Device state estimation device and device state estimation method

Abstract

A device state estimation device (10) is provided with: a determination unit (110) that determines whether an actual measurement signal obtained by measuring the operating state of a device (20) is normal or abnormal; a simulation unit (130) that generates an abnormality-time generation signal by performing a simulation of an abnormality in the device (20); a synthesis unit (123) that generates an abnormality-time synthesized signal by synthesizing a normal-time measured signal, which is the measured signal determined to be normal by the determination unit (110), and the abnormality-time generated signal; a learning unit (140) that generates a learning model for estimating an abnormal part of the device (20) by performing machine learning using the abnormal-time composite signal; and an abnormal location estimation unit (150) that estimates an abnormal location of the device (20) on the basis of the learning model and an abnormal-time measured signal, which is the measured signal determined to be abnormal by the determination unit (110). A simulation unit (130) generates an abnormality-time generated signal using the abnormality-time composite signal.

Description

Technical Field

[0001] The present disclosure relates to an apparatus for estimating a device state and a method for estimating a state. Background Art

[0002] In industrial equipment, industrial machinery, industrial robots, power generation equipment, etc. used in factories and the like, multiple motors or gears are used. Needless to say, sudden device failures, as well as device abnormalities caused by aging deterioration and / or wear deterioration, can also cause a production line to stop. In addition, there is concern about a decrease in productivity or the occurrence of accidents accompanying the stoppage.

[0003] Therefore, there is a need for a method that uses a machine learning model or the like to estimate the internal state based on the characteristics of signals from sensors installed in the device, and enables efficient planned maintenance corresponding to the state of the device.

[0004] In particular, in recent years, for the purpose of fully automating maintenance work, the need for a technology that not only estimates the normality or abnormality of a device but also reaches a technology for estimating an abnormal part has been increasing.

[0005] When estimating not only the normality or abnormality of a device but also an abnormal part, the lack of data at the time of abnormality becomes a technical problem.

[0006] In the case of only estimating the normality or abnormality of a device, it is only necessary to learn the characteristics of a large amount of available normal-time data and evaluate the change from the normal state. On the other hand, in order to estimate an abnormal part, abnormal-time data corresponding to each abnormal part is required. However, generally, it is rare to obtain a sufficient amount of abnormal-time data for learning from on-site devices.

[0007] In response to this, technologies for generating data by simulation are disclosed in Patent Documents 1 to 3.

[0008] For example, in Patent Document 1, it is disclosed to learn the relationship between parameters calculated when an abnormal phenomenon is reproduced in a simulator and the abnormal phenomenon through AI (Artificial Intelligence) processing.

[0009] In Patent Document 2, it is disclosed to generate simulated data corresponding to historical data of abnormal operation patterns based on a physical model. In Patent Document 2, it is also disclosed to update the parameters of the physical model based on the difference between the simulated data and experimental data.

[0010] In Patent Document 3, it is disclosed to generate an artificial signal composed of a voltage waveform indicating an abnormality. In addition, in Patent Document 3, it is disclosed to generate a simulated signal by synthesizing an actual signal obtained from an actual machine and the artificial signal.

[0011] Prior art documents

[0012] Patent documents

[0013] Patent Document 1: Japanese Patent Application Laid-Open No. 2022-20555

[0014] Patent Document 2: Japanese Patent Application Laid-Open No. 2018-10636

[0015] Patent Document 3: Japanese Patent Application Laid-Open No. 2016-50826 Summary of the invention

[0016] Problems to be solved by the invention

[0017] The present disclosure provides an apparatus state estimation device and an apparatus state estimation method capable of accurately estimating an abnormal part of a device, etc.

[0018] Means for solving the problems

[0019] The apparatus state estimation device according to one aspect of the present disclosure includes: a determination unit that determines whether a measured signal obtained by measuring the operating condition of the device is normal or abnormal; a simulation unit that generates a generated signal during an abnormality by simulating the abnormality of the device; a synthesis unit that generates a synthesized signal during an abnormality by synthesizing the measured signal determined to be normal by the determination unit, i.e., the measured signal during normal operation, and the generated signal during an abnormality; a learning unit that performs machine learning using the synthesized signal during an abnormality to generate a learning model for estimating the abnormal part of the device; and a first estimation unit that estimates the abnormal part of the device based on the measured signal determined to be abnormal by the determination unit, i.e., the measured signal during an abnormality, and the learning model, wherein the simulation unit uses the synthesized signal during an abnormality to generate the generated signal during an abnormality.

[0020] The apparatus state estimation method according to one aspect of the present disclosure includes: a step of determining whether a measured signal obtained by measuring the operating condition of the device is normal or abnormal; a step of generating a generated signal during an abnormality by simulating the abnormality of the device; a step of generating a synthesized signal during an abnormality by synthesizing the measured signal determined to be normal, i.e., the measured signal during normal operation, and the generated signal during an abnormality; a step of performing machine learning using the synthesized signal during an abnormality to generate a learning model for estimating the abnormal part of the device; and a step of estimating the abnormal part of the device based on the measured signal determined to be abnormal, i.e., the measured signal during an abnormality, and the learning model, wherein in the simulation, the synthesized signal during an abnormality is used to generate the generated signal during an abnormality.

[0021] In addition, these general or specific manners can be implemented by a system, a method, an integrated circuit, a computer program, or a recording medium, or can be implemented by any combination of a system, a device, a method, an integrated circuit, a computer program, and a recording medium.

[0022] Advantageous Effects of the Invention

[0023] According to one aspect of the present disclosure, it is possible to accurately estimate an abnormal part of a device. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a block diagram showing the configuration of a device state estimation system according to an embodiment.

[0025] Figure 2 is a block diagram showing the configuration of a composite signal generation unit of a device state estimation device according to an embodiment.

[0026] Figure 3 is a flowchart showing an example of the operation of a device state estimation device according to an embodiment.

[0027] Figure 4 is a flowchart showing the processing related to the identification of parameters among the operations of a device state estimation device according to an embodiment.

[0028] Figure 5 is a diagram showing an example of a signal generated at the time of an abnormality generated by a simulation unit of a device state estimation device according to an embodiment.

[0029] Figure 6 is a flowchart showing the processing related to the synthesis of signals among the operations of a device state estimation device according to an embodiment.

[0030] Figure 7 is a diagram showing an example of a composite signal at the time of an abnormality generated by a composite unit of a device state estimation device according to an embodiment.

[0031] Figure 8 is a diagram showing an example of an input / output screen of physical parameters displayed by an input / output unit of a device state estimation device according to an embodiment.

[0032] Figure 9 is a diagram showing an example of a GUI object for inputting a synthesis ratio and a display screen of a synthesis result displayed by an input / output unit of a device state estimation device according to an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] (Understanding Underlying the Present Disclosure)

[0034] The inventor of the present invention has found that the following problems occur with respect to the prior art described in the "Background Art" section.

[0035] Generally speaking, it is difficult to accurately grasp the physical parameters used in simulations. Therefore, simulation errors are likely to occur in the data generated by simulations. In the technology disclosed in Patent Document 1, abnormal misdetections or misclassifications of normal and abnormal cases may occur.

[0036] In addition, in the technology disclosed in Patent Document 2, by updating the parameters of the physical model, it is possible to make the parameter values closer to the actual values. However, the characteristics of the data generated by phenomena that were not originally considered in the simulation model cannot be reproduced. It is impossible to construct a simulation model by anticipating all the physical phenomena that may occur in the device in advance. Therefore, misdetections or misclassifications may also occur in the technology disclosed in Patent Document 2.

[0037] In the technology disclosed in Patent Document 3, actual signals and artificial signals are synthesized, so it is possible to incorporate the characteristics of the data generated by phenomena that were not considered in the simulation model into the simulation data. However, the characteristics of non-linear data generated by the superposition of the characteristics of the artificial signals generated by the simulation and the characteristics inherent in the actual signals cannot be reproduced.

[0038] Thus, in the technologies disclosed in Patent Documents 1 to 3, the abnormal part of the device cannot be estimated with high accuracy.

[0039] To solve the above problems, a device state estimation apparatus according to a first aspect of the present disclosure includes: a determination unit that determines whether a measured signal obtained by measuring the operating condition of a device is normal or abnormal; a simulation unit that generates a generated signal during an abnormality by performing a simulation of the abnormality of the device; a synthesis unit that generates a synthesized signal during an abnormality by synthesizing the measured signal determined to be normal by the determination unit, i.e., the measured signal during normal operation, and the generated signal during an abnormality; a learning unit that performs machine learning using the synthesized signal during an abnormality to generate a learning model for estimating the abnormal part of the device; and a first estimation unit that estimates the abnormal part of the device based on the measured signal determined to be abnormal by the determination unit, i.e., the measured signal during an abnormality, and the learning model, wherein the simulation unit uses the synthesized signal during an abnormality to generate the generated signal during an abnormality.

