Learning method, learning device, diagnosis device and diagnosis system

Abstract

To reduce deterioration in detection accuracy caused by a time-dependent change in the state of a target being monitored.SOLUTION: A learning method includes: acquiring first sound data generated in a plant in operation when an event occurs in the plant; acquiring second sound data generated in the plant in a normal state at one or both of a timing before the event and a timing after the event; acquiring first information related to the event; acquiring second information related to the normal state; and causing a learning model to perform first machine learning using the first data and the first information and second machine learning using the second data and the second information. The learning model uses, as input data, information related to sound data and uses, as output data, information related to the event or the normal state corresponding to the input data.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a learning method, a learning device, a diagnostic device, and a diagnostic system. [Background technology]

[0002] For example, Patent Document 1 discloses an abnormality monitoring device for a power plant. The abnormality monitoring device detects abnormal sounds in the monitored object by analyzing and stabilizing the measured values ​​of the sound of the monitored object detected by an acoustic sensor, and subtracting the analysis results of normal sounds measured in advance from the stabilized measurement results. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 9-166483 Summary of the Invention [Problem to be solved by the invention]

[0004] The anomaly monitoring device of Patent Document 1 may reduce the accuracy of detecting an anomaly when the normal sound changes due to a change over time in the state of the monitored object. The present disclosure provides a learning method, a learning device, a diagnostic device, and a diagnostic system that reduce the decrease in detection accuracy caused by a change over time in the state of the monitored object. [Means for solving the problem]

[0005] A learning method according to one embodiment of the present disclosure includes acquiring first data, which is data on sounds generated in a plant when an event occurs in the plant during operation, acquiring second data, which is data on sounds generated in the plant operating in a normal state at one or both of a timing before and a timing after the event, acquiring first information on the event, acquiring second information on the normal state, causing a learning model to perform a first machine learning process using the first data and the first information, and causing the learning model to perform a second machine learning process using the second data and the second information, wherein the learning model receives information on the sound data as input data and receives information on the event or the normal state corresponding to the input data as output data. [Brief description of the drawings]

[0006] [Figure 1] FIG. 1 is a diagram illustrating an example of a configuration of a diagnostic system according to an exemplary embodiment. [Diagram 2] FIG. 2 is a diagram showing an example of first sound data representing a sound signal by sound intensity and sound occurrence time. [Diagram 3] FIG. 3 is a diagram showing an example of the third sound data after processing by the diagnosis device. [Figure 4] FIG. 4 is a flowchart showing an example of the operation of the diagnostic device according to the embodiment in the learning mode. [Diagram 5] FIG. 5 is a flowchart showing an example of the operation of the diagnostic device according to the embodiment in the learning mode. [Figure 6] FIG. 6 is a flowchart showing an example of the operation of the diagnostic device according to the embodiment in the learning mode. [Figure 7] FIG. 7 is a diagram showing an example of the state of sound data when the plant is in a non-normal state. [Figure 8] FIG. 8 is a diagram showing an example of the state of sound data when sound is added by human action. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0007] In the following, exemplary embodiments of the present disclosure will be described with reference to the drawings. The embodiments described below are all comprehensive or specific examples. Among the components in the following embodiments, components that are not described in an independent claim showing a top concept will be described as optional components. Each figure in the attached drawings is a schematic diagram and is not necessarily illustrated precisely. In each figure, substantially the same components are given the same reference numerals, and duplicated descriptions may be omitted or simplified. In this specification and claims, "apparatus" may mean not only one apparatus but also a system including multiple apparatuses. "Apparatus" may also include "facility".

[0008] [Diagnostic system configuration] The configuration of a diagnostic system 100 according to an exemplary embodiment will be described with reference to FIG. 1. FIG. 1 is a diagram showing an example of the configuration of the diagnostic system 100 according to an exemplary embodiment. The diagnostic system 100 diagnoses the presence or absence of an abnormality in the plant 1 by monitoring sounds emitted by the equipment and machines constituting the plant 1 during operation of the plant 1. The diagnostic system 100 includes a sound detector 110, a diagnostic device 120, an input device 130, and an output device 140 as components. The diagnostic system 100 does not need to include all of the above-mentioned components. The diagnostic system 100 may include the plant 1. The plant 1 is an example of a monitored object, and the diagnostic device 120 is an example of a learning device.

[0009] The plant 1 may be any plant that generates sound during operation. Examples of the plant 1 may include industrial plants, chemical plants, and environmental plants. Examples of industrial plants may include food factories, pharmaceutical factories, steel plants, cement factories, power plants, etc. Examples of chemical plants may include chemical plants, petrochemical plants, oil refineries, gas refineries, etc. Examples of environmental plants may include waste treatment plants, sewage treatment plants, etc.

[0010] The present embodiment will be described by taking the plant 1 as a waste treatment plant as an example. The plant 1 includes a power generation device 11, an incinerator 12, a boiler 13, and a building 14 that includes these devices. The incinerator 12 receives and burns fuel or waste, including biomass. Exhaust gas, which is a combustion gas generated in the incinerator 12, is sent to the boiler 13. In the boiler 13, water flowing through the boiler 13 is evaporated by the heat of the exhaust gas. The steam is sent to the power generation device 11 and rotates a turbine 11a in the power generation device 11. As a result, the power generation device 11 generates electricity. The plant 1 may further include a storage tank that stores the fuel or waste, and a transfer device such as a crane that transfers the fuel or waste in the storage tank to the incinerator 12.

[0011] The sound detector 110 is disposed in a position where it can detect sounds generated during operation of equipment or machines of the plant 1, such as the power generation device 11, the incinerator 12, and the boiler 13. For example, the sound detector 110 is disposed in an operation space in the building 14 where the above-mentioned equipment or machines are disposed. For example, the sound detector 110 may be attached to the above-mentioned equipment or machines, or may be disposed in the vicinity of the above-mentioned equipment or machines.

[0012] The sound detector 110 is connected to the diagnostic device 120 via wired communication, wireless communication, or a combination of wireless communication and wired communication. Any wired communication or wireless communication may be used. The sound detector 110 can transmit and receive commands, information, data, etc. to the diagnostic device 120. For example, the sound detector 110 performs sound detection according to the received command and transmits the detection result to the diagnostic device 120.

