prediction processing unit into a second learning model that has been trained by machine learning to understand the relationship between the first and second information of the learning target and a third information of the learning target that is different from the first and second information of the learning target and includes one or more third parameters, thereby predicting the third information of the prediction target for the first and second information of the prediction target. The second information mentioned above is, It includes multiple of the above second parameters, The aforementioned information acquisition unit, Along with the first information of the subject to prediction, some of the second parameters included in the second information of the subject to prediction are obtained, The first prediction processing unit, The first information of the prediction target acquired by the information acquisition unit is input to the first learning model, thereby predicting the second parameters of a plurality of second parameters, excluding the portion thereof, as second information for the first information of the prediction target. The second prediction processing unit, The first information of the target to be predicted, acquired by the information acquisition unit, and the second information of the target to be predicted, which includes some of the second parameters acquired by the information acquisition unit and the remaining second parameters predicted by the first prediction processing unit, are input into the second learning model to predict the third information for the first and second information of the target to be predicted. Information prediction device.

8. A plant structure corrosion prediction method for predicting corrosion rate information of internal corrosion of a structure installed in a plant through which a predetermined fluid flows, An information acquisition step involves acquiring operational information, including one or more operational parameters, at the plant where the structure to be predicted is installed. A first prediction processing step involves inputting the operation information to be predicted, obtained in the information acquisition step, into a first learning model that has been trained by machine learning to determine the relationship between the operation information to be learned and corrosion environment information, which includes one or more corrosion environment parameters that are factors for internal corrosion of the structure installed in the plant operated with the operation information to be learned, thereby predicting the corrosion environment information for the operation information to be predicted. The second prediction processing step includes inputting the operation information of the target to be predicted, obtained in the information acquisition step, and the corrosion environment information of the target to be predicted, predicted in the first prediction processing step, into a second learning model that has been trained by machine learning on the relationship between the operation information and the corrosion environment information to be learned and corrosion rate information representing the rate of internal corrosion of the structure installed in the plant operated with the operation information to be learned, thereby predicting the corrosion rate information for the operation information and the corrosion environment information of the target to be predicted. A method for predicting corrosion in plant structures.

9. The operation information is This includes the aforementioned operating parameters at multiple points in time, The aforementioned corrosion environment information is, This includes multiple corrosion environment parameters at the aforementioned time points, The aforementioned information acquisition process is as follows: Along with the operational information of the subject to prediction, the corrosion environment parameters for some of the multiple corrosion environment parameters for the subject structure at the aforementioned time points are acquired. The first prediction processing step is: By inputting the operation information to be predicted, including the operation parameters at the time point other than the part obtained by the information acquisition step, into the first learning model, the corrosion environment parameters at the time point other than the part of the multiple corrosion environment parameters at the time point are predicted as corrosion environment information for the operation information to be predicted, The second prediction processing step is as follows: The operation information of the target to be predicted, acquired by the information acquisition step, and the corrosion environment information of the target to be predicted, including some of the corrosion environment parameters at the time obtained by the information acquisition step and the corrosion environment parameters at the time other than those predicted by the first prediction processing step, are input into the second learning model to predict the corrosion rate information for the operation information and corrosion environment information of the target to be predicted. The method for predicting corrosion of plant structures according to claim 8.

10. The corrosion environment information is, This includes multiple aforementioned corrosion environment parameters, The aforementioned information acquisition process is as follows: Along with the operational information of the subject to prediction, some of the corrosion environment parameters among the multiple corrosion environment parameters of the structure of the subject to prediction are acquired. The first prediction processing step is: By inputting the operation information of the target of prediction obtained in the information acquisition step into the first learning model, the corrosion environment parameters other than the partial set of corrosion environment parameters are predicted as corrosion environment information for the operation information of the target of prediction. The second prediction processing step is as follows: The predicted operating information obtained by the information acquisition step, and the predicted corrosion environment including some of the corrosion environment parameters obtained by the information acquisition step and the other corrosion environment parameters predicted by the first prediction processing step. By inputting the information into the second learning model, the corrosion rate information for the operation information and corrosion environment information of the target of prediction is predicted. The method for predicting corrosion of plant structures according to claim 8.

11. An information acquisition step of acquiring first information including one or more first parameters, The first information of the prediction target obtained by the information acquisition step is A first prediction processing step in which the relationship between the first information to be learned and the second information to be learned, which is different from the first information to be learned and includes one or more second parameters, is input into a first learning model that has been trained by machine learning, thereby predicting the second information for the first information to be predicted. The first information of the target to be predicted obtained by the information acquisition step and the second information of the target to be predicted predicted by the first prediction processing step are used. The system includes a second prediction processing step which predicts the third information for the first and second information to be predicted by inputting the relationship between the first and second information to be learned and the third information to be learned, which is different from the first and second information to be learned and includes one or more third parameters, into a second learning model that has been trained by machine learning, The first piece of information mentioned above is, This includes the first parameter at multiple points in time, The second information mentioned above is, The second parameter includes multiple of the aforementioned time points, The aforementioned information acquisition process is as follows: Along with the first information of the subject to prediction, the second parameters of some of the multiple second parameters of the time points included in the second information of the subject to prediction are obtained. The first prediction processing step is: By inputting the first information of the target to be predicted, including the first parameters of the time period other than the part obtained by the information acquisition step, into the first learning model, the second parameters of the time period other than the part among a plurality of second parameters of the time period are predicted as second information for the first information of the target to be predicted. The second prediction processing step is as follows: The first information of the target to be predicted, obtained by the information acquisition step, and the second information of the target to be predicted, which includes the second parameters of a portion of the time obtained by the information acquisition step and the second parameters of a portion of the time predicted by the first prediction processing step, are input into the second learning model to predict the third information for the first and second information of the target to be predicted. Information prediction methods.

12. An information acquisition step of acquiring first information including one or more first parameters, The first information of the prediction target obtained by the information acquisition step is A first prediction processing step in which the relationship between the first information to be learned and the second information to be learned, which is different from the first information to be learned and includes one or more second parameters, is input into a first learning model that has been trained by machine learning, thereby predicting the second information for the first information to be predicted. The first information of the target to be predicted obtained by the information acquisition step and the second information of the target to be predicted predicted by the first prediction processing step are used. The system includes a second prediction processing step which predicts the third information for the first and second information to be predicted by inputting the relationship between the first and second information to be learned and the third information to be learned, which is different from the first and second information to be learned and includes one or more third parameters, into a second learning model that has been trained by machine learning, The second information mentioned above is, It includes multiple of the above second parameters, The aforementioned information acquisition process is as follows: Along with the first information of the subject to prediction, some of the second parameters included in the second information of the subject to prediction are obtained, The first prediction processing step is: By inputting the first information of the target to be predicted obtained in the information acquisition step into the first learning model, the second parameters other than the partial set of second parameters are predicted as second information for the first information of the target to be predicted. The second prediction processing step is as follows: The first information of the target to be predicted, obtained by the information acquisition step, and the second information of the target to be predicted, which includes some of the second parameters obtained by the information acquisition step and the other second parameters predicted by the first prediction processing step, are input into the second learning model to predict the third information for the first and second information of the target to be predicted. Information prediction methods.

Images