[0040] Thus, since the simulation unit receives feedback from the synthesis unit (specifically, the synthesized signal during abnormality) to generate a generated signal during abnormality, it is also possible to reproduce the characteristics of non-linear data generated by the superposition of the characteristics of the artificial signal generated in the simulation and the characteristics inherent in the actual signal. Therefore, the reliability of the synthesized signal during abnormality that can be used as training data in machine learning can be improved and diversified, and thus the accuracy of the learning model can be improved. Therefore, according to the apparatus state estimation device according to the present mode, the abnormal part of the device can be estimated with high accuracy.

[0041] Further advantages and effects in one mode of the present disclosure will be clarified from the specification and the drawings. The advantages and / or effects are respectively provided by several embodiments and the features described in the specification and the drawings, but it is not necessary to provide all of them in order to obtain one or more of the same features.

[0042] In addition, for the apparatus state estimation device according to the second mode of the present disclosure, for example, in the apparatus state estimation device according to the first mode, the apparatus state estimation device includes: a second estimation unit that estimates the frequency and phase of the harmonic components included in the measured signal during normal operation as a first frequency and a first phase; and an identification unit that identifies a first physical parameter of the device based on the first frequency and the first phase, and the simulation unit performs the simulation using the first physical parameter identified by the identification unit.

[0043] Thus, by using the frequency and phase of the harmonic components, the identification of the physical parameter can be performed with high accuracy. For example, by using only the frequency and phase, an increase in the amount of calculation required for identification can be suppressed, and at the same time, the identification of the physical parameter can be performed with high accuracy.

[0044] In addition, for the apparatus state estimation device according to the third mode of the present disclosure, for example, in the apparatus state estimation device according to the second mode, the second estimation unit also estimates the frequency and phase of the harmonic components included in the generated signal during abnormality as a second frequency and a second phase, and the identification unit identifies the physical parameter such that the frequency difference, which is the difference between the first frequency and the second frequency, and the phase difference, which is the difference between the first phase and the second phase, are each less than a threshold value.

[0045] Thus, by making the frequency difference and the phase difference less than the threshold value, the reproducibility of non-linear data generated by the superposition of the characteristics of the artificial signal and the characteristics inherent in the actual signal can be improved when the synthesized signal is generated.

[0046] In addition, for the device state estimation apparatus according to the fourth aspect of the present disclosure, for example, in the device state estimation apparatus according to the second or third aspect, the device state estimation apparatus includes: an input unit that receives an input of a search range of the first physical parameter, and the identification unit identifies the first physical parameter within the search range.

[0047] Thus, since the search range can be set, it is possible to avoid searching for values that clearly cannot be obtained, thereby suppressing an increase in the amount of computation.

[0048] In addition, for the device state estimation apparatus according to the fifth aspect of the present disclosure, for example, in the device state estimation apparatus according to the fourth aspect, the input unit receives an input of a value of a second physical parameter different in type from the first physical parameter, and the second estimation unit uses the value received by the input unit to estimate the first frequency and the first phase.

[0049] Thus, it is possible to set the value of the physical parameter that does not require searching. By receiving inputs of two types of physical parameters, namely, the physical parameter that requires searching and the determined physical parameter, it is not necessary to search all physical parameters, thereby suppressing an increase in the amount of computation and the complexity of the simulation.

[0050] In addition, for the device state estimation apparatus according to the sixth aspect of the present disclosure, for example, in the device state estimation apparatus according to the fourth or fifth aspect, the device state estimation apparatus includes: an output unit that outputs the first physical parameter identified by the identification unit.

[0051] Thus, since the identification result of the physical parameter can be presented to the user, the user can evaluate the appropriateness of the identification result. For example, when the identification result is inappropriate, it is possible to re-identify the physical parameter. Therefore, since a physical parameter with a higher identification accuracy can be obtained, the accuracy of the simulation can be improved.

[0052] In addition, for the device state estimation apparatus according to the seventh aspect of the present disclosure, for example, in the device state estimation apparatus according to any one of the fourth to sixth aspects, the input unit receives an input of a setting range of an abnormal parameter of the device, and the simulation unit performs the simulation within the setting range of the abnormal parameter, thereby generating a signal generated during an abnormality.

[0053] Thus, by setting the setting range of the abnormal parameter, it is possible to avoid simulating an abnormality that clearly does not occur, thereby suppressing an increase in the amount of computation.

[0054] In addition, in the device state estimation device according to the eighth aspect of the present disclosure, for example, in the device state estimation device according to any one of the first to seventh aspects, the simulation unit performs the simulation using the abnormal synthesis signal at the first time, thereby generating the abnormal generation signal at the second time after a given period from the first time.

[0055] Thereby, it is possible to generate the abnormal generation signal while receiving the feedback of the abnormal synthesis signal for each given period. Therefore, it is possible to improve the reproducibility of non-linear data generated by superimposing the characteristics of the artificial signal and the characteristics inherent in the actual signal.

[0056] In addition, in the device state estimation device according to the ninth aspect of the present disclosure, for example, in the device state estimation device according to any one of the first to eighth aspects, the device state estimation device includes: an input unit that receives an input of the synthesis ratio of the normal measurement signal and the abnormal generation signal, and the synthesis unit synthesizes the normal measurement signal and the abnormal generation signal at the synthesis ratio, thereby generating the abnormal synthesis signal.

[0057] Thereby, it is possible to set the synthesis ratio so as to obtain a highly reliable abnormal synthesis signal. Therefore, since the reliability of the training data can be improved, the accuracy of the learning model can be improved, and the abnormal part of the device can be estimated with high accuracy.

[0058] In addition, the device state estimation device according to the tenth aspect of the present disclosure includes, for example, in the device state estimation device according to any one of the first to ninth aspects: an output unit that outputs the normal measurement signal, the abnormal generation signal, and the abnormal synthesis signal.

[0059] Thereby, by outputting each signal and presenting it to the user, the user can evaluate the appropriateness of the abnormal generation signal and the abnormal synthesis signal. For example, when the abnormal generation signal or the abnormal synthesis signal is inappropriate, it is possible to regenerate or synthesize the signal. Therefore, since the reliability of the training data can be improved, the accuracy of the learning model can be improved, and the abnormal part of the device can be estimated with high accuracy.

[0060] In addition, the device state estimation method according to the 11th aspect of the present disclosure includes: a step of determining whether a measured signal obtained by measuring the operating condition of a device is normal or abnormal; a step of generating a generated signal during an abnormality by simulating the abnormality of the device; a step of generating a synthesized signal during an abnormality by synthesizing the measured signal determined to be normal, i.e., the measured signal during normal times, and the generated signal during an abnormality; a step of performing machine learning using the synthesized signal during an abnormality to thereby generate a learning model for estimating the abnormal part of the device; and a step of estimating the abnormal part of the device based on the measured signal determined to be abnormal, i.e., the measured signal during an abnormality, and the learning model. In the simulation, the generated signal during an abnormality is generated using the synthesized signal during an abnormality.

[0061] Thus, it is possible to accurately estimate the abnormal part of the device with high precision in the same manner as the above-described device state estimation apparatus.

[0062] In addition, the program according to the 12th aspect of the present disclosure is a program that causes a computer to execute the device state estimation method according to the 11th aspect.

[0063] Thus, it is possible to accurately estimate the abnormal part of the device with high precision in the same manner as the above-described device state estimation apparatus.

[0064] In addition, one aspect of the present disclosure can also be implemented as a computer-readable non-transitory recording medium storing the program.

[0065] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings.

[0066] In addition, the embodiments described below represent general or specific examples, and the present disclosure is not limited to the following embodiments. The numerical values, shapes, materials, structural elements, arrangement positions and connection manners of the structural elements, steps, order of steps, etc. shown in the following embodiments are examples and are not intended to limit the present disclosure. In addition, among the structural elements in the following embodiments, the structural elements not described in the independent claims are described as optional structural elements.

[0067] In addition, each drawing is a schematic diagram and is not necessarily strictly drawn. Therefore, for example, the scales in each drawing are not necessarily the same. In addition, in each drawing, the same reference numerals are given to substantially the same structures, and repeated explanations are omitted or simplified.