[0013] The sound detector 110 includes a microphone and can detect ambient sound. The type of microphone is not particularly limited. Examples of the microphone type may include electrodynamic, piezoelectric, and capacitive types. The sound detector 110 may further include a converter that converts a sound signal detected by the microphone into sound data that can be transmitted to the diagnostic device 120, but this is not essential. The sound signal is a signal that represents sound and may represent sound intensity. The sound data may include a time series of the intensity of the sound signal within a measurement period. Examples of sound intensity may include sound pressure level, sound amplitude, etc. Sound intensity may also be referred to as sound power. The converter may include a circuit, a computer, or a combination of a circuit and a computer. The sound detector 110 may be an existing detector such as a sound level meter, or may be a detector designed specifically for the diagnostic system 100.

[0014] The input device 130 may be located anywhere. The input device 130 is connected to the diagnostic device 120 via wired communication, wireless communication, or a combination of wireless communication and wired communication. Any wired communication and wireless communication may be used. The input device 130 can transmit and receive commands, information, data, etc. to the diagnostic device 120. For example, the input device 130 accepts an input by a user and transmits commands, information, data, etc. corresponding to the input content to the diagnostic device 120. The input device 130 may include one or more of a button, a lever, a dial, a joystick, a mouse, a key, a touch panel, a microphone, and a motion capture device as a device to which an input is given through a user's operation. The input device 130 may be a terminal device. Examples of the terminal device may be smart devices such as smartphones and tablets, personal information terminals, other terminal devices, devices using these, and improved devices of these.

[0015] The output device 140 may be located in any location. The output device 140 is connected to the diagnostic device 120 via wired communication, wireless communication, or a combination of wireless communication and wired communication. Any wired communication and wireless communication may be used. The output device 140 may transmit and receive commands, information, data, etc. to the diagnostic device 120. For example, the output device 140 may output information, data, etc. received from the diagnostic device 120 visually, audibly, or both. The output device 140 may include one or more of a display, a projector, a light-emitting lamp, an indicator, a speaker, etc. The output device 140 may output an image via one or more of a display and a projector, and output a sound via a speaker. Examples of a display may include a liquid crystal display, an organic EL (Electro Luminescence), and an inorganic EL.

[0016] The diagnostic device 120 receives a sound signal or sound data indicating the detection result of the sound detector 110 from the sound detector 110, processes the sound signal or sound data, and diagnoses the state of the plant 1 indicated by the detection result. The diagnostic device 120 may output information and data including the diagnosis result to one or more of the output device 140 and an external device. The external device may be a device external to the diagnostic system 100.

[0017] In this embodiment, examples of diagnostic device 120 may include electronic circuit boards, electronic control units, microcomputers, personal computers, workstations, smart devices such as smartphones and tablets, and other electronic devices.

[0018] Devices performing functions, such as the diagnostic device 120 and the sound detector 110, may include circuits or processing circuits. The circuits may include a processing circuit. The processing circuit or circuits include a processor P and a memory device M, etc. In this embodiment, the diagnostic device 120 includes a processing circuit, and includes a processor P and a memory device M. The processing circuit or circuits transmit and receive commands, information, data, etc. to and from other devices. The processing circuit or circuits input signals from various devices and output control signals to controlled objects.

[0019] The memory device M may include a memory, a storage, or both a memory and a storage. Examples of the memory may include a volatile semiconductor memory such as a random access memory (RAM) and a non-volatile semiconductor memory such as a read-only memory (ROM). Examples of the storage may include a semiconductor memory such as a flash memory, a hard disk, and a solid state drive (SSD). For example, the memory device M stores a processing circuit or a program executed by the circuit, various data, and the like.

[0020] At least some of the functions of the diagnostic device 120 and the sound detector 110 may be realized by cooperation between the processor P and the storage device M, etc. The processor P and memories including RAM and ROM may function as components of a computer system. For example, the computer system may realize the above functions by the processor P using the RAM as a work area to execute a program recorded in the ROM.

[0021] Some or all of the functions of the diagnostic device 120 and the sound detector 110 may be realized by a computer system, by a dedicated hardware circuit such as an electronic circuit or an integrated circuit, or by a combination of a computer system and a hardware circuit. Both the diagnostic device 120 and the sound detector 110 may execute processing under centralized control by a single processing circuit or circuit, or may execute processing under distributed control by cooperation of multiple processing circuits or circuits.

[0022] For example, the processor P may include, but is not limited to, one or more of a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), a microprocessor, a processor core, a multiprocessor, an ASIC (Application-Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), a reconfigurable processor, and the like. The processor P may realize processing by a logic circuit or a dedicated circuit, which is a hardware circuit formed in an integrated circuit such as an IC (Integrated Circuit) chip and an LSI (Large Scale Integration). The multiple functions of the diagnostic device 120 and the sound detector 110 may be realized by an integrated circuit individually integrated into one chip. The multiple functions of the diagnostic device 120 and the sound detector 110 may be realized by an integrated circuit integrated into one chip so as to include some or all of the functions.

[0023] The diagnostic device 120 can operate in a diagnostic mode and a learning mode. The diagnostic device 120 operates in the diagnostic mode or the learning mode according to an instruction received from the input device 130. The diagnostic mode is a mode in which a learning model LM is used to diagnose whether a sound generated by equipment or machinery of the plant 1 is a sound generated during normal operation, i.e., in a normal state. The learning mode is a mode in which the learning model LM is made to perform machine learning of event sounds generated in the equipment or machinery of the plant 1. In the learning mode, the learning model LM is constructed by machine learning. The learning model LM is included in the diagnostic device 120 and stored in a memory device M of the diagnostic device 120.

[0024] An event sound is a sound that is not generated in a normal state by the equipment or machinery of the plant 1. The event sound includes a sound that is generated in a non-normal state by the equipment or machinery of the plant 1, and a sound that is caused by human action.

[0025] Examples of non-normal states may include a state in which the equipment or machine of plant 1 has broken down, a state in which some of the equipment or machine of plant 1 is operating while other parts are stopped, a state in which components such as parts of the equipment or machine of plant 1 are due for replacement or have reached the end of their useful life, an operating state for starting or stopping the equipment or machine of plant 1, an operating state for inspecting the equipment or machine of plant 1, an operating state during calibration of the equipment or machine of plant 1, and an operating state during an emergency shutdown of the equipment or machine of plant 1.

[0026] Examples of human actions may include an emergency stop of equipment or machinery of plant 1 by a user, stopping a part of an equipment or machinery of plant 1 during operation by a user, application of impact or vibration to the equipment or machinery of plant 1 by a user, and application of sound to the equipment or machinery of plant 1 by a user. Examples of application of impact or vibration may include striking with a tool such as a hand-held hammer, contact with a vibrating tool or a machine or device that generates vibration, etc. Examples of application of sound may include generation of sound by a machine or device that generates sound other than the equipment or machinery of plant 1, etc.

[0027] The generation of event sounds caused by human action is performed repeatedly on a regular or irregular basis by users such as managers, operators, or inspectors of the plant 1. This is to obtain information on event sounds corresponding to changes over time in the equipment or machinery of the plant 1. Examples of regular time periods include monthly, several months, half-yearly, yearly, several years, five years, ten years, etc.

[0028] In this embodiment, the learning model LM is a program that functions by being executed by the processor P, but is not limited to this. For example, the learning model LM may be a storage area of ​​the storage device M that stores and accumulates information related to event sounds. In this embodiment, the information related to event sounds is accumulated in the storage device M as learning data LD that the learning model LM uses in machine learning.

[0029] In this embodiment, the learning model LM is a classification model, and performs supervised machine learning. The learning model LM performs machine learning using the feature amount of the sound and a classification group into which the sound is classified. The classification group is teacher data. The classification group includes a group including an event sound and a group not including an event sound. The group including the event sound can be classified into a plurality of groups. Examples of the group including the event sound can include a group in which the equipment or machine of the plant 1 is in an abnormal state, a group in which the equipment or machine of the plant 1 is in a faulty state, a group in which components of the equipment or machine of the plant 1 are due for replacement, and a group in which the above-mentioned components are at the end of their useful life.

[0030] The learning model LM takes the sound features as input data and the classification group of the sound as output data. The learning model LM determines a classification group corresponding to the input sound features based on the features of the sound features that have already been learned. In machine learning, the learning model LM is adjusted and updated so as to reduce the error between the classification group into which the learning model LM outputs the sound features as input data and the classification group of the teacher data corresponding to the sound features. In other words, the learning model LM is optimized. The updated learning model LM improves the accuracy of the output data. The classification group is an example of the first information and the second information, and an example of the first group and the second group.

[0031] The learning model LM is not limited to a classification model, and may be another learning model. For example, the learning model LM may be a clustering model. The clustering model performs unsupervised machine learning. The clustering model groups sounds based on the similarity between the features of the sounds.

[0032] The learning model LM may be a learning model using a neural network. A learning model using a neural network can perform supervised machine learning. A learning model using a neural network can set the feature amount of a sound as input data, and set various data such as the probability that the sound corresponds to a candidate classification group, the probability that the sound corresponds to a candidate sound content, or the probability that the sound corresponds to a candidate sound as output data. For example, the content of the sound may be the state of the equipment or device of the plant 1 indicated by the sound. Even in a learning model using a neural network, the learning model is adjusted and optimized so as to reduce the error between the output data of the learning model and the teacher data.

[0033] The sound feature quantity is generated by the diagnostic device 120 processing the sound signal or sound data. A method for generating the sound feature quantity from the sound signal by the diagnostic device 120 will be described below. The diagnostic device 120 converts a sound signal such as the sound signal detected by the sound detector 110 into first sound data, which is two-dimensional sound data represented by sound intensity and sound generation time. The first sound data includes the intensities and generation times of a plurality of sound signals detected over time by the sound detector 110 or the like, which are associated with each other. The sound detector 110 may convert the sound signal into the first sound data. An example of the first sound data is shown in FIG. 2. FIG. 2 is a diagram showing an example of the first sound data representing a sound signal with sound intensity P and sound generation time T. The first sound data is an example of the first data and the second data.

[0034] Furthermore, the diagnostic device 120 converts the first sound data to generate second sound data, which is three-dimensional sound data that represents a sound signal in terms of frequency, intensity, and occurrence time. The second sound data includes a set of sound intensity and occurrence time that forms a sound of each frequency, in association with the frequency. For example, the diagnostic device 120 uses a transform method such as a wavelet transform and a short-time Fourier transform in the conversion process from the first sound data to the second sound data. As the wavelet transform, a continuous wavelet transform (CWT) may be used. In the second sound data, the frequency, intensity, and occurrence time of the sound signal are represented, for example, by a three-dimensional coordinate space in which the X-axis is time T, the Y-axis is frequency F, and the Z-axis is intensity P.

[0035] Furthermore, the diagnostic device 120 divides the second sound data into a plurality of frequency bands, and two-dimensionalizes the second sound data of each frequency band to generate third sound data. Although not limited, in this embodiment, the diagnostic device 120 divides the frequency F of the Y axis into a plurality of frequency bands. Furthermore, for each of the plurality of frequency bands, the diagnostic device 120 two-dimensionalizes data included in the same frequency band onto the XZ plane. That is, the diagnostic device 120 projects data included in the same frequency band onto the XZ plane, and determines the two-dimensional data on the projection plane as the third sound data of the frequency band. In this embodiment, as shown in FIG. 3, the diagnostic device 120 converts the second sound data into third sound data of three frequency bands FB1, FB2, and FB3. The frequency bands of the frequency bands FB1, FB2, and FB3 increase in this order. FIG. 3 is a diagram showing an example of third sound data after processing by the diagnostic device 120. The three frequency bands FB1, FB2 and FB3 include a high frequency band, a medium frequency band and a low frequency band. The method of dividing the frequency bands is arbitrary, and for example, the frequency band including the entire second sound data may be divided into a plurality of frequency bands. The third sound data is an example of the first conversion data and the second conversion data.

[0036] The diagnostic device 120 detects sound features using the first sound data and the third sound data of all frequency bands. Although not limited thereto, in this embodiment, the diagnostic device 120 detects a time waveform feature and a frequency characteristic feature as sound features. The time waveform feature represents the feature of the first sound data, and the frequency characteristic feature represents the feature of the third sound data of each frequency band.