[0068] In addition, in this specification, ordinal numbers such as "first" and "second" do not mean the number or order of structural elements unless otherwise specified, but are used for the purpose of distinguishing between the same structural elements to avoid confusion.

[0069] (Embodiment)

[0070] [1. Structure]

[0071] First, with reference to Figure 1 the structure of the equipment state estimation system and the outline of the processing will be described.

[0072] Figure 1 is a block diagram showing the structure of the equipment state estimation system 1 according to the present embodiment. As Figure 1 shown, the equipment state estimation system 1 includes an equipment state estimation device 10, an equipment 20, and a sensor 30.

[0073] The equipment state estimation device 10 is a device that estimates the operating state of the equipment 20. Specifically, the equipment state estimation device 10 detects an abnormality of the equipment 20 and estimates the abnormal part of the detected abnormality. The abnormal part is information indicating the occurrence part of the abnormality, the cause of the abnormality, and / or the degree of these abnormalities.

[0074] As Figure 1 shown, the equipment state estimation device 10 includes an input / output unit 11, a storage unit 12, and a state estimation unit 100. In addition, the state estimation unit 100 includes: a determination unit 110, a composite signal generation unit 120, a simulation unit 130, a learning unit 140, and an abnormal part estimation unit 150. Specific processing of each structural element of the equipment state estimation device 10 will be described later.

[0075] The equipment 20 is a machine whose state is to be estimated. The equipment 20 is, for example, a rotating machine such as a motor or a generator. Alternatively, the equipment 20 may also be a mechanism in which a plurality of rotating machines are connected by a gearbox, a load, a chain, etc., or may be a mechanism such as a robot arm or a moving body having a plurality of rotating machines built therein.

[0076] The sensor 30 measures physical quantities such as vibration, electromagnetic waves, and / or heat generated in the equipment 20, and converts the measured physical quantities into signals that can be processed as electronic information. The sensor 30 inputs the converted signal into the determination unit 110 of the equipment state estimation device 10. The sensor 30 is, for example, assembled in the equipment 20, but is not limited thereto.

[0077] In addition, the equipment state estimation system 1 may include a plurality of sensors 30. For example, the plurality of sensors 30 measure physical quantities at different measurement parts of the equipment 20. In addition, the plurality of sensors 30 may also measure physical quantities of different types.

[0078] [1-1. Equipment State Estimation Device]

[0079] Next, each structural element of the equipment state estimation device 10 will be described.

[0080] The input / output unit 11 receives inputs such as setting information used in parameter identification and data synthesis from the user. In addition, the input / output unit 11 outputs information such as search results of parameters and the estimated state of the device. In the present embodiment, the input / output unit 11 has a display function of visualizing information and presenting it to the user. In addition, the user is a manager of the device state estimation system 1, a manager of the device 20, an operator, etc., but is not particularly limited.

[0081] The input / output unit 11 is implemented, for example, by an integrated input / output device having an input function and a display function such as a touch panel display. In addition, the output of information by the input / output unit 11 may be replaced by or in addition to display, and may be output as sound. Alternatively, the input / output unit 11 may have a communication function of outputting information to other machines (for example, a portable terminal or a display machine).

[0082] In addition, the input / output unit 11 may be implemented by two or more devices including an input device and an output device. The input device is, for example, a keyboard, a mouse, a touch sensor, a microphone, etc. The output device is, for example, a display device, a speaker, a communication interface device, etc.

[0083] The storage unit 12 stores data such as information and programs used by the state estimation unit 100. In addition, the storage unit 12 stores information generated by the state estimation unit 100 and data such as a machine learning model. The storage unit 12 is implemented by a non-volatile storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

[0084] The storage unit 12 may not be provided in the device state estimation device 10. For example, the storage unit 12 may be provided in an external server device accessible by the state estimation unit 100.

[0085] The state estimation unit 100 detects an abnormality of the device 20 based on the setting information input from the input / output unit 11, the signal acquired from the sensor 30, and the information read from the database stored in the storage unit 12. When the state estimation unit 100 detects an abnormality of the device 20, it estimates the abnormal part using a learning model. The state estimation unit 100 outputs various information such as the detection result of the abnormality and the estimation result of the abnormal part to the input / output unit 11.

[0086] The state estimation unit 100 is implemented by, for example, a computer machine including a processor. For example, the computer machine includes a non-volatile memory storing a program, a volatile memory which is a temporary storage area for executing the program, input / output ports, a processor for executing the program, and the like. The program executed by the processor may also be stored in the storage unit 12. Each processing unit included in the state estimation unit 100 is implemented in software by the processor. Alternatively, each processing unit included in the state estimation unit 100 may be implemented in hardware such as a dedicated or general-purpose integrated circuit.

[0087] The determination unit 110 determines whether the measured signal obtained by measuring the operating condition of the device 20 is normal or abnormal. The normality of the measured signal means that the device 20 being measured by the sensor 30 is in a normal state. The abnormality of the measured signal means that the device 20 is in an abnormal state. That is, in the present embodiment, the determination unit 110 uses the measured signal input from the sensor 30 to determine whether the device 20 is in a normal state or an abnormal state.

[0088] In the determination, for example, a machine learning model that has learned the characteristics of the measured signal in the normal state may also be used. In machine learning, various well-known algorithms can be used, such as autoencoders based on neural networks, support vector machines, random forests, and ensemble learning that combines them.

[0089] Based on the determination result, the determination unit 110 assigns a label of "normal time" or "abnormal time" to the measured signal. The determination unit 110 outputs the measured signal determined to be normal, that is, the measured signal at normal time, to the composite signal generation unit 120. In addition, the determination unit 110 outputs the measured signal determined to be abnormal, that is, the measured signal at abnormal time, to the abnormal part estimation unit 150.

[0090] The composite signal generation unit 120 generates an abnormal-time composite signal. The abnormal-time composite signal is a signal assumed to be output when the device 20 is abnormal and is used for machine learning by the learning unit 140. Specifically, the abnormal-time composite signal is a signal generated by synthesizing the measured signal at normal time input from the determination unit 110 and the abnormal-time generated signal input from the simulation unit 130. By performing the synthesis, it is possible to include the measured-based characteristics not considered in the simulation in the abnormal-time composite signal, thereby improving the accuracy of machine learning.

[0091] In addition, the composite signal generation unit 120 outputs physical parameters to the simulation unit 130. The composite signal generation unit 120 assigns a label of the abnormal part to the abnormal-time composite signal according to the simulation conditions and outputs it to the learning unit 140. In addition, the composite signal generation unit 120 outputs the generated abnormal-time composite signal to the simulation unit 130. The specific structure and processing of the composite signal generation unit 120 will be described later.

[0092] The simulation unit 130 generates a signal generated during an abnormality by simulating the abnormality of the device 20. Specifically, the simulation unit 130 generates a signal generated during an abnormality based on the physical parameters and signals provided from the composite signal generation unit 120. In the simulation, for example, an equivalent circuit model using dq conversion, a more detailed FEM (Finite Element Method) model, or a behavior model obtained by reducing the dimensionality of the FEM model can be used. The simulation unit 130 outputs the generated signal generated during an abnormality to the composite signal generation unit 120.

[0093] In the present embodiment, the simulation unit 130 uses the composite signal during an abnormality input from the composite signal generation unit 120 to generate a signal generated during an abnormality. That is, the simulation unit 130 receives feedback from the composite signal generation unit 120 and performs simulation to generate a signal generated during an abnormality. Specific generation processing of the signal generated during an abnormality will be described later.

[0094] The learning unit 140 performs machine learning by using the composite signal during an abnormality output from the composite signal generation unit 120 to generate a learning model for estimating the abnormal part of the device. Specifically, the learning unit 140 learns the characteristics of the composite signal during an abnormality to generate an abnormal part estimation model as an example of the learning model. In addition, the actually measured signal during normal operation, the actually measured signal during an abnormality, and the composite signal during an abnormality can also be used for learning.

[0095] The abnormal part estimation model is a mathematical model that takes a signal as an input and outputs an abnormal part. In the learning of the characteristics of the composite signal during an abnormality, for example, an autoencoder based on a neural network or the like, a support vector machine, a random forest, or ensemble learning combining them can be used. The learning unit 140 outputs the generated abnormal part estimation model to the abnormal part estimation unit 150.