[0037] The time waveform feature is a feature of the sound intensity included in the frame FLA in the first sound data. The frame FLA is a predetermined time period. Although not limited, in this embodiment, the time waveform feature is a statistical value of the sound intensity included in the frame FLA. The statistical value may represent the sound intensity characteristics of the first sound data included in the frame FLA. Examples of the statistical value may include the average value, standard deviation, maximum value, minimum value, median value, mode value, etc. of the sound intensity. The frequency characteristic feature is a feature of the sound intensity included in the frame FLB in the third sound data of each frequency band. The frame FLB is a predetermined time period, and in this embodiment, is the same as the frame FLA, but is not limited to this. In this embodiment, the frequency characteristic feature is a statistical value of the sound intensity included in the frame FLB. The statistical value may represent the sound intensity characteristics of the third sound data included in the frame FLB. Examples of the statistical value may include the average value, standard deviation, maximum value, minimum value, median value, mode value, etc. of the sound intensity. Examples of the average used in the time waveform feature and the frequency characteristic feature may include the arithmetic mean, the power mean, the geometric mean, the harmonic mean, and the weighted mean.

[0038] The diagnostic device 120 sets a plurality of frames FLA or FLB at different times in each of the first sound data and the third sound data. In the present embodiment, the time timing of the plurality of frames FLA and the time timing of the plurality of frames FLB are the same, but are not limited to this. The frame FLA may be continuous with the frame FLA adjacent in time, or may be separated in time from the frame FLA adjacent in time. The frame FLB may be continuous with the frame FLB adjacent in time, or may be separated in time from the frame FLB adjacent in time. In two frames that are continuous in time, the start point of the subsequent frame may coincide with the end point of the preceding frame, or the start point of the subsequent frame may be located in the middle of the preceding frame. In the latter case, a part of the subsequent frame overlaps with a part of the preceding frame. This makes it possible to set many frames while increasing the time period of the frames.

[0039] As described above, the diagnosis device 120 generates, from the sound signal, a time waveform feature amount and a frequency characteristic feature amount as sound feature amounts.

[0040] The diagnostic device 120 operates in each mode as follows. In the learning mode, the diagnostic device 120 detects sound signals or sound data including an event sound and sound signals or sound data not including an event sound from sound signals or sound data, and stores a combination of these sound signals or sound data in the memory device M as learning data LD for the learning model LM. Furthermore, the diagnostic device 120 stores in the memory device M a classification group of sound signals or sound data including an event sound and a classification group of sound signals or sound data not including an event sound in association with the corresponding sound signals or sound data.

[0041] The diagnostic device 120 may receive sound signals or sound data in real time from the sound detector 110 in operation, and detect sound signals or sound data including an event sound and sound signals or sound data not including an event sound from the received sound signals or sound data. The diagnostic device 120 may receive classification group information related to the above sound signals or sound data from the input device 130. The diagnostic device 120 may receive sound signals or sound data from an external storage device that stores sound signals or sound data previously detected by the sound detector 110, and detect sound signals or sound data including an event sound and sound signals or sound data not including an event sound from the sound signals or sound data. The diagnostic device 120 may receive classification group information related to the above sound signals or sound data from the external storage device or from the input device 130.

[0042] In the case of an event sound caused by human action, the diagnostic device 120 may determine that a sound signal or sound data detected at one or both of a timing before and a timing after the event sound is a sound signal or sound data that does not include the event sound. In the case of an event sound in a non-normal state of equipment or machinery of the plant 1, the diagnostic device 120 may determine that a sound signal or sound data detected at one or both of a timing before the non-normal state occurs and a timing after the non-normal state is resolved is a sound signal or sound data that does not include the event sound.

[0043] When the start timing and end timing of the event sound are not preset in the sound signal or sound data, the diagnostic device 120 may determine the start timing and end timing of the event sound according to a command input by the user to the input device 130. For example, the diagnostic device 120 may process the sound signal or sound data to generate first sound data or third sound data, and display the sound data in a chart on the output device 140. The user can determine the start timing and end timing of the event sound based on the chart of the sound data displayed on the output device 140.

[0044] Furthermore, the diagnostic device 120 processes a sound signal or sound data including an event sound to generate a time waveform feature quantity and a frequency characteristic feature quantity of the sound including the event sound as sound feature quantities. The diagnostic device 120 processes a sound signal or sound data not including an event sound to generate a time waveform feature quantity and a frequency characteristic feature quantity of the sound not including the event sound as sound feature quantities. In addition to or instead of the time waveform feature quantity, the diagnostic device 120 may generate a time differential value that is a change amount per time of the time waveform feature quantity. In addition to or instead of the frequency characteristic feature quantity, the diagnostic device 120 may generate a time differential value that is a change amount per time of the frequency characteristic feature quantity.

[0045] The diagnostic device 120 may store in the memory device M one or more of a time waveform feature, a frequency characteristic feature, a derivative value of the time waveform feature, and a derivative value of the frequency characteristic feature, together with the sound signal or sound data, or instead of the sound signal or sound data.

[0046] The diagnostic device 120 inputs the feature amount of the sound including the event sound into the learning model LM, and adjusts the learning model LM so as to reduce the difference between the data output by the learning model LM and the classification group of the above-mentioned event, which is the teacher data. The diagnostic device 120 inputs the feature amount of the sound not including the event sound into the learning model LM, and adjusts the learning model LM so as to reduce the difference between the data output by the learning model LM and the classification group not corresponding to the event, which is the teacher data. The diagnostic device 120 inputs the feature amount of the sound including the event sound of one or more adjacent frames into the learning model LM, and inputs the feature amount of the sound not including the event sound of one or more adjacent frames into the learning model LM. The number of frames to be input to the learning model LM is arbitrary, but can be appropriately set according to the time period of the frame. The number of frames to be input to the learning model LM may be a quantity that includes one or more of the behavior of the sound intensity before and after the start timing of the event sound, the behavior of the sound intensity before and after the end timing of the event sound, and the behavior of the sound intensity before and after the local maximum point of the intensity of the event sound in the third sound data.

[0047] The diagnosis device 120 updates the learning model LM by storing the adjusted learning model LM in the memory device M. The learning model LM can learn features of the feature amounts of sounds that include an event sound and features of the feature amounts of sounds that do not include an event sound.

[0048] In the diagnosis mode, the diagnosis device 120 determines whether or not the sound signal or sound data of the detected sound received from the sound detector 110 corresponds to a sound generated during normal operation of the equipment or machine of the plant 1. In the following, the sound generated during normal operation of the equipment or machine of the plant 1 may be referred to as the "normal state sound of the plant 1."

[0049] The diagnostic device 120 processes the sound signal or sound data of the detected sound received from the sound detector 110, and calculates a time waveform feature and a frequency characteristic feature. The diagnostic device 120 inputs the time waveform feature and the frequency characteristic feature included in each of the frames of the input target frame count to the learning model LM. In this embodiment, the diagnostic device 120 inputs the time differential value of the time waveform feature and the frequency characteristic feature to the learning model LM. The learning model LM outputs a classification group corresponding to the sound signal or sound data of the detected sound received from the sound detector 110. The diagnostic device 120 determines whether or not the detected sound corresponds to a normal state sound of the plant 1 based on the output classification group. If not, the diagnostic device 120 may determine the state of the plant 1 corresponding to the classification group.

[0050] The diagnostic device 120 transmits the determination result to the output device 140, and causes the output device 140 to output the determination result visually, audibly, or both. The diagnostic device 120 may cause the output device 140 to output the determination result only when the sound signal or sound data of the detected sound does not correspond to the normal state sound of the plant 1. The diagnostic device 120 may cause the output device 140 to output such a determination result as an alert.

[0051] The operation of the diagnostic device 120 in the learning mode will be described with reference to Fig. 4 to Fig. 6. Fig. 4 to Fig. 6 are flowcharts showing an example of the operation of the diagnostic device 120 in the learning mode according to the embodiment. The following operation will be described as being performed when the plant 1 is in operation and the sound detector 110 is in operation.

[0052] In step S101, a user of the plant 1 inputs an execution command for the learning mode to the input device 130 while visually checking the output device 140. The input device 130 transmits the execution command, and the diagnosis device 120 receives the execution command.

[0053] In step S102, diagnostic device 120 activates a learn mode.

[0054] In step S103, the user inputs a method for acquiring an event sound to be used for the learning data to the input device 130. The input device 130 transmits a command for the acquisition method, and the diagnosis device 120 receives the command for the acquisition method.

[0055] If the method of acquiring the event sound is a method of acquiring the event sound using the sound detector 110 (Yes in step S104), the diagnostic device 120 proceeds to step S105. If the method of acquiring the event sound is a method of acquiring an already stored event sound (No in step S104), the diagnostic device 120 proceeds to step S106.

[0056] If the event sound is a sound of equipment or machinery in the plant 1 that is not in a normal state (Yes in step S105), in step S107, the user inputs, to the input device 130, a command to acquire the sound and information on the classification group that corresponds to the non-normal state of the plant 1. The input device 130 transmits the command and information, and the diagnosis device 120 receives the command and information.

[0057] Next, in step S108, the diagnostic device 120 associates the sound signal data received from the sound detector 110 with the classification group information and stores it in the memory device M. After storing the sound signal data for a predetermined time period in the memory device M, the diagnostic device 120 stops acquiring the sound signal data.

[0058] Next, in step S109, when the abnormal state is resolved and the plant 1 returns to a normal state, the user inputs a command to acquire sound and information about the classification group corresponding to the normal state of the plant 1 to the input device 130, and the input device 130 transmits the command and information to the diagnostic device 120.

[0059] Next, in step S110, the diagnostic device 120 stores the sound signal data received from the sound detector 110 in the storage device M in association with the classification group information. When the diagnostic device 120 stores the sound signal data for a predetermined time period in the storage device M, it stops acquiring the sound signal data. For example, in FIG. 7, the state of the sound data in the case of a non-normal state is illustrated using the third sound data of three frequency bands FB1, FB2, and FB3. FIG. 7 is a diagram showing an example of the state of the sound data when the plant 1 is in a non-normal state. In FIG. 7, the sound intensity is shown in absolute value, and further shown by a waveform line connecting the points of the sound signal. The solid line NL indicates the line of the sound signal in the normal state, and the dashed line IL indicates the line of the sound signal in the non-normal state. FIG. 7 shows the state of the sound data before and after the timing TR at which the non-normal state is resolved. After step S110, the diagnostic device 120 proceeds to step S118.

[0060] By steps S107 to S110, data of sound signals in abnormal states and data of sound signals in normal states that occur at the same time are stored in the storage device M.

[0061] If the event sound is a sound caused by human action (No in step S105), in step S111, the user inputs, to the input device 130, a command to acquire the sound and information on the classification group corresponding to the normal state of the plant 1. The input device 130 transmits the command and information, and the diagnosis device 120 receives the command and information.

[0062] Next, in step S112, the diagnostic device 120 stores the sound signal data received from the sound detector 110 in the memory device M in association with the classification group information. After storing the sound signal data for a predetermined time period in the memory device M, the diagnostic device 120 stops acquiring the sound signal data. The diagnostic device 120 may notify the user of the stop of acquisition via the output device 140.

[0063] Next, in step S113, the user inputs a command to acquire a sound and information on a classification group corresponding to a sound caused by a human action to the input device 130. The input device 130 transmits the command and information, and the diagnosis device 120 receives the command and information.

[0064] Next, in step S114, the user applies a sound due to human action to the equipment or device of the plant 1, and in parallel with this, the sound detector 110 detects the sound and transmits sound signal data to the diagnostic device 120. The diagnostic device 120 stores the sound signal data received from the sound detector 110 in the memory device M in association with the classification group information. For example, in FIG. 8, the state of the sound data when sound is applied due to human action is illustrated using third sound data of three frequency bands FB1, FB2, and FB3. FIG. 8 is a diagram showing an example of the state of the sound data when sound is applied due to human action. In FIG. 8, the sound intensity is shown in absolute value, and further shown by a waveform line connecting the points of the sound signal. The solid line NL indicates the line of the sound signal in a normal state, and the dashed line IL indicates the line of the sound signal when a human action is applied, that is, the line of the sound signal due to the human action.

[0065] Next, in step S115, after adding the artificial sound, the user transmits a command to stop acquiring the sound to the diagnostic device 120 via the input device 130. The diagnostic device 120 stops acquiring the sound signal data. After step S115, the diagnostic device 120 proceeds to step S118.

[0066] In steps S111 to S115, data on the sound signal in the normal state of the plant 1 is acquired, and then data on the sound signal in a state to which a sound due to an artificial action is added is acquired. The data on the sound signal in the normal state and the data on the sound signal to which a sound due to an artificial action is added are stored in the storage device M as data generated at the same time.

[0067] The order of the set of steps S111 and S112 and the set of steps S113 to S115 may be reversed from that described above. In this way, data on the sound signal in a state where a sound due to human action is added is acquired, and then data on the sound signal in a normal state of the plant 1 is acquired. In this case as well, the data on the sound signal to which a sound due to human action is added and the data on the sound signal in a normal state are stored in the memory device M as data generated at the same time.