[0096] The abnormal part estimation unit 150 is an example of the first estimation unit, and estimates the abnormal part of the device 20 based on the actually measured signal during an abnormality output from the determination unit 110 and the abnormal part estimation model output from the learning unit 140. Specifically, the abnormal part estimation unit 150 uses the abnormal part estimation model to evaluate the actually measured signal during an abnormality, thereby estimating the abnormal part of the device 20. That is, the abnormal part estimation unit 150 diagnoses the abnormal part of the device 20. The abnormal part estimation unit 150 outputs the estimation result (diagnosis result) to the input / output unit 11. The estimation result is presented to the user through the input / output unit 11.

[0097] [1-2. Composite Signal Generation Unit]

[0098] Next, refer to Figure 2An example of the structure of the composite signal generation unit 120 according to the present embodiment will be described. Figure 2 FIG. is a block diagram showing the structure of the composite signal generation unit 120 of the device state estimation device 10 according to the present embodiment. As Figure 2 shown, the composite signal generation unit 120 includes a feature estimation unit 121, a parameter identification unit 122, and a synthesis unit 123.

[0099] The feature estimation unit 121 is an example of the second estimation unit, and estimates a plurality of feature amounts of the input signal. The plurality of feature amounts include the frequency and phase of the harmonic components included in the signal. Specifically, the feature estimation unit 121 estimates the frequency and phase of the harmonic components included in the normal-time measured signal input from the determination unit 110 as the first frequency and the first phase. In addition, the feature estimation unit 121 estimates the frequency and phase of the harmonic components included in the abnormal-time generated signal input from the simulation unit 130 as the second frequency and the second phase. The feature estimation unit 121 outputs the calculated frequency and phase to the parameter identification unit 122.

[0100] The harmonic component may be, for example, a harmonic component caused by rotation. Here, the harmonic component caused by rotation refers to a harmonic component detected by the sensor 30 due to the rotation of a rotating mechanism such as a motor provided in the device 20.

[0101] Physical parameters required for estimating the frequency and phase of the harmonic component caused by rotation are provided by the user as "determined physical parameters" via the input / output unit 11, for example. The determined physical parameter is an example of the second physical parameter of the device 20. The determined physical parameter is a parameter whose value can be grasped in advance by the user according to the characteristics of the device 20 and the type of the sensor 30, etc. For example, when the sensor 30 is a current sensor, the power supply frequency, the number of poles, the number of rotor bar roots, etc. are used as the determined physical parameters. The determined physical parameter may also be stored in advance in the storage unit 12. The determined physical parameter is also used in the simulation performed by the simulation unit 130.

[0102] The parameter identification unit 122 identifies the "search physical parameter" of the device 20. The search physical parameter is an example of the first physical parameter of the device 20 and is a parameter of a different type from the determined physical parameter. The search physical parameter is a parameter whose value is uncertain according to the operating condition of the device 20 or cannot be measured by the sensor 30. For example, the search parameter is the rotor bar resistance, the motor radius, etc. The parameter identification unit 122 identifies the search physical parameter specified by the user via the input / output unit 11. The user sets the search range (upper limit value and lower limit value) of the search physical parameter via the input / output unit 11.

[0103] The parameter identification unit 122 identifies the search physical parameters based on the frequency and phase input from the feature estimation unit 121. Specifically, the parameter identification unit 122 performs the identification of the search physical parameters such that the difference in frequency, i.e., the frequency difference, between the frequency of the measured signal during normal operation (the first frequency) and the frequency of the generated signal during abnormal operation (the second frequency), and the difference in phase, i.e., the phase difference, between the phase of the measured signal during normal operation (the first phase) and the phase of the generated signal during abnormal operation (the second phase) are each less than the threshold value. The parameter identification unit 122 outputs the identified values of the search physical parameters to the simulation unit 130. The search physical parameters are used in the simulation performed by the simulation unit 130.

[0104] In addition, the parameter identification unit 122 determines the "abnormal physical parameter" of the device 20. The abnormal physical parameter is an example of the third physical parameter of the device 20 and is the physical parameter corresponding to the abnormal part of the device 20. For example, the abnormal physical parameter is the number of damaged rotor bars, the damage depth of the bearing, the damage depth of the gear, etc. The parameter identification unit 122 determines the value of the abnormal physical parameter within the set range of the abnormal physical parameter set by the user via the input / output unit 11.

[0105] The synthesis unit 123 synthesizes the measured signal during normal operation output from the determination unit 110 and the generated signal during abnormal operation input from the simulation unit 130, thereby generating a synthesized signal during abnormal operation. The synthesis is performed, for example, by weighted average as shown in the following equation (1).

[0106] (1) I fusion (t)=r×I sim (t)+(1-r)×I real (t)

[0107] Here, the time t is an example of the first time, and 0 can be used as the initial value, for example. I fusion (t) is the synthesized signal during abnormal operation at time t. I sim (t) is the generated signal during abnormal operation at time t. I real (t) is the measured signal during normal operation at time t. r is the synthesis ratio.

[0108] The synthesis ratio r is input via the input / output unit 11, for example. Alternatively, the synthesis ratio r can also be stored in the storage unit 12 as a predetermined fixed value. The synthesis unit 123 outputs the generated synthesized signal during abnormal operation to the learning unit 140 and the input / output unit 11. In addition, the synthesis unit 123 outputs the generated synthesized signal during abnormal operation as feedback to the simulation unit 130.

[0109] The method of synthesizing the normal-time measured signal and the abnormal-time generated signal by the synthesizing unit 123 is not limited to the foregoing method. For example, the synthesizing unit 123 may also sample and use the synthesis ratio according to a normal distribution with the synthesis ratio r input by the user set as the mean. Alternatively, the synthesizing unit 123 may use the synthesis ratio r input by the user as the initial value, and then evaluate the difference between the normal-time measured signal and the abnormal-time generated signal at each simulation step, and dynamically calculate the synthesis ratio for use.

[0110] The synthesis performed by the synthesizing unit 123 is so-called data assimilation. That is, the characteristics of the normal-time measured signal are assimilated with the abnormal-time generated signal. Thereby, the characteristics based on the phenomena not considered in the simulation can be included in the abnormal-time synthesized signal after synthesis.

[0111] [2. Operation]

[0112] Next, the operation of the device state estimation device 10 according to the present embodiment will be described. First, an example of the overall operation of the device state estimation device 10 will be described with reference to Figure 3 to describe an example of the overall operation of the device state estimation device 10. Figure 3 is a flowchart showing the operation of the device state estimation device 10 according to the present embodiment.

[0113] As Figure 3 shown, first, the input / output unit 11 receives the input of the setting information required for the processing of the state estimation unit 100 (S10). Specifically, the input / output unit 11 receives the input of the values of one or more determined physical parameters, the search ranges of one or more searched physical parameters, and the setting ranges of one or more abnormal physical parameters, etc. For specific input examples, use Figure 8 will be described later.

[0114] Next, the determination unit 110 determines whether the measured signal obtained by measuring the operating condition of the device 20 is normal or abnormal (S12). Specifically, the sensor 30 measures physical quantities such as the vibration and / or current of the device 20 and outputs them as measured signals to the device state estimation device 10. The determination unit 110 of the state estimation unit 100 acquires the measured signal output from the sensor 30 and determines whether the acquired measured signal is normal or abnormal, that is, determines whether the device 20 is in a normal state or an abnormal state. Based on the determination result, the determination unit 110 assigns a label of either "normal" or "abnormal" to the measured signal. The determination unit 110 outputs the normal-time measured signal to the synthesized signal generation unit 120. The determination unit 110 outputs the abnormal-time measured signal to the abnormal part estimation unit 150. In addition, the determination unit 110 outputs the normal-time measured signal and the abnormal-time measured signal to the input / output unit 11.

[0115] In addition, multiple measured signals may be output from the sensor 30. In this case, normal or abnormal determination (S12) and subsequent processing are performed separately for the multiple measured signals.

[0116] Next, the input / output unit 11 outputs the determination result (S14). Specifically, the input / output unit 11 displays the determination result indicating either "normal" or "abnormal" based on the tag information assigned to the measured signal. Thereby, the determination result can be presented to the user.

[0117] Next, when the determination result is normal (Yes in S16), the composite signal generation unit 120 performs identification of the search physical parameters (S18). Specifically, the composite signal generation unit 120 identifies the search physical parameters such that the frequencies and phases of the harmonic components of the normal-time measured signal output from the determination unit 110 and the abnormal-time generated signal output from the simulation unit 130 are respectively the same. For a specific example of the identification process, use Figure 4 will be described later.

[0118] Next, the composite signal generation unit 120 synthesizes the normal-time measured signal and the abnormal-time generated signal, thereby generating an abnormal-time composite signal (S20). For a specific example of the synthesis process, use Figure 6 will be described later.

[0119] Next, the learning unit 140 performs machine learning using the abnormal-time composite signal, thereby generating an abnormal part estimation model (S22). After generating the abnormal part estimation model, the equipment state estimation device 10 makes a determination for the next measured signal (S12) and executes the processing after step S12.

[0120] On the other hand, when the determination result is abnormal (No in S16), the abnormal part estimation unit 150 estimates the abnormal part of the equipment 20 based on the abnormal-time measured signal and the fault part estimation model (S24).

[0121] Next, the input / output unit 11 outputs the estimation result of the abnormal part (S26). Specifically, the input / output unit 11 can present the abnormal part of the equipment 20 to the user by displaying the estimation result. Since the user is informed of the occurrence of the abnormality and the abnormal part, the user can take measures such as repairing and restoring the equipment 20. Since the abnormal part is presented, the study and execution of the countermeasure method can be carried out quickly, thereby shortening the time required for the countermeasure, increasing the operation time, and improving the production efficiency.

[0122] After outputting the estimation result, the equipment state estimation device 10 ends the processing. Alternatively, the equipment state estimation device 10 may also make a determination for the next measured signal (S12) and execute the processing after step S12.

[0123] [Identification of parameters (S18)]

[0124] Next, with reference to Figure 4 the identification of parameters ( Figure 3 S18) in the operation of the equipment state estimation device 10 according to this embodiment will be specifically described. Figure 4 is a flowchart showing the processing related to the identification of parameters in the operation of the equipment state estimation device 10 according to this embodiment.

[0125] In addition, as Figure 4 the outline of the processing related to the identification of parameters shown, steps S181 to S182 are processing related to initial settings, steps S183 to S184 are processing related to the determination of physical parameters, steps S185 to S188 are processing related to the evaluation of physical parameters, and step S189 is processing related to the storage of physical parameters.

[0126] First, the feature estimation unit 121 estimates the frequency and phase of the harmonic components caused by rotation included in the measured signal during normal operation (S181). The measured signal during normal operation can be any one selected from the multiple measured signals during normal operation output by the determination unit 110, or multiple of them can be selected. When multiple signals are selected, the processing of steps S181 to S189 is performed for each of the selected multiple signals.

[0127] Next, the parameter identification unit 122 determines the value of the abnormal physical parameter within the set range (S182). The set range is the range set by the user via the input / output unit 11. The parameter identification unit 122 adopts one value from within the set range as the value of the abnormal physical parameter. At this time, the parameter identification unit 122 can also use the value of the abnormal physical parameter adopted in the past as a reference to determine the value of the new abnormal parameter. In addition, the value of the abnormal parameter can also be probabilistically determined from the range set by the user.

[0128] In addition, 0 can also be included in the set range of the abnormal physical parameter. The value of the abnormal physical parameter being 0 means that no abnormality has occurred. That is, when the value of the abnormal physical parameter is set to 0 for simulation, data generated during normal operation can be generated instead of data generated during an abnormality. The equipment state estimation device 10 can also learn the data generated during normal operation through machine learning and generate a machine learning model used in the determination of normal and / or abnormal conditions.

[0129] Next, the parameter identification unit 122 determines the value of the search physical parameter within the search range (S183). The search range is the search range set by the user via the input / output unit 11. The parameter identification unit 122 adopts a value from within the search range as the value of the search physical parameter. At this time, the parameter identification unit 122 may also use the value of the search physical parameter adopted in the past, or the frequency difference and / or phase difference calculated in step S187 described later as a reference to determine the value of the new search parameter. Additionally, the value of the search parameter may be probabilistically determined from the range set by the user.

[0130] Next, the parameter identification unit 122 evaluates the distance between the determined value of the search physical parameter and the value of the search physical parameter saved in step S189 in the past (S184). When the distance is lower than a preset threshold (No in S184), the parameter identification unit 122 determines the value of the search physical parameter again (S183). Thereby, a value that significantly deviates from the value determined in the past can be used for subsequent processing. That is, by not performing processing on values close to the value determined in the past, it is possible to comprehensively process various conditions while suppressing an increase in the amount of computation.

[0131] When the distance exceeds the preset threshold (Yes in S184), the parameter identification unit 122 outputs the values of the determined search physical parameter and the abnormal physical parameter to the simulation unit 130. The simulation unit 130 performs a simulation using the search physical parameter, the abnormal physical parameter, and the determined physical parameter, etc., to generate an abnormal-time generated signal (S185). The simulation unit 130 outputs the generated abnormal-time generated signal to the feature estimation unit 121.

[0132] Next, the feature estimation unit 121 estimates the frequency and phase of the harmonic components caused by rotation included in the abnormal-time generated signal (S186).

[0133] Next, the parameter identification unit 122 calculates the frequency difference and phase difference of the harmonic components caused by rotation included in the normal-time measured signal and the abnormal-time generated signal, respectively (S187).

[0134] Next, the parameter identification unit 122 determines whether each of the calculated frequency difference and phase difference is less than the threshold preset for each frequency difference and phase difference (S188). When at least one of the frequency difference and phase difference is not less than the threshold (No in S188), the parameter identification unit 122 returns to step S183 to determine the value of the search physical parameter again, and repeats the subsequent processing.

[0135] When both the frequency difference and the phase difference are less than the threshold value (Yes in S188), the parameter identification unit 122 stores the value of the search physical parameter determined in step S183 in the storage unit 12 (S189). Thus, when the search physical parameter stored in the storage unit 12 is used for simulation, the simulation unit 130 can generate an abnormal-time generated signal including harmonic components having frequencies and phases equivalent to those of the harmonic components included in the normal-time measured signal. In other words, in the abnormal-time generated signal, the harmonic components caused by the rotation of the device 20 can be reproduced.

[0136] Figure 5 FIG. is an example of an abnormal-time generated signal generated by the simulation unit 130 of the device state estimation apparatus 10 according to the present embodiment.

[0137] In Figure 5 the normal-time measured signal 201, the abnormal-time generated signals 211 to 213 before identification of the search physical parameter, and the abnormal-time generated signals 221 to 223 after identification of the search physical parameter are shown. Here, an example will be described in which the sensor 30 is a current sensor, and both the measured signal and the generated signal are signals representing the time change of the current value. Figure 5 Each of the signals shown is represented as a graph that defines the horizontal axis by time and the vertical axis by the current value. In addition, the abnormal-time generated signals 211 to 213 are signals generated by performing simulations of different abnormalities. The same applies to the abnormal-time generated signals 221 to 223.

[0138] The normal-time measured signal 201 includes a fundamental wave component 202 and a harmonic component 203 caused by rotation. Generally, the motor included in the device 20 operates through non-linear interactions such as induced electromotive force, magnetic flux density, and / or current. Therefore, even in the case of the device 20 operating normally, harmonic components and distortion due to the rotation of the motor appear in the normal-time measured signal 201.

[0139] In the abnormal-time generated signals 211 to 213 before identification of the search physical parameter, at least one of the frequency and the phase of the harmonic component 203 caused by rotation is different from that of the normal-time measured signal 201. In addition, the frequency of the harmonic component 203 corresponds to the number of times the harmonic component 203 appears in a given period (for example, 1 second), and is represented by the interval of the harmonic component 203 in Figure 5 The phase of the harmonic component 203 corresponds to the time when the harmonic component 203 appears, and is represented by the position on the horizontal axis in Figure 5

[0140] ​When at least one of the frequency and the phase is different, when the synthesizing unit 123 synthesizes the normal measured signal 201 and the generated signals 211, 212, or 213 during an abnormality, the characteristics of the frequency components inherent to the abnormality of the generated signals 211, 212, or 213 cannot be fully reflected in the synthesized signal during an abnormality. This is because the characteristics of the abnormal part of the device 20 are manifested in the frequency components formed by the superposition of the harmonic components caused by rotation and the frequency components inherent to the abnormality.

[0141] Among the generated signals 221 to 223 during an abnormality after identifying the physical parameters, the frequencies and phases of the harmonic components 203 caused by rotation are substantially the same as the frequencies and phases of the harmonic components 203 included in the normal measured signal 201, respectively. Specifically, as Figure 4 shown in step S188 of, the frequency difference and the phase difference between the normal measured signal 201 and the generated signals 221, 222, or 223 during an abnormality are each less than a threshold value. In this case, by synthesizing the normal measured signal 201 and the generated signals 221, 222, or 223 during an abnormality by the synthesizing unit 123, the frequency components formed by the superposition of the harmonic components 203 caused by rotation and the frequency components inherent to the abnormality can be reproduced.