[0068] In step S106, the user connects an external storage device containing data of existing event sounds, which are already stored event sounds, to the diagnostic device 120 and operates the input device 130 to store a copy of the data of the existing event sounds in the storage device M.

[0069] Next, in step S116, the user operates the input device 130 to cause the diagnostic device 120 to process the existing event sound data in the memory device M and extract sound data corresponding to the event sound and sound data preceding or following the event sound.

[0070] Next, in step S117, the user operates the input device 130 to cause the diagnostic device 120 to associate data of a sound corresponding to an event sound with information of a classification group corresponding to the event sound and store the associated data in the storage device M. The user operates the input device 130 to cause the diagnostic device 120 to associate data of a sound preceding or following the event sound with information of a classification group corresponding to the associated data and to a normal state and store the associated data in the storage device M.

[0071] In steps S116 and S117, data on the event sound that occurs at the same time and data on sounds in a normal state that precede or follow the event sound are stored in the storage device M.

[0072] After step S118, the diagnostic device 120 uses the data stored in the memory device M to perform machine learning on the learning model LM. Specifically, in step S118, the diagnostic device 120 processes data on event sounds and data on sounds in a normal state that occurred at the same time, and calculates the feature amounts of each sound. At this time, the diagnostic device 120 performs a conversion process on the sound data from first sound data to third sound data, and calculates the feature amounts of the sound using the first sound data and the third sound data.

[0073] Next, in step S119, the diagnosis device 120 adjusts the learning model LM to be optimized using the sound feature amount of the event sound data and the information on the classification group corresponding to the event sound.

[0074] Next, in step S120, the diagnostic device 120 adjusts the learning model LM to be optimized using the sound feature amounts of the sound data in the normal state and the information on the classification group corresponding to the normal state.

[0075] By steps S101 to S120 as described above, the learning model LM is machine-learned using data on event sounds acquired by various methods and timings and data on sounds of the plant 1 in a normal state acquired at the same time as the event sounds. By repeatedly acquiring data on event sounds over time and repeating steps S101 to S120, even if the state of the facilities and devices of the plant 1 changes over time, the learning model LM is optimized by machine learning in response to the change over time. Therefore, even if the state of the facilities and devices of the plant 1 changes over time, the diagnosis device 120 can determine with high accuracy whether the plant 1 is in a normal state or not.

[0076] In the description of steps S101 to S120, the diagnostic device 120 uses the data of the event sound and the data of the sound in a normal state preceding or following the event sound for machine learning of the learning model LM, but is not limited to this. The diagnostic device 120 may use the data of the event sound, the data of the sound in a normal state preceding the event sound, and the data of the sound in a normal state following the event sound for machine learning of the learning model LM. This improves the detection accuracy of the learning model LM for the classification group corresponding to the normal state.

[0077] [Other embodiments] Although exemplary embodiments of the present disclosure have been described above, the present disclosure is not limited to the above-described embodiments. In other words, various modifications and improvements are possible within the scope of the present disclosure. For example, the scope of the present disclosure also includes embodiments in which various modifications have been made and embodiments constructed by combining components in different embodiments.

[0078] For example, in an embodiment, when generating data for machine learning, the start timing and end timing of the event sound in the sound data including the event sound are set by the user. That is, the event sound in the sound data including the event sound is identified by the user. However, the identification of the event sound may be performed by the diagnostic device 120. For example, a database of event sounds may be constructed, and the diagnostic device 120 may identify the event sound in the sound data including the event sound by evaluating the similarity between the sound data including the event sound and the data of the event sound included in the database.

[0079] For example, the diagnostic device 120 may determine sound data whose similarity is equal to or greater than a threshold value as event sound data, but is not limited thereto. For example, the diagnostic device 120 may evaluate sound data included in a predetermined time period in the evaluation of similarity, but is not limited thereto. For example, the above-mentioned predetermined time period may be a time period that includes the number of frames that are input to the learning model LM in machine learning. For example, the above-mentioned predetermined time period may be a period that includes the time period of the event sound and one or both of a time period preceding the event sound and a time period following the event sound.

[0080] For example, in the embodiment, the diagnostic device 120 uses data of normal state sounds acquired at one or both of the timing preceding and the timing following the occurrence of the event sound as data of normal state sounds used for machine learning of the learning model LM. That is, the diagnostic device 120 uses data of normal state sounds acquired at a timing related to the occurrence of the event sound for machine learning of the learning model LM. However, the data of normal state sounds used for machine learning of the learning model LM is not limited to the above. For example, the diagnostic device 120 may use data of normal state sounds acquired at a timing unrelated to the occurrence of the event sound for machine learning of the learning model LM. For example, the diagnostic device 120 may continuously acquire data of normal state sounds periodically or irregularly, and use the sound data to perform machine learning on the learning model LM. This improves the detection accuracy of the classification group corresponding to the normal state in the learning model LM.

[0081] For example, in the embodiment, the diagnostic device 120 causes the learning model LM to perform machine learning in the learning mode, but may also cause the learning model LM to perform machine learning in the diagnostic mode. When the output data of the learning model LM in the diagnostic mode is inappropriate and the diagnostic device 120 erroneously issues an alert, the diagnostic device 120 may correct the output data of the learning model LM, and adjust the learning model LM so as to reduce the difference between the output data of the learning model LM and the teacher data, using the corrected output result as teacher data. Information on a false alarm of an alert may be notified to the diagnostic device 120 by a user via the input device 130. When the diagnostic device 120 receives a notification of a false alarm, it may autonomously correct the classification group, which is the output data of the learning model LM. The diagnostic device 120 may receive the notification of a false alarm and information on the classification group corrected by the user using the input device 130 from the input device 130, and correct the classification group output by the learning model LM.

[0082] Examples of each aspect of the technology of the present disclosure are given as follows: A learning method according to a first aspect of the present disclosure includes acquiring first data, which is data of a sound generated in a plant when an event occurs in the plant during operation, acquiring second data, which is data of a sound generated in the plant operating in a normal state at one or both of a timing before and a timing after the event, acquiring first information on the event, acquiring second information on the normal state, causing a learning model to perform a first machine learning using the first data and the first information, and causing the learning model to perform a second machine learning using the second data and the second information, wherein the learning model uses information on the sound data as input data and information on the event or the normal state corresponding to the input data as output data.