[0142] In addition, in step S188, only the frequencies and phases of the harmonic components 203 caused by rotation are evaluated. Thus, the generated signals 221 to 223 during an abnormality with various characteristics can be generated, which helps to generate various training data for the abnormal part estimation model.

[0143] [2-2. Signal Synthesis (Assimilation)]

[0144] Next, with reference to Figure 6 the specific processing of the synthesis ( Figure 3 S20) in the operation of the device state estimation apparatus 10 according to the present embodiment will be described. Figure 6 is a flowchart showing the processing related to synthesis in the operation of the device state estimation apparatus 10 according to the present embodiment.

[0145] In addition, as a summary of the processing related to synthesis shown in Figure 6 steps S201 to S202 are processing related to initial settings, steps S203 to S207 are processing related to signal synthesis, and steps S208 to S209 are processing for verifying the validity of the synthesized signal.

[0146] First, the synthesizing unit 123 determines the synthesis ratio r (S201). As the synthesis ratio r, the default value stored in the storage unit 12 is adopted. Alternatively, the value input by the user via the input / output unit 11 may be used as the synthesis ratio r.

[0147] Next, the synthesis unit 123 obtains Figure 3 Step S18 (specifically, Figure 4 Specifically, by performing Figure 4 , thereby identifying multiple values ​​for each of the multiple search physical parameters in the storage unit 12. The synthesis unit 123 selects the identified value for each search physical parameter by referring to the storage unit 12, and sets the selected value group as a group of search physical parameters. In addition, when there is only one type of search physical parameter, the synthesis unit 123 determines the value of the one type of search physical parameter.

[0148] Next, the synthesis unit 123 advances the time t by ΔT (S203). Here, ΔT is an example of a given period, which is the step time when the simulation is executed. The time t+ΔT is an example of the second time after the given period from the first time. By reducing ΔT, the accuracy of the simulation can be improved. By increasing ΔT, the number of executions of the simulation can be reduced, and the amount of calculation can be reduced. A reduction in power consumption due to the reduction in the amount of calculation can be expected. ΔT is pre-set and stored in the storage unit 12. Alternatively, ΔT can also be set by the user via the input-output unit 11.

[0149] Next, the simulation unit 130 performs a simulation to generate an abnormality generation signal at time t+ΔT from the abnormality synthesis signal at time t ( S204 ). The simulation is performed based on the following equation (2), for example.

[0150] (2) I sim (t + ΔT) = f (I fusion (t)

[0151] Here, I fusion (t) is the abnormal synthetic signal at time t. sim (t+ΔT) is the abnormality generated signal at time t+ΔT. f() is a simulation model. The simulation model is created based on the types and values ​​of the determination physical parameters, search physical parameters, and abnormal physical parameters. By using the abnormality synthesized signal at time t, it is possible to generate an abnormality generated signal including a frequency component formed by superimposing the harmonic component included in the normal measured signal and the frequency component inherent to the abnormality.

[0152] Next, the synthesis unit 123 synthesizes the abnormality generated signal at time t+ΔT and the normal measurement signal at time t+ΔT at a synthesis ratio r ( S205 ). Synthesis is performed by weighted averaging as shown in the following equation (3), for example.

[0153] (3) I fusion (t + ΔT) = r × Isim (t + ΔT) + (1 - r) × I real (t + ΔT)

[0154] Here, I real (t + ΔT) is the measured signal at normal time at time t + ΔT.

[0155] Next, the synthesizing unit 123 determines whether the time t has exceeded a preset threshold (S206). The threshold here is preset to be the length of the abnormal-time synthesized signal required in the machine learning performed by the learning unit 140 and stored in the storage unit 12. Alternatively, the threshold can also be set by the user via the input / output unit 11.

[0156] As long as the time t has not exceeded the threshold (No in S206), the processes of steps S203 to S205 are repeatedly executed. When the time t has exceeded the threshold (Yes in S206), the synthesizing unit 123 determines whether the processes of steps S202 to S206 have been performed for all the sets of search physical parameters identified in Figure 3 step S18 (specifically, Figure 4 the process shown) (S207). When there is even one unprocessed set (No in S207), return to S202, select the unselected set, and execute the subsequent processes (S203 to S206).

[0157] When the processes have been performed for all the sets (Yes in S207), the synthesizing unit 123 outputs the abnormal-time synthesized signal to the input / output unit 11 (S208). The input / output unit 11 displays the abnormal-time synthesized signal generated by the synthesizing unit 123. At this time, the input / output unit 11 can also display the abnormal-time measured signal or the normal-time measured signal and the abnormal-time synthesized signal arranged or overlapped with the abnormal-time synthesized signal. Thus, it is possible to easily compare the abnormal-time synthesized signal with the measured signal. An example of a specific display will be described using Figure 9 will be described later.

[0158] Next, the synthesis unit 123 determines whether the generated abnormal synthesis signal is appropriate (S209). Specifically, the synthesis unit 123 determines appropriateness based on the user's judgment result input via the input / output unit 11. For example, the user confirms the abnormal synthesis signal displayed in the input / output unit 11 and evaluates the appropriateness of the abnormal synthesis signal. This evaluation is performed, for example, based on the user's experience and / or comparison with the specification of the device 20, etc. The user determines whether the abnormal synthesis signal is appropriate and inputs this judgment result via the input / output unit 11. The synthesis unit 123 determines appropriateness based on the judgment result input by the user. When it is determined that the abnormal synthesis signal is inappropriate (No in S209), the synthesis unit 123 returns to step S201, determines the synthesis ratio r again, and repeatedly executes the subsequent processing (S202 - S209). When it is determined that the synthesis signal is appropriate (Yes in S209), the synthesis process ends.

[0159] Figure 7 FIG. is an example of an abnormal synthesis signal generated by the synthesis unit 123 of the device state estimation apparatus 10 according to the present embodiment.

[0160] In Figure 7 shows the results of spectral decomposition of each of the normal-time measured signal 301, abnormal-time generated signal 311 without feedback, abnormal-time synthesized signal 321 without feedback, and abnormal-time synthesized signal 331 with feedback. Each signal is represented as a graph with the horizontal axis defined by frequency and the vertical axis defined by signal intensity.

[0161] As Figure 7 shown, the normal-time measured signal 301 includes a fundamental wave component 302 and a harmonic component 303. In addition, the abnormal-time generated signal 311 includes a fundamental wave component 302 and a frequency component 312 inherent to the abnormality.

[0162] In Figure 7 here, the so-called "without feedback" means that the analog unit 130 does not use the abnormal synthesis signal. That is, "without feedback" is a case where the abnormal-time generated signal 311 is generated by performing simulation without using the abnormal synthesis signal at time t in the Figure 6 step S204. On the other hand, "with feedback" means that the analog unit 130 uses the abnormal synthesis signal. That is, "with feedback" is a case where the abnormal-time generated signal is generated according to the Figure 6 processing shown.

[0163] When there is an abnormality without feedback, the abnormal synthesized signal 321 includes a fundamental component 302, a harmonic component 303, and a frequency component 312 inherent to the abnormality. On the other hand, when there is an abnormality with feedback, the abnormal synthesized signal 331 includes a fundamental component 302, a harmonic component 303, a frequency component 312 inherent to the abnormality, and a superimposed frequency component 333. The superimposed frequency component 333 is a frequency component formed by superimposing the harmonic component 303 included in the normal measured signal 301 and the frequency component 312 inherent to the abnormality included in the abnormal generated signal 311.

[0164] In Figure 6 In the process of step S204 shown, when the abnormal synthesized signal generated by the synthesizing unit 123 is not fed back to the analog unit 130, the harmonic component 303 included in the normal measured signal 301 and the frequency component 312 inherent to the abnormality included in the abnormal generated signal 311 are simply added together to obtain Figure 7 the abnormal synthesized signal 321 shown.

[0165] On the other hand, as in the present embodiment, when Figure 6 In the process of step S204 shown, when the abnormal synthesized signal generated by the synthesizing unit 123 is fed back to the analog unit 130, an abnormal generated signal can be generated. This abnormal generated signal includes, in addition to the harmonic component 303 included in the normal measured signal 301 and the frequency component 312 inherent to the abnormality included in the abnormal generated signal, a frequency component formed by superimposing the harmonic component 303 and the frequency component 312 inherent to the abnormality. As a result, as Figure 7 shown, the abnormal synthesized signal 331 including the superimposed frequency component 333 is obtained.

[0166] In this way, according to the device state estimation device 10 according to the present embodiment, an abnormal synthesized signal 331 having a non-linear characteristic formed by superimposing a harmonic component and a frequency component inherent to the abnormality can be generated. Therefore, it helps to generate various training data for the abnormal part estimation model.

[0167] [3. Input / Output Screen for Physical Parameters]

[0168] Next, an example of the input / output screen for physical parameters will be described with reference to Figure 8 to.

[0169] Figure 8 is a diagram showing an example of the input / output screen for physical parameters displayed on the input / output unit 11 of the device state estimation device 10 according to the present embodiment. As Figure 8 shown, the input / output screen 400 includes a determination physical parameter setting area 401, an abnormal physical parameter setting area 402, and a search physical parameter setting area 403.

[0170] In the physical parameter setting area 401 for determination, a parameter value input box 411 is displayed. The parameter value input box 411 is set for each type of physical parameter to be determined. The parameter value input box 411 is a text box that receives input of text (numerical value) from the user. For example, the user inputs the value of the physical parameter to be determined into the parameter value input box 411. In addition, various GUI objects such as a list box, radio button, slider, etc. can be used instead of the text box.

[0171] In the abnormal physical parameter setting area 402, a lower limit value input box 421 and an upper limit value input box 422 are displayed. The lower limit value input box 421 and the upper limit value input box 422 are respectively set for each type of abnormal physical parameter. The lower limit value input box 421 and the upper limit value input box 422 are respectively text boxes that receive input of text (numerical value) from the user. For example, the user inputs the acceptable setting range (specifically, the upper limit value and the lower limit value) of the abnormal physical parameter into the lower limit value input box 421 and the upper limit value input box 422. In addition, various GUI objects such as a list box, radio button, slider, etc. can be used instead of the text box.

[0172] In the physical parameter setting area 403 for search, a lower limit value input box 431 and an upper limit value input box 432 are displayed. The lower limit value input box 431 and the upper limit value input box 432 are respectively set for each type of physical parameter to be searched. The lower limit value input box 431 and the upper limit value input box 432 are respectively text boxes that receive input of text (numerical value) from the user. For example, the user inputs the search range (specifically, the upper limit value and the lower limit value) of the physical parameter to be searched into the lower limit value input box 431 and the upper limit value input box 432. In addition, various GUI objects such as a list box, radio button, slider, etc. can be used instead of the text box.

[0173] In addition, in the physical parameter setting area 403 for search, a search result 433 is displayed. The search result 433 is a histogram representing the distribution of the physical parameter to be searched identified by Figure 3 step S18. In addition, the display form of the search result 433 is not limited to a chart such as a histogram.

[0174] By displaying the search result 433, the user can evaluate whether the identification result of the physical parameter to be searched is appropriate. In the case where the identification result of the physical parameter to be searched is inappropriate, the upper limit value or lower limit value, etc. of the physical parameter to be searched is changed, and thus the identification of the physical parameter to be searched can be performed again.

[0175] Thus, in the present embodiment, the input / output unit 11 receives an input for determining the value of a physical parameter via the input / output screen 400. In addition, the input / output unit 11 receives an input for searching the search range of a physical parameter. In addition, the input / output unit 11 receives an input for setting the range of an abnormal physical parameter. In addition, the input / output unit 11 outputs the identified physical parameter to be searched. Furthermore, Figure 8 The structure of the input / output screen 400 shown in [is] only an example and is not limited to the illustrated example. In addition, an example of input / output using the GUI displayed on the screen is shown here, but the input / output unit 11 may also receive an input of information based on other input means such as voice input. In addition, the information received and output by the input / output unit 11 is not limited to the above examples.

[0176] [4. Display Screen of Synthesis Result]

[0177] Next, with reference to Figure 9 an example of the display screen of the synthesis result will be described.

[0178] Figure 9 An example of a GUI object for input of the synthesis ratio and a display screen of the synthesis result displayed by the input / output unit 11 of the device state estimation device 10 according to the present embodiment is shown. As Figure 9 shown, the display screen 500 includes a synthesis ratio setting area 501, a frequency spectrum display area 502, and a signal feature amount display area 503.

[0179] In the synthesis ratio setting area 501, a synthesis ratio input bar 511 is displayed. The synthesis ratio input bar 511 is provided for each type (abnormal part) of the abnormal physical parameter. The synthesis ratio input bar 511 is a slider that receives an input of the synthesis ratio r from the user. For example, the user can observe the content displayed in the frequency spectrum display area 502 and / or the signal feature amount display area 503, and use the synthesis ratio input bar 511 to set the value of the synthesis ratio r for each abnormal part. In addition, various GUI objects such as a text box, a list box, and a radio button may be used instead of the slider.

[0180] In the frequency spectrum display area 502, the frequency spectra of the actually measured signal 521 in the normal state, the generated signal 522 in the abnormal state, and the synthesized signal 523 in the abnormal state are displayed for each abnormal part. The actually measured signal 521 (solid line) in the normal state and the generated signal 522 (dashed line) in the abnormal state are overlapped and shown in the same graph.

[0181] The user can confirm by comparing the frequency spectra displayed in the frequency spectrum display area 502 that by Figure 6The synthesis performed in step S205 can reproduce, in the abnormal-time synthesis signal 523, the frequency components formed by superimposing the harmonic components caused by rotation and the frequency components inherent to the abnormality.

[0182] In the signal feature quantity display area 503, the distributions of the signal feature quantities of the signal feature quantity selection box 531, the actually measured signal 532 during abnormality, and the synthesized signal 533 during abnormality are displayed. The signal feature quantity selection box 531 is a check box, and the user can select the signal feature quantity to be displayed. The signal feature quantity is a feature quantity determined based on the time change of the physical quantity (here, current) measured by the sensor 30. For example, the user displays the distribution of the signal feature quantity of the actually measured signal 532 during abnormality and the distribution of the signal feature quantity of the synthesized signal 533 during abnormality according to the feature quantity selected in the signal feature quantity selection box 531. The actually measured signal 532 during abnormality can be displayed according to each abnormal part, or the same distribution can be displayed at each abnormal part regardless of the abnormal part.

[0183] By comparing the distributions of these feature quantities, the user can confirm whether various abnormal-time synthesized signals can be generated by synthesizing the actually measured signal during normal time and the generated signal during abnormality. Since the abnormal-time synthesized signal is used as training data for machine learning, in the case where various abnormal-time synthesized signals can be generated, various abnormalities can be assumed, and the accuracy of the abnormal part estimation model can be improved.

[0184] In this way, in the present embodiment, the input / output unit 11 receives the input of the synthesis ratio via the display screen 500. In addition, the input / output unit 11 outputs (displays) various signals such as the generated signal during abnormality, the synthesized signal during abnormality, and the actually measured signal during normal time. In addition, Figure 9 The structure of the display screen 500 shown in is only an example and is not limited to the illustrated example. In addition, an example of input / output using the GUI displayed on the screen is shown here, but the input / output unit 11 can also receive the input of information based on other input means such as voice input. In addition, the information received and output by the input / output unit 11 is not limited to the above examples.

[0185] As described above, according to the equipment state estimation device 10 according to the present embodiment, even in a situation where data during abnormality is insufficient, various abnormal-time data reproducing non-linear feature quantities can be generated. Therefore, the equipment state estimation device 10 can estimate the abnormal part with high accuracy.

[0186] (Others)

[0187] As described above, the apparatus state estimation apparatus and the apparatus state estimation method related to one or more aspects have been described with reference to the accompanying drawings and based on various embodiments. However, it goes without saying that the present disclosure is not limited to these embodiments. As long as it does not deviate from the gist of the present disclosure, the aspects obtained by applying various changes or modifications conceived by those skilled in the art to these embodiments and the aspects constructed by combining the structural elements in different embodiments are also included in the scope of the present disclosure.

[0188] For example, the communication method between the devices (e.g., between the apparatus state estimation apparatus 10 and the sensor 30) described in the above embodiments is not particularly limited. In the case of wireless communication between the devices, the wireless communication method (communication standard) is, for example, short-range wireless communication such as ZigBee (registered trademark), Bluetooth (registered trademark), or wireless LAN (Local Area Network), etc. Alternatively, the wireless communication method (communication standard) can also be communication via a wide-area communication network such as the Internet. In addition, wired communication can be performed between the devices instead of wireless communication. Specifically, the wired communication is power line transmission communication (PLC: Power Line Communication) or communication using a wired LAN, etc.