[0083] According to the above aspect, the learning model performs machine learning using sound data when an event occurs and machine learning using sound data in a normal state before and after the event occurs. That is, the learning model performs machine learning using sound data in a normal state at the same time as the event occurs. This allows the learning model to improve the output accuracy of information about the normal state at the time of the event occurrence. The learning model can update the output accuracy of information about the event when an event occurs. Thus, the learning method reduces a decrease in the detection accuracy of the learning model due to a change over time in the state of the monitored object. The learning method may be realized, for example, by a CPU, a circuit such as an LSI, an IC card, or a standalone module.

[0084] In the above first aspect, the learning method according to the second aspect of the present disclosure may further include acquiring a plurality of first data related to a plurality of events that occur repeatedly in the plant at intervals of a period of time, acquiring a plurality of second data related to a plurality of normal states at one or both of a timing before and a timing after the plurality of events, acquiring the first information related to the plurality of events, acquiring the second information related to the plurality of normal states, and subjecting the first machine learning and the second machine learning to the learning model for each of the plurality of events.

[0085] According to the above aspect, the learning model performs machine learning using the first data and the second data acquired from the repeatedly occurring events. Each time an event occurs, the learning model can update the output accuracy of information about the normal state and the output accuracy of information about the event in accordance with the normal state of the plant at the time the event occurs. Therefore, the learning method effectively reduces a decrease in the detection accuracy of the learning model caused by a change over time in the state of the monitored object.

[0086] In the above first or second aspect, the learning method according to a third aspect of the present disclosure may further include generating the event in which a sound is generated by an artificial action.

[0087] According to the above aspect, an event can be generated at any time, and therefore the learning model can be updated at any time.

[0088] In any of the above first to third aspects, the learning method according to a fourth aspect of the present disclosure may further include converting the first data including the intensity and occurrence time of a sound signal to generate first converted data representing the sound signal in terms of intensity and occurrence time for each of a plurality of frequency bands, converting the second data including the intensity and occurrence time of the sound signal to generate second converted data representing the sound signal in terms of intensity and occurrence time for each of a plurality of frequency bands, using one or both of the first data and the first converted data in the first machine learning, and using one or both of the second data and the second converted data in the second machine learning.

[0089] According to the above aspect, the first conversion data and the second conversion data can indicate the characteristics of the sound signal for each frequency band. This makes the characteristics of the sound signal clear. Machine learning using such first conversion data and second conversion data makes it easy to build a learning model.

[0090] In the above fourth aspect, a learning method according to a fifth aspect of the present disclosure further includes generating a first feature which is a feature of sound including a feature related to a time waveform of sound obtained by processing the first data and a feature related to frequency characteristics of sound obtained by processing the first converted data, generating a second feature which is a feature of sound including a feature related to a time waveform of sound obtained by processing the second data and a feature related to frequency characteristics of sound obtained by processing the second converted data, using the first feature in the first machine learning, and using the second feature in the second machine learning, and the learning model may use the sound feature as input data.

[0091] According to the above aspect, the sound feature clarifies the characteristics of the sound signal. The sound feature can clearly indicate the difference between sound data when an event occurs and sound data in a normal state. Machine learning using such sound feature simplifies the construction of a highly accurate learning model.

[0092] In any of the above first to fifth aspects, a learning method according to a sixth aspect of the present disclosure further includes obtaining, as the first information, information of a first group into which the event is classified, and obtaining, as the second information, information of a second group into which the normal state is classified, and using, in the first machine learning, the information of the first group included in the first information, and using, in the second machine learning, the information of the second group included in the second information, and the learning model may use information of the group corresponding to the input data as output data.

[0093] According to the above embodiment, the first group of information can be associated with the contents of the event, and the second group of information can be associated with the normal state. The learning model can output, as output data, not only information indicating whether the event corresponds to the normal state, but also information indicating the contents of the event.

[0094] In any of the first to sixth aspects above, the learning method according to a seventh aspect of the present disclosure may further include using a sound signal or sound data detected by a sound detector arranged in the plant for the first data and the second data.

[0095] According to the above aspect, the learning model can perform machine learning using the sound signal or sound data detected by the sound detector. The learning model can perform machine learning using the sound signal or sound data of sounds actually occurring in the plant.

[0096] In any of the first to seventh aspects above, the learning method according to an eighth aspect of the present disclosure may further include acquiring the first data and the second data from a storage device that accumulates sound data generated in the plant in the past.

[0097] According to the above aspect, the learning model can perform machine learning using sound signals or sound data of sounds that have occurred in the plant in the past. This makes it possible to increase the amount of learning data used for the machine learning of the learning model, thereby improving the accuracy of the learning model.

[0098] A learning device according to a ninth aspect of the present disclosure includes a processing circuit including the learning model that is machine-learned by the learning method according to any one of the first to eighth aspects. According to the above aspect, the same effects as those of the learning methods according to the respective aspects of the present disclosure can be obtained.

[0099] A diagnostic device according to a tenth aspect of the present disclosure includes a processing circuit including the learning model that is machine-learned by the learning method of any one of the first to eighth aspects described above, and the processing circuit executes a diagnosis of the state of the plant using the learning model, and the processing circuit inputs information on a sound signal or sound data detected by a sound detector arranged in the plant to the learning model, diagnoses the state of the plant based on output data of the learning model, and outputs a diagnosis result.

[0100] According to the above aspect, the same effects as those of the learning method according to each aspect of the present disclosure can be obtained. Furthermore, the diagnostic device can diagnose the state of the plant based on the output result of the learning model and output the diagnosis result.

[0101] In the tenth aspect, the processing circuit of the diagnostic device according to the eleventh aspect of the present disclosure may output a command to issue an alarm when the diagnostic result indicates that the plant is not in a normal state. According to the above aspect, the diagnostic device can notify that the plant is not in a normal state.

[0102] In the eleventh aspect, the diagnostic device according to the twelfth aspect of the present disclosure further includes an input device that accepts information input by a user, and when the processing circuit receives false alarm information indicating that the alarm is a false alarm from the input device, the processing circuit may perform machine learning on the learning model using the input data of the learning model that outputs output data corresponding to the alarm and the false alarm information, thereby updating the learning model. According to the above aspect, the diagnostic device can improve the output accuracy of the learning model so as to reduce false alarms. In other words, the diagnostic device can improve the accuracy of the diagnosis result.