[0189] In addition, in the above embodiments, the processing performed by a specific processing unit can be performed by other processing units. In addition, the order of multiple processes can be changed, or multiple processes can be executed in parallel. In addition, the distribution of the structural elements included in the apparatus state estimation system 1 to multiple devices is an example. For example, the structural elements included in one device can be provided in other devices. In addition, the apparatus state estimation system can also be implemented as a single device.

[0190] For example, the processing described in the above embodiments can be implemented by centralized processing using a single device (system), or can also be implemented by distributed processing using multiple devices. In addition, the processor that executes the above program can be single or multiple. That is, centralized processing can be performed, or distributed processing can also be performed.

[0191] In addition, in the above embodiments, all or part of the structural elements such as the control unit can be constituted by dedicated hardware, or can also be implemented by executing software programs suitable for each structural element. Each structural element can also be implemented by a program execution unit such as a CPU (Central Processing Unit) or a processor reading and executing the software programs recorded in a recording medium such as an HDD or a semiconductor memory.

[0192] In addition, each functional block used in the description of the above embodiments is typically implemented as an integrated circuit, i.e., LSI (Large Scale Integration). The integrated circuit controls each functional block used in the description of the above embodiments and may also have an input section and an output section. They may be individually monolithic or may include some or all of them monolithically. Here, as LSI, depending on the degree of integration, it is sometimes referred to as IC, system LSI, super LSI, or ultra LSI.

[0193] In addition, the method of integrating into an integrated circuit is not limited to LSI, and dedicated circuits or general-purpose processors can also be used for implementation. FPGA (Field Programmable Gate Array) that can be programmed after LSI manufacturing, or a reconfigurable processor that can reconfigure the connection or setting of circuit units inside the reconfigurable LSI can also be utilized.

[0194] Furthermore, if an integrated circuit technology that can replace LSI emerges with the progress of semiconductor technology or other derived technologies, then of course, this technology can also be used for the integration of functional blocks. For example, it may be possible to apply biotechnology, optical integrated circuits, etc.

[0195] In addition, the integrity or specific manner of the present disclosure can also be implemented by a system, device, method, integrated circuit, or computer program. Or, it can also be implemented by a computer-readable non-transitory recording medium such as an optical disc, HDD, or semiconductor memory storing the computer program. In addition, it can also be implemented by any combination of a system, device, method, integrated circuit, computer program, and recording medium.

[0196] In addition, the above embodiments can be variously modified, substituted, added, omitted, etc. within the scope of the claims and their equivalents.

[0197] Industrial Applicability

[0198] The present disclosure can be used as a device and method for estimating the state of a device and is useful, for example, in a diagnostic system for detecting abnormalities such as device failures and estimating the abnormal parts.

[0199] Symbolic Explanation

[0200] 1 Device state estimation system

[0201] 10 Device state estimation device

[0202] 11 Input / output section

[0203] 12 Storage section

[0204] 20 Device

[0205] 30 Sensor

[0206] 100 State Estimation Unit

[0207] 110 Judgment Unit

[0208] 120 Composite Signal Generation Unit

[0209] 121 Feature Estimation Unit

[0210] 122 Parameter Identification Unit

[0211] 123 Composition Unit

[0212] 130 Simulation Unit

[0213] 140 Learning Unit

[0214] 150 Abnormal Location Estimation Unit

[0215] 201, 301, 521 Normal-Time Measured Signals

[0216] 202, 302 Fundamental Wave Components

[0217] 203, 303 Harmonic Components

[0218] 211, 212, 213, 221, 222, 223, 311, 522 Abnormal-Time Generated Signals

[0219] 312 Abnormality-Inherent Frequency Components

[0220] 321, 331, 523, 533 Abnormal-Time Composite Signals

[0221] 333 Superimposed Frequency Components

[0222] 400 Input / Output Screen

[0223] 401 Physical Parameter Setting Area for Decision

[0224] 402 Physical Parameter Setting Area for Abnormality

[0225] 403 Physical Parameter Searching Area

[0226] 411 Parameter Value Input Box

[0227] 421, 431 Lower Limit Value Input Boxes

[0228] 422, 432 Upper Limit Value Input Boxes

[0229] 433 Search Results

[0230] 500 Display screen

[0231] 501 Synthesis ratio setting area

[0232] 502 Spectrum display area

[0233] 503 Signal characteristic quantity display area

[0234] 511 Synthesis ratio input bar

[0235] 531 Signal characteristic quantity selection box

[0236] 532 Measured signal during abnormality.

Claims

1. A device state estimation apparatus, comprising: a determination unit that determines whether a measured signal obtained by measuring the operating condition of a device is normal or abnormal; a simulation unit that generates a generated signal during an abnormality by simulating the abnormality of the device; a synthesis unit that synthesizes the measured signal determined to be normal by the determination unit, i.e., the measured signal during normal operation, and the generated signal during an abnormality, thereby generating a synthesized signal during an abnormality; a learning unit that performs machine learning using the synthesized signal during an abnormality, thereby generating a learning model for estimating the abnormal part of the device; and a first estimation unit that estimates the abnormal part of the device based on the measured signal determined to be abnormal by the determination unit, i.e., the measured signal during an abnormality, and the learning model, wherein the simulation unit uses the synthesized signal during an abnormality to generate the generated signal during an abnormality.

2. The device state estimation apparatus according to claim 1, wherein the device state estimation apparatus comprises: a second estimation unit that estimates the frequency and phase of a harmonic component included in the measured signal during normal operation as a first frequency and a first phase; and an identification unit that identifies a first physical parameter of the device based on the first frequency and the first phase, wherein the simulation unit performs the simulation using the first physical parameter identified by the identification unit.

3. The device state estimation apparatus according to claim 2, wherein the second estimation unit further estimates the frequency and phase of a harmonic component included in the generated signal during an abnormality as a second frequency and a second phase, and the identification unit identifies the first physical parameter such that a frequency difference, which is a difference between the first frequency and the second frequency, and a phase difference, which is a difference between the first phase and the second phase, are each less than a threshold value.

4. The device state estimation apparatus according to claim 2 or 3, wherein the device state estimation apparatus comprises: an input unit that receives an input of a search range of the first physical parameter, and the identification unit identifies the first physical parameter within the search range.

5. The device state estimation apparatus according to claim 4, wherein the input unit receives an input of a value of a second physical parameter of a different type from the first physical parameter, and the second estimation unit uses the value received by the input unit to estimate the first frequency and the first phase.

6. The device state estimation apparatus according to claim 4, wherein the device state estimation apparatus comprises: an output unit that outputs the first physical parameter identified by the identification unit.

7. The device state estimation apparatus according to claim 4, wherein the input unit receives an input of a setting range of an abnormal parameter of the device, and the simulation unit performs the simulation within the setting range of the abnormal parameter, thereby generating the generated signal during an abnormality.

8. The device state estimation apparatus according to any one of claims 1 to 3, wherein the simulation unit performs the simulation using the synthesized signal during an abnormality at a first time, thereby generating the generated signal during an abnormality at a second time after a given period from the first time.

9. The device state estimation device according to any one of claims 1 to 3, wherein the device state estimation device includes: an input unit that receives an input of a synthesis ratio of the normal-time measured signal and the abnormal-time generated signal; the synthesis unit synthesizes the normal-time measured signal and the abnormal-time generated signal at the synthesis ratio, thereby generating the abnormal-time synthesized signal.

10. The device state estimation device according to any one of claims 1 to 3, wherein the device state estimation device includes: an output unit that outputs the normal-time measured signal, the abnormal-time generated signal, and the abnormal-time synthesized signal.

11. A device state estimation method, comprising: a step of determining whether a measured signal obtained by measuring the operating condition of a device is normal or abnormal; a step of generating an abnormal-time generated signal by simulating an abnormality of the device; a step of generating an abnormal-time synthesized signal by synthesizing the measured signal determined to be normal, i.e., the normal-time measured signal, and the abnormal-time generated signal; a step of performing machine learning using the abnormal-time synthesized signal, thereby generating a learning model for estimating an abnormal part of the device; and a step of estimating the abnormal part of the device based on the measured signal determined to be abnormal, i.e., the abnormal-time measured signal, and the learning model, in the simulation, the abnormal-time synthesized signal is used to generate the abnormal-time generated signal.

12. A program that causes a computer to execute the device state estimation method according to claim 11.

Images