[0103] In any one of the above tenth to twelfth aspects, the processing circuit of the diagnostic device according to a thirteenth aspect of the present disclosure may further execute constructing the learning model by machine learning, and the processing circuit may update the learning model by subjecting the learning model to machine learning using the learning method. According to the above aspect, the diagnostic device can diagnose the state of the plant while updating the learning model. Therefore, the diagnostic device can improve the accuracy of diagnosis.

[0104] A diagnostic system according to a fourteenth aspect of the present disclosure includes the diagnostic device according to any one of the tenth to thirteenth aspects, the plant, and a sound detector disposed in the plant. According to the above aspect, the same effects as those of the diagnostic device according to each aspect of the present disclosure can be obtained.

[0105] The present disclosure may be a computer program that causes a computer to execute the learning method according to each aspect of the present disclosure. Such a computer program can achieve the same effect as the learning method according to each aspect of the present disclosure. The computer program may be, for example, a program recorded on a non-transitory computer-readable recording medium, and may be configured to be read from the recording medium using a drive device for the recording medium and installed on a computer. The computer program may be, for example, a program that can be distributed via a transmission medium such as the Internet, and may be configured to be downloaded and installed on a computer.

[0106] The functions of the elements disclosed herein can be performed using circuits or processing circuits, including general purpose processors, special purpose processors, integrated circuits, ASICs, conventional circuits, and / or combinations thereof, configured or programmed to perform the disclosed functions. A processor is considered a processing circuit or circuit because it includes transistors and other circuits. In this disclosure, a circuit, unit, or means is hardware that performs the recited functions or hardware that is programmed to perform the recited functions. The hardware may be hardware disclosed herein or other known hardware that is programmed or configured to perform the recited functions. Where the hardware is a processor, which is considered a type of circuit, the circuit, means, or unit is a combination of hardware and software, and the software is used to configure the hardware and / or the processor.

[0107] All numbers such as ordinal numbers and quantities used in this specification are exemplified to specifically explain the technology of the present disclosure, and the present disclosure is not limited to the exemplified numbers. The connection relationships between components are exemplified to specifically explain the technology of the present disclosure, and the connection relationships that realize the functions of the present disclosure are not limited to these.

[0108] Since the present disclosure may be implemented in various forms without departing from the scope of its essential characteristics, the scope of the present disclosure is defined by the appended claims rather than the description in the specification, and therefore the exemplary embodiments and modifications are illustrative and not limiting. All modifications within the scope of the claims and their equivalents are intended to be embraced by the claims. [Explanation of symbols]

[0109] 1. Plant 100 Diagnostic Systems 110 Sound Detector 120 Diagnostic Equipment 130 Input Device LM Learning Model

Claims

1. acquiring first data, the first data being data of a sound generated in a plant when an event occurs in the plant during operation; acquiring second data, the second data being data of sounds generated in the plant operating under normal conditions at one or both of a timing before and a timing after the event; obtaining first information about the event; obtaining second information regarding the normal state; performing a first machine learning process on a learning model using the first data and the first information; and performing a second machine learning process on the learning model using the second data and the second information; The learning model uses information about sound data as input data, and uses information about the event or the normal state corresponding to the input data as output data. How to learn.

2. acquiring a plurality of first data relating to a plurality of events that occur repeatedly in the plant over a period of time; acquiring a plurality of said second data relating to a plurality of normal states at one or both of timings before and after said plurality of events; obtaining the first information regarding the plurality of events; obtaining the second information regarding the plurality of normal conditions; and executing the first machine learning and the second machine learning on the learning model for each of the plurality of events. The learning method according to claim 1 .

3. and generating the event in which sound is generated by an artificial action. The learning method according to claim 1 .

4. converting the first data including the intensity and occurrence time of the sound signal to generate first converted data representing the sound signal in terms of intensity and occurrence time for each of a plurality of frequency bands; converting the second data including the intensity and occurrence time of the sound signal to generate second converted data representing the sound signal in terms of intensity and occurrence time for each of a plurality of frequency bands; In the first machine learning, using one or both of the first data and the first converted data; The second machine learning further includes using one or both of the second data and the second transformed data. A learning method according to any one of claims 1 to 3.

5. generating a first feature which is a feature of a sound, the first feature including a feature related to a time waveform of the sound obtained by processing the first data and a feature related to a frequency characteristic of the sound obtained by processing the first converted data; generating second features which are features of the sound, the second features including a feature related to a time waveform of the sound obtained by processing the second data and a feature related to a frequency characteristic of the sound obtained by processing the second converted data; In the first machine learning, using the first feature amount; The second machine learning further includes using the second feature amount, The learning model uses sound features as input data. The learning method according to claim 4.

6. acquiring, as the first information, information of a first group into which the event is classified; acquiring, as the second information, information of a second group into which the normal state is classified; In the first machine learning, information of the first group included in the first information is used; The second machine learning further includes using information of the second group included in the second information; The learning model outputs information of a group corresponding to the input data as output data. The learning method according to claim 1 .

7. The method further includes using a sound signal or sound data detected by a sound detector disposed in the plant as the first data and the second data. The learning method according to claim 1 .

8. The method further includes acquiring the first data and the second data from a storage device that accumulates sound data that has been generated in the plant in the past. The learning method according to claim 1 .

9. A processing circuit including the learning model machine-learned by the learning method according to claim 1. Learning device.

10. A processing circuit including the learning model machine-learned by the learning method according to claim 1, the processing circuitry performs a diagnosis of a condition of the plant using the learning model; The processing circuit inputs information of a sound signal or sound data detected by a sound detector disposed in the plant to the learning model, diagnoses a state of the plant based on output data of the learning model, and outputs a diagnosis result. Diagnostic equipment.

11. The processing circuit outputs a command to issue an alarm when the diagnosis result indicates that the plant is not in a normal state. The diagnostic device of claim 10.

12. further comprising an input device for accepting input of information by a user; When the processing circuit receives, from the input device, false alarm information indicating that the alarm is a false alarm, the processing circuit performs machine learning on the learning model using input data of the learning model that has outputted output data corresponding to the alarm and the false alarm information, thereby updating the learning model. The diagnostic device of claim 11.

13. The processing circuitry further performs the steps of: constructing the learning model through machine learning; The processing circuit updates the learning model by subjecting the learning model to machine learning using the learning method. The diagnostic device of claim 10.

14. A diagnostic device according to claim 10; The plant; a sound detector disposed in the plant; Diagnostic system.

Images